1985_Siliconix_MOSPOWER_Applications 1985 Siliconix MOSPOWER Applications
User Manual: 1985_Siliconix_MOSPOWER_Applications
Open the PDF directly: View PDF 
.
Page Count: 512
| Download | |
| Open PDF In Browser | View PDF |
H
MOSPOWER
Applications
Handbook
Editor in Chief
Rudy Severns
Senior Staff Engineer
Assistant Editor
Jack Armijos
Senior Applications Engineer
Siliconix
incorporated
Santa Clara, California 95054
Library of Congress Cataloging in Publication Data
Main entry under title:
MOSPOWER applications handbook.
Bibliography: p.
Includes index.
1. Metal oxide semiconductor field-effect transistors-Handbooks, manuals, etc. 2. Power transistors--Handbooks,
manuals, etc. I. Severns, Rudolf P. II. Armijos.
Jack, 1949III. Title: M.O.S.P.O.W.E.R. applications handbook.
TK7871.95.M67 1984
621 31'7
84-14140
ISBN 0-930519-00-0
Editorial production/supervision: Robin Berliner
Interior design: Robin Berliner
Cover design: Thomas Washburn Associates
©
1984 by Siliconix incorporated. All Rights Reserved. This book, or parts thereof, may not be
reproduced in any form without permission from Siliconix.
Information contained in this book is believed to be accurate and reliable. However, we cannot
accept responsibility for its use, nor for any infringement of patents at rights of others which may
result from its use. Publication of this information does not imply any authority or license for the
use of any patented feature.
Printed in the United Stales of America
10 9 8 7 6 5 4 3 2 I
ISBN 0-930519-00-0
ii
Contributing Authors
Edward R. Abramczyk
Mark Alexander
George Allen
Dr. Richard H. Baker
Dr. Richard Blanchard
Gordon (Ed) Bloom
Dr. Dan Y. Chen
Robert R. Cordell
C.P. Der
Phil Ekstrom
Alton Eris
Z.D.Fang
W. E. Frank
David Giandomenico
James Harnden
Barry Harvey
Michael Herrick
John G. Kassakian
Steve Kent
DavidLau
Dr. Fred C. Lee
DavidMele
iii
Dr. R. D. Middlebrook
Ramesh Oruganti
Ed Oxner
George Riehm
William Roehr
Gary Rothrock
Ray Ruble
Rudy Severns
P. D. Wesel
Richard Williams
Acknowiedgement
This book is the result of the efforts of many people.
We have elected not to identify every contribution by
scattering the authors' names throughout the text.
Rather we have grouped them here for convenience.
The following is an alphabetical listing of the authors
and their contributions:
Contributing Authors
Sections
Edward R. Abramczyk
Sr. Design Engineer
Siliconix incorporated
Santa Clara, CA
2.9
2.9.1
Mark Alexander
Applications Engineer
Siliconix incorporated
Santa Clara, CA
2.9
2.9.1
2.12
3.1
4.2
5.1
5.2
5.6
5.6.2
6.6.4
6.10
6.11
6.14.2
A.1
A.2
George Allen
Testing Engineer
Siliconix incorporated
Santa Clara, CA
7.1
iv
Dr. Richard H. Baker
Gould Inc.
Andover,MA
6.1.1
Dr. Richard Blanchard
Vice-President of Design and
Advanced Technology Development
Siliconix incorporated
Santa Clara, CA
1.3
2.9
2.9.1
2.11
4.2
5.6
5.6.2
7.1
Gordon (Ed) Bloom
E. &J. Bloom Associates
Arcadia,CA
6.1.2
6.1.3
Dr. Dan Y. Chen
Assistant Professor of Engineering
Virginia Polytechnic Institute
and State University
Blacksburg, VA
5.3.4
5.5
6.7
6.8
6.9
Robert R. Cordell
Tinton Falls, NJ
6.6.3
C.F.Der
Westinghouse Electric Corp.
Sykesville, MD
6.2
PhUEkstrom
Chief Scientist
Northwest Marine Technology
Shaw Island, WA
6.12
Alton Eris
Litton Industries
Woodland Hills, CA
6.1.2
Ramesh Oruganti
Massachusetts Institute of Technology
Cambridge, MA
2.13.1
Z.D.Fang
Virginia Polytechnic Institute
and State University
Blacksburg, VA
6.9
Ed Oxner
Staff Engineer
Siliconix incorporated
Santa Clara, CA
W.E.Frank
Westinghouse Electric Corp.
Sykesville, MD
6.2
David Giandomenico
Graduate Student
University of California
Berkeley, CA
5.3.1
James Harnden
FineH Systems
SanJose,CA
6.12
6.14.1
6.14.2
1.1
1.2
2.2.1
2.3
2.4
2.5
2.6
2.7
2.10
2.13.2
3.2
5.3
5.6.1
A.3
A.4
A.5
6.1.5
George Riehm
NEC Electronics
Sunnyvale, CA
6.1.4
Barry Harvey
Advanced Micro Devices
Santa Clara, CA
Michael Herrick
NewYork,NY
6.1.6
William Roehr
Applications Manager
General Semiconductor Industries
Tempe,AZ
6.3
6.6.1
6.6.2
John G. Kassakian
Laboratory for Electromagnetic
and Electronic Systems
Massachusetts Institute of Technology
Cambridge, MA
5.3.2
5.3.3
Gary Rothrock
Fairchild
Sunnyvale, CA
6.13
Steve Kent
Cypress Semiconductor
SanJose,CA
7.2
7.4
Ray Ruble
Sr. Applications Engineer
Siliconix incorporated
Santa Clara, CA
2.8
6.5
Rudy Severns
Sr. Staff Engineer
Siliconix incorporated
Santa Clara, CA
Preface
2.1
2.13.2
4.1
5.3
5.4
5.6
5.7
6.1.3
6.1.4
P.O. Wesel
Virginia Polytechnic Institute
and State University
Blacksburg, VA
6.8
Richard Williams
IC Design Engineer
Siliconix incorporated
Santa Clara, CA
5.2
DavidLau
Student
Massachusetts Institute of Technology
Cambridge, MA
·5.3.3
Dr. Fred C. Lee
Professor of Engineering
Virginia Polytechnic Institute
and State University
Blacksburg, VA
2.13.1
5.3.4
5.5
6.7
6.8
6.9
Dave Mele
MacLeod Labs, Inc.
SanJose,CA
6.1.6
Dr. R. D. Middlebrook
Professor of Engineering
California Institute of Technology
Pasadena, CA
2.2
v
A book is a creation not only of the authors but also
of the support people who made the finished book a
reality and without whose help the book would never
have been printed. Special thanks are due to Susan
Scott for her endless hours of editing for English
(engineers really can't spell!); to Ray Lubow, Ray
Ruble, and Ed Oxner for technical editing; to Pam
Lewis, Susan Hamel, and Linda Cosner for their
seemingly unlimited patience in typing and retyping
and retyping ... the manuscript; and to Robin Berliner
who took the manuscript and the drawings (which
most closely resembled chicken scratchings) and turned
them into a beautifully finished text.
A substantial portion of the credit for the book
belongs to Jack Armijos who did the real grunt work
of collating material, hounding the authors and editors to finish their work, and in general, seeing to it
that the project kept moving.
Rudy Severns
July, 1984
vi
Introduction to MOSPOWER FETs
I11III
Power MOSFET Structures _
MOSFET Electrical Characteristics _
Thermal Design and SOA _ _
Practical Design Considerations _
Applications Information _
I
MOSFET Testing and Reliability _
Appendices _
i
Index _
vii
viii
Table of Contents
Chapter 1 Introduction to MOSPOWER FETs
1.1
1.2
1.3
Page
General Remarks .............................................................. 1-1
A Little History ............................................................... 1-4
Fundamental Limitations of Power MOS Transistor Performance ....................... 1-6
Chapter 2 Power MOSFET Structures
2.1
2.2
2.2.1
2.3
2.4
2.5
2.6
2.7
2.8
2.9
2.9.1
2.10
2.11
2.12
2.13
2.13.1
2.13.2
Principles of MOSFET Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 2-1
FETs and BITs as Charge-Controlled Devices. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 2-4
Charge Transfer Characteristics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 2-11
Planar MOS ................................................................. 2-17
DMOS (Double Diffused MOS) ................................................. 2-19
V-Groove MOS .............................................................. 2-21
Lateral Power DMOS (LDMOS) ................................................. 2-23
Vertical DMOS (VDMOS) ..................................................... 2-24
The Bilateral MOSFET ........................................................ 2-26
Depletion Mode Power MOSFETs: New Devices to Solve Old Problems ................ 2-33
Fabrication of Depletion Mode Power MOSFETs ................................... 2-39
Other Structures .............................................................. 2-41
Power Control with Integrated CMOS Logic and DMOS Output. . . . . . . . . . . . . . . . . . . . . .. 2-43
Drain-to-Source Breakdown and Leakage in Power MOSFETs ........................ 2-49
Modeling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 2-56
SPICE 2 Simulation of High Power MOSFETs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 2-56
The High Frequency Model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 2-70
Chapter 3 MOSFET Electrical Characteristics
3.1
3.2
Linear Operation ............................................................... 3-1
Switching Characteristics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 3-9
Chapter 4 Thermal Design and SOA
4.1
4.2
Safe Operating Area and Thermal Design for MOSPOWER Transistors . . . . . . . . . . . . . . . . .. 4-1
HP-41CV Power MOSFET Thermal Analysis Program .............................. 4-17
ix
Chapter 5 Practical Design Considerations
5.1
5.2
5.3
5.3.1
5.3.2
5.3.3
5.3.4
5.4
5.5
5.6
5.6.1
5.6.2
5.7
Page
High Speed Gate Drive Circuits .................................................. 5-1
Temperature Compensated Biasing for Power MOSFETs in Linear Amplifiers ............ 5-9
Parallel Operation of Power MOSFETs ........................................... 5-15
Anomalous Oscillations and Turn-Off Behavior in a Vertical Power MOSFET ........... 5-33
Thermally Forced Current Sharing in Paralleled Power MOSFETs ..................... 5-43
An Analysis and Experimental Verification of Parasitic Oscillations in Paralieled
Power MOSFETs ........................................................... 5-48
Power FET Paralleling ............ : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 5-55
dVDsldt Turn-On in MOSFETs ................................................. 5-57
Inverse Diodes of Power MOSFETs .............................................. 5-65
MOSFETs Move In on Low Voltage Rectification .................................. 5-69
Using Power MOSFETs as High Efficiency Synchronous and Bridge Rectifiers in
Switch Mode Power Supplies ................................................. 5-87
Use MOSPOWER Transistors as Synchronous Rectifiers in Switched
Mode Power Supplies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 5-95
Power MOSFETs and Radiation Environments .................................... 5-101
Chapter 6 Applications Information
6.1
6.1.1
6.1.2
6.1.3
6.1.4
6.1.5
6.1.6
6.2
6.3
6.4
6.5
6.6
6.6.1
6.6.2
6.6.3
6.6.4
6.7
6.8
6.9
6.10
6.11
6.12
6.13
6.13.1
6.13.2
6.13.3
6.13.4
6.13.5
6.13.6
6.13.7
6.13.8
6.13.9
Power Supply Applications ...................................................... 6-1
High Frequency Power Conversion with FET Controlled Resonant Charge Transfer . . . . . . .. 6-1
Practical Design Considerations for a Multi-Output eUK Converter ..................... 6-8
The Generalized Use of Integrated Magnetics and Zero-Ripple Techniques
in Switch Mode Power Converters ............................................. 6-25
Siliconix Switch Mode Power Supply Kit ......................................... 6-49
A 500 kHz Switching Inverter for 12V Systems ................................... , 6-59
A Low Cost Regulator for Microprocessor Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 6-62
Solid State RF Generators for Induction Heating Applications ......................... 6-71
Using Power MOSFET Transistors to Interface from IC Logic to
High Power Loads ................................................ , ......... 6-81
MPP500; The First Single Package Complementary MOSPOWER Device ............... 6-89
Applying 240 V MOSPOWER Transistors and Current Limiting Diodes to
Electronic Pulse Dialer Circuits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 6-93
Audio Amplifiers ............................................................. 6-97
The Autobias Amplifier. A New (AC Coupled) Topology for Automatically Biased Audio
Amplifier Using Power MOSFETs ............................................. 6-97
A Simple Direct-Coupled Power MOSFET Audio Amplifier Topology
Featuring Bias Stabilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 6-105
A MOSFET Power Amplifier with Error Correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 6-111
Boost Op Amp Output Power With Complementary Power MOSFETs . . . . . . . . . . . . . . . .. 6-125
Bipolar-FET Combinational Devices ............................................. 6-130
Power FET Base Drive Circuit for High Power Darlington Transistors ................. 6-135
A High Power High Frequency FET Inverter for a Low Frequency Transmitter .......... 6-139
Linear Voltage and Current Regulators Using Power MOSFETs ...................... 6-145
Frequency Response Analysis of the MOSFET Source-Follower ...................... 6-151
Stepper Motor Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 6-156
Design Tips. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 6-165
Solving the Stepper Motor Interface Problems ..................................... 6-165
The D469, an Optimized CMOS Quad Driver for MOSPOWER FET Switches .......... 6-169
14 V, 4 A Battery Charger/Power Supply ........................................ 6-176
400 V, 60 Watt Push-Pull Power Supply ......................................... 6-177
Self Oscillating Flyback Converter . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . .. 6-178
Positive Input/Negative Output Charge Pump ..................................... 6-179
20 Watt, Class A Audio Amplifier .............................................. 6-180
Astable Flip-Flop with Starter .................................................. 6-181
High Current Analog Switches ................................................. 6-182
x
6.13.10
6.13.11
6.13.12
6.13.13
6.13.14
Laser Diode Pulsers ..........................................................
A One-MOSPOWER FET Analog Switch ........................................
Stepping Motor Driver ........................................................
Constant Speed Motor Controller ...............................................
Voltage-to-Frequency Converter with Digital Line Driver ...........................
Page
6-184
6-185
6-186
6-187
6-188
Chapter 7 MOSFET Testing and Reliability
7.1
7.2
7.3
7.4
Understanding MOSPOWER Transistor Characteristics Minimizes Incoming
Testing Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 7-1
Accelerated Testing ............................................................ 7-8
Reliability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 7 -11
Reliability Report Example ..................................................... 7-13
Chapter 8 Appendices
A.l
A.2
A.3
A.4
A.5
A.6
A.7
Calculating the RMS Value of MOSFET Drain Current Using the HP-41CV .............. 8-1
HP-41CV Transient Thermal Impedance Program. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 8-3
Table of Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 8-5
Glossary of Terms ............................................................. 8-7
Additional References .......................................................... 8-9
MOSPOWER Selector Guide ................................................... 8-16
Worldwide Sales Offices. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 8-24
Index ............................. ........................................... 9-1
xi
Preface
Since the introduction of the first practical power
MOSFETs, in October 1975 by Siliconix, these devices have undergone major performance improvements and are now widely accepted and used in
power electronics equipment.
criticisms, and contributions. If you do this and we
are diligent in correcting and expanding the text,
future editions of this handbook should provide a rich
source of information for the user of power
MOSFETs.
Along with improvements in the devices, an understanding of how to use these devices in practical
circuits has gradually evolved. Like any new semiconductor device, the process of understanding has
been slow and the progress uneven. Even though our
knowledge of these devices is far from complete,
much has been learned which would be helpful to
circuit designers. Unfortunately, this information
has, until now, been scattered through a wide variety
of publications or in some cases was unpublished.
In this first edition, approximately 25 percent ofthe
material is new, and a number of important issues are
treated here in detail for the first time. Another 35
percent of the material is a rewrite and expansion on
earlier work with an emphasis on putting the information in context by interrelating the various subjects.
The remaining material is a collection of application
examples taken from a variety of industries.
The book is divided into four sections. Chapters 1
through 3 discuss the basic operation ofMOSFETs.
Chapters 4 and 5 deal with the practical problems of
using MOSFETs from a device point of view. Chapter6 is a collection of applications examples selected
to demonstrate how to use MOSFETs from a circuit
point of view. Chapter 7 provides an introduction to
the testing and reliability of MOSFETs.
The purpose of this handbook is to solve this problem
by providing a source of detailed, up-to-date information on device characteristics and circuit applications.
This is an ambitious goal, and total success in the first
edition is not possible. What is available here is
essentially a snapshot in time. This book is intended
to be a beginning, not an ending, and subject to revision and updating as our understanding grows. If
the book is to continue to grow and become the central reference work for MOSFET applications, it is
vital that you, the reader, send us your comments,
The intent throughout is to provide a balance of basic
theory and practical design information since it is our
belief that this leads to the best designs. The balance
should make this book useful to technicians while still
providing plenty of meat for the advanced worker.
Rudy Severns
July 1984
xiii
Introduction to MOSPOWER FETs _ _
Silicon
Chapter 1
Introduction to MOSPOWER FETs
1.1 General Remarks
Introduction
There are two basic field-effect transistors (FETs):
the junction FET (JFET) and the metal-oxide semiconductor FET (MOSFET). Both have played important roles in modern electronics. The JFET has found
wide applic.ation in such cases as high-impedance
transducers (scope probes, smoke detectors, etc.),
and the MOSFET in an ever-expanding role in integrated circuits where CMOS (Complementary MOS)
is perhaps the most well-known.
Before embarking on the differences between JFETs
and MOSFETs, note that operationally there are
three classes of FETs easily identified in Figure 1:
depletion-mode JFETs, depletion-mode MOSFETs,
and enhancement-mode MOSFETs.
"-CHANNEL
CLASS
p-CHANNEL
v••
o
+
1 T
1 T
IT
11
-A-
All of these applications are best described as smallsignal where signal levels are measured in a few volts
and where the relative power levels range from picoto micro-watts. Such FETs are microscopically small
with the potential of placing thousands ofthem upon
a single semiconductor chip.
DEPLETION-MODE
J-FET
o
v••
v••
o
This handbook is about MOSFETs-power MOSFETs-and they are not microscopic. Before the
reader plunges into the text, it is wise to review the
differences between a small-signal FET and a power
MOSFET. Also, the reader should identify our definition of power since this handbook is for power
applications.
+
-B-
DEPLETION·MODE
MOSFET
o
v••
FETs and MOSFETs-How They Differ
v••
o
Junction FETs (JFETs) and small-signal MOSFETs
differ in their cross-section as well as in their operation. Both are majority-carrier transistors in that they
do not rely upon the mechanism familiar to the users
of bipolar transistors: injection, diffusion, and collection. While the bipolar transistor needs an injection
of base current to initiate transistor action, the FET is
controlled by a gate-to-source potential. Current,
hence power, is not a requirement for FET action.
ENHANCEMENT-MODE
MOSFET
-c-
o
v••
+
The "ABC's" of FET Bias Control
Figure 1
1·1
- 'D
+
l1li
Class A is defined as the depletion-mode class; all
JFETs perform in this class. Class B is also defined as
being in the depletion-mode class, but, as can be seen
from the figure, it can support enhancement current.
Class C is strictly enhancement. MOSFETs perform
in either Class B or Class C. No JFET is capable of
performance in either of these latter classes because
it would require forward biasing of the gate junction-a condition strongly discouraged.
Figure 3, the output characteristics for a typical nchannel JFET are shown with IDSS and Vp clearly
identified.
.
U 290
OUTPUT CHARACTERISTICS
lOSS
/'
400
The JFET and the MOSFET have fundamental fabrication differences which are easily visualized from
the cross-sectional view in Figure 2. Here the JFET
has a diffused gate electrode, but the MOSFET has a
gate electrode electrically isolated from the channel.
In this illustration, the MOSFET is a Class C MOSFET-more commonly known as an enhancementmode MOSFET.
GATE
-
( /
!~
I
I 200
9
100
DRAIN
/1
I
-PlT"""FF
I/
{t/
LOCUS
I
Yp
•
,.
12
20
Vos - DRAIN-SOURCE VOLTAGE (V)
Typical Output Characteristics of a J FET
Showing the Saturation Drain Current, lOSS, and the
Pinch-off Voltage, Vp
Figure 3
CLASS 'A' DEPLETlON·MODE
J·FET
If the JFET gate is forward biased (positively for an
(a)
SQURCE
GATE
n-channel JFET and negatively for a p-channel JFET) ,
the pn-junction formed by the gate diffusion would
be in current conduction. Two things occur: first,
substantial gate current would flow resulting in a
lowering of the gate input resistance; and, second, a
slight increase in drain current would be observed.
The latter results from both the contribution of gate
current as well as from some further reduction of the
depletion field which increases the available channel
current.
DRAIN
CONDUCTIVE GATE
GATE INSULATOR
~:::b~~~~gJL-r-...,
CLASS 'C' ENHANCEMENT·MODE
MQSFET
(b)
SOURCE
cross-Sectional Comparison between a J FET (a) and an
Enhancement-Mode MOSFET (b)
Figure 2
GATE
DRAIN
CONDUCTIVE GATE
GATE INSULATOR
Operation is as follows. Gate control for the JFET
depends upon the manipUlation of a depletion field
about the gate diffusion. The gate is biased with an
abundance of negative charge (using an n-channel
JFET as an example) by tying the gate terminal to the
source. As the drain-to-source voltage rises, a depletion region forms and almost all the electrons in the
channel are swept away. The most depleted area is, of
course, between the drain and gate; the least depleted
area is between gate and source. As the drain potential increases, current will increase but only to a limit
where beyond, with increasing voltage, no additional
current passes. This current limit is called IDSS
(saturation drain current at zero bias), and the drain
voltage limit is called Vp (pinch-off drain voltage). In
CHANNEL
Class 'B' Depletion-Mode
MOSFET
Figure 4
Gate control for the MOSFET differs from that of the
JFET. A Class B depletion-mode MOSFET, shown in
Figure 4, may be compared with a Class C enhancement-mode MOSFET shown earlier in Figure 2. Note
in the Class B MOSFET there is a channel connecting
1·2
the source and drain. Although gate control is essentially the same for both, operationally there is a
difference. For an n-channel depletion-mode (Class
B) MOSFET, a negative potential upon the gate will
deplete the electrons within the channel (like charges
repel, unlike charges attract), and drain current will
be proportional to the gate potential. In fact, the
output characteristics of a Class B MOSFET are
remarkably similar to those of a Class A JFET as
shown in Figure 3. If, however, a positive potential is
applied to the gate, free electrons within the p-body
will accumulate under the oxide and existing n-channel inverting the p-region. The enlarged n-channel
allows additional drain current to flow over and above
that current flow at zero bias. Thus, the depletionmode MOSFET takes on the characteristics of an
enhancement-mode MOSFET.
MOSFETs and MOSPOWER-The Successful
Solution
Both the Class B (depletion-mode) MOSFET and the
Class C (enhancement-mode) MOSFETs shown in
Figures 2 and 4, respectively, were originally developed for small signal applications requiring, at most,
a few milliamperes of drain current with attendant
power requirements of a few hundred milliwatts.
There were several interrelated problems inhibiting
high-power performance that were eventually solved
by the introduction of the vertical double-diffused
MOSFET structure. As the reader continues through
this handbook, a clearer understanding of MOSPOWER theory, design, construction, and applications will evolve.
POWER: Its Definition Used in This Handbook
Conversely, the Class C enhancement-mode MOSFET has no current flow until inversion occurs.
Consequently, for an n-channel enhancement-mode
MOSFET, a positive gate potential is required. Closer
examination of Figure IC shows an offset before
conduction which is called the threshold voltage
(VT). Before drain current flows (before inversion),
an electric field must be established closely dependent
upon such variables as the thickness of the gate
insulation (silicon dioxide), the doping of the body
(which, in turn, controls the availability of free
electrons), and the gate material itself-whether
metal or polysilicon. Threshold voltages generally lie
between 1 and 6 volts.
World-wide, the semiconductor industry has established a 1 watt power dissipation to set apart smallsignal transistors from power transistors. At the time
of this writing (1984), it is premature to establish an
upper limit although we can, without embarassment,
suggest that the upper limit may not be totally the
responsibility of the semiconductor itself but more
probably that of the package. Today packages, such
as the ubiquitous T03 (TO-204), limit performance
to 250 watts. New packages just now emerging show
promise to 500 watts.
1-3
1.2 A Little History
The power MOSFET, as a practical commercial device, has been available since 1976. However, the
history and attempts to produce power FETs go back
much further. It may come as a surprise, but Shockley,
Bardeen, and Brattain were actually trying to fabricate a field-effect transistor when they stumbledyes, stumbled-upon the bipolar transistor. Actually
the field-effect transistor preceded the invention of
the bipolar transistor by nearly 20 years! Although
some try to contest who actually made the discovery,
the U. S. Patent Office credits the invention to Dr.
Julius Lilienfeld-the date of his patent? 1930! Unfortunately, the good doctor did not build the field-effect
transistor (or FET), for in his day semiconductor
material was not sufficiently advanced to be of much
use. After Shockley and his crew developed their
Nobel-prize winning bipolar transistor, Shockley went
on to develop the FET. He announced the FET to the
world in his classic paper [1] where he offered a
general theory of operation and his predictions of
performance.
FETs were self-defeating and offered excessively high
parasitic capacitances for the little gain in performance.
Undoubtedly, the breakthrough came in 1959 when a
paper authored by Wegener [2] identified the fundamental failing of the classic approach and offered a
remarkable solution. Wegener's solution was simplicity itself. He proposed that rather than build planar
FETs, we could achieve higher current densities if we
built cylindrical junction FETs. Within a few years of
this remarkable discovery, a plethora of published
papers appeared, scattered through several different
technical journals which announced the successful
fabrication of moderate-power, field-effect transistors. Notable among these early researchers were
Teszner, Zuleeg and Nishizawa [3]. The latter scientist
is well-known for his Static Induction Transistor (SIT)
which made its commercial entry during the mid1970's in Japan.
Unfortunately, their time had not yet come, and little
was heard of these power FETs. A few would appear,
undergo some preliminary evaluation, possibly attain
some publicity, and then quickly fade into oblivion.
Part of the problem was the difficulty in manufacture
and the problem of needing two power sources-one
for drain voltage and the other for gate voltage (they
were depletion mode devices). Finally, without support engineering, they simply never caught on. Consequently, although occasionally discussed, power
FETs remained dormant until 1976.
But Shockley's work was still 20 years from the
introduction of a successful power FET although he
was very close to the eventual understanding of what
it would take to build a power FET. Many followed
Dr. Shockley, making numerous experimental devices and hoping for power, but they all failed-and
they all failed for the same reason. The path these
unsuccessful experimenters followed was generally to
parallel many cells. What they disregarded, forgot or
tossed off as inconsequential was that paralleling for
additional current handling may drop the channel
resistance, but the parasitic elements increase at a
higher ratio. As a consequence, these early power
In 1976, Siliconix announced the world's first commercially-available power MOSFET in volume production with the registered name MOSPOWER.
14
Nearly simultaneously with this announcement by
Siliconix, Hitachi of Japan announced the availability
of complementary pairs of power MOSFETs ... and
the race was on. Unlike previous attempts at achieving power, these devices were first, MOSFETsMetal Oxide Semiconductor Field-Effect Transistors-and second, they utilized a novel technology
called double-diffusion, which will be further explained in a later chapter.
GATE
CHANNEL LENGTH
The first Siliconix power MOSFET was of novel
vertical construction, utilizing a four-level semiconductor process and an easily identifiable anisotropicetched V-groove shown in Figure 1. Hitachi, on the
other hand, followed the classic planar design with
source, gate, and drain accessible on the top surface
ofthechip.
SOURCE
GATE
SOURCE
GATE
DRAIN
Planar Vertical Double-Diffused Power MOSFET
Figure 2
It is important for the reader to note that despite the
various trade names (TMOS, HEXFET, SIPMOS),
the basic operation and construction of the power
MOSFET are, for all practical intent, identical.
SOURCE
References
1,
f
I--------==--------j
[1] Shockley, W., "The Unipolar 'Field Effect' Transis-
tor," Proc. IRE 40 November 1952) pp. 1365-76.
[2] Wegener, H.A.R., "The Cylindrical Field-Effect
Transistor," IEEE Trans. Electron Devices, Volume 6 1959, pp. 442-449.
CHANNEL '" _ , _
SUBSTRATE
n+
LENGTH
SIN 53°
DRAIN
[3] Teszner, S. and R. Giquel, "Gridistor-A New
Field-Effect Device," Proc. IEEE 52 (1964)
V-GROOVE DMOS
Figure 1
pp. 1502-13.
As the power MOSFET technology improved and
competitors appeared, the technology evolved into
what is now the standard-a vertical, planar fourlayer semiconductor process commonly called DMOS
or double-diffused MOS, shown in Figure 2.
[4] Zuleeg, R., "Multi-Channel Field-Effect Transistor
Theory and Experiment," Solid State Electronics
10 (1967) pp. 559-76.
1-5
1.3 Fundamental Limitations of
Power MOS Transistor Performance
The goal of power MOS transistor suppliers is to
produce devices with the best performance and the
lowest manufacturing cost When die manufacture
alone is considered (i. e., the cost of assembly and test
are ignored) this goal becomes the production of
devices with the lowest on-resistance per unit area for
a given breakdown voltage. While one might conclude that this statement is obvious, the method for
obtaining this performance is not. There are many
practical considerations to be made before the
"optimum" design for even a single voltage is obtained Various engineering approaches have led to a
large number of device designs in the growing power
MOS transistor field.
Before discussing the practical issues faced when
optimizing device performance, it is appropriate to
identify the theoretical maximum performance per
unit area as a standard against which various manufacturer's devices can be measured. This limit,
shown in Figure 1, is obtained when the product of
the resistance of a block of silicon and its surface area
are plotted as a fraction of its breakdown voltage.
This "normalization procedure" assumes that 100
percent of the surface of the silicon is injecting carriers and that 100 percent of the theoretical breakdown voltage ofthe silicon is obtained. Both ofthese
assumptions can never be achieved in real devices,
but it is possible to obtain accurate estimates of the
limits imposed by practical considerations.
The utilization of the transistor chip surface may be
examined in two steps. First, the percentage of the
surface area available for the active device is determined. The relationship between the chip area and
,00
i
~
10-2
I
"m 10-3
c
10-4
'0-5 L--L...J....J...J..J..I..u.L_L.......L....I....L.LJ..U":--'--'-'....I....Lu.J.I
101
103
104
BREAKDOWN VOLTAGE (BVossI, (VOLTS)
The Theoretical Limit of
MOS Power Transistor Performance [1]
Figure 1
the potential active chip area may be derived from
knowledge of the width of the region at the perimeter
of the chip required for device termination, the number and size of the bonding pads, and the spacings and
tolerances required for the active area.
improved by another 5 to 7 percent over the utilization factor shown in Figure 2.
A graph of this relationship consists of the three
regions outlined below for a device with source and
gate bonding pads on the surface.
Region I:
The chip size is dominated by the need
for the two bonding pads placed far
enough apart to allow for bonding.
Region II:
Minimum size bonding pads and minimum conductor widths are sufficient
for the current densities involved.
The surface geometry and layout dimensions chosen
by a device manufacturer also impacts the device
efficiency. The surface geometry requires optimization to obtain the greatest amount of injecting source
perimeter per unit area. Figure 3 compares the layout
efficiency of various surface geometries. While the
geometries differ in efficiency by only a few percent,
the "square-on-hex" pattern is most efficient. The
width of the gate region must also be optimized.
Region III: The current density increases to a level
that requires either greater than minimum conductor widths and bonding
pad sizes.
Figure 4 shows the cross section of a contemporary
MOSPOWER transistor. If the gate width is too
large, surface utilization is poor; while if it is too
small, the JFET formed by the body region results in
an increase in device resistance. Optimum gate width
is a function of the resistivity of the drain region and
decreases as the voltage rating of the device
decreases.
Figure 2 shows the percentage utilization versus chip
area for MOSPOWER transistors with two bonding
pads and a nominal width for the region at the edge of
the chip. For a chip 20 mils on a side (500 /Lm on a
side), the surface utilization is below 40 percent, but
rises rapidly to approximately 80 percent for a chip
40 mils on a side. If a manufacturer chooses to eliminate the area of the chip dedicated to the source pad
and to "bond over active area," the chip utilization is
The voltage rating of the MOSPOWER transistor
also sets limits on device performance. At high voltages, devices are designed to obtain a breakdown
SOURCE AND
BODY
GATE
I ~REGIONI
100 II
LREGION II
I
aD
60
~
_ _ e __
•
~REGIONI""'------~====~
.....
l
I
40
~CELL
SIZE------'I
N-
DRAIN
20
oL-_ _
o
~
_ _ _-L_ _ _-L_ _ _
1 x 107
2 x 107
3 x 107
N+
~
»
4 x 107
CHIP AREA (/-'m2)
DRAIN
Surface Utilization vs. Chip Area
for MOS Power Transistors
Figure 2
LINEAR
GEOMETRY
SOURCE
GEOMETRY
AND GRID
UNIT CELL
Illllill
Cross Section of a Power MOS Transistor
Figure 4
SQUARE ON
CIACLE ON
HEXAGON ON
saUARE
GRID
SQUARE
SQUARE
GRID
GRID
SQUARE ON
HEXAGONAL.
GRID
CIRCLE ON
HEXAGONAL
GRID
DOD 000 000 q-p
Q-P
Orol0 oic)10
oi61 0 [(0'0 0[0 10
L_J
0-6
DOD 000 000 0-6
~-~
l.r--·'!
r--,!
S
8
NOT
APPLICABLE
10
: ! IDI :0:
I I , f--Jl
L--.Jl
--151-
r--·,j
:O~s
! ~--i
51
<{
Ds)
L--.Jl aT\ '--- f'
HEXAGONAL
GRID
~-9
o!o~o
/--i,.
0-0
r--i
\--f
\~--f
09523
10
L_~
..... S+a--+i
I
I
I
I
I
HEXAGON ON
( OS) (OS)
COEFFICIENT
G FOR
CELLULAR
GEOMETRIES
08862
09306
10146
The Efficiency of Various MOS Power
Transistor Surface Geometries [2]
Figure 3
\-7
,
,,
,00
voltage close to the theoretical minimum value set by
the silicon. As the voltage is increased, it becomes
difficult to obtain greater than 80 percent of the theoretical maximum. As the voltage rating decreases,
another limit is encountered. The decrease in gate
width and other device dimensions cannot continue
indefinitely as the voltage rating decreases. The requirements for low resistance ohmic contacts to the
source and body regions set a minimum source
and body contact dimension, while the lateral
diffusion of the body region beneath the gate
sets another minimum dimension. These minimum dimensions establish a limit for low voltage
devices.
I
,.
10-1
4
3•• ,
5.
I
••
e.
2.,
I
I
,
I
I
7.
••
1
2
3
4
5
6
7
8
9
10
11
12
When all of the limits discussed in this section
are imposed on a power MOS transistor, the
"practical limit" of Figure 5 results. This figure
accounts for all of the overhead requirements for
edge terminations, bonding pads, device dimensions, surface efficiencies, and voltage effects
discussed. (This figure assumes an 85 percent
surface utilization value.) The performance of
various contemporary devices are also included
on this figure for information.
IRF308
GEVMOS
GEVMOS
1ft HEXMOS
SlUCONIX VN64GA
SlLICONIX VN1200A
SIUCONIX SYH
IR HEXMOS
AEG
SlPMOS
GEDMOS
SILICONIX VNDK
'0-5 "--'--'--''-'--'..LU.''--'--'-L...U-'-LIJ':--,--,-",-,-,.u.U
101
103
BREAKDOWN VOLTAGE (BVass). (YOLTS)
References
[1]
Adler, M.S. and S.R. Westbrook, "Power
Semiconductor Switching Devices - A
Comparison Based on Inductive Switching," IEEE Transactions on Electron Devices, ED-29, No.6, June 1982, pp. 947952.
[2]
Chi, M. and C. Hu, "Some Issues on Power
MOSFETs," PESC Record, June 1982, p.
392.
.
Theoretical MOS Power Transistor
Performance Showing the Practical Limit [1]
Figure 5
1-8
104
o
Power MOSFET Structures _ _
Chapter 2
Power MOSFET Structures
2.1 Principles of MOSFET Operation
One of the attractive features of MOSFETs is the
ease with which the basic principles of operation can
be understood. In many ways, the MOSFET is a solid
state equivalent to a vacuum tube and, at least for first
order behavior, is as easy to understand.
(a)
GATE
s, ..
A simplified model will be used to explain how a
MOSFET works. The model chosen does not represent how practical devices are actually built, but it
does operate in the same manner and makes the
explanation easier. This discussion will address an n
channel device. A p channel device would be just the
opposite, with n and p regions interchanged. The
basic operation is, however, still the same.
p
BODY
(e)
Figure la shows an npn junction structure. In many
ways, this structure is identical to a bipolar junction
transistor; the differences arise from the connections
made to this basic structure. As shown in Figure lb,
for a MOSFET (Metal-Oxide-Semiconductor Field
Effect Transistor), three ohmic contacts and an insulated capacitor plate are added to the npn structure.
As long as the potential between body and gate is not
positive, this device is essentially two diodes back-toback (Figure lc), and only a small junction leakage
current will flow if a + or - potential is applied
between the n region ohmic contacts. In this state,
the device is OFF.
s/D~./D
•
n-Channel MOSFET Structure
Figure 1
to the gate structure, as shown in Figure 2a. Due to
the presence of the gate oxide insulator, the gate
metal and the body semiconductor form a capacitor
which accumulates charge. Even if the potential
between the gate and body is small (O-3V), the
electron density in the body side of the capacitor will
be less than the hole density and the device acts like
two diodes back-to-back with a moderate increase in
leakage current.
The p region is doped so that there are more holes
than electrons. This is, by definition, what makes it a
p region. Even though the holes outnumber the
electrons, there are still plenty of free electrons in the
p region, and if the gate is made positive with respect
to the body, some of these electrons will be attracted
As the gate to body potential is increased, however,
the charge density in the body, immediately adjacent
to the gate oxide, will increase to the point where the
electron density exceeds the hole density, and a
2-1
l1li
portion of the body region (the channel) inverts to
become n rather than p. This is shown in Figure 2b.
The semiconductor structure is now n-n-n and has
become simply a silicon resistor through which current
can flow easily in either direction (Figure 2c). This
is a variable resistor in which the resistance of the
channel is controlled by the potential from gate to
body.
RDS(ON)
___________
I
~
U
'''___......_--.J
~
vas
(a)
(b)
SID
SID
SID
At low values of drain-source voltage (VDS), the
MOSFET, when turned ON, acts much like a normal
resistor. However, as VDS is increased, RnS(on)
becomes a function of VDS as illustrated in Figure 4.
For high values of VDS (> 10-20 V), the MOSFET is
no longer a resistor but acts as a very good current
source if the gate-body voltage (VGS) is fixed. The
transition from resistor to current source will depend
on VGS, as indicated. The actual behavior in the
resistor region will vary depending on the design
compromises adapted for a particular device.
SID
..
(c)
MOSFET TUrn-On Characteristic
Figure 3
SID
lOS
..
<>o-----~-----oo
NORMAL
RESISTOR
SID
YGS3 > YGS2 > VOS1
RDS(ON)
Channel MOSFET In the "ON" State
Figure 2
~-------------~u
There is a limit to the minimum resistance of the
channel. As the gate-body potential is increased,
more charge collects in the channel region. This
charge acts like an electrostatic shield to reduce the
field in the rest of the body. This action very closely
resembles the effect of space charge in a vacuum tube
which acts to reduce the electron emission from the
cathode. In a MOSFET, this "space charge" acts to
limit the additional charge in the channel so that the
channel resistance quickly reaches its minimum value.
t
9
~-------------~u
~-------------VGS1
Vos_
This charge scenario can be used to explain the
RDS(on) versus VGS characteristic shown in Figure
3 which is typical of all power MOSFETs. Region I
corresponds to the condition when the accumulative
charge is not sufficient to cause an inversion. Region
II corresponds to the condition where sufficient charge
is present to invert a portion of the p region, forming
the channel, but not enough that the "space charge"
effect is important. Region III corresponds to the
charge limited condition where RDS(on) does not
change appreciably as the gate-body potential is
raised.
Comparison Between a MOSFET and a Normal Resistor
Figure 4
For example, some power MOSFETs have a linear
RnS(on)NDS 'curve, like a normal resistor. Other
designs, where a minimum RnS(on) at low Vns is
desired, will have a curved charactenstic as indicated
by the dashed line in Figure 4.
Both of these effects are caused by a narrowing or
"pinch off" of the channel as Vns is increased. Figure
5 illustrates the channel narrowing or pinch off.
2·2
The body-source connection places a short between
the base and the emitter of the parasitic npn BJT as
shown in Figure 6b. The base-collector junction is,
however, still present so that the equivalent circuit for
the MOSFET is a MOSFET in parallel with a diode
as shown in Figure 6c.
For most applications, this equivalent circuit is adequate, but in some cases, a more accurate model is
needed. Inherent in the structure of the MOSFET
are parasitic resistances and capacitances. These parasitic elements are shown in Figure 7a and the equivalent circuit in Figure 7b.
Channel Plnchoff In a MOSFET
Figure 5
For a variety of reasons, a practical power MOSFET
does not have four terminals. The normal practice is
to connect the body directly to one of the n regions as
shown in Figure 6a. When this is done, the n region to
which the body is connected is defined to be the
source connection. The other n region becomes the
drain connection. To turn the device ON, the gate is
made positive with respect to the source.
G
RG
eg.
Cg_
~
c~
0-n
_T_
51 . .
T
(a)
egb
p
Rb
i~
f--o
n
G
,~·ti±]
(b)
oo-_vv-.......-H
Rb
(b)
Parasitic Elements Present In a MOSFET
Figure 7
(c)
The effect of these parasitic elements on the device
and circuit behavior is discussed in Sections 2.2,3.1,
3.2,5.4, and 5.5.
The Body Diode In a MOSFET
Figure 6
2-3
2.2 FETs and BJTs as
Charge-Controlled Devices
"Field-effect transistors are voltage-controlled
devices, and bipolar junction transistors are currentcontrolled devices."
However, it will be seen that this difference is of
only secondary importance in both switching and
analog amplifier applications.
The implication of these familiar statements is that
FETs and BITs are fundamentally different, and must
therefore be incorporated differently into a circuit
environment. To the contrary, the point made in this
introductory chapter is that both device types are
fundamentally the same, in that they are chargecontrolled. This point of view is particularly useful
with respect to switching (digital) applications, and
even more important for high-power switching applications such as switched-mode power conversion.
The main conclusion is that, as switches, FETs must
be driven just as "hard", both ON and OFF, as BITs
in order to achieve comparable switching speeds.
The Principle of Charge-Controlled Devices
Current is rate of flow of charge, expressed in the
basic relation
current = charge x rate of flow
Therefore, for a current to exist, there must be:
1. charge present capable of moving
2. a controlling quantity to make the charge
move
In the simple device represented in Fig. I, there is
a "channel" terminated at two contacts by which the
device may be connected through a switch to an
external circuit containing an emf potentially capable
of controlling the circuit current. If the channel is an
insulator, there is no mobile charge available, so no
current flows even though there is a controlling quantity available when the switch is closed.
In the broad sense of this chapter, all low-frequency
electronic devices are charge-controlled, including
that old-fashioned device, the TUBE. The term "lowfrequency" here means that "transit-time" devices,
such as the travelling-wave amplifier and the klystron,
are excluded.
There are, of course, many differences between the
various types of charge-controlled devices, differences
which are of greater or lesser importance depending
upon the application, and most of the balance of this
book is concerned with these differences. Only one
of these will be discussed in this chapter, namely,
the requirement for a steady state input current in
order to maintain an ACTIVE or ON condition in
the BJT; the absence of this requirement in the FET
is the principal reason why the FET is commonly
thought to be fundamentally different from the BIT.
If the channel is a conductor or an n-type semicon-
ductor, as represented in the corresponding structure
of Fig. 2, there are two types of charge originally
present: positive charge on the crystal lattice ions,
and negative electron charge. The channel is electrically neutral, and the total negative charge equals the
total positive charge.
When the external circuit switch is closed, the emf
sets up an electric field in the channel which exerts
a force on both charge types. However, only the
24
/
controlling quantity is applied. The essence of a threeterminal active device such as the FET or the BJT
is the separation of control and collection functions
so that control is exercised at an input terminal and
the resulting controlled current is collected at an
output terminal.
ZER OCURRENT'i7
CLOSED
SWITCH
{
The FET
NO CHARGE,
THEREFORE
NO CURRENT
(
Consider again the device of Fig. I containing the
insulating channel. How can mobile charge be introduced into the channel? That is, how can the device
be converted into an electronically operated switch
in which current collected from the channel can be
controlled by a third terminal?
__
L....
(
\
Charge can be induced in a previously insulating
channel by induction across a dielectric. Suppose such
a dielectric is introduced in a layer between a third
(control) terminal and the side of the channel, as in
Fig. 3(a). The control terminal, the dielectric, and
the channel connected to the lower terminal constitute
a capacitance: if a positive charge Q+ is inserted at
the control terminal, an equal negative charge Q- is
induced in the channel on the other side of the dielectric. This negative charge is capable of carrying
current, so that if a positive potential is applied to
the upper channel terminal by closing the output circuit switch, a current flows as in Fig. 3(b). This is
the principle of the FET, in which the control terminal
is designated the gate, and the lower and upper channel terminals are the source and drain, respectively.
A channel containing no lJIobile carriers (an Insulator) can
carry no current
Figure 1
CURRENT FLOWS
-----.
CONTROL AND COLLECTION
AT SAME TERMINr-AL_ _
",,=-_--.
FIXED posnVE
CHARGE,
MOBILE NEGATIVE
CHARGE,
THEREFORE
CURRENT
CAN FLOW
I
ELECTRON
Ft.OW
Quantitatively, the device turn-on process from the
OFF into the ACTIVE mode is as follows. A charge
Q+ is placed on the gate by an input current
lin = dQ+ Idt; the gate-channel capacitance is thereby
charged up and an equal and opposite charge
Q- = - Q+ of electrons is induced in the channel.
If a positive potential is placed on the drain, an output
current results that is given by lout = - Q- ITt, where
Q- is the charge in transit, and Tt is the mean transit
time of the charge Q- that is in transit between the
source and drain.
A channel containing mobile carriers (a conductor) is able to
carry current
Figure 2
The most significant features of this process are:
(1) To establish mobile charge in the channel, an
input transient current (capacitive charging current) is required.
(2) After a steady state is established, the controlling
charge Q+ remains static (there is no more input
current), but the controlled charge Q- is'in continuous motion clockwise around the output circuit, constituting a steady state (DC) output
current. Thus, even though the total charge in
transit in the channel remains constant at Q-,
the individual negative charges are constantly
being replaced, entering the channel at the
negative electron charge is mobile and capable of
carrying current; the positive charge is constrained to
the fixed lattice ions. In the diagram of Fig. 2, a
(clockwise) flow of electrons results, constituting a
counterclockwise steady state (DC) current in which
the electrons in the channel may be visualized as
moving continuously past the fixed positive charges
while retaining the same density distribution; that is,
the channel is still essentially electrically neutral.
In the two-terminal device of Fig. 2, the controlled
current is collected at the same terminal at which the
2·5
-
(a)
,I
.(
INSERTED POSITIVE " "
CHARGEQ+
'
.... 8
8
+
~
CONTROL TERMINAL ~
1-
'*
+
8
8
8
8
l-
(
8~
+ 88
+
+ 8
dO+
Iln::dt"
the mobile charge was made available; in practice,
the output potential is usually present before the controlled charge is introduced, and the device itself
becomes a (controlled) switch. In this case, the output
current rises as the controlling charge is placed on
the control terminal: thus, the rise-time Tr of the
output current is determined by how fast the controlling charge Q+ is applied, that is, by the magnitude of the input current lin = dQ+ /dt. Therefore,
to tum the device on fast, a large input charging
current is needed.
+ e
+ ee
+
INDUCED NEGAnYe
CHARGE (ELECTRONS)
Q- = -Q+, CAPABLE
OF CARRYING
CURRENT
--
)
An equivalent circuit model of the PET and associated
input and output current waveforms are shown in
Fig. 4. The essential features of this charge-control
representation are the dependent current generator
controlled by the input charge, and the tum-on waveforms showing that the rise time to a given output
current is shorter if the required controlling charge
is applied faster, requiring a larger transient input
current (for a shorter time).
DIELECTRIC
(b)
COLLECTED CURRENT FLOWS
lout::: -
~
DRAIN
DRAIN COLLECTION TERMINAL
..,_,"----------,!!!!!!
CONTROLLED CURRENT
Q-
~~
GArEO
Q+
"'"'1'1
dt
ELECTRON FLOW,
GATE
lout = -T[
MEAN TRANSIT
TIMErt
0+
C>---------------+-------~ SOURCE
SOURCE
AREA Q+
CONTROL
Principle of a charge-controlled device (FET): (a) An insulating
channel can have mobile carriers Induced in It by a control
charge Inserted at a control (Input) terminal (gate); (b) the
Induced channel charge Is capable of carrying (output) current
when a potential is applied to the collection terminal (drain).
Figure 3
CURRENT
'," ____.£-__.1-.1-______________- , . . . . . - _ . . - - , -___
AREA Q+
source and leaving at the drain after a mean
transit time Tt.
(3) The magnitude of the output current
lout = - Q- ITt is determined by the controlling
charge Q+ = - Q- .
These three statements describe the principle of a
charge-controlled device, in which a current in an
output circuit is controlled by the charge placed on
a control terminal.
Basic equivalent circuit model of a charge-controlled device
(FET). The same control charge inserted or extracted over a
shorter time lowers the rise or fall time, respectively.
Figure 4
A steady state, or DC, output current continues to
flow as long as the controlling charge remains (statically) at the control terminal. Equally important
as tum-on is the reverse, tum-off, requirement: to
For clarity in the above explanation of the control
and collection functions, the potential at the output
terminal was asssumed to be applied by a switch after
2-6
I
stop the output current, the controlling charge must
be extracted, requiring a reverse gate current.
To turn the device off fast, which means a short fall
time 7f, the controlling charge must be withdrawn
fast, requiring a large gate extraction current
lin = - dQ+ Idt. The turn-off waveforms are also
shown in Fig. 4.
(a)
INSERTED POSITIVE
CHARGE Q+
(HOLES)
The essential charge-control quantitative relations may
be summarized as:
CONTROL TERMINAL
dO+
Iln""di
lin = dQ+
dt
lout =
(b)
The charge-control model represented by Fig. 4 emphasizes that, to realize the potential fast speed of
the PET, the gate drive source must be low impedance
and capable of both supplying and sinking the required
transient insertion and extraction control charge
currents.
COLLECTED CURRENT FLOWS
lout =
COLLECTOR
The input capacitance shown in the model of Fig. 4
merely accounts for the fact that a voltage appears
at the input: it is a nonlinear capacitance, and so the
input voltage is a nonlinear function of the control
charge. However, the charge-control description of
the device operation makes it clear that it is the input
charge that does the work, and the input voltage is
of strictly secondary importance.
BASE
-~
COLLECTION TERMINAL
r
ELECTRON FLOW,
MEAN TRANSIT
TIME 7t
IIIF.JI
The BJT
The charge-control description of the BJT almost exactly parallels that of the FET, with appropriate
change of terminology.
Principle of a charge-controlled device (BJT): (a) Insertion of
control charge (holes) at the base terminal induces injection
of electrons into the base region (channel) from the emitter;
(b) the induced electron base charge is capable of carrying
(output) current when a potential is applied to the collection
terminal (collector).
Figure 5
In an npn junction transistor, the channel is the base
p region, and the lower and upper contacts are the
emitter and collector n regions, respectively. A conducting contact on the side of the base region becomes
the third, control terminal, as shown in Fig. 5(a).
The most significant features of the BJT are identical
with those of the PET:
(1) To establish mobile charge in the base, an input
transient current lin = dQ+ Idt is required.
(2) After a steady state is established, the controlling
charge Q+ remains static, but the controlled
charge Q- is in continuous motion around the
output circuit constituting a steady state DC output
current. Thus, even though the total charge in
transit in the base remains constant at Q- , the
individual negative charges are constantly being
replaced.
An input current lin = dQ+ Idt introduces positive
charge (holes) into the base region, which induces
an equal negative charge Q- of electrons to be injected into the base region from the emitter. This
negative charge is capable of carrying current so that,
if a positive potential is applied to the collector terminal, an output current lout = - Q-Irt flows where,
again, Q- = - Q+ is the total charge in transit
through the base region, and 7t is the mean transit
time between emitter and collector.
2-7
(3) The magnitude of the output current
lout = -Q-1Tt is determined by the controlling
charge Q+ = - Q- .
COLLECTOR
Also, if the collector potential is previously present,
the collector current rises as fast as the controlling
charge is introduced into the base input terminal, so
again a large input current lin = dQ+ /dt is necessary
to achieve fast turn-on. Similarly, fast turn-off
requires a fast extraction of the control charge by a
large negative input current lin = -dQ+ /dt.
I,n
BASE
0-_.--.....---..,
CONTROLLED CURRENT
Q-
Q+
lout=--=-
" "
So far, the properties of the BIT are seen to be
identical to those of the FET, and so the same chargecontrol model and associated input and output current
waveforms as in Fig. 4 are applicable. There is,
however, one difference in the physical device structure that leads to an additional effect.
EMITTER
CONTROL
CURRENT
lin
In the FET, the controlling charge Q+ and the (equal)
controlled charge Q- remain separated on opposite
sides of the gate dielectric. In the BJT, on the other
hand, the controlling and controlled charges occupy
the same physical volume, namely, the base region;
nevertheless, the one-to-one ratio still exists, and again
the resulting base-emitter voltage is (nonlinearly) related to the control charge Q+ by the diffusion or
storage capacitance that replaces the gate-channel
capacitance of the FET in Fig. 4.
- a- =~
CONTROLLED
CURRENT
lout
The presence of positive charge (holes) and negative
charge (electrons) in excess of their eqUilibrium concentrations, in the same physical volume, leads to a
gradual loss of both by the process of recombination.
Therefore, after a BIT is turned on, the collected
(output) current gradually decays back to zero as the
controlling and controlled charges are lost to
recombination.
'.
/
"
Basic equivalent circuit model of a charge-controlled device
(BJT). A maintenance DC base current 10ut/,8 is required to
make up for the hol_lectron recombination loss of the stored
charge.
Figure 6
The essential charge control quantitative relations are
correspondingly augmented to become
lin = dQ+ +~+
dt
T
Conversely, if the collected current is to be prevented
from decaying, the recombination charge loss must
be replaced: thus, a steady state (DC) maintenance
control current must be provided to keep feeding in
control charge Q+ at the same rate that it is lost by
recombination. This loss rate is Q+ IT, where T is
the mean lifetime of the hole-electron pairs in the
BIT base region.
Q- = -Q+
Q-
lout = - -
Tt
In the steady state condition in the ACTIVE region,
only the DC maintenance component of the input
current remains, and the ratio of the DC output to
DC input current is
To account for recombination loss, the charge-control
model and associated current waveforms of Fig. 4
may be augmented, as in Fig. 6, to represent an
additional feature of the BIT:
(4) To maintain the original DC output current at
turn-on, a maintenance input DC current must be
provided, represented in the model by the current
through a resistance in parallel with the storage
capacitance.
I
lout
T
= == f3 or hFE
lin DC
Tt
-
2·8
where {3 and hFE are the familiar symbols for the
"current gain" of the BJT.
at which all the induced charge can be collected, the
device is being overdriven and excess or uncollected
charge is present in the channel. The waveforms are
shown in Fig. 7, in which the saturated ON condition
replaces the final ACTIVE condition as a result of
input current being present beyond the time needed
for saturation.
It is clear that from a switching point of view the
FET and the BJT are identical in principle. Both are
charge-controlled devices whose input is capacitive,
and to make the device operate fast the input capacitance must be charged and discharged fast, that is,
the drive must be capable of sourcing and sinking
sufficiently large transient currents.
In contrast, the most obvious difference between the
BJT and the FET, the presence of DC input current
in the BJT, is irrelevant as far as the switching properties are concerned. It may be noted that the same
conclusion applies to the BJT operated as an analog
amplifier: for high frequencies the input admittance
is dominated by the diffusion susceptance, and the
{3 is irrelevant; only for low frequencies, down to
and including the DC bias conditions, is the input
admittance dominated by the {3-determined conductance. From this viewpoint, an "ideal" BJT would
have an infinite {3, as essentially does the FET.
lout
CHARGE·CONTROLLED
DEVICE (FET OR BJT)
INPUT
CURRENT
'"
Charge-Controlled Devices as Digital or
Power Switches
All the above discussion has been related to how a
charge-controlled device is driven from the OFF to
the ACTIVE region, and back, in order to emphasize
the underlying simplicity of the device nature. In
practice, when operated as a switch, either in digital
signal processing or in switched-mode power processing applications, the device is driven from OFF
through the ACTIVE to the ON or saturated condition, when additional effects come into play. Nevertheless, the charge control description continues to
be useful and, qualitatively at least, the additional
effects can be accounted for by superposition.
OUTPUT
CURRENT
lout
OUTPUT
VOLTAGE
Charge-controlled device operated as a switch: overdrive is
needed to cause output saturation, resulting in excess stored
charge that must be withdrawn before the switch begins to
turn off (causing the storage time).
Figure 7
When a device is used as a switch, the initial tumon phase is the same as already described: when the
collection potential is already present on the output
terminal, the channel charge Q- begins contributing
output current as the input current lin = dQ+ Idt
induces the corresponding charge - Q- = Q+. A
load carrying the output current causes the output
terminal collection potential to fall. If this potential
remains high enough to continue to collect all of the
channel charge after the input current ceases, the
device is turned on into the ACTIVE region as already
discussed.
In the saturated ON condition, the channel contains
two components of charge: the charge necessary to
contribute the saturated output current as if the device
were in the ACTIVE condition, plus the excess
charge. From a practical point of view, there must
always be some margin of excess charge to ensure
that saturation is maintained, and also, the larger the
excess charge the smaller is the saturation voltage
drop along the channel, which is a desirable circuit
property in the ON condition.
If the device is a BJT, there must be a suitable steady
state input current to maintain not only the active
but also a suitable excess channel charge.
However, if the output terminal potential falls sufficiently low that it ceases to collect all the induced
channel charge before the input current ceases, then
the saturated ON condition results in which the output
current is limited to a value determined by the output
circuit. Since input current continues beyond the point
When a charge-controlled device is to be turned OFF,
a qualitatively new effect occurs when the initial condition is saturated. As also shown in Fig. 7, the tum-
2-9
l1li
off input current must first extract the excess charge
before the output current begins to fall. This interval
is known as the storage time TS, familiar in BJT
devices, and the greater the excess charge the longer
the storage time for a given tum-off input current.
However, a quantity analogous to storage time also
exists in FETs, although quantitatively it is much
shorter than in BITs because the excess charge is
much smaller. Nevertheless, it is notable that even
this (charge-control) property is qualitatively the same
in both FETs and BJTs.
ON voltage drop (large excess charge) and short storage time (small excess charge). "Proportional base
drive," in which the ON maintenance base current
is made proportional to the output current (by a positive feedback transformer connection) is often used to
alleviate this conflict.
The FET is "easier" to drive than the BJT, not
because one is "voltage-controlled" and the other
"current-controlled", but because in the FET a
maintenance ON gate current is not required, and the
"storage time" is much smaller, so the design tradeoff
between ON drop and storage time is essentially
eliminated without the need for proportional drive.
Basically, the FET and BIT are both charge-controlled, and both must be driven from a low-impedance source capable of both sourcing and sinking
sufficient current to provide the desired tum-on and
tum-off times.
Optimization of base drive circuits for BJTs involves
choice of suitable tradeoffs between the three conditions tum-on, saturation, and tum-off. The prime
consideration is, of course, fast insertion and extraction of the controlling charge to obtain fast rise and
fall times. However, a compromise must be found
between the conflicting requirements of low saturation
2-10
2.2.1 Charge Transfer Characteristics
Introduction
Casual readers might question beginning the study of
hybrid power FET circuits by examining the charge
transfer characteristics. Indeed, some readers unfamiliar with power FETs might erroneously believe
that no energy, ergo no charge, is needed to operate a
power FET. Furthermore, these same readers might
anticipate that the coupling of a high gate impedance
power FET to a bipolar transistor or thyristor would
obviate any need for an intermediate power driver
other than, perhaps, a level shifter. Under some
conditions, these might be proper assumptions, but
first examine exactly how to turn ON a power FET.
Surprisingly, it is similar to that of a bipolar transistor
since both the power FET and the power bipolar
transistor are charge-coupled devices!
to rise (more positively for an n-channel MOSFET
and more negatively for a p-channel MOSFET), two
events begin: As the gate potential rises, channel
inversion begins (see Figure 1), and the input capacitor, Cgs , begins to charge. In reality, the second event
preceaes the first; the order is reversed here for
emphasis since the first event is recognized without
explanation. The current necessary to charge this
input capacitor is computed using the following _
equation.
:::=:;;kz~~~~~~:::O:X:'D~'
Although the power FET and the power bipolar
transistor are both charge-coupled devices, much of
the similarity stops there. Operation of a power
bipolar transistor is principally concerned with current gain whereas the power FET is, in effect, a
voltage-controlled current source. If, indeed, this is
true then how can a power FET be called a chargecoupled transistor?
INVERSION
LAYER
ega'" ega + Can +
,PI
-
- n+
-
SUBSTRATE
DRAIN
n- CHANNEL
Positive Gate Bias Inserts the p-Reglon Immediately Beneath
the Oxide to Effect Majority-Carrier Current Flow
Figure 1
The answer lies in the way a power FET is turned ON.
More properly it is identified as an enhancementmode MOSFET. This section concentrates on the
operation of an n-channel, enhancement-mode power
MOSFET remembering that the p-channel enhancement-mode power MOSFET obeys all the same
laws but in complement. Any enhancement-mode
MOSFET requires a finite gate voltage before conduction occurs. The level at which it conducts is called
the threshold voltage, VT. As the gate voltage begins
iN
Cilt
where, ilv is the gate-to-source voltage, Vgs
ilt is the rise time of the incoming signal
C is the input capacitance, Cgs , prior to
turn-ON.
2-11
(1)
but
400
Q=
cav
I
~
(2)
/
2
~
:;:: ~- rffl
so combining Equations (1) and (2) yields
.-~ - - - I
~
"'
Q = iat
/
(3)
-j
I- V. '
What is unique about equation (3) is that it is easily
measured! However, before looking into the mechanics of how to measure charge, note that the discussion
up to this point only involves the gate charge prior to
turn-ON [see equation (1)]. The entire switching cycle
must be examined. The portion thus far is known as
the subthreshold region.
12
vGG
10
I
VDS
where, Q is the charge necessary to raise the gate to
Vgs.
V
"
,.-c--
r-..
/
v ••
02;----=r- 3
I
II 0,
II
5
.~
,,~ ~O%
10
15
20
25
C(pe)
Gate Charge Characteristics Showing the 3 Regions
Figure 3
(a) The Dynamic Characteristics of Gate Charge
(a.l) Driving the power MOSFET into Conduction
Up to this point, the power MOSFET is OFF. The
input capacitor, Cgs , has a charge, Q, so the voltage
across this capacitor is equal to the threshold voltage,
VT, of the MOSFET. At this point, it might seem
that only a small additional charge is necessary to
drive the power MOSFET into action, that is, into
conduction. When this occurs, the MOSFET is in the
dynamic region-the transition region that demands
attention.
In Figure 3, the complete charge-transfer characteristics of a typical vertical planar double-diffused
(VDMOS) MOSFET, region 1 is the area previously
discussed, the sub-threshold region.
Tho very noticeable events occur in region 2. First, in
time, the drain-to-source voltage begins to drop;
second, what appears to be an incredible increase in
input capacity, Cgs , occurs as the fast-rising gate
potential apparently stalls [see equation (2)]. What is
happening is simply the Miller effect brought about
through the interaction of the feedback, or gate-todrain capacitance, Cgd (more commonly recognized
in FET terminology as Reverse Transfer Capacitance,
Crss), and the common-source forward voltage gain,
AV·
Generally, whenever power MOSFETs are used as
switches, they operate in what is commonly called the
common-source mode as shown in Figure 2. As this
transition region is entered, the charge characteristics
are greatly altered by a phenomenon called the Miller
effect.
Cin = Cgs
+ Cgd(1 - A V)
(4)
+VDD
where
AV = ilVds
ilVgs
Remember that the quantity, ilV ds, is, for an
n-channel MOSFET, a negative value, and for a
p-channel MOSFET, the quantity, V gs, is a negative
value. Thus in either case, this equation is calculated as (1 + voltage gain).
1----0 OUT
Thus, either considerably more time is needed
(assuming that the charging current, i, is a constant), or, to keep time, ilt, a constant considerably
more current is needed. One thing must be kept in
mind: the execution of equation (4) is not easy! If
the charge-transfer characteristics for the particular
Common-Source Configuration. Source Is "Common" Both
to Input Circuit as Well as Output Circuit
Figure 2
2-12
Next the Miller effect must be considered: the apparent flattening of the charging cycle. Since the calculation of input capacitance is, from equation (2), merely
the slope of the charge curve, it appears to be a
tremendous increase in input capacity. Fortunately,
equation (4) can be bypassed; Cin is easily determined by an obvious extension of equation (2).
power MOSFET are available, use them! Poised to
move into the conduction or transient region, first
take time to see how to use these charge-transfer
characteristics.
(a.2) Using the Charge-Thansfer Characteristics
It takes a finite charge to energize a capacitor and that
charge, 0, can be calculated using equation (2). If
ao
Cin= av
equation (2) is merged with equation (3), a constantcurrent source will provide a linear rate of increasing
voltage with time, at, as seen in Figure 4 as well as in
region 1 of Figure 3. If the capacitance changes, for
example, or if it increases because of the Miller effect
then with a fixed charging current, i, the charge time,
t, increases. Additionally, the effect of increasing
capacitance (Miller effect) upon the voltage, av, can
be examined. Where, with a fixed capacitance the
rate of change was linear, now there is an abrupt
slowing. All of these effects may be seen by careful
examination of Figure 3.
02 - 01
(7)
Vgs2 - Vgs l
This value of Cin obtained from equation (7) equals
that value that might have been obtained from equation (4) after a lot of labor. Finally consider the
calculation of input capacitance for region 3, and an
equation similar to equation (7):
03 - 02
Cin= - - - - Vgs3 - Vgs2
(8)
Whether these equations are used to calculate the
various input capacities or simply observed from
Figure 3, it is clear that Cin of the subthreshold region
(region 1) differs from the Cin of region 3-the
former being somewhat lower. The reason for this is
that capacitances within a power MOSFET-as with
any FET-are depletion-field dependent. Being depletion-field dependent, they are voltage dependent
as is evident from Figure 5. Since in the subthreshold
region a power MOSFET is OFF, the drain voltage is
at the rail, but when driven into full conduction, the
voltage across the MOSFET is simply its VSAT, at
worse only a few volts.
+v
CONSTANT
CURRENT
SOURCE
A Linear Charging Rate Results When a Capacitor Is
Charged from a Constant-Current Source
Figure 4
(a.2.1) Calculating Switching Times from Charge
Thansfer
These charge-transfer characteristics can also be used
to determine switching times with remarkable precision. Aside from turn-ON and turn-OFF switching
To determine the amount of energy necessary to move
a power MOSFET from a normally OFF state to a
normally ON state, the difference in charge provides
the answer.
1000
C g,
W = 1/2 (ao) (aV)
(5)
........
Using data extracted directly from Figure 3, we can
calculate, with precision, the input capacitance states
in all three regions: the subthreshold region (region
1), the turn-ON region (region 2), as well as after
V SAT has been passed (region 3).
"'"
From equation (2) calculate the capacitance in the
subthreshold region remembering that at the beginning of our initial tum-ON charge, Vgs = 0, which
simply means that aV = aVgs.
""
c~
Cgd
10
01
10
10
100
Vos - VOLTS (V)
C=~
Vgs
Voltage-Dependent Depletion Areas Also Affect DepletionDependent Capacitances, Cds and Cgd
Figure 5
(6)
2-13
l1li
times, the delay times must be considered. The first
delay is that time required for the input capacity to
charge to the threshold voltage at which tum-ON
begins. In actual practice, tum-ON time (or more
properly rise time) is generally specified as that
interval between the 90% and 10% rise time, so
region 1 really extends a bit further than the threshold
voltage and, conversely, region 3 begins at the 10%
level. A careful study of Figure 3 shows this quite
plainly. The second delay is at tum-OFF when the
input capacitor must be discharged to the voltage
where the channel begins to revert to the nonconducting state. This latter state is akin to storagetime delay caused, not by minority carriers as might
be anticipated with a power bipolar transistor, but by
overcharging the input capacitor when the power
MOSFET switch was turned-ON.
contained in Figure 3, the method used to generate
the charge-transfer characteristics must be considered.
(a.2.2) Developing the Charge-Transfer
Characteristics
In Section (a.2), the charge-transfer characteristics
were derived by observing the gate voltage with
respect to time with a constant-current drive. That, in
effect, is how the characteristics shown in Figure 3
are developed. Achieving the drain voltage characteristics involves nothing more than a simple monitor on
the drain terminal of the power MOSFET under test.
The circuit for this characterization is shown in Figure
6. Although admittedly simple, a few rules must be
observed if these will be used to evaluate the charge
characteristics. The first rule is to make absolutely
sure that the base line-the abscissa-is in easy-todivide integers. If the current source is fractional, for
example 4.7 rnA, then the sweep rate should be such
that the product (iot) is 5 nC/cm and not 4.5 nC/cm
which would occur at a sweep speed of 1 p..slcm. The
second rule, which regrettably is often times beyond
control, is to seek the largest, most easy to read chart
possible to facilitate accurate measurements. A digitized oscilloscope with x and y output ports attached
to a suitable recorder would be highly desirable. The
third and most obvious rule is to duplicate the switching circuit. That is, simply be careful that the load
impedance used in the test circuit (Figure 6) duplicates that which is used in the actual circuit. The final
and potentially most important rule is to remember
that the data derived from this test provides the turnON charge-transfer characteristics which means-as
seen in Section (a.2)-that they only provide turnON delay, rise time, and tum-OFF delay [equations
(9), (10), and (11)]. To calculate fall time, a new
switching model is needed.
There are three equations which can be used with the
charge-transfer characteristics as depicted in Figure 3
to determine the three switching times: tum-ON
delay, td(on); rise time, tr; and tum-OFF delay,
td(off). A fourth, fall t,ime, tf, cannot be calculated
using the charge-transfer characteristics of Figure 3,
Q1
VGG
td( on) = Rgen Qn
Vg1
VGG - Vg1
tr = Q2 - Q1 Rgen Qn VGG - Vg1
Vg2 - Vg1
VGG - Vg2
td(off) =
Q3 - Q2
VGG
Rgen Qn VGG - Vg2
Vg2
(9)
(10)
(11)
where the first two terms within each equation, aQ
aVg
Rgen, represent the classic RC time constant where
Rgen is the driving impedance as' seen from the gate
terminal.!
(a.2.3) Developing a Switching Model for Fall Time
Already many commercially available power MOSFET
data sheets include the charge transfer characteristics
The correlation of calculated switching time using
these equations with that of measured data can be
precise if data is taken carefully from the chargetransfer characteristics such as has been done in
Figure 3.
Before addressing the so-called problem of fall time,
tf, and why it cannot be calculated from the data
+---+
!Occasionally data sheets with erroneous generator
impedance listed with the switching time parameters
are discovered. If a data sheet offers a switching time
test circuit, check whether the gate shunting resistor is
considered in parallel with the generator impedance,
viz., the correct value of
Rl x R2
R
-'--gen - Rl + R2
TO SCOPE
INPUT PULSE
..JL
CONSTANT
CURRENT
DKlOE
Circuit for Monitoring Charge Transfer Characteristics
Figure 6
2-14
1
in graphical form similar to what is shown in Figure 3.
Offering such a graph is not particularly difficult for
most vendors. For the most part, they can anticipate
most turn-ON switching configurations based on such
power MOSFET parameters as Breakdown Voltage,
BVDSS; ON-Resistance, rDS(on); and Operating
Drain Current, ID. When a power MOSFET is turned
ON, the action is well defined: the fast-dropping ONresistance eventually crowbars the output capacitance
ofthe MOSFET reducing the voltage drop to VSAT.
All of this action can be reasonably well-defined on
the vendor's data sheet. What he cannot do is to
predict turn-OFF since this depends to a large measure on the type and size of the load and whether the
load is resistive, reactive, or both. Consequently, the
burden of developing a Fall-Time Charge-Transfer
Characteristic lies with the end user and not with the
supplier. The difference between turn-ON and fall
time is shown in Figure 7. Note that this figure is
elementary and does not represent a true model of
the power MOSFET. The reader is encouraged to
review Section 2.13 for specific modeling information.
It should be obvious to the reader that to develop a
fall time characteristic, the current (charge) must be
controlled as it leaves Cgs , in other words, as the input
capacitor is discharged. The circuit might look much
like that in Figure 8. This circuit, however, presents a
very fundamental problem: how to configure the
circuit shown in Figure 6 (the turn-ON charge-transfer characteristics) to that shown here (the turn-OFF
charge-transfer characteristics). The solution is simply to switch between the two using the circuit shown
in Figure 9. Instead of a turn-ON charging diode (our
constant current source) limiting the current from the
pulse generator (Figure 6), the input capacitor of our
power MOSFET is charged from a pre-set voltage.
An analog switch is used to switch the circuit configuration from that of a turn-ON charging network to a
turn-OFF discharging network. The finite ON-resistance inherent in the analog switch is not significant
since a constant-current source or load represents a
very high resistance. What is important is to have the
voltage drop across these constant-current diodesand in particular, the discharge diode-as low as
possible to ensure that the gate charge truly drops to
as near zero as practical.
C"nA:S(O~
VDD
LtURCE
-
Cal TURN-ON
1·
J
Cd
•
(b) TURN-oFF
During turn-ON (a), Cds begins fully charged and Is discharged
through the low rDS(ON) of the MOSFET. However, at turn-OFF
Cds begins fully discharged and charges to the rail through
the load.
Figure 7
Simple Turn·OFF Charge Transfer Circuit
Figure 8
+VOD
(a.2.4) Developing the Fall-Time Characteristics
In Section (a.2.2) and the accompanying Figure 6,
the gate of the power MOSFET is charged from a
constant-current source. In the initial pre-drive
state, prior to any attempt at turning ON the power
MOSFET, the input capacitor, Cgs , is uncharged, and
the gate voltage is zero. On the other hand, the draingate capacitance, Cgd, has the full drain-gate potential, Vgd, impressed across it. When considering the
fall time, quite the opposite is true. The input capacitance, Cgs , is charged-perhaps overcharged-and
the gate-arain capacitance, Cgd, rather than having
the full drain-gate potential, enters the fall region
with a very small potential which is discussed in
Section (a.3).
----
I
I
PULSE
IN
''''
5V..JnL
TTL
I
TO SCOPE
I
r~C_:
!
I
I
I
~V:-ITOSCOPEJ-'
1
DG186
iI
I
_
NO
I
~ ~---~f-!
-
Automatic Turn-ON, Turn-OFF Charge Transfer Test Circuit
Figure 9
2·15
(a.2.5) Calculating the Fall-Time
excruciatingly long time that VSAT is finally achieved.
The question that needs to be asked is: "Why? What
makes this obvious slowdown in turn-ON occur?"
Since this slowdown occurs after the measured (and
calculated) rise time, the effect and its cause are
generally disregarded.
The results obtained from testing a power MOSFET
in the circuit just described (Figure 9) can provide
any of three views depending on how the viewing
oscilloscope is synchronized. It can be set to view
only the rise-time characteristics with results similar
to what was seen earlier in Figure 3, or the fall-time
characteristics as shown in Figure 10 can be seen,
which, incidentally, also provide the turn-OFF delay,
tct(off) , information. In the calculation of tum-OFF
delay [equation (11)], data can be extracted from
either Figure 3 or Figure 10. A third option would be
to view the entire switching cycle: tum-ON delay, rise
time, the steady ON state (which, depending upon the
test conditions, could also be the tum-OFF delay),
and the fall-time.
Section (a.2.4) began by discussing the effects of turnON on the various parasitic capacitances paying
attention to the rapidly changing voltages impressed
across the gate-drain capacitance, Cgd. When the
MOSFET is OFF, the voltage impressed across this
gate-to-drain capacitor equals the full rail voltage,
VDD. But as the MOSFET begins to switch ON, the
drain voltage drops along with the impressed voltage
across this capacitor. Figure 5 shows that as this drain
voltage drops, the capacitance rises dramatically. In
fact, as the drain voltage slowly settles toward VSAT,
the drain-to-gate voltage, V dg, actually reverses polarity! Where, at the beginning of the rise time there
was a drain voltage many times higher than the gate
voltage, now the gate voltage is more positive than the
drain! Since Figure 5 cannot show the effects of either
zero or negative drain voltages, it is necessary to view
the effect of this change in polarity on capacitance in
Figure 11.
'00
vGG
I
II
1\
\
II
~
v.'
I-- I---a,
90%
I----
v.'
I
./
''''
a~ I
,."
:III 11
I
L a, Lt
!::~
\
V
r-r""-
\
I:::::
51015202530
Q CHARGE ene)
Turn-Off Charge Decay Characteristics
Figure 10
'"F33D
.,...,
I.....
~
VN4000A
I' t'-
MTM56S
With the assistance of the discharge-transfer characteristics provided in Figure 10, it is now possible to
calculate the fall-time using the following equation:
•
1
Q2- Ql
~
tf =
Rgen Qn
Vg2 - Vg1
Vg1
-6 -5 -4 -3 -2 -1
0
1
2
3
4
5
8
7
8
9
Yoo - DRAIN-YO-GATE VOLTAGE (VOLTS)
(12)
Gate-Drain Capacitance versus Drain-Gate Voltsge
Figure 11
Using these four Equations-(9), (10), (11), and
(12)-with the data obtained from Figures 3 and 10,
all of the switching times can be calculated quite
accurately.
During the initial rise time (region 2 of Figure 3) ,
there is the contribution of a gradually rising gate-todrain capacitance, Cgd, on the Miller effect which,
with some difficulty, could have been determined
from equation (4). Now after having completed the
turn-ON cycle (VDS dropping to within 10% of its
original value), the voltage gain, AV, has also dropped
(iJVDsliNoS --> 0), but because of the reversal of
polarity across the gate-drain capacitor, its capacitance has soared to astounding heights. As a consequence, there is an extended Miller effect and
slow decay of the drain-to-souree voltage to VSAT
(region 3):-
(a.3) Region 3: A Closer Examination
In attempting to show how to correlate switching
speed with the charge-transfer characteristics of power
MOSFETs, an important point was by-passed. By
closely examining region 3 of Figure 3, one should
question why the power MOSFET, during its turnON cycle, does not turn-ON fully. Not only does it not
turn-ON, but the Miller effect holds down the gate
voltage build-up. It is only after what amounts to an
2-16
/
2.3 Planar MOS
When small signal MOSFETs (such as those found in
FM and TV receivers) were in vogue, someone might
have asked if they could be used for a higher power
than the few milliwatts obtained, and the answer
would have been a resounding "No!"
The construction of a planar small-signal MOSFET
reflects a unique symmetry in that both the drain and
source diffusions are, for all practical purposes, identical. What actually differentiates source from drain
is the biasing arrangement; the gate is generally
biased with respect to the source.
The small-signal MOSFET was, in hindsight, simplicity itself. The structure is shown by the n-channel
depletion-mode device in Figure 1. If our desire was
to turn-ON this MOSFET, a positive gate voltage
would accomplish it. When the gate receives a positive charge, the capacitor effect forces an accumulation of electrons to migrate under the oxide. But since
there is a p-doped channel beneath this oxide, the
migration of electrons effectively inverts this region
into what might be termed p-doped. Once inversion
occurs, current can flow from source-to-drain since
the entire path including both the source and the
drain has n-accumulation throughout. Conversely, a
negative gate potential depletes the region, enhancing
the p-type characteristics of the channel and preventing any current flow aside from any leakage that exists
between the pair of back-to-back pn-diodes over
which the gate has no control.
SOURCE
GATE
The p-channel MOSFET is the exact inverse of the
n-channel version. Where the n-channel MOSFETwas
n-doped, the p-channel is p-doped. All that remains
identical to the n-channel MOSFET is the oxide, the
gate structure, and of course, the metallic contacts to
the gate, source, and drain. Operationally it, too, is
reversed. Now a negative bias enhances current conduction through the channel.
The obvious extension combines n-channel and pchannel MOSFETs on the same substrate into what is
popularly known as CMOS (complementary MaS).
It is one of the foremost technologies today extending
its influence from the fabrication of analog switches
to microprocessors.
Why the small-signal MOSFET is a small-signal
MOSFET and incapable of handling power can be
simply explained if the basic equations for current are
examined both in the linear region and in the saturation region.
DRAIN
In the linear region, the drain-to-source current is
given by:
W
[
VDS]
IDS = "LJ.IoCOX VDS - -2- VDS
3-5 pm
Planar Small-Signal MOSFET
Figure 1
2·17
-
where:
W = device channel width
L = device channel length
fJ. = electron mobility in the inversion layer
Although this restriction was well understood, there
were, nonetheless, a few attempts during the 1960's to
achieve power by paralleling multiple chips. These
attempts failed for many reasons: an incorrect understanding of the contribution of parasitic elements
(notably capacitance), improper headers and die
attach methods to maintain low thermal resistance,
and prohibitive costs.
In the saturation region, it is given by:
IDS
fJ.CoxW
(VGS-VT)2
2L
= --
In the early 1970's, investigators began in earnest to
resolve the conflict of high voltage and current in their
efforts to overcome the problem. The novel technology that evolved was called double-diffused MOS
which made possible the revolution in discrete power
MOSFETs that is occurring today.
The ~ndamental problem lies in the single term L,
the effective device channel length. If a MOSFET is
designed to withstand high voltages, the channel
length must be increased to reduce the deleterious
effects of the body-drain diode (pn junction) depletion region that, because of the increasing voltage
gradient, moves progressively further into the channel
region. If the channel length is not increased, punchthrough breakdown results, causing catastrophic destruction of the MOSFET. However, careful inspection of these two equations identifies that current is
inversely proportional to L; consequently, increasing
the breakdown characteristics reduces the power
handling capability!
References
Richman, Paul, MOS Field-Effect Transistors and
Integrated Circuits, New York, John Wiley and Sons,
Inc., 1973.
Sigg, Hans J., et. aI., "DMOS Transistor for Microwave Applications," IEEE Trans-Electron Devices,
Vol. ED19, 1 January 1972, pp. 45-53.
2-18
,I
2.4 DMOS Double-Diffused MOS
Realizing the fundamental limitation of conventional
planar MOS-a channel length proportional to
breakdown voltage but inversely proportional to current-researchers focused upon a new structure which
simultaneously accomplished both high voltage and
higher current. During the decade of the 1970's, the
literature described progressively advanced designs
using this new structure-what became known as the
double-diffused MOS.
(a)
SOURCE
GATE
SOURCE
p+
n+
DRAIN
Simultaneous with the progressive development of
the double-diffused MOS technology was the evolution of the MOS gate. Previously the gate consisted of
a metal overlay on oxide, but this new technology
used a layer of polycrystalline silicon as shown in
cross-section in Figure 1. Among the advantages
offered by this polycrystalline silicon gate structure
was the ease in interconnection between cells which
now could be done in diffusion rather than by the
more awkward techniques of metalization and bonding. A more significant advancement was the "selfaligning" polycrystalline structure that now could be
controlled by a mask rather than by using metalization. Consequently, interelectrode capacitances fell,
thus affording higher frequency (faster switching)
performance for the MOSFET.
(b)
SOURCE
GATE
SOURCE
Cross-Sectional ComparIson: Metal-Gate (a)
vs. Silicon-Gate (b) OM OS
Figure 1
doped region followed by a more heavily doped nregion. A typical profile showing this double-diffused
structure is shown in Figure 2. The channel length is
controlled by careful monitoring and control of the
doping levels and the subsequent diffusion cycle which
must consider both time and temperature.
A double-diffused structure consists of a sequentially
introduced set of impurities into the epitaxy that for
an n-channel MOSFET consists of first a deep p-
2-19
-
GATE
CHANNEL LENGTH
SOURCE
These two differences allow a MOSFET that has both
a short channel length and the ability to withstand a
high drain potential without fear of punch-through.
GATE
DMOS transistors exhibit a difference in their drain
current-to-gate voltage relationship. With the simple
planar MOS structure, a square-law relationship is
identified, and in the short-channel DMOS structure,
the accelerating field potentials can easily reach values exceeding 104 V/cm. When this occurs, the drain
current becomes linearly related to the gate voltage,
as given by the equation:
DRAIN
Planar Vertical Double-Diffused Power MOSFET
Figure 2
(1)
One obvious difference between the DMOS structure
and the conventional planar MOS structure is the
channel length, L, determined by the difference
between two sequential diffusions. Unlike the conventional planar MOS where the channel length is
achieved through the photolithographic process and
limited to 4 to 5 f.Lm, the DMOS channel length can
be reproducibly controlled to values in the 1 to 2 f.Lm
range!
Furthermore, as this equation shows, drain current is
no longer dependent upon channel length, L. Additionally, we can manipulate equation (1) to show that
transconductance, gm, is independent of Vas.
dIDS
gm = - - = Cox· VSAT
(2)
dVas
Operationally, the DMOS transistor works similarly
to the planar MOS. An n-channel enhancement-mode
DMOS transistor requires the application of a positive gate potential to cause surface inversion of the
p-region. Once the surface has inverted, a conducting
channel then bridges the source and drain-drift region
allowing conduction to occur.
A less obvious difference, but one of paramount
importance, is what happens to the pn depletion field
that formerly forced the construction of a long channel in the conventional planar MOS. In the DMOS
process, the body region (for the n-channel MOSFET,
the p-doped area) is more heavily doped than the ndrain (epitaxy) region. Consequently, the depletion
field extends further into the drain region than into
the body region when a reverse bias is placed across
the drain-to-body (np) junction. This not only allows
significantly higher voltages to be placed across the
junction without forcing a longer channel, it also
maintains a fixed threshold voltage with varying
drain-to-source potential.
This fundamental double-diffused process opened the
door for the fabrication of at least three basic highpower MOSFET designs: the V-groove transistor, the
lateral DMOS (LDMOS), and the popular vertical
DMOS structure (VMOS), usually called DMOS. All
of these power MOSFETs perform similarly as we
shall now examine.
2-20
2.5 V-Groove MOS
The first commercially available power MOSFET had
a unique way to achieve the performance goals that the
double-diffused MOS was expected to offer. It was,
after a fashion, a truly double-diffused transistor despite its somewhat unconventional design. With the
conventional double-diffused technology, the channel
length was controlled by precise lateral diffusion;
however, with V-groove, the control was vertical. It
was critical to the performance of the device that a
narrow channel be available; as a result, the structure's
design began closely resembling the four-layer construction typical of the high-frequency transistor as we
see in Figure 1. To gain access to the channel, it was
necessary to anisotropic ally etch a groove that, once
completed, could have an oxide overlay, and then
have the gate metal sputtered. The finished V-groove
cross-sectional view is shown in Figure 2.
BASE
EMlnER
SOURCE
OXIDE
OXIDE
II1II
DRAIN
V-Groove OM OS Power MOSFET
Figure 2
SOURCE
GATE
SOURCE
BASE
CHANNEL LENGTH
OXIDE
OXIDE
= SIN\30
n+ SUBSTRATE
n+
DRAIN
V-Groove OMOS
Figure 3
COl.LECTOR
Bipolar Transistor
Figure 1
that this sharp-edged trough made reliable high voltage performance difficult; consequently, a truncated
V-groove was developed. This new design is shown in
Figure 3.
Early in the development of these devices, the Vgroove was etched to a sharp trough. Once higher
voltages were demanded, it was quickly recognized
2-21
Before beginning the discussion of V-groove technology, it is necessary to recognize that silicon is a
crystalline allotrope element with three well-known
crystallographic planes - < 100>, < 110>, and
< Ill>. To build a V-groove structure, the starting
silicon is always < 100>. The V-shaped groove results from what is known as preferential etching that
follows the silicon crystallographic plane. Without
interference, the etching process terminates when the
sharp V-groove is formed; however, since a truncated
etch is desired, it is critical to the fabrication that
accurate timing be employed. Nevertheless, since the
etch is "controlled" by the crystallographic plane of
the silicon, the sides of the V-groove form a 54.7°
angle with the surface and terminate at the < 111>
plane. Because of the extreme care which is required
in the manufacture of V-groove transistors, it is only
used when the desired parameters cannot be achieved
in any other way.
Performance of this V-groove transistor follows that
of the DMOS transistor closely with the following
notable exceptions:
1. If the same diffusion schedule is maintained for the
V-groove structure as for the DMOS structure, the
channel length is longer by the geometrical factor
associated with the 54.7° angle. The difference is
approximately 1.7 times the channel length of an
equivalent planar DMOS transistor.
2. Electron mobility is dependent upon the crystallographic plane, and for V-groove, the < 111>
plane at the surface (channel) of the V -groove
results in a reduction of this mobility of approximately 25 percent.
3. Threshold voltages tend to be higher for the same
body dopant than a comparable DMOS structure as
a result of a higher oxide charge inherent in
< 111> structures.
The channel region of an n-channel V-groove transistor is the p-diffusion (shown in Figure 1). To achieve
electron velocity, this region must be short; it is in the
design of high-frequency bipolar transistors to
effectively reduce transit time. However, unlike the
high-frequency bipolar transistor, reducing the length
of this p-diffusion does not affect the operating breakdown voltage of the V-groove transistor. Access to
this channel is accomplished using the V-groove. It is
crucial to the success of the VMOS that the V-groove
(truncated or not) extend slightly beyond the pdiffusion depth.
References
[1] Fuors, Dennis and Verma Krishna, "A Fully
Implanted V-Groove Power MOSFET," Technical Digest, IEEE Electron Devices International Meeting, Washington, D.C., 1978, pp.
657-60.
[2] Heng, TMS, et. aI., "Vertical Channel
Metal-Oxide-Silicon Field Effect Transistor,"
Final Report, Westinghouse Research and Development Center, Pittsburgh, PA, November
1, 1976.
2-22
,
2.6 Lateral Power DMOS (LDMOS)
Although lateral small-signal DMOS structures appeared in the closing years of the 1960's, power
devices only became available commercially in 1977.
LDMOSFET can exhibit an IDZ of 100 rnA compared to a more conventional VDMOS IDZ of 3
amps.
The lateral DMOS transistor, shown in cross-section
in Figure 1, has all three terminals-source, gate,
and drain-on the top surface. This topology offers
unusual advantages as well as several disadvantages
that have limited the application ofthe LDMOS.
2. The gate-to-drain capacitance, Crss , is low. A low
C rss offers the advantage of high switching speeds
combined with high-frequency operation.
The following advantages are offered by the lateral
construction:
The disadvantages of lateral construction are:
3. The drain is isolated from the case.
1. It has a severe geometric constraint and an ineffi-
cient utilization of chip geometry. This constraint
tends to limit the high voltage capabilities because
limitations are imposed by the topology on the use
of field shaping techniques.
1. It has a very low zero-temperature coefficient drain
current, IDZ. For example, an 8 amp, 200 volt
2. As with the small-signal lateral MOSFET, the
depletion region of the pn body-drain diode spreads
laterally and thus adds to limitations in topology.
Reference
Richman, Paul, MOS Field-Effect Transistors and
Integrated Circuits, New York, John Wiley and Sons,
Inc., 1973.
CHANNEL
Cross-Sectional View of Idealized Lateral OM OS Structure
Figure 1
2-23
-
2.7 Vertical DMOS
The disadvantages recognized as inherent in the
lateral DMOS structure have paved the way for
industry to examine the vertical DMOS FET (DMOS).
a series-connected JFET in the drain because the
physical relationships between adjacent p-diffusions,
acting as gates, pinch the current flow in its vertical
descent to the backside drain. A simplified view of
this restriction is shown in Figure 3. The spacing
between the adjacent p-diffusions can be crucial in
the performance of a high-power (current) DMOS
PET. A more detailed analysis appears in Section
The DMOS structure is conceptually similar to the
LDMOS transistor except the drain contact lies on
the backside ofthe wafer. Consequently, current flow
is more vertical than horizontal as the cross-sectional
view in Figure 1 illustrates.
SOURCE
GATE
2.13.
SOURCE
GATE
cr---J
9A
1
LoGATE
DRAIN
Figure 2
Figure 1
The advantages of the DMOS structure, when compared with the lateral (LDMOS) structure, are simply
the elimination of the disadvantages outlined previously for the LDMOS. However, although we gain in
chip topology utilization, we do add another topology-related limitation which becomes increasingly
more pronounced as the current flow rises.
DEPLeTION
BOUNDARY
The electrical model of the vertical DMOS, although
basically the same as any other DMOS structure, does
offer an anomaly that distinguishes it from all other
structures. As shown in Figure 2, there appears to be
Figure 3
2-24
References
Hu, C., "A Parametric Study of Power MOSFETs,"
Conference Record PESC, San Diego, CA, 1979.
Lidow, A., T. Herman and H. Collins, "Power MOSFET Technology," Technical Digest, IEEE Electron
Devices International Meeting, Washington, D.C.,
1979, pp. 79-83.
Lisiak, K. and J. Burger, "Optimization of Nonplanar
Power MOS Transistors," IEEE Trans-Electron Devices 25 (1978), pp. 1229-34.
-
2-25
2.8 The Bilateral MOSFET:
A New Device Sees New Applications
For many years there has been a need for a semiconductor capable of general purpose AC control. While
there have been, and continue to be, some special
purpose devices available which will do part of the
job, no truly general purpose device for AC control
was available. This situation has just changed. Advances in the field of MOSFET technology have now
produced the first truly general purpose AC semiconductor: The Bilateral MOSFET.
DRAIN
J
l
GAIT~9-aSOURCE
Until now, engineers wishing to control AC had
numerous problems and few solutions. For low frequency AC switching, one could use thyristors, but
they do not work at high frequencies and are hard to
turn OFF. For switching both ON and OFF, one could
use a bipolar transistor inside a diode bridge, but this
is inefficient, cumbersome, expensive, and slow. For
linear low-level applications, special attenuator and
MUX ICs are available, but these function at low
power levels only. For either linear or switching
control of high level, high frequency AC, you were
generally out of luck. Now, the bilateral MOSFET
can handle all these jobs.
DRAIN
Bilateral MOSFET-Symbol
Figure 1
New bilateral switches, such as the Siliconix BLSlOO
family, give the designer requiring analog switching,
subscriber line switching, or AC power conduction, a
nearly ideal combination of characteristics which
evades many of the compromises required by the
approaches now in use. These n-channel enhancement mode devices have no inherent offset voltage
and can transmit AC or DC signals with equal facility.
Furthermore, the impedance in the ON-state is resistive, providing very low harmonic distortion.
Bilateral MOSFET switches are a new product resulting from recent advances in power MOS technology.
In terms of construction, the bilateral switch may be
thought of as two n-channel DMOS power FETs with
their source and gate leads connected in common. In
fact, the first bidirectional devices were made by
connecting two discrete power FETs in this fashion.
The device that resulted from this pairing had such
impressive characteristics that unified bilateral switches
are now being manufactured.
In operation, when a positive voltage greater than the
threshold voltage (VT) is applied to the gate (with
respect to the source), both FETs turn ON and
conduct in series via their channels (see Figure 1).
When the gate signal is removed, both channels turn
OFF and conduction ceases. The source-to-drain
diodes of the individual FETs do not conduct in
reverse as they do in a unidirectional MOSFET
because such conduction is blocked by the diode in
the other half of the device.
Reprinted with permission from Electronic Design, September 6, 1984;
copyright Hayden Publishing Co., Inc., 1984.
2-26
Today, isolated gate drive presents few problems,
both because of a proliferation of devices meant to
accomplish this function and because FETs require
so little static drive power. The designer has a choice
of optical, transformer, or capacitive isolation, as well
as combinations thereof. Each type of drive has its
own properties. The choice of drive will depend on
the system's requirements.
Either of the diodes (depending on the drain-to-drain
voltage polarity) may conduct when the device is
turned ON but it will only conduct when the voltage
drop across half the channel resistance exceeds the
forward voltage of the diode. When one of the diodes
does turn ON, it clamps the voltage across its associated half of the channel. This feature both reduces
power loss in the bilateral switch at high power levels
and serves as a surge protection in signal applications.
Both diodes will conduct from the source to their
respective drains if the source is biased above drain
potential, a condition which should be avoided. Note
that despite the fact that the two halves of the device
are n-channel and operate in the enhancement mode,
the resulting device is symmetrical and conducts
identically in both directions.
Transformer isolation may be either linear or switching in nature. Linear transformer drive is usually
accomplished by rectifying a variable amplitude high
frequency tone sent through the transformer. Transformer switching drive is accomplished by driving the
transformer at two opposite polarity voltage levels:
one for ON, the other for OFF. Note that proper voltsecond balance must be maintained in the transformer, or the core will saturate. To prevent saturation, the ExT applied to the primary in one direction
must equal the ExT applied in the opposite direction.
By making the OFF voltage greater than the ON
voltage, ON intervals greater than 50% of total period
can be achieved. Using 40 V gate breakdown FETs,
this technique allows ON intervals to reach 80% of
total period while maintaining proper volt-second
balance in the transformer. Automatic control of voltsecond balance in the drive transformer may be
achieved by placing a suitable capacitor in series with
either the primary or secondary of the transformer.
This will also reduce the reverse voltage stress on the
gate oxide when duty cycles are below maximum.
Good gate drive for the bilateral MOSFET switch
requires some new, yet not difficult techniques. The
gate-source terminal pair may be viewed as a voltagesensing capacitive load exactly as in a unidirectional
MOSFET. Like the unidirectional MOSFET, input
capacitance (Ciss) will vary with chip size, and thus,
with voltage and current ratings. Bilateral devices will
have approximately double the input capacitance of
unidirectional devices of equal rating. The available
threshold voltage (VT) range will also be the same as
for unidirectional devices and is well controlled on
the bidirectional devices to assure symmetry.
There is one striking difference between the bilateral
MOSFETswitch and any other device (see Figure 2).
When the device is ON, the source terminal (gate
return) is electrically in the center of a resistive divider
formed by the two gate channels, and when the device
is OFF, the source is referenced one diode drop above
the more negative drain terminal. In many cases it
will be necessary to use an isolated gate drive in order
to avoid common mode effects. This seems a fair
trade for true symmetrical AC capability and may, in
some cases, be a convenience.
Dl
J
50% ON TIME
DR:=JII~~J""
P'
+ 10 v--
63385
75%
ON TIME
Dl
- 30
v---------
Extending the Duty Cycle
of a Transformer Isolated MOSFET Drive
Figure 3
Rl
R2
Transformer drive, especially when using shielded
transformers, offers the best common mode rejection
ratio (CMRR). However, transformers tend to be
bulky and expensive, and if it is necessary to build
them to UL or VDE insulation requirements, performance (primarily coupling) will suffer significantly,
especially at higher frequencies.
(R1 "" R2)
D2
D2
EaUIVALENT DC CIRCUIT
OFF STATE
EQUIVALENT DC CIRCUIT
ON STATE
Source Reference
Figure 2
2-27
-
ing special purpose ICs) has evolved making old
methods more practical. Modern capacitively isolated switching drives consist of a gated oscillator
driving a rectification circuit through two matched
capacitors (see Figure 5). The frequency of the
oscillator must be high, perhaps 100 times the desired
switching frequency. ON/OFF control of the MOSFET is achieved by gating the oscillator ON and
OFF. The isolator in Figure 5 has a 1 MHz oscillator frequency and works well switching signals of
:;;; 10 kHz and:;;; 30 VAC. Some telecommunications
companies are studying this method to control bidirectional MOSFET switches for signal switching.
Linear control could be achieved similarly by amplitude modulation of the oscillator.
AC
1N5615
r --LOGOC{:'l
INPUT
--f..T. .
L __ _
68 K 1 W
AC
Parasitically Powered, Optically Isolated Drive
Tr < 75 nsec Tf < 50 nsec 120 VAC Switch
Figure 4
There are still limitations to the technique, however.
CMRR is low, as would be expected. Capacitors,
after all, conduct AC equally well in both directions.
Presently available systems are better suited for linear
controls, or for switching under tightly constrained
circumstances than for high power switching systems
where large common mode voltage transitions can be
expected. If the load frequency being switched is too
high, spurious turn-ON of the bilateral MOSFET may
result. Present devices, which use a 1 MHz oscillator,
can control frequencies of 5 to 10 kHz at levels suitable for telephone switching, but a 50 VRMS, 20 kHz
signal, for example, would make some of the present
units appear very leaky. The necessity of running the
oscillator at frequencies much higher than the maximum frequency of the switched device also places a
constraint on either the switching frequency of the
main switch or the cost of the oscillator. Faster chips
can be expected to alleviate this problem in the
future. High DC isolation voltages are also a problem.
The dielectric materials available for capacitors included in integrated circuits now place a limit on the
maximum DC standoff voltage of such capacitors at
500 volts DC. Improvements in standoff voltage can
be expected to come slowly.
Optically coupled gate drive, likewise, may be either
linear or switching in nature (see Figure 4). Generally, optical drives are smaller and less expensive than
transformer drives and somewhat slower. Modern,
shielded, optocouplers do, however, have good
CMRRs, approaching the best transformers. Furthermore, units with sufficient voltage swing to drive
a FET with switching times in the tens of nanoseconds
are now available. Even using such high-speed devices, optocoupler speed remains the primary limitation on system switching speed. Sophisticated optical
drives made from discrete PIN photo diodes and fast
amplifiers can improve switching speed significantly
when the extra speed is found to be worth the extra
cost.
Linear optical drives suffer primarily from nonlinearities in the optocoupler. These nonlinearities may be
evaded by several means. The most common system
is executed by using two matched optocouplers, each
of which serves as a local feedback path to compensate for the nonlinearities in its companion. Similar,
but more elaborate, systems can be made which are
adequate for precision applications. Again, the primary limitation on the small signal gain bandwidth of
linear optocoupling in terms of FET drives is cost.
To recap briefly, with optically isolated drive, you can
perform any drive function except very rapid switching. With transformer isolated drive, you can perform
any drive function except keeping the switch in the
Capacitively isolated drives are both an old system
and a very new innovation. New technology (includ-
20 pF, 500 V
1/44011
1/6 4049
1/& 4049
AC
Capacitor Isolated Gate Drive Driving a Bilateral Switch
FigureS
2·28
Until now, the Venturini converter has had two
horrendous drawbacks which limited its applicability.
First, an N-phase to M-phase Venturini converter
requires NxM AC power switches with individually
isolated drives. In practice, this means using 2(NxM)
unidirectional switches, each with its own drive and
protective power diode. For a 3-phase to 3-phase
converter, this adds up to 18 assemblies, each consisting of a power transistor, its drive electronics, a fast
power diode with the same voltage and current rating
as the transistor, and an isolated driver. The resulting
system is incredibly complex, and only those who
truly need the multifunctional capabilities of the
Venturini are willing to bear the cost.
ON state indefinitely. Transformer drive requires the
most volume. By using capacitor-isolated drive, you
can reduce costs, but only if you can accept limitations
on load frequency and slower switching.
One of the most obvious applications for the bilateral
MOSFET is in the Venturini converter. The Venturini
converter is an elegant voltage-independent, frequency-independent, bidirectional N-phase to Mphase switching converter. For example, a Venturini
converter can be built to convert 3-phase 60 Hz power
to 3-phase 400 Hz power bidirectionally, in a single
stage, at well over 80% efficiency while regulating
output voltage and input power factor. Venturini
converters can also be built to convert 3-phase power
to 2-phase, 5-phase, or any other number of phases.
They can be built to change frequency to either a
higher, lower, or variable frequency, including DC.
Because Venturini converters are bidirectional, they
will also function to synthesize multiphase sinusoidal
power from DC or single-phase AC. They may also
be used as regulators and power factor correctors.
They may also be built as bidirectional synchro-toresolver converters, or, with an extra A to D stage, as
an either-to-digital converter. The Venturini converter is probably as close to being a power anythingbox as any currently known device.
JO.
,.,
,.,
,.,
TO
ROTAnON
AND
AMPLITUDE
PROCESSOR
Y
JO.
,.,
,.,
Y
This problem is now solved. Bilateral MOSFET
switches conduct AC naturally and require a much
less complex drive (though isolation is still necessary).
In a 3-phase to 3-phase converter using bilateral
MOSFET switches, several reductions take place:
nine power components replace 36 power components; nine drivers replace 18; and drive complexity,
cost, and volume are reduced.
The second drawback to the Venturini converter is
the lack of optimized control algorithms. Progress in
solving this problem is more likely now that a practical
J
on.
3-,
,.,
,
Y
I
I
I
I
I
T
I
JO.
J
on.
T
,-3
.JO.
I
J
3-,
SYMMETRIC
MODE
\
-'N
ANTISYMMETRIC
)
3-Phase to 3-Phase Venturini Switch Matrix
Figure 6
2·29
MODE
--
-D.2Y
I-""
-DAY
~ ~ 10-
--
r-
-o.ev
-Day
-1.0~
YDS - DRAIN TO SOURCE VOLTAGE (V)
1.0
~~
A ~V
0 .•
0.'
0.7
0••
Ib)
Enhancement·Mode
Operation
05
,
OA
I')
0.2
0.1
+~
~
+ Day
+ 0.2 V
YGS-OY
~~
03
DepletLon-Mode
..4~ ;...--- r~ ........
~ ~ /"
I#}. /"
,/
J
1/
Vr
VDS - DRAIN TO SOURCE VOLTAGE (V)
The NOS100 Functions Both as a Depletion-Mode
MOSFET (a) with VGS < 0, and an Enhancement-Mode
Device (b) with VGS > 0
Figure 2
It is important to note that the on-resistance
(RnS(on» of a NOS100 may vary substantially
from device to device with zero applied VGS. This
variation can be as large as 100%. Enhancementmode devices with similar specifications, on the other
hand, may show only a 20 to 30% maximum variation in RDS( on).
Ib)
Enhancement-Mode
The Difference Between an n-Channel Depletion-Mode MOSFET
and an n-Channel Enhancement-Mode Device, is that IDSS is
Substantially Greater than Zero for the Former, and Equal to
Zero for the Latter
Figure 1
Unexpectedly, this behavior results from the differing test conditions used when measuring RDS( on)
for each type of device. Enhancement-mode power
MOSFETs generally have a threshold voltage of2 to
4 volts, and are tested for RDS( on) at VGS = 10
volts. The effective enhancement voltage (V GS-VT)
can therefore be as large as eight volts, but is no less
than six volts.
Unlike a JFET which is a depletion-mode only device and can accept only a single polarity of applied
VGS, the NOSIOO can accept either a negative or
positive VGS. With a negative applied VGS, the
NOS100 works as a depletion-mode device with
ID < IDSS. But with VGS positive, the NOS100
The NOS100 and other depletion-mode power
MOSFETs, however, are normally tested for
2-34
a standard 1 rnA current-regulator diode (CRlOO)
that supplies a constant collector current for the
bipolar reference transistor Q 1. This transistor sets a
constant voltage drop of one VBE across resistor
RS. By varying the value of RS, adjustable output current is obtained. This circuit works well
for currents up to several hundred rnA, but its
sustaining voltage is only equal to the breakdown
voltage of the current-regulator diode.
RDS(on) with VGS equal to zero volts. Since
the typical threshold voltage variation of the
NOSIOO is from -2 to -4 volts, the effective
enhancement voltage (V GS - VT) is then only
+2 to +4 volts.
For most power MOSFETs in the 100 to 150 volt
range, the resistance of the conducting channel comprises 25 to 40% of the total device on-resistance.
The conducting channel resistance (RCHAN) can
be shown to be inversely proportional to the effective
enhancement voltage (V GS-VT). Consequently, an
enhancement-mode MOSFET will have an RcHAN
that varies only by about 30% over its 2 to 4 volt
threshold range. A depletion-mode MOSFETwith a
- 2 to - 4 volt threshold, however, will exhibit an
RCHAN that varies by 100% over this range. It is
this variation in RCHAN that accounts for the wider
spread in RDS(on) for the NOSIOO and other
depletion-mode power MOSFETs.
By connecting the gate terminal of an NOS 100 back
to the other end of a source degeneration resistor, as
shown in Figure 4, a simple adjustable current regulator is obtained. This two-component regulator
performs just as well as the circuit in Figure 3, but
uses fewer components and requires less board space.
Applications
1. A two-terminal adjustable current r~gulator
Many analog circuits require a constant current to be
generated that is independent of supply voltage. This
current may be used to bias an amplifier stage, such as
a differential transistor pair, or may be used to charge
a timing capacitor which generates a sawtooth voltage waveform.
The NOS100 and One External Variable Resistor Form a Simple
but very Useful Two-Terminal Current Regulator
Figure 4
Two-terminal current regulator diodes such as the
Siliconix CR series of devices have been available for
a number of years. While these devices are very
useful in certain applications, they are limited by
their maximum voltage and current ratings (100 volts
and 4.7 rnA, respectively). Clearly, if constant currents larger than 4.7 rnA are required, and at higher
supply voltages, the designer must choose some form
of discrete regulator circuit.
Because the negative gate bias for the NOSIOO is
generated by the drain current flowing through RS, _
one can solve analytically for the value of this resistor
given the desired ID and the two parameters VT and
lOSS. To accurately determine RS, however, both VT
and lOSS must be known accurately as well. This
necessitates measurement of VT and IDSS for every
device, since these parameters may vary widely from
device to device.
Until now, the best way to implement a high-voltage
two-terminal adjustable current-regulator was with
an enhancement-mode MOSFET and a bipolar
transistor as shown in Figure 3. This circuit employs
Rather than using the analytical approach to determine the required value ofRS given a desired value of
ID, a simpler method would be to replace RS
with a multi-turn trimpot. This trimpot can then be
adjusted until the desired operating current is reached.
With RS being the gate bias resistor, the maximum
possible output current will be equal to IDSS. This
current is reached when RS equals zero. It is not
possible to obtain values ofID in the low f.J-A range,
unless the maximum value of RS is quite large
(around 1M a ). This behavior occurs because the
VGS of the NOSIOO is equal to - IORS. For
VGS to approach VT, progressively larger values
of RS are required. But VGS can never be equal
to VT because ID would then be equal to zero
and there would be no gate bias-a paradoxical
situation.
This Two-Terminal Adjustable Current-Regulator Uses Four
Components Including an Expensive Current Regulator Diode
Figure 3
2·35
Certain applications, particularly those in the telecom domain, often require a logic controllable AC
switch with high breakdown voltage as well. The
major difference here is that the switch must remain
on whenever there is a failure in the gate drive circuitry or its power supply. Until now the only way of
obtaining normally-ON switching action, combined with high breakdown voltage and moderate
current handling capability was with electromechanical relays. The disadvantages of relays are
numerous and include large drive power requirements, excessive EMI generation, finite contact
lifetime and slow switching speed. Depletionmode power MOSFETs with their normally-ON
resistive characteristics provide a solution to the
problem of building a normally-ON solid-state
AC switch.
In most real applications for the circuit of Figure 4,
the required output current will probably be in the
range of a few rnA to several hundred rnA. The
lkO trimpot chosen for RS, provides an ID adjustment range of approximately 5 rnA to 500 rnA.
To illustrate the performance of the NOSIOO as a
current regulator, the 1kO trimpot used in the circuit
of Figure 4 was adjusted until ID was equal to
30 rnA at a VF offive volts. Figure 5 shows how the
output current varied when VF was varied between 0
to 200 volts. As can be seen, the output current
changed less than 1 rnA ( or 3.3 % ) resulting in a high
output resistance of approximately 200 kO.
401-+-+_+-+-+-+-+----1_t___l
The VQ5000 depletion-mode MOSFET array contains four devices. These transistors can be used as
the basic switching elements in a dual normallyON solid-state AC switch shown in Figure 6.
When the sources and gates of each pair of
MOSFETs are connected together, two bilateral
switches are formed. This switch configuration
can conduct current and block voltage in both
directions.
1
1301r=+=~~=F~=t=1=-r_t_l
13
~
~20~+-+-+-+-+-+-+--1-t---l
I
9
10 J-+-+_+-+-+-+-+---1I_t___l
40
80
120
160
VF - FORWARD VOLTAGE DROP (V)
200
The rest of the circuitry shown in Figure 6 comprises
a capacitive isolated driver stage that generates logic
controllable DC bias for the two bilateral switches.
Because the MOSFET's gate and source terminals
are isolated by two capacitors from the logic circuitry, the AC signal being switched need not be referenced to the logic power supply ground. This is a very
useful feature of the circuit
The Two-Terminal Current Regulator Using a NOS100 Performs
Very Well OVer an Applied VoHage Range of 0 to 200 V
Figure 5
When the current regulator is to be operated at
moderate current levels with large sustaining voltages, a small clip-on heatsink should be attached to
the NOSIOO because the temperature coefficient of
ID is positive. This recommendation was followed
for the 30 rnA regulator described above.
Each half of the driver circuit in Figure 6 is identical,
and its operation can be understood if the signal path
is traced from the LM555 output to the gates of the
MOSFETs.
2. A normally-ON solid-state AC switch
The LM555 is connected as a free running oscillator
and generates a rectangular waveform at approximately 100 kHz. This signal is fed to the inputs of a
Siliconix D469 CMOS quad driver Ie. Each pair of
drivers within the D469 buffers the 555 output, and
generates antiphase 100 kHz outputs that swing between ground and 12 volts. The remaining inputs of
the two pairs of drivers are connected to control logic
that allows the D469 outputs to be gated on or off as
required.
Solid-state AC switches that can handle reasonable
amounts of power have been implemented with a
variety of semiconductor devices such as bipolar
transistors, SCRs, triacs and MOSFETs. Of these
devices, only MOSFETs offer high breakdown voltage, low leakage, and linear on-resistance down to
zero volts.
If a solid-state AC or "bilateral" switch is built using
enhancement-mode MOSFETs, then the switch is
by nature normally-open or normally-OFF. This
means that when the gate bias voltages for the
MOSFETs are lost due to failure in the gate-drive
circuitry ( or. its power supply), the switch turns off. In
many AC switching applications this characteristic
behavior is acceptable.
The antiphase D469 outputs are fed into a pair of AC
coupling capacitors that drive a bridge rectifier comprised offast small-signal diodes. Both outputs of the
bridge rectifier are connected to the bilateral switch,
such that when the outputs of the D469 are pulsing
2·36
SWITCH 2
1/2 VQ5000
SWITCH 1
1/2 VQ5000
5,
'3
5,
52
52
10
10 kn
10kO
l"~"
.5,1.1F
8 9
5 •
TANT
SWITCH 2
CONTROL
.nJlJl
110kHz
'---"0.0'---+--------0
OFF
01 -ON
1/6CD4D69
1/6 CD4069
+ 12V
+12V
SWITCH.
CONTROL
CD4069
'kfl
PINS 5,7,9,11,13 "" GND
PIN 14
"" + 12 V
2kn
LM555
00022
I"F
A VQSOOO Quad Depletion-Mode MOSFET Array, a 0469 Driver
and a 555 Timer Form the Basis for an Isolated, Dual Normally
On-Solid State Switch That Is Logic Compatible
Figure 6
the VGS of the MOSFETs is negative. Under these
conditions the bilateral switch is OFF.
constant value. Not surprisingly, this value of
limiting current is equal to the IDSS of the
MOSFETs. The bilateral switch thus has built in
current limiting which is a useful self-protecting
feature.
When the logic control signal gates the antiphase
0469 outputs off, the output of the bridge rectifier
drops to zero. Consequently, the effective gate to
source capacitance of the MOSFETs discharges
through the 10 kf! resistor, and the bilateral switch
turns on. The switch also turns on when the logic/
driver power supply fails, since the output of the
bridge rectifier is then zero as well.
0.5
0.4
--
03
02
!E
V
0.1
~
Figure 7 shows the on-state characteristics of the
depletion-mode bilateral switch. The vertical axis
in Figure 7 is the switch current in A and the horizontal axis is the voltage drop across the switch in V.
One can see that the switch on-resistance is equal to
the reciprocal of the slope of the trace in Figure 7, and
is approximately 7.50. It should be noted that
the on-resistance remains very linear through the
zero-crossover region in the center of the photograph. As the voltage across the switch increases in either direction, the current increases
linearly up to a point but then levels off at a
il
'"~
/
-0.1
/
rn _ 02
- 03
V
I"""" F""
-0.4
- 0.5
-5
-4
-3
-2
-1
0
VOLTAGE DROP ACROSS THE SWITCH (V)
The On-State Transfer Characteristics of the Normally-On
Bilateral Switch Show a Constant RDS(on) Through the Zero
Crossover Point
Figure 7
2-37
-
The switch off-state characteristics are shown in
Figure 8. The vertical axis here is the switch leakage
current in IlA and the horizontal axis is the sustaining
voltage of the switch in V. As can be seen the switch
leaks negligible current until the breakdown voltage
ofthe MOSFETs is exceeded in either direction. In
this case the switch breakdown voltage is ± 180
volts.
as shown in Figure 9. The turn-on time is approximately 50 ILsec, while the tum-offtime is substantially
less than this value (about 10 JLsec). The observed
difference is explained by the fact that during switch
turn-on, a finite amount of time is required for the
effective gate to source capacitance (CGS( eft) to
discharge through the 10 kn resistor. During switch
turn-off, however, more than enough current is available to rapidly charge CGS(eff) to a negative voltage
less than VT.
sao
5 Y/DlV
l ...
ffi
~
w
---
t.oo
LOGIC CONTROL
SIGNAL
1/
-800
- 200
- 100
0
100
SWITCH OUTPUT
I NTO 1 Kfl LOAD
1\
200
SUSTAINING VOLTAGE (V)
2y/DW
The Off-State Transfer Characteristics of the Bilateral Switch
Show That It Can Block AC Voltages Up to 180V Peak
Figure 8
50 IAoS/DIV
The Turn-On and Turn-Off Delay TImes of the
Bilateral Switch Relative to the logic Control Signal
(Top-Trace) Are Not Bllndlngly Fast, but Nonetheless
Adequate for Many Applications
Figure 9
The turn-on and turn-off times for the driver and
switch configuration described are greatly different,
2·38
,
2.9.1 Fabrication of Depletion-Mode
Power MOSFETs
Combining low voltage depletion-mode MOS technology with high voltage MOSPOWER technology
has produced a new type of power device-the depletion-mode power MOS transistor. This new class of
transistors represents a merging of the strengths of
both of its parents. Its announcement is the result of
an I8-month effort to obtain a delicate balance between the need to tum the device off when a low
negative voltage is applied to its gate (i.e., to pinch
off), while having a low on-resistance when no voltage is applied to the gate. The hurdles that were
overcome in the development program may be understood by referring first to Figure 1. This figure shows
the cross section of both a low voltage enhancementmode and a low voltage depletion-mode MOS transistor. The threshold voltage of a MOS transistor is
determined by a number of process parameters. For a
given gate conductor (usually aluminum in a metalgate process and phosphorus-doped polycrystalline
silicon in a silicon-gate process), the threshold voltage is set largely by the dopant profile in the body
region. The low voltage enhancement-mode transistor of Figure I(a) is converted to a depletion-mode
transistor by locally adjusting the body dopant using
an ion-implantation. For the n-channel transistor of
this figure, a phosphorus implant is used.
from a common origin. The channel length L is
reproducibly controlled to values in the 1 to 2 /Lm
range. A small value of L results in both a large
current per unit area and a low value for the
device channel resistance.
The development of a high voltage, n-channel MOSPOWER transistor requires considerably more
process optimization than is shown in Figure 1. The
dopant profile through the double diffused source,
body and drain regions of a vertical DMOS device is
shown in Figure 2. In this structure, the channel
length L is determined by the difference between two
sequential diffusions moving in the same direction
In the DMOS power transistor structure, the body
region is more heavily doped than the n-drain region
as shown in Figure 2. With a reverse bias across this
junction, the resulting depletion region extends further into the drain region than the body region. This
behavior allows high voltages to be placed across the
body-to-drain junction without significantly affecting
the channel length of the DMOS transistor.
SOURCE
GATE
DRAIN
l1li
P·TYPE BODY
a. Enhancement-Mode
SOURCE
GATE
DRAIN
P-TYPE BODY
b. Depletion-Mode
The Cross Section of an NMOS Silicon Gate Transistor
Figure 1
Reprinted with permission from Electronic Design, Vol. 32, No. 13;
copyright Hayden Publishing Co., Inc., 1984.
2-39
diffusion cycles and the specific ion-implantation
dose and energy. These profiles were verified experimentally on existing device geometries and the resulting transistor characteristics were measured.
This information was used as a starting point for
another round of simulations and experiments.
DRIFT REGION
SOURCE
OXIDE
=~~:6CNE
N+
BODY
SUBSTRATE
DOPING
PROFILE
N+ SOURCE
P CHANNEL
N- DRIFT
While the iterations were taking place in the process
area, device layout was started in design. The device
was designed to meet a 150 volt breakdown and a
5 ohm on-resistance specification. The design was
completed as the process experiments were drawing
to a close, so the mask set was run using the final
process sequence. The culmination of this work is
shown in cross section in Figure 3 where an enhancement-mode device is compared to a depletion-mode
transistor. A photograph of the device is shown in
Figure 4. This device was designed for a specific
application, but the process was developed so it
would be applicable to the complete line of power
DMOS transistors. This goal was met, and the
NOSIOO is the first device in a complete family
of depletion-mode DMOS power transistors.
N+ DRAIN
The Doping Profile in a Vertical DMOS Transistor
Figure 2
The first attempt to produce a depletion-mode power
MOS transistor added only an ion-implantation step
to the normal process sequence. Transistors produced in this fashion did exhibit the required depletion-mode characteristics, but suffered from two
problems. First, these transistors could not be turned
off with less than-lO volts from gate to source. In
addition, the breakdown voltage was considerably
less than desired. The use of only an ion-implantation
with a dose sufficiently large to produce a depIetionmode device altered other characteristics in an unacceptable manner. Analysis ofthese results lead to an
understanding of the problem. The lateral nature of
the dopant profile shown in Figure 2 and the vertical
dopant profile resulting from an ion-implantation do
not produce the required body region profile without
considerable optimization. The peak concentration
in the body must be reduced while the remainder of
the body region concentration is impacted in a minimal fashion. This balance was arrived at through an
iterative procedure. Computer simulations were first
used to home in on the body dopant profile. The
results of these simulations were used to devise both
SOURCE &
BODY
~..,.
GATE
...... 1
..."r-
l""
l ~F
P
N-
BODY
J
DRAIN
-------------------------N+
SUBSTRATE
6
DRAIN
The Cross Section of a Depletion-Mode Vertical DMOS Power
Transistor Shows the Conducting Channel with VGS
OV
Figure 3
=
Photograph of the Depletion-Mode Device (NOS100)
Figure 4
240
2.10 Other Structures
There are many alternative solid-state devices that
can handle high voltages, high power, and respectable
switching performance although perhaps not at the
speed of the power MOSFET. Since the early 1950's,
the Darlington pair has offered quasi-MOS-like performance for high gain and reasonably high input
resistance. Their nemesis has been their poor VSAT
and long turn-OFF time. Thyristors, such as the
ubiquitous SCR, have tremendous voltage standoff
and power handling capability, but triggering and
switching time have been their shortcoming. Early in
the 1980's, a concentrated development effort appears
to be rewarding industry with innovative power devices that, in some cases, have often encircled MOS
technology and MOS performance in their designs.
DRAIN/COLLECTOR
-~~
EMITTER
GATE
EMlnER
EPITAXY
SUBSTRATE
n+
DRAIN/COLLECTOR
MOS-Bipolar Structure
Figure 1
One of several useful criteria for judging the effectiveness of solid-state power devices is to compare their
current capability per unit area of silicon. Another
criterion, of course, is simply to compare performance.
in Figure 2. Although they exhibit an impressive
current density, their principal shortcoming is, as with
all composite MOS-bipolar structures, an excessive
storage time or turn OFF delay.
The composite MOS-bipolar transistor structure as
shown in Figure 1, whether the Darlington configuration or one of several alternative configurations, has
found widespread industrial acceptance. Although
often touted as exhibiting ultra-low ON-resistance,
much like the power bipolar transistor, their slow
switching speed and attendant high switching losses
(again, much like the power bipolar transistor) have
severely limited their usefulness above 30 kHz.
SOURCE
GATE
SOURCE
p+
More recent developments have focused attention on
MOS-gated SCRs-variously called the InsulatedGate Transistor (IGT), the High-Conductance MOSFET, or Insulated-Gate Rectifier (IGR)-as shown
COLLECTOR
Insulated-Gate Rectifier (IGR)
Figure 2
2·41
Reference
Ghandi, S. K., Semiconductor Power Devices, New
York, Wiley Interscience, 1977.
Schultz, Warren, "High Current FETs-A New Level
of Performance," Power Conversion International,
March 1984, pp. 43-6.
Smith, Marion, et. aI., "Insulated Gate Transistors
Simplify AC-Motor Speed Control," EDN (February
9,1984).
242
'I
2.11 Power Control with Integrated
CMOS Logic and DMOS 'Output
Abstract
sistors are being fabricated simultaneously with
CMOS control logic to obtain DMOS/CMOS or
D/CMOS power integrated circuits. This paper
begins by quickly reviewing the history of DMOS
or double-diffused discrete MOS transistors. Next,
the development of the circuits presently available
using D/CMOS technology is covered. D/CMOS
circuits may be manufactured using different
process sequences. The characteristics of each of
the D/CMOS IC technologies are compared. Finally, the expected directions of technology and
circuit development are discussed. In this section,
unexploited aspects of D/CMOS devices are
pointed out for specific applications.
This paper examines the evolution and the present
status of a new class ofpower integrated circuitsDMOS/CMOS or D/CMOS ICs. This type of circuit is a natural extension ofthe development ofthe
discrete MOS power transistor that has occurred in
the last few years. However, there are significant
differences between D/CMOS ICs and either conventional MOS or bipolar circuits. The major difference is their ability to withstand and control voltages that are considerably greater than those of
either circuit type. A second difference is their low
power dissipation. Less obvious advantages also
make D/CMOS ICs attractive in other applications. These advantages and their effect on future
D/CMOS circuit trends are also discussed
History of Discrete Power DMOS Transistors
Introduction
A detailed description of the development of contemporary power MOS transistors was given in a
previous paper [11. The flow of discrete MOS power
transistor development is shown in Figure 1. As
shown in this figure, a ten-year span has witnessed
large multiples in both current and voltage rating.
The next few years will see continued advancement
in discrete power MOS transistors, but not the large
increases in current and voltage ratings that have
occurred in the last few years. The on-resistance
characteristics of these devices restrict their use in
most cases to applications requiring less than 1000
volts. Larger chip sizes are possible, but yield considerations make it more attractive to parallel die or
packages to obtain higher currents. The main focus
for discrete technology in the coming years will be
manufacturability and economics. Large reduction
The term "integration" as used in electronics
means the incorporation of more functions in a
physical unit. The trend to integrate more devices
on a piece of silicon is one manifestation of this
trend. In integrated circuit development, progress
within a given technology is usually made in a
very predictable fashion. An example of this type
of progress is the storage capability of the largest
commercially available RAM. Over the last fifteen
years, RAM capability has doubled every one
and one-half to two years from lK in 1969 to 256
K today. Breakthroughs occur when new technology is injected into the mainstream of Ie development, allowing previously unfeasible circuits
to be developed. Such a breakthrough is occurring
today as high current, high voltage DMOS tran(This manuscript was prepared for and presented at Electro 84)
2-43
l1li
Table 1
A Comparison of Power MOS and Bipolar Transistor
Electrical Characteristics [11
in price will lead to greater device acceptance. The
economics of scale resulting from increases in MOS
power transistor use will enable them to compete
more effectively with bipolar transistors.
Paramete,
Input Impedence
The most significant development in discrete D MO S
devices in the last two years is the announcement of
MOS-gated regenerative (Le., latching) and conductivity modulated devices. These two similar types
of devices promise to significantly enhance the CUfrent ratings of D/CMOS ICs.
Current/Voltage
1970-
High (11)5.101,
Intormedl... (,0'.,02)
SwHchlng Frequency
HIgII(I_kHz'
Intermedl... (20-10 kHz)
On Re.I_
High
Low
OIfRHI_.
High
High
1..-1... (500 V)
High (1200 V)
~bliity
Rugged .....
Excellent
Good
Coot
High
Intormedlolw
M..lmum Opoming
High (2OO"C)
Intermedlolw (ISO'C)
Temper.tu,.
- - Metal Gate Lateral DMOS TranSlStor
1975 ...... Metal Gate V-groove Transistor
30V/SO rnA
circuits. The ease with whichDMOS transistors may
be driven when compared to bipolar transistors is one
obvious advantage. Ruggedness, particularly the
absence of secondary breakdown in DMOS transistors, is another advantage. However, the voltage
drop per unit area for transistors with equivalent
voltage ratings is considerably greater for DMOS
transistors. This disadvantage dictated that DMOS
transistors be utilized first in ICs with voltage ratings
beyond those ratings of bipolar transistors. The first
application of D/CMOS technology was for high
voltage display drivers. The development of these
circuits took place in the 1977/78 time frame.
60V/1 Amp
...... Silicon Gate V-groove Transistor
100V/S Amp
- ... Metal Gate Vertical DMOS Transistor
400V/IOAmp
1980 ...... Silicon Gate Vertical DMOS Transistor
1985
BIpolar Performance
High (lot-lOll ohm.) Intennedl... (1CJ3.105 ohm.,
Cumonl GoIn
VoHogo
Maximum Ratings
TECHNOLOGY
MOS Performance
SOOV/20Amp
...... High Voltage/High Current Technologies
10OOV/SO Amp
_ -Regenerative & Conductivity Modulated
10OOV/IOO Amp
Two classes of displays-plasma and electroluminescent panels-required voltages from 80 to 200 volts
for successful operation. Line density and allowable
cost per driver also dictated that as many outputs as
possible (in multiples of8) be fabricated on one chip.
Different drive schemes were developed to address
and solve the problems presented by each panel type.
This work resulted in the three different D/CMOS
technologies which are the starting point for contemporary D/CMOS ICs. These three D/CMOS IC
technologies are discussed in the next section.
Devices
A Chronology of Discrete DMOS Power Transistor
Development
Figure 1
The advent of MOS power transistors has lead to
numerous comparisons between them and bipolar
transistors. At first, the goal of the comparison was to
show that one type of device was superior to the other
for a given application. One such comparison is contained in Table 1. A more enlightened approach is to
examine the characteristics of each type oftransistor,
and then to make a selection based on the resulting
match. Each transistor type has a range of applications which match the device requirements.
Self Isolated D/CMOS IC Technology
In one driver configuration, the output drive transistors were required to withstand 200 volts in an open
drain configuration. The drive circuitry was required
to dissipate minimum power, so CMOS logic was an
appropriate choice. These two requirements lead to
the development of "self isolated" D/CMOS technology as shown in Figure 2. The CMOS logic in this
technology operates with a power supply up to 15
volts, while the DMOS output transistors are capable of 400 volts. Using silicon-gate technology
with a 4 JLm feature size, the clock frequency of
the logic is 5 MHz.
The Development of ICs with Low
Voltage CMOS Control and High Current,
High Voltage DMOS Outputs
The comparison of characteristics presented in Table
1 compels IC designers to consider the benefits derived from the incorporation of ICs in monolithic
2-44
/
/
/
StB
5,8
G
G
G
p- Substrate
Low Voltage n-channel
High Voltage n-channellateral
CMOS FEY
Enhancement and DepletIon Mode FEYs
Cross-Section of a Self-Isolated D/CMOS I.C.
Figure 2
Junction Isolated D/CMOS IC Technology
Junction Isolated D/CMOS Technology
with Complementary Outputs
Other displays required lower voltages (60 to 120
volts) with totem-pole, n-channel outputs. These requirements were met usingjunction isolation to stand
off the voltage between adjacent devices. DMOS
transistors, because of their configuration, have a
higher breakdown voltage than simultaneously fabricated bipolar transistors. A cross section of this
technology is shown in Figure 3. Because ofits added
complexity, junction isolated technology is capable
of providing a variety of additional components. The
most useful of these components is the npn transistor,
though pnp transistors and a variety of diodes and
resistors are also available.
8G
SG
D
8
A third D/CMOS configuration was required when
high voltage push-pull outputs were needed. Junction
isolation was essential but the process sequence had
to be modified to accommodate the high voltage pchannel device. Size constraints made it advantageous to use DMOS technology for the p-channel
device. This combination of constraints led to the
junction isolated complementflry output structure
shown in Figure 4. Both the junction isolation and the
DMOS output transistors are capable of withstanding 120 to 150 volts. The low voltage CMOS logic is
designed to operate in the 5 to 15 volt range.
E
BG
S G D
BG
5
G
D
~---D~M~O~S--~~----~NP~N--~!L!----~NM~O~5------~L--~P~MO=S~~
Cross-Section of a Junction Isolated D/CMOS I.C. [1]
Figure 3
n-Substrate
High Voltage n-channel
DMOS FEY
High Voltage p-channel
DMOS FEY
Low Voltage "-channel
Enhancement and Depletion Mode FEY.
High Voltage CMOS Process
Cross-Section of a Junction Isolated, Complementary Output
D/CMOS I.C. [2]
Figure 4
2-45
-
-.'30
These three D/CMOS technologies are compared in
Table 2 with respect to fabrication sequence and
electrical performance. Table 2 contains the output
configurations possible with each of the three major
technologies. This table also provides the paralleled
resistance and the current capability of DMOS output transistors.
introduced and represents only an additional, well
characterized process sequence. The growth of quality gate dielectrics is also well known. The gate conductor most often chosen is either aluminum or doped
polycrystalline silicon. Aluminum as the interconnect metallization offers process simplification but
results in larger device sizes. The use of poly silicon
as the gate conductor adds a process step to the
fabrication sequence, but it also allows another level
of process interconnect.
Status of Circuits Containing CMOS
Logic and DMOS Output Transistors
Circuits fabricated with each of these three
D/CMOS technologies are becoming commercially
available. Considering the design flexibility and
economics of IC manufacture, the second technology - junction isolated D/CMOS - is likely
to become the most popular industrial choice. This
technology has a number of advantages for both
IC manufacturers and users.
Device Characteristics Similar to
Those of Other Technologies
The similarity between the bipolar and the junction
isolated IC fabrication leads to a second significant
advantage. Many of the available circuit devices
have characteristics that are similar to those of bipolar technology. The circuit designer has to become
familiar with the characteristics of only a few new
devices as opposed to unlearning and relearning a
whole set of new characteristics.
Similarity to Existing Bipolar Process
Technology
The process cross section shown in Figure 3 differs
only in a few ways from the cross-section of a bipolar
IC. The most significant of these differences are the
need to include a deep, lightly doped region called a
p-well for the n-channel transistors, and the requirement for a gate dielectric (usuallv thermallv grown
Si02), and a gate conductor. The p-well is easily
Circuit Design Techniques Are Familiar
Design techniques using the grounded substrate configuration of the junction-isolated D/CMOS technology are also well known. A designer is not faced
with learning a set of new design techniques. Only an
extension of existing knowledge is required.
Table 2
Comparison of the Features of the D/CMOS Processes
D/CMOS TECHNOLOGY
SELF ISOLATED
JUNCTION ISOLATED
JUNCTION ISOLATED WITH
COMPLEMENTARY OUTPUTS
Epi Required
No
Yes
Yes
Masking Steps
9 - II
10 - 12
II - 13
Logic Voltage
5 - 15V
5 - 15V
5 - 15V
Clock Rate
5 MHz
5 MHz
5 MHz
200- 400V
150V
ISOV
CHARACTERISTIC
FABRICATION
Output Voltage
Output Configuration
Single Ended
(open collector)
Single Ended,
Totem Pole
Single Ended,
Totem Poie. Push-Pull
ELECTRICAL
Parallel
Resistance of
Output Devices
(200 x 200 mil chip)
3-5n
Total On-Chip
Current Capability
2-4 amps
i- 2 n
10-20 amps
2-46
2- 4
n
5-10 amps
Circuits Are Similar to Other Device Families
The negative ground configuration with CMOS compatible inputs is common in the electronics industry.
Little, if any, systems modification is necessary to
begin taking advantage of circuits fabricated with the
junction isolated D/CMOS technology.
For these reasons, the junction isolated D/CMOS
technology is likely to gain industry-wide acceptance
more rapidly. An example of a circuit fabricated
using this technology is shown in Figure 5. This chip
contains analog and digital control functions implemented using silicon-gate CMOS technology and
two large output drive transistors in the totem-pole
configuration. The input circuitry is TTL compatible
with the low voltage section operating from a 30 volt
supply. Level shift is accomplished using high voltage lateral DMOS transistors. The two large DMOS
transistors are rated at 80 volts, and the large transistor supplies 12 amps while the small one supplies 8
amps. Resistors are placed in series with the two
output transistors for sensing and controlling the current in each transistor. The chip size is 195 mils
x 335 mils (4.9mm x 8.4mm). This junction
isolated IC represents the state-of-the-art in power
D/CMOS circuits.
Photograph of a Driver Circuit Fabricated Using the Junction
Isolated D/CMOS Technology
Figure 5
environments with considerable electrical noise. The
high currents through the transistors coupled with the
presence of various four layer (and hence potentially
latching) structures require care in both circuit layout
and circuit use. The sensitivity of the IC to latch-up
can be minimized through proper circuit design, but it
cannot be eliminated in conventional junction isolation. Use of dielectric isolation instead of junction
isolation as shown in Figure 6 will eliminate latch-up,
but this technology adds considerable cost to the
circuit The dielectric isolation technique may be
applied to any of the three D/CMOS technologies,
but it is probably more applicable to the two junction
isolated versions.
Directions for Future Development in
D/CMOS Integrated Circuits
The initial enthusiasm resulting from the promise of
D/CMOS technology has tended to obscure two
things-problems that need addressing before greater acceptance is earned, and growth directions to
fully take advantage ofthe promise of this new technology. Concerns in both of these areas are essential if
D/CMOS technology is to realize its full potential.
These issues are discussed in this section.
A major potential concern for D/CMOS circuit designers and circuit users is circuit performance in
DRAIN
SOURCE
GATE
.
-.::'.
.. '
.:~:. :):/:.:.:,.:
. : ""J
::::
"
:";":',
Cross-Section of a Dielectrically Isolated D/CMOS I.C. [1]
Figure 6
2-47
-
l1li
Careful circuit use is as important as the design of the
circuit in systems with large currents flowing. For
this reason, it may be necessary to have more than
one ground on the chip. A logic ground should be
capable of providing stable current sinking. The presence of a separate high current ground for the output
devices means that some ground "bounce" may be
allowable without interfering with the operation of
the low voltage section of the chip.
of these device types promise to control more power
per square mil of silicon surface than DMOS transistors. The conductivity enhanced device is gate
controlled during both tum-on and tum-off, while the
voltage across the regenerative device must go to zero
before conduction stops. It is possible to integrate
both of these structures in D/CMOS ICs. However,
this integration cannot proceed without some caution.
The presence of excess minority carriers in the drain
region may increase circuit sensitivity to latch-Up.
Considerable care must be taken to guarantee latchup free operation.
One major hurdle facing high power D/CMOS IC
manufacturers and users is the availability of packages, heat sinks, etc., that allow circuits to operate at
or near their intrinsic limits. The last few years have
seen continued progress in the high power package
area in both plastic and hermetic packages. However, this area is key for future improvements in
circuit performance and reductions in circuit costs.
Summary and Conclusions
This article has examined the evolution, present
status, and future directions of D/CMOS IC technology. Three different specific technologies were
discussed, and their features were compared. Based
on this comparison, one D/CMOS technology, the
junction isolated D/CMOS, was selected as the
technology most likely to become an industry standard. One circuit that pushes junction isolated
D/CMOS technology to its limits was examined as
an example of what is possible. Finally, future directions for D/CMOS IC circuits were indicated by
discussing both areas of concern and advantages of
this technology. The theme running through this
paper is the dynamic nature of this field. The promise
of D/CMOS technology is being realized as both
users and manufacturers begin to understand its
limits.
The DMOS transistor's ability to operate at both
high and low temperature extremes offers significant
advantages for many applications. The package
power dissipation limits discussed in the previous
paragraph set the ultimate limit, but both the logic
and the output stage of a D/CMOS are capable of
operating at 150°C and above.
The positive temperature coefficient of resistance of
DMOS transistors has both advantages and disadvantages. The output device resistance does increase
with temperature resulting in increased power dissipation and decreased efficiency, but DMOS transistors are capable of stable operation at temperatures as high as 200°C and above. This behavior is in
contrast with that of bipolar transistors which have a
negative temperature coefficient. The low temperature performance of D/CMOS circuits also offers
significant advantages over bipolar circuits. The
speed and current handling capability of both the low
voltage CMOS transistors and the output DMOS
transistors increases as the temperature decreases
from room temperature down to 77°K (liquid
nitrogen) [3]. The dual advantages at lowered temperatures are enhanced device performance and
increased power dissipation capability.
References
Another area in which D/CMOS circuits may offer
advantages is in radiation environments. The major
effect of ionizing radiation on MOS transistors is to
change the threshold voltage due to charge generation and trapping in the gate dielectric. Techniques
for "hardening" conventional MOS transistors apply
as well to DMOS transistors. It should be possible to
obtain D/CMOS integrated circuits capable of withstanding a total dose of 105 to 106 (rads silicon).
The announcement of discrete DMOS-gated conductivity enhanced [4] and regenerative [5] structures foreshadows future circuit developments. Both
2-48
[1]
RA Blanchard, "MOSFETS IN ARRAYS
AND INTEGRATED CIRCUITS," Proceedings ofElectro 83 Paper 7/4 (1983).
[2]
RA Blanchard and W.G. Numann, "A High
Voltage Chip Set for Use as Electroluminescent
Panel Drivers," Proceedings of the Society of
Information Display, April, 1982.
[3]
RA Blanchard and R Severns, "Designing
Switched-Mode Power Converters for Very
Low Temperature Operation," Proceedings of
Powercon 10, D-2 (1983).
[4]
J.P. Russell, AM. Goodman, L.A Goodman,
and J.M. Neilson, "The COMFET - A New
High Conductance MOS-Gated Device,"
IEEE Electron Devices Letters, P63, EDL-4
No.3 (March, 1983).
[5]
A Pshaenich, "The MOS SCR, A New
Thyristor Technology," Motorola Engineering
Bulletin EB-103 (1982).
2.12 Drain to Source Breakdown and Leakage in
Power MOSFETs
An examination of almost any power MOSFET data
sheet reveals values for both the drain to source
breakdown voltage, BVDSS, and leakage current,
IDSS, with zero applied gate-to-source voltage. At a
cursory glance, these two parameters might seem to
be unrelated until one looks at the test conditions for
each. The breakdown voltage is measured at a specified leakage current and leakage current is measured
at a specified voltage, often a percentage of the
breakdown voltage. Specifying these two parameters
separately on a data sheet is accepted practice (by
many manufacturers), but is somewhat redundant.
•
)r-i--r-i--t-i-i+-i--t-i--r-1~
5OY!dlv
A statement relating breakdown voltage and leakage
current would certainly help to clarify matters. In the
ideal case, the BVDSS of a MOSFET is reached when
the leakage current begins to increase drastically with
a small change in applied voltage. In a properly
designed and manufactured power MOSFET this
effect is as shown in Figure 1. Beyond a certain
voltage, the drain current suddenly begins to increase
uncontrollably if not limited.
Normal Breakdown Occurs in an n.Channel DMOS Power
FET When the Applied Drain to Source Voltage Exceeds
the Device's BVDSS
Figure 1
Not all power MOSFETs show such predictable behavior as that illustrated in Figure 1. An unusual
effect known as "latch-back" has also been observed
in certain types of MOSFETs [1] (notably metal gate
V-groove devices), and is shown in Figure 2. Here
normal breakdown begins as BVDSS is reached, but
as the leakage current rises towards a critical value,
the sustaining voltage of the device suddenly drops or
"latches-back" to a lower value. This latched-back
breakdown voltage can be as much as 80% less than
BVDSS·
\
taVldlV
Latchback Occurs in a Power MOSFET When the Parasitic
Bipolar Transistor Begins to Conduct Collector Current
Figure 2
2-49
-
,
Both the normal and latched-back modes of breakdown (Figures 1 and 2 respectively) are non-destructive if the increased leakage current is limited to a
safe value. This prevents excessive power dissipation
in the MOSFET, which could force the junction
temperature above the maximum allowable limit and
cause device destruction.
n+
keeping the heavily doped nand p regions sufficiently far apart that there is no interaction between them.
5. Dielectric breakdown; here the electric field in a
dielectric layer such as silicon dioxide (Si02) or
silicon nitride (Si3N4) exceeds the dielectric strength
of the material. This field causes very large currents to flow through the dielectric and can permanently alter its insulating properties. Commercially available power MOSFETs use sufficiently
thick dielectric layers such that the electric fields
within them never reach the critical value necessary to initiate breakdown.
Since both zener and dielectric breakdown have been
eliminated as potential causes of normal breakdown
in power MOSFETs, we will examine the first three
in more detail. Following this section, the phenomenon of latch-back and how modern power MOSFETs
are designed to avoid it is discussed. Finally the
temperature dependence of breakdown voltage and
leakage current is examined.
SUBSTRATE
This Cross Section of an n-Channel VDMOS Power FET
Shows How the Polysilicon Gate is Located Relative to
the Source and Body Diffusions
Figure 3
Causes of Normal Breakdown in Power MOSFETs
1. Avalanche breakdown [2]
This phenomenon, as its name implies, is a sudden
avalanche of mobile carriers caused by the increased
electric field present within the depletion regions at
the body drain pn junction. Electrons or holes that
enter the depletion regions acquire sufficient energy
from the electric field to knock bound valence electrons out of the silicon lattice atoms in that region. If
one electron or hole produces on the average less than
one additional carrier then the leakage current is not
increased.
There are five different phenomena that can cause
normal breakdown to occur in semiconductor diodes
and transistors under excessive applied voltage. Using
Figure 3 as a guide, these five effects are [1,2]:
1. Avalanche breakdown; this breakdown is the dominant effect in commercially available power MOSFETs, and occurs when the electric field in the
vicinity of the body-drain pn junction increases to
a critical value.
If, however, one or more additional carriers are
produced and these extra carriers each produce one
or more additional carriers then avalanching ensues.
The depletion regions around the body-drain pn
junction must be wide enough so that the mobile
carriers can gain sufficient energy from the local
electric field to initiate this process.
2. Reach-through breakdown; this breakdown is a
special case of avalanche breakdown which is
caused by the depletion region in the n - type
epitaxial layer reaching the heavily doped n +
substrate.
3. Punch through breakdown; this breakdown occurs
when the depletion region of the reverse biased
body to drain junction reaches the heavily doped
n + source diffusion.
The mathematics for describing this effect are already
known from the analysis of gaseous breakdown phenomena. When this analysis is applied to the semiconductor case, the carrier multiplication factor can be
derived. Basically it is [2]
4. Zener breakdown; a high electric field (on the
order of 106 Vlcm) initiates this type of breakdown. There is a finite probability that electrons
in their covalent bonds will be excited directly into
the conduction energy band at the junction. These
high electric fields can only be reached in heavily
doped p + In + junctions, with resulting breakdown voltages < 6 volts. This type of breakdown
is avoided in commercial power MOSFETs by
(1)
where VB is the junction breakdown voltage, VR is
the applied reverse voltage and N is a numerical factor
which depends upon the type of semiconductor crystal
2·50
I
used (N = 3 to 6 for silicon). As VR approaches VB,
M approaches infinity and the junction breaks down.
is a major factor in the design of an optimized power
MOSFET. If the epitaxial layer is too thick, the device
will have an excessively high on-resistance for its
breakdown voltage specification. Conversely, if the
epitaxial layer is too thin, reach-through breakdown
occurs below the desired breakdown voltage rating.
Avalanche breakdown occurs when the maximum
electric field in the vicinity of the junction reaches a
critical value. Since the critical field in silicon is about
3 x 105 V/cm for impurity concentrations less than
10 16 atoms/cm3 [1], the breakdown voltage of a planar
diffused pn junction is determined to a first order by
the doping concentration on the lightly doped side [2].
This result follows from the derivation of the breakdown voltage for the one-sided step-junction (OSSJ)
model. In this model, the acceptor doping concentration on the p - type side is much greater than the
donor doping concentration on the n - type side. To
a good approximation, the breakdown voltage of a
power MOSFET is determined by the impurity concentration in the lightly doped epitaxial layer.
2. Reach-through breakdown
This mode of breakdown occurs when the expanding
depletion region in the n - type epitaxial layer reaches
the heavily doped n + substrate. The spreading of the
junction depletion region is inversely proportional to
the square root of the impurity doping concentration [1]. As the depletion region enters the substrate,
it begins to spread much more slowly, and the electric
field rises rapidly towards the critical value necessary
to initiate avalanching. When it reaches this critical
value, localized avalanching occurs, and the device
breaks down.
A 100 volt power MOSFET requires a background
(epitaxial layer) concentration of 3.5 to 6 x 10 15
atoms/cm3 , while a 500 volt MOSFET requires a
lower concentration in the region of 1 x 10 14
atoms/cm3[3]. Since the resistivity of the epitaxial
layer is inversely proportional to the dopant concentration, a high voltage MOSFET has a higher onresistance than a low voltage counterpart with the
same die size. In almost all available power MOSFETs,
the on-resistance is approximately proportional to the
breakdown voltage raised to the 2.5 power [4]. This
relationship is shown in Figure 4.
1
Commercially available power MOSFETs are designed by first choosing the background concentration
to give the required avalanche breakdown voltage.
The epitaxial layer thickness is then chosen so that at
the avalanche breakdown voltage, the depletion region in the epitaxial layer just reaches the substrate.
This epi layer selection process minimizes the device
on-resistance for a given breakdown voltage and die
size.
3. Punch-through breakdown
This effect occurs when the widening depletion region
at the body-drain junction reaches the heavily doped
n + source diffusion. The field in the body depletion
region sweeps electrons from the source across the
reverse biased body-drain junction. The leakage current increases rapidly as a consequence.
1000
i
I
iI:
~
This breakdown mode is rarely observed in commercial power MOSFETs. They are normally designed to
be avalanche breakdown limited.
100
~
iii0:
W
iii
~
o
,.
The Cause of Latchback and Its Avoidance [1]
In power DMOS structures, the source-body contact
is made some distance from the channel region. This
contact shorts the emitter and base of the parasitic
npn transistor (pnp for a p-channel MOSFET), formed
by the source, body and drain regions. Figures Sa, b
and c show this for the VMOS, LDMOS, and VDMOS
structures, respectively.
~-~.~~~~~~~~,~.~~~~~,~~
BREAKDOWN VOLTAGE, BVDSS M
The Observed Relationship Between R'DS(on) Multiplied by
Chip Size (R'DS(on» and Breakdown Voltage for DMOS
Power FETs, Shows That BVDSS is Proportional to
R'DS(on) Raised to the 2.5 Power
Figure 4
The source and body diffusions form a distributed
series of npn bipolar transistors with their bases
separated by the resistance of the body region below
the source. To explain the phenomenon oflatch-back,
a simplified model of the DMOS structure is used.
This model, shown in Figure 6, consists of the gate
It is important to ensure that the epitaxial layer is no
thicker than necessary, if the MOSFET is to have
minimum on-resistance. The epitaxial layer thickness
2·51
-
SOURCE
•
BODY
GATE
-------------------------\
n+
a) VMOS Structure
DRAIN
SOURCE
•
BODY
GATE
DRAIN
b) LDMOS Structure
SOURCE
•
GATE
BODY
c) VDMOS Structure
DRAIN
Cross-Section and Circuit Models
This Diagram Shows Distributed Parasitic npn Bipolar
Transistor Inherent in the Three Common Power DMOS Structures
Figure 5
2-52
DRAIN
r--'VV'.,---£.---A/VI"---fE:--~---+!
GATE
1---0
IT
-----
BVCEO
BVCES
DRAIN·TQ.SOURCE VOLTAGE (V)
SOURCE
•
This Curve Shows Idealized Latchback Occurring in a DMOS
Power FET Where BVCEO is 50% of BVCES
Figure 7
BODY
The Simplified n-Channel DMOS Model, Used to Explain the
Phenomenon of Latchback, Includes Only the Parasitic
Bipolar Transistors at the Extreme Edges
of the Body Diffusion
Figure 6
curve, where the BVCEO is 50% of the BVCES or
BVOSS. The trigger current IT is the value of avalanche current at which the transition to BVCEO
occurs. Not surprisingly, this current can be shown to
be [1]
controlled channel and only the two npn transistors at
the extremes of the body diffusion [1]. The bases of
these two npn transistors are separated by a resistance
Rp. The body contact resistance is modeled by a
resistance RB. The body-drain diode is also included,
and serves as the avalanche current generator in the
model. This model is equally applicable to all three
power MOSFETstructures mentioned above, despite
their cross-sectional differences.
(2)
where VBE (Q2) is roughly 0.6 volts.
Now that the mechanism which causes latchback in a
power MOSFET has been described, at least two
methods for preventing it, based on the model shown
in Figure 6, can be listed. These are [1]:
With the gate short-circuited to the source, the
leakage current through 01 increases as the applied
VOS reaches its avalanche breakdown voltage. This
current causes the voltage drop across resistors Rp
and RB to increase. If this drop is allowed to reach
0.6 volts, 02 will begin to conduct current. When this
happens, the sustaining voltage of the MOSFET
suddenly drops from the avalanche breakdown voltage of diode 01 to a lower value. This lower sustaining
voltage is equal to the collector-emitter breakdown
voltage of 02 under forward bias conditions.
1. Reduce the value of Rp + RB so that any lateral
body-current flow does not result in the 0.6 volt
drop necessary to tum on 02.
2. Change the shape of the body region such that the
current flow in avalanche breakdown bypasses Rp
(Rp is usually> > than RB).
Most of the modem commercially available OMOS
power FETs use technique #2 to prevent latch-back.
Figure 8 shows how this can be accomplished in both
vertical OMOS and V-groove type devices. The
heavily doped p + regions that extend more deeply
into the epitaxial layer form an integrated bypass
diode that effectively diverts avalanche current away
from the p --: type body region (and hence Rp). The
equivalent model for this modified structure is shown
in Figure 9. It has been experimentally verified that if
the avalanche breakdown voltage of 02 is less than
that of 01, the circuit will not suffer from latch-back.
Most bipolar transistor manufacturers do not usually
specify the collector-emitter breakdown voltage under
forward bias conditions. They specify two other values of collector-emitter breakdown voltage. These
are the BVCEO (with the base left open circuit), and
the BVCES (with the base short-circuited to the
emitter). The BVCES of transistor 02 is equal to the
avalanche breakdown voltage of diode 01, which sets
the BVOSS limit of the power MOSFET.
The BVCEO of 02 approximately equals its collectoremitter breakdown voltage under forward bias conditions. Most bipolar transistors, including the parasitic device in a power MOSFET, exhibit a BVCEO
that can be as low as 20% of their BVCES. Thus the
latched-back sustaining voltage of a power MOSFET
can be considerably lower than its BVOSS. Figure 7
shows an idealized latched-back breakdown voltage
The Temperature Dependence of Breakdown
Voltage and Leakage Current
As mentioned previously, the breakdown voltage of a
power MOSFET is determined to a first order by the
doping concentration in the epitaxial layer. Since this
2·53
-
SOURCE
•
BODY
GATE
p+
(a)
n+
DRAIN
SOURCE
•
BODY
GATE
(b)
n+
DRAIN
A Practical Device Modification to Prevent Latchback Is the
Inclusion of an Integrated Bypass Diode, 02, as Shown In
the Modified VDMOS and VMOS Structures Respectively
Figure 8
2-54
"
DRAIN
D,
~________~____~~____~____~__~~E
RB
SOURCE & BODY
The Circuit Model of CMOS Power FET with an Integrated
Bypass Diode Is Very Similar to That in Figure 6,
but Now Includes Diode D2
Figure 9
concentration does not vary substantially with junction temperature, one would expect the device breakdown voltage not to vary with temperature either.
which works out to be approximately + 8% per °C, or
a doubling of leakage current for every 12°C rise in
junction temperature.
In practice however, there is a slight variation in
breakdown voltage with temperature (about a 5%
increase per 100°C rise in junction temperature). This
is due to a second order effect known as "phononassisted electron scattering." At elevated junction
temperatures, the increased vibrational energy of the
silicon lattice atoms causes the mobile charge carriers
in the epitaxial layer to move in an increasingly
random pattern, relative to the direction of the
electric field.
The body-drain junction leakage in a power MOSFET at a junction temperature of 150°C can therefore
be more than three orders of magnitude greater than
that at room temperature. This can limit the usefulness of a power MOSFET that exhibits excessive
junction leakage current (say several hundred /-LA) at
room temperatures, when operated at temperatures
near its maximum data sheet limit (typically 150°C).
Frequently, however, transistors with high leakage
currents at 25°C do not have the same temperature
coefficient as one with low leakage.
Thus, the mobile carriers do not attain as high an
average kinetic energy for a given electric field.
Maximum kinetic energy is attained when the carriers
move parallel to the field. Consequently, the value of
the critical field required to initiate avalanching increases with temperature, and so does the device
breakdown voltage.
There are additional sources of leakage in a power _
MOSFET originating from the package in which the
die is mounted. In a properly packaged device, these
leakages are of the order of a few nA and do not
constitute a significant part of the total leakage. In a
device which has been defectively packaged, however,
the additional leakage can reach several hundred nA
and mayor may not be temperature sensitive. This
situation is rarely encountered in properly manufactured and tested devices.
The drain-to-source leakage through the body-drain
junction, on the other hand, is highly temperature
sensitive. The leakage current in a reverse biased pn
junction is mainly proportional to the square of the
intrinsic (i.e. undoped) carrier concentration in silicon (Ni). The value of Ni2 can be written as [2]:
References
[1] R.A. Blanchard, Optimization of Discrete High
Power MOS Transistors, PhD. Thesis Dissertation, Department of Electrical Engineering at
Stanford University, December 1981.
(3)
where K is a constant independent of temperature,
Eg is the energy band-gap of silicon (1.12 eV) and K is
Boltzmann's constant. One can see that the T3 term
completely dominates the leakage current over a wide
temperature range. The approximate fractional change
in the leakage current of a reverse biased silicon pn
junction is thus given by [2]
[3] S.M. Sze and G. Gibbons, "Solid State Electronics," September 1966.
=1
- (3- + ~) \VR = Constant
IL1(aIL)
aT
2 T Kl""
[4] Chenming Hu, "A Parametric Study of Power
MOSFETs," PESC 1979 Conference Record.
-- -
rrn
[2] David M. Navon, Electronic Materials and Devices, Houghton Mifflin Company, 1975.
(4)
2·55
2.13 Modeling
2.13.1 SPICE 2: Simulation of High Power MOSFET
Introduction
The power MOSFET offers superior performance
due to its high switching speed, low power voltagedriven gate requirement, ease of paralleling, and
absence of secondary breakdown phenomenon that
troubles the Bipolar Junction Transistors. Herren Jr.,
Nienhaus and Bowers developed a computer model
for high power MOSFET[l]. This model takes into
account the deviation ofthe device transfer characteristic (Figure 4c) from the ideal square law characteristic by the inclusion of a lumped source resistance.
One of the chief attractive features of the model is
that its parameter determination procedure requires
only the manufacturer's published device data.
constant), and KD (the channel length modulation
constant). In the SCEPTRE version of the model,
CGD and CDS are non-linear capacitors. However,
CGD is modeled as a constant capacitor in the SPICE
2 version ofthe model[l]. CDS is given by:
In the following article, the computer model is briefly
described, and the model parameter determination
procedure for the SPICE 2 program is explained. A
computer model of the device type IRF330 was
generated based on this procedure. Simulation results
of both resistive and inductive switching circuits using
this device are presented.
In the SPICE 2 version ofthe model values of = 1.0
and n = Y2 have been assumed. Thus, the model
parameters to be determined are 1) VT. 2) (3. 3) KD,
4) RS. 5) RD. 6) RG. 7) CDO. 8) CDG. 9) CGS.
CDO
CDS
(1 + v~sr' where
=
CDO = capacitance value at zero VDS
= junction contact potential = PB
n
= an exponent between 'h and Y2
RO
SPICE 2[2] is a network analysis program using nodal
analysis techniques and has the advantages of programming ease and relatively fast execution time.
Version 2E2 was used in the following programs.
VGO
~~.
a
Ra
Jo
tas
Vas
+
High Power MOSFET Model[l]
_
Figure 1 shows the high power MOSFET model for
an n-channel MOSFET. A similar model exists for
the p-channel MOSFET device. The substrate and
the source are assumed to be shorted internally. JD is
a non-linear current source depending on VGS, VGD,
VT (the threshold voltage), [3 (the device conductance
CDS
Rs
High Power MOSFET Model (Ref. 1)
Figure 1
2-56
Figure Z shows the SPICE Z built-in level-l n-channel
MOSFET model[Z] based on the model proposed by
Schichman and Hodges. This model is adapted to the
requirements of the high power MOSFET model as
indicated in the next section. n-channel and p-channel
devices are taken care of by using different built in
SPICE Z models, viz., NMOS and PMOS. For
n-channel MOSFETs VTO, the gate threshold voltage is considered positive for enhancement mode
devices and negative for depletion mode devices. For
p-channel MOSFETs, VTO is negative and positive
for enhancement mode devices and depletion mode
devices, respectively.
-VDS
Output Characteristics - To Locate P1 Thru P6
Figure 3
RD
CGD
e) P6, a point on the saturation (ohmic
region) line. It may be more convenient
to select this point from the saturation
(low VDS) characteristics provided by
the manufacturer.
CBD
DBD
G
B
10
STEP 3: Choose VTl, the approximate intersecting
point of the transfer characteristic with the
I D = 0 axis (Figure 4c), as the starting value
ofVT·
DBS
CGS
CBS
CGB
RS
80 ,",I PULSE TEST
410v
VG~=J5V
SPICE 2 Built-In MOSFET Model (Ref. 2)
Figure 2
Determination of Model Parameters[l]
~=
50V= i=='
Following is a step by step procedure to determine
the model parameters based on Reference 1.
F=
J
'5V
~
STEP 1: Figure 3 shows a set of output (dc) characteristics drawn in an exaggerated manner.
Choose three characteristic curves Ll, LZ,
and L3 corresponding to gate-to-source voltages VaSl, Vasz, and VaS3, respectively,
according to the following guidelines.
a) Ll-curve with the highest drain current
b) L3-curve close to I D = 0 line
c) LZ-curve approximately half-way between the above two curves.
1
4j"=
50
100
150
200
1= F=
250
300
Yos. DRAlN·yo-SOURCE VOLTAGE (VOLTS)
a) Typical Output Characteristics
5
.1 . .1
.l'';~v.t--
aD ,:£8 PULSE TEST
JV
4
STEPZ: Choose points PI (IDl, VDSl), PZ (IDZ,
VDSZ), ... P6 (lD6, VDS6) as per the following guidelines.
a) PI, any point well into the pinch-off
region on line L1.
b) PZ, on line LZ such that VDSZ = VDSl(VaSl-VaSZ)
c) P3, any point well into the pinch-off
region on line L3.
d) P4 and P5, two points as far apart as
possible on the near horizontal straightline segment of Ll curve.
VGS=SOY
W
3
I
I
JI
2
1
I
J
45V- ~
40V_
II
-
2
4
6
8
Yos. DRAIN-YQ-SOURCE VOLTAGE (VOLTS)
b) Typical Saturation Characteristics
IRF330 Characteristics (Ref. 3)
Figure 4
2·57
10
-
2000
Jas =
80 #,S PULSE TEST
VDS = 10DV
' .. 1 MHz
1600
TJ "" +125"C
TJ
= 25"C ................
TJ = -55"C'-...
'NJ
I\\'
_\
1\
... ,J...
rH.
Clas
rN
/'/
..a
'\
t--
1
•
•
Vas. GATE-TO-SOURCE VOLTAGE (VOLTS)
~10
H
H
~
9
VDSo DRAIN TO SOURCE VOLTAGE (VOLTS)
c) Typlcal1l'ansfer Characteristics
STEP 4: Rs
d) Typical Capacitance Vs. Draln-ta-Source Voltage
IRF330 Characteristics (Ref. 3)
Figure 4
( Im)Vz(VOS2 - VT) - (VOSl - VT)
= ....>..I..:::D'-"'2"--_ _ _ _ _ _ _ _ __
(1m ID2)Vz- 1m
For the SPICE 2 model, RS = Rs.
STEPS: Let
a
X
x (VOSl - ID4 * Rs - VT)
y = (VOSl - IDS * Rs - VT)
KD = ID4 y2 (VDSS - V OSl + VT) - IDS x2 (VDS4 - V OSl + VT)
IDS x 2 - ID4 y2
For the SPICE 2 model, LAMBDA
=
l/KD.
If the characteristic curves are horizontal in
the pinch off region (without any discernible
slope), this step may be skipped and KD
=
00.
STEP 6' 13 =
.
1m
{
(VDSl- VOSl + VT)}
(VOSl-IDl Rs- VT)2 1 +
KD
IfKD
= 00
then 13
,
=
IDl
(VOSl- 1m Rs - VT)2
For the SPICE 2 model, KP = 2*{3
STEP 7: VT to be obtained by solving
(VDS3 -KVDOS3+ VT)}
ID3 = 13 (VOS3-I D3 Rs - VT)2 {l +
One way to achieve this would be to solve for
VT, iteratively, the rearranged equation.
ID3
VT = VOS3 - ID3 Rs-
(VDS3- V OS3+ VT)}
KD
2-58
IfKD
= 00, then
(iD3
VT = VGS3-ID3Rs-,,~
For the SPICE 2 model, VTO
=
VT.
STEP 8: Repeat steps 4, 5, 6, and 7 until parameters
VT, R s , KD, and 13 converge. It may be
necessary to carry out a number of iterations
before the parameters converge.
STEP 9: Letb ~ 2(VGS6- VT-VDS6)
ID6
c ~ _1__ ~(V--,D=S6~-----2ID~6~R:..:::s~)~(2:....:.V~G~S~6--:-~ID~6:.:R~s_--=2V..:..T~-~V~D~S~6)
(3ID6
ID62
RD=
-b +Jb2-4c
2
--~--
For the SPICE 2 model, RD = RD.
STEP 11: Refer to the switching time test circuit
(Figure 5). A simplified equivalent circuit
during turn-on and turn-off is shown in
Figure 6.
Note: Since P6 lies on the ohmic region line, determining V GS6 is somewhat arbitrary. However, RD is
not found to be critically dependent on this.
STEP 10: Refer to the capacitance vs. VDS curves
(Figure 4d).
a)CGD = Crss(athighVDS)
For the SPICE 2 model, CGS = CGD
b) CGS = Ciss-C rss (at high VDS)
For the SPICE 2 model, CGD = CGS
c) CDO = Coss-Crss (at VDS = 0)
For the SPICE 2 model, CBD = CDO/
(lxlO- lO ) and PB = 1.0
-
+25V
RL
PULSE
11.n
GENERATOR
r----------------,I
!
I
I
I
I
I
R1
I
v.
II
t_______
Note 1: CGD and CGS are approximated
by constant value capacitances. Also, the
units for CGS and CGD in SPICE 2 are
farads/meter. This, however, does not create any error since the default values for
channel length and width in SPICE 2 are
1.0 m each.
50n
I
OUT.
,
II
,
,
I
TO SCOPE
R'
50n
--------~
'0+
001 SHUNT
Switching Time Test Circuit (Ref. 3)
Figure 5
R11IR2=R'
Note 2: As pointed out in Reference 1,
CGS and CGD have been interchanged in
SPICE 2 MOSFET model, apparently an
error.
RG
'h~_ '
( .!:IR1+R2VP
Note 3: The unit for CBD in SPICE 2 is
faradlmeter2 . The default drain area in
SPICE 2 is 1 x 10-10 m2 .
Simple Equivalent Circuit of Gate Input
Figure 6
2-59
By adopting the above method, a computer model for
the device type IRF330 was generated. The model
listings may be seen in Figure 11 which shows a SPICE
2 program of a circuit using IRF330. Here, the NMOS
model IRF330 along with resistor RF constitutes the
device model.
Let
R' ~ R1R2
R1 + R2
V'p ~ VpR2
R1 + R2
RG1 = _ _T...oO::.::N~_ -R'
V'p
Ciss In
V
V'p- T
The capacitor Crss is a highly non-linear function of
VDS, especially at low VDS values, whereas it has
been approximated as a constant value capacitor. This
could be a source of error when using SPICE 2 with
this model. The dynamic performance (rise and fall
times) was found to be critically dependent upon the
value of Crss chosen.
where TON is the turn-on delay.
Simulation with the Model
To consider the turn-off delay, plot the load
line on the output characteristics as in
Figure 7. This intersects the ohmic region
line at point Q. Find the gate-to-source
voltage, VGS = VA, such that the corresponding pinch-off region characteristic
passes through Q when extrapolated as
shown.
DC Characteristics
Considering the turn-on process,
The output characteristics, saturation characteristics,
and the transfer characteristics were simulated with
this device model. Comparing these simulated curves
(Figures 8a, 8b, and 8c) with the characteristic curves
provided by the manufacturer (Figures 4a, 4b, and
4c), we note that the model simulates the performance of the real device accurately both in the ohmic
region and in the pinch-off region. However, in the
area in which transition from one region to the other
occurs, the model is less accurate.
Then
RG2 = _ _T...oO:::.;:FF,-=::---;- - R'
V'p)
Ciss In ( VA
Switching Characteristics
RGl
The real test of a dynamic model is its ability to
simulate dynamic behavior of the device in a circuit in
a fairly accurate manner. With this in view, the model
was tested with resistive loads and with an inductive
load under switching conditions.
a) Resistive Loads:
i) Switching Time Test Circuit:
This circuit, shown in Figure 5, is the test circuit
adopted by the manufacturer for measuring the
switching times at low voltages (Reference 3). The
results of the simulation (Figure 9) indicate that,
while the details of the waveforms (Figure 7C,
Page 16, Reference 3) are not simulated accurately, yet the 'ON' and 'OFF' delays have the
correct values. The rise and fall times are between
20 and 30 nsec while the actual device has typical
values of 50 nsec. This discrepancy is probably
due to the error introduced in simulating the nonlinear capacitor Crss with a fixed value capacitor.
In fact, by changing Crss from 20 pF to 40 pF
rise and fall times close to the manufacturer's
specifications may be obtained. However, since
IRF330 is a high voltage device, the value of 20
pF for Crss corresponding to higher VDS values
was retained.
+ RG2
For the SPICE 2 model, RG = - - - - 2
Note: RG has to be included as an external resistance,
since SPICE 2 MOSFET model has no built-in element for this purpose.
-
"os
To Determine VA
Figure 7
2·60
OUTPUT
LfGENO:
+:
*:
ves
~:
4. ';V
VGS
#: VGS
VGS
@:
VG~
5.tlV
5.6"V
10.11V
VOS
(VOLTS)
!l.tI
? flllll[J+{I0
4.0000+00
6.0000+00
8 onOD+oo
1.0000+01
1.2000+01
1 4000+{)1
1.6000+01
~
6.51)110-01
6. SOIID-Ol
6. 51l /1LJ-O 1
6.5U40-"1
6.5fI JW-tJl
J!
6.50 110-n1
m
2.6000+01
2.8000+01
3.0000+01
6.5040-01
..
'" _.
N
6. 'JlIlln-lIl
6.51l4D-Ol
6.5fJ4D-1l1
6.50 /'0-01
2 2000+01
2.40flO+01
::!llll
<:3
3.2000+111
3.4000+01
3.6000+01
3.8000+1)1
;5.
~g.
.!!!."
4.0000+01
4. 2110D+() 1
4.4000+01
4.6000+01
4.8000+01
5.0000+01
5.2000+01
5.4000+01
5.6000+01
5.8000+01
6.0000+01
6.2000+01
6.4000+01
6.6000+01
6.80UO+01
7.0000+01
7.20UO+Ol
7 ,'1000+01
7.6UOO+Ol
:II
IX
<:
iii'
<
t:I
III
<:
o
1u:::::::::
8.2000+rll
8. "000+01
8.6000+01
8.8000+01
9. OOOO+() 1
~
Uggg!g~
-
9.6000+01
9.8000+01
6.504D-Ul
6. SOliD-Ol
6. S(JliIJ-Ol
6.50,,0-01
6.5040-01
6. 5(J/,O-O 1
6.50 LID-01
6. SO/tO-Ol
6.5040-01
6.50'10-01
6.50"0-fll
6.5()1ID-()1
6.5040-01
6.51)1,0-01
6.5040-01
6.5040-01
6.50"0-01
6.50 1'0-01
6 50',D-lll
6.50 110-01
6.50,,0-01
6.50"0-01
6.5040-01
6. 501iD-0 1
6.5""0-01
6.50"0-01
6.501,0-01
6.50"0-01
6.501'0-01
6.50 1'0-01
6.5040-01
6.5040-01
6.5040-01
6.501'0-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
2.0000+00
0.0
2. (981)-42
1 • aOOD+Ol
2.0000+01
0
10 (AMPS)
10
(MIPS)
(+'#@=) -------------
III
CHARACTrRISTIC~
'I.OV
4. (1000+00
6.0000+00
8.0000+00
X
+
* .#@= *
+
,
,
+
+
+
,
,
+
+
+
+
+ <
+1:)
+111
+ II
+ ...
+ b
+ <
+
+
+
+
,
• <
,I:)
,III
, II
· ...
,• in
<
#
'"
,
,
,
·
*,
+
+
+
+
@
@
@
@
@
@
@
u~
Z III
@
@I:)
H
@III
@ II
Z<
#
#
#
H
#
H
#
H
H
#
H
H
#
H
#
H
#
H
H
H
H
H
I
II
#H b
'"
H
H
H
+
+
+
+
+
+
+
+
+
@
@
#
#
H
H
H
H
H
H
H
H
H
+
+
+
+
+
+
+
+
+
@
#
H
H
<
@ '"
@
in
:<
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@
SAfURAT I ON CHARACTER I ST ICS
VOS
(VOLTS)
(AMPS)
(+*N@=I------------0.0
1. ooon-o 1
2.0000-01
3.0000-01
4.0001l-01
5.0000-01
6.0000-01
7. 0000-01
8.0000-01
9.0000-01
1.0000+00
1.1000+00
1.2000+0fJ
1.3000+00
1.4000+00
1.5000+00
1.6000+00
1.7000+00
1. 8000+00
mm
e~
...,
~
1.900D+1H1
iil m
Jl=N
cagle
2.0000+00
2.1000+00
2.2000+00
2.3000+00
2.4000+00
2.5000+00
2.6000+00
2.7000+00
2.8000+00
2.9000+00
3.0000+00
3.1000+00
3.2000+00
3.3000+00
3.4000+00
3.5000+00
3.6000+00
3.7000+00
3.8000+00
3.9000+00
4.0000+00
If. 1000+00
!; (')3
CD:::r C
.,ii~
-!l g
!:a
al
~~
t:~ggg:gg
~
m
_
0<
r.
'-I
!!J
I
4.4000+UO
,.. 5000+00
4.6000+00
4.7000+00
4.8000+00
4.9000+00
5.0000+00
5. 1000+00
5.2000+00
5.3000+00
'0 (AMPS)
10
?0980-42
9.2090-02
1.8210-01
2.6970-01
3.5-111 0-0 1
lL 34110-01
5.0910-01
5. "1570-01
b.2830-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.50'10-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-0 (
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
6.5040-01
x
x
5.0000+00
1.7500+00
2.5000+00
1. 2500+00
0.0
x
+XX
+XX
+·X
+ *X
+ *X=
+ *X=
+
+
+
+
+
+
+
+
+
+
+
+
+
<+
G) +
m
+
II +
~
Q
+
+
<+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
*#X
*#X.
*#x.
*#@=
. *#@~
·N@=
* /IX
*
•
I/@-
•
#@c
#@=
* #f!:::.
• #@=
#@ =
•
•
•
#~.
# @=
# @=
# @
*
#
"
·"
<.
G) "
m"
II "
... "
in·
<"
"
·"
"
··
"
"
"
=
@
#
=
@
#
=
@=
#
@=
#
@
N
#
=
@
#
=
@
=
@
#
=
~
m
@.
#
@
N
@
#
.#
=
@
II
=
@
I
....
=
?Q
@
#
@
I
@
I I
=
@
<
=
@
@
1#
~
~'\
m I
m
I
i:>
III
i:>
<
II
GO
~ <
I
@
@.
DC TRANSrrR CURVE
VGl
(VOLlS)
10 (AMPS)
I(Vll )
(AMPS)
~
n.n
fJ.n
l).n(JOD-O?
1.00(11 1-11'
3.0001}+{JO
3.05UO+00
3. 1DOD+OU
3. 1500+110
3.200D+1I1I
3.25lH}+(UI
3.301J[)+IlO
3.31)011+110
3.1101l11+00
3.45I1D+OO
3. suulHon
J.55(1)+(10
3.6IlilO+OO
III
3.6500+00
mo
"'m
3. Boon+un
.... "11
3.700{H(JO
3.7500+11U
!!l,,,,
..,
0.
'"
1. 8~1I0+f1n
3.90110+00
1.9 r)IJIl+(JO
11.00(J[)+OO
!!!!cn
cg.. ::rc
03'
CDD>-
II.O')OD+IIII
... iiI!!!.
'1.1000+(1)
4 15110+110
4.200D+1I1I
.. !I. 0
-CD'"
::!.:u
~=
'1.2500 i UO
11.300U+1I0
ll.35UO+00
".400()+on
4.45110+01 1
11.5000+1H1
'L 5500+01J
'1.600D+CJlJ
!;l~
II>
~.
6500+00
4.700D+00
11.7500+00
11.8000+00
II. 850D+(J1)
11.9000+(10
11.950D+OO
5.UOOO+Oll
<
I:)
en
"<
o
!:i
III
-
j'"'""'""
5.100D+Qn
5.1500+00
5. 2000+01J
5.2500--111
2. ~991l-(I1
3.91180-111
5 19~D-Ol
6.5(1110-01
7.90?1J-1I1
9.189D-(I1
1.09 t )[)+1111
1.25BD+nO
I. ~~ 71l+f)O
1.6010+011
1 7810+IH)
1 9661HIHl
2.156[J+0[)
2.3500+(JO
2.548D+01I
2.75UO+OIl
2. S'~5D+OO
3.1b'ID+OO
3.376D+OO
3.5900 i OO
,.8080+00
~. 0290+no
11.2520+00
4.117/D+IIIJ
~. 7050+0 f J
~. 'J360+0n
5.168D+()O
5.4030+0U
5.6390+00
5.8780+UO
6.1180+00
6.360D+OO
6.6n1.JD+()1)
6.81190+00
+
<
C
III
II
~
+
0
0
<
+
+
+
+
+
+
+.
---------------------------------------------------
I
LEGEND
+
*
_
(It
CD..
:
'1(2)
1('12.)
TIME
\1(2)
- - - - - - ----- ---- - -- - -- - -- 501)00+110
-- - - - - - - - - -- - --- - - - --
(+)-----------------
-5 0000+00
(*)-----------------
-30000+00
00
10000-08
20000-08
30000-08
40000-08
50000-08
60001;)-08
70000-06
80000-08
90000-08
I 0000-07
11000-01
12000-01
I ]OOO-l17
, 4000-07
15000-01
16000-01
I 7000-01
18000-fl'
19000-01
20000-01
2 1000-01
22000-01
2]000-01
24000-07
25000-07
26000-07
2 7000-07
28000-07
29000-07
] 0000-07
1 1000-07
3 2000-07
3 3000-07
34000-07
3 5000-07
3.6000-07
3 7000-0,
3 8000-07
3 9000-0r
40000-07
4 1000-01
4.2000-07
43000-07
44000-01
4 5000-07
46000-07
4 7000-07
4.6000-07
49000-07
5.0000-07
51000-01
52000-07
53000-07
511000-07
5 5000-01
56000-07
5.7000-07
5800D-07
59000-07
60000-07
61000-07
62000-07
6300D-07
64000-07
6 5000-01
66000-07
6 7000-07
6 8000-07
6 J}OOO-07
70bOO-07
7 1000-07
72000-07
73000-07
74000-07
75000-07
7 6000-07
1 7000-07
7 8000-07
7 9000-01
80000-01
8 1000-07
82000-01
8 3000-07
8 11000-07
8 5000-07
86000-07
87000-01
88000-01
89000-07
90000-07
9 1000-07
9 2000-07
9 3000-01
94000-07
95000-07
96000-07
97000-07
9 6000-07
9.9000-01
10000-06
00
-1 2630-16
22440-16
24670-16
2 3260-16
-21310-16
76750+00
80200+00
8 3120+no
811850+110
8 6790+00
86750+00
90420+00
9 16110+00
930')0+00
94080+00
9 4960+110
9 Hl0+00
96350+01)
96890+00
9 1350+00
97150+00
98080+00
98310+00
98610+00
988:<'0+00
98990+00
991110+00
99270+00
99380+UO
991170+00
9 9,)50+00
'I
96~O+fJO
99670+00
9 97~0+00
9 9760+00
99800+00
99830+00
99650+00
99880+00
9 9890+00
99910+00
99920+00
9 9930+00
999110+00
9 9950+00
99960+00
99970+00
99970+00
9 9980+00
99980+00
99980+00
99980+00
<} 9990+00
<;I 9990+00
9 9990+00
23270+00
19610+00
I 6860+00
1 4360+00
I 2:<'110+00
1 1010+00
98650-01
8 11390-01
7 1870-01
61200-1)1
52100-01
411360-01
3 7770-01
3 2160-01
2 7380-01
23320-01
19850-01
1 6900-01
111390-01
1
..
:
~260·01
1 0430-01
8 8850-02
7 '630-0~
6111120-02
511820-02
46700-02
391110-02
33860-02
28810-02
24550-0:<'
2 0880-02
t 1800-02
1 5130-02
1 2910-0:<'
10970-02
9 3630-03
7 91180-03
6 7910-03
5 1580-03
49270-03
41110-03
35160-03
3 0200-03
:<' 5960-03
:<' 2140-03
SPICE 2 Simulation Waveforms Switching Time Test
Figure 9
2·64
-- - - --
I 5000+01
.. _." .. It'II""'''•• ''_'' ••
ii) Switching Efficiency
:*!~~~!*~!~!!~~.:
One of the important considerations for the operating frequency in switching applications is the
efficiency of the switch. Reference 3 (Page 18)
describes a test performed to determine the efficiency of the switch at different frequencies with
resistive load. Figure 10 shows a SPICE 2 diagram
which simulates the test circuit. Figure 10 also
shows additional SPICE 2 circuits to measure the
device dissipation and the source power delivered.
The measurement principle is as follows.
TEST Of MOSfET MODEL
----------------------------------------------------------------------
:~~~~~~!~~-~:~!:!~~:~-~:~:~~!~~~!~~-~~-~~~-~~~-:-~:~!::-~~~::!~~~~~---
VO 1 0 PULSEjO.
Rl 1
MOS1
RL "
VO 0
01 0
Cl 6 7 2.ur
C2 8 9 2
Itltl +
flV
=...!... Itl +
C2 C2 tl
=...!...
C1
ur
~~~-~-~-~:-~-~--------------------------------------------------------
RG23665
.MODEL IRF310 NMOS(V10"3.60'18 KP=ll Illl lAMBDA=O. RS=.1293 CGS=20 PF
!------------------:~~:~~~-~~-:~~:~.:~-~~:~--~~:.:~:~~-~:~:~:~!--------,TRAN 20 NS 6 2US 1.8US UIC
,PLOT TRAN V(41 V(21 V(ll I{VOI I{VZll I (VZ21
PRINT TRAN I(VZ1) V(6.7) IIVZ21 V(8,91
.. ~~!~~~~-~~~~!~:~~~~-------------------------------------------------[NO
SPICE 2 listing for the Circuit to Test SWitching Efficiency
Figure 11
Figure 12 shows the simulation results, along with
experimental data from Reference 3. It is found
that the model predicts the efficiencies at different
frequencies in a reasonably accurate manner.
T
T
Note: The 1000 Mil resistors across C1 and C2
are included to meet SPICE 2 programming
requirements.
By selecting Cl = C2 = T,
fl VC1 = PI = average power dissipated in
MOSFET
ID.
..
fl V C2 = P2 = average power delivered by
source
il
Switch
.
.
Power mput-MOSFET losses
Efficlency
= ---....!....-------Power input
~
10 - 3 AMPS
9.
•
96
I
I
SIMULATED
I
M
o
/~,
R1
sn
0 \R? . ~
~OS1
0
-
I'r--
95
93
R:on
r-r-..
--EXPER!MEJAL (Raf 3)
97
9.
IRF33~...... _........
01
RC2 6 9 lOOO.MEG
In one period, T,
Cl
10 0
1NS ,INS .9999U5 2 US)
5.
3 0 0 IRrl30
100
DC -300 V
POL VIZ) 4 0 5 " 0 0 0 0 .01
ReI 6 7 1000 MEG
VZ170DCO.V
G2 0 8 POLY(21 5 0 5 " 0 0 0 0
Gl = VDSiD = instantaneous power dissipated in MOSFET
G2 = VDiD = instantaneous power delivered by source
flV
2
4
5
5
6
'0
2.
.
CONDITIONS
50% DUTY CYCLE
TJ = 10O"C
RG = 5 OHMS
= 300 ViLTS
ro
'DO
200
5DO
'.DO
FREQUENCY (kHz)
0
Switching Efficiency vs Frequency
Figure 12
b) Inductive Load Switching:
Figure 13 shows a circuit (Page 18, Reference 3)
that was used in order to study the ability of the
model to simulate switching circuits with inductive
loads. The diode was modeled as a near ideal
device, and stray inductances were not included in
the simulation model. A capacitive speed-up network was included in the MOSFET gate drive
circuit. The inductance value was chosen to be
large so that the current remains essentially constant over the interval of interest. The circuit was
simulated both under ON and OFF transient
conditions. SPICE 2 listings for switching off are
given in Figure 14. The ON and OFF simulated
waveforms are shown in Figures lSa and lSb,
respectively.
YO
-=-3OQV
68 50'1 .. _ ......
va
SPICE 2 Simulation Diagram to Test Switching Efficiency
Figure 10
2·65
then at approximately 300 volts. VDS now falls at
such a rate as to maintain the internal gate-tosource voltage, VOS, (across COS) ofthe device,
equal to the value required to sustain the load
current. In this case, this value of VOS would be
4.65V as may be seen from the output
characteristics.
D1
In Figure l5b, the gate voltage is switched off at
500 nsecs. Initially, a negative voltage appears
across the gate-to-source, though the final voltage
would be zero. This is due to Cl, the speed up
capacitor. VDS now rises to maintain the internal
gate to source voltage, VOS, once again at approximately 4.65 volts. ID continues to flow through
the MOSFET. When VDS reaches a value a little
more than 300V, the diode takes over conduction
and the device current drops quickly from full
value to zero.
Switching Circuit with Clamped Inductive Load (Ret. 3)
Figure 13
*****************
:*!~~~!*~!~!!~~*:
!:~~~~:-~~~::~-~~~:~-------------------------------------------------
:~:~:~~~~~~-~:-~~~~~!~~:-~~~!~~~~~-~~~:~~~~-:-~:~~~:-!~~::~~:~~~-----Rl 1 2 50.0HMS
R2 2 3 1.K
Cl
R3
Vl
Ll
2
3
1
6
3 OlUr tC=10 v
0 1.K
0 PUlSEI20 5 O. 500 NS .1N5
5 5.Htt Ie".. 3.0A
In Figure l5a, the rise times of ID and VD are 0
nsec and 50 nsec, respectively. The corresponding
values from Reference 3 are 80 nsec and 30 nsec.
In Figure l5b, the fall times of ID and VD and the
time delay (off) are 0 nsec, 50 nsec, and 25 nsec,
respectively. The corresponding values from Reference 3 are 40 nsec, 40 nsec, and 60 nsec. A
major discrepancy is observed in the rise and fall
times of the drain current. This may be attributable to the likely presence of stray inductances in
the experimental circuit.
lNS 1.5 1 15)
01 5 6 0
V2 0 6 DC -JOO.V
MOSl 5400 IRF330 IC:::2.3. 10., O.
HODEL 0 0
*-----------------------------------------------------------------.--.
RG 3 4 66.50HMS
,HODEL IRF330 NHOSIVTO:::3 60146 KP=13 44 LAMBDA=O
RS= 1293 CGS"'20 PF
:------------------~~~:~~~-~:-:~~:~:~-~~:~--~~::.~~::-~:~:~:~!--------.TRAN S.NS 950.N5 450.N5 ute
.PLOT TRAN V13) VIS} IIV21
*~~!~~~!-~~~~~~~;,~~~~!-----------------------------------------------END
SPICE 2 Listing for Inductive Load Circuit
(Switch-Off Conditions)
Figure 14
In Figure l5a, the gate voltage is applied at 500
nsecs. The voltage at the device gate quickly rises
above the final value of 10V due to the presence of
the speed-up capacitor, Cl. Until the gate input
capacitor, Ciss, charges up to the threshold voltage, VT, the device remains in the cut-off region.
The device starts to conduct at about 515 nsecs
commutating the free-wheel diode. Since no parasitic elements are considered, this transfer takes
place almost instantaneously. VDS remains until
Note: When simulating a switch in series with an
inductor using SPICE 2, the voltages across the
switch and the inductor are sometimes inaccurately computed under switch off conditions (i.e.,
switch current and inductor current = 0). The
voltages may even, erroneously, appear to be
oscillatory. The more stringent relative error tolerance (RELTOL) value used in the program in
Figure 14 is to avoid this source of error.
2..fi6
~~;
lEGEND:
+: VI 3)
': VI 5)
0: IIV2)
TIME
fl.
n
5.0000+00
1.IlQnUt()l
1.5000+01
? nouo+nl
(*) -----------------
o. (l
1.0000+n2
2.OIHIU~I)?
3.o0QO+()?
II. (j{)OD+02
0.0
?()(J(J() I [)"
II.OClIIO+llli
fi .IJOUO+OO
+
+
+
4.7000-111
I . 3990-1)3
1.396U-[l3
1.39,,0-03
I. 3911J-03
11.75110-07
1.3SQU-(J3
4 800U-1I1
4 8500-01
1.38l0-03
1.3850-03
1.383U-01
~.6511D-07
3en
".9110D-07
ID-
4.950[)-1l7
~~
:IN
'"~
5. OUOo-o 7
1.3810-113
I. J 191J-03
5.0500-07
1.181111+01
r
...,~!!!
5. llIUU-O
5.1S0D-OJ
~~iii'
5.2500-0 I
5.3IJOO-07
.!!!3
5.351)()-flT
5.4001J-lJl
cS' g, ~
5.2o0D-Or
~Qi1f
::I
5. 4~IUD-07
5. ,000-07
~ . 55(1)-(17
.!.$:;:
~<
~lr
S'3
7",
o
a.
-7 IH'OO+IJII
1.40 )0-03
',.6uoO-O"
'0'11
~.600D-07
5 650IJ-f17
5.7000-117
5.75l10-07
5.8000-Cl7
5.8500-07
5. 900U-IJ I
5.9500-07
6.(JUOD-07
6.0500-07
G.10 I )O-07
6.1500-07
6. 2[IIJI)-(1 I
6.250[)-07
6.3000-07
6. 15OfJ-fJ7
<
+ I:)
,en
+<
+ 0
+ !:j
+
en
+-
#
#
# -
• <
0>
# ;;:
'<
, 0
#
*, !:j
en
* -
U "II
# en
0#
H
+
,1
1.3111f)+-()1
1.311}()+1J1
1.3Ub(J+\I1
l.302D+II1
•
o +
o +'
# +
o+
# +
U +
1 • 299()+fll
o+
o+
1.2950+01
1.292U+UI
0+
#+
1. 28QO+1) 1
1.2870+01
# +
/I +
1. 293D+1) 1
1 • 32~UtfJ1
o
1.3530tfJl
#
#
#
#
1.381LHfJ1
1.4011J+Ol
1.430D+1)1
*,en
0
#0
o
I 22lU+{ll
1.267D+Ol
1.3160+111
1.1153U+01
1.1173U+01
1.493U+01
1.5110+01
,
,
.
o
o
1.~2/DHll
*
#
1. 543D+01
I. 5581J+O I
6. 'WUD-07
1.5710+01
1.5840+01
6.450U-07
~ 16.5000-07
6.5501J-f17
n 6.6000-07
.!!. 6,650U-07
6.7000-07
6.7500-U7
6.8000-07
1.5950+01
1.6U6U+UI
1.616U+01
1.6260+01
I. 634U+0 I
1.6430+111
1.6501J+OI
1.6570+01
:c
f URNS
VI 3)
I,. SOOD-Ol
4.5501J-UI
Q
wl\vr
1+)-----------------
(# 1-----------------
.
I I (It-or~
o
o
#
#
*
o
o
o
,*
*
o
#
o
o
*
*
#
#
I
+ .
+.
+
.+
+
+
+
+
+
+
+
+
+
+
+
+
LEGENO:
SHII(,H-OFF \"AVEHmN5
+: V( 3)
*: V(5)
#:
I (V2)
TIMf
VI,)
(+ )-----------------
-1.01100+01
(*)-----------------
(H )----------------11.500D-07
..3111
CD-
:>",
5.1000-07
O
c.
_m
&l
1. 0020+01
L!. 9~UD-fl'
S .0000-07
"CI"CI
.,
1.0020+01
'1.600D-1I7
11.6IjOD-07
'I. lOOD-lIl
lL 15UO-l17
I). fl50D-!)7
~III
'J 15f}O-07
.a'~Uj"
5.2000-117
'J.2500-07
5.31100-07
5. 35(1[)-U i
c-
" t--
I"01
K=I./lb
Variation In Irms as a Function of the Trapezoidal Current
Ration (K)
Figure 21
The Converter Circuits Used for the Switch Current Calculations
Figure 22
on ROja and RDS. As a rough guide, L should be at least
twice the critical inductance (Lc) required to maintain
continuous inductor current at full load over the range of
Vs. Making L much larger than 3XLc usually isn't worthwhile since most of the benefits have already been achieved.
We can calculate the rms values for the switch current (i I)
in each circuit, first by assuming L is sufficiently large so
that the switch current waveforms are rectangles, and second by assuming L is smaller so that the waveforms are
triangles, but with the same value of D. The results of this
calculation are listed in Table 3.
Operation in discontinuous inductor current mode raises
the power losses rapidly to levels even higl1er than indicated
in Table 3. In general, operation at full power in the discontinuous mode is unacceptable from a power loss point
of VIew and is usually avoided at powers above IOOW for
off-line operation and 25 to SOW for lower voltages.
In all cases il(rms) is lower when L is made larger or when
circuit B is used in preference to A. The relative losses are
proport!onal to [il(rms)12 plus a factor to allow for the
thermal regeneration. Currents can be compared by exponentiating the ratio by a factor of 2.5 giving a more
realistic comparison of the actual losses. The exponent will
always be greater than 2, just how much greater depends
100
~Tc
en...
Table 3
Comparison Values for rms Switch Current
:;
~
I-
CIRCUIT
L
Vs
A
Small
20V
3.0A
B
2.79A
:V·5{~V·5
20V
2.65A
2.42A
1.25
Small
25V
2.74
2.35
1.47
Large
25V
2.37
2.03
1.47
Small
30V
2.51A
1.77A
2.39
10
a:
a:
~-BL
~tt=B
()
c
1.43
I
1.0
~
L- RDS(on)
1.44
"'
Power Limit
Limit
C
.P
Breakdown
0.1
1.0
,Willi
10
100
D
1000
V DS - DRAIN SOURCE VOLTAGE (VOLTS)
1.44
1.77A
TTl
Current Limit
:::l
«
a:
1.44
2.17
~.Bs} 2.5
1.20
L.arge
30V
zw
25°C
150°C
III
Z
1.36
Large
t=Tj
1
Typical MOSFET SOA Curve
Figure 23
1.66
4-12
two straight lines (dotted lines Figure 24) in the data
sheets since this gives a more accurate definition of this
boundary.
100
Tj
enc..
f - - RDsLlmit DC
:2:
I-
2
10
w
cr
cr
~;;..
The data sheet SOA curve is not unique, even for one
particular device. The reason is the SOA boundaries do
not generally represent limits which, if exceeded, will
result in immediate destruction, but rather these boundanes are limits beyond which the service life of the device
becomes unacceptable. What is acceptable in one application may not be in another. To illustrate this point, we
have added a third aXIs to the SOA curve (Figure 25) to
represent useful life. As the boundaries are reduced, the
useful life of the MTBF increases dramatically at first and
then more slowly. This is why in applications where especially long service life or very high MTBF is required,
significant derating may be needed beyond manufacturer's
specifications.
~ ~DS limit
Pulsed
..
.'
~.,
(J
.'
2
0
./
'
::J
«
cr
I'
I
~
25°C
1.0
I
V' 1/"
FTj 150°C
0
....... - Siliconix Approximation
0.1
10
100
VOS - DRAIN SOURCE VOLTAGE (VOLTS)
4.
Snubbers can be used to remove some or all of PS.
There are many different snubber circuits, and the literature is quite extensive [2-6]. The reduction in Ps will
depend on the particular circuit chosen.
The SOA curve is also a function of the Tc, Tjmax, tp
and D. The data sheet SOA curve will normally be for
Tc = 25°C, a single pulse, and a few different values of tp.
Few actual applications enjoy these particular conditions.
The manufacturers should not be criticized since there are
an infinite number of possibilities, and they have merely
chosen to standardize on one, to allow a direct comparison
between different devices. The MOSFET SOA curve is also
patterned after the traditional Bipolar Junction Transistor
SOA graph.
5.
An HP-41 program for calculating ID(rms) is given in
appendix A.l. This will greatly simplify the rms calculation
and make it easier to optimize the circuit.
For each application, the user must create an SOA curve
that reflects his specific requirements. Fortunately this is
not a difficult task.
RDS(on) SOA Boundaries
Figure 24
The SOA Curve
Useful life
A typical example of a MOSFET data sheet SOA curve is
given in Figure 23. The curve exhibits four limiting boundaries: maximum current (A-B), maximum power (B-C),
maximum voltage (C-D) and the RDS(on) limit (E-A).
The current limit is set at a level which limits the current
density in the bonding wires and in the die surface metali·
zation to provide reliable operation. The power limit is that
power dissipation which will raise Tj from Tc to Tjmax.
The voltage boundary is determined by the designed breakdown voltage of the device. Actual breakdown voltage of
the device is higher than this limit but varies from one
device to another and varies with temperature. The SOA
boundary is selected to lie inside of all normal variations.
The RDS(on) limit is due to the minimum "on" resistance
of the device. For example, it is not possible to have IDA
flowing through a 2Q resistor when only I V is applied.
or MTBF
Vos
At the present time, all manufacturers do not draw the
RDS(on) boundary the same way. The solid line in Figure
24 shows the actual limit allowing for the heating of the
device due to the power being dissipated. The dashed lines
represent the boundary for RDS(on) at 25"C and 150'C.
Any of the three boundaries may appear on a data sheet.
Siliconix has elected to approximate the actual limit with
A Three Dimensional Representation of SOA for a MOSFET
Figure 25
4-13
How to Create Real-World SOA Curves
2.
Tjmax = 125°C
Figure 26 is a typical data sheet SOA graph. As pointed
out in the previous section, this graph is valid only for one
set of conditions which in this case is:
3.
Tc
85°C
4.
tp
200 psec
5.
D
0.25
6.
The device is an lRF440
7.
Rejc = IOC/W
8.
ret) = 0.27 (from Figure 6)
2.
Tc
3.
D
4.
tp
o
STEP 1: Re-draw the RDS(on) limited boundary.
A typical user might wish to make the following changes:
Points on the boundary for Tj < 125"C can be calculated
from the following expressions:
I.
Tc higher
2.
Tj lower
3.
Repetitive pulses, D = 0
4.
Derating of VDS and ID
5.
A tp other than that shown on the graph
(33)
ID=
ROjc ret) RDS(on) RDSN
(Tj - T cJ RDS( on) RDSN
VDS =
(34)
Rejc ret)
For Tj = 125°C the boundary is defined by:
The new SOA graph can be created with a simple step by
step procedure as shown in the following example:
ASSUME:
A plot of the RDS(on) boundary for this example is shown
in Figure 27. The values for RDSN are taken from Fig. 28.
Voltage and current derating factor = 0.8
I.
100
100
,
en
0-
::;:
~
I2:
." "
10
100 j1sec
"
w
2:
;;:
I
.P
1.0
/
::::::
-
0.1
1.0
Tj
ETj
;;:
= 150°C
a:
Tc = 25°C
tp = 100 j1sec and DC
0
100°C
,/'
(,)
2:
E::=::
.,.-:
,0 t?'
10
~
DC
",
125°C
~
a:
a:
~
(,)
0
Tj
0-
::;:
~
I2:
w
a:
a:
a:
en
/"
1.0
I
eTj
0
V
85°C
Single Pu Ise
II,I!!!
!I
10
100
0.1
1000
1
V DS - DRAIN SOURCE VOLTAGE (VOLTS)
10
100
V DS - DRAIN SOURCE VOLTAGE (VOLTS)
IRF440 SOA
Figure 26
The Modified ROS(on) Boundary
Figure 27
4-14
Two points corresponding to this power are:
VDS
40
60
80
101)
120
140
160
180
10
A
SOV
2.97A
B
200V
0.74A
Now plot these two points and draw a straight line between
them. This becomes the pulsed power boundary as shown
in.Figure 29.
TEMPERATURE (OC)
The Effect of Temperature on ROS(on)
Figure 28
STEP 2: Re-draw the ID and VDS boundaries parallel to
the original boundaries but reduce them by a factor of
0.8 as shown in Figure 29. Note in this case the 10
boundary does not exist because the ROS boundary takes
precedence.
The graph in Figure 30 (solid line) is the final SOA curve
for the IRF440, given the specified operating conditions. In
comparison to the original SOA curve (dashed line), the
new curve is very different. Clearly, if you use the unmodified curve, it is highly probable that device failure
would occur if the device is operated anywhere near the
original limits.
Conclusion
STEP 3: Re-draw the thermal boundary. The total power
for the boundary can be determined from:
TJ·-T c
PT=
As we have shown, the thermal calculations for a power
MOSFET are not particularly complex or time consuming.
It has also been shown how important these calculations
are to achieve an efficient and reliable design. Most design
parameters are under the designer's control, and by juggling
the variables around a bit, an efficient design can usually
be aclueved. If the thermal design is overlooked or given a
short shnft, there is small chance for a successful design.
(36)
148W
ret) Rojc
100
100
Data Sheet SOA
...
::E
if
::E
~
I-
zw
a::
a::
Ul
~
r\..
10
I-
zw
a::
a::
Derated Power=
:>
:>
z
({
a::
z
({
a::
CJ
C
I
"
1.0
.9
C
I
C
1.0
10
100
==
_J
- =r
_ Tc-
II
0.1
1.0 ~~
-
Derated
BVDSS
1000
V DS - DRAIN SOURCE VOLTAGE (VOLTS)
Rt
,'.
CJ
0.1
1.0
125°C
85°C
_ tp = 200 11 sec
D = 0.25
I
II1II
" -,
~
10
SOA
"
I
10
100
1000
V DS - DRAIN SOURCE VOLTAGE (VOLTS)
Modified Boundaries for BVOSS and Power
Figure 29
Final SOA Graph for the IRF440
Figure 30
4-15
References
[1] W. Roehr, "Mounting Techniques for Power Semiconductors," Motorola Application Note AN778,
February 1978.
[2] R. Walker, "Circuit Techniques for Optimizing High Power Transistor Switching Efficiency ,"
Proceedings of POWERCON 5, May 1978.
[3] T. Undeland, "Stress Reduction in Power Transistor Converters," IEEE-lAS Annual Meeting
Proceedings, October 1976.
[4] Harada, Ninomiya and Kohno, "Optimum Design of an RC Snubber for a Switching Regulator by
Means of the Root Locus Method," Proceedings of the IEEE-Power Electronics Specialists
Conference, June 1978.
[5] G. Edwing and R. Isbell, "A New High Efficiency Tum-Off Switching Aid for Power Transistors in
Switching Regulators," IEEE Journal of Solid State Circuits, Volume SC-17, November 6,
December 1982, pp. 1210-1213
[6] E. Calkin and B. Hamilton, "Circuit Techniques for Improving the Switching Loci of Transistor
Switches in Switching Regulators," IEEE Transactions on Industry Applications, Volume IA-l2,
No.4, July/August J976.
4-16
4.2 HP-41CV Power MOSFET Thermal
Analysis Program
Introduction
MOSPOWER FETs are being used in more and more
switching circuits now that high voltage and high
current devices are available at reasonable cost. One
of the most misunderstood facts about MOSFETs is
that they are not immune to thermal runaway when
driven with gate-to-source voltages between 10 and
20 volts. Under these conditions, the drain-to-source
on-resistance (RDS(on» increases exponentially with
junction temperature and, consequently, so does the
power dissipated by the MOSFET for a given RMS
drain current.
8. Thermal resistance junction to case (Rojd
9. Thermal resistance case to heatsink (Rocs)
10. Thermal resistance heatsink to ambient (Rosa)
Regression constants "a" and "b" are used by the
program to determine the value of RDS(on) at different values of Tj. These constants can be obtained
from an exponential curve-fit to data points taken
from the normalized RDS(on) versus the Tj curve.
This curve shows RDS(on) as a function of Tj normalized to its value at 25"C, and is of the form y =
aebx . Most manufacturers now include the curve as
part of their MOSFET data sheets.
Proper heatsink design, as with bipolar transistors, is
essential. Reference 1 gives an excellent account of
thermal design for MOSPOWER FETs and presents
an accurate graphical approach for selecting a heatsink
with the correct thermal resistance.
The curve fitting can be done using the exponential
curve-fit program EXP which is included in the HP41 Statistics Pac. If this Pac is not available, then
the thermal analysis program can be instructed to
calculate constant "b" given the value of constant
"a". This method is much faster than a separate
curve-fit operation, but the accuracy is reduced
slightly. In many cases though, this reduced accuracy
is more than adequate for the thermal program to
calculate useful results.
This application note presents a program for the HP41CV programmable calculator that computes the total
dissipated power (PT) and the operating junction
temperature (Tj) of a MOSPOWER FET. The program features prompting for the ten input variables
required and fast convergence to the solution (if one
exists).
Theory
The ten input variables required are:
1.
2.
3.
4.
5.
6.
7.
The program uses two iterative algorithms to determine the points on the Pr versus Tj curve at which
the tangent line touches it and the Roja line intersects
it (refer to Figure 1). Roja is the total thermal resistance from junction to ambient and is equal to ~c
+ Roes + Rosa· By finding the point on the P'f
curve at which the tangent line touches, the range of
Tj for which a solution will occur can be determined.
This range will extend from Ta to either Tj(maximum)
Regression constant a
Regression constant b
Fixed MOSFET power loses (PI)
RMS drain current (ID(RMS»
RDS(on) at 25"C
AmbIent temperature (Ta)
Maximum operating junction temperature
(Tj(maximurn) )
4·17
_
I
and so:
I'T(W)
4.
Tj = Ta
4.
3.
3.
2.
(4)
KlabebTj
By choosing an initial value of Tj = Ta + lib, the
first iterative algorithm evaluates approximations of
Tj(tangent) according to:
20
15
,.
PI
+ lIb + --"..,.-
Tj(i)
I'T
.
..
7'
(To)
100
125
150
Ta + lib
PI
+ ---..:....,.,--,-Klabe{bTP-I)}
(5)
The convergence. condition chosen for this algorithm
is that Tj(i)-TjY?
228 GTO 09
229 RCL 11
230 STO 14
231 GTOOS
232.LBL 09
233 RCLll
234 STO 15
23S GTO 08
236+LBL 10
237 "PT="
238 ARCL 17
239 PROMPT
240 ''TJ=''
241 RCLll
2421NT
243 ARCL X
244 PROMPT
245 GT002
246 END
-
Practical Design Considerations _
Chapter 5
Practical Design Considerations
5.1 High Speed Gate Drive Circuits
Introduction
Many design aids and application notes have been
written about driving power MOSFETs directly from
CMOS or open-collector TTL logic. These schemes
are very attractive because of their simplicity (see
"Using Power MOSFET Transistors to Interface from
IC logic to High Power Loads" in Section 6.3), but
the switching times of MOSFETs driven in this manner are often too slow for many applications. This is
not by any means a result of the MOSFET having
inherently slow switching times; in fact, any power
MOSFET can be switched ON or OFF in less than
10 nsec if desired. When a MOSFET is driven by
logic, the loss of its potential high speed switching
characteristics is actually due to the limited peak output current available to charge or discharge the effective gate-to-source capacitance.
(Qg) required to change the gate-to-source voltage
from 0 to 10 volts, or vice-versa, also increases.
Logic gates, with their very limited peak output current (or iJQ/{)t) capability, are inevitably the limiting
factors if used as gate drivers for modem, large power
MOSFETs.
Additionally, a phenomenon known as the "Miller
effect" causes the effective gate-to-source capacitance
of a common-source connected power MOSFET to
assume a much larger value than its static specification
during a switching transition. This phenomenon,
illustrated in Figure 1, is well known to the designers
of analog circuits and is often used to advantage.
However, in high-speed gate drive circuits, it necessitates a substantial peak output current capability.
Since Cgs(~ff) ~ iJQiiJVgs , from Figure 1 it can _
be seen that this parameter is essentially constant at
about 700 pF below VGS = 6 volts. Cgs(eff)
The larger the die area of the power MOSFET being
driven, the larger its effective gate-to-source capacitance (Cgs(eff». Consequently, the total gate charge
;;;
!:i
400
Vos=o
w
Cgs(eH)
= 1900 pF .......... ~
!1 .....
= 700 pF _
I
3UDY
200
I;i!
z
~
c
...!'!
I
~
0
4.
o
P;w
,
I
I
W400V
I I I
I I I
111-
7 l"tO'1
L
CHARGE (ne)
400
..
RL = 10011
VGG = lOY
w
IN
~
Ii
Vos = 0
w
~ 12
g
!;! •
5 10 1520 25 30 35
Q-
TURN-OFF CHARGE
~ 16
Uhf
2OOV,-HA'
300
Cgs(etf)
[
TURN·ON CHARGE
~
.
4
~
0
I
300
I I I
II ~'"'
I II
~lJ
II
300V
\i-
200V
c~
II
051015202530
Q-
CHARGE (ne)
The Q g Versus VGS Charge Transfer Curves for
the Siliconix VNDA40 Geometry Show the Occurrence
of Maximum Effective Gate to Source Capacitance
at a VGS of Approximately 7 V
Figure 1
5-1
200
YS=400V
,
00
Rgen = 2 n, the four switching times for the MOSFET depicted in Figure 1 are as follows:
increases rapidly when the MOSFET turns on or off
at VGS = 7 volts, as indicated by the approximately
horizontal portions of the tum-on and tum-off VGS
versus Qg charge transfer curves. Above VGS = 7
volts, Cgs(eff) returns to a lower constant value of
approximately 1900 pF. One can see that, due to the
large capacitance in the "horizontal" region of the
curves, a relatively large amount of charge is required
to change VGS by only a fraction of a volt.
td(on) =
tr
td(off) =
tf
=
ON
10.2 nsec
4.3 nsec
4.3 nsec
For a low value of Rgen, we see that the MOSFET
will switch very rapidly. However, a low value of
Rgen also means that the peak MOSFET gate current
can be large during the switching transition. Increasing
the value of Rgen reduces the peak gate current drawn
by the MOSFET but degrades the switching speed
of the device. It is up to the designer to trade off
switching speed against peak MOSFET gate current
for a particular application.
It is difficult to quantify the actual value of C gs (eff)
in the "horizontal" region of the charge transfer
curves because it varies with VDS. In practice, one
really doesn't need to know the value of Cgs(eff) in
the "horizontal" region anyway. The curves shown
in Figure 1 provide enough information to accurately
determine how fast a MOSFET will switch when
dri ven by a particular circuit.
OFF
1.7 nsec
OFF
ON
Vos
Vos
90%
90%
VGG
'---+-1::1--
Vg,
Vg '
0 9 1 0 92
Og3
0 9 209 1
Qg3
TURN-ON
TURN-OFF
This Circuit May Be Used to Model the Current Source Driver
Circuits Such as the Siliconix 0469 IC
Figure 4
Switching Time Equations
Q,l
ld(on) ""
Vgi
t.
2-
Qg
QgJ
If = - y
y
Rgen In
g2 -
Qg3 - Qg2
VGG
)
Rgen 1n \VGG - Vgl
gl
(I)
(VGG
- Vgl)
- y - - y - (2)
GG -
g2
(VGG)
td(off) = VGG _ Vg2 Rgen In
Qg2 - QgI
(Y,,)
Vg2 - Vgl
Ygi
If=--Rgenln
Vg2
-
(3)
Certain gate drive circuits will appear to a MOSFET
not as a voltage source with a series output resistor
Rgen, but rather as a current source (or sink). This
is the case for both the 0469 and OS0026 IC drivers
to be discussed later. Figure 4 shows a circuit that
can be used to model these two devices. If the peak
output capability of a current-source type driver is
known, the total tum-on or tum-off time for a power
MOSFET driven by it can be very easily determined.
(4)
These Four Equations Enable One to Calculate the Switching
Times of a Power MOSFET, Based on Classical Circuit
Analysis of the Driver Circuit Model Shown in Figure 3
Figure 2
Rgen
VGG
t
v
-=
1
,J
By definition:
09
Cgs(eH) = d Vgs
~t
~Q
= -1-
(5)
If the MOSFET depicted in Figure 1 is driven by a
current-source driver with a 3 A output sink or source
capability, the total tum-on or tum-off time, assuming
a VGS swing of 0 to 10 volts and vice versa, is:
This Simple Circuit Can Be Used to Model a Voltage Source
Type Gate Drive Circuit
Figure 3
25 nC
ton (or toff) = ~ = 8.33 nsec
In Figure 2, the pertinent equations for determining
the four switching times td(on), tr , td(off), and tf
are given, based on classical circuit analysis of the driver
circuit model shown in Figure 3. In order to minimize
the switching times, it can be seen that the output
resistance Rgen must be made as low as possible.
Doing this also increases the immunity to possible
mode 1 (aVOS/at) breakdown (reference Section
5.4). As an example, if we let VGG = 10 V and
This is a very fast switching time for any power
MOSFET regardless of the application, but in many
cases, more moderate switching times and hence
lower gate currents will suffice.
Let's examine three groups of high-speed gate drive
circuits suitable for driving most of the larger power
MOSFETs available when the required switching
times are less than 50 nsec.
5-2
draws with one output at zero volts (30 mA) means
its average power dissipation at low MOSFET duty
cycles will be quite large (as high as 0.5 watts).
Additionally the DSOO26 is configured as an invertingonly driver and requires two external TIL inverters
when non-inverting drive is required.
Direct Coupled Drive Circuits Using ICs Alone
Until now, the most popular IC for driving MOSFETs
in high speed switching applications has been the
DS0026 MOS clock driver made by National Semiconductor. This dual-channel bipolar IC can sink or
source a relatively large peak output current (1.5 A)
and has inherently small propagation delays within
its internal circuitry. It also operates from a wide
supply voltage range (22 V).
As an alternative, the D469 CMOS quad power
MOSFET driver from Siliconix can be used to replace
the DS0026 in just about all but the higher speed
switching applications. Each driver in the D469 may
be configured as being either logically inverting or
non-inverting and can sink or source a peak current
of 0.5 A. To obtain greater peak output current capability, necessary when driving large capacitive loads,
each of the D469s outputs may be connected in
parallel if desired.
Unfortunately, the DS0026 has some drawbacks
which prevent it from being the "perfect" IC driver.
Since the DS0026 requires a large logical "1" input
current (10 mA), it is unsuitable for use in applications
where the control logic is LS TTL or high-speed
CMOS. The large quiescent supply current the device
1 DS0020
+12 V
+60Y
46.11
NON-INDUCTIVE
o 1 fLF CERAMIC
100pF
BNe
IRF
243
INPUT
470n
DSOO26
6211
20469
+12V
+60 V
...
~
46U
NON-INDUCTIVE
I.
~----------~l)T~
13
BNe
IRF
243
12
INPUT
0469
11
I.
SOil
2,4,6,7,9
The Schematic Diagram of the Test Fixture Used to Evaluate
the SWitching Performance of the DS0026 and 0469 Shows
Identical Load Conditions for Each Device
FIgure 5
5·3
l1:li
Both the DS0026 and D469 were incorporated into
a test fixture driving identical power MOSFETs with
identical loads so that their switching performance
could be compared. Figure 5 shows the schematic
diagram of this fixture. The power MOSFETs used
were Siliconix IRF243s with a maximum static Ciss
of 1600 pF and a maximum static Crss of 300 pF.
44 nsec, respectively, while the DS0026s were 26
nsee and 42 nsee. It appears that the DSOO26 was
somewhat faster (by 10 nsec) on the rising edge of
its waveform than the D469, but both had very similar
fall times.
Some further tests were conducted to see how the
D469, with two or more outputs connected in parallel,
compared to the DSOO26 with both of its outputs
connected in parallel.
The load resistance for both MOSFETs was approximately 4.6 ohms (non-inductive), and the power
supply voltage was 60 V. Both the DS0026 and the
D469 were powered from a separate 12-volt power
supply and driven by a TTL level pulse train. The
pulse train had a pulse width of 150 nsec and a
repetition rate of 1 kHz.
DSOOJv--
V--
01&
r
,
/7
r'== 'Y
~
~ 7'
0469/J
'l
J
0469
-
/
/
This Graph Shows the Gate Waveform Rise Time for 2
Paralleled DS0026 Outputs Versus 4 Paralleled 0469 Outputs
(Vert: 2.4 V/Dlv. Horlz: 20 ns/Div)
Figure B
J
This Graph Shows the Gate Waveform Rise Times for 1
DS0026 Output Versus 2 Paralleled 0469 Outputs
(Vert: 2.4 V/Div, Horiz: 20 ns/Div)
Figure 6
'"
~
~
\\
........;
0469
1\\
1\
\
DSj"\ D\ I>..,;
'"
t\
~
-
The Fall Times of the Gate Waveforms for 2
Paralleled DS0D26 Outputs Versus 4 Paralleled 0469 Outputs
(Vert: 2.4 V/Dlv, Horiz: 20 ns/Dlv)
Figure 9
040.
DS0026~ \\
Figures 8 and 9 show the rise and fall times of the
gate waveforms for the D469 with four outputs in
parallel versus the DS0026 with two outputs in parallel. Here it can be seen that the DS0026 with rise
and fall times of 22 nsec and 30 nsec, respectively,
was still the faster of the two - but not by much.
The D469 came in a very close second with rise and
fall times of 32 nsec and 42 nsec, respectively. Interestingly enough, the rise and fall times of the D469
with four outputs connected in parallel were virtually
the same as those with only two outputs in parallel.
The rise and fall times for the DS0026, on the other
hand, showed some improvement.
The Fall Times of the Gate Waveforms for 1 DS0026
Output Versus 2 Paralleled 0469 Outputs
(Vert: 2.4 V/Dlv. Horiz: 20 ns/Dlv)
Figure 7
Since the DS0026 contains two independent drivers
and the D469 has four, a fair test would be to compare
the two devices as dual drivers. Thus, two D469 outputs were paralleled and compared to a single DSOO26
output in drive capability. Figures 6 and 7 indicate
that on a "per chip" basis, the D469 was essentially
the equal of the DS0026. To be precise, the D469
gate waveform's rise and fall times were 36 nsec and
5·4
The main difference observed between the D469 and
the DS0026 was the propagation delay-time each
device exhibited. For the D469 this parameter was
approximately 50 nsec while that of the DS0026 was
in the range of 5 to 10 nsec. Figure 10 clearly show~
this difference. It should be noted that Figure 10 is
the only one in which the difference in propagation
delay is shown. Figures 6 through 9 do not show
this because it is easier to compare switching waveforms when the oscilloscope traces are overlaid with
one another.
/
/
'r---- I(
J
050026
--../
/""
I
r--
/0469
l
When it is desired to drive very large single chip or
multichip power MOSFETs with switching times in
the range of 50 nsec to 20 nsec, a current booster
stage may be added to either a D469 or a DS0026
with no loss of switching speed. Figure 11 shows
how a complementary MOSFET booster stage can
be connected to one output of a D469. The additional
output buffer functions in exactly the same manner
as a CMOS logic inverter, and its output voltage
swings all the way to either power rail. Current is
only drawn by the buffer during switching transitions;
consequently, the average power it dissipates is proportional to the switching frequency. The buffer may
also be used by itself if the TTL input level compatibility of the D469 is not required.
'\
VDD
\
1\
\
...."'
"...../
J.~1
J
J
~~
This Graph Clearly Shows the 50 ns Difference in
Propagation Delay Between the DS0026 and the 0469
(Vert: 2.4 V/Div, Horiz: 20 ns/Div)
Figure 10
MOSFET BEING
DRIVEN
I
~
NOTE: 01 AND 02 COMPRISE 1/2 A VQ7254 POWER MOSFET ARRAY
A Simple CMOS Current Booster Stage Can Be Added
to a 0469 Output to Increase Its Output Capability,
or May Be Used on Its Own
Figure 11
In summarizing the results of this comparative test,
we can say that the D469 is really a DS0026 in a
"CMOS disguise." Although the DS0026 was faster
than the D469 in both cases, its speed advantage was
no more than about 10 nsec in any of the rise and
fall time measurements. As a two-channel driver the
D469 is the obvious choice, especially when one
considers its low power consumption compared to
the DS0026. In the test fixture shown in Figure 5,
the DS0026 drew an average supply current of 30
rnA and thus consumed about 360 mW of power
while the D469 drew an average current of slightly
more than 2 rnA and consumed about 25 mW of
power. Needless to say, the DS0026 ran quite hot to
the touch while the D469 was cold. The 50 nsec
propagation delay of each driver in the D469 should
not pose any problem at all as long as the MOSFETs
in a particular circuit are all driven by D469s.
Depending on the MOSFETs used in the buffer stage,
the available output current can be as high as several
amps. For the VQ7254 dual complementary pair
shown in Figure II, the pulsed output current specification is three amps. This is a six-fold improvement
over the 0.5 A peak output sink/source, capability of
each channel of the D469 alone.
For very high speed applications where MOSFET . '
switching times of less than 20 nsec must be achieved,
both small internal propagation delays and high peak
output current capability are the necessary characteristics of a gate driver circuit. Figure 12 shows a
circuit which works very well with large capacitive
loads. When the input to the driver is a logical "0,"
QI is held in conduction by one half of the DS0026,
and Q2 is clamped off by QI. When a logical "I"
input occurs, QI is turned off and a current pulse is
applied to the gate of Q2 by the other half of the
DS0026 through TI. After about 20 ns, TI saturates,
and Q2 is held on by its own egs and the bootstrap
circuit comprised of CI, DI and RI. For pulses less
than 50 ILsec wide, the bootstrap circuit may not be
needed because the input capacitance of Q2 discharges
very slowly. At the end of the logical "I" input
pulse, QI turns on, shutting off Q2.
Direct Coupled Drive Circuits Using
Discrete Components
If the IC gate drivers described above begin to become
the limiting factor in the performance of MOSFET
switching circuits that use them, it is inevitably the
result of two factors:
(1) The peak output current capability of the IC is
not high enough.
(2) The device has inherent switching times which
are too slow.
5-5
PERFORMANCE SUMMARY
V+
'-'"F
5011000pF
20
15
225
7
15
4700pF
20
15
225
1fi
16
15.5
511
20
15
23
11
14
45
SU
50
20
22
18
14
1IX1RG2231U
TERMINATED
IN51 n
20
15
20 5·
19
13·
V+ +
1 - 0 001 ",F
1-0.01 ",F
} PARALLEL
IN914
0,
'000 pF
65
C,
• MEASURED AT INPUT TO CABLE
100pF
Sl
PWM
.
,
c>---+---A/IIv--4
5 V...
,/2
OS
0'"
T1 -IS THREE TURNS 30 BIFILAR ON A FERRITE BEAD
A Circuit Capable of Driving Very Large Capacitive Loads Is
Shown in This Diagram Along with Its Tabulated
Performance Under Varying Load Conditions
Figure 12
If a wide range of duty cycles must be accommodated
then the MOSFET gate will be overdriven at low
duty cycles and underdriven at high duty cycles. This
can lead to variations in the total MOSFET switching
time as is evident from equations (1) to (4). However,
for those applications where only a moderate variation
in duty-cycle is to be accommodated, this simple
circuit can work very well.
One can see that the switching performance of the
above mentioned driver is extremely good with large
capacitive loads, and it is suitable for driving even
the largest currently available power MOSFETs.
ffigb Speed Isolated Drive Circuits
Many switching applications involve driving power
MOSFETs from control logic or other circuitry that
is ground referenced. The question that inevitably
arises is: how does one drive power MOSFETs whose
sources are not ground referenced? The solution, of
course, is to use some form of DC isolation between
either the control logic and the driver or between
the driver and the MOSFET(s).
VDD
~VSl. .
rt/4'"7OIT,lqJ
F
D469
DRIVEN
MOSFETBEING
°
~
"':"
There are basically two viable methods that can be
used to provide DC isolation between a MOSFET
and its associated control circuitry when necessary.
These are:
"':'"
a) This Simple Transformer Coupled Circuit Provides Isolated
Drive for the Power MOSFET
1---.
I
I
(I) Transformer isolation
(II) Opto-coupler isolation
I
'
VGS
o~~~--+-4--'----------~
Each method has its own inherent advantages and
disadvantages as far as performance/cost trade-offs
are concerned. We will examine them in more detail.
b) The Resulting Gate Voltage Can Be Seen to Be an
Asymmetrical AC Waveform, Whose Amplitude Varies with
Duty Cycle
Figure 13
(I) Transformer isolation
Figure 13(a) shows a simple transformer drive circuit
that effectively isolates the ground referenced D469
from the n-channel MOSFET with a floating source.
Unfortunately, this circuit has a major problem: the
AC waveform across the primary winding can be
asymmetrical. This follows from the fact that the
average of the transformer winding volt-seconds
product must equal zero as shown in Figure 13(b).
Therefore, the gate enhancement voltage of the
MOSFET will vary with duty cycle, being greater at
low duty cycles and smaller at high duty cycles.
There are transformer isolated drive circuits where
VGS does not vary with duty cycle. Figure 14 shows
a transformer coupled drive circuit in which the primary winding sees only a symmetrical AC waveform.
VGS is fixed by the voltage swing at the output of
the D469 drivers and the turns ratio of the transformer.
The AC voltage swing across the primary of the
transformer can be seen to have a peak to peak value
of 2 VDD since one D469 output is at 0 volts while
the other is at VDD and vice-versa.
5·6
PRIMARY
DRIVE
CIRCUITRY
II
This Transformer Coupled Drive Circuit Generates a
Symmetrical Gate to Source Voltage Waveform for Each
MOSFET, Whose Amplitude is Independent of Duty Cycle
Figure 14
A Bipolar Transistor and Its Associated Components Reduce
the Effective Drive Impedance Presented to the Gate
Terminal of the MOSFET
Figura 15
C 1 is added to the circuit to prevent a DC component
from appearing across the primary of T 1 due to a
potential lack of symmetry in the switching times of
the two drivers used. The value of Cl is determined
by the magnetizing current of Tl. It should be large
enough so that the voltage drop across it due to the
magnetizing current during one half a cycle is small
compared to VDD.
adding a bipolar transistor Ql and its associated components as shown in Figure 15, these problems can
be overcome. At turn-off, Ql is driven on by the
energy in Tl, clamping the gate of Q2 to its source.
Even after all the energy in Tl is discharged, Ql
still presents a relatively low impedance to the gate
ofQ2·
(II) Opto-Coupler Isolation
Of course, current booster stages may be added to the
D469 outputs in Figure 14 to increase the drive current to the primary winding and hence the gates of
the MOSFET(s). Additional isolated windings may
also be added to Tl so that totem pole MOSFETs
in bridge circuits may be driven as illustrated in
Figure 14.
Figure 16 shows a different method of providing DC
isolation in a totem pole MOSFET "H-bridge" circuit
that uses n- and p-channel devices. Here a D469
driver is referenced to the positive power supply rail
and an opto-coupler is used to provide DC isolation
between it and the low voltage control logic. A - 12
volt power supply, referenced to the positive rail, is
generated for the D469's ground pin with a zener
diode, capacitor, and resistor.
Often the impedance of the secondary winding(s) that
drive the MOSFETs may well be high enough to
cause mode 1 (lJ VDStat) triggering problems. By
112 DUAL
OPTOCOUPLER
.-"-..
"r.o
LE~~
OR~ ~
.
1
1/40469
NOTE D IS A 12 VOLT ZENER
This Schematic Diagram Shows a Different Method of
Providing Isolation - Floating a D469 Driver and Optically
Coupling the Logic Drive to the Upper MOSFETs
Figure 16
5·7
~~E.
-
The 0469 in this example provides gate-drive voltages
for the p-channel MOSFETs that swing from the
positive rail to 12 volts below it. The reservoir capacitor (C) must be large enough to ensure that sufficient
charge is available for transfer to the gates of the pchannel MOSFETs.
References
[1] Rudy Severns, "Using the Power MOSFET as a
Switch," Intersil Applications Note A036, 1981.
[2] Mark Alexander and Jim Harnden, "The D469:
An Optimized Quad Driver for MOSPOWER FET
Switches," Siliconix Applications Note AN8311, 1983.
Here again a complementary MOSFET current booster
stage can be added to the floating D469 as well as
to the ground referenced one to increase the output
current capability.
5-8
5.2 Temperature Compensated Biasing for Power
MOSFETs in Linear Amplifiers
Introduction
tial drain current (and hence gate-drive) is required,
the drain current decreases as the junction temperature increases, causing the device power dissipation
to decrease as well. If adequate heatsinking for the
power MOSFET is provided then thermal runaway
will not occur.
Many of the highest performance audio amplifiers
available today use power MOSFETs in their output
stages. The advantages of using power MOSFETs
instead of bipolar transistors in an amplifier output
stag,.e have been well described in numerous artic1esL1,2,3]; however, important thermal characteristics of MOSFET output stages are often overlooked.
The purpose of this application note is to familiarize
the designer with these characteristics and to show
how to compensate for them when necessary.
Linear amplifiers, on the other hand, generally require a certain value of quiescent current in the output
stage to minimize crossover distortion. This value of
quiescent current is normally made less than 100 rnA
per device to minimize power dissipation with no
output signal. The problem here is that the quiescent
gate-to-source bias voltage of each MOSFET may
only be a few tenths of a volt greater than its threshold
voltage (VT). Under these conditions, the temperature coefficient of drain current is positive, and if not
compensated for, can result in thermal runaway in the _
output stage.
It is a misconception that power MOSFETs have a
negative temperature coefficient of saturated drain
current at all values of gate-to-source voltage. Figure
1 shows that for large gate-drive voltages (six volts
and above) the temperature coefficient of a power
MOSFET is negative. In applications where substan-
VDS~15V
1//
-55"1V V
25°C
5
2.
1.5
ZERO
TO
POINT
1.
~
~
4
47
An obvious solution might be to bias the MOSFETs
in an output stage at the zero TC point indicated in
Figure 1. This would be acceptable if the zero TC
point occurred at a drain current value within the
required quiescent bias current range of interest.
Unfortunately, since most power MOSFETs exhibit a
zero TC point at a gate-to-source voltage of between
four and seven volts, the corresponding drain current
(depending on the device) can be as high as several
amps. Such a large value of quiescent bias current is
unacceptable because the power dissipation at idle
would then be very high. The output devices would
inevitably fail due to excessive junction temperature
if called upon to deliver any appreciable power to a
load.
IIV
/; /"
1~
5
/
150"C
1.
VGs(V)
The Saturated 10 Versus VGS Transfer Characteristics
of an IRF521 n-channel OMOS Power FET Clearly Show
the Occurrence of the zero TC Point at a VGS of
Approximately 4.7 V
Figure 1
5·9
100
Figure 2 shows a typical complementary MOSFET
source-follower output-stage and its associated bias
and driver circuitry. The biasing network is generally
an adjustable DC voltage source connected between
the bases of the complementary Darlington bipolar
transistor driver stage. If this bias voltage generator
is fixed with respect to the junction temperature of
the MOSFETs, their bias currents will rise as they
heat up. As the bias currents of the MOSFETs rise,
however, the quiescent power dissipation increases
and causes the MOSFET junction temperatures to
rise even further. This is clearly an unstable situation
and must be avoided in a viable amplifier design.
v++
v++
SILICONIX
1
+--'IIIfV-
"G
v--
V
..
/
30
20
/
,.
zo
l'L
.
80
80
100
120
Trel
Figure 3
ZL
provide a reasonably accurate description of the IN
characteristics. For VDS > Vas - VT and VQS >
VT, the drain current of a MOSFET is given byL 4J
DEVICE
v--
i
~ 5.
v+
BACK
/
80
~~1
I
I
I
I
I
I
I
I
I
I
I
V
70
9
V
/
IRF521
80
JI~".~,
INPUT AND
GAIN STAGES
OF AMPUFIER
/
90
v-
ID(SAT) =
A Commonly Used Driver and Output Stage Configuration In
MOSFET Audio Power Amplifiers
Figure 2
~
(1)
(Vas - VT)2
where
This paper investigates the physical phenomena that
govern the behavior of power MOSFETs with temperature and presents two simple biasing circuits that
effectively compensate for the positive temperature
coefficient of drain current at low bias levels.
( . 1 - ILCoxW
..
,.. - -L-'- (the trans1stor gam)
and
Theory
IL
carrier mobility
{cm2N.sec}
The experimentally observed temperature dependence of saturated drain current in a power MOSFET
;s illustrated in Figure 3. Clearly, under low bias
conditions the temperature coefficient of drain current is positive. In order to compensate for such a
temperature dependence, we need to know which
electrical characteristics of a MOSFET vary with
temperature and which ones are dominant at low
valuesofID·
Cox
gate capacitance per
unit area
{F/cm2}
W
total gate width
{cm}
L'
electrically effective
channel length
{cm}
threshold voltage
{V}
VT =
Despite their structural differences, the conventional
planar MOSFET and a DMOS power FET exhibit
similar electrical behavior in the saturated region of
operation. Consequently, conventional device models
It is important to accurately determine the true
threshold voltage of a MOSFET in order to obtain
agreement between equation (1) and the actual device
characteristics .
5·10
Most manufacturers specify a convenient, fixed value
of drain current (usually 1 rnA) at which the "threshold voltage" is measured. True threshold, however, is
measured at a current proportional to the transistors
W/L ratio[51, and it is often substantially larger than 1
rnA. Consequently, the threshold values obtained by
measurement of V GS at an ID of 1 rnA and at a value
proportional to W/L can differ by as much as 0.5 volts.
that the temperature coefficient of drain current is as
follows[7];
The graph of JID(SAT) versus VGS shown in Figure
4 provides an accurate method for determining VT as
well as 13[6]. The straight line portion of the graph
corresponds to the "square-law" operating region
described by equation (1) and has a slope of 13/2 and
an x-intercept equal to the true threshold voltage of
the MOSFET. One can see that the equation closely
agrees with measured data over a wide range of drain
current, including the bias range of most audio power
amplifiers.
(2)
tempco (ID) =
aID (SAT)
aT
= IIDI
2
aVT)
( 1 aIL
~ aT - (VGS-VT) aT
From device physics, it can be shown that over the
temperature range of interest[4,8]
a) for an n-channel MOSFET,
aIL
aVT
.
• - and - - are negative
aT
aT
• (VGS-VT)ispositive
Yiii(SjiT)
ID(SAT) (A)
(A112)
- 1A
0.950
0.900
f
D.'"
0700
aIL
.
• - and (V GS - VT) are negative
aT
I
O.BOO
0.750
b) for a p-channel MOSFET,
'""'" t--
II
SlUCONIX
IRF521
aVT
• - - is positive
aT
05 A
0650
I
0600
0550
/
05DD
I
SLOPE
=~
The two terms in equation (2) can be seen to have
opposing signs for both n or p-channel devices.
(083 A/V2)
I
0450
At low values of the effective gate drive (VGS - VT)
and correspondingly low values of ID(SAT), the
second term in equation (2) is dominant, and the
tempco is positive .
7
0.350
.7
0.300
.
0250
100mA
II
In conclusion, at low values of (V GS - VT) the temperature coefficient of drain current is proportional
to the temperature coefficient of threshold voltage.
Figure 5 shows the experimentally determined _
threshold variation with temperature of a power
MOSFET. Typically this change in VT is in the range
.j
0.200
0150
lSmA
II II
0.100
. ...
0050
2.0
2.2
2.4
lOrnA
~
II
2.6
28
1 mA
3.0
TRUE Yy
3.2
3A
36
3.8
40
42
4A
= 2.92 V
2.
VGS(V)
A Method to Determine the True Threshold VT (and Transistor Gain B/2) from the Saturated Drain Current Transfer
Characteristics (@ VDS = 20 V)
Figure 4
,
SILICONIX
IRF521
........
2S
~
'" "'"
27
By differentiating equation (1) with respect to temperature, some insight can be gained into the temperature dependence of saturated drain current. The two
terms in this equation which exhibit a variation with
temperature are the carrier mobility and the threshold voltage VT. Differentiating equation (1), we see
W""-3SmVI"C
"'-
~
2.
20
40
60
..
TEMPERATURE T (OC)
Figure 5
5-11
""I'
100
120
y++
of -2.5 to -6 mV/o e[4,9]. To achieve temperature
compensated biasing then, the effective gate drive
(Vas - VT) must remain nearly constant. This can
be realized by controlling Vas so that it is varied by
an amount equal to the change in threshold voltage
with temperature.
We now examine two practical circuits that provide
output stage bias stabilization by reducing the MOSFET gate to souree voltages as their junction temperatures increase.
INPUT AND
GAIN STAGES
OF AMPLIFIER
r
I
Ir A~UST
T
Compensated Biasing Circuits
I
I
r
r
There are basically two means of providing thermal
bias compensation in a linear amplifier. The first
method is to electrically sense the output stage idle
current and use some form of feedback control circuitry to ensure that the bias remains constant. The
second method also uses feedback control, but instead of sensing the idle current, the case temperatures (Te) of the MOSFETs are sensed by a thermal
sensor, and the quiescent gate bias is reduced accordingly as Te increases. The first method works well to
keep the idle current in a MOSFET output stage
constant, but it often degrades the electrical performance of the amplifier. What one achieves is a very
stable amplifier over its operating temperature range;
however, the resulting distortion specifications or
output volta.&e swing will very likely be
unacceptable[lUJ.
y--
y--
y-
NOTE Q4. 08 AND Qg ARE MOUNTED ON A
COMMON HEATSINK
Superior Bias Stability Can be Obtained In a Complementary
Source-Follower Output Stage Using Dual VBE MuHlpliers
Figure 6
with temperature divided by the variation in baseemitter voltage with temperature. This multiplication
factor is typically greater than one.
The other VBE multiplier generates the difference
between the voltage of the upper VBE multiplier and
the required DC bias voltage between the bases of
the emitter-followers. By adjusting the VBE multiplication factor of the upper multiplier, adjustable bias
for the MOSFET output stage is achieved.
The second method, on the other hand, in no way
electrically interferes with the drain-to-souree current paths of the MOSFETs in the output stage. As
well, it can provide comparable bias stability to the
first method, if implemented properly.
For this circuit, the total gate bias for the MOSFETs
is given by
Two simple, thermally compensated biasing circuits
for MOSFET power amplifiers using either complementary source-follower output stages or complementary common-source output stages will now be
described. Both circuits take advantage of the - 2.2
mvr C variation in base-emitter voltage of a bipolar
transistor to decrease the gate bias voltage of the
MOSFETs in an output stage as they heat up.
VBIAS == (1
+ R3/R4) VBE(Q4)
+ (1 + R1/R2) VBE(Q3) - 4VBE
(3)
and
iJVBIAS
aT
Bias Stabilization of the Source-Follower Output
Stage Using Dual VBE Multipliers
If (1
Figure 6 shows a complementary source-follower
output stage driven by complementary darlington
emitter-followers that have two VBE multipliers in
series between their bases. One VBE multiplier
transistor (Q4) is thermally coupled (Le., glued) to
the heatsink on which the MOSFETs are mounted.
R3 and R4 are then adjusted to provide a VBE
multiplication factor equal to the total threshold
voltage variation of the nand p-channel MOSFETs
=
(1
+ R3/R4)
aVBE(Q4)
aT
(4)
+ R3/R4) == 3, then
aVBIAS == -6.6mVloC
aT
This decrease in MOSFET gate bias voltage with
temperature effectively compensates for the decrease
in VT of both the nand p-channel devices with
temperature.
5-12
Bias Stabilization of the Common-Source Output
Stage Using a Temperature-Compensated Current
Sink
This effective change in p-channel MOSFET gate to
source voltage compensates for the change in its VT
with temperature (which is positive, not negative).
Another means of biasing a complementary power
MOSFET output stage is shown in Figure 7. The
circuit uses a fixed gate resistor and an adjustable
current sink to provide variable bias for the p-channel
MOSFET. Transistor Oz is thermally coupled to the
heatsink on which the MOSFETs are mounted. As
Oz heats up, its VBE decreases and causes the current
drawn through the collector of 01 to decrease as well.
This reduces the gate drive voltage of the p-channel
MOSFET as it and the n-channel device heat up. The
op-amp automatically controls the n-channel MOSFET to mirror the bias current of the p-channel
device, which keeps the output centered at zero volts.
Conclusions
.
v+
It has been shown that in linear power amplifiers,
where the MOSFET bias current levels are low, the
temperature dependence of VT is the dominant factor
in the variation of ID with temperature. By compensating for this effect alone, a high degree of thermal
stability over an uncompensated bias arrangement
can be achieved.
Figure 8 compares the temperature stability of drain
current for the two biasing circuits described with the
uncompensated thermal characteristics of the n-channel MOSFET used. Both of the bipolar transistor
VBE referenced compensation circuits are extremely
effective in reducing bias sensitivity to temperature .
v+
,
275%
1
IBIAS::: 50 mA
..00/0
II
225%
I
~~~TB~~~ Iy+ OUTr>-~-'VII'r--r-_ _--1I--_VO;:;:UT:.:.,
CIRCUITRY
I v-
200%
/
ZL
175%
t
~
lofT} X 100%
~
04
n-CHAHNEL DEVICE
150%
/
VUNCOMP~NSATED
/
/
I
v-
v-
125%
NOTE: Q2. 03 AND 04 ARE MOUNTED ON A
COMMON HEATSINK
This Complementary Common-Source Output Stage Uses a
Temperature Compensated, Adjustable Current Sink for
Bias Stability
Figure 7
100%
.......
I
v-
IDEAL
~
/
75%
-r----::::
-
CURRENT SOURCE
VBE MULTIPLIER
In this circuit, the gate bias ofthe p-channel MOSFET
is given by
VGS(P) = -RZ I 1 + VBE(OZ)
R1
(5)
20
40
60
80
T(OC)
'00
'20
Figure 8
and
aVGS(p)
aT
IfRZ/Rl
_ RZ x aVBE(OZ)
R1
aT
(6)
These circuits provide an inexpensive but necessary
solution to the problem of bias stabilization in a
MOSFET power amplifier without compromising
distortion or output voltage swing.
== 1.5, then aVGS(p) == +3.3mV/oC
aT
5-13
-
References
[1] Robert R. Cordell, "A MOSFET Power Amplifier With Error Correction," presented at the
72nd convention of the Audio Engineering Society, October 1982 (AES preprint).
[6] Richard S. Muller, Theodore L. Kamins, "Device Electronics for Integrated Circuits," John
Wiley; New York, 1977, pp. 360-361.
[7] William N. Carr, Jack P. Mize, "MOS/LSI Design and Application, "McGraw-Hill; New York,
1972, pp. 87-88.
[2] Erno Borbely, "High Power High Quality Amplifier Using MOSFETs, " Wireless World, March
1983.
[8] Edward Yang, "Fundamentals of Semiconductor
Devices," McGraw-Hill; New York, 1978, p. 23.
[3] Tohru Sampei, Shin-ichi Ohashi, "100 Watt Super
Audio Amplifier Using New MOS Devices," IEEE
Trans. Consumer Electronics, Volume CE-23 ,
August 1977.
[9] William Penney, Lillian Lau, editors, Staff of
American Microsystems Incorporated, "MOS
Integrated Circuits, Theory, Fabrication, Design
and Systems Applications of M os LSI," Krieger
Publishing; Malaban FL., 1972, p. 84.
[4] Paul Richman, "MOS Field Effect Transistors
and Integrated Circuits, "Wiley: New York, 1973,
pp.84-86,pp.124-133.
[10] Bill Roehr, "A Simple-Direct Coupled Power
MOSFET Amplifier Featuring Bias Stabilization," IEEE Trans. Consumer Electronics, Volume CE-28 No.4, November 1982, pp. 546-552.
[5] Paul Richman, "Characteristics and Operation
of MOS Field-Effect Devices," McGraw Hill:
New York, 1967,p.142.
5·14
5.3 Parallel Operation
of Power MOSFETs
(TA 84-5)
Introduction
Motivation for Paralleling
There are many reasons for operating power MOSFETs in
parallel, either as multiple die on a common substrate or as
individually packaged devices. Compared to paralleled bipolar
junction transistors (BJTs), paralleled MOSFETs present
fewer problems, require less derating, and provide higher
performance. Parallel operation, however, is still not trivial.
Problems can and do occur. To parallel devices successfully, the
designer must understand both the causes and the cures for the
problems. Fortunately, most of the cures are simple and are
more a matter of attention to detail than of exotic or complex
design.
There are many possible reasons for paralleling multiple
devices:
This discussion attempts to identify all the problems that users
potentially could experience and provides an explanation of
each problem. Then it will tell how to recognize these problems
and will give practical means for eliminating them.
•
•
•
•
Lower RDS(on)
Lower circuit inductance
Improved thermal performance
Compensation for radiation effects
•
•
•
•
Lower cost
Redundancy
Higher ID
Derating
In switching applications, the majority of the power dissipation is due to RDS(on). If RDS(on) is made smaller, the
power loss will go down. RDS(on) for a given device is
determined by the active die area and the breakdown voltage
(BV DSS). RDS(on) decreases more or less linearly with _
increasing die area and increases exponentially (factor ~ 2.5)
Although the discussion is quite extensive, the final conclusions will show that all of the problems may be easily
avoided and that paralleling MOSFETs is reasonable and, in
many cases, highly desirable. However, because of the lack of
prior, in-depth discussion of paralleling MOSFETs, it is
necessary to substantiate this contention.
with BVDSS. In theory, RDS(on) may be made as small as
desired simply by increasing the die area, but the cost per unit
area of the die increases exponentially when the die dimensions
are greater than about 0.125" x 0.125". This increase in cost
puts a practical upper limit on die size. Presently this is in the
range of 0.25" to 0.30" square. In addition, as die are made
larger, fewer and fewer package choices are available. When
RDS(on) must be reduced even further, paralleling is the best
alternative.
Throughout the discussion, the object will be to maintain a
balance between simple, practical advice and reasonable,
theoretical explanations.
Many designers experienced in BJT applications have a well
merited aversion to paralleling devices unless it is absolutely
necessary. While this is certainly appropriate for BJTs, such
thinking is a distinct handicap when using MOSFETs because
useful opportunities for improved performance or reduced cost
may be missed. It is important to keep an open mind and to read
carefully the following arguments.
Because of the rapid increase in cost of the larger die and the
restriction on suitable packages, there are often compelling
reasons to use smaller die in cheaper packages even when a
suitably low RDS(on) device is available in a larger die. For
example, a size five die (0.25" x 0.25") will have one half the
5-15
RDS(on) of a size four die (0.18" x 0.18"). However, the size five
die will not fit in a TO-220 package. It requires a TO-3. At
current prices, the cost of two TO-220s would be twenty to
twenty.five percent less than one TO·3. As prices drop further,
production increases and competition grows, this price differential should grow wider.
(io
DRAIN
LO
V1
+
It has been shown (1) that paralleling is an effective means for
----I
reducing the junction-to-heatsink thermal impedance (Rejs)'
To use the previous example, Rejs for a size five die in a TO-3
package is about 1.2°C/W; for a size four die in a TO·220
package, Rejs = 1.3°C/W; but two TO-220's in parallel would
reduce the effective Rejs to 0.65°C/W. For the same ID(rms)
and heatsink, RDS(on) will be lower because Tj will be lower.
This may aHow a further reduction in die size or as an alternate,
the heatsink may be made smaller thus reducing its cost. The
thermal design issues are treated in detail in Reference (1). For
high power pulse applications, the limitation may be on
GATE
V2
VOS
VOO
LS
+
SOURCE
Mechanism Responsible for Internal Voltage
Spiking at Turn-off
ID(peak) rather than RDS(on)' Again paralleling provides a
means for increasing the allowable peak current. Very often in
pulse applications, rapidly rising (high dildt) current waveforms are required. Sometimes the limiting factor on speed is
the package inductance. One means of reducing this inductance is to parallel several smaller devices so that the total
effective inductance is decreased in proportion to the number of
devices paralleled. Also smaller packages will tend to have
lower inductances.
FIGURE 1
Concerns about Paralleling Devices
When considering a design using parallel FETs, a number of
concerns and possible problems arise. Generally these concerns falI into four categories:
1. Steady·state current sharing
2. Thermal stability and maximum Tj
3. Dynamic current sharing during switching
4. Parasitic oscillations
Another problem which can arise in fast pulse applications- if
low voltage high current devices are used - is a degradation in
the external BVDSS capability caused by voltage spikes
generated by the internal connection inductance (2). Figure 1
shows a MOSFET with the internal parasitic inductances. At
turn-off, ID is flowing. As the device is turned off, the voltage
polarities will be as indicated. The actual voltage across the
j unction will be
d
As is so often the case, many problems that exist in theory are
not significant in the actual hardware. Nevertheless, in the
following discussion, many possible problems wilI be explained.
Steady-State Current Sharing
VDS=VDD+(LD+LS)~
dt
For example, if a 50 ampere, 100 V device with 20nH of internal
parasitic inductance is switched in 20nsec, the internal voltage
spike between the drain and source terminals (which is not
visible externally) will be 50V! The usable value for BVDSS has
been reduced to 112 its data sheet value. Paralleling several
smaller devices, possibly in smaller, lower inductance packages, can reduce the effective value of parasitic inductance
and increase the usable operating voltage.
The distribution of current among paralIel devices is a concern
to the designer. With regard to the effects of asymmetrical
current sharing, three questions need to be answered:
What is the maximum junction temperature among the
devices?
2. Does the asymmetry cause a significant increase in total
dissipation?
3. Is any device operating outside of its safe operating area
(SOA)?
1.
When MOSFETs are subjected to large neutron fluxes (> 1013
neutrons/cm 2), RDS(on) may rise dramatically. One means to
ensure that the post irradiation RDS(on) is not unacceptably
high is to start with a low value for RDS(on) (much less than
required) in the unirradiated circuit. To achieve this, it is
usually necessary to parallel devices.
When the current is not distributed equalIy, some devices may
run hotter than others. Because the operating reliability is
directly related to Tj' it is important to identify the maximum Tj
which can occur. Obviously if any device is operating outside of
its rated SOA, the reliability will also be greatly reduced.
5-16
Furthermore, it is important to know if the conduction asym·
metry is creating a power loss penalty.
10
The answers to these questions depend on the operating state of
the devices. There are two possibilities: linear or switching
operation. When the devices are used as switches, Vas will be
large (lO·15V), and the devices will be fully enhanced. In this
mode the device acts like a positive temperature coefficient
resistor. TYpical ROS(on) versus Tj characteristics are shown
in Figure 2. Note that ROS(on) has been normalized (ROSN) to
25° C for comparison purposes.
8
VDS =50V
u;-
a.
~
I-
zW
6
II:
II:
:::l
0
Z
«II:
Zero Temperature ______
Coefficient Point
(OTC)
4
C
I
9
2
24
J
2.2
V
18
1.6
/
1.4
iii
0
c:
12
10
0.8
06
04
°OL-~--~2--~~~4~CL--~6L-~--~8
500V deVIce
20
I-
-
-~
~~
/.. ,/'
V
V
VGS-GATE SOURCE VOLTAGE (VOLTS)
/'
Typical Transfer Characterlsllcs for an IRF330
FIGURE3
V"
/'OOV devlce-
than average current to heat up more than the other devices,
increasing ROS(on)' which in turn reduces its 10 , The degree of
thermally forced current sharing present has been analyzed (3),
and an outline of the analysis is presented here.
~
02
o
·60 ·40 ·20 0
20
40 60
(A)
80 100 120 140 160 180
Vos
!
The Relationship Between ROS(on), TJand BVOSS
FIGURE 2
Yos
In-~
When the devices are operating in the linear mode, the behavior
is quite different. Figure 3 shows a typical transfer charac·
teristics graph (10 as a function of Vas) with Tj as a
parameter. The interesting feature of this graph is that above
4.5A, the temperature coefficient (TC) is positive, but below
4.5A, the TC is negative. For this device Rejs= I °C/W, and if a
perfect heatsink is assumed such that tc = 25° C then the
maximum value for VOS at 4.5A is
(8)
T· -T
VOS = _l__C_ = 27.8 V
RejcIO
(2)
Current Sharing In Parallel MOSFETs
FIGURE4
Since this is a 400V device, it is unlikely that the device would
be used in the linear mode with such a low value of VOS' Most
linear applications would use the device at currents well below
4.5A to exploit the BVOSS capability (400V), and therefore,
they would be operating in the negative TC region. Since this is
exactly the opposite of the switch mode, the two types of
applications will be treated separately.
A thermal model for two parallel devices is given in Figure 5A.
The normal thermal impedances for junction to case (Rejc)'
case to heatsink (Racs) and heatsink to ambient (Resa)' are
present as well as the thermal coupling between the devices
(Rc)' Three possibilities are indicated:
1. Rc between nodes Tcl and Tc2 represents die mounted on a
common case or header.
2. Rc between nodes Tsl and T s2 represents the situation in
which separate packaged devices are paralleled on a
common heatsink.
Current Sharing While Fully Enhanced
When VGS is large (6·8 V above Vth), a MOSFET is essentially
a positive TC resistor, and the current will divide among the
paralleled devices in proportion to their individual ROS(on) as
illustrated in Figure 4. As was shown in Figure 2, the TC will be
positive, so there is a tendency for a device which has a greater
3. The situation where there is no common coupling between
devices, i.e. separate heatsinks, is represented by Rc = "'.
5·17
-
(A)
T,.
TI'
Assume' for example, that lOA, 100V devices are being used
which have the following characteristics:
(B)
T,.
TI'
p.
.,-
T.1
.....
_~_
Rsa
Ta ~
Po
Ts2
...
.,-
p.
Po
.,-
RiO = 0.12 n
R20 = 0.16 n
I
= 20A
Tc = 0.67 %/OC
Rejc = 1.67° C/W
Recs = 1.00° C/W
Resa = 1.47°C/W
R,.
7
:-
Assume further that the die are mounted on the same header:
:-
Ra = 1.67° C/W
Rd = 2.47°C/W
":' Ta
Thermal Models for Two Parallel MOSFETs
FIGURE 5
If Equations (5) and (6) are solved for values of Rc from 0 to
100° C/W, the graph shown in Figure 6 results.
'02
Ra = Rejc + Recs
Rd = Resa
TI'
..
.00
For analytical purposes, the model in Figure 5A can be
simplified as shown in Figure 5B where the values for Ra and
Rd depend on the position of Rc. For example, for devices
paralleled on a common heatsink,
.
..
..
96
IT
'--
...-
(3)
92
90
(4)
86
84
From this network, it is possible to calculate the values of
RDS(on) for each device (R1 and R2). Given the values for R1
and R2, one can then calculate the values of In (11 and 12),
power dissipation (PI and P 2), and the junction temperatures
(Tj1 and Tj2). Kassakian (3) has derived the following ex·
pressions for this network.
0
•
3456789101112100
Rc(OC/W)
FIGURE 6
Notice that when Rcis small the difference between Tj1 and Tj2
is also smali"{about 9°C) even though R20 is one-third larger
than RiO! The effect on current sharing is almost non·existent
(about 0.3A).
As long as Rc is comparable to or smaller than Ra , the
difference in junction temperatures should remain small. In
practice this should not be difficult to achieve. For multiple die
on a common header, this requirement holds very well. For
multiple individual packages mounted on a common heatsink,
this requirement can be met by having a reasonably heavy
common web between devices and by mounting these devices
relatively close to each other.
Where
In the past, many people (this author included) have touted
thermally forced current sharing as a major feature of parallel
device operation in MOSFETs. Yet thermally forced current
sharing is not prominent except in the case of very high
voltage devices with large TCs and separate heatsinks.
RiO = 25° C value for RnS(on) in device #1
R20 = 25° C value for RnS(on) in device #2
A = Temperature coefficient of resistance. This can
vary from 0.5 to 3%/° C depending on the device.
= 11 + 12
The most common applications for MOSFET switches involve
a load impedance that is large compared to RnS(on). In these
applications, the total current (I) through all of the devices is
determined by the load. In this case, the total power loss is
By itself, this set of equations is not generally useful since the
equations are non·linear: i.e. they contain cross products and
powers of R1 and R2 (the dependent variables). Using numer·
ical methods, this type of equation is usually solved on a
computer. However, in most cases this is not necessary as the
following example solution demonstrates.
(7)
where RT = the total on·resistance of the parallel devices.
5·18
In this case, the distribution of current between the devices is
not considered; all that matters is the final value for Ror. The
fact that some devices will be slightly warmer than others will
increase Ror slightly, but this is usually a second order effect.
2. The quiescent operating or "Q" point is ill-defined and
varies with temperature.
3. Thermal runaway and subsequent device failure are
possible.
From these analyses and other work (5), the following general
observations can be made:
The following discussion will examine these proble.ms and
demonstrate a simple cure. For the purposes of this discussion,
two parallel devices will be used, but the principles exposed
apply to multiple parallel devices. Figure 7 shows the reference
model for the discussion.
1. The current sharing between parallel devices is in pro-
portion to RnS(on)'
2. When the devices are well coupled thermally, the differ-
o
ences in Tj are small even when the difference in RnS(on) is
substantial.
3. For thermally coupled devices, forced current sharing is
insignificant.
4. For high voltage thermally uncoupled devices, some forced
sharing can occur, but since this is achieved at the expense
of higher Tj and Ror, in most cases thermally uncoupled
parallel operation is undesirable.
5. Parallel devices should be mounted on a common heatsink
or substrate with a minimum common thermal impedance
G 0--1--1
(Rc)'
6. Matching of devices for RnS(on) is usually not necessary
unless the range of variation (±2D%) would allow too large a
value for Ror. Rather than matching or screening devices
s
for RnS(on)' it is frequently cheaper and easier to add
another device in parallel.
7. If a limit on the maximum value for Ror is desired and
Two MOSFETs In Paraliel, Reference Circuit
FIGURE 7
matching is acceptable, the matching for RnS(on) should
be done at the anticipated average current for each device
When two non-identical devices are paralleled, a variety of
situations can arise depending on the differences between the
devices. Figures 8, 9, and 10 show several possibilities as well as
the resulting current imbalance. For the moment, the effect of
and the planned value for VGS'
8. To minimize differences in RnS(on) and the final value for
Ror, it is important to enhance the devices fully. A V GS oflD
to 15 V is sufficient.
9. The relevant issue in paralleling is not current sharing, per
se, but rather the junction temperature differences and any
additional power losses. If the ATj and APT are small, the
asymmetry in the current is irrelevant.
Tj changes will be ignored.
Figure 8 illustrates the effect of differences in gm' In this case,
the current differential increases as I D is increased.
10
r---------------------~_,_,
Current Sharing During Linear Operation
In the linear mode of operation, the temperature coefficient of
ON-resistance is negative. This means that if one device, in a
group of parallel devices, is conducting more than its share of
current, its temperature will rise. This will further increase this
device's share of the total current. This process can lead
to thermal runaway and is very similar in nature to the thermal
instability present in BJTs. In the case of the MOSFET, the
transconductance (gm = aIn/a Vas) is much lower, and the
tendency towards instability is correspondingly less.
7
6
10 (A)
,
,,
,,
,
,
4
3
2
,
,,
I
Thermal regeneration of this type combined with normal
device characterization variations can cause three problems:
1.
,
(11- 12)
5
o
2
4
5
6
7
10
VGS(V)
Large differences in current sharing can occur. The current
distribution among parallel devices will vary with temperature and, in some cases, with total current (I).
Current Difference between Two MOSFETs where
gm is the same but Vth Is different
FIGURE 8
5-19
-
peak value of the difference between 11 and 12 is increased, and
the range of V GS over which Ql takes all or most of the current
is expanded. In most applications, the effect of differential
heating in linear operation is to make the problem worse. One
way to minimize the effect of asymmetrical currents is to
maximize the thermal coupling between the devices (make Rc
as small as possible). This is the same conclusion reached for
parallel devices operating as switches!
Figure 9 illustrates the effect of a one volt difference in Vth. In
this example, all of the current flows through one device until
V GS reaches 5 V. At that point, the second device begins
picking up some of the current. The current asymmetry remains
essentially constant for Vth > 6.5V. The problem here is that if
the "Q" point is below 3A, one device will hog nearly all of the
current!
10
9
10
9
7
10 (A)
6
4
10 (A)
3
4
3
o
7
4
8
9
10
VGS(V)
Current Difference between Two MOSFETs where
gm Is the same but Vth dille rent
FIGURE 9
o
8
7
10 (A)
4
I
I
I
10
I
A better solution would be to take randomly selected devices
and force them to share. This can be done most easily by using
small source resistors to provide negative feedback, as shown
in Figure 12A. As an added advantage, this will also stabilize
the "Q" point.
I
I
I
9
5
Good thermal coupling by itself, however, is not enough to
assure good current sharing; it merely reduces the degree of
mismatch. For linear applications, some further means to force
sharing must be taken. One obvious solution is to match
devices. Unfortunately to get really good current sharing, it is
necessary to match the entire transfer characteristic. This level
of matching would rarely be practical. A useful compromise
would be to match the devices at the "Q" point. This is fine in a
stable thermal environment, but if the ambient temperature
(Ta) varies over a wide range then it is unlikely the devices will
remain matched. In any event, matching of devices can be
costly and a considerable nuisance in production.
10
I
4
Current Imbalance from Figure 9 when Heating
Ellects are taken into account
FIGURE 11
Now it is necessary to include the effect of the iunction
temperature on the current imbalance. Looking again at Figure
9, notice thatI l > 12. Itis reasonable to assume that Ql will heat
up and Q2 will cool down. An approximation of what will occur
is shown in Figure 11. From this figure, it can be seen that the
I
3
VGS(V)
Most practical applications will represent some combination of
the two previous examples. A typical combination is shown in
Figure 10. In this particular example, the device carrying the
greater portion of the current will depend on the "Q" point.
5
2
I
I
o
I
I
The effect ofthe source resistor can be quantified by examining
its effect on the "Q" point of a single device as indicated in
Figure 12B. It has been shown (6) that the effective gm (gm')
will be
-1
-2
Current Difference between Two
Mismatched Devices
FIGURE 10
Rs+ 1
gm
5-20
(8)
10
To minimize the effect of differences in gm' it is necessary that
(9)
8
Typical devices will have values for lIg m in the range of 0.1 to
1.0 ohms.
7
RS=Oj
VGG = S.4V
6
10 (A)
(A)
5
4
o
2
3
VGS(V)
Operating Point Stabilization
FIGURE 13
1.667A. From Figure 12 we know that
VGS = VGG- InRs
(10)
so that V GS in this case is 4 V.
What we are seeking is a simultaneous solution to Equation (10)
and the equation represented by the graph of the transfer
characteristic. This is done by graphing Equation (10) (the
straight line) on the transfer characteristic graph and by
noting the intersection of the graphs. In Figure 13, the
(8)
intersection representing the "Q" point is at V GS = 5.4 V, and
In = 1.2A.
This same technique can be applied to parallel devices since
stabilization of the "Q" point also stabilizes the current
sharing. Using the example of Figure 9, the effect on sharing
by adding a 2 ohm resistor in series with each device can be
determined. A series of parallel lines are drawn for several
different values of VGG. each having a slope corresponding to 2
ohms. Each line establishes a "Q" point with a given value of
t.I. The values for t.I corresponding to the different values for.
+
10 RS
Forced Current Sharing using Source Resistors
FIGURE 12
V GG can now be plotted and compared to the values for
th~
original example (Figure 9). This is shown in Figure 15 wherein;
can be seen that t.Iis reduced from 3.2to 0.5A! Clearly this is an
effective means for equalizing the current distribution as well
as stabilizing the operating point.
The effect of Rs on the "Q" point stability (with fixed V GG) is
shown in Figure 13. If the operating point is selected as V GG =
V GS = 5.4V and Rs = othen In= 1.2AatT1, butit will rise to In
= 2.7Aat T2 -an increase of more than 2:1! By making Rs = 3
ohms (as indicated by the sloping line), the change in In(t.I)
when the temperature changes from Tl to T2 is reduced to
0.25A! This is a very great improvement. In this example, gm is
approximately 2.5S so that Rs ~ 8/g m . Even if Rs is reduced to
1 n, t. I is still only 0.6A.
Source resistors can also improve thermal stability by reducing
differences between the transfer characteristics as Tj is varied.
This effect can be illustrated by regraphing the example in
Figure 13 as a function of V GG rather than V GS' This is done in
Figure 16. In effect, the thermal regenerative gain is much
lower and the stability greatly enhanced.
For those readers not familiar with the graphic technique just
used, a few words of explanation may prove helpful. The
Note that Figure 16 is also an alternative method for determining t.I between non-identical devices; however, itis usually
straight line representing Rs = 3 and V GG = 9V is shown in a
graph representing the values of V GS for given values of ID'
For example, if In = 1.667 A, the drop across Rs is 5V, viz. 3 x
simpler to draw the resistive load lines for R s ' as was done in
previous examples (Figures 13 and 14), than to redraw the
graph in the form used in Figure 16.
5-21
-
10
The use of source resistors has several advantages:
1. The operating point is stabilized.
2. Excellent current sharing and equalization of power
dissipation is achieved.
3. Transconductance differences are minimized so that the
small signal gain is nearly the same in each device.
4. Thermal stability is enhanced.
5. The small signal linearity is improved.
10 (A)
Q
1
2
8
3
9
10
11
12
13
14
VGsM
Rs Load Lines Plolted on the Transfer
Characteristic Graph
FIGURE 14
Unfortunately, these benefits are not gained without some cost.
The use of source resistors has the following disadvantages:
1. Additional components are needed: i.e. the resistors.
2. The large signal dynamic range is reduced by the voltage
drop across the resistors, but since this voltage drop is
usually small compared to VDS this is not a severe penalty.
On the other hand, the source resistors tend to extend the
lower limit of the large signal linear region (Figure 16), and
this compensates by extending the dynamic range.
3. The voltage gain of the stage is reduced. In the common
source configuration, the voltage gain (A) is
A=gmRL
(11)
where RL is the load resistance.
RS"'O
From Equation (6), the gain will be reduced to
A = gmRL
1 + Rsgm
when source resistors are used.
(12)
From the foregoing discussion, the following general obser·
vations regarding parallel devices operating in the linear mode
can be made:
1. This mode of operation is very different from the switching
mode in most respects.
2. To avoid excessive current asymmetry and thermal
instability, some positive means must be provided to
stabilize the operating point and force current sharing.
3. The simplest means to achieve the above goals is to use
small resistors in series with each device's source lead.
4. Good thermal coupling between devices will greatly im·
prove thermal stability and current sharing and help to
minimize the size of Rs.
5. Matching of devices, while useful and effective, is usually
not necessary. Small source resistors are usually a cheaper
and simpler solution. The size of the resistors can be
reduced by prescreening devices to eliminate those with
larger than average characteristic deviations.
6. Multiple die devices will normally (at least at Siliconix)
have the die selected from adjacent positions on the same
wafer and will be well matched. In addition, the thermal
coupling will be very good. The user cannot, however, add
individual source resistors because the die are sealed within
the case. If the degree ofintrinsic matching in multiple die
devices is not adequate, the single die devices can be
paralleled using individual source resistors.
The Effect of Source Resistors on Current Sharing
FIGURE 15
10r--------------------i--i---------,
10CA)
The Effect 01 Rs on the Transfer Characteristics
over Temperature
FIGURE 16
5·22
Current Sharing During Switching Transitions
impedances. Using symmetrical layouts is one way to doit.
An example of a symmetrical layout for a parallel switch
connection (push·pull) with three paralleled devices is
gi ven in Figure 18. Many other practical symmetrical
When parallel MOSFETs are used as switches, several ques·
tions arise concerning device behavior during the switch
transitions from OFF to ON and from ON to OFF:
layouts are possible.
3. Good layout techniques are vital. The effect of Ls on the
switching waveform is shown in Figure 19. Obviously Ls
should be made as small as possible. Further examples of
good layout technique are given in Figures 20 and 21.
1. What is the current distribution?
2. Can current asymmetries be severe enough to damage a
device?
3. Is the switching time affected by current asymmetries?
When a MOSFET makes a transition from one state to another,
the device must pass through the linear region, if only momen·
tarily. This means that much of what has been discussed
concerning the linear mode of operation applies to dynamic
switching. However, there are some differences:
1. The operating time in the linear region is very short
(typically 10 to 100 nsec).
2. Because of the rapid transitions, thermal heating effects
are usually negligible.
3. The use of source resistors is unacceptable because of the
substantial increase in the effective value of RDS(on)'
4. The external parasitic and intentional resistances, induc·
tances, and capacitances must be taken into account.
Symmetrical Layout Example
FIGURE 18
An equivalent circuit representing two parallel FETs is given
in Figure 17. Differences in current during switching can be
caused by:
1. Differences in the external circuit elements.
2. Differences in capacitive or inductive device characteristics.
3. Differences in device characteristics during linear
operation.
VGG
The Effect of Common-Source Inductance
on Switching
FIGURE 19
Differences in package inductance and inter·terminal capaci·
tance are amazingly small (~±5%) and usually can be ignored.
This is fortunate since without a complex matching procedure,
there is little that can be done anyway.
GATE
DRIVE
Equivalent Circuit for Two Parallel MOSFETs
During Switching Transitions
FIGURE 17
If for example, Q1 and Q2 are identical, but the drive circuit
impedances are different, Q1 and Q2 will not turn on simulta·
neously. This problem, however, is under the control of the
designer. The following steps can be taken to minimize this
problem:
1. Minimize the values for R, Lg , Ls ' LD,C p and C s '
2. There will be practical limits on how small the external
impedance can be. When that limit is reached, every effort
should be made to equalize the values for the remaining
·'nduclance
(
• Noise Pickup
• Parasitic Oscillation
lnneel Direct To MOSFET Tonnlnal
MInimizes Common·Source Inductance
Low Impedance Onva For.
• Fasl Switching
• Mmimlze Noise Pickup
• Minimize dV/dl Turn·On
Suggested Drive Circuit Layout
FIGURE 20
5·23
l1:li
Figure 23. If V GS is a linear function of time (a ramp) then the
graphs for 11 and 12 become their current switching waveforms
as a function of time! The salient feature ofthese waveforms is
that even though the two devices have very different threshold
voltages, the current spike for Ql is small (about 1. 7A). In most
cases, when unmatched devices are paralleled, only relatively
small current differentials are observed. A good series of actual
oscillographs is given in Reference (5) to illustrate this point.
Applications do exist where from 20 to 50 devices are paralleled.
In these applications, it is theoretically possible (but relatively
unlikely) to generate destructive current spikes. Such spikes
can be avoided by prescreening devices to eliminate those
devices that are radically different from the median.
Differences in device characteristics, on the other hand, can
cause differences in current distribution during switching
transitions. To illustrate what occurs, the techniques pre·
viously used for linear operation can be applied to the example
in Figure 9. If a clamped inductive load is assumed, Figure 22
shows the switching waveforms. IfID = 11 + 12 = 12A then the
current during switching will assume the value shown in
1
I
"(
""
.
.
'
r"
TI ~
153 (
(S4
'MINIMIZE THE AREA
12
High dl/dt Paths In an H-Bridge
FIGURE 21A
r----------~,~~~"I
,
ft
10
:
--I
'1+12 :
,I
,1
1
+
10 (A)
1
,
6
1
,
I
1
9
10
vGS(V)
Values for Drain Current during Changed Inductive
Switching when 11 + 12 is limited to 12A
FIGURE 23
TWISTED PAIR,
COAX OR STRIPLINE
I.e. LOW L CONNECTION
The following observations can be made regarding switching
in parallel MOSFETs:
Low Inductance Connections for High dl/dt Paths
FIGURE 21B
1. Careful circuit layout is needed to minimize and equalize
parasitic impedances.
2. Minimizing differe\ltial gate impedances will equalize VGS
for each device.
3. Except for large numbers of devices, the current spikes
during switching transitions will usually be well within the
limits of the device capability.
4. If many devices are paralleled, simple prescreening which
measures ID at a given value of V GS in the transition
region (i.e. not fully enhanced) should be sufficient. This
will eliminate the unusually different devices.
'"
lL
I
I
I
I
I
I
I
I
I
I
Parasitic Oscillations in MOSFETs
Most power MOSFETs presently available are very fast devices
with appreciable gain at frequencies up to 300 MHz. This high
frequency gain, coupled with the internal and circuit parasitic
Clamped Inductive Load Switching Waveforms
FIGURE 22
5-24
inductances and capacitances normally present, make it possible for unwanted parasitic oscillations to occur. The fre·
quency of oscillation can range from 1 to 300 MHz.
VGS
The oscillations occur while the device is in the active mode
where the transconductance is large. When the device is off or
when V GS is large and the device is fully on, oscillations do not
occur. This means that in a switching application, the oscilla·
tion will occur on a transient basis during the turn·on and
turn-off transitions. In an application where the device is
biased on to some fixed point in the linear region, the oscilla·
tions can be continuous.
VOS
An example of a parasitic oscillation during a turn-on transition
is given in Figure 24. Figure 24A shows the normal drain·to·
source (VDS) and gate·to·source (VGS) voltage waveforms
when no oscillation is present; Figure 24B shows the same
waveforms when an oscillation is present. An expanded portion
ofthe oscillation envelope is given in Figure 25. In this case, the
oscillation frequency is approximately 85 MHz. Although the
oscillation amplitude shown is not very high, voltages of 100V
or more are possible at the peak of the envelope.
FIGURE 24B
The existence of parasitic oscillations can have the following
consequences:
1. Gate rupture due to overvoltage.
2. Gate rupture due to overheating of the gate structure.
3. Increased power dissipation in the device.
4. Increased voltage and current stress in associated circuit
components.
5. HF to VHF electromagnetic interference (EMI).
Expanded Time Base Showing Free Running
Parasitic Oscillation
FIGURE 25
None of these are normally acceptable, and, in nearly all
cases, the parasitic oscillations must be eliminated. Many so
called "mysterious" failures in MOSFETs are due to parasitic
oscillations.
A single device can oscillate by itself. Parallel devices can also
oscillate as if they were a single device; this is referred to as a
"common mode" oscillation. In addition, paralleled devices can
oscillate in a "differential mode." The analysis for each mode is
very similar, but the parasitic circuit elements controlling the
oscillations are different.
Common Mode Oscillation
Figure 26 gives a model ofthe internal and external parasitic
capacitances and inductances in a typical application. To
analyze this circuit for oscillations, a simplified incremental
model (Figure 27) can be used. Figure 27 is not exact but will
still give useful answers. The following assumptions have been
made and are normally valid:
1. LGe» LGi
5. C 1 = Cgse + Cgsi
2. LDe » LDi
6. C2 = Cgde + Cgdi
3. LSe » LSi
7. C3 = Cdse + Cdsi
4. RGe» RGi
A more complex model could, of course, be used at the cost of
greatly increased computational difficulty.
Non-Oscillatmg MOSFET
FIGURE 24A
5-25
-
In some applications, the model may be further simplified if
either Ln or LS is very small. The coefficients of the charac·
teristic equations for these cases are
.".,..--",/
I
I
I
/
ROI
/
(20)
a1 = LGLSC~
(21)
a2 = C~(LSRG + LGRn + LsRn) + gm LGLSC2 (22)
aa = RnRGC~ + gmC2(LSRG + LGRn + LsRn) + (23)
LG(C 1 + C2) + LS(C 1 + C3)
a4
=
gmRn RGC2 + RG(C 1 + C2) + Rn(C2 + Ca ) + (24)
gmLS
Equivalent cIrcuit
FIGURE 26
LS = 0
L~
RO
= LnLG
(25)
a1 = LnLGC~
(26)
a2 = C~(LnRG + LGRn) + gm LnLGC2
(27)
a3 = RnRGC~ + gm(LnRG + LGRn)C 2 +
(28)
LO
LG(C 1 + C2) + Ln(C2 + Ca)
a4 = gmRn RGC 2 + RG(C 1 + C2) + Rn(C 2 + Ca) (29)
Incremental Model
FIGURE 27
The practical question which must be answered is: "noes the
circuit oscillate or not?" The characteristic equation can an·
swer this question. If any of the roots of the polynomial are
negative (i.e. lie in the right half-plane or have the term s-a)
then the circuit will oscillate. There are several ways this can be
determined.
The characteristic equation for the incremental model is (7)
a1S4 + a2S3 + a3S2 + a4S + 1 = 0
(13)
Where
a1 = C~L~
(14)
a2 = C~[Ra 100nA). The best means for
detecting oscillations is to place a scope probe between the gate
and source terminals, directly on the device. The scope should
havea bandwidth of at least 200 MHz sinceitis very difficult to
detect VHF oscillations with an instrument that responds to
only a few MHz. A low capacitance probe should be used, and as
Circuit for Monitoring Charge Transfer
Figure 1
little stray inductance as possible should be introduced. It is
Since we are looking at high-frequency parasitic
oscillation, it is vital that the circuit use direct pointto-point wiring-a few misplaced nanohenries may
upset the results.
possible for the measurement process itself to alter the circuit
operation and to suppress or induce the osciliatlOn!
Prevention of Oscillation in MOSFETs
Examining the Conditions for Oscillation
From the theoretical and experimental work on this problem, it
is clear that preventing parasitic oscillations is not a major
problem and can be accomplished by observing the following
In Figure 2, the resulting charge-transfer characteristics are divided into three well-defined regions. In
region 1, the gate voltage has charged the input
capacitor of the power MOSFET to where turn-ON
just begins (VT); in region 2, we witness not only the
completion of the turn-ON cycle but also the effects
of Miller capacitance upon the charge characteristics.
In region 3, we see the drain-to-source voltage settling
slowly to VSAT followed by a resumption of the gate
charging cycle.
guidelines:
1. Minimize the parasitic inductances and capacitances. In
particular LG' LS' and Cgd should be made as small as
possible. Making the parasitic elements smaller raises the
resonant frequency. As the frequency is increased, the
gain ofthe device will decrease, and the resistive damping
present will become more effective. The net result is a
reduced likelihood of oscillation.
2.
Use small (1-5 ohm) differential resistors in the gate lead of
400
each device. Because of the silicon gate structure of Siliconix devices, most of the needed resistance will already be
Z
present. RG should, of course, be non-inductive. Carbon
composition resistors are particularly good.
3.
u;
ti
.
2Q
g
~
(30)
r---ir-- Vg1
~ 100
the oscillation will not appear even though the circuit is
potentially unstable.
.-~ - - t
~
~
r-...
VGG
J
Q2;----1- 3
Vg2
I.
Q,
I'I--~O%
10
vide both RG and increased LG and can be very effective in
suppressing oscillations.
II
I
I
For the differential mode of oscillation, ferrite beads pro-
I
V
~
f---'
I
1/
4. Minimize the differential values of LS and Ln.
5.
~
15
I
then
;::: s.!- r-
~
VDS
200
"
T,
z
~
"~
W
If the switch transition time is short compared to
300
~
Any resonant circuit has a non-zero value of Q. The
higher Q is, the slower the oscillation will build up. The
time constant (T) will be
T=--
I
N
15
20
25
A Typical Charge-Transfer Characteristic
Figure 2
5-29
-
Here we must point out that if we were to parallel
power MOSFETs in such a manner as to ensure
parasitic oscillation, we would discover a well-defined
envelope of oscillation only in the immediate vicinity
of the transition between regions 2 and 3. This welldefined envelope is shown in Figure 3 where we
should carefully note its position on the charge curve.
The gate voltage has passed threshold (VT), and the
drain-to-source voltage has decayed to a point approaching but not having reached VSAT Oscillation
found anywhere else is caused by other phenomena,
perhaps excessive wiring inductance in the test circuit.
Causes of Oscillation
Oscillation results from a resonance induced by the
combination of parasitic capacitive and inductive
elements including, in particular, the loop inductance
between discrete power transistors coupled with the
characteristically high gain of the power MOSFETs.
Since the interelectrode capacitances of a power
MOSFET are voltage dependent, behaving as we see
in Figure 5, this parasitic oscillation becomes conditional upon the applied drain voltage.
In Section 3.2 we resolved that the prime cause ofthe
extended Miller effect on the gate charge as well as
the slow settling of the drain-to-source voltage to
VSAT after turn-ON, was the result of a many-orderof-magnitude change in the gate-to-drain capacitance, Cgd as shown in § 3.2-Figure 6 (and again in
Figure 6). The meteoric rise in gate-to-drain capacitance only begins as the power MOSFET gate charge
enters into the transitional area between regions 2
and3.
To understand the circumstances resulting in oscillation where it appears on the charge-transfer characteristic, we need to closely examine what changes take
place in the power MOSFET.
1000
Cg•
.........
"'c•• '
""
Envelope of Parasitic OSCillation Situated within Region 2 of
the Charge Characteristics
Figure 3
Cg.
By expanding our oscilloscope time base, we can
arrive at a close approximation of the frequency of
oscillation. The result is given in Figure 4 for a pair of
parallel-coupled Siliconix VN4000A MOSPOWER
FETs.
10
01
10
10
100
Vas - VOLTS IV)
The Interelectrode Capacitance of a VN4000A MOSPOWER
FET as a Function of Drain Voltage
Figure 5
-,-
-I- f'~
1000
\
VN4000A
IRF33VT; and c) Vgs»VT.
Upper Traces: Vsupply = 10 V/div;
Middle Traces: Vdraln 10 V/dlv; Lower Traces: Vgs 2V/div
Figure 2
=
5-34
=
The interaction of the parasitic bipolar transistor by
dV/dt breakdown initially seemed to be a likely cause
of the oscillation. This suspicion was strengthened
by the anomalous turn-off behavior in which the
MOSFET turns back on for an extended period. Interaction of the parasitic bipolar transistor requires
that a source of current be injected across the baseemitter junction to turn it on. This situation can occur
when a fast rising voltage is applied to the drain!
collector of the transistor that causes a rapid expansion
of the base-collector depletion region. The charge
displaced from the base depletion region moves towards the source contact causing a resistive voltage
drop in the base region. If this voltage drop exceeds
approximately 0.6 volts, the emitter base junction
becomes forward biased and the bipolar transistor
turns on, shunting the cut-off MOSFET.
Oscillation of the Power MOSFET when a 0.1 J-LF Capacitor is
Connected -3 Inches from the Gate of the MOSFET
Figure 3
For the above type of mechanism to be responsible,
the oscillation can only exist in a large signal state
since a high dV/dt at the drain is required to activate
the parasitic bipolar transistor. An experiment to test
this hypothesis demonstrated that the parasitic transistor is not involved but that the oscillation is due
to a small-signal instability. The experiment involved
removing the exciting transient introduced by the rising edge of the input pulse by shunting the gate of
the MOSFET with a 0.1 JLF capacitor. This reduced
the tendency of the MOSFET to oscillate as might
be expected by AC grounding the MOSFET gate.
However, when the capacitor is decoupled from the
gate by moving it 3 inches towards the pulse generator, the circuit oscillated readily. The slowly increasing oscillation amplitude seen in Figure 3 started
at near zero amplitude, and this indicated that the
oscillation is indeed due to a small-signal instability.
Thus, a small-signal analysis should reveal the conditions for oscillation for the power MOSFET.
CGS
RD
I
LD
Small Signal AC Model of the Power MOSFET Inverter Circuit
Shown in Figure 1
Figure 4
Writing KVL around the three loops and KCL at the
three nodes for the circuit shown in Figure 4 yields
ig Lg -ics Lcs =-Ig Rg + Vgs
(1)
id - ics Lcs = - Id Rd + Vds
(2)
Vgd + Vds = Vgs
(3)
Cgs Vgs +Cgd Vgd =-Ig
(4)
Cds Vds - Cgd Vgd = - Ig - V gsgm
(5)
Routh-Hurwitz Analysis
Having established that the oscillation is due to a
small-signal instability, the simple inverter circuit,
shown in Figure 1, is modeled using the standard
small-signal AC model as seen in Figure 4. A set of
state equations can be written to describe this circuit,
from which the characteristic polynomial can be
found. The location of the roots of the characteristic
polynomial in the complex frequency plane then will
determine whether the circuit is stable.
Ig +Id +Ics = 0
(6)
These equations can be cast into a matrix state
equation:
5·35
-
(Lg + Lcs) Rd
Lcs Rg
Leff2
Leff2
[~:}
Vds
Lcs Rd
Leff2
Cgd
- Ceff2
Cgs
+ Cgd
Ceff2
(Ld
+ Lcs) Rg
Leff2
Cgd + Cds
Lcs
- Leff2
Lg + Lcs
Ld + Lcs
Leff2
Lcs
- Leff2
Cgdgm
- Ceff2
Ceff2
Cgd
- Ceff2
(C gs
+ Lcs Ld + Lcs Lg
Ceff2 == Cds Cgs
(9)
The characteristic equation is the determinant of the
matrix A- sl where A is the state equation matrix
written above. The characteristic polynomial is then
P(s} = aos4 + als 3 + a2s2 + a3s + ~
(10)
(11)
+ Lcs) Rg + (Lg + Lcs) Rrl]
+Cgdgm Leff2
(12)
Rrl Rg + Cgdgm [(Ld + Lcs) Rg
+ (Lg + Lcs) Rd] + (Cgs + Cgd) Lg
+ (Cgd' + Cds) Ld + (Cgs + Cds) Lcs
a2 = Ceff2
0
01 = al
(13)
02
a3 = Cgdgm Rrl Rg + (Cgs + Cgd) Rg
+ (Cgd + Cds) Rd + gm Lcs
(14)
~=1
(15)
Vds
Although there is no analytic solution for the roots
of a fourth order polynomial, stability only requires
that all the poles lie in the left half s-plane. The
Routh-Hurwitz criterion states that a necessary and
sufficient condition for all of the roots of an nth order
equation to lie in the left half of the s-plane is that
all of the Hurwitz deteI;IIlinants, Ok (k= 1,2, ... n),
must be positive. The Hurwitz determinants for a
fourth order polynomial are given by
where
al = Ceff2 [(Ld
+ Cgd) gm
(7)
form a loop, and the inductors form a cutset. Kirchhoff's Voltage Law constrains one of the capacitor
voltages to be the sum of the other two; hence,
the three capacitors contribute only two state variables. Likewise, although there are three inductances,
there are only two independent state variables because
Kirchhoff's Current Law constrains the current in one
of the inductors to be the sum of the other two. Thus,
the six energy storage elements have only four independent state variables resulting in a fourth orde"
characteristic polynomial.
(8)
+ Cds Cgd + Cgd Cgs
~~
0
Ceff2
where Leff2 and Ceff2 are defined as
Leff2 == Ld Lg
Leff2
=
03 =
The order of the characteristic polynomial is determined by the number of energy storage elements
which contribute a state variable. Note that although
there are a total of six energy storage elements, the
characteristic polynomial is only fourth order. This
is because the three capacitors, C gs , Cds, and Cgd,
03 =
5·36
Ialao
(16)
a31
a2
al a3
ao a2
0 al
al
ao
0
0
a3
a2
al
ao
(17)
0
(18)
~
a3
0
~
a3
a2
0
0
0
~
=
~03
(19)
To simplify the evaluation of the roots of the characteristic polynomial, it would be desirable to reduce
the order of the polynomial by setting some of the
capacitors or inductors to zero. Unfortunately, the
removal of one inductor and one capacitor will not
reduce the order of the polynomial because the number of independent state variables is not decreased.
However, the algebra in solving these equations is
still simplified enormously if some of the circuit elements can be set to zero, and thus, Rg and Lcs are
set to zero. Setting Rg and Lcs to zero represents a
typical situation since these quantities are usually very
small (Rg«wLg and Lcs«Ld). By inspection,
D1 is always a positive quantity since all of the
component values are positive. An algebraic evaluation of D2 reveals that when Lcs is zero, D2 is
always positive. Further, since D4= a4D3, and a4= I,
the circuit's stability is indicated by the sign of the
third Hurwitz determinant, D3.
The algebraic expression for D3 when Lcs= Rg= 0 is
D3
= {
(Cgd + Cds) Rd [Cgd gm Lg Rd+ (Cgs + Cgd) Lg + (Cgd + Cds) Ld 1
-C eff2 Lg Rd - Cgd gm Ld Lg} (C eff2 Lg Rd + Cgd gm Ld Lg)
(20)
- Ceff2 (Cgd + Cds)2 Ld Lg RJ
This equation is second order in all variables except
Lg which fortunately is one of the variables which can
control the oscillation. Solving the above equation for
Lg with the constraint that D3 > 0 yields
Lg<-------------------------------------------------(Ceff2 Rd + Cgd gm Ld) (gm Ld - Cgd gm R~ - Cds gm RJ - Cgd Rd)
(21)
for
(22)
Note that when the equation on the bottom is negative,
the sign of the inequality is changed. Then the condition for stability is always satisfied because Lg is a
positive quantity.
uations of equations (21) and (22) for a conventional
MOSFET show that these constraints are both easily
satisfied.
An examination of equation (23) reveals that the critical value of Lg is roughly inversely proportional to
the gate area. Note that the capacitances and transconductance scale proportional to the area while
both the on resistance and the load resistance is inversely related to the area. Substituting these dependencies into equation (23) shows that the
maximum gate inductance for which the circuit will
be stable decreases with increasing device size. Thus,
as MOSFETs grow larger in size, they have a greater
oscillation tendency.
For a typical circuit, the load inductance terms dominate in equations (21) and (22). The condition for
stability as given by equation (21) can then be reduced
to
Lg <
+ Cds)2 Rd
--='------
(Cgd
(23)
Cgd gm
For the inverter circuit studied here, this condition is
a relatively severe constraint. Typical numerical eval-
5·37
-
Computer Simulation Using SPICE
........UUD8/S5 ••••- •• SPICE 2G I5A (2SNOVB2) .. ······11 32
os·····
OLe lESS TF.AN CRlTICAL
0.. •• INPUT LISTING
T£JlPERA""'l'"RE = 2., OCC DEG C
0LG
.. ••• ..03
•••••••••••
ION .. ••• .. ··"····························· ............ .
The analytic solution in the previous section has limitations because of the assumptions that Rg= 0 and
Lcs= 0 and because of the general complexity of D3.
More significantly, the analysis does not show how
close the circuit is to oscillation but only if the
circuit is or is not in oscillation.
LD
CCS
CCD
2~
200P
200P
1000P
20U
VG
40
ACI
aJS
80
82
20
RD ~o ?
Gil eo 66?{;e-C2
8708e+-05 '1 SC3e-C2
II 9S0e+Oo e 6-;.~-C2
12::2e+0~ 9 7G3e 02
14ele+0:; l1CZe-Cl
o727e+05 1 22Ce (,.1
-I CCOe+05 I 344e el
6296e+CO
,. 738e+CC
98S2e+::::
13Z6c+CI
19S4e...Cl
:3 156e+Cl
o4531e+Cl
3246e+Cl
2114e+Cl
I riSGeH'l
1215e+Cl
I CICe+C)
B e97e+CC
,. 6B3e+C:'
6 BIBe+C::
632Ce+C:
5843e+:::r.
5452e+CC
5128e+CC
485-4e+CC
462Ie+CC
4 .;20e +CC
4 24!ie+OO
1 C28e+OS
1 057e+03
I 087e+05
.1I7e+05
J 14ge+05
1 lele+05
• 214e+O-3
J 246e+05
.2:!Se+D5
1.320e+05
1,35'e-t09
1..39!5e+-D5
1.434e+0:3
1,4?4e"05
UIiUle+O!':
UIi58e+D5
• .102e+05
Edu..ed. fJ'om the SPICE program Ol...tput.
1 44'5e-CI
I 523e-C 1
1 572e-Cl
1 eS.;e-Cl
I 5ti~e-Cl
157cl;-Cl
I ~Se-C 1
152':e-CI
14'1iuc-Cl
146ge-C1
1444e C)
I 42~e-CI
14C5e-Cl
1 SBle-Ol
1.5!:c-CI
1 S73e CI
I S6ce-C 1
Edlted from the SPlCEproU&D o\.t.p\.t.
SPICE Simulation of the Inverter Circuit's Loop Gain
to Show the Effects of Gate Resistance
(Note That the Gain Is Less than Unity at the Frequency
Where the Phase is Zero, and Hence the Circuit is Stable)
Figure 9
SPICE Simulation of the Inverter Circuit to Show the Effects
of Common Source Inductance or the Loop Gain
(Note That the Maximum Loop Gain Is Decreased (0.159)
Which Decreases the Tendency to Oscillate)
Figure 10
5·40
output capacitance and then returned to the inductor
with the opposite current polarity. Neglecting the
nonlinearity of the output capacitance and the damping
of the output load resistance, the time for 1/2 cycle
is
Turn-off Behavior
The turn-off behavior of the power MOSFET is shown
in Figure 11. Initially, the gate voltage is high and
the MOSFET is fully turned on. When the gate voltage falls, the drain voltage rises quickly and then
abruptly drops to the "on-voltage," where it remains
for several microseconds before ringing and settling
to the supply voltage.
t
= 7T-v'LC
(24)
After this time, the parasitic diode is forward biased
and the voltage across the inductor is
If the damping losses are neglected, then
Id
= VDD/RL
(26)
and
VL = 2VDD
+ VD- fwd
(27)
The drain voltage remains near zero during the time
the diode is conducting. The current through the diode
is dictated by the L-R circuit and is
Id = 2VDD
a)
+ VD-fwd (l-e- tR / L) _ VDD (28)
R
R
The time at which the diode current is zero is
[2VDD
ton = (L/R) In
+ VD-fWd]
LVDD + VD- fwd
(29)
For a reasonably large supply voltage, VDD» 0.7
volts, the above equation can be expanded in a Taylor
series to give
ton =L- [ 1 R
VD-fWd]
In 2
2VDD
(30)
Neglecting the reverse recovery of the diode for the _
moment, the drain voltage rises at this time and then
the drain voltage rings and settles to the supply voltage.
b)
Oscillograph of the Turn-off Behavior
of the Power MOSFET Inverter;
a) Upper Trace: Vo = 10 V/dlv; Lower Trace: Vgs = 5 V/div;
b) Upper Trace: Vo = 10 V/div; Lower Trace: 10 = 0.5 A/div
Figure 11
Second Order Effects
A comparison of experimental data with the calculated
results of equation (30) reveals a large discrepancy
in the "on-time." Also, the oscillograph of Figure
11 shows that the power MOSFET appears to turn
on a third time. These discrepancies were found to
be due to the effect of the reverse recovery of the
diode (see Figure 12). Instead of turning off when
the forward diode current crosses zero, the diode
continues to conduct in the reverse direction before
The sequence of events is shown in Figure 12. Before
the MOSFET is turned off, energy has been stored
in the inductor. When the device is turned off, the
output capacitance of the power MOSFET and the
inductor cycle for slightly more than one half cycle
in which the inductive energy is dumped into the
5-41
the drain voltage begins to rise. Consequently, the
total tum-on time expressed by equation (30) must
be modified to include the reverse recovery time trr:
ton"" -L [ 1 R
VD-fWd] In 2
2VDD
+
termine if the circuit was stable, a set of expressions
was derived that places a constraint on the maximum
value of gate inductance for which the circuit will
be stable. From this expression it may be seen that
as the area of a MOSFET increases, the maximum
allowable gate inductance decreases, which may explain why conventional MOSFETs do not have as
great a tendency to oscillate as power MOSFETs.
(31)
trr
Since the inductor once again has a downward current
direction after the diode's reverse recovery, the situation is the same as when the MOSFET first turned
off but with a lower initial current level, and thus
the cycle repeats again.
Experiments and computer simulations of the power
MOSFET inverter circuit reveal that the addition of
small amounts of gate resistance (Rg>woscLg) or
common source inductance effectively stops the oscillation tendency while not severely degrading the
performance of the MOSFET. Unfortunately, there
appears to be no way of reducing the tendency of
the power MOSFET to oscillate without degrading
its performance.
n
I \
I \
z
"
I
I
I
~ Voo
\
\
1~:J---t====-C:=-=J-t::=-==j-I-
The anomalous turn-off behavior was found not to
be caused by dVldt breakdown of the parasitic bipolar
transistor. Instead, it is caused by the intrinsic diode,
clamping the output voltage to ground. The reverse
recovery of the diode both extends the time during
which the MOSFET is apparently on and causes the
power MOSFET to seemingly tum "on" repeatedly
as the output voltage settled to its final value.
- - EFFECTS OF TRR
---~
T1ME-
-TRR-~
The tum-off behavior will not be ob~erved in a typical
power MOSFET circuit because when the MOSFET
is used to drive an inductive load, the drain is usually
clamped to the positive supply voltage. However,
when the power MOSFET is used to drive an unclamped inductive load, the parasitic diode can extend
the time required for the drain voltage to settle at its
final value.
Diagram of the Power MOSFET's Turn-off Behavior
Showing the Effects of trr
Figure 12
Another discrepancy is the dependency of the turnoff time on the supply voltage. From equation (31)
the time is independent of the supply voltage when
the supply voltage is much greater than 0.7 volts.
However, when the peak voltage during turn-off is
near the breakdown voltage, the MOSFET will conduct and dissipate some of the stored energy, resulting
in a decrease in the tum-on time. In our experiment,
even with a low supply voltage of ten volts and a
load inductance of 17 ILH, the peak voltage sometimes
exceeded 450 volts and resulted in a decreased turnon time with increasing supply voltage.
References
[1] D.K. Sharma and K.V. Ramamathan, "Modeling
Thermal Effects on MOS IV Characteristics,"
IEEE Electron Dev. Let., Vol. EDL-4, p. 362,
Oct. 1983.
[2] Yasuhisa Omura, Eiichi Sano, and Kuniki
Ohwada, "A Negative Drain Conductance Property in a Super-Thin Film Buried Channel MOSFET on a Buried Insulator," IEEE TIan. E1ec.
Dev., Vol. ED-3~, Jan. 1983.
Conclusions
It has been found experimentally that the power
MOSFET oscillates due to a small signal instability.
The small signal instability was confirmed by analyzing a small signal model of a power MOSFET
inverter circuit both analytically and with computer
simulations. Using the Routh-Hurwitz criterion to de-
[3] D.S. Kuo, C. Hu, and M.H. Chi, "dV/dt Breakdown in Power MOSFETs," IEEE Electron Dev.
Let., Vol. EDL-4, p. 1, Jan. 1983.
542
5.3.2 Thermally Forced Current Sharing in
Paralleled Power MOSFETs
Abstract
A static thermal model for a pair of paralleled devices
is proposed and analyzed. The model includes the
parametric variability necessary to represent both chip
and package paralleled devices, and to represent various degrees of thermal coupling between the two
devices. Analysis of the model produces a set of
coupled non-linear equations which are solved
numerically using an iterative algorithm. The drain
currents and "junction" temperatures of both devices
are then determined for a variety of conditions.
A thermal model is proposed and analyzed to determine the extent of thermal forcing of current sharing
in paralleled power MOSFETs. Both package paralleled and chip paralleled devices are examined. It is
shown that the effects of thermal forcing on the current
distribution among paralleled devices are relatively
insignificant. The most reliable operation of paralleled devices is achieved if they are tightly coupled
thermally.
Introduction
The drain-source resistance of a power MOSFET is
dominated by the bulk spreading resistance of the
drain region, and not the inverted channel resistance.
The temperature dependence of this resistance is
caused principally by the variation of the drift mobility. As a consequence, the resistance exhibits a positive temperature coefficient whose value is dependent
upon conductivity type, but relatively independent of
device construction. The TCR itself does exhibit some
temperature dependence, which is more pronounced
for high voltage devices. For n-channel, 400 V
devices this coefficient varies from approximately
0.6%/oC at 25°C, to a value of approximately
1.2%rC at 150°C. For purposes of this analysis, the
non-linear behavior of the TCR is not considered,
and the high telnperature value is assumed for all
devices. The drain-source resistance as a function of
temperature, Rds(T), is thus
The popularity of the power MOSFET arises from its
low drive requirement, fast switching speed, and the
relative ease with which it may be paralIeled to
achieve higher power ratings. Since manufacturing
yields are presently such that single devices with ratings above approximately 5 k VA are not economicalIy
justifiable, it is this latter advantage which permits
employing the device in applications above a few kW.
Paralleling can be done by the manufacturer at the chip
level, or by the user at the packaged device level.
This paper considers the influence of the temperature
dependence of the drain-source resistance on static
current sharing among paralleled devices. The greatest
effect of thermal forcing is achieved when the paralleled devices are thermally isolated from one another,
thus permitting them to equilibrate at different temperatures. In practical applications complete isolation can
only be approximated, in part due to the physical
constraints on equipment size and component
arrangement, and in part due to the conflicting
requirements imposed by rhe need to eliminate electrical parasitics.
»
RdsCT) = RdsoCI+ACTj-Ta
(1)
where Rdso is the drain-source resistance measured at
Ta , and A is the TCR.
5-43
-
Because the current distribution among paralleled
devices is governed by their relative drain-source resistances, it is often believed that the positive temperature coefficient of resistance forces current sharing
among paralleled devices. [1,2,3] In order to determine the extent to which such forced sharing occurs,
it is necessary to assume a thermal system and solve
the resulting set of non-linear equations for the individual drain-source resistances.
tions. Therefore the sources PI and P2 in Fig. I each
depend on both Tjl and Tj2.
The Thermal System
Electrical Connection of Paralleled Devices
Figure 2
Figure I shows the static thermal model for two
devices in parallel. The resistances Rjc, Rcs , and Rsa
represent the "junction" to case, case to sink, and
sink to ambient thermal resistances, respectively.
(Although there is no heat generating junction in a
MOSFET, the term is used here generically to represent the heat source.) The voltages Tj, Tc , T s , and Ta ,
represent the junction, case, sink, and ambient temperatures, respectively. The resistance Rc represents
the thermal coupling between devices. For mUltiple
chips on a common header, the coupling occurs between the case nodes in the model. For two packaged
devices on a single heat sink, the coupling occurs
between the sink nodes in the model. The current
sources PI and P2 represent the rate at which thermal
energy is being dissipated in devices I and 2, respectively.
The input parameters for this problem are the net drain
current, I, and the reference drain-source resistances
Rdsol and Rds02. The variables Tjl, Tj2, II, 12, PI:
and P2 are then determined. For symbolic simplicity,
the subscripts "ds" are not carried forward in the
following analysis and discussion. The operating values of the drain-source resistance are referred to as RI
and R2, while the initial, or reference, values are Rio
and R20. Also, literal subscripts are used for thermal
resistances, while numerical subscripts are reserved
for electrical resistances.
The model of Fig. I can be generalized slightly by
recognizing that the point of thermal coupling can be
moved by varying the relative values of Rjc, Rcs , and
Rsa. Therefore, in the equations that follow, Rjc, Rcs
and Rsa have been combined into an equivalent resis:
tance Rct which represents the thermal circuit between
the point of coupling and ambient, and an equivalent
resistance Ra which represents the thermal resistance
between the heat source and the point of coupling.
This more general thermal model, in which chip paralleled or package paralleled devices may be represented by simply changing the relative values of Ra
and Rd, is shown in Fig. 3.
Ric
Rc
- -'\NV-Res
T.,
Rc
- -¥I\r- Rsa
Ts2
Rsa
T"
Static Thermal Model for Two Paralleled Devices
Figure 1
P2
The electrical connection of the devices is shown in
Fig. 2, where Vds is the on-state drain-source voltage,
and I is the net drain current. The non-linearity in this
system arises because the distribution of the net current, I, between the devices depends upon the relative
values of Rdsl and Rds2, as do the relative dissipa-
Rc
Generalized Static Thermal Model
Figure 3
544
The model of Fig. 3 may now be analyzed. In terms 'of
the temperatures of the coupled nodes, Tel and Te2,
the junction temperatures are
Tjl = PIRa + Tel
(2a)
Tj2 = P2Ra + Te2
(2b)
The electrical dissipation in each device, as a function
of net drain current is
(3a)
Te2 = [P2(Rc + Rd) + PIRd]
[(Rc/Rd) + 2] + Ta
(3b)
(4a)
P2 = [I(RIII R2)]2/R2
(4b)
As indicated earlier, the on-state drain-source resistances are
The temperatures of the coupled nodes, relative to
O°C, may now be determined in terms of PI and P2,
[PI(R c + Rd) + P2Rd]
T I - --:-----e [(Rc/Rd) + 2] + Ta
PI = [I(RIIIR2)]2/Rl
RI = Rlo(1 + ATjl)
(Sa)
R2 = R20(1 + ATj2)
(Sb)
Finally, (2), (3), (4), and (S) can be combined to yield
R I and R2 as functions of the net drain current, I, and
the thermal environment,
R, A[R{~+ R'~ ~,+ 1)+ R'}
+ R]
R-R{, "R, R, A [ f~+R')( ~'+ ')+ R'}
+ R2
RI = RIo
2-
20
{
]+
hi
(R] + R2)
2
R]
+
(R] + R2)2
d
~+2
Rd
Rd
Rc+ 2
Rd
Rd
R + 2
_c
Rd
]}
]}
(6a)
(6b)
Results
For a given set of input parameters, (6a) and (6b) are
solved numerically using an iterative algorithm. Both
T0220 and T03 packages were considered for individually packaged devices in parallel, and for individual chips in parallel a Rcs value of one-half that for
the package was assumed. This latter is a convenient
assumption since it is satisfied automatically as Rc is
"moved up" toward the "junction" in the model of
Fig. 3.
Case 2: (400 V, 3.S A, T0220)
Rjc = 1.6rC/W
Rcs = I.OO°C/W
Rsa = 1.47°C/W
RIo = 0.7S ohm
R20 = 1.00 ohm
I=7A
A= .011
Although a wide variety of different conditions were
studied, the essential conclusions of this work can be
demonstrated by the results of a few specific cases. In
each case it is assumed that the devices are paralleled
at the chip level, Le., Ra = Rjc. The parameters for
these cases are:
Case 3: (400 V, 11 A, T03)
Rjc = .83°C/W
Rcs = .20°C/W
Rsa = .60°C/W
RIo = 0.2 ohm
R20 = 0.3 ohm
I = 22 A
A= .012
Case I: (100 V, 10 A, T0220)
Rjc = 1.67°C/W
Res = I.OO°C/W
Rsa = 1.47°C/W
Rio = 0.12 ohm
R20 = 0.16 ohm
I = 20 A
A= .0067
Equations (2)-(6) were solved for Rc varying from 0
to 100°C/W. The effect of forced current sharing is
maximized for large Rc, which permits the necessary
temperature difference between devices. This result
is, of course, independent of the relative values of Ra
and Rd.
545
-
T('c)
TI('c)
'6.
'05
'50
__----------t
~T"
------------r
~T,2
95
90
'20
85
80'--------L..-----.....J,O-~ ~
RcrcJw)
'0
(a)
(a)
I(A)
I(A)
:8i!~~;:',:
L - - - - - - - ' - - - - - - - L1'o--?
'2
11
'0
~RcfcJw)
12
'--------L..------L,o-~ ~
(b)
(b)
Variation of "Junction" Temperature (a),
and Drain Current (b)
with Thermal Coupling - Case 1
Figure 4
Variation of "Junction" Temperature (a),
and Drain Current (b)
with Thermal Coupling - Case 3
Figure 6
'20
Figures 4, 5, and 6 show the variations of "junction"
temperature and drain current with thermal coupling.
The results show the effect of forced current sharing to
be practically non-existent. In fact, if the Tj's are
constrained to be equal (by setting Ra = 0, Rd = Rjc
+ Rcs + Rsa , and Rc = 0), the drain currents differ
by less than 5% from their values for the case of total
thermal isolation (approximated by Rc = 100°C!W).
Even though current sharing is not enhanced by thermal isolation, the figures show that the "junction"
temperatures can be substantially different under such
conditions. Reliability of the hot device is thus compromised with no apparent benefit.
110
'05
~T,2
'OO'-------+------:l'O:--~
100 II'
"
RcC°clw)
(a)
I(A)
42:~,------+-~~ ~I,
t
L.._------------T ~
':0
<---" •
1Jo II Rcf'cIw)
(b)
Variation of "Junction" Temperature (a),
and Drain Current (b)
with Thermal Coupling - Case 2
Figure 5
546
Conclusions
References
Since the continuous drain current specification for a
power MOSFET is a reflection of the maximum junction temperature limit, even though II in Figs. 4(b),
5(b), and 6(b) exceed this specification the device is
quite happy as long as Tj < Tjmax. One is therefore
led to the conclusion that forced current sharing is an
illusion, and furthermore, that devices to be paralleled
need not even be carefully matched for Rdso. Also, it
is clear from the analysis that paralleling can be used
to its best advantage if chips are paralleled by the
manufacturer using techniques which result in the
minimum value of Rc.
[1] HEXFET Databook, International Rectifier Corporation, 1981, p. 20.
[2] Oxner, E.S., Power FETs and Their Applications,
Prentice-Hall Inc., N.J., 1982, p. 8.
[3] MOSPOWER FET Design Catalog, Siliconix,
1982, p. 6-4.
5·47
5.3.3 An Analysis and Experimental
Verification of Parasitic Oscillations
in Paralleled Power MOSFETs
©
1984 IEEE. To be published in IEEE TRANSACTIONS ON ELECTRON DEVICES, Vol. ED-31, No.
7, July 1984. This appears here, with pennission of the IEEE.
Abstract
An analysis of the small signal dynamic model of the
power MOSFET is presented which predicts the existence of high frequency parasitic oscillations when
these devices are electrically paralleled. It is shown
that the existence of these oscillations is a strong
function of the small signal transfer admittance, gm,
and the differential mode drain, gate and source resistances. The sensitivity of the oscillations to these
parameters is determined. Experimental data verifying
the qualitative aspects of the analytical results is
presented. It is concluded that the problem is potentially most severe for devices which are paralleled
by the manufacturer at the chip level. A practical
solution to the problem is the introduction of differential mode gate resistance, either as lumped components, or by the use of polysilicon overlays.
and have had observed amplitudes exceeding the gatesource breakdown voltage. The existence of the oscillation is a strong function of the gate drive source
resistance, but for reasons to be shown, this is believed to be due to the extended duration of the
saturation region transition for larger drive resistance.
It will also be shown that, given a saturation region
residence of sufficient duration to support an oscillation, the inception of this oscillation is a function
of only the differential components of the gate and
drain resistance, and the parasitic drain and source
inductances.
An earlier papert: 4] has shown the theoretical effect
of gate and drain impedances on the parasitic phenomenon. This paper extends the previous analysis
to include the effect of source impedance, and verifies
the theoretical results experimentally.
Introduction
Anecdotal user reports have indicated "mysterious"
failures of paralleled power MOSFETs. At the same
time, the appearance of oscillations during the
switching transitions of such paralleled devices has
been reported in the literatureJi,2,3] The understanding of this behavior is of importance not only
for reliably paralleling packaged devices, but also for
successfully fabricating large devices using paralleled
chips. In this latter case, since the user has fewer
remedial options available (because of the lack of
multiple gate access), the responsibility rests with the
manufacturer to assure that the device is stable under
all operating conditions.
The Dynamic Model
Figure 1 shows two electrically paralleled power
MOSFETs, Ql and Q2. It is assumed that these
devices are identical, and that the layout of their
interconnection is symmetrical. Since oscillations occur only when the devices are in their active (saturation) region, it is the incremental model of the
circuit of Fig. 1 operating in this region which is
important. This model is shown in Fig. 2, which also
includes parasitic elements necessary to the creation
of an oscillatory system. The inductance, LG, and
resistance, RG, represent the parasitic impedance between the common drive source connection and the
gates of the two devices. The inductances LD and
LS represent the parasitic drain and source impedances, respectively. The resistor RD represents the
Phenomenologically, the oscillations occur at frequencies between 50 and 250 MHz, are observable
at both the gate and drain tenninals of either device,
5·48
incremental drain-source resistance, which is primarily
the resistance of the drift region of the drain. Because
of the assumed symmetry of the problem, the parasitic
elements are disposed symmetrically between the two
devices. The absence of parasitic elements in series
with RS and RL is justified below.
The same is true of any parasitic elements which
might have been included in these branches in the
model of Fig. 2.
LO
Incremental Model, Including Parasitic Elements lor Saturation
Region Operation 01 the Paralleled Devices of Figure 1
Figure 2
A Pair 01 Electrically Paralleled MOSFETs
Figure 1
The differential mode circuit model of Fig. 3 may
now be analyzed to determine its characteristic polynomial. However, rather than confront this rather
complicated analysis, two separate situations are considered. The first, shown in Fig. 4(a), assumes no
parasitic source inductance, and the second, shown
in Fig. 4(b), assumes no parasitic drain inductance.
In both cases the effects of gate resistance and parasitic gate inductance are determined. In this way it
is possible to gain some insight to the oscillatory
behavior as a function of the parasitic elements. The
results of the detailed analysis of the effects of drain
resistance is presented only for the case LD = O. It
is believed that this situation will be more commonly
encountered in practice because the physical location
of the drain contact (chip substrate) results in very
low differential drain inductance (compared to the
corresponding differential source inductance) when
devices are paralleled. For example, the heat sink
will frequently be the common drain connection for
T03 or T0220 packages. This produces an electrostatic
geometry exhibiting a low inductance.
Because of the symmetry and linearity of Fig. 2, the
problem may be transformed into differential and
common mode coordinates and the method of half
circuits used to obtain the relevant characteristic
equations. Establishing the plane of symmetry and
opening all circuit branches cutting this plane quickly
establishes the fact that, to the extent that a single
device operates stably, the common mode response
of the model cannot include oscillatory behavior. Essentially, the two devices are decoupled in the common mode. Thus the oscillations must be a result of
the differential mode circuit.
The differential mode circuit may be derived by recognizing that all nodes on the plane of symmetry
are at incremental ground potential. The resulting
model is shown in Fig. 3. It is now clear that since
RS and RL are common mode elements that do not
appear in the differential mode circuit, they will not
have any effect on the circuit's unstable behavior.
LG
C,
v,
+
RG
RO
RG
Lo
LO
l1li
Lo
C,
RO
c,
C,
RL
LS
Differential Mode Circuit for the Model of Figure 2
Figure 3
5-49
+
V2 •
(b)
(a)
Oifferential Mode Circuits for the Model of Figure 2 Assuming: (a) LS
Figure 4
= 0, (b) LO = 0
The differential mode circuit model of Fig. 4(a) is
analyzed to determine its characteristic polynomial,
which is
(la)
S4(LDLGC~) +
S3(RGLDCe2+gmLDLGC2) + S2(LG (Cl + C2) + LD(C2 + C3) + gmRGLDC2)
+ S (RG(CI + C2» + 1 = 0
where
C~ =
Cl C2 + C2C3 + C3Cl
A similar analysis of Fig. 4(b) yields
(lb)
S4 (LGLSC~) + S3 «RL)eC2 + gmLSLGC2) + S2(LG (Cl + C2) + LS (CI + C3) + gm (RL)eC2
+ RGRDC~ + S (RG(CI + C2) + RD (Cl + C3) + gm (LS + RGRDC2» + 1 = 0
where (RL)e
=
RDLS + RGLS + RDLG
of (la) is plotted in Fig. 5 as a function of gm for
several combinations of LD and LG.
These polynomials may now be evaluated for the
existence of right half plane zeros by use of the
Routh-Hurwitz criterion.
'00
Theoretical Results
Characteristics typical of a 100 V, 10 A power MOSPET were selected for use in evaluating (1). These
characteristics are:
CI = 700 pF
C2 = 100 pF
C3 = 300 pF
The Routh-Hurwitz criterion was then applied for
values of gm between 0.1 and 10 mhos, parasitic
inductances between .05 and 50 nH, and parasitic
gate resistances between .01 and 100 ohms. It was
felt that the selected range of parasitic inductance
encompassed both wiring inductance for package paralleled devices, and bond wire inductance for chip
paralleled devices.
0"0"'".'---'0'-::2--:':0.:--7:,0:---:2=-0--:''::-0--;'::00~2o.0
gm(S)
If the drain resistance, RD, is assumed to be zero,
Minimum Value of RG Necessary to Assure Stability of the
Model of Figure 4(a) with RO
0, as a Function of gm'
for Various Values of LO and LG:
(a) LO = 5 nH, LG = 5 nH
(b) LO 20 nH, LG 5 nH
(c) LO 5 nH, LG 20 nH
(d) LO = 20 nH, LG = 20 nH
Figure 5
it was found that for any combination of LD, LG,
and gm, some value of RG could be found below
which (la) possesses a pair of rhp zeros. This minimum value of RG necessary to assure no rhp zeros
Figure 6 shows the results of solving (lb) (non-zero
source inductance), again assuming RD = O. The interesting conclusion of this calculation is that for a
Behavior with RD
=
=
=
=0
5·50
=
=
which rhp poles can exist as a function of RD for
RG = o. Since the assumed device capacitance parameters are typical for devices with widely differing
RDS(on) specifications, the results of this analysis
imply an aggravation of the oscillation problem for
low voltage devices, where RDS(on) is typically
fractions of an ohm.
'
ratio LG/LS greater than approximately 2.6, (Ib)
possesses no rhp roots.
'00
'0
Effect of RS
In a system exhibiting a pair of rhp poles, the resulting
oscillation will grow exponentially with a time constant proportional to its Q. The higher the Q, the
slower will be the growth of the oscillation amplitude.
Therefore, the duration of validity of the model of
Fig. 2 is critical in determining whether these oscillations will be observed. In this respect, the gate
drive source resistance, RS, of Fig. I is important
because it, together with the device input and reverse
transfer capacitances, Ciss and Crss , will determine
the rise and fall times of the drain variables, and thus
the duration of time spent in the saturation region.
This is consistent with experimental results which
have shown oscillations to appear in the presence of
relatively large gate drive source resistance, but to
disappear when the drive resistance was reduced.
02
0'
0050' - , -....
02--'"'"'--','-,-"'"20--5'-0--'''-'----'200
gml S)
Minimum Value of RG Necessary to Assure Stability of the
Model of Figure 4(b) with RD = 0, as a Function of gm,
for Various Values of LG and LS:
=
=
=
(a) LG 5 nH, LS 5 nH
(b) LG = 5 nH, LS 20 nH
(c) LG = 20 nH, LS = 20 nH
Figure 6
It is clear that under the assumption RD = 0, the
model of Fig. 2 predicts unstable behavior for a wide
range of practical values of the parasitic parameters.
Experimental Results
Behavior with RD :f. 0
A series of experiments was performed to verify the
qualitative aspects of the analytical results presented
above. It is first shown that the oscillations appear
primarily in the differential mode circuit. A critical
value of the ratio LG/LS is observed, and the effect
of RD to restrict the range of gm over which oscillations are possible is verified.
Figure 7 shows the results of applying the RouthHurwitz criterion to (Ib) for LG = LS = 20 nH.
The minimum value of RG necessary to assure no
rhp roots is plotted as a function of gm for various
values of RD. These curves show that as RD increases, the region of gm over which oscillations are
possible decreases, until at some critical value of RD
no rhp roots will exist even if RG = O. For this
specific example, this critical value is 1.9 ohms. Figure 8 shows the behavior of the range of gm over
14
\
\
\
--
12
"
em(S)
Rolli)
2'
Theoretical Range of gm Over Which Oscillations Persist,
as a Function of RD for RG = 0 and LG = LS = 20 nH
Figure 8
Minimum Value of RG Necessary to
Assure Stability of the Model of Figure 4(b),
as a Function of gm for Various Values of RD
Figure 7
5·51
Figure 9 shows the schematic of the experimental
circuit. A pair of 1RF530's were placed adjacent to
each other on a common heat sink. The pulsed gate
source consisted of a DS0026 clock driver being
driven by a 74LS123 monostable. The load resistor
was a 2 ohm non-inductive Dale NH250 and the
clamping diode was a fast recovery UES1403. The
gate drive source resistor, RS, was adjusted to ensure
that the saturation region residence time was long
enough to permit the oscillations to develop. In addition to switching tests, steady-state oscillatory behavior was observed by placing the devices in their
saturation regions using a DC gate drive.
211
(a)
(b)
UES
1403
(c)
Circuit Used to Experimentally Verify Theoretical Conclusions
Figure 9
To verify that the oscillations are differential mode
in nature, an explicit differential source inductance
was created by connecting the source terminals together with a small loop of wire (r = 0.35"). Connection was then made to common mode ground from
the middle of this loop with a relatively long (3")
piece of wire. The voltage waveform along the loop,
with respect to common mode ground, was then observed while the devices were switching and exhibiting the parasitic oscillation. Figure 10 shows these
waveforms and the corresponding measurement location referenced to the connection at the middle of
the loop (100% is the source lead at the package).
It is apparent that only a small fraction of the parasitic
voltage shows up across the common mode source
inductance, even though it is substantially larger than
the differential mode source inductance, thus confirming the theoretical conclusion that the parasitic
oscillation is a differential mode phenomenon.
(d)
Differential Source Voltage Measured at
Different Points Along the Differential Source Inductance;
(a) 10%, (b) 25%, (c) 75%, (d) 100%
Figure 10
The source loop was next removed and replaced by
a short, direct connection. The gate-to-gate connection
was then made with a loop whose size was successively increased until oscillations ceased, thus confirming the theoretical prediction of a critical ratio
LolLS above which there are no rhp poles. Although
a quantitative measurement of the differential source
and gate inductances was not attempted since a component of these inductances is internal to the package
and not accessible, geometric approximations suggest
that the observed critical ratio was approximately 2.
5·52
proximately constant. The measurements, however,
were made at relatively low values of VD where the
manufacturer's data shows Ciss, Crss , and Cos s
changing quite rapidly as a function of VDS. Analysis
has shown that both of these phenomena will give
rise to an increase in the range of gm over which
oscillations can occur as VD is increased, a conclusion
consistent with the qualitative aspects of Figs. 8 and
11.
The source and gate connections were next both made
with short straight wires and the pulsed gate drive
was replaced by a DC source. The gate voltage was
then increased from zero until oscillations were observed and the corresponding drain current recorded.
Increasing the gate voltage further eventually caused
the oscillations to cease. The drain current at this
point was also recorded. The transconductance, gm,
at these points was then obtained from measurements
or the manufacturer's data sheets, depending on the
current level. It was observed that the resulting range
of gm was sensitive to the applied drain voltage,
Vcc. The range of gm supporting oscillations was
thus determined as a function of Vcc and is shown
in Fig. 11.
The quantitative discrepancy between the analytical
and experimental results can be attributed to several
causes. The first is that the parameters for the model
of Fig. 2 were measured at a frequency of 1 MHz,
whereas the observed frequency of oscillation was 65
MHz. Although the distribution and value of the small
signal energy storage elements are not likely to vary
much between I and 65 MHz (because they are dominated by the MOS and junction structures, neither
of which has strong frequency dependencies in this
range), parasitic loss mechanisms within the device
will likely exhibit a strong dependence upon frequency, with most of them producing losses which
increase with frequency. For instance, the channel
body resistance was not taken into account in the
model, although Oxner has shown this to be an important component of Ro at high frequencies. [5]
These additional parasitic losses will decrease the
analytically predicted range of gm over which oscillations may persist, as can be inferred from Fig. 7
for the case of an increasing RG. An additional reason
for the discrepancy is that measurements showed LG
and LS to be approximately 50 nH, whereas the
analysis resulting in Fig. 8 assumed values of 20 nH.
9m(S)
Experimental Range of gm Over Which Oscillations Persist,
as a Function of Vcc
Figure 11
Conclusions
Discussion
An incremental, differential mode circuit model has
been used to analytically predict experimentally
observed oscillations in paralleled power MOSFETs.
The oscillations depend upon the differential mode
parasitic gate, drain, and source impedances, and not _ _
upon the common mode elements most accessible to
the circuit designer. The region of potential oscillation
increases as RD decreases, suggesting that the parasitic oscillation problem may be most severe for low
voltage, high current devices. It is shown that the
introduction of differential mode gate resistance can
eliminate the potential for instability, and that increasing switching times can aggravate the tendency
to oscillate.
The experimental results reported above are essentially
consistent with the theoretical analysis. The major
discrepancy is between the theoretically predicted and
experimentally observed ranges of gm over which
oscillations can occur. For the IRF530 the maximum
specified value of RDS(on) is 0.18 ohms, which,
from Fig. 8, implies a very large range of gm based
on the theoretical analysis. However, Fig. 11, which
is based on measurements made for RG = 0, shows
that the observed range is quite restricted, especially
for low values of drain voltage. The range also increases rapidly as the drain voltage is increased. There
are two plausible explanations of this latter behavior.
The first is that the lightly doped drain region of the
device is partly depleted when VDS > O. This depleted region increases as VDS increases, resulting
in a reduction in the contribution of the n-drain region
to the bulk drain resistance. The second is that the
assumed values of the capacitances in the model of
Fig. 2 were those for the devices operating deep in
the saturation region where the capacitances are ap-
These results are particularly significant for chip paralleled devices, for in this case the small parasitic
inductance could give rise to very high frequency
oscillations which could build up quickly relative to
even "fast" switching times. In addition, these oscillations may not be observable because all the elements of the differential mode model of Fig. 3 are
5·53
inside the package. This suggests that devices to be
paralleled at the chip level be fabricated using silicon
gates of an appropriate sheet resistivity, or that small,
discrete resistors be inserted in series with the gates
of chips fabricated using metal gate technology.
[1]
For package paralleled devices, these results support
the use of differential gate resistors [2,3] and gate
drives which achieve fast switching times. In addition,
the existence of a critical LGILS ratio supports the
use of ferrite beads on gate leads to increase Lo.
[3]
References
[2]
[4]
Acknowledgement
The authors are grateful to Kurt Ware and Tom Lee,
who as undergraduate students contributed to initial
aspects of this work. This research was supported. in
part by unrestricted grants from Lutron Electromcs
Co., and the Gould Foundation.
[5]
5·54
HEXFET Databook, International Rectifier Corporation, 1981, p. 20.
Sloane, T.H., H.A. Owen, Jr., and T.G. Wilson, "Switching Transients in High Frequency
High Power Converters Using Power MOSPETs," IEEE PESC Record, 1979.
Oxner, E.S., Power FETs and Their Applications, Prentice-Hall Inc., N.J., 1982, p. 129.
Kassakian, J.G., "Some Issues Related to the
Behavior of Multiply Paralleled Power MOSFETs," Proceedings of IPEC-Tokyo, March
1983.
Oxner, E.S., "Meet the V-MOSPET Model,"
RF Design, Jan/Feb 1979, pp. 16-22.
5.3.4 Power FET Paralleling
Introduction
Paralleling of power MOSFETs is not aLways as easy
as it may seem. When all the devices are fully
enhanced (fully ON), the probLem is minimaL both
because the toLerance of RDS(on) is small and because the positive temperature coefficient of FET
conduction resistance tends to enforce proper current
sharing through junction temperature differentiaLs.
During the switching period, however, current sharing can be a probLem. If this probLem is not attended
to, the surge current in one of the devices during
switching may exceed the ratings of the device. This
is especially true for high-frequency high-puLse-current applications. In this paper, the causes of dynamic
current imbaLance will be presented and recommendations for alleviating dynamic current imbaLance will
be suggested.
constant will take more than its share of current
during switching.
Device Parameter Values
Typical gate-to-source threshold voltages VGS(th)
for commercially available devices range from 2 to 4
V. Transconductance, gfs, depends on the device
current rating. In general, the larger the current
rating, the larger the gfs. For 5-A devices, the values
range from 1 to 3 S. For 10-A devices, they range from
5 to 10 S. The magnitude of gate capacitance Ciss,
which is the sum of gate-source and gate-drain capacitances, depends on the chip size. The higher the
voltage and current ratings of the device, the larger
the chip size, and, consequently, the larger the gate
capacitance. Typical values of Cjss for commercially
available devices range from approximately 5000 pF _
for a 500 V, 20 A device to 150 pF for a 100 V, 3 A
device. The tolerance of Ciss for a given device is
normally within ±30%.
Dynamic Current Sharing
The causes of dynamic current imbaLance can be
classified into two categories: device reLated and
circuit related. Device parameters that directly affect
dynamic current imbalance include gate threshold
voltage (VT), transconductance (gfs), and input capacitance (C g). Circuit parameters critical to dynamic
current imbalance include the parasitic inductances
of both the gate-drive circuit (LG) and the power
circuit (Lp) as well as the gate-drive circuit source
resistance (RG). The reason both VT and gfs are
involved in current imbalance during switching should
be obvious. Devices with lower VT and/or larger gfs
will carry more current at any V GS in their linear
region. Gate parameters Ciss, RG, and LG affect the
time required to charge the gate capacitance from a
gate-voltage source with a fixed impedance. A FET
with a gate-drive circuit that has a smaller time
Temperature Effects
Device gate-source threshold voltage decreases with
temperature. A typical device threshold decreases
aproximately 5 mV per 0c. Transconductance, gfs,
also decreases with temperature. The temperature
coefficient is - 2% per °C.
It can be seen from the above characteristics that if
one of the paralleled devices heats up due to current
imbalance, the negative temperature coefficient of
VT tends to aggravate while the negative temperature
coefficient of gfs tends to alleviate current sharing
problems during switching. This is different from
5·55
conduction state current sharing in which both the
positive temperature coefficient of channel resistance
RDS(on) and the negative temperature coefficient of
gfs contrIbute to reduction of current imbalance.
duration of current imbalance during switching. To
increase the drive circuit speed, one should reduce
the gate-drive impedance (see Figure 1). It should be
noted, however, that extremely small values of RG
may not be enough to damp undesirable oscillation in
the paralleled gate circuits. Parasitic inductances
around the gate circuit must also be minimized.
Recommendations for Controlling Current
Sharing Problems
There are basically two approaches to control dynamic current imbalance: by means of device parameter matching and by means of circuit techniques.
Device Parameters Matching
Dynamic current imbalance can be minimized by
matching the three device parameters described in
the previous section, i.e. VT, gfs, and Ciss. The
sensitivity of current imbalance to each parameter is
different, however. Of the three parameters mentioned, matching of gfs and Ciss is less critical. For
transconductance mismatch, the worst case current
imbalance is limited to the percentage of tranconductance mismatch. Furthermore, if a device heats up due
to mismatch, the negative temperature coefficient of
transconductance tends to reduce the imbalance. The
effect of input capacitance mismatch depends on the
external gate-drive impedance as described in a following paragraph.
...
Lp
LS
FET Switching Circuit with Two Devices In Parallel
Figure 1
Of the three device parameters to be matched, the
most critical one is the gate threshold voltage VT.
The amount of current imbalance due to VT mismatch
is essentially unlimited. The negative VT temperature
coefficient further worsens dynamic current imbalance.
Conclusions
The causes of current imbalance in paralleled MOSFET
switching power circuits have been examined, and
some techniques have been recommended for improving imbalance. It has been shown that some
parameters are more critical than others in affecting
the imbalance. Threshold voltage VT mismatch, for
example, can be very critical to dynamic current
sharing. The amount of current imbalance depends
both on device parameters and circuit parameters.
Reference 1 gives several numerical examples of the
degree of current imbalance for some specific power
MOSFETs.
Circuit Means for Controlling Imbalance
There are several circuit techniques recommended
for minimizing dynamic current imbalance. All, however, result in some additional alterations to circuit
performance. Thus, trade-off decisions must be made
by the circuit designer.
Gate Drive Circuit
The optimum circuit technique for improving current
sharing during switching is merely to increase switching speed. By increasing the gate-drive circuit speed,
one can reduce the time required to charge and
discharge the gate capacitance and thus reduce the
Reference
Forsythe, James, "Techniques for Controlling Dynamic Current Balance in Parallel Power MOSFET
Configuration," 1983 Powercon 8.
5·56
5.4 dVns/dt Turn-On in MOSFETs
(TA84-4)
Introduction
Under certain conditions, when a rapidly nsing voltage waveform
IS applied from drain to source of a supposed "off" MOSFET,
the device will turn-on. dV/dt turn-on is well known in SCRs
and bipolar junction transistors (BITs). This phenomenon in
MOSFETs is similar in many respects except that dV/dt turn-on
occurs at much higher rates than it does in either SCRs or BITs.
Because spurious turn-on can be destructive, it is of serious
concern in those applications where it might occur.
DRAIN
eSTRAY
GATE
Several different modes of dVDS/dt turn-on have been identified,
and in most cases, turn-on may be eliminated or the effect of turnon may be reduced to an acceptable level.
npn
RG
Ro
Lo
To eliminate this problem, the mechanisms of the turn-on modes,
the consequences of spurious turn-on, the means for reducing or
eliminating turn-on, and the applIcations where turn-on is likely
to be a problem, must be understood. The following discussIOn
intends to provide the basic mformation on dVDS/dt turn-on
needed by the design engineer.
Vo-=-
)~,
;-
SOURCE
N Channel MOSFET Equivalent Circuit
Figure 1
Equivalent Circuit Model
drive circuit, an equivalent senes battery (VD) IS present. In some
cases, a DC bias may be present, so V D could be either positive or
negative. Cstray represents the parasitic capacitance from drain to
gate due to the circuit layout. This capacitance may be as large or
larger than the intrinsic Cgd if careful layout is not used.
Figure I is an equivalent circuit model for a power MOSFET
which includes the parasitic elements within the device and withm
the drive circuit (R D , LD, Vd ).
The parasitic NPN BIT, an integral part of the device structure, is
present in all VMOS and DMOS devices. To minimize the effect
of the BIT, Rb is made as small as possible, and HFE is low.
Nonetheless, in some circumstances the BIT can be turned on.
RG is the mternal gate resistance. The value of this resistance varies
from 0.05 to 5.00 depending on the manufacturer and the size of
the device. Siliconix devices typically run from 0.5 to 4.00.
The impedance of the drive circuit (Zgs) may take many forms, but
the series R, L, battery is representative. RD comes from several
sources: resistance deliberately placed in series to limit the peak
current or reduce the rise time of Vgs , bypass capacitor ESR, wiring resistance, the on resistance of drive switches, etc. LD is due to
the physical size and layout of the drive circuit. If a transformer is
used for dnve isolation, the pnmary-to-secondary leakage inductance will add to LD. When series diodes or BITs are used in the
The mtraterminal capacitances of a MOSFET are functions of
VDS as shown in Figure 2·. Because Cgd and Cds are functions of
VDS , spurious turn-on will depend on both dVDS/dt and VDS ,
whIch complicates the analysis. In the following discussion, the
body-drain diode will be mentioned. This diode is the basecollector junction of the parasitic BIT.
• NOTE: Because C be » C db , Cdb - Cds
5-57
l1li
1000
NO TURN-ON
VTH2
"-..".,.
I'
J
......
LL
E;
W
o
zc:(
!:: 100
.......
Cgs
1""0"
, I
oc:(
D..
c:(
o
o
"
1
VOS
"
'"
10
r--...~
(A)
Cds
IB)
Mode 1 Equivalent Circuit
Figure 3
Cgd
100
VOLTS
Intratermlnal Capacitance
Figure 2
The following modes of dVDs/dt turn-on have been observed:
I. Device quiescent: Zgs high, MOSFET turns on.
2. Device quiescent: Zgs low, BJT turns on.
3. Body-drain diode conducting: Zgs low, BJT turns on.
4. Body-drain diode conducting: Zgs low, BJT goes into
avalanche breakdown.
Mode 1 Spurious MOSFET Turn-On. Rg = 10 ohms
Figure 4
Mode 1 Turn-On
The equivalent circuit for mode I turn-on is given in Figure 3 along
with a generalized waveform for Vgs. The exact nature of the Vgs
waveform will be a function of Zgs and Cstray, but in general, the
lower the impedance presented by Zgs and the lower the value for
C stray , the lower the peak value of Vgs. If Vgs exceeds V'h, the device
will turn on. In most applications, the amplitude of I D , during
turn-on will be limited by RDS(on). This mode of turn-on is not
usually destructive, although it will increase the power losses.
Figures 4 and 5 show actual turn-on waveforms when Zgs
is resistive.
10Vi
Vos ,...
1m
~
- - L \,
,...
-
Even when the FET is not turned on, a pulse of drain current will
be present due to the intraterminal capacitances. An idealized
waveform for this case is shown in Figure 6A, and a more realistic
waveform is shown in Figure 6B. There will always be some series
inductance which, when combined with the nonlinear capacitances, produces a distorted, damped sinusoid.
~ h..
~
A.
,...
500lmV
SinS
Mode 1 Spurious MOSFET Turn-On. Rg
Figure 5
This current pulse will be reflected into the switch which is generating the positive dVDS Idt.
= 100 ohms
Jl
One additional complication to mode I turn-on has been suggested
but not observed. If the FET is turned on, it is possible for parasitic
oscillations to occur. This would raise Vgs and increase the power
dissipation. It is possible for the gate to be ruptured ifVgs becomes
large; however, if oscillation is not present during normal switching, it is unlikely that it will occur during mode I turn-on.
The best way to detect potential mode I turn-on is to observe the
gate-to-source voltage waveform with an oscilloscope. If a significant voltage spike appears during the drain voltage transition then
some positive action will have to be taken to reduce the pulse
amplitude.
(A) Ideal
(8) Realistic
Pulse Drain Current Waveforms
Figure 6
5-58
200
Mode 2 Turn-On
The equivalent circuit for mode 2 turn-on IS given In Figure 7,
along with VDS and ID waveforms. The mechanism forthis mode
of turn-on is quite simple. As VDS slews, current will flow through
Cdb • If enough current flows so that the voltage across Rb exceeds
-O.65V, some of the current will flow into the base, turning
on the BJT. In contrast to mode I, the acceptable d VDs/dt rate
is only a function of the device design, not of the external circuit
elements.
OJ
150
'§
100
~
;:;
50
a
The result of mode 2 turn·on can be harmless or cata~trophic
depending on VDS and the magnitude of the base current.
a
100
300
400
IV)
dVosldt Limits in Power MOSFET
Figure 9
DRAIN
+
VDD-
....
Z
w
C!
W
~
'j
200
VOLTAGE
Mode 3 and Mode 4 Turn-On
0:
0:
0
::l
U
>
Frequently the body-drain diode is used as a catch diode for an
inductive load. In a few applications, such as sinewave synthesis or
some motor control schemes, the positivedVDS/dt is applied while
the diode is conducting. Figure 10 shows the equivalent circuit for
this condition. Just priorto the application the voltage ramp, IR is
flowing through the BJT in a reverse direction. Much of the current will flow through Rb and the collector-base diode. Even in the
inverted mode, however, the BJTwill have some current gain, and
a portion of IR will flow in reverse through the emitter, as indicated. As a result of this current flow, the device is saturated with
charge, i.e., it is hard on at the moment the dVDS/dt is applied.
z
BVCEO and base current is injected, the device will go into
avalanche breakdown. Furthermore, if ID is not limited by the
external circuit, the device can go into secondary breakdown and
will be destroyed.
BVCEO
BVCER
BVCEX
BVCBO
IC
Mode 3 Equivalent Circuit
Figure 10
Typical voltage and current waveforms for mode 3 operation, with
and without turn-on, are shown in Figure Ii. Figure llA shows
the waveforms for the case where the BJT does not turn-on and
VCE
Breakdown in a BJT
Figure 8
So much for the bad news. Now for the good news. In modern
MOSFETs, the dVDS /dt rate at which mode 2 turn-on can be
initiated is generally very much higher than most applications will
ever see. The dVDs /dt limits for Siliconix devices are shown in
Figure 9. Realize that the rates shown correspond to transition
times of one to three nsec. Very few applications require transitions
this rapid. Even testing at these rates is a formidable problem.
TURN-ON
NO TURN-ON
(8)
(A)
Mode 3 Waveforms
Figure 11
5-59
-
-3---,
thus acts like a normal p-n junction diode. Looking at the waveforms, the sequence IS initiated by a switch closure at t = to' The
current through the diode begins to reverse at a rate (dI/dt) which
is determined by the switch transitIOn time and/orthe circuit series
inductance. If the switch transItion is very rapid, the circuit
Ind uctance will dominate; conversely, if the transition is slow, the
SWItch dominates. During the interval to-t I> Vos changes only
slightly (-I to + I volt). VOS does not change significantly until 10
reaches its maximum value and begins to fall. At t= tl, Vos begins
rising rapidly. The rate of dVos/dt at this point is determined by
the circuit series inductance, the peak current and the recovery
characteristic of the diode. The recovery characteristic is in tum
a function of the initial dI/dt. This relationship is illustrated by
Figure 12. The more rapid dI/dt, the greater will be the peak
amplitude ofIo , Q,. will be larger and trr shorter. More important,
the dID/dt during the interval t l -t2 becomes greater as the initial
dl/dt is increased. In some diodes, dIoIdt can increase dramatIcally and can form a sharp step or snap as indicated in Figure 13.
At lower levels of initial dIo/dt, the step may not be present, but
as the initial dIo/dt is increased, the step appears. This current
step during recovery can generate extremely high rates of dV/dt in
the circuit.
-2---~
-1----0\
t
+
------------~~------------~~~--_
IF
+
VarIatIon In Trr with IF
Figure 14
failure. Notice that the BJT turn-on occurs during the interval tl-t2
and is triggered by the dVDs!dt resulting from the recovery
characteristic of the diode. The dVDS/dt during the interval trt2 is
related to the switch transition time, but it is not a direct function.
This is very different from modes I and 2 where the current
waveforms are coincident with and directly proportional to the
transition time.
+
Mode 3 turn-on is an indirect function of transition time and is
more directly controlled by the circuit and the recovery characteristic of the device. In terms of rating the device for dVos /dt
capability, the appropriate voltage transition is not the switch
transition but rather the transition which appears across the diode
during the second portion of the recovery interval.
Or-------~~----~~~-------T
This complex interaction deserves a more detailed examination.
The following discussion uses a successive approximation to
explain what happens during mode 3 operation.
An equivalent circuit for mode 3, where turn-on does not occur
and the device can be viewed as a diode, is given in Figure 15. Note
that the inductive load current is assumed to be constant during
the recovery interval and is represented by a current source. L I
and L2 are the circuit parasitic inductances.
Variations In Trr and Q rr with di/dt
Figure 12
Or-------~--------
_ V3 +
r-"---,
__~~---------T
SOFT RECOVERY
SNAP RECOVERY
Soft Versus Snap Diode Recovery Waveform
Figure 13
Mode 3 Equivalent Circuit
FIgure 15
A further complication exists: the diode recovery characteristics
are also influenced by the magnitude of 10 during reverse conduction. This relationship is shown qualitatively in Figure 14.
The reverse recovery current waveform (II) can be approximated
by triangular waveforms as shown in Figure 16. From this
approximation, the voltage (V3) across the diode is easily determined, as shown. The critical point in the waveform is the rapid
and large positive transition which occurs when the waveform
Figure II B shows the waveforms when turn-on is present. Two
cases are shown: recovery without damage and catastrophic
5-60
The device could respond to the voltage transient in several ways:
I. !fthe amplitude is small and dVos /dt moderate, the device
recovers normally.
2. Mode 3 turn-on could occur without avalanche breakdown.
3. Mode 4 turn-on could occur; i.e. the device avalanches
before the BJT turns on.
4. Mode 3 turn-on could occur, lowering BVos s , leading to
avalanche breakdown and perhaps second breakdown.
Each of these turn-on scenarios is potentially destructive.
~::::1
dl
21
o'~~-------------------------------
+Vo
The complexity of the mteraction between the parameters makes
it difficult to characterize FETs for mode 3 and 4. Figure 18
displays the results of preliminary testing of Siliconix devices.
Competitive devices were also tested and found to be no better at
best, and much worse in some cases. While this curve is hardly
definitive, it does show that modes 3 and 4 turn-on levels are an
order of magnitude lower than mode 2. The dl/dt levels used in
th~ testing were in the range of 150 A/ usec. This is representative
for motor control or low frequency AC synthesizer; however, LF
transmitters and sonar transducer drivers would be expected to
have much higher dl/dt rates which would further reduce the
acceptable dV/dt.
"2
VI
~
V2
-Vo
2"
-Vo
1-
--------------.
, ,
,, ,,,
,,
,,
,
,,
,
4Vs
,,,
,,
,
___________ I
I
~-~-----------
+1V
Fortunately, mode 3 operatIon is restricted to a few special
applications such as:
I. Motor drives and sinewave inverters where multiple
repetitIve pulsing of the sWItches occurs.
2. Highly reactive resonant loads, such as found in LF
transmitters, sonar transducer dnvers, and induction
heating equipment.
3. Synchronous rectifiers where the body-drain diode is
allowed to conduct.
-1V
Triangular Waveform Approximation
Figure 16
di/ dt reverses. The amplitude of the transition is a function of
di/dt dunng recovery-a diode characteristic. A diode with a
snap characteristic would generate a very rapid, high amplitude
voltage spike.
Figure 16 indicates an infinite dVos /dt, which is not what is
observed. A better approximation would be to use a trapezoidal
waveform as indicated in Figure 17. Figure 17 also shows the
actual waveforms. From this figure, we see that the large voltage
transient still exists. Its amplitude is essentially the same, but the
dVos/dt is finite. We can see that the reverse recovery characteristic of the diode determines both dVos /dt and the amplitude of
the transients.
dV/dl
SLEW RATE
(V/nS)
20
10
o Lo----.J...----.,.10,LO"..%----L-----2...JOLO.-V.----L-----:3...JOO%
PERCENT OF RATEO FORWARD
BODY-DRAIN DIODE CURRENT
Vl'V2{+~~L-------~-'-'-----------------
-vo
Mode 3 Performance
Figure 18
The Effect of Radiation and Tj on dVDS/dt
"2 ____________ ::
-vs
L_.J
4Vs
The effects ofTj and radiation on dVos/dt behavior have not been
discussed. While little testing has been done, trends can be identified. Vth has a negative temperature coefficient of four to eight
mV/cC, so that at high temperatures Vth is lower. This will lower
the rate at which mode I turn-on occurs. At higher temperatures,
Vbe will decrease in the BJT, and the gain will increase. Both of
these effects will reduce the acceptable dVDS /dt rate for modes 2
and 3.
-----------r/;.l
p
f:
V3
"
2Vo ----------,"
,"
Vo ---------+1V
Ionizing radiation has two effects for total doses above
IOKRad(Si). First Vth is reduced, and second the gain and diode
recovery time of the BJT are reduced. This implies that mode I
performance is degraded, but modes 2 and 3 should be improved.
-1V
Trapezoidal Waveform Approximation
Figure 17
5-61
How to Avoid dVos/dt Problems
,1 I "f . .". . f"
T I ~( ,t
In most circUIts, dVosjdt turn-on may be avoided. There are
many ways to accomplish this:
I. Reduce dVosidt
2. Low impedance drive circuits
3. Good circuit layout practice
4. Negative gate bias
5. Series drain diode
6. Snubbers
7. Turn-on of the FET during commutation
8. Use MOSFETs with good dVjdt performance
9. Current fed topologies
e
\
The most obvious way to reduce dVjdt problems is to slow the
switching speed. This can be accomplished easily by adding resistance in the gate as shown in Figure 19. The diode is added to
provide a low impedance when the device is off so that mode I
problems do not arise.
+
TWISTED PAIR,
COAX OR STRIPLINE
ie LOW L CONNECTION
dV/dt Slowdown Circuit
Figure 19
Low Inductance Connections
Figure 21B
The best means to avoid mode I turn-on is to use a low impedance
driver and good circuit layout. Figure 20 illustrates how to design
a good drive circuit. It is also very helpful to reduce the noise and
ringing in the overall circuit.
Mode 3 performance may be improved by reducing or eliminating
the current in the body-drain diode. Two ways to do this are
shown in Figure 22. In (A) the FETchannel is turned on to divert
current away from the body diode. This can be very effective as
shown in Figure 23. Figure 23 is a superposition of the FET and
diode conduction characteristics in a VNE003A. In this particular
device, the diode would not start conducting until 10>30 Amps.
In very noisy environments, it may be necessary for Vgs to be
negative during the switch off interval. The noise coupled to the
gate will have to overcome a potential of Vth + Vbla,.
MInimize Parasitic
Feedback Capacitance --..
.n..
Another method which eliminates diode conduction entirely is
shown in (B). A low voltage diode is placed in series with either the
drain or source, and a normal diode is added in parallel to carry
the commutation current. This approach will eliminate mode 3
turn-on, but there is a penalty in the power cost in the series diode
and in the extra cost of the external diodes.
i
v~v-","---v~,1
~wisted
(
MINIMIZE THE AREA
High di/dt Paths
Figure 21A
PaIr Reduces'
• Inductance
• NOIse Pickup
• Parasitic OscIllatIon
(nneCI Direct To MOSFET Termonal
1
f"'qf
Minimizes Common-Source Inductance
Low Impedance Dnve For
• Fast Switching
• MInimiZe Noise Pickup
• MInimIZe dV/dt Turn-On
Circuit Layout to Reduce Mode 1 Turn-On
Figure 20
Figure 21 shows how to layout an H-Bridge switch-mode power
converter primary to minimize noise, ringing, and coupling from
one point of the circuit to another. The principle is quite simple;
first Identify those paths in the circuit where the switched currents
are flowing and then minimize the inductance and radiating area
of these current loops. These techniques will not only reduce
dVosjdt turn-on but will be beneficial in reducing EMI, eliminating voltage spikes and generally cleaning up the circuit
waveforms.
-
S
I
I
I
(A) CHANNEL ON
SHUNTING CURRENT
+
\..
SERIES DRAIN
DIODE
-
-"\
I
I
I
I
D
~.
r--....I
I
(6) SERIES DRAIN DIODE
Schemes for Reducing Body-Drain Diode Conduction
Figure 22
5-62
Appendix
Another scenario for mode 2 turn-on has been postulated but not
observed. Figure 25 shows a simplified boost converter where the
MOSFET is represented by a perfect switch in parallel with the
intrinsic BJT. When the switch is opened, some of the inductor
current (ILl will flow into Cdb . If IL is large enough, it is conceivable that the BJT could be turned on. The effect on the turn-off
waveform would be similar to storage time in a BJT, i.e. extended
turn-off time.
Body Diode/Forward Conduction Comparison
Figure 23
In extreme cases, snubbers may be used to reduce the dVos/dt,
the transient amplitude and to limit current pulses if turn-on
should occur.
The effect of dV/dt turn-on will depend on the impedance of the
circuit in which the device is being used. For example, the circuit
shown in Figure 24A is a voltage fed switch-mode power converter. When S I is closed, a positive dVos/dt will appear across
S2. If S2 is turned on then a short circuit will appear across the
transformer windmg. There is little to limit the current because the
source impedance is very low. GIven the same circumstances in
Figure 24B, the series inductor will limit the rate of current rise.
The probability of damaging the switches will be reduced.
Postulated Mode 2 Turn-On Mechanism
Figure 25
Conceivably, several turn-off pulses could occur as the BJT turns
on stealing base current. Like all mode 2 turn-on, this effect will
only be possible for very rapid transitions.
A mode of anomalous tum-off ringing has been identified by
Giandomenico (6). This ringing could easily be mistaken for the
mode turn-on just discussed or for parasitic oscillation. The
ringing occurs when there is excessive unclamped drain inductive
and is described by Giandomenico, et aI, as follows:
"When the power MOSFET inverter circuit shown in Figure 26 is
suddenly turned off, the drain voltage rises quickly and then
abruptly drops to the "on-voltage" where it remains for several
2SV
~~'25
N"
QuasI squareware Con~erter
Voo:25V
I
Vo~
-25V
Wirewound
Resistor
117 pH
L
1
1
170
RL
-.J
Voltage and Current Fed Converters
Figure 24
Conclusions
dVos /dt turn-on does exist in power MOSFETs, so the following
conclusions may be drawn:
I. Mode I is primarily a circuit problem which can best be
eliminated by usmg low impedance drives and good circuit
layout techniques.
2. Mode 2 occurs only for very high values of dVos/dt;
therefore, this is seldom a problem.
3. Modes 3 and 4 are potentially the most troublesome but
occur only in specialized applications. Modes 3 and 4 may
be reduced or elimmated WIth circuit modIfications.
The Power MOSFET Connected In an Inverting Configuration
Figure 26
5-63
-
microseconds before ringing and settling to the supply voltage (see
Figure 27 A). This behavior was initially thought to be due to
dV/dt breakdown and was thought to be related to the "on-state"
oscillations. However, this turn-off behavior is in fact due to the
clamping diode between the drain and source which is intrinsically
incorporated into the vertical MOSFET structure. A typical
display of the drain voltage and current during turn-off is shown
in Figure 27B from which the turn-off phenomenon can be
explained.
"Neglecting reverse recovery, the time at which the diode current
is zero is:
ton =
= 7T v'LC
In [2VDD+VF]
RL
(6)
VDD + VF
"For a reasonably large supply voltage, Vdd> > 0.7 volts, the
above equation can be expanded in a Taylor series to give:
ton =
•'When the device is turned off, the magnetic energy stored in
the load inductance is transferred into the output capacitance and
then returned to the inductor with the opposite current polarity.
Neglecting the non-linearity of the output capacitance and the
damping of the output load resistance, the time for the current
reversal is:
t
(~)
(~)
RL
[1 -
~]
ln2
(7)
2V D D
"Neglecting the reverse recovery of the diode for the moment,
the drain voltage rises at this time and then rings and settles to
the supply voltage.
"A comparison of experimental data with the calculated results
of eq. (7) reveals a large discrepancy in the "on-time." Further,
the oscillographs of Figure 27 show that the power MOSFET
appears to turn on a third time. These discrepancies are due to
the reverse recovery of the diode (see Figure 27B). Consequently,
the total turn-on time expressed by equation (7) must be modified
to include the reverse recovery time, trr:
(1)
"After this time, the parasitic diode is forward biased, and the
voltage across the inductor is:
(2)
where VF is the forward diode voltage. If the damping losses are
neglected, then:
ton =
(3)
~
RL
and
[1 -
~]
ln2 + trr
(8)
2V DD
"Since, after the diode's reverse recovery, the inductor once
again has a downward current direction, the situation is the same
as when the MOSFET first turned off but with a lower initial
current level, and thus the cycle repeats again."
(4)
Bibliography
A
OV
1.
J. Redonfey, "The Dv/DT Phenomenon in Power Transistors," The power transistor in its environment, ThompsonCSF, Semiconductor Division, Applications book, 1978.
2.
J. Rockot, "Reverse Transistor Action In Transistor Inver-
OV
ters," Westinghouse Power Semiconductor Division, Tech
Tip, pp. 1-13.
OV
3.
R. Severns, "dV/dt Effects in MOSFET and Bipolar Junction
Transistor Switches," Proceedings of the IEEE Power Electronics Specialists Conference, June 1981, Boulder CO.
4.
D. Kuo, C. Hu & M. Chi, "dV/dt Breakdown in Power
MOSFETs," IEEE Electron Device Letters, Volume EDL-4,
No.1, January 1983.
5.
W. Slusark et aI., "Catastrophic Burnout in Power VDMOS
Field-Effect Transistors," Proceedings of the 21st annual
IEEE, Reliability and PhYSICS Symposium, 1983, pp. 173-177.
6.
D. Giandomenico, et aI., "Analysis and Prevention of
Anomalous Oscillations and Turn-Off Behavior in a Vertical
Power MOSFET," private report done for Siliconix Inc.,
later published in the proceedings of POWERCON 11, April
1984, ~allas.
B
OA
Oscillograph of the Turn-Off Behavior of
the Power MOSFET Inverter
a) Upper Trace: Vo = 10V/Oiv; Lower Trace: Vgs = 5V/Oiv.
b) Upper Trace: Vo = 10V/Oiv; Lower Trace: 10 = O.2A/Oiv.
Figure 27
7.
"The drain voltage remains near zero during the time the diode
is conducting. The current through the diode is dictated by the
L-R circuit and is:
F (
2VDD +-VID=_-Rl_e-tRt/ L)
L
DD
- VRL
(5)
5-64
E. Oxner, "Static & Dynamic dV/dt Characteristics of Power
MOSFETs," PCI Proceedings April, 1984, pp. 132-43.
5.5 Inverse Diodes of Power MOSFETs
Introduction
A parasitic anti-parallel diode is inherently built in
the process of fabricating any power MOSFET transistor. This diode has voltage and current ratings
equal to the FET, but has, in general, a slow reverse
recovery. Depending upon circuit application, the
parasitic diode could be a bonus or a problem, for the
user. The slow recovery speed of the diode may result
in a large recovery current spike and cause high power
dissipation in an opposing FET switch. In this paper,
first the formation of the parasitic diode will be
described, then the problems associated with the
diode in power circuits will be explained. Remedies
for the problems will also be suggested.
G
OXIDE
CHANNEL
Epl
n+
SUBSTRATE
D
DMOS Device Structure
Figure 1
Formation of Parasitic Diode
A DMOS device structure is used to explain the
formation of a parasitic diode in the power MOSFET.
As shown in Figure 1, source metalization overlaps
both the p and the n+ regions. The reasons for this
are, first, to stabilize the voltage of the p region so it
does not drift when the device switches and, second,
to suppress the action of the parasitic bipolar junction
transistor in the structure thereby ensuring proper
operation of the MOSFET. This metallization, however, connects an anti-parallel pn diode from source
to drain. All MOSFET device structures, such as
VMOS and HEXFET, have similar diode formation
and, therefore, power MOSFETs will not block
reverse voltage.
considered to be a bonus for the user. However, the
reverse recovery speed of the diode is normally
between 200 and 400 nanoseconds. This can present
problems in some circuit applications. .
Problems of Using Parasitic Diodes in the Circuit
In a power circuit with a single FET, the parasitic
diode is seldom used and seldom presents any problems. However, in a power circuit with a loop consisting of two serially connected FETs and a voltage
source, the diode may present a problem. Commonly
used half bridge and bridge inverters are examples.
Under certain operating conditions with these converters, the recovery of the diode causes problems as
will be explained below.
Because the parasitic diode uses the same Epi and p
regions as the MOSFET, the voltage rating of the
diode is the same as that of the companion MOSFET.
The diode current rating is also the same as that of the
MOSFET. Therefore, the parasitic diode is often
Figure 2 shows a half bridge in which two FETs are
used. In first quadrant operation, inductive current
free-wheels through D2 and the load when 01 turns
.
5·65
l1li
OFF. When Q1 turns on again, there is a temporary
short circuit formed by the loop of VS , Q1, and D2
during the reverse recovery time of D2. This will
cause a large power spike in Q1 and D2 and may
damage Q1. The reverse recovery time of D2 depends, among other parameters, on the amount of
forward current flowing at the instant Q1 turns ON.
If a discontinuous mode of operation is chosen, there
will not be current flowing in DZ when Q1 turns ON;
therefore, the problem described will not exist. A
similar situation, of course, applies to QZ and D1.
Va
-=-
D,
v.
n n
l:v.J U L
D,
Va-=-
Ib:'
BACK EMF
v,
0 0
n
Ie)
Two-Quadrant DC Chopper Circuit
Figure 2
-v.__________
Similarly, in a bridge inverter circuit, the problem
with the diode mayor may not be present depending
on the nature of the load and inverter operating
conditions. Figure 3 shows the inverter circuit diagram and several possible output voltage waveforms.
Depending upon the gate drive timing sequence, the
load voltage waveform can be classified into three
categories:
O
Bridge Inverter and Three Basic Output Voltage Waveforms
(a) Bipolar Square Wave
(b) Unipolar Square Wave
(c) Quasi-Square Wave
Figure 3
Tn n
1. Bipolar square wave
2. Unipolar square wave
3. Quasi-square wave
IL
_v,J
LJ
I
I
I
I
I
I
I
I
I
I
I
I
An analysis of whether the diode will present problems will be given for each of the above three cases.
Generally speaking, however, if the inverter circuit
requires commutation of current from a diode to the
opposing FET in the same totem pole then a severe
current spike will occur. If the current is commutated
from a diode to its companion FET then the problem
will not occur. This is because the diode in the first
case is subjected to reverse bus voltage, but in the
second case, it is subjected to essentially zero voltage
when the diode current is terminated. The next
paragraph will discuss the load conditions under
which the diode problem will exist for the three basic
load voltage waveforms mentioned.
10104
L·
I
I
I
I
I
RLLOAD
LEADING
RLe LOAD
LAGGING
RLe LOAD
(DIODE PROBLEM EXISTS)
Figures 4, 5, and 6 show the load current waveforms
and associated switch timing for different loads.
Figure 4 shows the waveforms for the bipolar square
wave output voltage. As can be seen, for R-L loads,
Bridge Inverter Load Current Waveforms for the Case of
Bipolar Square Wave Load Voltage
Figure 4
5·66
v.
I
I
I
I
I
I
-v.--r----r---Il
I
I
I
I
I
I
I
: Q,Q4
:
I
I
I
I
I
I
I
1041>3
: 0,04
I
I
~
I
I
I
I
:
I
Q,04
I
RL LOAD
I
RL LOAD
(DIODE PROBLEM EXISTS)
Il
LEADING
RLe LOAD
LEADING
RLC LOAD
Il
LAGGING
LAGGING
RLe LOAD
RLe LOAD
(DIODE PROBLEM EXISTS)
(DIODE PROBLEM EXISTS)
Bridge Inverter Load Current Waveforms for the Case of
Quasi-Square Wave Load Voltage
Figure 5
Bridge Inverter Load Current Waveforms for the Case of
Unipolar Square Wave Load Voltage
Figure 6
diode current is always commutated by the companion
FETs and, therefore, a diode recovery problem would
not occur. For an R-L-C load, the current waveform
could either lead or lag the voltage waveform. If the
natural frequency of the RLC is higher than the
inverter switching frequency then the current waveform will lead the voltage waveform. Figure 4, shows
that the diode problem will not exist in this case
because diode current is commutated by the companion FET. If, however, the current lags the voltage,
which means the natural frequency of the RLC load
is less than the inverter switching frequency, then the
diode current will be commutated by the FET on the
opposing part of the totem pole, and the diode
recovery problem will occur. This has direct ramifications on the diode requirements for any resonant
bridge converter.
Circuit Remedy
Since the diode is inherent in the MOSFET and its
recovery time is much longer than the FET's switching
time, a circuit remedy must be found in order to use
the FET effectively. Figure 7 shows a scheme using a
center-tapped inductor in each totem pole to limit the
current surge during the diode reverse recovery time.
This arrangement requires additional free wheeling
diodes around this inductor to release inductive energy when the FET is cut OFF. Another circuit
remedy is to use a Schottky diode, DS, in series with
-
Figure 5 shows the waveforms for the case of unipolar
square-wave output voltage. For an R-L load, the
load current may be continuous and, when it is, the
diode current will be commutated by the opposing
FET, and the diode recovery problem will occur. If
the load current is discontinuous then the diode
problem will not occur. Again, for the R-L-C load,
only the lagging case presents problems. This is
similar to the situation in the case of the bipolar
voltage waveform. Figure 6 shows the waveforms for
a quasi-square wave output voltage. The situation is
similar to Figure 4. Among the three load conditions
discussed, only the RLC "lagging" condition presents
problems.
Vs
-=-
Inverter Circuit Using Center Tapped Inductors to Reduce
Temporary Current Spike
Figure 7
5-67
each FET and use a fast recovery diode as the freewheeling diode. Figure 8 shows a circuit diagram of
this arrangement. The purpose of the Schottky diode
is to prevent current from flowing through the parasitic diode. Reverse current now has to flow through
diode DF. Because of the fast reverse recovery nature
of DF, the current spike during recovery is considerably reduced.
Conclusion
Due to the slow recovery of FET inherent diodes, a
temporary short circuit could occur in several commonly used switching power circuits. The effects of
the temporary short on the circuit are high power
dissipation particularly at higher switching frequencies and possible damage to the FET. This should be
avoided.
It is noted that only certain operating conditions and
load types. lead to diode reverse recovery problems
and require a circuit remedy. Two solutions were
recommended for this problem. One requires the
addition of two inductors in the circuit. The other
requires the addition of Schottky and high speed
diodes.
Va ..:...
Reference
Lee, F. C., D. Y. Chen, M. Smith, and G. Carpenter,
"Characterization of High Power Darlington Transistors," 1982 Powercon.
Inverter Circuit Using SchoHkys to Prevent the Temporary
Circuit Short
Figure 8
5-68
5.6 MOSFETs Move In On Low Voltage
Rectification (T A84-2)
The efficiency penalty imposed by the offset voltage
inherent in Schottky and pn junction diodes has been
a perennial thorn-in-the-paw for designers of low
voltage, high current power supplies. Recent power
MOSFET design advances allow designers to economically replace Schottky diodes and provide higher
conversion efficiency in many applications. Using
power MOSFETs will remove that thorn.
a broad range, and then to optimize devices for
certain applications such as low voltage rectification.
MOSPOWER transistors may be used as synchronous
rectifiers for any voltage, but they are receiving the
most attention in the 10-50 volt breakdown range.
Many designs for power MOSFETs are presently in
use. In most cases these designs were originally
optimized for the 100V to 400V range and, as a result,
do not represent the practical limits of ON resistance
for low voltages.
To exploit the opportunity provided by the MOSFET,
several questions must be answered:
Figure 1 shows the performance of contemporary
transistors normalized by multiplying total chip area
(including all inactive chip area) by its on-resistance.
1. How is a MOSFET optimized for synchronous rectifier service?
2. What are the performance limits?
This figure allows comparison of different size devices
on the basis of the normalized resistance. The figure
includes both theoretical and practical limits. From _
these limits, it is clear that below 100V, significant
improvement remains to be made.
3. How is an optimized device used in a circuit
to take advantage of its characteristics?
4. What is the basis for choosing between a
MOSFET or a Schottky diode in a given
application?
Figure 2a shows the cross section of a contemporary
MOS transistor. Device on-resistance can be minimized in several ways. First, the thinnest and lowest
resistivity epitaxial layer consistent with the breakdown voltage is employed. Second, gate width must
be optimized. If it is too wide, surface utilization is
poor, while if it is too narrow, the JFET formed by
body regions results in an unwanted increase in total
resistance. The surface geometry also requires optimization to squeeze maximum performance from the
devices. Figure 2b compares layout efficiency of
various surface geometries. While the geometries
vary only a few percent, one geometry (square-onhex) is more efficient than the rest. Keep in mind that
The following discussion addresses each of these
questions.
When we have answered those questions, it will be
quite clear that MOSFET synchronous rectifiers are a
practical alternative to Schottky rectifiers, particularly for output voltages below 5Y.
MOSFET Structures
The present trend of power MOS manufacturers is to
first provide transistor selections that perform across
5-69
SquarE> on CirclE' on
1.
SourcE'
geometry
and grid
Unll cell
Coef'ICIE'nt
of cellular
squarE'
grad
square
grll:S
0 0 0
0 0 0
O:~:O 0:0:0
0 00 0 '0' 0
000
o 0
000
[?i
~~~
;0:
10
08862
I
0.0
0.0
0(0'10
°0'0
LC.JO 0(0'10
0'0 ci--o
,.--
ref
gpomE'tru,'s
Hl'ugon Square 01'1 Circle on He-lagon on
on square heugonal helagonal heUigonal
grId
grid
gnd
grid
[9]
09306
...
" ... 0__ ...
10746
,.-- .
>
"0
t,
r--"
>
'-_oJ
<0...--- ,
09523
10
The Efficiency of Various MOSFET Surface Geometries
Figure 2b
Low voltage, high current devices using some of these
optimizing techniques are beginning to appear. Two
such devices presently available are listed in Table l.
These are practical synchronous rectifiers, but they
are by no means fully optimized. A factor of two or
more improvement should be possible.
R,,, • (BVDSS)2 5
/
,/
[
IRF
.~O{l
Gl VMOS
GL VMOS
IR IILXMOS
5
10 •
8
q
\0
11
l~
SlllCQNIX VN{,4GA
SIUCONIX VNI.:!OOA
SILICONIX SVH
IR tll:XMOS
Gl DMOS
SIUCO:\IX V;-.,lJI...
10'
10'
Br~Jl..lIo"n
Table 1
The ratings of the lowest on· resistance commercially available
MOSPOWER transistors.
AlG
SIPMOS
DEVICE:
PACKAGE:
ON-RESISTANCE:
BREAKDOWN:
CURRENT:
10'
Voh,.gc IBVI>St;1 IVolh)
Normalized ROS(on) Comparison
Figure 1
Source and
Body
SILICONIX
VNCOO3A
MOTOROLA
POWER MODULE
TO·3
.035
60V
60A
Epoxy Module
.018
60V
100A
Electrical Characteristics oCPower MOS
Transistors as Synchronous Rectifiers
Gate
The performance of a MOSPOWER transistor as a
synchronous rectifier differs significantly from its use
in either linear or switching applications. As shown in
Figure 3, the MOS transistor will conduct current in
an opposite direction to normal. The transistor should
be operated so the intrinsic diode is not turned on, or
the performance of the circuit may be altered, as
shown in Figure 4 where current continues to flow for
some time after the voltage is reversed. When the
diode conducts, minority carriers exist. The presence
of these minority carriers may have a second effect
on circuit performance. The limit of a power MOS
transistor is reduced when carriers are present in the
body·drain junction region. When a MOSPOWER
transistor is rapidly switched off, displacement current flows to discharge the body-drain junction capacitance. Figure 5a shows this behavior schematically.
Figure 5b shows the dV/dt behavior of Siliconix
devices without minority carriers present at the moment of transition. The dVldt turn-on threshold is
greatly lowered when minority carriers are present.
Reduction can be one to two orders of magnitude
below that shown in Figure 5b.
NN+
Dram
Cross Section of a MOSPOWER Transistor
Figure 2a
the optimum design depends on breakdown voltage
and drain current (in high current devices). Another
means for reducing resistance is to increase chip area.
In practice this means paralleling several die.
5-70
..
Reverse Conduction
300
Forward ConductIOn
iii
~
150
.;;
~
100 t - - - - + - - - 7 " T...~
50
v
O~_~_L-
a
100
__L-__L-~L-~L-~L-~
~OO
300
400
VOLTAGE (V)
The dV Idt Limits of Siliconix MOSPOWER Transistors When the
Diode is Not Conducting
Figure 5b
Region of
OperatIOn for
Synchronous
Rectification
Several other considerations must also be made when
using power MOS transistors for synchronous rectification. Gate-to-source voltage must be limited to
manufacturer's maximum data sheet specifications.
This requirement sounds straight forward, but the
dynamic behavior of circuits, particularly switching
circuits such as those found in switch mode power
supplies, often results in unexpected transients. Careful design and analysis will prevent any unexpected
voltage excursions.
The Current versus Voltage Behavior of a MOSPOWER
Transistor in Both the Forward and Reverse Directions
Figure 3
Current
The resistor-like characteristics of power MOS transistors are not maintained as the current increases for
two distinct reasons. First, increase in current forces
more carriers through the channel region, causing a
transverse voltage drop. The channel region narrows,
and the current that flows through the device saturates. This increase in ON-resistance as a function of
gate voltage and channel current is shown in Figure 6.
An increase in current also produces an increase in
channel resistance due to device heating. Figure 7
shows the positive temperature coefficient of power
MOS transistor resistance.
Time
The Reverse Recovery Characteristics of the Integral Diode of a
MOSPOWER Transistor
Figure 4
Designing in MOSFET Synchronous Rectifiers
Gate Drive Timing
Gate drive timing is critical to proper circuit operation. What constitutes proper timing varies with the
type of converter circuit used. The popular quasisquare wave converter circuit (Figure 8) is a good
example. Voltage and current waveforms are given in
Figure 8b, and the proper timing for 11 and 12
conduction is a 50% duty cycle drive which is in phase
with VS. While this is the most obvious timing
scheme, it is not the only workable one. Figure 9 shows
the waveform for 11 and 12 which exists in this circuit
if a normal diode is utilized instead of a MOSFET.
G
S,B
A Schematic Showing the Effect of Displacement Current (Id) on
MOSPOWER Transistor Performance
Figure 5a
5-71
l1li
PW~l
ON Resistance Characteristics
1.&
fi)
'oLoA
:IE
::c
1.6
u
z
1.4
e.
w
~
U)
&1\
~
6A
enw
\
1.2
a:
3~~~
z
0
r'--
""
1.0 f--2A ~ ........
I
"C
~
c
PWM
0.&
Quasi-Square Wave Converter
Figure 8a
<
a:
o
o
2
4
6
Vas -
VOS(th)
&
10
12
GATE ENHANCEMENT VOLTAGE (VOLTS)
II
The Device Resistance as a Function of VGS and ISO
for the IRF 730
Figure 6
12
TEMPERATURE EFFECTS ON rDS(on)
2.4
""Z
/"
u
..:
.....
2.0
1/
en
Vi
""e>::Z
1.6
0
1.2
/
:i
..:
::E 0.8
",:
0
Z
I
14
V
0
u.J
N
13
V
".
1/
./
0.4
VI~------t-------+-------;--------r----
~
o
-50 -25 0
25 50 75 100 125 150 175
Tj - JUNCTION TEMPERATURE (0C)
V2~------t-------+-------;--------L----
The Normalized Resistance of a MOSPOWER Transistor as a
Function of Temperature for the IRF 730
Figure 7
Quasi-Square Wave Converter Waveforms
Figure 8b
5·72
time the secondary will be short circuited, and a large
current spike will appear in the primary switches.
This event is clearly undesirable. Fortunately, MaSFETs can switch rapidly and are easy to drive. These
problems can be avoided with a little care when
designing the drive circuit.
In other circuits, timing waveforms and consequences
of mistiming may be different. Two other converter
circuits and their idealized waveforms are shown in
Figures 11 and 1Z. In the buck-derived converter,
there is no advantage to extending the conduction
interval beyond 50% as there was in the quasi-square
wave converter. On the other hand, ifthe conduction
interval is too long, it does not cause current spikes in
the primary switch due to the current limiting action
of the input inductor. This performance is definitely
an improvement over the quasi-square wave circuit.
This circuit also accommodates too short a conduction
interval by allowing the integral diode to conduct.
Diode Rectifier Current Waveform
in a Quasi-Square Wave Converter
Figure 9
Notice that during the intervals when both S3 and S4
are off, IL divides more or less equally between the
two rectifiers. In a diode this change in waveform isn't
significant from a loss point of view. If we change the
MOSFET gate drive timing as shown in Figure 10, the
same current waveforms (without trr spikes) in the
synchronous rectifiers will reduce power losses. The
improvement occurs because conduction losses in a
MOSFET are proportional to the rms current, and
the extended conduction interval provides a lower
rms waveform for the same average current.
(a) Schematic
50')
1
V,
rcdll~cd
T
Jo~~
Imllllg
V~
(b) Waveforms
V,
O~
____
~~
____
~
-+______-L_
______
-
norm.1I
Imllng
Gate Drive Timing for Reduced Losses
Figure 10
Consider what happens if the gate timing does not
correspond to the ideal waveforms. There are at least
two possibilities worth analyzing-too short and too
long conduction intervals. If the conduction interval
of S1 is terminated early, 11 must continue to flow (the
switch is in series with an inductor!), and it will,
through the integral diode within the MOSFET. This
is not catastrophic, but it will increase the losses
because of the higher forward drop of the diode. It
will also introduce reverse recovery current spikes
and may aggravate the dV/dt problem. If the conduction interval is too long so that S1, SZ, and either S3
or S4 are on simultaneously, then for some period of
Buck-Derived Current Fed Converter Example
Figure 11
The boost-driven circuit is even more tolerant of long
conduction intervals. The conduction interval may be
increased up to 50% with no effect on circuit operation. For drive circuit simplicity, 50% drive is usually
used. For duty cycles beyond 50%, the primary
5·73
(a) Schematic
D> 5
SI
-_--G +
M ......--_-'1oJ L ....
S2
Auxiliary Winding Gate Drive
Figure 13
D> 5
twice each switching cycle), VI =0, and there is no
gate drive! 11 continues to flow, but now it goes
through the integral diode-increasing the power
losses. There is another drawback to using auxiliary
winding drives in the quasi-square wave family of
converters. Transformer winding voltages are directly
proportional to input voltages and will vary with the
input voltage. For example, if the gate drive voltage is
12V when the input voltage is low, then when the
input voltage is doubled (not an uncommon requirement), gate voltage will also double to 24V. Except
for Siliconix MOSPOWER devices, most MOSFETs
are limited to ± 20V on the gate. Exceeding this level
will destroy the device (Siliconix rates their gates at
±40V with 100% testing to SOY). Adjusting the
winding so the gate drive is below 20V at high line
may produce a condition where there is not sufficient
gate enhancement at low line, thereby increasing the
conduction losses.
(b) Waveforms
II
I,
---,
I
I
---,
v,
I
I
Boost-Derived Current Fed Converter Example
Figure 12
inductor will again limit switch current transients. In
fact, in this circuit, synchronous rectifiers may be
used to control output if their conduction duty cycles
are greater than SO%. In this mode of operation, the
output voltage is controlled by varying the conduction
duty cycles of SI and S2 so that both switches are on
simultaneously for part of the switching sequence.
This allows the regulation function to be done on the
secondary without having to couple a control signal
back to the primary. This is a significant simplification.
The winding voltage dependence on input voltage
also increases the peak voltage seen by the rectifiers.
For a SV output, the rectifier commonly sees a peak
voltage of20 to 2SV, not including noise or transients.
This requirement for a relatively high rectifier voltage
rating (in proportion to output Voltage) means that
the RDS of the MOSFET must be higher for a given
device size.
All of these problems may be overcome by using
another circuit topology. The circuits in Figures 11
and 12 both have correct timing and relatively constant gate voltage when used with an auxiliary winding. They also subject the rectifier to a reverse voltage
of twice the output voltage (lOV in SV supply)
regardless of the input voltage.
Clearly, gate drive timing and degree of tolerance to
mistiming depend on which converter is used. The
designer must take this into account when designing
the drive circuits.
Gate Drive Circuits
The simplest way to provide gate drive is to use an
auxiliary winding on the transformer secondary (Figure 13). The advantage ofthis method is its simplicity,
but there are some disadvantages. For example, if this
scheme is used in the quasi-square wave converter,
when S3 and S4 are both off (a condition which occurs
Even the best designed converter has some noise,
ringing, and/or transients appearing across the transformer windings. When auxiliary winding drive is
used, these transients are coupled directly into the
gate where they may cause improper or mistimed
5·74
An alternative to using an auxiliary winding is to use
a separate drive circuit derived from the primary
switch drive circuits as shown in Figure 15. The
example shown is for a quasi-square wave circuit. This
drive scheme allows for proper timing of the gate
drive.
The advantages of using independent gate drive are
several:
1. The amplitude of VGS may be optimized at a
predetermined value.
2. The negative part of the drive may be eliminated, reducing drive power and gate stress.
Gate Protection Methods
Figure 14
3. Higher switching speeds are usually possible.
turn-on and turn-off or even gate destruction. Frequently it is necessary to provide some gate protection
(Figure 14). It is vital that gate protection be right at
the devices with a minimum of series inductance and
be capable of nsec response time. Not all transient
suppressors and zeners will respond quickly enough.
Notice also in Figure 14 that the source connection
for the gate drive is brought out separately. This
technique reduces pick-up from the high current in
the source lead. Further protection can be provided
by bringing the drive winding to the device as a
twisted pair.
4. Gate protection is much easier.
5. Precise drive timing is easily achieved.
6. The rectifiers may perform control functions
in some circuits (Figure 12).
7. In some applications, bi-directional power
flow is desired. Use of a synchronous rectifier
with independent drive allows bi-directional
power flow [3,4,51.
The most obvious disadvantage is increased circuit
complexity. This additional complexity may not be
great, so it is worthwhile to consider the advantages.
Figure 14 also shows damping resistors (RD) across
the gates. These resistors are added because drive
circuit and winding leakage inductance can resonate
with gate capacitance to generate a ringing voltage on
the gate. Some form of damping should be used since
this ringing may damage the gate or cause spurious
turn-on or turn-off. The R-C network on the gate
performs this function. Either series or shunt damping resistances may be used. The series resistance,
however, reduces switching speed. If, by careful
layout and transformer design, the ringing frequency
is 1 MHz or above, a ferrite bead may be used for
damping. This technique also helps to prevent high
frequency parasitic oscillation.
Avoiding dV/dt Problems
Whatever drive scheme used, the integral diode
should not be allowed to conduct. This will reduce
conduction loss and eliminate diode reverse recovery
current spikes. Even more important is the preservation of static dV/dt characteristics. Most MOSFETs
remain quiescent with an applied dV/dt of 50V-lOOVl
nsec as long as there is no reverse diode current. It is
extremely unlikely for a low voltage power supply to _
even approach this limit. However, if diode current is
present, the dV/dt capability may fall to a value as low
as O.5V/nsec. Such rates are possible even in low
voltage supplies. If a device is triggered on when the
voltage is reversed, the transformer secondary is
effectively short circuited and some damage could
result.
Thermal Design Considerations
As shown in the previous section, RDS is a positive
function of junction temperature (Tj). When ID is
controlled by a current source (the inductor), positive
thermal feedback is present. As the device becomes
hotter, RDS increases which in turn increases power
dissipation, and the junction temperature increases
even further. For this reason, it is imperative to
observe good thermal design practices. The achieva-
Independent Gate Drive Circuit Example
Figure 15
5·75
ble efficiency is very much dependent on the thermal
design. A discussion of this subject is available in a
Siliconix Applications Note [6].
lLe
Equipment using cryogenic cooling has become relatively common. For the designer of power supplies to
be used with such equipment, low temperature cooling provides a unique opportunity to drastically lower
conduction losses. A MOSFET will work just fine at
liquid nitrogen temperatures ( -196°C) and even at
liquid helium temperatures (-265°C). For Tj =
-196°C, RDSis20-25% ofitsvalueatTj = + 125°C.
As an added bonus, heat transfer is much better
because of the higher thermal conductivity of silicon
at these temperatures. The switching properties are
not significantly affected. If cryogenic cooling is
available, it should be used.
= the transconductance of the
MOSPOWER transistor
= carrier mobility
The equations predict an increase in device performance proportional to the increase in carrier mobility.
Figures 16 and 17 show the improvement in the characteristics of a VNEOO3A MOSPOWER transistor as
the operating temperature is reduced from 2SO C to
-196"C. These figures show that for this transistor,
the gm increases almost a factor of 2, while the onresistance decreases the same amount.
The operation of a MOSPOWER transistor does not
depend on the injection of carriers across a junction,
nor is the minority carrier lifetime a factor. As long
as sufficient majority carriers are present in the source
region, drain current will flow when a bias is present
on the drain and voltage is applied to the gate. Carrier
freeze-out in the source and drain regions does not
impact device performance until the conduction electron (or hole) concentration is reduced to a small
percentage of its original value.
The Characteristics of a VNE003A MOSPOWER Transistor at
25"C (gm = 6.5S, ReSlon) = 56 mO)
Figure 16
The main factor affecting MOSPOWER transistor
operation is the mobility of the carriers as they flow
through the body and drain regions. This mobility
increases, so the device on-resistance should decrease.
The drain current of a vertical DMOS transistor is
given by
IDS
=
Co Z VSAT (VGS - VGS
(TH»
(1)
while its on-resistance is
1
1
The Characteristics of a VNE003A MOSPOWER Transistor at
-196"C (gm = 12S, ReS(on) = 31 mO)
Figure 17
(2)
RDS(on)= {3 (VGS - VGS(TH»
where
MOSPOWER transistors are also beginning to be
used in their third quadrant of operation. In this
quadrant, the carriers flow from the drain to the
source, opposite to normal current flow. The intrinsic
body-to-drain diode is in parallel with the MOSPOWER transistor operated in this direction, as shown
by Figure 18. A MOSPOWER transistor operated in
this fashion has high current capability with a low
forward voltage drop in one direction, and the normal
breakdown voltage characteristic in the other direction. This device performance makes MOSPOWER
transistors attractive for use as synchronous rectifiers.
The characteristics of the VNE003A MOSPOWER
transistor in the third quadrant are shown in Figures
19 and 20. These photographs are of particular interest, because of the offset voltage present when the
Z
(3 = TILe Co
In these equations,
L
= the channel length
IDS
= the drain to source current
Co
= the capacitance per unit area of
the gate oxide
Z
the amount of active source
perimeter
VSAT
the saturation velocity of electrons
in silicon
VGS
the gate-to-source bias voltage
VGS(TH) = the threshold voltage of the
MOSPOWER transistor
5-76
MOSPOWER transistor is operated at - 196"C. This
offset voltage is present in Figure 17, but is more
easily measured here.
To demonstrate the possibilities of low voltage operation an actual power converter was built and tested.
Source
Gate
Source
HALF· BRIDGE
A 5 V, 100 W Half Bridge SMPS Using MOSPOWER Transistors
as the Switching Elements and a Synchronous Rectifiers
Figure 21
Assuming that the majority of the power dissipation
occurs in the rectifiers, the power saved by using
MOSPOWER transistors in place of junction devices
is given by
N+
Dram Current (10)
The Carrier Paths in a MOSPOWER Transistor Used as a
Synchronous Rectifier
Figure 18
PSA VED
=
2
I LOAD [ R bulk- RDS(on) ]
+
(4)
LOAD Voffset
For a 6045 Schottky diode and the VNE003A the
percentage of the output power saved as a function
of IL has been calculated at 25"C and - 196"C. The
results of these calculations are shown in Figure 22.
This figure shows that at room temperature, a MOSPOWER transistor is a slightly more efficient rectifier
at low currents, but is less efficient as the load current
increases. However, at liquid nitrogen temperatures,
the MOSPOWER transistor is more efficient until the
load current approaches 25 amperes. As MOSPOWER
transistors optimized for low voltage operation are
introduced, the percentage of power saved will increase significantly. For SMPSs operating below 5
volts or greater than 100 volts, MOSPOWER transistors are particularly attractive as the output rectifiers. Below 5 volts the offset of a Schottky diode
at very low temperatures results in a significant power
loss. For outputs above 100 volts, fast switching pn
diodes have a large offset voltage and a high bulk
resistance.
The Characteristics of a VNE003A MOSPOWER Transistor in
the Third Quadrant at 25'C (Voffset = 180 mY,
R bulk
29 mil)
Figure 19
=
+20
+1.
PERCENTAGE OF OUTPUT
POWER
SAVED
+12
The Characteristics of a VNE003A MOSPOWER Transistor
in the Third Quadrant at -196'C (77"K)
(Voffset = 460 mY, R bulk = 18.8 mil)
Figure 20
+. r -
+4
The 5 V, 100 watt power supply shown schematically
in Figure 21 was constructed for operation at liquid
nitrogen temperatures. For ease of testing, only the
MOSPOWER transistors were cooled to this temperature. All other components were kept at room temperature above the nitrogen container. (The PWM 25
was used as the control I.C., but it was found to
stop functioning properly at liquid nitrogen temperatures.) The MOSPOWER transistor operated as expected at LN2 temperatures.
~·CI
II
~
-4
-.
(25' C)
--
-... --....
-r---
.............. 1-...,
...............
_-12
-1.
-20
.............
PERCENTAGE OF OUTPUT
POWER
LOST
•
I.
15
2.
25
The Percentage of the Output Power Lost or Saved by Using
a VNE003A MOSPOWER Transistor to Replace a 6045
Schottky Diode at -196'C and 25°C.
Figure 22
5-77
-
It is possible to operate the entire power converter
in LN2. The effect of low temperature operation on
the other components is discussed in Appendix 1.
MOSFET-Schottky Comparison
In low voltage output power supplies, overall efficiency (Tj) is dominated by losses in the rectifiers. The
relationship [7] between efficiency and the rectifier
forward drop (Vr ) can be expressed by
ID(rms)Reduction
In a MOSFET, conduction loss (Pc) is due to RDS. It
can be expressed simply as
Pc = ID2 (rms) RDS
Tj
=
1-~
(7)
------~-----
1 + (Vr)/(Vo)
(5)
Where ~ is the percentage of power dissipated with
the supply excluding the rectifier losses, and V0 is the
nominal output voltage.
Notice that it is the rms value of ID which counts.
Since RDS is also a function ofID.
(6)
A graph of equation 7 is given in Figure 24, for several
values of ~. The efficiency range for 2V and 5V
outputs and for Vr = O.4V to 0.6V is also shqwn.
From Figure 24, it is clear that with a low output
voltage, the rectifier forward drop is a major factor in
overall efficiency regardless of other circuit losses.
Where a > 2. Obviously high efficiency requires that
ID (rms) be as small as possible. There are practical
ways to do this. One method is to make the averaging
inductor larger which changes the shape of 11 and 12.
Figure 23 gives examples of typical waveforms for
large and small inductors. In both examples, average
current is the same, but rms current is lower when a
larger inductor is used. If we take the ratio of the two
values and raise this ratio to the 2.5 power, the power
loss difference is almost 40%! Clearly the value of the
averaging inductor is critical.
100~-----+
______-r______+-______
8A
90~~~~~
____
~~
____+-______
~
~
____
____
~
~
"~I
I
I
I
I
I
I
I
I
I
60~--~r;-+-----+------~----~~~--~
I
(b)
I
SA
4A
-: -
-
-
-
-
Vo =
la\lg
sv
~
Vo= '2V
3A
S0r------;-------+------~------~----~
Large L -
I rms = 4 04A
40~
The Effect of Current Waveform on RMS Value
Figure 23
00
The rms current also depends on the topology chosen.
For example, the circuits shown in Figures 8 and 11
have a lower rms rectifier current than the circuit in
Figure 12, for the same output current. This occurs
because of the pulsating nature of the output current
in any boost derived converter.
____
~
______- L_ _ _ _ _ _
01
~
_ _ _ _ _ _L __ _ _ _
04
~
05
Efficiency as a Function of Forward Drop (Vr)
Figure 24
In a Schottky diode, there are other losses in addition
to forward drop. The most prominent of these losses
is reverse leakage current which can be substantial
when the junction temperature is in the range of
100°C to 150°C (the normal device operating range).
Another problem arises from the large capacitance of
As previously indicated, in some cases gate drive
timing may also be used to reduce I D(rms).
5-78
these diodes. Losses due to charging and discharging
this capacitance are relatively small, but in high
frequency (> 100kHz) inverters where the primary
switches may have transition times of 20nsec or less,
large current spikes may be generated. These spikes
may increase dissipation and peak power stress in the
primary switches.
Psaved
(Watts)
20
15
10
In a MOSFET, reverse leakage current is significant
and capacitance may be as much as an order of
magnitude lower. This reduction is a significant improvement particularly for high frequency conversion. Given the very fast transition times possible with
a MOSFET, they should be usable as synchronous
rectifiers at frequencies up to I MHz in switch mode
converters, and at even higher frequencies in resonant
converters.
0.6
o"
-s
-10
+ IL Vos + PI
(8)
--
,.....-::::::
--......
~ ~
ROS(on)=IO
~
-
RDS(ony2 Rbulk
.........
30
~40
IL
50 (amps)
RDs(on) =3 Rbul~
RbUI~ ios(on)=k~
5 Rbulk
~
Putting a Price on Efficiency
To this point, we have been emphasizing efficiency.
For many applications, the decision on the device
selection is based on the following economic
considerations.
The power lost in the MOSFET (PM) is expressed as
(9)
1. Direct device cost.
where the switching loss is assumed negligible. To
determine which device is superior in a particular
application, take the difference between equations 8
and 9. The result in terms of power saved when a
MOSFET is used may be expressed as
2. Power converter selling price.
3. Life cycle cost of the system converter.
On a device-to-device basis, MOSFETs have historically not looked very promising due to high prices and
high RDS. However, prices have drastically decreased, and further reductions are expected. Manufacturers now know how to build very low RDS
devices, so the cost gap between MOSFETs and _
Schottkies will be much narrower in the future.
Psaved = I2L(rms) (RB - RDS)
+ IL Vos + PI
~
~
.--...-
The temperature effect on RDS, RD, and Vos and
PI is also not included. RDS and RB increase with
temperature as does PI, and Vos increases as Tj is
reduced. Consequently as Tj is reduced, Psaved
increasingly favors the MOSFET. Consider this behavior when comparing devices.
where: IL = average output current
I2L(rms) = RMS diode current
RB = diode bulk resistance
Vos = diode offset voltage
PI = loses due to leakage
PM = I2L(rms)RDS(on)
V
= RbU~ ~
Graph of Power Comparison
Figure 25
The power lost (Ps) in a Schottky diode is expressed
as
Ps = I2L(rms)RB
I
ROS(on)
(10)
A graph of equation 10 can be used to determine
which device is more efficient. If Psaved is positive,
then the MOSFEThas the edge; if however, Psavedis
negative, then the Schottky is better. A graph of
equation 10, using typical values for RB and Vos and
a range of values for RDS, is shown in Figure 25. This
graph makes clear where each device is superior and
shows the critical role played by RDS.
A more useful comparison is the manufacturing cost
of a power supply using each type of device. In this
context, the improved efficiency possible with MOSFETs becomes important. With higher rectification
efficiency, less cooling is required. This efficiency
translates into either smaller heat sinks or (in higher
power units) the elimination of a cooling fan. The
lower circuit currents due to the higher efficiency also
allows cost and size reduction in switches, filter
capacitors, and magnetics. The net result is an overall
cost reduction.
There are additional factors not taken into account by
the graph. As rectifier efficiency is changed, the RMS
currents in the transformer windings, the primary
switches and the filter capacitors will change. Most
of these losses are 12(rms) dependent, and as rectifier
efficiency is improved, the value for 13 will also
improve significantly. Relating this to Figure 25, the
zero crossing points will not change, but the degree
of improvement in the areas where each device is
superior will be enhanced.
For telephone companies and large computer manufacturers, a third level of comparison (life cycle cost)
becomes important. For these types of users, im-
5-79
proved efficiency translates into lower utility bills for
both the direct load power and the power used to cool
the equipment. Reduction in air conditioning loads
reduces overall operating costs as well as the capital
invested in the cooling equipment. In such systems, a
higher initial cost for power conversion equipment is
acceptable when improved efficiency reduces operating cost proportionately.
Appendix 1. Operation of Other Power
Converter Components at Low Temperature
The Electrical Behavior of Semiconductor Devices
The behavior of semiconductor devices such as diodes
and transistors and of monolithic integrated circuits
made up of these and other components varies greatly
with temperature. A basic understanding of the effects
of temperature may be gained by examining the variations of three key semiconductor parameters with
temperature. These three parameters are:
(1) Carrier mobility: The ease with which conduction
electrons and holes move through the semiconductor crystal varies greatly with temperature.
The mobility depends on the total number of
impurities and the effect of crystal lattice vibrations on carrier movement. As the temperature
of silicon is lowered from room temperature, carrier mobility increases as shown in Figure 26.
The carrier mobility peaks between lQa'K and
20a'K, depending on doping concentration, but
is typically a factor of 2 or 3 above room temperature mobility.
(2) The number of dopant atoms ionized: As the
temperature is reduced, the percentage of dopant
atoms that remain ionized (and hence contribute
to the conduction process) begins to decrease
when KT approaches the ionization energy of the
dopant atoms. For common dopant atoms in silicon, the temperature at which dopant "freezeout" begins is approximately 125"K, as seen in
Figure 27.
(3) Minority carrier lifetime: The relative position of
the Fermi level and centers that control the recombination rate of carriers varies significantly
with temperature. As the temperature is reduced,
the probability that a trapping center will capture
both a hole and a conduction electron increases,
and minority lifetime therefore decreases.
Clearly there are several ways to compare costs
between Schottkies and MOSFETs as rectifiers. The
best comparison method depends on the particular
application, but a direct device-to-device comparison
is rarely useful.
Conclusion
Changes in pricing and lower RDS make the MOSFET competitive with Schottky diodes in low voltage
rectification applications. To gain the most advantage
from using MOSFETs as synchronous rectifiers, the
user should carefully consider circuit implementation
and overall economic issues. A simple substitution of
one device for anotht::r is definitely not the key to
success.
References
[1] M.S. Adler and R.R. Westbrook, "Power Semiconductor Switching Devices - A Comparison
Based on Inductive Switching," IEEE Transactions on Electron Devices, ED-29 , Number 6,
June 1982, pp. 947-952.
[2] M. Chi and C. Hu, "Some Issues on Power
MOSFETs," PESC Record, June 1982 (392).
[3] G. Cardwell and W. Neel, "Bilateral Power Conditioner," IEEE Electronics Specialist Conference Record, June 1973, pp. 214-221.
200r--l7"<~r--r--r--r--r---'
[4] H. Matsuo and K. Harada, "New DC-DC Converters With an Energy Storage Reactor," IEEE
Transactions on Magnetics, Volume Mag-13,
Number 5, September 1017, pp. 1211-1213.
150
[5] R.D. Middlebrook, S. Cuk and W. Behen, "A
New Battery Charger/Discharger Converter,"
IEEE Power Electronics Specialists Conference
Record, June 1978.
[6] R. Severns, "Safe Operating Area and Thermal
Design for MOSPOWER Transistors," Siliconix,
inc. Applications Note AN83-1O.
[7] R. Kagan, M. Chi and C. Hu, "Improving Power
Supply Efficiency with MOSFET Synchronous
Rectifiers," Proceedings of POWERCON 9, 1982,
p.D4.
700
The Effect of Temperature on the Mobility of Holes In Silicon [6]
Figure 26
5-80
TlK)
POWER transistor at 25"C, while Figure 29 shows
the voltage drop of the same diode at -196"C. The
photographs show that the bulk resistance (R bulk)
of this diode is approximately the same at both temperatures, but that the offset voltage (Voffset) increases by a factor of two from .66 volts to 1.33
volts. This behavior results from the freeze-out of
donor and acceptor atoms. As the temperature is further decreased, this offset will increase until the diode
stops functioning altogether.
~nr~~~200~__~100~__~7~5________~~
INTRINSIC RANGE
SLOPE· Eg
SATURATION RANGE
Schottky Barrier Diodes
\ nl
Reducing the operating temperature of a Schottky
barrier diode from 25"C to - 196"C increases its forward voltage drop as shown in Figure 30 and 31. It
can be seen from the photographs that the offset
voltage increases by about a factor of two while the
bulk resistance decreases. The decrease in bulk resistance is the result of an increase in carrier mobility
at the lower temperature. However, the forward voltage drop at any current level increases as the temperature is reduced.
I
lOI30:-----LJ--4~--L--L---L--L---L--L---L---1
8
12
16
20
10001 T (K-I)
Conduction Electron Concentration in a Silicon Substrate
Doped with 1015 Donor Atoms per cm 3 [7]
Figure 27
pn Junction Diodes
The reduction in temperature increases the forward
voltage drop of a pn junction diode as seen in Figures
28 and 29. Figure 28 shows the forward voltage drop
of the body-to-drain diode of a VNE003A MOS-
The Current vs. Voltage Characteristics of a 6045 Schottky
Barrier Diode at 25°C (Voffset = .34 V, R bulk = 10 mO)
Figure 30
The Current vs Voltage Characteristics of the Body-ta-Drain
Diode of a VNE003A MOSPOWER Transistor at 25°C
(Voffset = .66 V, R bulk = 43.5 mO)
Figure 28
l1li
The Current vs. Voltage Characteristics of a 6045 Schottky
Barrier Diode at -196°C (77"K) (Voffset
.61 V,
R bulk 7 mO)
Figure 31
=
=
Bipolar Transistors
The Current vs. Voltage Characteristics of the Body-ta-Drain
Diode of a VNE003A MOSPOWER Transistor at -196°C (77"K)
(Voffset = 1.33 V, R bulk = 45.5 mO)
Figure 29
Bipolar transistors operate through the injection of
minority carriers from the emitter region through the
base region to the collector region. The current gain
5-81
of a bipolar transistor depends upon both the injection
efficiency of the emitter-base junction and the lifetime
of the injected carriers as the flow across the base
region. Both of these phenomena grow less efficient
as the temperature is lowered, so the performance of
a bipolar transistor should decrease as it cools.
pacitors fabricated in the silicon substrate. The behavior of diodes and transistors over temperature has
been discussed, but not that of integrated resistors
and capacitors. Diffused resistors decrease in absolute
value with temperature, though the temperature coefficient varies with dopant concentration. The change
of diode, resistor and transistor characteristics does
not track with temperature; integrated circuits specified
for the commercial or military temperature range operate poorly as the temperature decreases, finally failing to function as designed. The temperature at which
an I.C. ceases to function and whether the failure is
recoverable upon reheating or not, is circuit
dependent.
This predicted performance is verified by Figure 32
and 33. The two figures show the electrical performance of a 2N6579 bipolar transistor at 25"C and at
- 196'C. The significant decrease in hFE and the
accompanying increase in VCE(SAT) are evident.
Design of integrated circuits to operate at very low
temperatures is possible; however, the performance
of bipolar transistors at reduced temperatures makes
them less attractive than MaS transistors for I.C.
design. Table 2 shows that n-channel and p-channel
MaS transistors operate to liquid helium temperatures
(4.2"K). The relative stability of their threshold voltage and predictable increase in gm allow optimization
of the device geometries for low temperature
operation.
The Current vs. Voltage Characteristics of a 2N6597 Bipolar
Transistor at 25°C (hFE = 7)
Figure 32
Table 2
The Characteristics of Various MaS Transistors Over
Temperature [5]
The Current vs. Voltage Characteristics of a 2N6597 Bipolar
Transistor at -196°C (7rK) (hFE = .9)
Figure 33
Other Electrical Characteristics
Two parameters, device leakage and breakdown voltage, vary in a similar fashion for all of the devices
discussed above. Device leakage decreases as the
temperature is reduced, since this current is thermally
generated. This current is an exponential function of
the temperature. The breakdown voltage also decreases as the temperature decreases though not nearly
as rapidly. The breakdown voltage decreases only a
few percent per lOO"C.
Vth
(Volts)
TYPE
TK
SD10ll
Lateral
DMOS n-channel
2N6661
V-groove
DMOS n-channel
293
77
4.2
293
77
4.2
3408
7776
3.18
4.33
3362
7442
2783
1.15
2.54
1.82
IRDll01
Vertical
DMOS n-channel
293
77
4.2
3678
7938
8712
2.65
3.91
4.17
IRD9120
Vertical
DMOS p-channel
293
77
4.2
3698
6498
5408
-3.16
-3.83
-4.1
M0405
Metal Gate
NMOS
293
77
4.2
42
128
149
1.1
1.7
1.4
M0405
Metal Gate
PMOS
293
77
4.2
35
169
149
-0.7
-1.1
-1.8
CD4007
CMOS
Metal Gate
293
77
4.2
752
576
171
2.2
2.7
2.2
CD4007
CMOS
Metal Gate
p-channel
293
77
4.2
907
1290
984
-1.3
-1.4
-2.0
VN30A
V-groove
n-channel
293
77
4.2
1048
8712
9248
0.4
2.4
1.3
~
--
--
n-channel
Integrated Circuits
Both bipolar and MaS monolithic integrated circuits
are made up of transistors, resistors, diodes, and ca-
5-82
Thermal Characteristics and Safe Operating
Area of Components
bond wire size increases because the conductivity of
aluminum-the common lead wire material-increases with decreasing temperature as shown by Figure 43.
The ability of electronic components to dissipate heat
at very low temperatures is considerably enhanced.
The power dissipation of a component is determined
by the thermal characteristics of the device multiplied
by the temperature difference between the power dissipating region of the component, and its ambient
temperature. Operation of circuits or systems at liquid
nitrogen or liquid helium temperatures multiplies the
power dissipation capability of a typical device by a
factor of 3 to 6. An effect that considerably enhances
the power dissipation of silicon devices is its significant increase in thermal conductivity at low temperature. As seen in Figure 34, the thermal conductivity
of silicon increases a factor of 10 between room
temperature and liquid nitrogen temperature. Other
materials including copper show an increase in thermal
conductivity over this temperature range as well.
Chlp/Bondwlre Limitation on Current
Package Dlsslpallon
Limitation
iE
:!.
Safe Operating Area
9
E
~
U
c:
e
c
Increase in MOSFET SOA
Performance Over Bipolar
Transistors - due to No
Thermally I••duced
Secondary Breakdown
Drain-source Voltage, VDS (Volts) -
The Safe Operating Area of a Typical Bipolar and
MOSPOWER Transistor
Figure 35
10 3
Component Reliability
~
I
E
u
f::
I«
10'
!
..........
:..:
V
>
t
>
i=
The mechanisms that lead to the failure of semiconductor devices are characterized as having a rate of
change determined by the Arrhenius equation:
I
1'\
I
Ii
10
R= A e -EAiKT
I
where
u
::l
a
z
a
R
A
EA
K
T
"I,si
I
/
u
j
I
100
10
300
=
=
=
=
=
specific rate of change
constant determined by the component
the activation energy (eV)
Boltzmann's constant
temperature in degrees Kelvin
Activation energies for failure mechanisms in semiconductors have been found to range between .3 and
1.1 e V. Reduction of the operating temperature to
liquid nitrogen or liquid helium levels significantly
increases device lifetime when it is limited by thermally activated mechanisms.
I
.11
(11)
1000
T (K)
The Thermal Conductivity of Silicon Over Temperature [8]
Figure 34
Nature is unlikely to be too kind to the designer of
a power conversion system for operation at very low
temperatures. Reducing the operating temperature is
likely to increase mechanical stress caused by differential expansion rates with temperature, causing a
new class of failures.
The safe operating area of a typical bipolar or MOSPOWER transistor must be modified when the transistor is operated at very low temperatures. A
comparison of the SOA of a bipolar and a MOSPOWER transistor is shown in Figure 35. The secondary breakdown characteristics of a bipolar
transistor do not allow its operation at full current
and voltage regardless of the power handling capability of the package. A MOSPOWER transistor with
its absence of secondary breakdown may operate at
full current and voltage with adequate power dissipation. The limit set by the breakdown voltage is
still present, but as discussed earlier, is reduced with
decreasing temperature. The current limit set by the
The Behavior of Passive Components at Very Low
Temperatures
The performance of each of the passive component
types used in the design of an SMPS changes in its
own fashion as the temperature is lowered. In this
section, the behavior of passive devices is discussed
and related to the underlying physical mechanism.
5·83
l1li
Magnetic Components
A typical power converter uses magnetic components
such as inductors and transformers. To successfully
design a low temperature converter the designer must
know the characteristics of the core, winding, and
structural materials.
V/
/ /
~
V. . . .
Theory predicts that in a ferromagnetic material, as
the temperature is reduced, Bmax will increase and
in -practice this is exactly what-happens. As shown
in Figure 36, as the temperature is reduced Bmax
increases, rapidly at first and then leveling off as T
approaches O"K. For example, reducing the temperature of a 50% Ni-Fe core from + 100"C to - 200"C
increases Bmax by 18%.
Core Losses of TDK H5B2 at 25'C
Figure 37
/"
I
I
V
..:
IQ
T.... penoIunC·q
Core Losses of TDK H5B2 at -196'C (77'K)
Figure 38
•
l/"
I /
Typical Variation of Flux Density (Bm) with Temperature [3]
Figure 36
Theory also predicts that the core losses will increase
as T is reduced. This is due to increased conductivity
of the core material and increased hysteresis losses.
Again, theory and practice agree. However, the
amount of increase in loss varies greatly from one
material to another, as shown in Figures 37 through
42. In a ferrite material such as TDK H5B2 (Figures
37 and 38) the core loss increases dramatically. On
the other hand, in the 80% Ni-Fe material (Figures 41
and 42) the increase in core loss is barely discernible.
i/
Core Losses of .5 mil Supermalloy at 25'C
Figure 39
If
Copper or aluminum is usually used for the winding
material. As shown in Figure 43, both materials have
a large positive temperature coefficient, resulting in
greatly reduced resistance at low temperatures. The
resistance of copper, for example, drops by a factor
of 13 between + l00"C and - 200"C. This increase
in conductivity can be used to reduce the winding
resistance and subsequent losses. One point should
be kept in mind, however, for AC currents the skin
depth varies as the square root of the resistivity. In
a high frequency converter, as the resistivity of the
windings is lowered, less of the material is conducting. Thus, the effective winding resistance does not
drop as rapidly as the resistivity of the winding
material.
Core Losses of.5 mil Supermalloy at -196'C (77'K)
Figure 40
5-84
it appears that by lowering the operating temperature
from + 10erC to - 20erC, the insulation voltage stress
may be increased by a factor of 2 to 5 for the same
life in many film insulators.
I
12.000
10.000
j
1\
~
~ _f-"
8000
"-......
~"0
Core Losses of .5 mil 80% Ni-Fe at 25°C
Figure 41
/
6000
"'r
4000
..
2000
.liu
~
6
j'-...,
"~
\
-200
I
-1 MIL
>
u
..:
a
-100
200
100
300
Temperature lOCI
V
J
Temperature Effect on the AC Dielectric Strength of Type H
Kapton Film [2]
Figure 44
Core Losses of .5 mil 80% Ni-Fe at -196°C (77°K)
Figure 42
'0'
/
'0'
/
w
3
10'
V
~
0
~
~
~
~
'0'
- - .-.. ~.. - t - - - t - - - - 1
'0'
10
1 MIL Kapl0n H
,3'c
3500 VOLTS
BREAK OWN
/
3 _~'.L.~7_---,1~~'_ _'...
3IL'_...:.J.'Yc:...
14_---",11°C
16
1.8
2.0
,.,
,.
,.•
RECIPROCAL OF ABSOLUTE TEMPERATURE
1
Resistivity of Copper, Aluminum, and Iron as a Function of
Temperature [4)
Figure 43
x 10 3
TEMPERATURE
The Effect of Temperature on the Life of 1 mil Kapton H [2)
Figure 45
For high voltage applications, the properties of the
insulating material become critical. The net effect of
low temperatures is to improve the electrical properties
of the insulation, at the expense of the mechanical
properties. The effect of temperature on the dielectric
strength of .001" type H Kapton film is shown in
Figure 44. We see that by reducing the temperature
from + 10000C to ~ 20<1'C the dielectric strength almost
doubles, a most desirable result. In a practical application, the usable voltage rating of an insulator is
as much a function of its resistance to corona-induced
deterioration as it is of dielectric strength. An indication of how strong a function of temperature the
life of an insulating film is, can be seen from Figure
45. While the available data is certainly incomplete,
The principal problem in magnetic components caused
by lowering the operating temperature is increased
brittleness in the materials. In particular the plastic
bobbins and some insulating films can become very
brittle at low temperatures. A large temperature range
can also bring about problems due to the different
coefficients of expansion of the various materials in
the magnetic structure. These mechanical problems
can, however, be overcome by careful design, handling, and mounting.
Given appropriate materials, low temperature magnetic operation is advantageous, since it allows lower
losses and small size to be achieved.
5-85
-
An important benefit of low temperature operation in
film capacitors is the improvement in voltage capability of the insulating film, as outlined in the magnetics discussion. From an energy storage point of
view, low temperature operation is a big plus, since
U = CV2 and the voltage capability is improved. A
factor of 5 increase in V, increases U by 25 times.
This is very attractive in those applications, such as
pulsed loads, when large amounts of energy must be
stored.
Capacitors
Many types of capacitors are used in power converters; some of these are usable at low temperatures and
others are not. The temperature characteristics of several types of capacitors are shown in Table 3. The
film and mica capacitors are very good, showing
only small drops in capacitance. More surprising is
the relatively good behavior of the solid tantalum
types, some of which hold up very well even at
4.2"K! The wet tantalums are not usable at low temperatures due to large capacitance changes and very
high ESR values. Very little information is available
concerning the ESR of film capacitors at low temperatures, but we do know that the dissipation factor
goes up somewhat, and that the resistivity of the foil
or metalization goes down. For high current filtering
applications the ESR is dominated by the metalization
resistance, so that lower temperatures should result
in a lower ESR. The lower ESR, coupled with the
increased thermal conductivity of the capacitor body,
should mean that the current rating of a film capacitor
is much greater at lower temperatures. Unfortunately
this is only an educated guess at the time.
TYPE
CeramiC
MANF
Corning
Value
300"K
I~FI
77"K
77°K
4.2"K
-559
-549
-635
-£2.6
1.036
1.001
0457
0451
0387
0374
Elmenco
1 D02
0997
0997
0993
0990
0990
- 05
- 04
- 12
- 07
Film
S& El
9923
9.955
9452
9.552
0309
9 444
- 48
- 41
- 62
- 51
S& El
1195
1246
1126
1176
1115
1161
- 58
- 58
- 67
- 68
S& EI
1017
1099
1012
1035
1 D09
- 05
- 58
- 08
- 65
Film
(polysulfone)
Film
(polysulfone)
While resistors are not used in the power conversion
process, they are needed for the control functions.
Many types of resistors exist and are readily available.
The only type not suitable for low temperature operation is the ubiquitous carbon composition resistor.
Besides having a large temperature coefficient, the
resistance values display considerable hysteresis during temperature cycling. Table 4 summarizes the
measured properties of several types of resistors. For
most purposes, the metal film resistors are more than
adequate, particularly at LN2 temperatures. For precision voltage dividers or similar uses, the bulk metal
resistors are remarkably good, and should be adequate
even at LHe temperatures.
% Change fro 3000K
4.2"K
Mica
(polysulfonel
Resistors
1027
Film
(polycarbonate)
S& EI
1221
'166
1170
1155
1233
Tantalum
Kemet
1045
0998
0973
0909
0936
0854
1158
- 45
- 51
- 54
- 61
- 69
- 89
-104
-144
References
[1] J.K. Watson, "Applications of Magnetism,"
Wiley - Interscience, 1980, p. 163.
[2] Boeing Aerospace Co., "High Voltage Design
Guide for Airborne Equipment."
[3] Magnetics Inc., "Design manual featuring tape
wound loops."
[4] Private communications with Dick Gilbert, LSMC.
[5] Private communications with Mike Dix, Staff
Scientist at NASA Ames Research Center, Sunnyvale, CA.
[6] G.L. Pearson and J. Bardeen, "Electrical Properties of Pure Silicon and Silicon Alloys Containing Boron and Phosphorus," Physics Review,
75, 865 (1949).
[7] S.M. Sze, Physics of Semiconductor Devices, 2nd
edition John Wiley & Sons, 1981, p" 26.
[8] C.Y. Ho, R.W. Powell, and P.E. Liley, Thermal
Conductivity of the Elements-A Comprehensive
Review, American Chemical Society and American Institute of Physics, New York, 1975.
[9] R.S. Kagan, M. Chi, and C. Hu, "Improving
Power Supply Efficiency with MOSFET Synchronous Rectifiers," Proceedings of Powercon
9, p. D-4 (1982).
Table 3
The Change in Capacitor Values Over Temperature [5]
Value (Ohms)
TYPE
MANF
Thick
film
Bulk
Cornmg
Leel
Vishay
metal
Metal
film
Metal
film
NC55
Fit
spares
0
42°K
77°K
42"K
09995
10231
1004
0999
10459
'0403
1295
1713
1927
+23
+46
+40
+296
+712
+927
9996
9974
99768
9972
99B71
-02
-02
- 02
- 01
1.0007
10015
10003
10008
10248
10361
10283
+01
+11
+10
+ 24
+ 35
+ 28
09973
26077
09989
10031
26173
10161
+06
+04
+05
+ 19
+ 16
+ 23
99987
Corning
% Change fro 300 K
77"K
300" K
10131
1.0098
10039
26489
10217
Table 4
The Change in Resistor Values Over Temperature
5-86
5.6.1 Using Power MOSFETs as High-Efficiency
Synchronous and Bridge Rectifiers in
Switch-Mode Power Supplies
(TA83-1)
Introduction
Switch-mode power supplies (SMPS) are well-known for
their many features: high efficiency, light weight (and
accompanying small size), and outstanding regulation.
These features are principally as a result of their high
switching frequency.
Since the Schottky-barrier diode fails at high voltages, we
are forced to consider alternate rectifiers. Heretofore, our
only choice was to use fast recovery rectifiers in either the
bridge or full-wave configuration, as shown in Figure I.
With the recent introduction of improved pUlse-width
modulator (PWM) integrated circuits capable of clock
frequencies to 500 kHz and beyond, and power MOSFETs
having saturation voltages superior to comparable power
bipolar transistors, we must now turn our attention to
the rectifiers.
Historically, when high-speed rectifiers were required, if
the voltage was low, say less than 30 V, the Schottkybarrier diode reclifier was the optimum choice. However,
because of their inherently low peak reverse voltage
rating, if our voltage was much higher we had no alternate
choice but to use a fast-recovery p-n diode whose PIV
ratings could be chosen to be more than adequate for most
applications.
Bridge & Full-Wave Rectifiers
Figure 1
There are, unfortunately, two fundamental problems
when using, or trying to use fast-recovery diodes in either
of these configurations (Figure I).
The first and major problem is that for high clock frequency SMPS applications these fast-recovery diodes are
simply not fast enough! To allay any argument to the
contrary, let us acknowledge that there are high voltage
and high current fast diodes with reverse recovery times in
the low nanosecond region. But, let us not forget that the
new, high-efficiency switch-mode power supplies now
appearing in the market as well as those under development are PWM switchers not high-frequency sinusoidal
switchers. In other words, the diodes must rectify square
waves, or at least something very close - perhaps trapezoidal. If we were to overlay the optimized SMPS squarewave output over a "typical" fast-recovery diode's conduction characteristic (Figure 2), our problem becomes
apparent: we discover that we have conduction when we
should have none!
We have now reached a barrier where the Schottkybarrier diode and the fast-recovery p-n diode limit performance. This barrier prevents high voltage and high
clock frequency SMPS design for two reasons: Schottkybarrier diodes are limited to a PIV generally below SOV;
and, fast recovery diodes have excessive reverse recovery
time.
Low ON-resistance, high voltage power MOSFETs break
through this barrier and offer the potential for higher efficiency,
high-voltage switch-mode power supplies.
The Problem Dermed
Before we address the problem, we must first acknowledge that
for low voltage, high-frequency SMPS applications the
Schottky-barrier diode offers satisfactory performance as a rectifier. It is when we attempt to achieve efficient service at high
voltages and high clock frequencies that the problems appear.
An issue might be raised that PWM switch-mode power
supplies do not deliver square waves as we find in Figure 2,
but, in fact, taking into account the minority-carrier
5-87
-
storage time of the typical switching bipolar transistor(s),
the SMPS generally exhibIt a "dead" time. That is, as we
see in Figure 3, we have a period where both the bipolar
transistor's storage time as well as the reverse recovery
time of the rectifying diode may recover without fear of
the crowbar effect.
Since we are discussing 'high-efficiency' SMPS design, we
can anhcipate an output square-wave without dead time,
or at least a very short dead time. Here we will discover
that the wave is directly affected by a diode's reverse
recovery time. In a bridge rectifier, for example, when one
diode in the totem pole begins conduction, its counterpart (in the same totem pole) enters into reverse recovery
(Figure 2) and for a brief period we have a crowbar. This
effect is clearly seen in Figure 5.
Our answer to this argument is that we are addressing
high-efficiency switch-mode power supply design. Dead
time seriously affects effIciency as we see in Figure 4.
A Square Wave Superimposed Over the
Diode Conduction Waveform Showing the Crowbar Effect
of Reverse Recovery Time (t rr)
Figure 2
Fast Recovery Diode Bridge Rectifier at 400 kHz Showing
Effect of Reverse Recovery Time on Rectification Waveform
Figure 5
,
-
~
Our second problem involves the voltage drop across the
rectifier. A bridge results in two (2) dIOde drops, as shown
in Figure 6, whereas a full-wave rectifier has but one diode drop.
Again, our efficiency is affected to our detriment.
r--
The Typical Output Waveform of a Switch-Mode Power Supply
Showing Dead Time
Figure 3
CLOCK 10
EACH OUTPUT 10
I
1 MHz
250 KHz
2 MHz 1 500 KHz 300 KHz
200 KHz
1500 KHz 1
1125 KHz 1
250 KHz 150 KHz
100 KHz
I~OMHZ
~
'"
b
i
Output (Positive) Waveform Superimposed Over Sinusoidal
Input Showing a 1.2 V Two-Diode Drop (Bridge Rectifier)
Figure 6
w
~
:::l
...
..."o
...
"-
:::l
:::l
Enter the Power MOSFET
..
Resolving Problem #1
W
III
II)
Because of the market penetration of power MOSFETs, many of
us are familiar with their outstanding features which we need not
review. All we need to remember is that power MOSFETs
exhibit no minority-carrier storage time.
:::l
OUTPUT PULSE WIDTH ~,.) Cr' MAX DUTY CYCLE
Dead Time Affects Usable Pulse Width
Figure 4
5-88
the channel current heretofore excited by the gate bias
potential. A clearer understanding of this phenomenon is
shown in Figure 9.
This statement does, however, require qualification. When we
categorically state, "no minority-carrier storage time," we are,
in fact, referring to the basic MOSFET, assuming no parasitic
elements. Intrinsic to every power MOSFET is a base-drain p-n
diode which is, indeed, a parasitic element often considered in
many applications as a "beneficial" diode (Figure 7). As we
continue our study of synchronous rectification, we perhaps need
to be reminded that any p-n diode will exhibit minority-carrier
storage time which, in tum, will result in a finite reverserecovery time.
N+
Drain Current (lD)
Simplified Third Quadrant "Output" Characteristics
Figure 9
The Cerrier Paths In a MOSPOWER Transistor
Used as a Synchronous Rectifier
Figure 7
Before moving on we need to pause to reflect on why we
need to understand third-quadrant operation. We are, in
effect, planning to replace the rectifier diode with a power
MOSFET. We need to review the electrical schematic of
a full-wave rectifier, shown in Figure 10.
Consequently, if our commission is to use power MOSFETs as rectifiers we must somehow avoid activating this
parasitic body-drain diode.
Now that we have "set the stage" so to speak, let's review
some perhaps not-so-familiarpower MOSFET characteristics.
Anoda
Cathode
...~
m
+
-I
!
~
1_----------, o·
It
.
;"
i
I_----------------~~i
ID
:0
(,)
'!
Q
Anode
,..
c
Cathode
g.=
~----------------------~g
Full-Wave Diode Rectifier
Figure 10
~
I
Here we have a pair of rectifier diodes in common-cathode
connection. Note the polarity of the diode pair: anodes
tied to the a-c source, cathodes tied in common to output
a positive potential. If we envision the electrical symbol of
the power MOSFET and superimpose this symbol in lieu
of the rectifier diodes of Figure 10, we discover that our
power MOSFET is installed "backwards," that is, we have
the source tied to the a-c source and the drain outputting the
d-c, as shown in Figure II.
Drain-source Voltage
First Quadrant Output Characteristics
Figure 8
MOSFET Characteristics
The classic first-quadrant saturation characteristics of a
power MOSFET, shown in Figure 8, are generally wellknown. However, as long as we have a gate bias enhancing
current flow, the first quadrant characteristics are also
duplicated in the third quadrant! In the familiar first
quadrant (Figure 8) the ultimate operational limit is breakdown voltage. Third quadrant operation differs in that
here we discover our operational limit is when the "parasitic" body-drain diode enters conduction and 'captures'
If our intent is to have the MOSFET act in lieu of the
diode, we now must synchronize the MOSFET's tum-ON
with the positive,going a-c cycle and tum-OFF when the
a-c swings negative. Furthermore, we must never allow
the parasitic p-n body-drain diode to participate. If we do,
its reverse-recovery time will crowbar the other (MOSFET)
rectifier. Remember, we're rectifying the output of a
5-89
-
I
current before we reach Vf -- forward voltage turn-ON of the
parasitic body-drain diode. Figures 12, 13 and 14 provide typical
characteristics taken from three popular power MOSFETs.
r-________~SO~u=r~c~e_t,:~._~D~ra~i-n------~----o+
Source
Drain
MOSFETs as a FUll-Wave Rectifier
Figure 11
high-efficiency SMPS square-wave with little or no dead
time. We'll focus on how to synchronize the MOSFET later.
Resolving Problem #2
Siliconix inc. Type VN64GA
"Third Quadrant" Output Characteristics
Superimposed Over Forward Diode Breakdown
Figure 12
Interestingly enough, the solution of our first problem is
also the solution to our second problem! Once we remove
the diode we no longer have a dIode drop. We do, however, have an IR drop, and it is here where we must now
focus our attention.
The power MOSFET shown in Figure 12 offers a 60V
BVOSS and a nominal rOS(on) of 0.4 Ohms at a drain
current of lOA. The topology is V -groove double-diffused
MOS.
Selecting the Power MOSFET
If, indeed the IR voltage drop is a problem needing resolution, it most certainly is if we expect to use power MOSFETs to resolve Problem #1. There is only one way for
us to prevent turn-ON of the parasitic body-drain diode:
we must maintain an IR drop across the power MOSFET
below Vf - the forward voltage of the diode.
Not only do we need a low IR drop across our power
MOSFET (which sImply means we need an ultra-low
rOS(on», we also need to select a MOSFET whose breakdown voltage ratings suffice to withstand the peak inverse
voltage (PIV) that we will expect from our SMPS output.
We are now approaching a paradoxical problem: as we seek
ultra-low roS(on) power MOSFETs while simultaneously seeking a reasonably high breakdown voltage, we may discover
that our costs get in our way! The relationship between roS(on)
and breakdown voltage is:
25
rOS(on) -- k BV .
International Rectifier Type IRF130
"Third Quadrant" Output Characteristics
Superimposed Over Forward Diode Breakdown
Figure 13
Consequently, the only way to achieve low rOS(on) is
to increase the area of the semiconductor chip which, in
turn, forces the price upward exponentially!
The IRF 130, whose characteristics are displayed in Figure 13,
is rated by the manufacturer at 100V with a typical roS(on)
of 0.14 Ohm at a drain current of 6A. The structure is a vertical
double-diffused MOS (VOMOS).
The characteristics shown in Figure 14 are for a VNE003A
VOMOS structure rated at 100 V with a typical roS(on) of
0.035 Ohm at a drain current of 60 A.
We are able to interpret these data by remembering that
we wish to pass as much current as possible WIthout exceeding Vf - the turn-ON voltage of the intrinsic parasitic
body-drain diode. For all three illustrations, the turn·ON
Since, as we learned earlier, the third quadrant output characteristics duplicate those of the first quadrant, we do have a
reasonably simple means to select suitable power MOSFETs
from the plethora of power MOSFETs available. All we need
is a curve tracer, such as the Tektronix 576. By superimposing
the first quadrant output characteristics over the reverse diode's
forward characteristics, we, in effect, can duplicate the critical
parameters from which we then can make a suitable selection
based on both breakdown voltage and maximum permissible
5-90
p-type material, we discover that for an equivalent COS(on)
p-channel MOSFET our costs are considerably higher than
for its n-channel equivalent (with the same voltage rating).
The Circuit
The basic synchronous bridge rectifier circuit using both
n- and p-channel MOSFETs follows that of the diode
bridge much like we did for the full-wave rectifier shown
in Figure II. The circuit is shown in Figure 15.
[]
Siliconix inc. VNE003A
"Third Quadrant" Output Characteristics
Superimposed Over Forward Diode Breakdown
Figure 14
voltage, V f, lies close to - 0.6V. Consequently, we can
make some reasonable approximations of the maximum
current that each of these MOSFETs IS capable of handling (at 25°C). For the VN64GA (Figure 12) 2A; for the IRF130 (Figure 13) 4.2A; and for the VNE003A (Figure 14) > 20
VNE003A (Figure 14) > 20.
In a practical sense we can dismiss the VN64GA as Schottkybarner diode rechflers are easily capable of handl!ng considerably more than 2A and withstand 60V. Were it not
for the high standoff voltage of the IRF-130 (Figure 13),
we might dismiss it as well. The VNE003A (Figure 14) offers
us a little more, but, again, if it were not for the 100V rating,
it, too, might be dismissed.
Synchronous Bridge Rectifier Using n- & p-Channel
MOSFETs
Figure 15
The operallon of thiS MOSFET bndge rectifier is quite
straight-forward. At the half-cycle when the upper rail
swmgs pOSitive the gate bias on FETs Band 0 is also
positive and the gate bias on FETs A and C is negative.
Since FETs A and Bare n-channel, only FET B turns
ON. However, since FETs C and 0 are p-channel, only
C turns ON. With PETs Band C, ON current flows with the right
terminal positive. Likewise, when the lower rail swings positive
the gate bias on PETs A and C is positive; the bias on FETs Band
o is negative and PETs A and 0 tum ON. Current continues to
flow in the same direction as before, and we have gone through
one cycle.
We should by now recogmze the fundamental limitation
of today's power MOSFETs: too high rOS(on)' But, on
the other hand, we must not lose sight of the I!mltation of
the Schottky-barrier rectifier diodes, as well as the limitations of
the so-called fast recovery p-n diode.
If we begin our examination of this bridge rectifier by
Synchronous Bridge Rectifiers
Using Power MOSFETs
first applying a smusoidal input and observing the output
from both the positive and negallve terminals with respect
to one rail we obtain a waveform as shown in Figure 16.
Setting aside our quest for low COS (on) n-channel power
MOSFETs, let's turn our attention to the basic bridge
rectifier (shown m Figure 1) and recall that the diode
bridge recllfier suffers from a 2-diode drop (see Figure
6). If this bridge consisted of Schottky-barrier diodes
across we might attain twice the normal voltage rating
seemg that we use a totem pole of diodes across the rails.
On the other hand, using conventional fast-recovery
diodes, voltage is secondary but we're again troubled
when rectlfying high efficiency SMPS square waves with
little or no dead time.
If we could implement a quad array of MOSFETs using
low COS(on) n- and p-channel devices, we could escape the
limitations of breakdown voltage, as well as offer high
efficiency, because of the lack of minority-carrier storage time
which is inherent in all MOSPETs.
Again, of course, we're strapped with the COS(on) problem
inherent in today's MOSFETs and especially with pchannel devices. Because of the reduced mobility of
Output Waveforms of Synchronous Bridge Rectifier
Figure 16
5-91
-
We originally faulted the diode bridge rectifier for having
a 2-diode drop, which regardless of current generally
amounted to at least 0.6V per diode and perhaps more,
depending upon the rectifier used. For MOSFETs, our
voltage drop is dependent upon rOS(on} and current flow;
in other words, the classic IR drop. Consequently, it is
definitely to our advantage to use MOSFETs with the
lowest ON resistance.
In both Figures 17 and 18, the output current was 1.2 A
into a pure resistance load. The total TOS(on) of the nand p-channel MOSFET pair was 0.6 Ohm for a total
voltage drop (at 1.2 A) of 0.72 V which accounts for the
slight 'thickening' of the trace in Figure 17. Because of
the present limitation of p-channel rOS(on), as the load
current increases we can expect a corresponding increase
in the voltage drop. The output waveform, showing both
positive and negative swing for a square-wave of 250 kHz
is shown in Figure 19. The slope of the positive wave was
the fault of the pulse generator used for driving the MOSFET
bridge. (Compare Figure 19 with Figure 5.)
We can obtain a better appreciation for the diode drop
effect by a simple experiment using the synchronous
MOSFET bridge. We first observe the output (positive
for this experiment) with the input waveform superimposed. Because of the limitations of laboratory equipment we were forced to go to a 60 Hz line to obtain
worthwhile current (the functIOn generator simply was
incapable of outputting current). The effect is noticeable if we compare Figure 17 with Figure 18. In Figure
17, we have the bridge operating as a quad MOSFET; but
in Figure 18 we have removed the gate drive to the
MOSFETs causing the bridge to rely entirely upon the
intrinsic body-drain diodes for rectification. The forward
voltage drop is readily apparent in Figure 18.
Output Waveform Showing Both Polarities with 250 kHz
Square-Wave Drive
Figure 19
Synchronous Full-Wave Rectifier
Using Power MOSFETs
The synchronous full-wave rectifier, shown in Figure 11,
depends upon low rOS(on) n-channel power MOSFETs
as the rectifier elements. For this illustration we will
implement the VNEOO3A whose characteristics are shown in
Figure 14.
Output Waveform (Positive) MOSFET Bridge Rectifier
Figure 17
To synchronize the MOSFETs we must tum-ON the gates
with a gate-to-source voltage of at least +1OV to ensure
optimum conductance. In a SMPS we are able to add
"tickler" windings to the secondary windings of the
output transformer to accomplish this, as shown in
Figure 20.
The Parts List for the SMPS shown in Figure 20 is as
follows:
R1
R2
R3
T1
5.6k
75k
56k
see text
R4
R5
R6
10k
7.5k
3.6k
CI
C2
C3
C4
QI, Q2 VN0600A
Q3, Q4 VNEOO3A
PWM25 - SilicoDix Pulse Width Modulator IIC
Output Waveform (Positive) Body-Drain Diode Bridge Rectifier
Figure 18
5-92
0.01
0.001
5uF
470uF
+ ~--------------------~
R3
15
R1
~Q1
L1
t-------'
+
Switch-Mode Power Supply Using Synchronous MOSFET Rectifiers
Figure 20
Transformer Design, Tl
Undoubtedly the transformer output stage represents the
major design problem for most SMPS designs. Although
this lies outside the general scope of this article, a few words
should suffice to give us a more comfortable feeling. Aside
from the secondary winding we have a pair or'tickler" windings
that provide the necessary gate voltage to 'synchronize' the
gate turn-ON.
The transformer has an Indiana General E-core #8031-1.
Over a nylon bobbin we fIrst wind a 16-turn pnmary
consisting of 8 strands of #30 AWG enamel (Litz wire)
over which is wrapped a copper fOIl acting as a Faraday
shield. The secondary has a total of 28 turns of #24 AWG
enamel with both a center tap as well as addItional taps at
±12 turns which provide tickler windings of 8 turns each.
With a pnmary of I OV /turn and a secondary of 3V /turn,
we have a transformer turns rallo of 1:3.33.
Output Waveforms of Tickler (Upper) and One-Half Secondary
(5 VDIV) with Direct Drive from SMPS (Figure 19)
Figure 21
In Figure 21 we are able to view the output waveform of
one tickler and one-half of the secondary (from centertap). We see that the tickler winding supplies 12.5V of
gate potential - adequate for complete turn-ON of the VDMOS
synchronous rectifier. We are now ready to attach
the VNEOO3A (Figure 14) to this secondary.
-
Another view of interest, shown in Figure 22, is the output from both secondaries of the transformer where we
see the 1800 phase reversal common to all center-tapped
secondaries.
Feeding both VNE003As, we exit through suitable filtering at
a fixed 5 V at lOA. In our experimental SMPS we were power
limited by our laboratory supplies used for primary power.
Nonetheless, from Figure 14, we can anticipate a reasonable
current before the intrinsic body-drain diodes tum-ON.
We have intentionally left the effects of temperature upon
performance until last. As our operating temperature in-
Waveforms Across Secondary
Figure 22
5-93
+1
07
Reverse Breakdown
u;
.......
0
2S'C
G
~
7S'C
a:
0
w
to
O.6V
I
-E
""
""
"
""
"
05
.......
03
0
a:
Z
;0:
0
"w..
Other Radiation Damage
-60
C
0:
III
-80
-90
Both ionizing and neutron irradiation will cause other parameter
shifts such as lower gm. Since in most cases. the shifts are small and
represent second order effects. we have chosen to ignore them in
this discussion. At very high levels ofirradiation(>1 MRad(Si) or
lOIS neutrons! em'), these other effects may not be second order.
Unfortunately at this time, very little test data is available at
extremely high irradiation levels.
Device C
4D0 V
-70
300Y
o
L
1~O
.L
2~O
1
~
4~Q
I
I
5~O
I
6~O
1
7~O
I
8~O
I g:O:OO:-L:-!;:;:-'--:-!
DDSE [krad (sl)]
BVoss Shift with
Radiation
Figure 14
© 1983 IEEE ref. #10
Strategies to Minimize the Effect of
Radiation Damage
Several strategies may be employed to minimize the effect of
MOSFET radiation damage on circuit operation.
The first and most obvious is to employ shielding. In a terrestrial
application, like a nuclear plant, the addition of lead shielding for
the instrumentation is quite practical. In a space application,
however, large amounts of shielding are out of the question
because of the weight yet some degree of shielding is usually
possible.
The second strategy is to employ devices specifically hardened for
radiation. Many changes are possible in MOSFET design and
processing which will improve the radiation resistance. The
feasibility of devices with reasonably stable characteristics up to 10
MRad(Si) has been demonstrated (3). It is interesting to note that
some of the first generation power MOSFETs, (which are no
longer in production), were much more resistant to radiation
damage than today's standard products.
VOS=20 V
~
w -1
." -.
60 V
Z
__==:::::::==---40 V
l:
U
w -3
~
~ -4
OV
g
Z -5
g
-6
"~
0:
III
-7
Device 0
150V
-8
-9~~~~~~~~~~~~~~~~~
100
200
300
400
500 600 700
DOSE (krads)
800
Figure 15
900
1000 1100
© 1983 IEEE ref. #10
Siliconix has begun development of radiation resistant devices
which will be available in the future.
It should be kept in mind that the graph shown in Figure 16 IS
representative of the change in R OS(on)' However. there can still
be large variations in the actual shift seen in a particular device.
The trend is clear: The higher the voltage rating. the greater the
susceptibility to neutron damage.
Until such time as "rad hard" devices are available, the most
practical strategy is to design the circuit so that changes in device
characteristics do not degrade circuit operation.
14
/
12
10
/
,..--------DEVICE A
450 V
~
DEVICEB
/
/
17
~
>-"'-
DEVICEC
150V
Pre-Rad
.
5x 10
~
'\.
10
NEUT/CM2
Plot of ROS(on) VB Neutron Fluence
Figure 16
© 1981
IEEE ref. #2
5-105
As was shown earlier, there are four distinct characteristic changes
which must be accommodated: Vth shift, R OS( on) increase,
BVOSS reduction and photo-current turn-on. One or more of _
these may occur in the same application depending on the
irradiation scenario.
Vth shift is essentially a gate-drive problem. To maintain proper
operation, the Vgs applied to the gate of an n-channel device must
be sufficiently positive (> + 10 V) to tum on the device fully
before irradiation and yet negative enough (2-5 V below the final
Vthl to tum off the device after the maximum irradiation, allowing
for device-to-device variations. In practical terms, this means
a drive waveform such as shown in Figure 17 must be used. One
advantage of the negative bias during the off interval is that it
reduces the shift in Vth at a given total dose. The exact amount of
reduction at a given total dose would depend on the bias values
and the duty cycle. Figure 18 gives an example of the effect of
bias conditions on the shift in Vth. In this example it can be seen
that even though the gate drive duty cycle is 50 % the shift in Vth
lies much closer to the - 10 volts OC bias curve than to the + 10
volts bias curve. To obtain an accurate value for the shift, test
devices should be irradiated with the actual circuit voltages, currents and duty cycle applied. The use of negative bias, while very
effective in maintaining normal circuit operation, does bring with
it some problems. In the circuit shown in Figure 19, for example,
Vgs can reach - 30 V. Most MOSFETs are only rated for ± 20 V
except for Siliconix parts which are rated for ± 40 V on the gate.
The gate to source breakdown voltage can become a limiting
factor in the design.
+10
-10
-20
+10V
-
-30
---------------
Variation {
in Vlh
~~----+_--------+_----t
Actual Gate Waveform
Figure 19B
------------_______ ~::::::::::;::::==;-'
-10V -
Of'.!
OFF
Gate Drive Waveform
Figure 17
Another problem in some circuits is how to couple the asymmetrical (from a volt-second point of view) waveform given in
Figure 17, to the gate in a transformer isolated circuit.
35,-----~------~------------~----_,
in
30
~
~
25
w
20
§!
15
~
.
If a power converter circuit is turned on and kept on during
irradiation, the negative bias voltage can be self-generated. I n the
pre-irradiation off-condition, the switches will be off. The bias is
not needed to hold off the switches nor for starting. lfhowever, the
converter is turned off after a high level of Irradiation, the bias
voltage will disappear, and the switches will be on. If the bus power
is still applied, the converter will be a short circuit across the bus.
There are two ways this may be avoided. First the power bus could
be disconnected before turning off the converter. This is usually
done with a mechanical relay. If the converter is to be turned back
on, the negative bias must be applied first. The second method
would be to supply a rad hard bias supply which is always present.
The bias supply would still have to be interlocked to a bus
disconnect relay in case of failure.
The degradation in BV DSS may be circumvented by using a
device with an initial BV DSS higher than the normal circuit
operation required. Another step would be to minimize the voltage
across the device during irradiation. There are a number of
topological means for doing this in power converters. For
example, instead of using a parallel (or push-pUll) converter
circuit, a half-bridge circuit could be used, reducing the voltage
across the devices by a factor oftwo.
:;:
";!...
Vo = HOY JlJ100 kHz
9
0
l:
f3
a:
-1DV
10
j!:
However, both ofthese schemes bring a penalty: RDS(on) will be
increased unless larger devices are used. If neutron radiation is also
present, the use of a higher voltage device will exacerbate the
increase in R DS( on) due to neutron damage.
YG = -10V
0.1
0.2
0.3
05
OA
Total Dos. [Mrad (sl)]
Photo-current turn-on is an easier problem to solve. To prevent
damage, the current in the device must be limited to a safe value. In
a power converter, the simplest approach is to use a circuit
topology which inherently limits the current even when all switches
and diodes are in conduction. An example of such a converter
circuit is given in Figure 20, and many other circuits of this type
exist. In some circumstances however, using this type of topology
may not be enough. The problem arises from the turn-on of the
parasitic BJT. The circuit shown will limit the current to the
normal switch current, but this may be too high for the BJT, which
does not normally carry any of the switch current. Second
breakdown may then occur. Very short pulses (<100 nsec) may not
be damaging, but longer pulses may destroy the FET. One means
to eliminate this problem would be to deliberately turn on both
FETs during the pulse interval so that much of the current is
diverted through the FET channels.
The Effect of Gate Bias on Vth Shift
Figure 18
© 1982 IEEE ref. #6
NEGATIVE GATE BIAS EXAMPLES
PWM
Drive
]
+
~SI
20V
VI
; II +
10V
-
I
82
[
Negative Bias In a Converter
Figure 19A
5-106
A not her protective scheme would be to place a BJT, as shown in
Figure20, to shunt the current. This device is shorted from base to
emitter and plays no role in normal circuit operation. It is assumed
that the shunt BJT is capable of absorbing the normal switch
current without failing.
shift in a given device at a given Irradiation level IS strongly
dependent on small variations in processing, the specific operating
conditIOns dunng irradiatIOn, and the character of the radiation
source.
Wide vanations in test results for the same type of device done at
different times or In different facilities is unfortunately typical of
radiation testing of all semiconductors.
One word of caution, not all "current fed" converters are self
protecting. For example, the circuit shown in Figure 21 will have
very high switch current due to the photocurrent in the diode.
If high levels of irradiation are anticipated and there IS doubt
concerning the adequacy of the deSign the best course of action is
to install deVices, from the same lot as those to be used in the flight
equipment, in the actual circuit-then irradiate them while the
circuit IS operating.
The total shift in Vth may be greatly reduced in long life, relatively
low irradiation missions by operating the FETs at elevated
temperatures (125 to 1400 C) to accelerate annealing. Obviously
higher operating temperatures compromise reliability therefore
this technique must be applied judiciously.
Conclusion
The increase in R DS( on) due to neutron damage is essentially a
thermal problem. The power dissipated (PT) in the FET is
proportional to:
PTa 12 D(rm,) RDS(on)
The electncal characteristics of power MOSFETs are altered by
radiation. The degree and type of charactenstic change depends on
the type, intensity, and duration of the radiation. In general, the
effect of changing characteristics can be circumvented by proper
Circuit design. Successful power converter designs have been
demonstrated for total doses greater than 200 Krad(Si). When
radiation hardened MOSFETs become available, circuits capable
of surviving total doses approaching ten Mrad(Si) should be
practical.
(I)
The junction temperature (TJ) may be expressed as.
(2)
Where:
= ambient temperature
Ta
Reja = thermal impedance from junction to ambient.
Clearly the MOSFET is suitable for use in high radiatIOn
environments.
As R DS( on) increases, Tj , will increase. The effect is, however, not
linear because R DS( on) is also a positive function of Tj so the
increase in R DS( on) due to radiation is magnified. The thermal
system has positive regeneration and can go into thermal runaway
if R DS( on) becomes too high. This dependence and the determination of thermal stability are treated in detail in a Siliconix
Application Note (4).
We recommend that the designer read the papers hsted In the
bibliography to obtain a more detailed understanding of radiation
effects in power M OS FETs.
There are several things that can be done to minimize this problem:
I) Use the lowest voltage MOSFET consistent with the circuit
requirements. In some cases, the use of series low voltage
devices may be helpful. The direct series of several devices is
not usually attractive from a circuit point of view, but there
exist(5) power converter circuits which automatically divide
the voltage across several switches. I n these topologies, the
use of lower voltage devices is quite practical.
2) Use a device with an initial RDS(on) much lower than the
power level requires. If one device is not adequate, several can
be paralleled.
3) Minimize the ambient temperature.
4) Minimize I D(rms)' In power converter circuits, this can be
accomplished by making the averaging inductor larger or by
using alternate topologies (4). In particular, the family of
converter circuits which includes that shown in Figure 19 are
very useful (5).
5) Minimize Reja' This may be done by using a lower impedance heatslnk, a non-isolated device mounting to the sink,
parallel devices (lower effective Reja)' and in general, good
mounting practices.
Radiation Activated Crowbar Examples
Figure 20
Unsuitable Current Fed Converter Example
Figure 21
Acknowledgements
Several of the figures used in the text were derived from the work
of others:
In a practical design, one or more of these techniques may be used
to arrive at a satisfactory solution.
Figures 2, 3, 4, 10 and II
Figure 12
Figures 13, 14 and 15
Figure 16
Figure 18
Some Thoughts on Radiation Testing
The radiation test data shown in earlier sections is, for the most
part, typical. However, the data from some devices, even of the
same type, can be very different because the degree of parameter
5·107
Reference I
Reference 9
Reference 10
Reference 2
Reference 6
-
References
6) D. Blackburn, et. al., "Ionizing Radiation Effects on Power
MOSFETs During High Speed Switching," IEEE Trans. on
Nuclear Science, Volume NS-29, No.6, December 1982.
7) W. Baker, Jr., "The Effects of Radiation on the Characteristics of Power MOSFETs," Proceeding of Powercon 7, 1980,
p.3.
8) S. Rattner, "Additional Power VMOS Radiation Effects
Studies," IEEE Trans. Nuclear Science, Volume NS-27,
1980.
9) S. Seehra and W. Slusark, "The Effect of Operating Conditions on the Radiation Resistance of VDMOS Power FETs,"
IEEE Trans. Nuclear Science, Volume NS-29, December
1982.
10) Blackburn, Benedetto and Galloway, "The Effect of Ionizing
Radiation on the Breakdown Voltage of Power MOSFETs,"
IEEE Trans. Nuclear Science, Volume NS-70, No.6, December 1983.
1) H. Volmerange and A. Witteles, "Radiation Effects on MOSPOWER Transistors," IEEE Transactions on Nuclear Science,
Volume NS-29, No.6, December 1982.
2) D. Blackburn, et. aI., "VDMOS POWER Transistor DrainSource Resistance Radiation Dependence," IEEE Transactions
on Nuclear Science, Volume No. 28, No.6, December 1981.
3) G. Roper and R. Lowis, "Development of a Radiation Hard
N-Channel Power MOSFET," IEEE 20th Annual Conference
on Nuclear and Space Radiation Effects, July 18-20, 1983.
4) R. Severns, "SOA and Thermal Design for MOSPOWER
Transistors," Siliconix Application Note AN83-1O.
5) R. Severns, "Switchmode and Resonant Converter Circuits,"
International Rectifier Application Note, 1981.
5-108
Applications Information . .
I
Chapter 6
Applications Information
6.1 Power Supply Applications
6.1.1 High Frequency Power Conversion with
FET -Controlled Resonant Charge Transfer
Abstract
The author discusses FET-controlled, high frequency,
sinusoidal, power conversion. An off line multi-output
power converter with output power exceeding 250
watts up to 500 kHz has been demonstrated. System
operation is such that current flow occurs as
successive packets of charge made from individual
sinusoids of current. The resonant charge transfer process where switching loss is independent of
frequency is described along with FET circuit
implementation.
lower component loss and reduced component stress.
Also there is no voltage or current overshoot.
Resonant Charge Transfer (RCT)
Figure 1 shows a diagram of the power-train for an
RCT converter. Operation is as follows:
Filtered AC is converted to DC between (160-250)
V. FET Ql and Q2 are alternately turned on for a
fixed period of 6 /LS but with a load-dependen~ variable PRF. The primary of the power transformer T
is connected to the common point of QI and Q2 and
in series with the resonant circuit consisting of L I
and C3 + C4. Diodes D3 and D4 constrain the voltage excursion across C3 and C4 thereby stabilizing
the resonant tank.
Introduction
Pressure to design cost effective reliable switching
power supplies in smaller space points to high frequency operation. Further, the thermal impact on
component reliability requires that supplies have
higher efficiency which implies reduced switching
loss.
The secondary sections of T are auto-connected to
give best cross-regulation and the main +5 V output
is filtered and fed back (Vsense) to control the _
switching PRF of QI and Q2.
Power conversion employing square waves has a
fundamental disadvantage, vis-a-vis sinusoidal operation, because squarewave operation requires that the
power switch (or its associated snubbing components)
dissipate energy whenever (each time) voltage and
current are handled (interrupted). Therefore, switching
loss is directly related to the operating frequency;
which is not the case using sine waves.
Low voltage differential series-pass FET regulators
are used to regulate the auxiliary outputs.
A separate low power transformer is used to supply
logic voltage but only during turn-on. After the system
is up and running the rectified + (13- 15) V secondary voltage is fed back to the input of a + 5V
regulator chip in the bias supply. This is done not
only to improve the system efficiency but in addition
to allow the logic system to be fed through the converter to utilize the energy stored on CI and C2 to
improve hold-up-time after loss of the AC input.
This paper presents a method that uses resonant charge
transfer for DC to DC conversion. Resonant charge
transfer gives low switching loss because each sinusoidal wave of current is allowed to terminate before
the voltage forcing function is withdrawn. As a consequence, with resonant operation, the controlling
switch turns both on and off at zero current to give
Reprinted by courtesy of PCI, April 1983 Proceedmgs
6-1
SERIES RESONANT TANK
AC REGULATION
AND SURGE CONTROL
5W
c,
'60 '""\..., 25••
AUTO SECONDARY
-(14-17) Y
\
II
c.
II
'.W
II
At POWER 250 W
OUTPUT
1BDW
EFFICIENCY 72%
II
'.W
RCT Converter Pseudo System Diagram
Figure 1
FIXED
EH "" 2 E-N5 V
+~o---~----------------------,
Ip .. EH
vr1/4 POWER
SA
I
I
..•• ..
I.
'2
I
G.•
5••
3.G
2.G
"., ,I..
RUN--..
G.
E~_~~
W -5
PIN
750
550
360
200
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
1"'6p.8~
Po (MAX)
420
375
260
'50(80)
'2'
I
__ ~ __ ~~~~ ____
D =1
I
:~:I
I
w=nVi:C
-ECo----4----------------------~
M A X _ '25
NORMAL.,.
MIN----..
HOLDUP ....
FULL POWER
'Om
I
POWER. EiPrI
I
I
n = 72
"
Resonant Waveforms
Figure 2
6·2
- _ . FULL POWER
Figure 2 shows the current waveforms for the resonant
tank. Relevant operating values are obtained as
follows:
the capacitive input Pi-filter, V(A) and V(B) have
the wave shape indicated. At 10 A output current the
I/LH choke has a current variation of ±0.5 A.
Assuming a minimum operating DC bus voltage of
E= ±SO V and that the junction of the two 0.05 /Lf
resonating capacitors is at -SO V, then when FET
S + is turned on a sinusoidal pulse of current will
flow. The peak of the sinusoid is equal to 6 A. This
is because the reflected voltage across the primary
winding of the 14 to I turns ratio transformer is about
SO V, and therefore, the voltage head EH for the
resonant loop is also equal to SO V (the effective
primary loop inductance including reflected secondary
leakage is = IS /LH). At full power output S+ and
S- are alternately triggered at 6/Ls (W+ D) intervals
and the 4 /Ls sinusoids are separated by 2 /LS intervals
(D). Under this condition the average current is 2.5
amperes and the power drawn from the DC source
is 200 watts.
In order to control the output voltage at
V(B)
is compared to a stable reference voltage such that
when V(B) goes below the reference the one-shot
multivibrator (OS) is triggered. The OS produces a
6 /LS (fixed width) pulse which retriggers the primary
power switches. Accordingly, a new packet of charge
is added to the 15 mf filter input capacitor whenever
it is required (up to a maximum PRF equivalent to
6 /Ls). This control strategy is termed fixed-pulseadaptive-variable-PRF because the PRF automatically
adjusts to compensate for input DC bus variations, as
well as load changes. This discontinuous format,
where the control loop gain is zero between pulses,
is quite stable because the 6 /LS pulses that drive the
primary switches are inverted and fed back to the
reference side of the comparator. This assures that
each pulse "uncrosses" the comparator input.
®,
At the maximum DC bus voltage of 250 V (E= ± 125
V) the voltage head is 170 V which gives a 12 A
peak pseudo-sinusoid pulse of current with a linear
fall rate of about 4.5 A//Ls (SO V + IS /LH). With
12 A pulses the maximum average current flowing
in the resonant tank is now about 6 A and the power
taken from the DC source is 750 watts.
The measured output regulation is 1 mViA (Ro=
1 mil) from zero to 16 A. Also, it has been found
that by adding a feedback path from V(A) to the
reference side of the comparator the output impedance
Ro can actually be made negative. The reason is that
with V(A) feedback, the current flowing through the
input resistance in the filter choke causes the average
value of V(A) to increase. Thus the output voltage
is a stable positive function of load current. One last
item, with 5 /LS sinusoids (100 kHz) the 1 /LS choke
has an impedance of 500 mil. This is sufficient for
effective voltage filtering yet allows an output current
slew rate of 1.0 A per mS per mV (very good transient
response).
RCT Control
Figure 3 shows how the RCT converter is controlled.
The resonant primary sinusoids are transformed and
rectified to give approximately SO A 5 /LS (nominal
width) current pulses. Using the values shown for
II
AC_
2mO
10mn
(t/4e = 1 :.ymv)
OIJA_r\
--'-+-~fIIV-"'VIB)
{
250mV
1 mU'--- STATIC
IDA
mS mY --DYNAMIC
System Waveforms and Control
Figure 3
VeAl
:
I
I
I
I
CHOKE
I
I
I
I
I
I
I'I
~
I
I
6-3
\-
~
I
FEED
BACK
!
VARIABLE PRF
(\
!lli!
AI = 1 OA
CONTROL FIXED PULSE ADAPTIVE
I mF
-I5J'8~
f--oV(A)
___
I(I.mo)
I
I
I
'Iii
tiN '" 10 mV
~
I
VIB)
-VREF
-
I
FET Regulator
Signal.SD, controlled from under/over voltage (uV/
o V) circuits, which are not shown, serves to tum off
the two series-pass PETs and crowbar the 4.3 V
output. When low, SD turns off the regulator. When
SD goes high the regulator will tum on, provided
the output current is below IL.
In the future the FET will find wide application in
the design of low voltage DC linear regulators. Figure
4 shows an example of a very low drop self-protected
switch for an isolated 2 A output.
Two individual SO m!l n channel PETs are connected
directly in parallel and controlled by shut down signal
SD which is obtained from a logic chip with open
collector output. When SD is high the two PETs
conduct current with a roS(on) of about 40 m!l (at
100'C). The maximum gate to source voltage is set
by the Zener diode to give current limiting. Note that
the PET type and Zener voltage can be selected to
give current limit (IL) at the temperature stable point
of the FET transconductance. When SD is Iowa third
PET, used as an output shunt, clamps the output
terminal to ground.
The nonlinear connections to capacitor C2 and to the
gate of the crowbar PET serve to ensure that when
SD changes state the crowbar PET turns on after the
series-pass PETs tum off and turns off before the
series-pass FETs turn-on.
Capacitor C introduces a small delay which provides
a transient over-current capability.
The regulator configuration of Figure S is simple
(mainly three PETs and one LM324 chip), has low
drop, and is flexible and functional in design (Le.,
additional FET units can be paralleled to give higher
current). Further, when using n-PETs for positive
voltage regulation or p-PETs for regulating negative
voltages, the basic regulator loop is unconditionally
stable and gives good static regulation (= 0.2S AI
p, s- m V). The regulator loop is unconditionally stable
because the series-pass PETs are used as a source
follower (voltage gain < 1.0) yet they have high
frequency response (even at low drawn-source voltage) and therefore the LM324 amplifier provides the
low frequency dominant pole.
Figure S shows an example of a low drop, low output
impedance, high current FET regulator.
The design provides for an isolated +4.3 V output
with a current limit at = 11 A. Two low voltage SO
mn PETs are connected in parallel and their source
voltage is compared to REFV. Feedback through amplifier A (type LM324) controls the FET gate voltage
to give an output regulation of = 1 mY/A. In order
to obtain a current limit function the output of A is
also fed back through B. Accordingly, when the PET
gate voltage exceeds REFI, the voltage level of the
REFV input to A is reduced. The values for REFV
and REF{ are set to current limit at about II.S A.
Also, REFC is set such that the negative going output
of B "trips" C to crowbar the output at a level just
above current limit.
The reliability of FET low voltage regulators should
prove quite high; the power components have low
internal voltage stress, the PET has inherent ability
to handle large short duration power peaks, and for
DC regulator application large device leakage currents
REGULATED
INPUT
+500 V o----+..----+--.
+492 Y
G
+12
+12
REGULATED
OUTPUT
~r-~----------~~~~2~A
vo-'VVI.---+---+---fCl-i
I
'uv
FEEDBACK
TolD
25OC-'/1..-100"C
~I
• SIMPLE
-FUNCTIONAL
• STABLE
IL - - - - - - -
Low Voltage FET Switch
Figure 4
6-4
~:~o-------~-----r--~ r~~-+--~----------------------~---QI~~P~
I
I
,
I
I
I
G
TO UV/OV SENSE
t -.......M-o REFy
CROWBAR
REFI
TURN-ON
DELAY
w------~--~--~
:1: \'"~ .~~"""
LM324
REF
• FLEXIBLE
·~~~~I~::~NSE
0
1-
-FET
STABLE
• REG 1 mY/AMP
RELIABLE
STABLE LOOP
Low Voltage FET Regulator
Figure 5
are pennissible. In fact in "test regulator circuits"
the FETs have been run so hot "the leads tum blue"
with no perfonnance degradation.
AC input-to-DC bus control is accomplished using a
conventional SCR half-bridge with a perhaps unconventional but flexible SCR controller. The control
method is shown in Figure 6.
Surge Protection; Feedforward Phase Control
Rectifier Control
For an input sinusoid of AC voltage a priori amplitude
infonnation is not available. Never use the conventional zero crossing time delay method for SCR trigger
if it is necessary to isolate AC surges for the DC
bus; instead, use a post-peak level-recross logic system where AC surges are automatically eliminated.
The control system operates as follows:
The goals for the power supply using RCT are shown
in Table I. From the outset it was appreciated that this
is a challenging specification especially for volume
production. The design requires low power component
stress, and AC surge control is mandatory.
Negative AC sinusoidal loops from the secondary of
the bias transfonner are used as input AC reference _
signals (the positive loops are distorted by the +5 V
I~ic load) and inverted to give positive AC loops at
®. The AC waveform at
is then co~ared to
a DC reference level
to give wavefonn \gJ. When
the ISD signal is high wavefonn @ causes the SCR
to be triggered at the trailing (rising) edge of @.
Noise protection is provided by setting the values of
Rand C so that insufficient trigger energy is available
to the SCR below a minimum width of waveform
Table I
Resonant Regulator Specification
Vin
{(80 - 150) Vnos' 200 Vrms surge
(160 - 300) Vnos' 400 Vrms surge
eEl
©
Po: 180 W; ± 12 V, ±5 V, +5 V (16 A), +4.3 V (II A)
Regulation: ±50 mV (total Matic + dynamic)
Hold Up Time: 30 mS, +5 V (16 A)
B & W: All Standard, Remote Sense, Sequenced Turn On/Off,
Self Protecting, u V/o V Clamp
@.
Size: 254 in3 (.147 ft 3)
©
The reference level at
is adjusted by the peak
synchronized squarewave @ where the resistor values
are set to allow the SCR to fire just after the peak
Reliability: 5 yrs.
Cost: minimum
6-5
•
SCRTRIGOER
' " CON::l':nOC
GAIN
FA~II
I
I
I
I
I ZERO CROSSING
~~t--.!!!!.t
®,
r
®A?TVn-r--=-"'1
1
-1I
I
REF@)
I
I
1+5VI
I
I
®
aior-,
r!
r;:..r--tHi
\V
I
I
I
I
I
160 V rmll-.l
L __ .J
L __ --I
: ....I t==i
~==::J
90 V rms -1
I---;
r---t
80 V rmll.-i---J
L.......J
AC AMPLITUDE DET
AND~
I
AC Surge Control
Figure 6
when the AC input is between 80 Vnns and 90 Vnns .
This is done to give maximum DC bus voltage at low
AC line.
resonant conversion should allow much higher
operating frequency than will square wave
operation.
Finally, because the output of a phase control rectifier
is sensitive to amplitude dependent voltage slope
(dV/dU; the trigger point is adjusted upward as the
input AC voltage increases. This is accomplished by
detecting the average AC ~ak amplitude and adjusting the reference level at (£Y. The 324 integrator also
gives a soft start function at "Power-up".
2. Reduced component stress.
In low power systems (say less than 100 W output)
component count may be a meaningful, perhaps
dominant measure of reliability. However, in
higher power ranges, especially when operating
in harsh environment, component stress is dominant. The resonant mode, operating without current
discontinuities, has little voltage overshoot, much
reduced di/dt, and requires much less snubbing
compared to square wave operation.
Conclusions and Recommendations
Several authors have written excellent papers regarding the do(s) and don't(s) of high power, high frequency design so this subject is not discussed here.
It suffices to say that the selection of magnetics and
filter capacitors are important as is circuit layout and
grounding. However, the following items should be
emphasized.
3. Use field effect transistors.
In applications below 500 V and 50 A the PET bipolar race is about over; the PET is fast, more
rugged (where it counts, i.e., SOA), more efficient
and cost effective. Indeed, for sinusoidal operation
the simple low cost proportional drive long used
with bipolar power switches is unavailable.
1. Sinusoidal operation at high frequency is an
important way to reduce the size of power
supplies.
The ease of design with FETs, their speed, voltage-activated low power drive requirements, unifonnity and stability now allows them to be an
item of commerce (almost anyone's equivalent
type will work without circuit tweaking).
Resonant operation, where the sinusoidal current
pulses are allowed to naturally tenninate, greatly
reduces switching loss in the power handling
components, especially in the power switches and
power diode rectifiers. Accordingly, in the future
6-6
4. Use FETs for low voltage linear regulation.
However, the trend is clear; the PET switching
speed is quite adequate at 1 MHz sinusoidal operation, as are film capacitors.
Low rDS(on) gives low voltage drop without saturation. The resulting high frequency response at
low voltage coupled with the ease of drive, ability
to parallel devices, the stability and uniformity of
electrical parameters (especially transconductance),
make them ideally suited for low voltage regulator
applications.
In the future the Gallium Arsenide FET technology
offers potential for factors of 5 to 10 improvement
in both speed and rDS(on). If this materializes,
coupled with superior high frequency magnetic
material such as Metglass, then sinusoidal operation in the 1.0 - 5.0 megahertz range appears
practical.
5. Technology.
For higher power systems, resonant charge transfer
(where current flow naturally terminates) is compatible
with the SCR switch.
The RCT system described has been operated with
1 JLs sinusoids (it did not work as well as with
5 JLS sinusoids-the magnetics require redesign).
6-7
6.1.2 Practical Desig~ Considerations
For a Multi-Output CUK Converter
©
1979 IEEE. Reprinted, with permission, from PESC '79 IEEE POWER ELECTRONICS SPECIALISTS
CONFERENCE, June 18-22, 1979, San Diego, CA, pp. 133-146.
Abstract
An experimental high frequency three output DC-DC
converter design is presented in circuit detail, emphasizing the practical design needs of its contemporary optimum topology (t:UK) output stages. Power
component selection criteria are also given along
with suitable load protection methods against output
voltage reversals at turn-on. Inductor coupling criteriafor ripple current reductions at input/output ports
are developed including an alternate method for external tuning inductance insertion. Results of corresponding empirical evaluations of a 200 kHz, 55-W
representative design are discussed.
yielding maximum performance. The current work to
date has been spearheaded by a team of researchers
at the California Institute of Technology who, in 1977,
revealed their fmdings [1], to the design community.
The topology was appropriately termed optimum as
it achieved all the desired goals stated above.
Since 1977, further extensions [2, 3, 4], modeling
methods [5, 6J and practical applications [7, 8, 9]
have been suggested by the developers of the new
topology which now make it an even more viable
and attractive circuit structure for power conditioning
designs. It is the intent of this paper to present a
practical application of many of the features of the
optimum topology (CUK) converter, including multiple outputs, isolation of input/output grounds and
coupling of output filter inductances. Section 2 contains a short review of the basic topology operation
and evolution of extensions for the reader's benefit.
While total output power control achieved is less than
100 watts, the design presented does demonstrate the
compelling simplicity of the topology and its many
advantages.
1. Introduction
This decade has seen many significant advances in
the art of power conditioning system design and contemporary electronic component technology. In the
area of DC-to-DC conversion methods, the designer
now has LSI components at his disposal for converter
control, amplification, and supervisory functions. Yet,
with all the current progress made in DC power system design, the main power translation stages have
remained relatively untouched. This lag can be easily
attributed to the need for this part of a DC-to-DC
network to control relatively large amounts of power
to perform its function. Consequently, the main power
conversion stages must be formed from discrete active
and passive circuit devices that contribute significantly
to the total design's complexity, size, reliability, and
power loss.
Because of its promise of significant and controllable
current ripple reduction at chosen circuit ports, a
three-output CUK converter power stage was included
in an experimental pulse-width modulation system
undergoing evaluation for high-frequency operation
under an in-house R&D study program. It must be
emphasized that the resultant total system design to
be described herein does not represent a truly optimum
network from power loss or bandwidth standpoints,
as the main goal of the study was design information
and education for system practicality and producibility. Nevertheless, it was proven that it is feasible to
For these reasons, a great deal of effort has been
expended on the search for a DC-to-DC power stage
topology that has the simplest possible circuit structure with the minimum number of components while
6-8
utilize the CUK topology in converter applications
requiring switching frequencies approaching 200 kHz
with predictable circuit responses assuming proper
layout and noise prevention practices [10] are
followed.
average current transfer ratio magnitudes for each of
the circuits of Figure 1 are simply (continuous inductor currents assumed)
7jD
I-D
Perhaps the most important achievement of this particular design is the simultaneous reduction of input
and one selected output port ripple currents to zero,
utilizing tightly coupled inductor windings and small
external inductances for ripple current adjustments.
From a system design standpoint, this implies no
need for EMI filtering for switching current noise at
the input to the converter. In addition, no output
filter capacitance for ripple voltage reduction is necessary at the selected output port.
= 7jD =.!.!.
D'
,D'
=I -
D
(1)
12
where D is the duty cycle of conduction of transistor
QI with respect to a constant timing period, TS, and
7j is the converter circuit efficiency.
Section 3 presents the details of the total power conditioning system, including a somewhat unusual ramp
slope control method of pulse-width modulation.
Section 4 is devoted to power stage component selection for proper ratings and electrical values. In
Section 5, the problem of output voltage reversal
with power application is explored when input and
output inductors are magnetically coupled, along with
simple preventive circuit measures.
(a) Basic Inverting Topology with Nonpulsating Input/Output
Current Waveforms
C1
Ripple current projections in a four winding inductor
assembly are considered in Section 6 with emphasis
placed on achievement of predictable coupling coefficients to increase the effective inductances of selected windings for current ripple reduction. This
subject is continued in Section 7 where it is demonstrated that the desired ripple current reduction can
be accomplished by selection of small external inductances working in conjunction with tightly-coupled
multi-inductor components.
(b) Coupling of Input/Output Inductances
(Single Core Construction)
(c) DC-Isolated Extension with Separated Coupling
CapaCitance
Laboratory evaluations of a representative 200 kHz
converter are presented in Section 8 as verifications
of the analytical predictions and circuit responses
given in earlier sections.
2. Background
(d) Coupled-Inductor, DC Isolated Extension
The operation of the basic CUK converter stage h~s
been covered in excellent detail in the references [I
through 4]; however, for those not familiar with the
topology, it is worthwhile to briefly review the fundamental design features and extensions.
Circuit Evolutions of the CUK Converter
Figure 1
Capacitor CI serves as the single energy transfer
device between the input and output ports of the
circuit in Figure I(a). During the interval D'TS when
QI is OFF, diode D is forward-biased and CI is
charging in the positive direction through inductance
Lp. The voltage across the collector-emitter junction
of the transistor is therefore positive and can be turned
ON for the subsequent interval DTS. As soon as QI
turns ON, CI becomes connected across the diode,
thus reverse-biasing it. Thus, during the time DTS,
Figure I illustrates the evolution of the new converter
topology from its basic form (Ia) to coupling of
input/output inductances (lb), introduction of transformer isolation (Ic), and fmally, recoupling of isolated input/output chokes (ld). The principal electrical
features of all of these converters are nonpulsating
input/output currents as shown graphically in Figure
I(a). It is easy to demonstrate that the voltage and
6·9
Cl discharges through the load RL and inductance
Lo, charging the output capacitance C2 to a negative
voltage. Circuit operation is repeated when interval
D'TS is reached.
of the topology [4]. As shown in Figure l(c), the
energy-transfer capacitor is simply broken into two
parts, Cp and CI, with an isolation transformer separating the two halves of the circuit. Since no average
or DC voltage can exist in the ideal sense across
either transformer winding or the two inductances,
the voltage on Cp is VQ and that on CI is V. Note
that the sum of these two capacitor voltages is VQ
+ V = VQ ID' = VID, the same potential as that
across the single coupling capacitance CI of Figures
lea) and (b).
Because inductances are present on the input and
output lines, the currents are nonpulsating with no
sharp transitions that contribute to increased EMI as
are present in conventional buck and boost designs.
Output voltage magnitude can be less than, equal to,
or greater than the input voltage value, depending on
duty cycle as implied by (1), thus achieving true generallevel conversion.
Since both windings of the transformer in Figure l(c)
are isolated by the two energy-transfer capacitors, no
DC transformer core magnetization can take place.
Automatic volt-second balance is therefore achieved
in the topology, requiring no special magnetic or
circuit designs for balancing, as might be the case in
a conventional "push-pull" converter. A smaller ungapped, square-loop core can be used for the transformer as a result of this circuit feature, with lower
core and winding copper losses than would be possible
with comparable converter topologies, as discussed
in length in [ 4] .
The first extension of the converter topology to come
to light was the feasibility of coupling of input and
output inductances [2], as illustrated in Figure l(b),
without impact on the basic DC-to-DC conversion
property. Examination of the ideal voltage waveforms
across Lp and Lo reveal that they are identical in
magnitUde and timing, as diagrammed in Figure 2.
Therefore, with proper choice of coupling coefficient
and turns ratio, both inductances could be wound on
a single core resulting in obvious reductions in converter size, weight, and component count. It has been
shown, in fact, that significant reduction in ripple
current magnitUdes can be achieved due to the magnetic coupling between the chokes if the turns ratio
and coupling coefficient maintain specific mathematical relationships [3]. DC bias effects on the core
material are, of course, additive because of the required phasing between windings. This fact must be
taken into account when selecting the induction core
material to prevent saturation due to DC magnetization
effects.
Because the basic converter operation of Figure lea)
is unchanged from that of Figure l(c), the desirable
feature of inductance coupling is still possible, as
shown in Figure led), reducing the number of circuit
magnetic assemblies to two.
The extension to multiple and isolated outputs [4]
should now be obvious, as shown in Figure 3 for a
three-output design. Again, by coupling of circuit
inductances on one magnetic structure, only two
magnetic assemblies are required for the total power
stage. However, one must judiciously choose the
coupling coefficients and turns ratio relationships be-
The problem of an inverted output not isolated from
input ground potential was solved by a later extension
V,-
--,-----,-- -------T
I-----~-;;---l
<'--II
I
r---------,
1 V1-Vc-~V2+t,=VJ! 1L
_________
J
!------'I-I
1 L
V,-Vl=V,-O'=-o'v,
I
I'
I
- - - - - -11f----...;- - - - - - -1----,>
~
I:----OT.
"'~!-----T.
Ideal Lp, Lo Voltage Waveforms for the Circuit of Figure 1a
Figure 2
6-10
1
_ _ _ _ _ _ _ _ _ _ -.l
+5 V, + 15 V, and -15 V DC, with corresponding
maximum current capabilities of + 5 A, + 1 A, and
-1 A DC, respectively. Since the +5 V load was
the highest and desired tolerance, the lowest of the
three outputs, the + 5 V output was chosen for converter pulse-width modulation (PWM) control with
the ± 15 V lines controlled by cross-regulation through
the converter transformer secondary turns ratio
relationships.
tween windings in order to steer and reduce ripple
currents in the directions chosen for design needs.
Transformer turns ratios are not restricted to 1: 1, but
the ideal transfer voltage relationship specified by (1)
must be modified to include the transformer turns
ratio factor. For example, for the second output V02
in Figure 3,
(lID )
V02
- = n2 X - - , n2 = NsZlNp
(2)
V£
I-D
One further observation about the topology of Figure
3 is needed. Note that the primary and secondary
coupling capacitances have been relocated to the other
side of the respective primary and secondary transformer windings in contrast to those shown in Figure
l(c) or l(d). Because the capacitances are transformer
coupled and isolated by T1, they can be placed on
either side of their respective winding with no change
in circuit operation. Since the cases of electrolytic
capacitors are usually common to their negative terminals this placement effectively grounds the cases
of these capacitors, thus reducing the high frequency
radiated noise due to the pulsating switching currents
in these capacitors.
Figure 4 is a schematic of the overall converter system, with each functional area identified. Like its
CUK output power stage, the PWM control system
is somewhat unique with respect to its control method.
Because currently available PWM LSI integrated circuits do not use this control and isolation approach,
SSI logic and discrete parts were employed.
The system as shown consists of an error amplifier,
individual clock and ramp generators, voltage comparator, amplification and buffer stages for correct
gate drive levels for the power VMOS converter
switch. Isolation between input/output grounds is
achieved by allowing the error amplifier, which is
powered by the + 5 V output, to interface with the
ramp generator control through a high speed optical
coupler. A simple shunt regulator, consisting of a
temperature-compensated reference diode (VR3) and
bias resistor (R4) provides the single regulated voltage
needed for all remaining PWM Controls. The
input + 28 V is used as the unregulated voltage source
for this low power regulator.
3. Experimental Circuit Designs
In order that the results of the development study of
the CUK converter operation and its control electronics could be applied to representative production
program needs, a three-output configuration was selected, each isolated from one another and from the
input DC power ground. Voltage levels were set at
Lp
r--_ _ _~--------f'rrYYY""----,.----o+v,
D'"" 1-0
"
N.,
01
+V02
....2
No>
DO
N.,
n1=Np
+
N",
C2
"2=Np
n3=
-
~3
Np
N03
Tl
D3
- C3 +
Three-Output, DC Isolated, Scaled Version ofthe CUK Converter (Coupled-Inductors)
Figure 3
6·11
IDII
ERROR AMPUFIER
,..---.------.--~-__t"-+5 V(sENSE)
+28V
r----o
+28V
Lp
+28 V RETURN
\7p
,..-------fYfl"'1-----'----<> +28 V INPUT
_ _--IT'-----'T]' TT,--..-...-...,..---o +15 vee
D4
{.!..!,1f-~_ _ _ __'____'L--+___o+15VRETURN
'--.--..---0 +5
'91f-~---HI_-'----JL-----o +5
VDe
vee RETURN
'-l11Z-'--------+----'----o -15 V RETURN
C3
CUK POWER STAGE
Exper]menta] Converter System
Figure 4
ramp slope control method is diagrammed in Figure
5 for duty cycle adjustment.
02, Q4:a YN1206N2
TO Lp
UIA, another Schmidt trigger gate, in conjunction
with Rl, Cl, U2A, and diode CRl, form a very
simple 200 kHz astable oscillator for fixed-frequency
PWM synchronization. When the output of UIA is
in a low state, U2A discharges both the oscillator
capacitor Cl and the ramp capacitor C2 through
diodes CRI and CR2 respectively, until the lower
input voltage threshold point of UIA is reached.
UIA's output then goes to the logic one state, allowing these capacitors to recharge.
D
.-'-~!
Rae =
RUM
=
100n
r
+
The output of the system comparator, UlB, is buffered and inverted by the remaining four Schmidt triggers within Ul, all connected in parallel, for lower
output impedance drive through CR4 for rapid charge
of the gate capacitance of the power stage VMOS
device, Ql. This ensures fastturn-on ofQl to minimize
switching losses. U2D discharges the stored charge
in the gate capacitance of Ql when the driver gates
return to the logic zero state concurrent with the
discharge of the ramp capacitance, C2.
Alternate Design Approach for Converter Power Drive
Figure 4A
The PWM operates as a fixed frequency, variable
+ 5 V potential is
detected by the error amplifier which adjusts the lightemitting diode (LED) current in the optical coupler,
OC-I. This current change alters the emitter current
of the light sensitive transistor within OC-l changing
the base voltage bias of the ramp control transistor,
U2B. This action, in turn, changes the slope of the
ramp voltage appearing across capacitor C2, causing
a corrective shift in time when this ramp voltage
reaches the fixed high level input voltage threshold
of a CMOS Schmidt trigger inverter gate (UlB), designated as the comparator for the PWM system. This
OFF time system. A change in
The error amplifier itself is a conventional design
with U2C as a buffer for current control of the LED
within OC-1. VRl is a low voltage (1.23 V) reference
diode selected for stable temperature characteristics
with low bias current. Variable resistor R8 is used
to remove output voltage variations due to initial
component tolerance within the control loop.
6·12
-
-
-
-
-
COMPARATOR THRESHOLD
j - - _ j -_ _ _-, +10 V
VG(Ol)
tON
tOFF
DUTY CYCLE = - = 1 - - : = : 0
tOFF
Ts
CONTROL
I
°l~---~--~
I
Ts
I
1
:.~-----T'", -----l.~:
Fixed-Frequency, PWM Converter Control Concept
Figure 5
Three diodes in series (CR5, CR6, CR7) clamp the
ramp control transistor current level to a fixed maximum value at converter input power application,
thus limiting the duty cycle of drive to QI to prevent
excessive power dissipation from high inrush currents.
These diodes, along with CR8, also provide a path
for discharge of capacitance C5 in the lead-lag frequency compensation network for the converter on
input power removal. Resistor Rl3 is the companion
component in this compensation.
removal controls. Note that an additional winding
(Tl-FB) was added to the power transformer to aid
in rapid tum-on and saturation of Q1. The position
of the winding in the source of Q4 also increases the
effective VGS (ON) of Q4, thus increasing the available base current for Q1. The lower VMOS FET
(Q2) turns on when Q4 is commanded off and removes the stored charge in the B-E junction rapidly
through its low ON resistance. Resistance RUM and
RBE provide for current limiting in the drive circuitry
when the voltage across TI-FB reverses as a result
of QI being turned OFF to prevent possible upset
andlor damage in the controlling CMOS logic.
Capacitors C3 and C4 are small noise filter capacitors
for the bias regulator. A small ferrite bead on the
gate lead of QI reduces high frequency noise problems due to gate drive. A suppression zener diode
clamp (VR2) limits the tum-off transient voltage
magnitudes across the drain-to-source terminals of
Q I to safe levels.
Switching losses in this alternate buffer and power
control method are limited primarily to those due to
Q 1. If a square loop core material is used for Tl,
one must ensure that the DC imbalance produced by
this non-capacitively coupled winding does not produce saturation. This scheme for the power switch
of the converter system is very attractive for switching
frequencies in the 40 kHz range should one choose
to reduce the converter frequency of operation.
The configuration of the CUK power stage shown in
Figure 4, including component rating considerations,
are the subjects of Sections 4, 5, and 6 to follow.
3.1 Alternate Converter Power Switch
4. Power Stage Component Selection
The limited number of vendor suppliers of high current VMOS power FETs with low drain-to-source
resistance plus their current high costs may prove to
be uneconomical in some high-volume or militarized
power system applications. Lower power VMOS is
currently more available at a proportionally lower
cost.
Because the CUK power stage and its components
were the key to an optimal converter design, much _
emphasis was placed on a proper understanding of
the component values and ratings needed to achieve
reliable and predictable operation.
In our early experiments with the breadboard of the
multi-output converter, we used an alternate design
in the area of buffer stage and power control. This
alternate scheme is shown in Figure 4A. A highfrequency power transistor (2N6277) is used in place
of the high current VMOS (Q1 in Figure 4) and
lower current VMOS devices used as base drive and
In the matter of the energy transfer capacitances,
because they are asked to transfer the entire power
in AC form from input to outputs, the rms ripple
current capability of these parts is a prime consideration. This subject was covered in [4] and [8] for
single output CUK converters; however, the results
can be used to predict the needs for a multi-output
situation. Referring to the scaled and triple output
6-13
design of Figure 3, ripple current needs for the four
energy transfer capacitances (Cp, CI-C3) can
be estimated to be
VQl (OFF)=Vll+ Vol =V,,+ V02
nl
"
n2
=y, + V03
II
n3
(3)
iCI (rms) = 101
In a similar fashion, reverse voltage levels appearing
across the three secondary commutating diodes are
_/Y02
iC2 (rms) = 102
(4)
V~
where s again is the secondary number (1,2, or 3).
(5)
V~
The peak current level in QI is the sum of the average
line current, Ill, and the reflected secondary load
currents.
3
iCp (rms) =
L nsics, or
3
IQl (ON) = 1£ +
s= I
y:;iVi
x los,
In terms of output voltages, (12) can be restated as
3
IQl (ON) =
V02, VC3
=
V03
IFDs (ON) =
DTS
Vllns
1]
los
(14)
s= 1,2,or3
Since the output filter portions of the converter of
Figure 3 are essentially that of a buck regulator,
inductance and capacitance values for L o l - Lo3
and Co l - C03 can be formulated from standard
formulae [11] based on ripple current and voltage
needs for each output. Ripple current reduction by
inductive coupling between the input and output
chokes of Figure 3 will be considered in Sections 6
and 7. Like the output filter networks, the design of
the isolation and scaling transformer (TI in Figure
3) can follow conventional techniques. For operation
at high frequencies, ferrite toroids are recommended
for the small core areas and low loss. Winding techniques for leakage inductance reduction are mandatory
to reduce transient voltage spikes across semiconductor junctions during circuit transitory periods.
Choice of turns ratios are based on desired limits of
duty cycle variations with line and load changes,
stress limitations on the surrounding semiconductors
and energy transfer capacitances, and, of course,
cross-regulation needs for the unregulated outputs.
(8)
where Pll is the total converter input power, and III
is the average input current at that power level. Similarly, for the secondary transfer capacitances,
100~
[~+
7j
Cp>
Cs >
(13)
II
Forward current stress levels in the secondary commutating diodes is related to the corresponding secondary load current by the following relationship.
Capacitance values of these parts are chosen to keep
their voltage ripple magnitudes to reasonable values
to minimize OFF voltage stresses on QI and the
secondary commutating diodes. A good first approximation is I % of the corresponding average voltage
stress for ripple voltage peak-to-peak magnitude. For
example, for capacitor Cp in Figure 3, using the
basic expression ill = CpdV/dt,
IllDTS
, or
0.01 Vll
P£ DTS
Cp> 100
2
V£
L [:~s +ns] los
s=1
(7)
=
(12)
s= I
VCp = Vll' VCl = Vol,
VC2
L nslos
s= 1
(6)
3
L V ns Vos
iCp (rms) = _1_
(11)
VDs (OFF) = Vos + ns Vll
_/Y03
iC3 (rms) = 103
(10)
(9)
Vos2
where s is the secondary number in question (1, 2,
or 3).
Peak OFF voltage Q 1 of Figure 3 is simply the line
voltage magnitude plus the reflected secondary capacitance voltage of TI.
6·14
Even though gapping of the core material is not required in the ideal sense, as mentioned in Section 1,
rapid load demand changes or sharp excursions in
line voltage may produce transient conditions such
that core movement into saturation is feasible. One
practical application [8] did experience such situations;
therefore, precaution must be exercised in evaluating
the need for a gapped magnetic core for T1 in light
of the intended converter application and dynamic
input/output electrical specifications.
by the time constants and magnitUdes of the power
stage impedances and output load values.
Figure 7 suggests some solutions to reversal protection
for loads and output filter capacitors. In Figure 7(a),
a blocking diode is placed in line with the secondary
inductance to ensure a proper undirectional current
path for steady-state output current, but prevents reverse current flow at turn-on. However, its presence
does add to the power waste in the converter during
normal operation. Figure 7(b) adds an output clamp
diode to limit reverse transient potential to less than a
volt, but does not conduct during normal operation.
Use of a Schottky diode here is recommended due to
the gray areas presently in prediction of damage
thresholds for ICs or polarized filter capacitors.
5. Transient Output Reversal
In eUK converter designs where magnetic coupling
of input and output inductances is desirable for ripple
reduction, one undesirable feature of operation is introduced; namely, that of transient voltage polarity
reversal on power application. For systems where
short term reversals of outputs could produce damage
to output capacitances or loads (e.g., monolithic ICs),
this phenomenon must be circumvented or reduced to
acceptable magnitudes.
The solution of Figure 7(a), while lowering converter
efficiency, may be more attractive when the problem
of transient currents in Tl producing core saturation
is considered as mentioned in Section 4. There may
be more sophisticated methods for reversal prevention;
however, the two noted in Figure 7 seem to remain
the simplest found to date.
Figure 6 is a diagram of this temporary reversal for
a single output C:UK converter with transformer isolation and coupling between input and output chokes.
Inrush current through Lp due to first tum-on of Ql
or by charging of the primary energy transfer capacitance Cp when VQ is applied produces a secondary
transient current in the direction shown in Figure 6.
This direction is opposite to that of normal current
flow in converter steady-state operation, and results
in a reversal of output voltage polarity for the transient
time period.
6. Ripple Reduction by Inductor Coupling
The proportionality of the two inductor voltage waveforms of the circuits of Figure l(a) led in [2] to the
idea of inductive coupling and to the subsequent coupled-inductor extension of the CUK converter shown
in Figure l(b). At first thought, the equal amplitUdes
of the inductor voltages implies a need for l: 1 turns
ratio only (or Lp = Lo for perfect coupling). However, in [3], it was shown that there is really no need
for such restriction and the coupling of inductors may
be achieved at any inductor values. In fact, proper
choice of the relationship between turns ratio of self
inductance and the coupling coefficient can result in
This condition will persist until the potential levels
of Cp and Cl reach values to counteract the voltage
across Lol produced by the coupled primary inductance voltage. The reversal period will be determined
.-------------.
I
INRUSH CURRENT (Q1 OR Cp)
L ______
/-
~
/
I
I
~~~~~-~_Lp •
+Vo
I
r------------..,
I
PRODUCES CURRENT IN
I
L_~~~~~~'~~~~_.J
illustration of Output Reversal on Turn-On
Figure 6
6·15
ADD
BLOCKING
T1
~
I~
DIODE
L.,
RL1
.,
C,
+
Vol
D'
~-~--~--------------~--~--~
.~[--.-:-+-c:;_~~~:f~D'~:~_"'_'_~~~~t~:-Co~'~f_R~L'~f~---D_C~~~·~Suggested Methods of Protection for Output Polarity Reversal on Turn-On
Figure 7
zero current ripple at either input or output ports of
the circuits of Figure l(b) or l(d), as dramatically
demonstrated in [3] for a single output design.
where Lp, Lol, L02, and L03 are the self or opencircuit inductances of the windings, and the various
M's represent mutual inductance values between the
four windings. We now define effective turns ratios
(no 1--' n03) and coupling coefficients (kl--' k6)
as
In our case, however, the problem of ripple current
steering and reduction is magnified due to the presence
of more than one secondary inductance and the corresponding complex interrelationships between turns
ratios and coupling coefficients. However, this problem is simplified if a first-order model, such as that
shown in Figure 8, is utilized for analysis As shown,
all windings are mutually coupled (M's), and the
assumption is made that the primary and secondary
voltages are related to the primary potential by proportionality factors a, b, and c. In the design of
Figure 4, these factors would be the secondary tum
ratios of Tl, or a = nl, b = n2, and c = n3,
assuming an ideal transformer model for Tl. Proceeding with the voltage matrix for the model of
Figure 8, we find
~
vPJ ~LP
avp
bvp
cvp
=
Ml2
Ml3
M14
A... lIP
no I =
V "La!'
A...!LP
Uo2 =
V Lo2'
A" lIP
n03 =
V Lo3 '
(16)
Ml3
M23
L02
M34
V1 = avp
'13
First-Order Inductor Model for Ripple Current Predictions
Figure 8
6·16
Z:::CYp
Substitution of these definitions into the matrix of
(IS) and solving for the various current derivatives
will then allow us to solve for each winding's effective
inductance value in relation to its self-inductance.
After a few lines of algebra and the aid of Cramer's
rule,
dip _
vp
di2
bvp
di3
cVp
I
NEGATIVE
NEGATIVE
POSITIVE
POSITIVE
POSITIVE
POSITIVE
NEGATIVE
NEGATIVE
c0 0 3 kS> aOol
@)
b00 2< k2
"°0 1 kS> c003
k'S
::::::>
COUPLING COEFFICIENTS"" 1(ALL WINDINGS)
First-Order Inductor Model with External Inductances
Figure 9
6-19
,
,
nol =
, no2=
ftp
,
, n03 =
V~
LpL02
' -k3
k3-
~P~02
where ~p = Lp + Qp, ~ol = Lol + Ql, ~02 =
L02 + Q2, and ~03 = Lo3 + £3. Repeating the
voltage matrix and solution for current rates, the effective inductance expressions for the four windings
become the same equations as (18 21), except
of course, for replacement of all k, n, and L symbols
with primes as indicated in (23) above. For the
condition
bn~2 1<.'2 = I and ~l = cn~3 1<.'5
(30)
when L' efp 00 and L' efsl- 00. Substitution for
prime values as defined by (23) into (30) and solving
for the external inductances yields
substitution of the values from (23) into (24) shows
that
Q2=b~ -L02
(25)
£3=
still must be satisfied within the multi-winding inductor itself to reduce the effective inductances to
the following simplified equations.
(I - ki2) ~p
(23)
~P~03
A case matrix similar to that of Table 2 can be
prepared using (26) through (29); however, that case
concerning itself with zero input and zero output ripple
on the first secondary is of direct interest to this
work. This situation according to (26) and (27) above
should occur for
(24)
LCfp =
~PLo3
--,
[:J~
for k2, k5
=
(31)
(32)
-L03
1 (tightly coupled windings).
Since the relationships of (26) and (27) do not include
Lol or Lp in their denominators, then we can set
(26)
1- bn'02 ki
(33)
(1 - kS2)
Lefsl =
1_
,
[
,
1,'
cn03, AS
anol
JLol
(I - ki2) ~02
J
and still satisfy the conditions to allow
and Lefsl ---. 00 •
(27)
,
(1 - kS2) L
[an'01 k5
03
J
00
Referring back to (24), it is mandatory that this condition be completely satisfied, as the contribution of
small differences from zero can produce large changes
in effective inductance predictions using (26) through
(29), as can be seen by examination of the basic
relationships of (18) through (21). Since it was proven
in (25) that (24) is still dependent on predictable and
uniform coupling properties of the tightly-wound inductor windings, a ripple current prediction problem
can still exist.
(28)
1- _2_
k
.[ bn~2
Lefs3 =
r.efp-
(29)
What is interesting, however, about condition (25) is
that it can be absolutely satisfied if
1---
cn'03
(34)
6-20
which implies two non-coupled inductor assemblies,
each with two windings, such that Lp and L02 are
wound on the same core and Lol and L03 on the
other. External inductances £2 and £3 are then added
in series with L 02 and L03, respectively, to reduce
both input and the first secondary (in our case, the
high current + 5 V section) ripple currents to zero.
In other words, the coupled-induct10.0
10
77
slope)
slope)
slope)
slope)
15.0
127.0
5.0
50.0
87.4
108.0
>77.0
<10.0
<77.0
102.0
111.0
50 mV
secondary voltages of Tl and T2 become unbalanced
(vI" VI, v2, and V3 of Figure 9), and the equations
(26 - 2 9 ) become invalid since they assume ideal
time phasing between all potentials.
+5V
OUTPUT
INPUT
"'"
The noise and ringing observed in the current waveforms are caused in part by the presence of inductive
and capacitive parasitics in the breadboard layout that
become significant at high switching frequencies
which modulate current levels, and by the non-ideal
nature of instrumentation with regard to grounding
and noise .
050mV
'"0
"Positive" Ripple Case
Figure 11
.. mV
8.2 Group Two: Line/Load Variations
+,V
These tests were mainly performed to evaluate the
steady state operating properties of the converter.
The tests were conducted while maintaining zero current ripple in the input and the + 5 V output. These
tests included line and load regulation and system
efficiency measurements.
OUTPUT
INPUT
I .....
22=5~.
---.
0.. mV
, .0
Table 5
Line Regulation with RL1 (5 V Output Load) = 1
RL2. RL3 (±15 V Output Loads) 15 n
"Negative" Ripple Case
Figure 12
=
VIN
VOLTS DC
The values of £2 and £ 3 for different current ripple
conditions are summarized in Table 4 along with the
theoretical values from (36). Also shown in Table 4
are experimental and theoretical values for L~fs2 and
LMs 3 under the zero ripple current condition.
22
28
32
+5V
VOLTS DC
+15V
VOLTS DC
15.02
14.58
14.37
4.98
5.00
5.00
n.
-15V
VOLTS DC
-14.92
-14.50
-14.29
Table 6
+5 V Output Load Regulation with Vln
Examination of Table 4 shows excellent correlation
between experimental and calculated values of circuit
inductances. Variations noted are in part due to the
differences between assumed and actual inductance
values of Lp, Lo l, L0 2, and Lo3, and to the measurement accuracy in determining £20 and £30 of the
fixed value, discrete chokes.
RL2
The slight change in the magnetization current of Tl
along with the input current level of Figures 10
through 12 is primarily caused by differences in
switching times of Ql and the three secondary commutation diodes. During these switching periods, the
= RL3 = 15 n
= 28 V DC,
RLl
(0)
+5V
(VOLTS DC)
+15V
(VOLTS DC)
-18V
(VOLTS DC)
10
2
1
5.00
4.99
5.00
12.05
13.15
14.54
-12.06
-13.14
-14.46
In Tables 5 and 6 we observe that the load and line
regulation of the + 5 V output are very satisfactory
and do not need any comments. It is interesting to
note that, although the line regulation of the + and
6-22
-15 volt outputs are well within required ranges,
their absolute value decreases with increasing line
voltage. This change is enhanced with respect to
+ 5 V load variations as noted in Table 6. This phenomenon can be explained when it is observed that
+ and - 15 volt output voltages are determined
mainly by the voltage impressed across the N 1 winding of T1 during the off time of Ql. This voltage is
made up of the voltage on Cl and the forward drop
of diode D 1. The voltage on C2 does not significantly
change with line variation; however, the forward
voltage drop of Dl increases with decreasing line
voltage due to increased value of the reflected current
from the primary, thus increasing the total impressed
voltage across N I. With + 5 V load variation, this
effect becomes more pronounced. Since now, in addition to the forward voltage drop of DI which
changes with load, the steady-state value of the voltage on CI also changes to make up for the changing
drops in the winding resistance of T1, Lol and in
its own ESR. As a net result, the voltage impressed
across NI during the off time of Ql increases with
increasing + 5 V load and decreases with decreasing
+5 V load.
0-
22
28
32
3.5
2.6
2.3
77.0
72.8
73.6
//
10,..8
(b) With Diode Clamp
+15 V Output
Output Waveforms at Turn-On
Figure 13
output effectively clamps this negative spike to a
diode drop below the + 15 V return line.
9. Conclusions and Remarks
A high frequency triple-output DC-to-DC converter
design utilizing a CUK output power stage is proposed
herein, with magnetic coupling between selected output filter inductances for attaining zero ripple currents
at both the input power port and one output. The
design also includes transient polarity reversal protection at the output port magnetically coupled to the
input line.
Po
%
(TOTAL) EFFICffiNCY
(WATTS)
55.0
53.2
52.4
10,u.&
./
Table 7
Efficiency Measurements at Full Output Loads
IQ
PQ
(AMP DC) (WATTS)
/
--'
(a) Without Diode Clamp
+15 V Output
In Table 7, efficiency figures are calculated for minimum, nominal and maximum line voltages while the
converter is delivering full output power. These figures compare well with practical efficiencies that are
presently being obtained from conventional switching
regulators operating at much lower switching
frequencies.
VQ
(VOLTS DC)
'"'" ---
b-
/
71.4
73.0
71.2
Power losses are attributed to those produced by
semiconductor switching actions, dissipation in the
energy transfer capacitances, IR drops in QI, and
resistances of magnetic windings. As mentioned earlier, the experimental design did not include optimization measures to minimize these losses.
Although the control electronics for converter pulsewidth-modulation were somewhat unconventional due _
to the use of ramp slope control, standard PWM LSI
integrated control circuitry could be employed if desired. By the use of a VMOS power device instead
of a bipolar transistor for the CUK converter primary
switch, the interface electronics are greatly simplified
and overall converter efficiency improved due to the
micropower gate drive required for the VMOS field
effect ~ansistor.
8.3 Group Three: Reversal Protection
In this test, output voltage polarity reversal of the
V output 'during tum-on was investigated. This
voltage polarity reversal occurs only at the + 15 V
output during tum-on due to magnetic coupling between Lp and L0 2. The photograph of Figure 13(a)
shows that negative going spike went as high as - 20
volts and recovered in about 60 Ihsec after tum-on.
Figure 13(b) demonstrates that a diode at the + 15 V
+ 15
Obviously, the converter could be designed to operate
at lower switching frequencies with corresponding
increases in energy transfer capacitor values and larger
magnetic cores for the isolation transformer and inductor assemblies. In fact, the need for no output
6-23
filter capacitor on the regulated + 5 V line should
simplify the frequency response of the converter as
was pointed out in [3]; therefore, an extensively high
switching frequency is not required in comparison to
the filter comer frequency for reduction of voltage
ripple on the regulated output.
[2]
Slobodan euk and R.D. Middlebrook, "Coupled-Inductor and Other Extensions of a New
Optimum Topology Switching DC-to-DC Converter," IEEE Industry Applications Society
Annual Meeting, 1977 Record, pp. 1110-1126
(IEEE Publication 77CH1246-8-IA).
With regard to frequency response, no mention was
made in this paper of breadboard converter performance in this area as these experiments are currently
in process. Modeling of the converter for response
predictions can follow published techniques [1, 2, 5]
with modifications made to include nonregulated output loads and their filter networks reflected to the
controlled output and the magnetic coupling of input!
output inductances. Investigations are being performed
in the areas of bandwidth improvement with increased
stability margins by the use of multiple feedback
control techniques such as those described in [6]
and [8].
[3]
Slobodan euk, "Switching DC-to-DC Converter
with Zero Input or Output Current Ripple,"
IEEE Industry Applications Society Annual
Meeting, 1978 Record, pp. 1131-1146, (IEEE
Publication 73CH1346-61A).
[4]
R.D. Middlebrook and Slobodan Cuk, "Isolation
and Multiple Output Extensions of a New Optimum Topology Switching DC-to-DC Converter," IEEE Power Electronics Specialists
Conference, 1978 Record, pp. 256-264 (IEEE
Publication 78CH1337-5 AES).
[5]
R.D. Middlebrook, "Modelling and Design of
the CUK Converter," Proc. Sixth National
Solid-State Power Conversion Conference
(Powercon 6), pp. O3.1-G3.13, May 1979.
[6]
Shi-Ping Hsu, Art Brown, Loman Rensink, and
R.D. Middlebrook, "Modeling and Analysis of
Switching DC-to-DC Converters in ConstantFrequency Current-Programmed Mode," IEEE
Power Electronics Specialists Conference, 1979
Record, paper 4A.6.
[7]
Slobodan euk and Robert W. Erickson, "A
Conceptually New High-Frequency SwitchedMode Amplifier Technique Eliminates Current
Ripple," Proc. Fifth National Solid-State Power
Conversion Conference (Powercon 5), pp.
03.1-03.22, May 1978.
[8]
Loman Rensink, Art Brown, Shi-Ping Hsu, and
Slobodan euk, "Design of a Kilowatt Off-Line
Switcher Using a CUK Converter," Proc. Sixth
National Solid-State Power Conversion Conference (Powercon 6), pp. H3.1-H3.26, May 1979.
[9]
R.D. Middlebrook, Slobodan euk, and W.
Behen, "A New Battery Charger/Discharger
Converter," IEEE Power Electronics Specialists Conference, 1978 Record, pp. 251-255
(IEEE Publication 78CH1337-5 AES).
The success of external control of ripple current magnitudes reported in Section 7 and 8 has implications
of corresponding circuit improvements to the basic
single output CUK converters of Figure 1 that could
simultaneously eliminate both input and output ripple
currents. In addition, it should be possible for converter design with more than three outputs to utilize
the techniques described in Sections 6 and 7 to attain
zero ripple currents on the input and more than one
output by proper choice of output inductances to be
magnetically coupled.
Acknowledgements
The authors would like to thank Morris Rauch for
his diligent work on the evaluation of the experimental
breadboard and Karl Dakteris for the many hours
spent in prototype magnetic designs. Thanks are due
also to Dusan Jocic for excellent magnetic prototype
construction work.
Most of ay, special personal gratitude is sent to Drs.
Slobodan Cuk and R.D. Middlebrook for their helpful
advice and inspiration in putting this work together.
References
[1]
Slobodan euk and R.D. Middlebrook, "A New
Optimum Topology Switching DC-to-DC Converter, " IEEE Power Electronics Specialists
Conference, 1977 Record, pp. 160-179 (IEEE
Publication 77CH1213-8 AES).
[10] R.D. Severns, "High-Frequency Switching
Regulator Techniques," IEEE Power Electronics
Specialists Conference, 1978 Record, pp.
290-298, (IEEE Publication 78CH1346-61A).
[11] A.1. Pressman, "Switching and Linear Power
Supply, Power Converter Design," Hayden
Book Co., Inc., 1977, pp. 289-315.
6·24
6.1.3 The Generalized Use of Integrated
Magnetics and Zero-Ripple Techniques
in Switch Mode Power Converters
©
1984 IEEE. Reprinted, with pennission, from IEEE Power Electronics Specialists Conference, PESC '84,
Gaithersburg, MD, June 18-21, 1984.
Abstract
There is little disagreement among power converter
designers that the magnetic parts of any switch mode
design are the major contributors to supply cost,
weight, and size. For these reasons, it is not uncommon for designers to select converter approaches that
have a minimum magnetic content, even though, in
many cases, one with more magnetic components
would be more suitable from an overall circuit performance standpoint.
Methods are presented that allow the discrete transformers and inductors of switch mode power converters to be unified in single magnetic structures. It is
demonstrated that unified magnetics and zero ripple
operation are general phenomena applicable to all
types of switch mode power converters.
Introduction
Seemingly, a day does not pass now without an
announcement of a new advance in microcircuit technology, bringing us closer to the age of ultra miniaturization of electronic products. For those of us who
must design and develop power processing-systems it
is a time of frustration and reflection, for as product
sizes shrink, so must their power conditioners.
Realizing that moving to higher switching frequencies
to reduce magnetic size has practical boundaries,
designers are now turning to another avenue of
investigation - magnetic integration. This rather
innovative-sounding, but accurate, term is used to describe magnetic design techniques whereby various
inductive and transformer elements of a power converter can be combined on a single core structure.
Twenty-five years ago, when the sizes of electronic
systems were measured in terms of room dimensions,
their power supplies could afford to be large. Today,
with electronic products, such as calculators smaller
than the dial of a watch, the power supply subsystems
are often as large or larger than the electronics.
Another example is the personal computer, where
power supply areas often occupy up to 50% of enclosure volume, with a cost approaching 45% of product
price.
If one accepts the above definition of magnetic integration as applied to switch mode DC-to-DC power
converters, then two converter topologies can be read- _
ily identified as integrated magnetic circuits, namely,
the buck-boost-derived flyback converter and a special variation of the "boost-buck-derived" CUK converter[2] .
Both of these circuits, together with their discrete
magnetic counterparts, are shown in the lower portions of Figure 1. In the case of the flyback converter,
the required inductance for energy storage is simply
"built-into" the isolation transformer by proper
choice magnetization inductance. Thus, the transformer of the flyback converter serves two important
purposes - as an isolation element for input and output grounds and as a means of primary energy storage
In a concerted effort to reduce power supply sizes,
recent years have seen switch mode conversion and
processing designs pushed higher and higher in
operating frequency. Theoretically, at least, converters with high switching rates imply that their circuits
will have smaller magnetic components. There are
however, very practical limits to the size reduction
obtainable from high frequency operation [1].
6·25
I
for supplying load power needs. In the case of the
integrated-magnetic version of the transformerisolated CUK converter, its single magnetic houses all
inductive functions of the converter and, by selecting
proper amounts of mutual inductance that exist between windings, input and output ripple current
magnitudes can be controlled, and even be reduced to
zero in special instances [3] .
magnetic buck and boost converters, it is worthwhile
to digress briefly and review some not-so-familiar
electromagnetic fundamentals and magnetic circuit
modeling methods using electric circuit analogs. For
those readers interested in a more comprehensive review than the one to follow, reference [4] is highly
recommended.
Recall that the similarity between Kirchoff's voltage
and current laws for linear electric circuits and Ampere's circuital laws related to magnetomotive force
and flux continuity in linear magnetic circuits permit
the use of electric circuit analogs for analysis purposes. Such analysis makes voltage, V, analogous to
magnetic potential, f; current, i, analogous to magnetic flux, cp; electrical resistance, R, analogous to
magnetic reluctance, R. Furthermore, because the
electric circuits derived by the use of these analogs are
linear, they can be manipulated into even more useful
forms by established duality relationships [5]. Transformation by duality then produces electric circuit
models that relate magnetic reluctances, R's, to inductances, L's; flux linkages in windings, A'S, to
voltages, v's; and flux levels in magnetic paths, cp's,
to currents, i's.
Both the buck-boost converter and its dual circuit, the
CUK converter, can be evolved from the appropriate
cascade arrangements of basic buck and boost circuits,
as demonstrated in [2]. It is interesting to note that
integ!:ated-magnetic versions of isolated buck-boost
and CUK converters are perfectly feasible and welldocumented circuit possibilities, but little is publicly
known about similar integration methods for
transformer-isolated versions of their parent circuits,
as visually emphasized by the vacant sections of
Figure 1.
In this paper, we will explore design techniques for the
integration of the transformer and inductive components of buck and boost-derived converter topologies.
To set the stage for this examination, a brief introductory section on basic electromagnetic modeling and
analysis is included. Discussions of magnetic integration of the inductive components of a forward converter are then presented, with extensions for reducing
the output ripple current significantly by external
means. By duality integrated-magnetic boost converter circuits are then evolved and extensions illustrated
for input ripple current reductions.
I
The rate of change of flux with time within a coil of
wire (with or without a ferromagnetic core) of N turns
can be related to A, as
(1)
by Lenz's and Faraday's Law. As shown in (1), a
similar relationship can be stated in terms of the inductance (L) of the coil and the instantaneous current
(i) thru it. In the case of a single coil of wire, cp is the
measure of flux linkage within the core produced by
self-induction. In an instance where there are multiple
coils with common magnetic paths, the total flux linkage of one coil would be the sum of that produced by
self-induction plus those produced by mutual interaction, or mutual induction, with the others.
Integrated-magnetic "push-pull" DC-to-DC converter
structures are then examined, including one based on a
special variation of the Weinberg circuit. Results of
laboratory tests of a representative integrated magnetic
converter are then shown, followed by discussions of
small-signal averaged modeling for stability and control analysis.
Tools for Magnetic Circuit Modeling
Integrated magnetics for converters brings to mind a
picture of magnetic structures that are highly complex
and unwieldy from design, analysis, and construction
viewpoints. To some degree, this concern is understandable, for most engineers today are only familiar
with magnetic design methods that address inductors
and transformers as individual converter components.
Just the thought of having to deal with a magnetic
assembly with more than one major flux path is often a
deterrent to an engineer to 'attempt such a design.
The magneto motive force, F, of an excited coil of wire
(with or without a ferromagnetic core) is defined as the
product of the instantaneous current through it, i, and
the number of turns, N, of the coil. In equation form,
F can be directly related to magnetizing force, H, and
its magnetic path length, 1, as
F
=
Ni
=
HI
(2)
The self-inductance, L, of a single coil of wire of N
turns, can be equated to the rate of change of flux with
current from (1) as
The attitude of designers in this regard is now changing. However, for many converter designers, the tools
to properly design and model integrated-magnetic
components have been unused over the years. Thus,
before beginning the exploration of integrated-
L = N dcp
di
6-26
(3)
INTEGRATED MAGNETICS
DISCRETE MAGNETICS
(a)
(b)
VolV~=nD
1'n:"
....>--<.--0+
?•
+
Do--t-l!4,
FORWARD CONVERTER
(BUCK-DERIVED)
(c)
(d)
?•
REVERSE CONVERTER
(BOOST-DERIVED)
(f)
(e)
V~
D
L-_~-~-o+
FLYBACK CONVERTER
(BUCK-BOOST-DERIVED)
FLYBACK CONVERTER
(BUCK-BOOST-DERIVED)
(g)
(h)
VolV~ = nD/D'
c+
V~
+
c
+
L-------~~---J1:nL---_'--------~--_'--~+
L__ _ _ _ _ _~~---J
ISOLATED CUK CONVERTER
(CUK-DERIVEDI
"r
1:n'---......_------...--........--<1 +
ISOLATED CUK CONVERTER
(CUK-DERIVED)
Discrete and Integrated-Magnetic Converter Possibilities-What About the Forward and Reverse Designs?
Figure 1
6-27
I
-
(,I
Assuming that a linear relationship between flux level
and magnetizing force is always maintained, R can be
defined as the ratio of a change in F produced by a
change in cp. From (2) and (3), reluctance can be
expressed in terms of inductance as
RIGHT-HAND RULE
FLUX DIRECTION
(4)
(bl
MEAN FLUX PATHS
Reluctance can also be expressed in terms of related
magnetic path length, cross-sectional area of the magnetic material, AC, and the permeability, /1-, of the
path in question. If AC is uniform in value throughout
the path, then
R = /l-Ac =
1
P
(5)
• - RELATIVE TO WINDING A-8
A- RELATIVE TO WINDING E-F
. - RELATIVE TO WINDING C-O
Determining Flux Directions (a). and Dotting of Windings (b)
Figure 2
where P is defined as material permeance, the reciprocal element of reluctance.
visually express voltage polarities. However, this is
easily done by the use of the right-hand rule and the
three dotting rules given above. An example of multiple "dotting" is shown in Figure 2(b), where three
sets of "dots" are needed to express winding
polarities relative to each one of the three windings of
the magnetic.
In magnetic circuits, arrows are used to indicate the
assumed directions of winding currents, rather than
polarity marks for magnetomotive forces (often shortened to "mmfs"). The most popular method for determining flux direction is the "right-hand rule"
illustrated in Figure 2(a). With the right hand positioned as shown, the flux direction will be indicated by
the direction of the curvature of the fingers. Note that
the thumb must be pointed in the assumed direction of
winding current when making this determination.
A Magnetic Circuit Modeling Example
All of the fundamental definitions related to magnetic
circuits given above, when combined with the techniques of electric circuit modeling and duality, produce a powerful set of analysis tools for quick and
accurate examination of any magnetic circuit
arrangement, no matter how complex it may be. The
electric circuit models that result are just that - circuit models - and do not depend on abstract mathematical relationships (such as mutual inductance expressions) for performance evaluations.
In representing multi-winding magnetic circuits in
electric circuit diagrams, it is customary to use dot
notation to convey the voltage polarity relationships of
each winding relative to the others, as shown in Figure
2(b). Three basic rules are followed in the "dotting"
of windings:
RULE 1. Voltages induced in any two windings due
to changes in mutual flux will have the same polarity
at "dot-marked" terminals.
As an illustration of the power of these modeling
techniques, consider now an example. For this exercise, we will derive an electric circuit model for a
two-winding transformer with parasitic leakage inductances, as shown in Figure 3(a). Here, an ungapped toroidal core provides the major magnetic path
between windings (cpm), and each winding has a leakage path (cp 1, CP2) for flux that is not contained in the
material path. We will assume for this exercise that the
core's cross-sectional area is uniform throughout its
body, and that a mean path length can be used to
define the reluctance of the core.
RULE 2. If positive currents flow into the "dotmarked" terminals of related windings, then the
mmfs produced in each winding will have additive
polarity.
RULE 3. If any related winding is open-circuited, and
if the currents flowing into the "dot-marked" terminals have a positive rate of change, then the voltage
induced in the open winding will be positive at its
"dot-marked" terminal.
When a magnetic circuit arrangement has more than
two windings and contains more than one major flux
path, then multiple' 'dotting" windings is necessary to
If we designate Np as the primary exciting winding,
we can first determine the voltage polarity of the
secondary winding using the right-hand rule. With a
6·28
primary current direction and the winding dotted as
shown in Figure 3(a), fluxes rp 1, and rp2 must have
positive directions as indicated in this figure. The dot
for the secondary therefore must be at the top of this
winding to produce a flux in the same direction as rp m.
turns of the winding we select as reference for the final
electric circuit representation. This is accomplished in
Figure 3(d). The major purpose of this scaling step is
to place all circuit permeances in a form that can be
directly related to inductance. Note that the scaled
model of Figure 3(d) also permits easy conversion of
flux linkages to primary and secondary voltage values.
With winding polarities established, we note that the
secondary voltage, Vo , must have a polarity as
shown, for the given polarity of exciting voltage, V s.
Because the resulting secondary current is "out of the
dot", we also note from our earlier transformer rules
that this current, io , will produce an mmf that opposes
that of the primary.
Finally, we convert the scaled permeance network to
one involving voltages and inductances. This conversion is shown in Figure 4(a). Note that we have added
an ideal transformer to this network to properly scale
terminal secondary voltage and current values. Using
impedance translation methods, we can also "move"
all or some primary inductances to the secondary side
of the final electric circuit model if desired. Two
versions of such impedance movements are shown in
Figures 4(b) and 4(c).
Next, we draw the equivalent magnetic circuit diagram for the arrangement of Figure 3(a). This is done
in Figure 3(b), using the analogous relationships defined earlier. Note that rpm and rp 1 have the same
polarity in the primary winding, since they are caused
by is. However, in the secondary winding, rpm andrp2
have opposite polarity, since one flux (rpm> produces
io while the other (rp2) is a consequence of io .
In just four easy steps, we have been able to develop a
realistic electric circuit model for a somewhat complex
magnetic circuit arrangement. Note that the final models of Figure 4 do not involve mutual inductances and
are in a form that a designer can easily relate physical
properties of the corresponding magnetic system of
Figure 3(a) to familiar electric circuit quantities.
From the magnetic circuit of Figure 3(b), we can now
develop a dual permeance network for it, resulting in
the equivalent circuit of Figure 3(c). Our next step is
to scale the network of Figure 3(c) by the number of
Looking at Figure 4(c), we see that inductances Ll
and L2 represent those produced by flux leakages, and
are often called leakage inductances. Inductance Lc in
the networks of Figure 4 represents that produced by
the primary turns (Np ) wound on the ferromagnetic
core of the transformer, and is not the mutual inductance (M) shared by primary and secondary windings.
Recall that mutual inductance is a mathematical measure of the degree of coupling between two windings of
a magnetic. In this case, we can easily find the value
1,1
'.
1,1
L,
[~r2
+
=0
. " m'O
_,.~
fo_
•
•
+
N,
LOAD
Ideal T ' )
>]
Ibl
'.
L2
'0
t
LOAD
'0
1-
10'
IdedlT)
'0
LOAD
Developing Reluctance and Permeance Circuit Models for a
Two-Winding Transformer with Leakage Fluxes
Figure 3
Electric-Circuit Equivalent Networks for the Two-Winding
Transformer of Figure 3(a)
Figure 4
6·29
-
D'
of M by writing the two nodal equations that relate
input and output currents of the circuit models, and
then isolate the common inductance terms within
them. If this is done, then M is found to be
NS
M= Np
X
Np
(6)
I.e
"Coupled-Inductor" Output Filter Network of a QuaslSquareWave Converter
Figure 5
Achieving Zero-Ripple Currents
Although the techniques of achieving zero-ripple currents in the windings of selected magnetic designs
have recently been thoroughly explored [3], there is
much historical evidence that the use of magnetic
means to lower ripple currents in converter outputs
is not a new discovery. For example, O.B. Crouse
disclosed in his 1933 United States patent [6] a magnetic method for lowering the ripple voltage of an
L-C filter for use in radio receiver power supplies.
This method, upon close scrutiny, appears to be identical to one reported some years later by S. Feng [7]
for reducing the' size of filter capacitors used across
the outputs of power converters. Several other examples [8,9,10] also exist.
age voltage across it equal to V0 in the presence of the
ripple current. Ideally, the mutual inductance shared
by the inductor windings is then chosen to completely
remove the ripple current from the load (R) and to
steer it to the inner winding (N3). For this reason, no
output fIlter capacitor is shown across R in Figure 5.
However, in practical designs, some capacitance is
usually added across R for decoupling of noise and for
additional energy storage for instantaneous load demands.
To understand how the inductor arrangement of Figure
5 can magnetically reduce ripple currents, we can use
the electric circuit model developed in Section 2.1.
Looking at Figure 5, we can assume that the voltages
impressed across the two windings of the inductor are
proportional in amplitUde and equal in dynamic
periods, such as those shown in Figure 6(a). For our
purposes, we will therefore assume that the "primary" voltage is of a value equal to Vs, with the
"secondary" excited by another proportional voltage,
avs. With these assumptions, we can impose these
voltages across the terminals of the electric circuit
model from Figure 4(c), and then analyze what model
values must be present to make the primary ripple
current (is) vanish.
In both cases, the method entailed the addition of
another winding to the filter inductor, whose mutual
inductance relationship to the original inductor winding was selected by design to "steer" the ripple current in the main winding to the added "secondary"
winding and, therefore, away from the output of the
filter. The resulting circuit arrangement is shown
schematically in Figure 5, for application as the secondary fIlter of a forward converter. Note that a fIlter
capacitor (C) is necessary for DC isolation of the
"inter" winding and is selected to maintain an averw
r-------------------,
'0
•
"
LZ
NS
1
I'
"'.
1
1
1
1d
L -
-
-
- - -
- - -
- - - - - -
lL:....~,;sf;mer
Modal
of Fig 41cl
(b)
o~U RE.
U
t,
--
",
, >
0
The Coupled-Inductor of Figure 5 Modeled Using the
Transformer Equivalent Circuit (a), Driven by Proportional Voltages (b)
(Note: a = 1 for Figure 5)
Figure 6
6-30
Comparing the constraints imposed on the values of
L2 and Ll by (9) and (10), respectively, one finds that
it is not possible to achieve zero values of is and io
simultaneously. This point is also made by the current
waveforms for is and io illustrated in Figure 8. Note
also from the equations accompanying Figure 8 that
the effective inductance seen at the "primary" terminals of the two-winding magnetic will be equal to its
open-circuit value (Le. "secondary" side open) when
L 1 is selected to reduce io to zero. Conversely, the
effective inductance seen at the "secondary" terminals will be equal to its open-circuit magnitude when
L2 is set to reduce is to zero.
Ib)
".
Circuit Conditions in the Model of Figure 6 for is
Figure 7
+0
In actual practice, it is difficult to design and manufacture magnetic assemblies with consistent and
specific values of parasitic leakag~ .inductance~ to
achieve the zero-ripple current conditIOns as defmed
by (9) or (10). One viable solution to this producibility
problem is to tightly wind primary and secondary turns
to reduce Ll and L2 to essentially zero, and then to
insert a small external "trimming" inductor [8] in
series with the input or output to emulate the required
inductance needed for Ll or L2. This trimming
method is shown in Figure 9, along with corresponding values of Lext necessary for zero-ripple current
conditions.
This analysis task is easily performed, as illustrated in
Figure 7. Writing the circuit equations for Figure 7(b)
gives
(7)
diO )
Lc ( dt
(8)
[aNP
]
NS)2
( Np
Lc NS - 1
(9)
= (NS)2
( NS)
Np Vs
Np
L2
=
Because of recent emphasis on application of zeroripple-current filters in CUK converter v~ation~ [9],
[10], it is now generally believed that thIS partIcular
family of converter topologies is the only one that can
benefit by their use. However, as we have just seen,
this is not the case. In boost converters, these same
principles of ripple current reduction can be applied to
their input inductor arrangements, one example of
which is shown in Figure 10. And, as we will soon
see this method of ripple current reduction can also be
ad;antageously used in both buck and boost derived
integrated-magnetic converter systems.
Therefore, if we select the leakage inductance of the
secondary winding to meet the needs of (9), then the
primary ripple current will be reduced to zero! Although not shown here, a similar analysis of the circuit
model of Figure 6(b) for zero-ripple secondary current
can be made. The value of primary leakage inductance
for io = 0 is then found to be
(10)
L2=(~:)\ [~-'l
L, =L, [~-'l
'.
....._-'"
-".f..= 0
Input and Output Cur-rent Waveforms of Figure 6(a)
When Selected Values of L1 or L2 are Present (Note: a 1 for Figure 6(a»
=
Figure 8
6-31
r _________
(.)
Laxt
al
1-T:n,::":r :0:1 ~r ~ ~"
L1 "'0
Ie
l2""'0
10
"",0
",
"
i
~ldealT
bl
[~_
Laxt .....
L-
- -
- - - -
- - - - - -
-
,d
- - - -
r ___________________ -,
- J
L (~)2 L, [;,.- -,]
oxt "
(b)
al
IS
=
l1"'0
Ie
L2"""0
0
Laxt
'0
•
",
N,
"
I"
L -
-
-
-
-
-
-
-
-
-
-
-
-
-
,-=-~r:Sf:me~ Model for h ... 1
Using Small External Inductances to Trim Ripple Current
Magnitudes In the Circuit of Figure 6(a)
Figure 9
r'· ~
,
I
I
OT....... OT----IoI
o
'
,
\.-T--l
o
"
~
\.-T
----..J
,
10
~,-,:="Q1
t
01
C2
L ...... tlghtly-coupled wmdmgs
L oxt = L, [
~~
-,]
Lc- magnetizing Inductance of L referenced to N2
Using a Coupled-Inductor Arrangement In a Basic Boost
Converter to Reduce Input Ripple Current Magnitude
Figure 10
this is accomplished, the voltage across L 1 is zero
and so is the ripple!
The preceding explanation for zero ripple, while correct, may leave one wondering what is really going
on and if the results are truly general or just a bit of
mathematical serendipity.
Remember that, in this particular example, a = 1
because both windings are excited by the same source.
This is not the case in general as the next example
will show. For the special case where a = 1, then
NpfNS > 1 (see Equation 9) if L2 is to have a
positive value.
A very simple intuitive explanation can be provided
by looking again at Figure 6. If the current through
Ll is to be zero, then the voltage across Ll must
also be zero. To the left of Ll is the voltage source
VS. To the right of Ll is a voltage divider formed
by LC and the transformed value for L2: L2' = (NSf
Np)2.L2' as well as the ideal transformer. The game
being played is to adjust the values for Le, L2' , and
NpfNS so that the v~ltage source aVS is transformed
to a value equal to Vs at the right side of L 1. When
We can now extend our zero-ripple example to a
multiple winding inductor in a multiple output quasisquare wave converter in which more than one output
has zero ripple. The use of a single mUltiple winding
inductor reduces the total weight and volume of the
6-32
01
R3
DOTS
Vo3
• - RELATIVE TO N,
• -
RELATIVE TO N2
... -
RELATIVE TO N
Coupled-Inductor Arrangement lor a Quasi-Square Wave
DC-DC Power Converter
Figure 11
inductive portion of the system, and also lowers cost
and enhances the dynamic cross-regulation properties
between the outputs of the converter. One of several
possible versions of a converter with "coupled" output inductances is given in Figure 11.
ideal conditions exist (Le., diode drops may be neglected and perfect coupling of transformer windings
has been achieved). With this assumption, we define
VNJ = avs, a = NSJINp
Those familiar with the design aspects of coupledinductors for converters such as this one will note
that the winding arrangements, as well as the core
structure itself, are somewhat unconventional. Here,
a three-legged core is employed, with two inductor
windings placed on the outer portion of the structure,
and the third winding contained on the "center" leg.
Air gaps are placed in each of the two outer legs.
As illustrated by the inductor current waveforms of
Figure 11, it is postulated that the ripple currents in
two of the three inductor windings can be made to
disappear!
VN2 = bvS, b = NS31Np
VN = CVS,
C
(lOA)
= NS21Np
where Vs is the dynamic waveform produced across
either of the primary windings of T1 by the switching
actions of Ql and Q2. By our earlier assumption,
we know that this converter is being operated at a
constant switching frequency, and that Ql and Q2
are being alternately turned on for a portion, D, of
a switching cycle.
The first three steps in evolving the model are shown
in Figure 12. In part (a) flux directions are found in
each of the three magnetic paths, using the righthand rule method and assuming that the center leg
is the controlling and dominant source of mmf.
At first glance, it may seem that the analysis effort
necessary to understand how the magnetic arrangement can achieve significant current ripple reductions
in selected windings is overwhelming. However, if
we use the modeling techniques just demonstrated,
the analysis becomes straightforward and its results
illuminating.
Three other minor flux paths are shown in Figure
12 (a). These represent "leakage" fluxes, Le., not
contained within the core material and are to be expected in any practical magnetic winding arrangement.
Therefore, we have added them to our magnetic system
to ensure that our final electrical model will include
leakage inductance effects associated with each of
the three windings.
Note that the dynamic voltages impressed across each
of the three windings of the magnetic are proportional,
always in phase with one another, and have the same
frequency. Voltage proportions are ideally set by the
ratios of the transformer secondary turns relative to
primary turns, and for this analysis, we will assume
6-33
-
I
{.I
• MATERIAL
PERMEABILITV=~m
• UNIFORM CROSS-SECTIONAL AREA (AJ THROUGHOUT
{bl
UNIFORM
.... ,.. .1 II,...CROSS-:SECTIQN
~,
~·II
ALL DIMENSIONS
IN METERS
{,I
.;
1-'
I:.:.-I-..:j
-1"'~
:l
: -1,,21--
I
I
I
I
I
I
I
I
I
I
I
I
I
L - - - - - - - r - I - - - - ____ .J
+
r
t
!
-,fJlEANPATH
(b-c + 28 + 2f + ig,l
LENGTHS
R
.'.
~
R.,=3L
.
JoIoAc
'fR'2
RCT
R'2"~
JoIoAc
RC'"
RC2
• 1'0.411'010-7
:::~
~~
·""....c
(b-c + 28'" 2h + 2g2)
(:I....C
."".,Ac
Po" PERMEABILITY OF FREE SPACE
Reducing the Inductor Assembly of Figure 11 to a
Reluctance Network
Figure 12
{.I
Next, in part (b) a two·dimensional view of the struc·
ture of (a) is given in order to be able to relate core
dimensions and air gap lengths to values for reluc·
tance. Because winding details are not important for
these determinations, they are omitted for clarity.
In part (c) an "electrical" network of reluctances is
constructed from the magnetic path details of part
(b). Also shown in part (c) are the equations relating
reluctance values to core dimensions, permeabilities,
and cross-sectional area.
(bl
We now add to our reluctance network of Figure
12(c) the mmf's produced by the three winding
currents as well as surrounding leakage reluctances.
This is done in Figure 13(a). The "polarity" of each
of the winding mmf's in our model is established
using Rule 2 in conjunction with the assumed flux
directions in each leg of the core in Figure 12(a).
Note that, because each winding current is directed
into its respective "dot," the mmf polarities are such
that they will produce fluxes that are in the same
direction as those assumed in Figure 12(a) in their
respective legs.
AS1' R92> AC1 ' RC2
RN » RCT
(a) Adding mmf's to the Network of Figure 12(c),
(b) Slmpllficalion of (a)
Figure 13
For the most part, air gap and leakage reluctances
will always be much greater than those presented by
magnetic material paths, since material permeabilities
are much greater than that of air. Therefore, the
magnetic model of Figure 13(a) may be reduced in
complexity to that shown in Figure 13(b) with little
loss in accuracy.
Although the magnetic circuit model of Figure 13(a)
is perfectly suitable for translation into an electrical
equivalent network, some simplification may be made.
6·34
'e
iii
(.(
+
0,
PN
-
+ .. 2
N,2 p,
+
~;\C
"
i\, =N, I¢A +0,1
N'e
0.
P"
N,ll
(b(
P,
- oN +
°A
~
N1
G=1
,e
=OA + "B +"N
P,
+
Ie
N,2pg1
LN
+
C!'S
-
-
P"
+
,
IDE~ eN
..l.=OB+02
N,
N2 12
N?P N
N,2P2
N,2 pg2
N,
N2
},2
Final Electric Circuit Model for the
MagnetiC Arrangement of Figure 12(a)
Figure 15
abilities of various parts of magnetic paths. The final
model gives excellent visibility into the effects of
winding voltages and currents; also, how we can
reduce ripple currents in selected windings to zero!
Referring to the converter circuit of Figure 11, we remind ourselves that we are interested in the model conditions that must exist if the dynamic currents, il and
i2, are zero. Looking now at our final model in Figure
IS, we simply set il and i2 to zero, and then establish
what circuit conditions prevail as a result of their
absence. These conditions are shown in Figure 16.
Developing an Equivalent Electrical Circuit Model for the
Reluctance Network of Figure 13(b)
Figure 14
The next step in our model development is establishing a dual network for the reluctance and mmf system
of Figure 13(b). Using the duality procedures outlined
in [5], the permeance circuit of Figure 14(a) is easily
obtained. A choice must now be made as to which
of the three windings we want the final elements of
the electrical model to be "referenced." We arbitrarily chose N 1, keeping in mind that we can always
alter our final electrical model later to refer to N2 or
to N if we so desire.
We now scale all of the elements of the model of
Figure 14(a) by Nl. This process gives us a model
relating flux linkages to winding current values, as
illustrated in Figure 14(b). We then convert the
permeance model of Figure 14(b) to one involving
inductances and winding voltages. This is done in
Figure 14(c), with ideal transformers added to account
for the model reference to N 1.
Circuit Conditions In the Model of Figure 15 for Ripple Currents
in N1, N2 Equal to Zero
Figure 16
With il and i2 at zero value, the voltages impressed
across the gap inductances, Lgl, and Lg2 will be
simple reflections of associated winding voltages, avs
and bvs, respectively. Also, the dynamic voltage ~
drops across the leakage inductances, Ll and L2, _ _
must be zero, because their respective branch currents
'
are zero. At this point, we can make an important
observation - the values of the leakage inductances,
Ll and L2, are not important in establishing a zeroripple current condition in either winding N 1 or in
winding N2!
Finally, using conventional impedance transformations, we move the leakage inductances associated
with N and N2 to the "primary" sides of their respective ideal transformers. The result of this last
model manipulation is shown in Figure 15.
In a few basic steps, we have been able to reduce a
rather complex magnetic system to an equivalent
electrical circuit. In reviewing our steps, we find that
we can easily relate all of the elements of the final
electrical model to the parameters of the magnetic
assembly, including physical dimensions and perme-
In order to discover what remaining parameters of
the magnetic arrangement are important in making i 1
and i2 zero, we now determine the describing voltage
6-35
and current relationships for the circuit conditions
illustrated in Figure 16
avs =
(~) Lg1 (~;)
Nl Vs = (N)
b N2
N 1 Lg2 (dic)
dt
diC\
cvS - vx= LN ( dtJ
Note that (15) and (16) are not interdependent. Therefore, for fixed values of a, b, c, N, Nt. N2, and LN,
air gap inductance levels may be selected by design
so that requirements of (15) and (16) are satisfied
simultaneously. For example, assume that we have a
converter application as shown in Figure 3 where
a = 1, b = 2, c = 3, and Nl = N2. Placing these
values into (15) and (16), one finds that gap inductance,
Lg2, must be twice that of Lg l, and that the choice
of N, N 1, and LN will determine the actual gap inductance levels required to produce zero-ripple current
conditions in windings N 1 and N2.
(11)
(12)
(13)
Developing an Integrated-Magnetic Forward
Converter
It is interesting to note that, contrary to popular belief
[3], integrated-magnetic concepts for power converters
are not innovations of research performed over the
past seven years. Like the magnetic methods of ripple
current reduction described in the last section of this
paper, the concept of integration of the magnetic
functions of power processing circuits has historical
roots in much earlier periods. Perhaps the best documented evidence of this fact is an obscure (and
mistitled) United States patent by Cielo and Hoffman
of the mM Corporation [14] in 1971, which discloses
possible circuit methods for integration of the transformer and inductor of so-called "push-pull" DC-toDC converters. Later in this paper, we will examine
the methods of this interesting patent by extending
the integrated-magnetic forward converter concepts
to include "push-pull" switch arrangements.
(14)
Combining (11) and (12) with (13) and (14) to eliminate the common diddt term gives
LN = :. (Nl) _ !:(N1 )-1
Lgl
a N
a N2
(15)
LN = ~(N2) _ ~(N2)_1
Lg2
b N
b Nl
(16)
With a brief exposure to magnetic circuit modeling
methods and a review of inductive ripple current reduction techniques behind us, let us proceed to systematically construct a single magnetic assembly
wherein the transformer and the inductor of a forward
converter can be contained. No rigorous synthesis
procedure will be followed here, but rather a path of
deduction and intuition based on the general operation
of this converter and knowledge of the flux change
relationships that must exist in its inductive
components.
Also, using (13) and (14), we can find the effective
inductance, Le , that is seen at the input to winding
N when il and i2 are zero. Performing this operation
gives
Looking at (15) and (16), we now realize that if their
relationships are satisfied, the ripple currents in both
N 1 and N2 will be entirely eliminated! These equations also predict that the zero-ripple current conditions are not dependent on external converter
operational conditions, such as line, load, or switch
duty cycle magnitudes! The phenomenon will be dependent, however, on winding turns, proportional
constants set by the transformer secondary-to-primary
turns ratios, leakage inductance of the center leg of
the magnetic assembly and the values of the two air
gap inductances.
The first step in our integration process will be to
redraw the forward converter circuit of Figure 1 in
a manner so as to emphasize both the electrical and
magnetic aspects of the converter. Then, assuming
that the converter is operllting in the continuous mode
of inductor energy storage, the equations describing
circuit conditions for each of the two switching states
are found, and then placed in a format relating corresponding flux changes in the transformer and inductor. Because we are interested at this time in only
major flux relationships, the describing circuit equa-
6-36
tions can be ideal, thus ignoring all parasltlcs including switch and diode drops, switching losses,
etc. The elimination of parasitics is not a requirement,
but it is a convenience for this discussion.
During D'T,
The redrawn forward converter is illustrated in Figure
17. The core of the transfonner is shown ungapped,
while the inductor core is shown with an air gap, as
is the usual case for a magnetic that must withstand
DC bias. Transfonner windings are drawn and dotted
to emphasize an assumed counter-clockwise direction
of flux in its core and to produce a positive voltage
(V 1> across the secondary winding (N s) when the
primary switch (Ql) is ON. Note also that we have
included a "core-reset" winding (Np2) on the transfonner for energy removal during intervals when Q I
is OFF. A core material with low residual flux is
presumed. For the inductor with its winding position
as shown, flux direction will be clockwise.
~
¢L= Vo
NL
(21)
¢T = (NL)CPL
NS
+
Vo
NS
•
cpT
(NL).
NS CPL
=
•
+ CPo
(19)
,..-------,
¢L
I
-1',1-
I
NL
A
I
I
V
I ~-P-f>-P-'f--'
L.. _
__
> ___
(23)
where CPo = Vo INs. Turning now to the first righthand tenn of (22) and (23), we note that its contribution to CPT is dependent on a fraction, NL INS, of
the flux change rate in the inductor of the converter.
Since our ultimate goal is to make the inductor an
integral part of the same magnetic assembly that also
contains the transfonner windings, it is logical to
assume that NL should be made equal to NS, so as
to contain all inductor flux in a single magnetic path.
I
v,
(22)
Note that the last tenn of (22) is of a fonn that could
be considered as defining a flux change in a magnetic
medium that is dependent on the value of the output
voltage, V0, of the converter and the number of
secondary turns on the transfonner, Ns . It follows
then, that if we make this consideration, that such a
magnetic medium must be a part of the transfonner
assembly to satisfy the conditions of (22). Therefore,
we can rewrite (22) as
(18)
_ dCPL _ VI _ V0
L- d t - NL
NL
(20)
Np2
Looking at the equations for interval DTs, we see
that we can combine them to remove the intennediate
secondary voltage value, V I:
Next, the ideal voltages that appear across the transfonner and inductor windings are found by simple
circuit analysis for each of the two switching states
of the converter, DTs and (l-D)Ts = D'Ts. Using
(1), we then relate the rate of flux change in the
transfonner core (CPT) and in the inductor (cpO to
these voltage magnitudes.
During DT,
• = _'1'_=
dA.T _s
V =VI
CPT
_
dt
Npl NS
Vs
¢T=
.J
v
o
=
[~l
NplJ
D
Vs
Conventional Forward DC-DC Power Converter (Idealized)
Drawn to Emphasize Magnetic Operations
Figure 17
6-37
¢L
As we will see later when we analyze the integrated
magnetic that results from this "synthesis" exercise,
NS must be equal to NL to realize a non-pulsating
output current waveform and to achieve an input-tooutput voltage transfer function that matches that of
a forward converter.
to release the energy stored in the center leg is needed,
and its turns (NU must equal those of NS from our
earlier discussions. Second, a winding (NP2) is
needed in the same leg of the core as that of the
primary for "reset" purposes. The magnetic arrangement that results is shown in Figure 18(b). Note that
these two windings are dotted so as to produce the
same flux directions in their respective core legs as
those in Figure 18(a).
Setting NL = NS in (23), we arrive at a final expression for ~T during interval DTs:
Our final task in this deductive "synthesis" process
is to combine the two magnetic arrangements of Figure 18 by the addition of switches and diodes to
establish the required winding voltage values for each
switching interval of the converter. From our knowledge of the forward converter, we would expect that
no more than one switch and three diodes would be
needed, and this is indeed the case.
(24)
Remembering our previous magnetic modeling, we
can interpret (24) as defining a magnetic assembly
in which there are three major flux paths. This equation also tells us that the flux change in an input
source related l2ath (~T) contributes to the change in
another path (cpU associated with the "inductor"
I'ortion of the assembly, as well as to flux change
(~o) in a third path. From the assumption made in
(23), this third path is related to the output voltage
value of the converter.
Figure 19 is the schematic of the integrated-magnetic
of the forward converter that results from our
efforts. Reflecting now on the steps that were necessary to place the transformer and inductor on a
single-core structure, we find that no exotic efforts were
really required and that, in reality, the deductive procedure followed was rather straightforward and ele-
~version
These general observations permit us now to sketch
a possible magnetic path arrangement that satisfies
the conditions of (24) for the switching interval DTs.
This sketch is shown in Figure 18(a). Note that the
"inductor" path includes an air gap, as we expect
that this leg of the magnetic core will need one to
establish the required amount of storage inductance
and to sustain DC bias without material saturation.
To the outer legs of the core arrangement, we have
added windings for primary and secondary in accord
with (18) and (19), dotted properly to produce the
required polarity of V0 for the indicated directions
of is and io .
Using the magnetic arrangement of Figure 18(a) as
a baseline, we now look at what must be added to
permit the conditions of (20) and (21) to be satisfied
for the other switching interval D'Ts. First, a winding
A Forward Converter with Integrated Magnetics
Figure 19
INTERVAL
INTERVAL
DT
1.1
r-<--'--<-~
y
D'T
Ibl
I
~-<--.----...,
I
I
I
I
I
I
I
'.
I
I
L
!/IT
I
I
I
I
_> __
I
I
I
J.. ____ ...J
!/IT· !/IL
I
A
'.
I
A
'l
I
Np1
I
I
+ 1/1 0
Developing Magnetic Core Arrangements for Each State of
the Converter in Figure 11
Figure 18
6·38
gantly simplistic. It now becomes clear that we can
use similar deductive methods to evolve integratedmagnetic versions of any transformer-isolated buckderived converter. Once this is done, we can then
use the principles of duality between converter circuits
[15,16] to formulate complementary boost topologies.
• UNIFOAM CROSS-SECTIONAL AREA (AC) THROUGHOUT
• MATERIAL PERMEABILITY OF ALL SECTIONS IS THE GAr,.1E
• DOTS RELATIVE TO Npl
Verification by Analysis
Following the evolution of an integrated-magnetic arrangement for a converter, it is worthwhile, if not
mandatory, that its structure be verified by analysis.
This is necessary in order that no major design aspect
has been overlooked, as well as to examine the effects
of any circuit parasitics (such as leakage inductances,
etc.) that may have been introduced by the integration
process.
"L
For the integrated-magnetic forward design of Figure
19, the analysis procedure begins with the extraction
of the magnetic from the overall circuit topology,
labeling each winding as to voltage polarities as well
as current directions that are imposed by the converter.
It is also advantageous to mark each winding terminal
and carry the identifying marks through the modeling
procedure, so that the final electric circuit model
found by the analysis can be directly substituted for
the magnetic assembly in the converter topology.
'L5"L
~
"'Je
6.
IDEAL T
NL
+
Ibl
'p
IDEAL T
l'S
~~
7
An isometric sketch of the extracted magnetic from
Figure 19 is shown in Figure 20(a). To this sketch,
we add flux directions in each leg, assuming that the
primary winding, NPl, is the exciting and dominant
winding of the magnetic. From this point on, the
analysis procedure follows the same steps of the twowinding transformer example of Figure 3(a), keeping
in mind that there are three magnetic paths and four
sources of mrnf in this case. Also, since we are only
interested in major flux paths, we have omitted all
leakages for this examination. However, they can be
easily added later and the corresponding electric circuit model changed to reflect their presence.
An Electric Model (b) for the Magnetic System of (a)
Taken from Figure 13
Figure 20
ideal transformer of turns ratio NPI :Ns , using standard
impedance tranSlatIon methods. This is done to position most of the induct!plces to the "right" of this
transformer in order that the final equivalent converter
circuit can be compared to that of a forward converter
with discrete magnetics. Note that this impedance
movement also required a change in the turns ratio
of the ideal transformer across Lg.
The electric circuit model of the magnetic of Figure
20(a) is shown in Figure 20(b). For the sake of brevity, we have not included the intermediate steps of
the modeling effort here, but they can be found in
[ 17] . This particular model is referenced to the main
primary of the magnetic, NPl, as evidenced by the
turns ratios of the three ideal transformers within it.
Inductance Lg represents that presented by the centerleg-winding, while the two inductances, Lc , represent
those presented by the outer-leg windings of the core
(referenced to NPl).
In practical designs, we would expect that Lc would
be much larger in inductance value than Lg, since
the permeability of free air is much less than that of
a ferromagnetic material. Therefore, we can simplify
our analysis at this point by eliminating the reflected
Lc across terminals 7 and 8 of the circuit model in
Figure 21. Also, we can assume that the Lc across
terminals 1 and 2 of the model can be viewed as the
"magnetizing inductance" of a real transformer of
ratio NPl:N s . It is also apparent that winding NP2
can be a part of this same transformer, since it parallels Lc.
The next step is to place the electric circuit model
into the converter circuit, as shown in Figure 21. In
making this placement, we have also "moved" Lg
and one of the two Lc inductances "through" the
With these assumptions in mind, we can now redraw
the circuit of Figure 21 in a slightly simpler form
shown in Figure 22(a). We also remember that the
6·39
mil
I
I
02
02
(a)
01
+
+
+
-C
+
+
R
V.
R
V.
1
1
Vs
Vs
PWM
la~It-o CONTROL
101
( NS) 2Lg
Placing the Model of Figure 14(b) Into the Topology of Figure 13
Figure 21
(b)
integrated converter of Figure 19 also constrained NL
to be equal to Ns and, therefore, the turns ratio of
the ideal transformer across L has been changed accordingly in Figure 22(a). With the turns ratio of this
transformer now 1:1, it can be completely removed
by simply reorienting the circuit positions of diodes
D1 and D2.
+0--+---4
L= Np1
;((·fl
I
r
\
(MAGNETIZING INDUCTANCE = Lei
NL " NS
Np1 • Np2 -+ TIGHTLY COUPLED
Further Circuit Manipulation (a) to Derive the Discrete
Magnetic Version of Figure 13 Shown in (b)
Figure 22
This final analysis step leads to the converter configuration of Figure 22(b) , which matches that of a
conventional forward converter! We see also that the
equivalent output inductance of this converter is equal
to that set by the gap of the integrated magnetic
multiplied by the squared ratio of "transformer" secondary-to-primary turns. If we had not chosen NL to
be equal to NS. this equivalent circuit and its predecessor in Figure 22(a) tells us that the equivalent
inductance would not be of the same value for each
switching state of the converter and, therefore, we
could expect the output current to be somewhat pulsating, just as would be experienced in a forward
converter with a "tapped" output inductor [ 17] !
as is evidenced by the dynamic current and voltage
waveforms of Figures 23 and 24, respectively. These
waveforms and their magnitudes were derived by inspection of 'the current and voltage conditions that
must exist in the electrical circuit models of Figures
21 and 22 for each converter state (continuous mode
of energy storage assumed). As we see from both
sets of waveforms, current and voltage stresses on
the switches and diodes remain the same as\ those
that would be experienced in a conventional f9rward
design.
It is evident from the results obtained above that the
analysis of an integrated magnetic by the use of electric model equivalents has great value and provides
much valuable design information. It is particularly
valuable in designing integrated-magnetic versions of
existing converter designs, where fIlter inductances
and transformer characteristics (turns ratios, etc.) are
known quantities. Using the equivalent circuit, such
as the one in Figure 22(b), these quantities can then
be directly related to the parameters of the integrated
magnetic, such as core dimensions and permeabilities,
winding turns, etc.
'01
'01
'02
'DC
'.
Voltage and Current Waveforms
From a "black box" standpoint, we would expect
an integrated-magnetic version of a converter to have
the same voltage and current characteristics as its
discrete-magnetic counterpart. For the forward converter design of Figure 19, this is indeed the case,
f.-
f-
f.-
l-
0
0
t--
I-
~
r--::::.
0
DT,
DT,
DT,
DT,
I-T,- I-T,-
N L • Ns
Idealized Currents In the SPC Topology of Figure 13
(Continuous mode)
Figure 23
6-40
is ON, energy is stored in the center-leg of the magnetic. During this same time period (DT s), load needs
are supported by the output capacitor, C, and by
magnetizing energy stored in the outer leg of the
magnetic from the previous switching cycle (via 03).
When Q2 turns OFF, QI is turned ON and the energy
stored in the center leg is magnetically routed to the
output load, R, via winding N s , with diode DI now
forward-biased and the other diodes non-conducting.
Like its discrete-magnetic contemporary in Figure 1,
the ideal input-to-output voltage transfer function is
simply
IJOS(Ql)
'NS
(7-B)
'NL
16-51
(25)
'DC
(CATHODE)
with the "inductor" turns equal to that of the "primary" winding of the magnetic assembly.
TS
TS
Adding Zero-Ripple Current Features
Voltage Waveforms in the SPC of Figure 19
(Continuous Mode)
Figure 24
Magnetic integration of the inductors and the transformers of buck and boost-derived converters does
not lessen the possibility of adding additional windings
to control ripple current magnitudes on output or
input lines.
Developing Integrated-Magnetic Boost Converters
It is possible to follow similar procedures of deductive
synthesis and analysis from the last section of this
paper to evolve integrated-magnetic versions of various boost-derived converters, such as the reverse
converter [16] in Figure 1. However, given a dual
buck-derived converter approach, it is much easier
to use duality methods [15,16] to derive these complementary circuits.
As we have seen in prior examples of integratedmagnetic forward and reverse converters, the inductive
part of their magnetic assemblies is isolated to one
path of the core arrangement. Therefore, by adding
another winding in these same paths and impressing
a voltage across it that is proportional in amplitude
and of the same frequency as that appearing across
the original inductor winding, we can steer the ripple
current from the inductor winding to the other winding
and, therefore, away from the output. We also have
the option of "trimming" the ripple current magnitude
by external means [11] if we so desire.
For example, given the integrated-magnetic forward
converter, duality manipulations then produce the integrated-magnetic boost equivalent of the reverse
converter of Figure 1. The unusual converter that
results is shown in Figure 25. In this case, when Q2
-
1
Vs
D
An Integrated-Magnetic Version of the Reverse Converter
Shown in Figure 1
Figure 25
6-41
Figure 26 shows an integrated-magnetic forward converter with output ripple current control capability.
Note that another winding has been added to the
center-leg section (NUV, along with a small series
inductor (Lx) and DC-isolation capacitor (CR).
Winding NL is the original inductor winding and the
new winding, NLR, is wound in close proximity with
it. This is done to minimize leakage inductance effects
and to maximize the mutual inductance that will exist
between the two windings.
Push-Pull Designs with Integrated Magnetics
Just as there are so-called "push-pull" versions of
the forward converter with discrete magnetics, integrated-magnetic arrangements are also possible. Two
viable approaches [14] are shown in Figure 28. The
first alternative, shown in part (a) of this illustration,
looks very similar to the single-switch forward design
of Figure 19, except that another set of primary and
secondary windings has been added on opposing sides
of the outer portion of the core structure. With the
windings dotted as shown (relative to the primary
winding, Np , controlled by Ql), diode DI will conduct when Ql is ON, and diode D2 will conduct
when Q2 is ON. During both conduction intervals,
energy is stored in the center leg of the magnetic,
and is then released to the load via diode D3 when
QI and Q2 are both OFF.
With the new winding positioned in the converter
circuit as shown in Figure 26, the value of inductance
Lx needed to cause the output ripple current to vanish
is found easily using (9) with a = I:
Lx = (
r;;~)2 Lg (~~ - 1)
(26)
where Lg is the inductance presented by the air gap
of the center leg with NL turns.
In the case of the second alternative shown in Figure
28(b), note that the phasing relationships (Le., dots)
of the two secondary windings (N s) have been
changed from those of Figure 28(a). This change
now prevents significant energy from being stored in
the center leg of the magnetic, since conducting secondaries lie on the same magnetic path as their corresponding primary windings. For example, in Figure
28(b), when Ql is ON, D1 will conduct and when
Q2 is ON, D2 conducts. Therefore, another winding
(NLl) has been added to the center leg to provide a
means of energy storage in the cent!!r leg when either
Ql and Q2 is conducting.
In a similar manner, a ripple-control winding can be
added to the integrated magnetic of a reverse converter, as is done in Figure 27. In this case, the value
of Lx to reduce the input ripple current to zero is
again the same as given by the solution of (26),
assuming that windings NL and NLR are tightly coupled together.
In actual designs, there will be some slight slope in
the "zero" ripple currents of these two converters
due to the presence of the inductances (Lc) of the
outer-leg primary and secondary windings. However,
since Lc is usually much greater in value than that
posed by the "inductor" winding, Lg, this deviation
from an ideally flat current waveform is normally so
small that it can be neglected for all practical
purposes.
Ideal Voltage Transfer Functions
The input-to-output voltage relationships for the continuous mode of inductive energy storage for the converters of Figure 28 can be easily found by equating
the volt-seconds appearing in the center leg of their
T+L
r----
----,
1
A
·0
1
1
1
v,
Q1
1J::"!I~tT_
DT,
Lg = gap mductance related
to NL
• - relative to Np1
Adding Ripple Control Features to the Integrated-Magnetic
Forward Converter of Figure 19 (NL. NLR tightly coupled)
Figure 26
6-42
D,
v.
Lg .. gap Inductance related
to NL
• dots relative to Np
Adding Ripple Control Features to the Integrated-Magnetic
Reverse Converter of Figure 25 (NL. NLR tightly coupled)
Figure 27
(bl
(.1
+C>------,
l1li
:
Two "Push-Pull" Versions of the Forward Converter
of Figure 19
Figure 28
integrated-magnetic assemblies during each of the two
switching intervals, DTs and D'Ts . In the case of
Figure 28(a), the DC voltage transfer is ideally found
to be
Vo _ NL {
D
}
Vs - Np 1 - D [I - (NUNs)]
For Ns equal to NL, (20) reduces to
Vo = NS x D
Vs
Np
(28)
which confirms that this integrated-magnetic converter
design is buck-derived. For the other circuit arrange-
(27)
6-43
ment in Figure 28(b), the DC voltage transfer function
is ideally
I -f-~
"oSIQ1I
DT,
Vo
NS[
1
](29)
Vs = Np 1 + (D'/D)(NLl/NL2)(NS/Np)
o~ 1 -f--I
"08 102)
In this instance, if
-
then (22) becomes
VANODEI D1 1
Vo _ NS x D
Vs - Np
~
Q1
~
DT,
--
Ql
Q2
Q2
Ql
Ia,
ON
02
ON
OFF
--Nl. =Ns
Voltage Waveforms In the Converters of Figure 28
(Continuous Mode)
Figure 30
Voltage waveforms also are quite similar to those of
a "push-pull" buck-derived converter, except that
the magnetic integration process has yielded another
potential benefIt-lower OFF voltage stress on the
two switches of the converter. Ideally, the OFF voltage stress on either switch in a conventional "pushpull" converter will be equal to a maximum value
of twice the input voltage magnitUde. However, in
its integrated-magnetic version, Figure 28(a), maximum OFF voltage stress is equal to the value of
source voltage plus the reflected output voltage. This
implies that, by proper choice of turns ratio, we can
signifIcantly reduce switch voltage stress by using
integrated magnetics as compared to a conventional
discrete-magnetic equivalent!
Practical Design Considerations
r-.
In order to reduce the possible harmful effects of
leakage inductances, the integrated-magnetic circuit
variation of Figure 28(b) is often preferred over that
of its contemporary approach in Figure 28(a). Note
that, in Figure 28(a) , corresponding primaries and
secondaries are positioned on the core legs so that
they cannot be tightly coupled together (Le., wound
tightly together). However, in the converter of Figure
28(b) , because corresponding primaries and secondaries lie on the same legs of the core structure, they
can be wound tightly together to maximize their magnetic coupling and to minimize parasitic leakage inductances. Therefore, for this reason, the design of
i '~ r--...
DT,
---
I -f--
OFF
"":m
D'T,
'--
I--T,- I--T,-
'm ~
o
-
--
0
ON
..!"'-
-
-
0
,.......
~
..-
-
~
VANODEfD2)
The typical current and voltage waveforms that will
be observed in either converter circuit of Figure 28
are shown in Figures 29 and 30, respectively. In the
case of the current waveforms, they look very similar
to those of a conventional "push-pull" quasi-square
wave converter, except that the magnetizing current
of the outer legs of the magnetic produces a minor
step in the otherwise continuous form of the output
current. As discussed earlier, these .. steps" will usually be very small because of the high values of
effective inductances of the windings on the outer
legs.
'.
-
-
Voltage and Current Waveforms
'02
~
(31)
which, like (28), confIrms that this circuit variation
is a member of the buck-derived converter family.
Interestingly enough, this converter, with its magnetic
replaced by its electric-circuit equivalent [ 17], is found
to be an integrated-magnetic version of the Weinberg
circuit [ 18] !
'D' o
-DT, ID'T,-f-- f - - -
0
-
'01 0
f-0
(30)
'.
D'T,
r----
v.
D'T,
I---T,-I-T,-
NL
-N.
Ideal Current Waveforms In the Converters of Figure 28
(Continuous Mode)
Figure 29
6-44
Figure 28(b) is usually chosen rather than the circuit
of Figure 28(a), even though the latter approach requires an additional winding to be added to the cel: 'ar
leg for energy storage.
tested by the authors over the past two years. Major
areas of interest for these empirical examinations were
verification of expected DC voltage transfer functions
and anticipated dynamic waveforms within the converters. In all cases, the results obtained were very
close to those predicted.
Another advantage of the converter in Figure 28(b)
is the presence of an inductor in series with both
primary switches. Thus, high instantaneous current
due to conduction time overlap is automatically eliminated. Also, this converter, like its discrete-magnetic
counterpart, can be made to operate as a boost converter, if the primary switches are purposely forced
to have overlapping conduction intervals [16].
Typical of the test results are the voltage waveforms
taken for a low-power version of the integrated-magnetic Weinberg converter of Figure 28(b), shown in
Figures 32(a) through (c). In this instance, the supply
voltage was set at 40 VDC, and PWM duty cycles
of Ql and Q2 manually adjusted to obtain an output
voltage of 5 VDC, across a load of 5 ohms and an
output capacitor, C, of 100 microfarads. PWM clock
frequency was set very close to 100 kHz, resulting
in a 50 kHz switching frequency of each primary
power switch. Also, to minimize voltage "spikes"
produced by leakage inductances, small R-C snubbers
(1.2 K ohm x 0.01 ILF) were added across each
primary winding and across NL2.
Small-Signal Modeling Considerations
Even though the magnetics of a buck or boost-derived
converter have been magnetically integrated, modeling
and analysis for stability and other dynamic studies
is straightforward, using proven techniques such as
the state-space averaging method [2]. All that is necessary is to derive an electric circuit model for the
integrated-magnetic system, and then to add it in
place of the magnetic in the converter topology. Following this, the normal procedures for deriving smallsignal averaged models are followed, accounting for
additional inductive elements [19] as presented by
the integrated-magnetic arrangement.
The integrated magnetic used for this particular experiment was composed of two joined Ferroxcube
E625 E-cores, with each of their center legs ground
to produce a total air gap length of 0.01". Three
bobbins were then placed on each leg of the core
structure, each with two sets of windings. Each
winding set was composed of 30 turns of # 20 AWG
wire and 15 turns of#20 AWG wire. The completed
magnetic was then connected into the converter arrangement of Figure 28(b), with windings phased as
shown. Windings with 30 turns were used as primaries
and for NLl, while the 15-turn windings were used
for the secondaries and for NL2.
For example, the averaged small-signal model for the
integrated-magnetic version of the forward converter
of Figure 16, using its equivalent electric-circuit topology from Figure 22, is illustrated in Figure 31.
Note that the averaged model accounts for all of the
inductive elements of the electric-circuit model, as
well as its isolation transformer. If we had chosen to
add leakage inductances to the electric-circuit representations in Figure 22, they could also have been
easily accounted for in the equivalent small-signal
model.
Using the relationships developed in Section 2, the
inductance of the 15T center-leg winding was calculated to be about 1 mh for an air gap length of
0.0 1". Inductances of the primary windings on the
outer legs were also calculated to be about 10 times
that of this center-leg winding, confirming the earlier
assumption that these latter inductances are indeed
much larger in value.
Laboratory Investigations
All of the integrated-magnetic versions of the various
converters shown in this paper have been built and
--
I
'0
"
1
1
D
n
Averaged Small-Signal Model of the Integrated-Magnetic
Forward Converter of Figure 19
Figure 31
6-45
(a)
(b)
o
o
o
o
(c)
(d)
o
o
o
o
Breadboard Test Photos of Voltages Within the IntegratedMagnetic Weinberg Converter of Figure 28(b)
Figure 32
Comparing the lab photos of Figure 32 to the ideal
waveforms of Figure 30 expected in this design, we
find that they closely match in all instances. Also,
the observed duty cycle of each primary switch is, in
close agreement with that predicted earlier by (31)
when correction is made for converter efficiency.
common operational property-flux change. Therefore, if relationships can be established between their
flux changes from a knowledge of their position in
a converter circuit, then it is possible to use them to
deductively synthesize a single magnetic system to
house all transformer and inductive functions.
Concluding Remarks
With integrated-magnetic versions of frrst-order buck
and boost converters established, the possibility of
magnetic integration of all of their family member
topologies becomes immediately evident.
In this paper, we have presented a broader and more
general insight into integrated-magnetic concepts for
switch mode DC-to-DC power converter circuits and,
hopefully in doing so, have dispelled a common belief
among designers that the transformers and inductors
of buck and boost-derived converters cannot be magnetically blended on a single core arrangement.
Many of these possibilities are shown in [17], while
others are still to be disclosed in the future. Although
we have presented integrated-magnetic versions of
single-output buck and boost converter arrangements,
the idea is easily extended to ~ccommodate multiple
outputs, or multiple inputs as the case may be. This
is easily done by the addition of more secondary or
primary windings on appropriate core legs, together
with contemporary switches and diodes.
Also contrary to popular thought, the demonstrated
ability to integrate the magnetics of these converters
also removes the restriction that impressed winding
voltages be completely proportional in all respects.
In general, this particular restriction is correct, if one
thinks only in terms of winding arrangements placed
on a single magnetic path. However, as we have
seen, multiple-path arrangements are perfectly viable
design approaches. It must be remembered that, even
though the winding voltages of the transformers and
inductors in converters may not be completely proportional in frequency and amplitude, they share a
In this paper, we have also shown methods whereby
more windings can be added to inductive core paths
to control input or output ripple current magnitUdes,
and significantly reduce them by proper internal and!
or external magnetic means. Even though we have
used single inductive paths in making this demonstration, it is also feasible to add more core legs with
646
appropriate air gaps to the integrated-magnetic systems
to control ripple current characteristics of contemporary windings, much in the same manner as was
shown in [13] for the two inductors of a CUK
converter.
References
The discussion here has been restricted to switch
mode power converters. Work presently in progress
indicates that these techniques can also be applied to
resonant converter circuits. Take for example the series resonant converter, with an inductor in series
with the transformer primary. This inductor can be
absorbed into the transformer simply by increasing
the primary-secondary leakage inductance. This produces an integrated magnetic.
Finally, we conclude our paper with the following
reflective thoughts. Although recent years have seen
an increased interest in integration concepts for minimizing the magnetic content of converters or for
enhancing their conversion properties, such techniques
have historical roots, reaching back into the early
years of this century. It is interesting to note that
many advances in magnetic concepts for electronics
were made in these earlier periods, with years following the introduction of the first electronic digital
computer seeing little or no new work being continued
in this important aspect of power electronics design.
Shortly thereafter, a sad decline in educational opportunities for engineers in the field of practical magnetics design began, and continues today. Ironically,
it appears now that magnetics could hold the key to
achieving smaller and more efficient power conversion
systems. As was noted in the introduction of this
paper, it is truly a time of frustration and reflection
for the power electronics engineer.
Acknowledgements
[1]
R. D. Middlebrook, "Reduction of Switching
Ripple in Power Converters," Proceedings of
the Sixth International PCI Conference, April
19--21, 1983, p. 6-14.
[2]
Siobodan Cuk, Modelling, Analysis, and Design
of Switching Converters, Ph.D. Thesis, California Institute of Technology, Pasadena, CA,
1977.
[3]
Siobodan Cuk, "A New Zero-Ripple Switching
DC-to-DC Converter and Integrated Magnetics," 1980 IEEE Power Electronics Specialists
Conference Record, p. 12-32 (IEEE Publication
80CH1529-7 AES).
[4]
J. K. Watson, Applications of Magnetism, John
Wiley and Sons, 1980.
[5]
C. Desoer and E. Kue, Basic Circuit Theory,
McGraw Hill, 1969, p. 453-457.
[6]
G. B. Crouse, "Electrical Filter," U.S. Patent
1,920,948 (August 1,1933).
[7]
S. Feng et.al., "Smail-Capacitance Nondissipative Ripple Filters for DC Supplies," IEEE
Transactions on Magnetics, Vol. MAG-6, No.
1, March 1970.
[8]
A. L1odye, "Choking Up on L-C Filters,"
Electronics Magazine, August 21, 1967 issue.
[9]
G. C. Waehner, "Switching Power Supply
Common Output Filter," U.S. Patent 3,916,286
(October 28, 1975).
[10] W. J. Hirschberg, "Improving Multiple Output
Converter Performance with Coupled Output
Inductors," 'Proceedings of POWERCON 9,
July 1982.
The authors would once again like to thank Morris
Rauch for his help in constructing the breadboards
of the various integrated-magnetic converters shown
in this paper. Also, a special thank-you is sent to
Colonel Wm. T. McLyman of CALTECH's Jet PropUlsion Laboratory for constructing one of the integrated magnetics for the breadboards, and to the
engineering directorate of Christie Electric Corporation
for the use of their laboratory facilities during the
breadboard testing efforts.
[11] G. E. Bloom and A. Eris, "Practical Design
Considerations of a Multi-Output CUK Converter," 1979 IEEE Power Electronics Specialists Conference Record, p. 133-146 (IEEE
Publication 79CH1461-3 AES).
To J. Cielo and H. Hoffman of mM, we send a very
special inspiration thanks for their exceptional 1971
patent concerning integrated magnetics for "pushpull" DC-to-DC power converter circuits.
[12] L. Rensink, Switching Regulator Configurations
and Circuit Realizations, Ph.D. Thesis, California Institute of Technology, Pasadena. CA.
1979.
647
[13] Slobodan Cuk, "Analysis of Integrated Magnetics to Eliminate Current Ripple in Switching
Converters," Proceedings of the Sixth International PCI Conference, April 19-21, 1983, p.
361-386.
[17] G. E. Bloom and Rudy Severns, Modern DC-
to-DC Switchmode Power Converter Circuits,
to be published in 1984 by Van Nostrand-Reinhold, Inc. (Catalog #0-442-21396-4), also
available from e/j BLOOM Associatp-s, 1312
Lovell Avenue, Arcadia, CA 91006.
[14] J. Cielo and H. Hoffman, "Combined Transformer and Indicator Device," U.S. Patent
3,553,620 (January 5, 1971).
[18] A. H. Weinberg, "A Boost Regulator with a
New Energy Transfer Principle," Proceedings
of the 1974 Spacecraft Power Conversion Electronics Seminar (ESRO Publication SP-103).
[15] Slobodan Cuk, "General Topological Properties
of Switching Structures," 1979 Power Electronics Specialists Conference Record, p. 109-130
(IEEE Publication 79CH1461-3 AES).
[19] R. D. Middlebrook, "Design Techniques for
Preventing Input Filter Oscillations in SwitchedMode Regulators," Proceedings of POWERCON 5, May 1978, p. A3-1 to A3-16.
[16] Rudy Severns, "Techniques for Designing New
Types of Switching Regulators," Proceedings of
the International PCI Conference, September
1979, paper 5.5.
6-48
6.1.4 Siliconix Switch Mode Power Supply Kit
Introduction
The Siliconix power supply kit is an educational tool
which will give you a hands-on introduction to both
switch mode power converters (SPC) and devices
from Siliconix's expanding line of power converter
components. The circuit operates at a switching frequency of 150 kHz and a ripple frequency of 300
kHz. The use of switching frequencies greater than
100 kHz represents the present trend in SPC design.
The benefits of high frequency operation are reduced
size, weight, and cost along with improvements in
transient response and input noise rejection.
SPC
Circuit Description
1------.,.-....
Vo
Power Supply Block Diagram
Figure 1
There are literally hundreds of different SPC circuits
from which to choose. For the kit we have chosen a
buck-derived circuit which is commonly used in offline applications. Normally, this type of circuit would
be operated from a 100 to 350 volt source, but for
reasons of safety and convenience, we have scaled
the source voltage down to 24 to 32 VDC. The
waveforms and circuit operation remain the same. It
is important to emphasize that the kit is designed as
a learning tool and is not intended to be used as a
production power supply.
A simplified schematic for the power converter is
S2) are connected in a half-bridge configuration. In
this configuration the switches see a maximum voltage
of Vs and a peak current of twice the input current
(Ii). The relatively low voltage stresses on the power
switches make this circuit suitable for high input
voltage (100 to 400 volt) operation; however, for a
given type of switch, the high switching currents
become a limiting factor, so this design is rarely used
at power levels above 500 watts.
A block diagram of the power supply is shown in
Figure 1. The power converter uses a control loop
which compares the SPC output voltage (V0) to a
regulated reference voltage (Vref) at the error
amplifier. The error amplifier generates an error voltage proportional to the difference between V0 and
Vref. The error voltage is then compared to a ramp
signal and converted to a pulse-width-modulated
(PWM) control signal for the power converter (SPC).
As stated earlier the kit SPC operates from 24 to 32
volts, a much lower than normal voltage for this configuration for reasons of safety. The effect of low
input voltage and high switching currents is a reduction in the circuit efficiency of 5 to 10%. This should
be kept in mind when testing the circuit.
6-49
Ll
L_
SWitching
Section
Level Translation
and Isolation
Low·Pass
Filter
Simplified Power Supply Schematic
Rgure 2
correspondingly higher losses. Schottky diodes also
have a fast reverse-recovery-time (trr) which reduces
the rectifier losses at very high switching frequencies.
Referring again to Pigure 2, the average value of V2 is
the output voltage (Vo), and the relationship to the
PWM is:
Vo = Vs
2
(NNps)
(1)
Where D is the portion of the switching cycle during
which the switches are conducting
t1
D= -
Ts
V2
(2)
where: t1 is the conduction time of the switches.
Ts is the switching period.
We can see from equation (1) that if V0 is to remain
constant, D must become smaller as Vs increases
and vice versa. This is the basis for the control action.
Idealized Waveforms
Figure 3
Switches S 1 and S2 conduct alternately for equal
periods of time. The actual "on" time (tI) of the
switches is controlled by the PWM drive to each
switch. The waveforms shown in Pigure 3 illustrate
how the circuit (in Pigure 2) works. The action of the
switches causes a pulsating DC voltage (VI), the AC
component of which is applied across the primary of
T1 (C5 blocks the DC component). The primary voltage (Vp) is stepped down to approximately 7.5 volts
by TI and rectified by BR2 producing a pulsating DC
voltage (V2).
Por most applications the switching noise and ripple
must be very low, so in addition to averaging, the
output filter must reject most of the AC component of
V2. This is accomplished by making the cutoff frequency (fc) of the fiiter lower than the ripple frequency (fr). In the kit converter, fc is so much less
than fr that the equivalent series resistance (ESR) of
the output capacitors greatly reduces the available attenuation. Pigure 4 shows an equivalent circuit for the
filter and attenuation curves with and without the
ESR.
BR2 is a dual Schottky diode assembly. Schottky
diodes are used in low voltage, high current output
applications because of their low forward voltage drop
(Vp). BR2 has a rated Vp of 0.5 volts and accounts
for about 30% of the circuit losses. A normal PN
junction diode has a Vp of 0.8 to 1.2 volts with
The filter inductor current waveform can have two
possible shapes: either the current goes to zero during
the switching interval, or it doesn't. Another way of
stating this is either all of the energy stored in Ll while
PET 1 or PET 2 are conducting is discharged into the
load when PET 1 or PET 2 are off, or it is not. At first
6·50
=
VBR2
performance. This will be discussed again in the control loop stabilization section.
Ll
The schematic for the complete power supply is given
in Figure 5. All control loop and drive functions are
performed by the Siliconix PWMl25 (ICI). The
PWM125 is a high performance pulse-widthmodulator IC designed specifically for high frequency
power converter applications. The PWMl25 has the
same pinout and is functionally similar to its predecessor, the SG1525A. The PWM125 features separate
analog and digital grounds, true TTL gate structures
for precise metering of pulse modulated signals, and
completely redesigned totem-pole output sections with
very low crossover current (120 rnA). In earlier
totem-pole designs (Le. l525A), crossover currents of
950 rnA caused catastrophic failure at high speeds.
Other features unique to the Siliconix PWM series
(PWM 25/27, 125) are high operating temperature and
high voltage capability. The PWM125 will be discussed in greater detail later in this text.
Gnd
100Hz
1kHz
10kHz
100kHz
OdBI------1:::
-IOdB
-20dB-------- - - - - - --- ---------
--- -1 FiOie -
"',
-30dB
.2
;'~~~b;R
Poie--":::;:--
Slope
,
without ESR "
-40dB ......~~~~......._~~~--U..'-'-'-_-'-........,'--'-"........
Figure 4
glance this seems like a trivial distinction, but it is not.
The conduction mode of the inductor current defines
the operating mode (continuous or discontinuous) of
the converter. The open loop static and dynamic properties of the converter will differ dramatically from
one mode to the other. This is characteristic of all SPC
circuits. The conduction mode must be known to
properly stabilize the loop and to predict the circuit
Actual Circuit Waveforms
The waveforms shown in Figure 3 are idealized. In the
actual circuit, the waveforms are somewhat different,
and it is worthwhile knowing why. Figure 6 shows the
actual waveform (Vp) across the primary ofT!. What
interests us are the spikes and ringing when one of the
FETs turns off.
24V
c.
toOOJ.l f
l 1
-=
Cll
'1011
2'5
13
C3 47IJf
14
10
Rl
r--,
75k
BR2 CTB34
12
L 1 121tH
'--_-...--_-o~'out
SV@8A
-
C13
47pf
R3
75k
R12
10k
Siliconix SMPS Kit Schematic
Figure 5
6·51
I
L£
:
Tl
:JIIC
Figure 78
n
L£
"off"
Vs
Vin = 24 Volts
lout=BA
Vertical Scale - 5 V/Div.
Horizontal Scale - 1 p,S/Div.
-=-
C5
IlL
"on"
Figure 7b
Primary Waveform
Figure 6
In a real transformer, the primary and secondary
windings are not perfectly coupled, and some of the
primary flux linkages will not be shared by the secondary. This creates an equivalent series inductance (referred to as leakage inductance) LQ shown in Figure 7a.
The mechanism for generating the voltage disturbance
is as follows:
1. Referring to Figure 7b, assume SI is off and S2
is on. The primary current is flowing through L Q
and S2.
2. Referring to Figure 7c, if we turn S2 "off" suddenly, the current in L Q must continue to flow.
What happens is that the voltage across L £ reverses and rises until 01 conducts (01 is the integral diode in the MOSFET). When the energy in
L£ has been discharged, 01 will cease to conduct.
The circuit now rings at a frequency determined by
L £ and Ca , where Ca is the total device and stray
capacitance at that node. If 01 were not present
then the voltage across S2 would rise until the
switch went into avalanche breakdown. 01 limits
the voltage rise to just above Vs and acts as a
clamp.
Figure 7c
To reduce the ringing, there is a simple RC network
made up ofRS and C12 across the primary ofT1. This
does not completely eliminate the ringing but greatly
reduces it.
Vin = 24 Volts
lout=8A
Vertical Scale - 5 V/Div.
Horizontal Scale - 50 nS/Div.
Another feature of this waveform which may appear
puzzling is the apparent change in turn-off switching
between heavy and light loads. This effect is shown in
Figures 8 and 9. Figure 8 is the turn-off waveform for
FET 2 with an 8A output load. The turn-off transition
S2 Turnoff Voltage Waveform
Figure 8
6-52
Vin = 24 V
lout = 0.5 A
Vertical Scale - 5 V/Div.
Horizontal Scale - 50 nS/Div.
Vln = 24 V
lout = 0.5 A
Vertical Scale - 5 V/Div.
Horizontal- 1 ItS/Div.
S2 Turnoff Voltage Waveform
Figure 9
FET2 Gate
Figure 11
Vgs
A1
~A2
~'-'----'
11
ov
12
Figure 12a
Vin = 24 V
lout=8A
Vertical Scale - 5 V/Dlv.
Horizontal- 1 ItS/Div.
FET 2 Gate Voltage Waveform
Figure 10
Figure 12b
is 50 nsec. Figure 9 is the same waveform with the
output load reduced to 0.5 A, and the apparent
switching time is 200 nsec. In fact, the switching time
of the PET has not changed at all; it is 50 nsec in both
cases. The reason for the slower transition time at light
loads can be explained by referring back to Figure 7c.
At tum-off the current flowing in S2 will be available
to charge Ca , and the rate of voltage rise across S2 will
be dependent on the current (ILQ) available to charge
Ca. When the output current is 8 A, ample current is
available to charge Ca , and the rate of voltage rise is
determined by the PET switching characteristics.
However, when the output load is reduced to 0.5 A,
the current available to charge Ca is reduced by a
factor of 16. In this case, the charging of Ca completely dominates the voltage rise time.
Another waveform of interest is the gate drive
waveform (Vgs) shown in Figures 10 and 11. Ideally
Vfe should see symmetrical waveforms independent of
D. A cursory examination of the oscillographs shows
that this is, not so; the positive and negative amplitudes
vary with D. Since this is an AC coupled waveform
(through capacitor C3 in Figure 12a), the enclosed
areas are equal as is shown in Figure 12b. From a
6-53
-
in series with the primary of Tl. The AC voltage
across C5, due to the transformer primary current,
subtracts from the applied voltage and causes the tilt in
the output waveform. As the power supply output load
increases, the voltage drop across C5 increases which
increases the tilt in V2. When this type of circuit is
used directly from the AC mains at the same output
power rating, the waveform tilt is insignificant. The
higher primary voltage results in a lower primary current which, in tum, results in a smaller voltage drop
across C5. By scaling down the input voltage for the
kit, the effect of C5 is magnified. The waveform tilt
could be reduced by increasing the value of C5, but
this is not necessary for the purposes of this kit.
- -~-;
FET 1
_ _ _ _ ...JI
Transformer Drive Clr. for Wide Variation of Duty Cycle
Figure 13
practical point of view, this means that as D is
increased, the amplitude of Vgs during t1 must decrease while the amplitude during t2 increases. In this
particular application, the variation in Vgs is not large
and is not a problem. In some circuits the variation in
D may be much larger causing Vgs to be too large at
small duty cycles and too small at large duty cycles.
This problem can be eliminated by modifying the
drive circuit as shown in Figure 13. For the kit this
modification is unnecessary because of the small variation in duty cycle.
Control Loop Stabilization
For the power converter to function as a regulated
power supply, some means is needed to maintain a
constant output voltage, as the output load and input
voltage vary. This function is provided by the feedback control loop. As was pointed out earlier, most of
the control loop functions are provided by the
PWM125 (IC1). The issue to be discussed now is that
of loop stability and transient response performance.
Figure 16 shows an equivalent block diagram of the
power supply control loop. To analyze the stability of
the loop, we will calculate the gain and phase characteristics of the loop and then graph them using a Bode
plot.
The voltage waveform at the output of BR2 (V2 of
Figure 2) is shown in Figures 14 and 15. It is clearly
not ideal. In addition to the ringing introduced on the
primary leakage inductance, there is a distinct tilt to
the top of the waveform which increases as the output
load increases. The tilt is due to the impedance of C5
The total loop gain (AT) is: AT = G1 G2 G3
Vln = 24 V
lout=1A
Vertical Scale-2 V/Div.
Horizontal Scale - 500 nS/Dlv.
Vin = 24 V
lout = 8 A
Vertical Scale-2 V/Div.
Horizontal Scale - 500 nS/Dlv.
Output of BR2 @ 1 A Load
Figure 14
Output of BR2 @ 8 A Load
Figure 15
NOTE: Current Goes Slightly Negative During Transition Period.
6·54
(3)
C13 47pl
To Output
R12
S.6k
Error Amp
>---4-..... ToPWM
Gain Blocks
Figure 16
R3
75k
G I represents the gain of the power converter. The AC
part of G I is determined by the output filter which, as
we saw in Figure 4, is a simple 2 pole LC filter with an
ESR zero. The result is a single-pole characteristic
with the cutoff frequency (fc) equal to 2.4 kHz. The
DC portion of Gl is determined by the duty cycle and
the transformer (Tl) turns ratio for D = 0.5. The DC
value for G I is:
Gl = 0.7
Figure 17a
(4)
This analysis presumes that the converter is operating
in the continuous inductor current mode. If the load is
reduced to the point where the discontinuous mode
exists then the converter will have a single, rather than
a two pole characteristic. Simultaneously, the loop
gain will drop. Fortunately, these two changes work
together in this application, and the loop remains stable.
1.9 kHz
41 kHz
Frequency
Figure 17b
G2 is the gain of the error amplifier, and we can
calculate it from Figure 17. We will assume the
amplifier gain is large. The DC value of G2 is very
large (60 dB). There will be a pole at the origin due
to C2, and a zero will appear in the transfer function
at a frequency determined by R6 and C2:
Where Vramp is the peak-to-peak amplitude of the
ramp voltage which appears on pin 5 of ICI:
Vramp = 3.2V p-p
(8)
So that:
1
fo = 27TRC = 1. 9 kHz
(5)
G3 = .31
The asymptotic gain at 1.9 kHz is:
R6
G2 = R12 = 14.6
We can now plot AT, as shown in Figure 18, in
asymptotic form. We can see that gain crossover is
about 9 kHz with a single pole roll-off which should
give us a phase margin close to 90 o. The ultimate test
for stability is the transient response to a step load.
The transient response of the kit is shown in Figure 19.
The response is well damped with a very small step
change (about 40 mY). Given that the load step is 1 A
superimposed on a 5 A DC load, the transient response
is excellent. The small high speed oscillations on the
leading edge of the step response are inherent in the
dynamic load and are not due to the control loop.
(6)
The combination of C 13 and R6 will produce a pole at
41 kHz. An asymptotic approximation for G2 is
shown in Figure 17b.
G3 is the gain characteristic of the pulse-widthmodulator. Given the particular modulator used in the
PWM125 (ICI), the gain expression is very simple:
G3=
Vramp
(9)
It is quite possible to have a much higher gain crossover frequency and an even better transient response,
(7)
6·55
l1li
I
100Hz
1 kHz
10kHz
100kHz
1 MHz
Figure 18
Clocking and Control Circuit (Figure 20)
The ramp and timing signals are generated by charging
a capacitor (Ct at pin 5) from an internal current source
which is programmed by an external resistor (Rt at pin
6) using a current mirror technique. A comparator in
the oscillator section is set to trip at 0.8 V (low) and
3.6 V (high). When capacitor Ct (pin 5) is charged to
the high trip point, the discharge transistor (Qd) turns
"on" and discharges Ct to ground through pin 7 (Qd
collector). Pin 7 is normally connected to pin 5 on the
PWM125. When the capacitor (Ct) is discharged to
approximately 0.8 V, Qd is turned "off," and the
cycle starts over again. The slow charge and fast
discharge cycle generates the ramp signal for the
pulse-width-modulator circuit. Other circuits in the
oscillator generate the clock pulse for the metering
circuits.
Vln = 24 V
lout= 5A
Vertical Scale - 50 mV/Dlv.
Horizontal Scale-100 /-~_
"177_ _
7.
PWMOUT
• •
DV
()!!._ _ _ _ _---l
SOFTSTART & SHUTDOWN
PWM125 Clock & Control Circuit
Figure 20
---1
V .......
/,,"--...-,,'3'0 COLLECTOR
11~PUT
OUTPUT ENABLE F/F
UU
fin
CLOCK
fin
PWM125 Metering & Output Circuit
Figure 21
amplifiers, with a load resistance of ~ 10 megohms
is 60 to 80 dB. Because the error amplifier is a transconductance design (Vo = gmRL Vi), as the output is
loaded the gain is reduced. The details on how to use
this amplifier were discussed in the section on control
loop stabilization.
in input voltage or output current are met with little or
no change in output voltage. Figure 20 shows the error
voltage vs. ramp at about 50% duty cycle.
Metering and Output_Circuits
The clock pulse generated by the oscillator circuit is
used to trigger the metering flipflop which controls
output gate selection. This ensures that the totem-pole
outputs at pins 11 and 14 are never on simultaneously.
The metering latch ensures that only one modulated
pulse per clock period is allowed and prevents doublepUlsing of the outputs.
The pulse-width-modulator compares the incoming
ramp and error voltages to form the pulse width modulated signal shown in Figure 20 (the shaded areas).
As the error voltage changes, the active time on is
varied according to the reference crossing points on
the rising and falling edges of the ramp, and the pulse
width is varied or "modulated" accordingly. As demand for output current increases, or as the power
converter input voltage changes, the change is sensed
by the error amplifier, and the error voltage changes.
These changes cause the ramp crossing points to shift
and the pulse width to vary proportionately. Changes
The output control gates are 3 input NOR gates (4
input in the PWM25, 27) controlled by the two previously described signals and the undervoltage lockout! .
shutdown circuits. When bipolar transistors are used
as the power switches, a period of dead time is re-
6-57
--
Regulation and Shutdown Circuits
VrefOUT 0"----1~-----,
The regulation circuit provides a 5.1 volt ± 1% source
for the internal circuitry and a 20 rnA output at pin 16
(Vref). Also included is the undervoltage lockout
which ensures a stable off condition when IC V+
(pin 15) is below 8 V.
V+IN
IC GROUND
0"'---1
. - - - - - _ TO FIG 3 PIN c
IPWMj
60llA
SOFT START CAP
The soft-start circuit provides protection by bringing
the pulse width up slowly. This feature prevents cold
start stresses on the SPC components when the power
supply is turned on under heavy load conditions.
0 - ' ' - - - - - - + - - &____ FROM V"f
SHUTDOWN 0"1O,---'\M~+--J::.
The shutdown circuit provides an external control for
turning the outputs "on" or "off." When the shutdown transistor (Qs& is "off," a current source from
the regulator circuit charges the soft-start capacitor at pin
8. The charge time determines the speed at which the
modulated pulse comes up to the required width.
When Ql is turned "on" by the undervoltage lockout
circuit or by an active high (2.4 V) on the shutdown
pin, the soft-start capacitor is discharged to ground
which disables the PWM circuit. When Ql is turned
"off," the soft-start capacitor charges again, and the
soft-start cycle is repeated.
PWM125 Regulator & Shutdown Circuit
Figure 22
quired to get rid of the stored charge inherent to
bipolar transistors. The maximum dead time allowed
in the PWM125 is 120 nsec. When MOSPOWER
FETs, which have no storage time problems, are
used as the power switches, the dead time is set to
minimum by connecting pin 7 (Qd collector) directly
back to pin 5 (Ct). Do not use the PWM125 with
bipolar transistors as there is no dead time control.
Use the Siliconix PWM25 , 27 which have the adjustable dead time feature.
6·58
6.1.5 A 500 kHz Switching
Inverter for 12 V Systems
(AN79-1)
The principles of operation may be described by the
basic circuit shown in Figure 1, which includes the
pulse width modulator (PWM) control circuit; the high
frequency MOSPOWER PET Switch; the flyback circuit inductor (L), diode (D) and capacitor (C); and an
error amplifier.
This note describes the design of a 12 V to ±20 V
inverter. The design is of the flyback type, made much
more practical by the use of DMOS devices running at
a high switching rate. It is an energy transfer circuit,
not a voltage or current transfer circuit; the output
power is maintained at a constant level for a given
pulse width-modulator (PWM) operating point. If the
load requires less current, the output voltage will soar.
Conversely, if the output current demand increases,
the output voltage will sag to maintain the constant
output power (V x I). This contrasts sharply with
conventional circuits that deliver a constant output
voltage per PWM operating point. Maintaining constant power permits simplified circuitry with reduced
magnetic and filter requirements.
At the heart of this design is the PWM control circuit
which provides the control pulse to the DMOS Power
Switch in the flyback circuit. The output of the PWM
is a pulse whose width is proportional to the input
control voltage and whose repetition rate is determined
by an external clock signal. To provide the control
input to the PWM and to prevent the output voltage
from soaring or sagging as the load changes the error
amplifier and reference voltage complete the design.
They act as the feedback loop in this control circuit
much like that of a servo control system. Pertinent
waveforms are given in Figure 2; 2a describes the
conditions that exist at 50% (maximum) duty cycle
and 2b describes those at a low duty cycle.
VSUPPL v 1Vsl
.....-otf-_-......_OUTPUT
Operation is as follows:
CONTROL
II1II
I
1. Between to and t} the DMOS is turned on and
applies the supply voltage across L. The drain
current is closely approximated as IL = t x V S/L
and the final current IL (peak) = (tl - to) Vs/L.
The energy stored in L is EL = (t} 2 - to 2)
Vs2/2L.
Figure 1
6-59
~OCK
E~GE
t
t
_ . L -_ _ _ _ _ _ _
L -_ _ _ _ _- - ' _ ' - -_ __
TIME-
I
r
PWM _ " ' -_ _........._ _ _. L -_ _........_ _ _L-_ _ __
OUTPUT
IL~~~
--tO~--~tl-~t2~t~3~---~~--~--, .. a.VOUT + VDlODE~A ..
VORAIN
J
~
!"II
-
r-1ftltl
Res
'0.-oJ
"'VI
Figure 28
CLOCK
EDGE
t
t
t
TIME-
OUTPUT
D
'L
.0.
PWM
0
0
.0.
.0.
VORAIN
Figure 2b
2. At the instant of tl the VN64GA is turned off and
its drain voltage increases to the voltage on C plus
the forward voltage drop of D because of the inductor L. This works well when VOUT is much
larger than VS and permits nearly all of the energy
stored in L to be transferred to C with no large
current or voltage spikes.
L will be 'TJ VS2 /8 fCLOCK POUT, where'TJ
is the overall power efficiency of the circuit.
The power losses in the circuit are, in order of
importance:
A. (lD)2RDS loss in the MOSFET
B. ILOAD VF loss in the catch diode D
C. Transient charge losses in D due to a slow tum-on
characteristic. This actually allows many volts of
forward bias on D just at time tl while the diode
is attempting to turn on. Tum-off losses are not
important in this circuit.
D. Loss in L due to hysteresis and saturation effects.
The inductor is required to pass all of I =
4POUTI'TJ VSUPPLY, which usually results in
magnetic saturation and losses. Fortunately, the
higher frequencies allow fewer turns on L, reducing the number of turns and ISUPPLY product,
and the core material of L can be high-frequency
ferrite, which is less prone to saturation (for a
given inductance).
E. Simple CSTRA Y V20UT fCLOCK losses, where
CSTRAY includes CDS of the FET and CD of
the rectifier.
3. Between t2 and t3, no net DC current flows
through L, and VDRAIN tends toward VSUPPLY,
although a ringing will occur between L and the
capacitance of the FET and D. The cycle is repeated at a frequency fCLOCK, such that the
power drain from Vs is (1/2(tl - to)2 VS2)/
fCLOCK·
If we let the duty cycle
8
= tl
- to
t3 - to
= (t}
- to) X fCLOCK,
1
where fCLOCK == - - - ,
t3 - to
8 VS2(t1 - to)
then PIN = --'-"'----2
If 8
Figure 3 is the schematic of a simple 35 watt inverter
designed to produce ± 20 V regulated outputs from
12-16 volt inputs. Ul is a simple Schmitt-trigger oscillator with a nominal 50% duty cycle. This duty
cycle waveform buffered into Q5 by Q4 and U2
through U6 runs the system at full output power.
VS 2
= 1/2 max, then PIN max = --:&,....=--8L lCLOCK
The peak current through the DMOS will
4 POUT
.
closely be - - - ; and the value of mductor
'TJVS
6-60
VSUPPLY
II
FERRITE BEAD(S)
UNITRODE
'~~fGNo
R21
100
R1
330
1/2W
NOTES CAPACITORS IN MICROFARADS UNLESS NOTED
RESISTORS IN OHMS, 1/4W, 5% UNLESS NorED
tTHESE PARTS MOUNTED ON HEAT SINKS
·T,. T2. T3. LT. L2. L3.l4 - SEE TEXT FOR DETAilS
C17
01
QPTC1
,..- .......
0,5
GEH11Al
I
L-+________________~.
lN4148
'.,
"
R,.
....-
20K
1K
.Lr~\--------I
:'L', ,
5,
f'
1K
R13
'5
1K
RlO
500
5V
..... ·UST
~
12K
C"
200 pF
R17
lOOK
100 kHz, 150 Watt Half·Bridge Switching Power Supply
Figure 1
R"
1K
All outputs are isolated from the AC power line. The 5
volt output was chosen to be the main regulated output
controlled by the pulse width modulator. Feedback
from this output is optically isolated from the line side
of the power supply. The complete supply is overcurrent protected by sensing the source current in the
lower MOSPOWER FET (Q6) and using this signal to
shut down the supply.
Circuit Board
The circuit board layout (Figure 2) uses double sided
construction. Most of the traces are on the bottom side
of the board while the top side is used as a ground
plane. Three ground planes are used - one for the
input and control circuitry, one for the ±15 volt outputs and one for the +5 volt output. If a common
ground is desired for all DC outputs, both output
ground planes may be connected together.
Construction Details
Plated-through holes are not necessary for making the
circuit board, but they would be useful. If platingthrough is not used, all of the components connecting
to top traces or the ground plane must be soldered on
top of the board in addition to any soldering necessary
on the bottom. There are a few connections from one
side of the circuit board to the other side that do not
have components mounted in them. If plating-through
is not used, a short piece of wire must be soldered in
these holes to both sides of the board. All solder points
on the top side are indicated by an 'X'. Table I shows
the recommended drill sizes to be used for drilling the
circuit board .
Careful circuit board layout is very important for the
proper operation of high power, high-frequency
switching regulators. Single point grounding is absolutely necessary to prevent ground loops from rendering the circuit totally unstable or inoperable. Ground
planes are also required to lessen the effects of electromagnetic interference on the circuit. Presented here
is a circuit board layout which is known to operate
correctly and reliably. Use of this layout will make the
construction of this power supply much simpler and
will speed your evaluation of the VN4000 series of
high-voltage MOSPOWER FETs.
•
•
10"
Circuit Board (Bottom Side)
Figure 2a
Circuit Board (Top Side)
Figure 2b
6-64
Secondary (2) -
Table 1
Recommended Drill Sizes
Make all connections to the bobbin according to the
parts placement diagram (Figure 3).
The following drill sizes should be used on the circuit board:
#66
ICI-IC3. 1/4 W resistors. disc capacitors. opto-isolator.
QI-Q5. 01-06. C6. Cl3
#60
TI bobbin leads. 1/2 W resistors. CS. C9. T3 secondary. TRI.
1'1. CIS. C21. C22. 02. 0I4
#57
I W resistors. T2 bobbin leads. 07-DIO. C7. 013. C13. CI9
#44
TO-3 lead sockets. line cord
#23
1'1. IC4. IC5 and TO-3 screw-mount holes. 5 V CT
3/16"
0ll.0\2
1/4"
C5
5/16"
Banana jacks
T2 -
Primary (1) - The primary is made up of a type .of
litz wire using several strands of regular enamel wire
(refer to Figure 4). Cut 8 strands of #28 enamel wire
(about 6 feet long) and place them together in parallel.
Twist the ends only together (not the whole length),
but do not solder. Fold this twisted bundle of wires in
half and wind 8 turns of this doubled over bundle onto
the bobbin. Cut the folded over end of the bundle so
that there are now 4 ends coming out of the bobbin.
Twist the ends of each newly cut bundle. Next, connect one of the beginning bundles (D) to the end of the
other (B). This effectively connects the bundles in
series, wound in the same direction, to fonn a single
16 tum primary. The purpose of winding in this manner is to equalize the flux across the transfonner core.
There should now be two ends of the wire free and two
ends connected to each other. Connect the free ends to
the bobbin as shown in Figure 3. Make sure all windings are wound tight and neat - do not waste any
space. Now put a layer of transfonner tape to cover the
primary.
Three transfonners are used in this power supply: 1)
DMOS drive transfonner, 2) power transfonner and 3)
current sense transformer. The winding details
explained here should be closely followed, especially
for the power transfonner T2.
DMOS Drive Transfonner
Using the correct bobbin and pot core (see parts list),
wind the following:
Primary (1) -
Power Transfonner
Using the appropriate bobbin, wind the following (pin
1 of the bobbin has a notch for identification):
Transformers
Tl -
30 turns of # 24 enamel wire
20 turns of #24 enamel wire
7915
7815
~
~c,~
C
0
I
0
I
C
---t
-15V
+
+1SV
:!:i...
C11,-
GND
1,11,i
GND
TTTT
+5V
~5V
~
L3RJR'{~:
e11
R'<4 -H- ,
,rt
R'9
IC2...1..
C'2OPT!'
L
ADJ
R'3~~gt
CA
3130
G
L4
R'8~~...L
~ ~,R17
-Ilc
,
'C'5
~R16
21
Parts Location Diagram
Figure 3
6-65
..
.
gSIIICOnIX
incorporated
l00KHZ-l50WATT
.-J
SWITCHING POWER SUPPLY
IDII
Secondary (4) - ± 15 volt secondary: Make another
litz wire similar to the primary but this time parallel 6
strands of # 28 enamel wire about 40 inches long
(refer to Figure 6). Twist the ends together and double
the bundle over itself. Wind 5 turns of this doubled
over bundle neatly on the bobbin (41/2 turns for ± 12
V outputs). Cut the double end and connect Band D
together and connect to bobbin pin # 11. Connect the
two free ends of the bundle to the other bobbin pins.
Put a layer of transformer tape over these windings.
Secondary (2) - Start-up Winding: Wind 4 1/2 turns
of #24 enamel wire (about 20 inches long) evenly on
top of the primary winding. Connect the ends as
shown in Figure 3. Put a single layer of transformer
tape over the start-up winding.
Faraday Shield - This is a shield used to minimize
radiated electromagnetic interference (EMI). Cut a
piece of 5/8 inch copper tape about 3 inches long and
wrap this around the existing windings (refer to Figure
5). Do not make a complete loop -leave about 1/4
inch between the ends so that they can't touch. Solder
a small stranded wire (#20) onto the shield and connect it to the bobbin as shown in Figure 3. Put a layer
of transformer tape over the Faraday shield.
Secondary (3) - 5 volt secondary: Make up some
insulated copper tape by placing transformer tape on
one side of a 10 inch long piece of 5/8 inch wide, 2 mil
copper tape (refer to Figure 7). Make two of these
insulated tapes. Make sure the transformer tape is
slightly wider than the copper tape so that the windings don't short to each other. Wind both of these
insulated tapes at once (like bifilar tape) for two turns.
Connect the beginning of one tape (A) to the end of the
other tape (D) - this is the 5-volt center tap. Connect three # 18 stranded wires (or # 18 ribbon cable) in
parallel to each of the free copper tape ends and to the
center tap. Spread out the stranded wires flat when
soldering to the copper tape - this makes a much
neater and less bulky connection. Connect the ends of
these paralleled wires to the output rectifiers and P.C.
boards as shown in Figure 3. Wrap a final layer of
transformer tape to hold everything together.
Step 1 - Parallel 8 strands #28 wire
I
z::::::::::::ss :
TWIST ENDS
Step 2 - Fold In half
·c. .----
A
c ....
·-----o
Step 3 - Wind on bobbin
T3 -
Current Sense Transformer
Place the windings directly on the toroid:
Secondary (2) - Wind the secondary first. Wind 7
turns of #20 enamel wire (about 10 inches long) onto
one side of the toroid (Figure 8).
Primary (1) - Form the primary by soldering 2
strands of # 16 enamel wire to the circuit board
connections (Figure 3). Run these strands through
the center of the toroid. This forms the one turn
primary. Solder the secondary into the board as shown
(Figure 8).
Step 4 - Connect C and A to pins 6 and 7
SOLDER WIRE TO THIS
SIDE AND CONNECT TO
BOBBIN PIN NO 1
Faraday Shield
Figure 5
Primary Power Transformer
Figure 4
6-66
Step 1 -
Parallel 6 strands #28 wire
A
Step 1 -
Make two insulated copper straps
--------..---
TAPE
#K
\
TWIST ENDS
Step 2 -
Fold In half
(
Step 3 -
Step 2 - Place one strap on top of the other with leads on
opposite sides.
Wind on bobbin
A
D
Step 3 - Wrap two turns of this double copper tape onto the
bobbin. Connect A to D and solder this center tap into the circuit
board. Connect the other free ends, Band C, to diodes D11 and
D12·
Step 4 - Connectthefree ends to bobbin pins #10, 12. Connect
doubled over end to pin 11.
C
(CONNECT TO
PIN 10)
5 V Secondary
Figure 7
Mount IC4 and IC5 onto their heat sinks with thermal
compound. IC5 should be insulated from the heat
sink. Use metal screws for mounting IC4 and IC5 and
use plastic screws for the other mounting screws for
the TO-66 heat sinks. Cut off the center lead of each
regulator and insert the other two pins into the circuit
board.
± 15 V Secondary
Figure 6
Output Inductors
Ll and L2 are identical. Wind one tum of # 18 enamel
wire through each core and solder to the circuit board.
Schottky rectifiers Dll and D12 mount directly on the
heat sink with thermal compound. Use star washers
for a good electrical connection when bolting rectifiers
Dll and D12 to the board.
L3 is also one tum, but use three strands of # 18 in
parallel.
L4 is an air core inductor. Close wind 10 turns of # 16
wire on a 5/16 inch diameter form.
Miscellaneous
Heat Sinks
Use star washers on both sides of the board when
mounting C5. IC sockets may be used for ICl, IC2
and Opto 1.
Mount the TO-3 heat sinks off the board with 1/4 inch
spacers (to make space for R5, R6, R8, and RI9). No
insulating washers are needed, but heat sink compound should be used.
ALWAYS use an isolation transformer when connecting an oscilloscope to look at waveforms on the primary side of the power supply.
Do not mount Q3 until after the initial test procedures.
6-67
will not change the output. No significant current
should be flowing from the high voltage supply at this
time. Check the ± 15 volt outputs for the correct voltage.
While monitoring the supply voltage on pin 12 of ICI
(using a floating voltmeter) connect a load to the 5 volt
output to draw about 1 ampere. The supply voltage on
pin 12 should increase to about 15-20 volts if the
power supply winding #2 on T2 is working correctly.
If everything works correctly so far, disconnect all
power supplies and install Q3 in the circuit board.
Using a variac or DC power supply and a floating
voltmeter (or an isolation transformer and a nonfloating meter), increase the input voltage to the line
cord to about 20 VDC or 40 Vrms while monitoring
the supply voltage on pin 12 of ICI. This voltage
should level off around 10-12 volts. Connect a
minimum load to the power supply (5 n, 5 W) and
increase the line voltage to full voltage. ICl's supply
voltage should increase to about 15-20 volts.
Current Sense Transformer
Figure 8
Power-Up Procedures
Even though this is a proven circuit board, the control
circuitry should be checked separately before powering up the complete supply. To do this, connect an
isolated + 12 volt supply through a diode between pin
12 of ICI and ground (Figure 9). Use an oscilloscope
to check the drive signals at pins 9 and 11 of ICI.
These signals should be in-phase, quasi-square waves
at a frequency of 100 kHz (Figure lOa). When these
signals check OK, look at the gate waveforms of the
MOSPOWER FETs. These waveforms should be out
of phase (Figure lOb). 07's waveform may have some
overshoot. With Q3 still out of the circuit board,
connect a variac or a high voltage DC power supply to
the AC input. Slowly increase this power supply voltage (the control circuitry is still running with the
floating 12 V supply) while monitoring the output
voltage. When the 5 volt output gets somewhere between 4.5 volts and 6.5 volts the supply should begin
regulating and further increases in the input voltage
The power supply is now ready for use. Adjust the
output voltage to 5 volts using RIO. Adjust the current
limit R24 with the ± 15 volts fully loaded and the + 5
volt output delivering about 25 amperes. A minimum
current of about 5 amperes must be drawn from the 5
volt output for the ± 15 volt outputs to be able to
deliver 1 ampere each.
The power supply may now be plugged in directly to
the power line for operation. The only requirement
necessary is to have a minimum load of about 1 ampere at all times on the + 5 volt output.
Power Supply Features and Specifications
+15 V FLOATING
SUPPLY
n
.,
~PINI2
5 Volt Output
• 20 amperes output current
• 0.2% line regulation (±20% line variation)
• 0.4% load regulation (no load to full load)
• < 100 m Vp-p ripple and noise at full load
• Output over-current protection
• :;;;0.5 mS transient response time (no load
to full load)
• Over-current protected
INPUT
GROUND
Power-Up Connection
Figure 9
15V~
I
I
I:
o
1
I I
I:
~
: I
II
~!
I:
I:
I
PIN 11
I
::
15 Volt Outputs
• 1 ampere output current each
• 0.2% line regulation
• 1.0% load regulation
• < 10 mV ripple
• Short circuit current limiting
!Hj !Hi
o-----LJ lLJJ ~9
15V
a) TL494 Drive Waveforms
o------.w!I: n11W! !t: n~
15V
15V~'!iiii
o
I"
I
"
II
VN4000A Features
• 4000 volt B VDSS
• < 1 ohm on-resistance
• < 100 nS switching times
• Rugged safe-operating area
• No secondary breakdown
Co
b) DMOS Gate Waveforms
Drive Circuit Waveforms
Figure 10
6-68
Parts List
Part #
Resistors
RI
R2. R(). R9. R12. R13. R15. RI9
R3. R7
~
R5. RS
RiO
Rll
R14. Rl7
R16. R23
RI8
R20
R21
R22
R24
Capacitors
CI
C2
C3. C14. C16. Cl7
C4
C5
C6. CS. C9
C7
CiO. Cll
Cl2
Cl3. CI9
CI5
C18. C2J. C22
CI9
C20
Integrated Circuits
ICI
IC2
IC3
Quantity
2
I
2
2
2
I
4
2
4
Description
5.1 KO ±5% 1/4 W Resistor
I KO ±5% 1/4 W Resistor
3300 ±1O% 1/2 W
33 Kn ±1O% I W
500 ±5% 1/4 W
500 O± 10% Trimpot
1.2 KO ±5% 1/4 W
100 KO ±5% 1/4 W
4700 ±5% 1/4W
20 KO ±5% 1/4 W
ISO ±1O% I W
1000 ±5% 1/4 W
ISO 0 ±5% 1/4 W
100 0 ±1O% Trimpot
Recommended Mfg.
Allen Bradley
0.001 fLf Ceramic Disc
0.05 fLf Ceramic Disc
0.1 fLf. 25 V Ceramic Disc
1.0 fLf. 25 V Electrolytic
710 fLf. 200 V Electrolytic
(320)
47 fLf. 25 V Electrolytic
1.0 fLf. 400 V TRW-35
2 fLf. 25 V Tantalum
200 fLf Mica
220 fL f. 10 V Tantalum
5 pf Mica or Ceramic
0.01 fLf. 5 K V Ceramic
10 fLf Electrolytic
6.S fLf Electrolytic
TRW
TlA94 PWM IC
CA3l30 Op-Amp
TlA30 Voltage Reference
Texas Instruments
RCA
Texas Instruments
6-69
Sprague
Mallory 227KOIOPIG
Parts List (continued)
Part #
Quantity
Integrated Circuits (continued)
IC4
IC5
Description
Recommended Mfg.
/LA7815UC+15 V Regulator
(JI,A7812UC+12 V)
/LA7915UC-15 V Regulator
(JI,A7912-12 V)
Fairchild
Fairchild
Dlodes/Rectlfters
D" D3 - D6, D14, DI5
D2
D? - DIO
Dl1, DI2
DI3
D14
Transistors
6
I
4
2
4
IN4148
IN4107
IN5804
IN5831
IN5406
IN5360
Diode
Zener
Fast Recovery
Schottky
Rectifier
Zener
Motorola
Motorola
Unitrode
Unitrode
Motorola
Motorola
QI,Q4
Q2,Q5
Q3
Q6,Q7
Ferrites & Accessories
TI
TI Bobbin
T2
T2 Bobbin
T3
LI, L2
L3
Miscellaneous
Optol
THI
FI
TO-3 Heat Sink
TO-220 Heat Sink
TO-3 P .C. Sockets
Dl1, DI2 Heat Sink
Output Banana Jacks
3-Wire Line Cord
Fuse Block for FI
Transformer Tape
Copper Foil
2
2
2N4400
2N4403
MPSUIO
VN4000A MOSPOWER FET
Motorola
Motorola
Motorola
Sihconix
F1146-1-06 Pot Core
B475-1
IR8031-1
B680-2
BBR7727-1 Toroid
F1146-I-TC9-315
F2037-J-TC9
Indiana General
Indiana General
Indiana General
Indiana General
Indiana General
Indiana General
Indiana General
Hl1AI Opto-Isolator
3D304 Thermistor
2 A Fast Blow Fuse
LAT03B5CB
LAD66A4CB
LSG-3DG2-1
E240-001
G_E.
Midwest Components, Inc.
Buss
!ERC
!ERC
Augat
IERC
Type 10 Epoxy Film
Type 1194
3M
3M
2
2
2
2
4
5
6-70
6.2 Solid State RF Generators for Induction
Heating Applications
©
1982 IEEE. Reprinted, with pennission, from Conference Record Industry Applications Society IEEE lAS - Annual Meeting, October 4-7, 1982, San Francisco, CA, pp. 939-944.
Abstract
Radio frequency power for induction heating has
traditionally been generated by vacuum tube oscillators. These generators use high voltage tubes and
generally operate in the frequency range of 100-500
kHz, 1-400 kW power output at approximately
50-60% efficiency. Although thyristor SCRs and bipolar transistors are marketed at this time, they are
unsuitable due to either slow switching or low power
performance. Until just recently, solid state devices
with both high-frequency and high-power capabilities
were not available.
times. For surface heating and non-magnetic parts,
high frequency power is required. Therefore, the
power source must be capable of supplying a
minimum of several kilowatts of power in the radio
frequency range. In addition, the power supply must
have adequate control, protection and monitoring circuits.
A new power semiconductor - the power MOSFET - a highspeed field effect transistor, can now
meet these criteria. The power MOSFET switching
time is in the order of 50-100 nanoseconds and can
generate many kilowatts of power at frequencies to
500 kHz.
Over the past 50 years, the high voltage vacuum tube
was the only device available for radio frequency
power generation. Semiconductors, such as SCRs and
bipolar and Darlington transistors, have appeared in
the last decade. These components work well in the
frequency range up to 10 kHz, but they do not have the
characteristics that are required for a radio frequency
power source necessary for induction heating applications. In addition, these solid state devices are limited
by slow switching speeds and/or low power ratings.
This paper describes a solid state RF generator using
MOSFET transistors for the power semiconductors.
The RF generator is a load tracking, resonant inverter
capable offull power output over a frequency range of
100 kHz to 500 kHz. Output power levels are in the
kilowatt range with induction coil KVAs to 200 KVA
(400 V and 500 A) and more. This solid state RF
generator has similar characteristics to the more
familiar low frequency power converters presently
used in the industry; namely, all solid state, high
efficiency (approaching 90%), small and compact,
and no vacuum tubes or moving parts.
A new switching device, the power field effect transistor, is capable of both higher power outputs and
significantly higher switching speeds. Some various . '
market names for these transistors are VMOS, MOSFET, HEXFET and DMOS (these are the manufacturer's distinct trade names). The availability of such
components have been disclosed and described in increasing detail since 1979 or so [1, 2, 3, 4]. As a
result, Westinghouse Electric Corporation has developed and manufactured a solid state, radio frequency
power supply using the power MOSFET in the inverter circuit design of the RF generator.
Introduction
Power MOSFET Transistor
The art of induction heating is to unifonnly heat
metallic workpieces at specified temperature and
The power MOSFET is a field effect transistor with
ratings presently ranging from 50 V/60 A to 1000 VI
6-71
3 A among the various manufacturers. Figure I
depicts an N channel and a P channel MOSFET transistor showing polarity of the terminal potentials and
current flow for normal transistor action. The MOSFET contains an internal, reverse P-N junction diode
(shown dotted in Figure I) which has the same current
rating as the transistor. 5 This diode is a viable circuit
element that can be used, just as any externally connected discrete diode, for a reverse current path
around the transistor.
D(+)
example, a MOSFET having an input capacitance
Ciss = 1000 pf, passes a gate current of I A when
driven with a gate-source voltage rise time of 20 V/20
ns. At RF frequencies, these drive requirements can
become formidable.
A limiting factor with the MOSFET is the recovery
time of the reverse diode. This internal diode is relatively slow, having a recovery time which can be in
the order of 200 ns or more compared to FET transistor switching times of 20-100 ns. These differences
in speeds must be recognized and dealt with in certain
switching circuits.
D{-)
._ IJ-* I- .~J-* 1'"
' '19--"
S(-)
N Channel
"'I9--
J
S(+)
P Channel
VDe
INPUT
Power MOSFET Symbol Shown with Internal Diode &
Polarities of Voltage & Current
Figure 1
The MOSFET transistor is a voltage controlled device
with an insulated gate (G) which controls current flow
between the drain (D) and source (S) terminals. A
positive gate-source potential applied to the N channel
MOSFET causes current to flow from the drain to the
source. For the P channel MOSFET, the opposite is
true: a negative gate-source potential causes current
flow from the source to the drain. The drain current
can be controlled linearly by the gate-source voltage or
can be "switched on" by gate . overdrive to a level
determined by the impedance of the external circuit.
With sufficient gate overdrive, the drain-to-source
characteristic appears resistive with "ON" resistances
for high current MOSFETs being less than an ohm.
This "ON" resistance has a positive temperature
coefficient which forces current sharing among paralleled devices, eliminating the need for external current balancing components and making paralleling
relatively easy.
e.g
=RFT
HEATED
_WORK
PIECE
L.
Basic RF Inverter
Figure 2
Solid State RF Generator
A. Basic RF Inverter
Figure 2 shows the basic circuit of a I kW resonant
inverter developed for induction heating. MOSFET
transistors QI-Q4 are configured as a full bridge,
voltage fed inverter with the internal diode of
each providing the return path to the DC source for
reactive currents. For higher powers, paralleled
MOSFET transistors are used; for example, four paralleled 6 A/450 V devices in each leg for a 3 kW
rating. The DC terminals of the inverter are tightly
coupled to RF bypass capacitor (RFBC) whose capacitance is sufficient to pass the AC component of
inverter input without substantially changing its DC
potential. The AC terminals of the inverter drive the
RF load circuit which is essentially a high Q series
resonant circuit formed by tuning capacitor. Ct. and
induction load coil, Lt. Transformer RFT matches
the load impedance to the VA capability of the inverter. while coupling capacitor RFCC prevents any
DC current from flowing in the primary winding and
Drain-source switching speeds can be very fast, as low
as 20-100 nanoseconds, depending on drive circuit
source impedance and gate capacitance. MOSFET
transistors do not exhibit minority carrier storage
effects as do bipolar transistors and switching times
are determined primarily by how fast the gate capacitance can be charged and discharged. The fast
switching speeds result in a low and thermally manageable switching energy loss per cycle which is requisite for RF operation.
The steady-state gate-source input impedance is very
high, being essentially capacitive. But dynamically
the gate must be sourced and sinked by a high
current source to produce fast switching speeds. For
6·72
Thus, MOSFET diodes I & 2 are conducting when
MOSFET transistors I & 2 are turned on eliminating
short circuit recovery problems.
saturating the core. These two components are sized
for minimal effect on the resonant circuit parameters,
namely:
RFCC
»(~~y
Ct
Figure 3 shows "light load" inverter waveforms resulting from driving of the series load at a frequency
(1)
I
Leakage Inductance of RFf
«(~~y
f> > 27T "LtCt
Lt
(2)
01,a2
I
Eo
Primary Inductance ofRFT » - - x -1- (3)
wmin
0
EGS
Output
Where:
Np
E&I
Primary-secondary turns ratio of RFT
FETI's
Eo
Rated RMS output voltage of inverter
Diode "s
10
Rated RMS output current of inverter
Input I
wmin
Minimum output frequency of inverter
Devices
Ns
oj
Conducting
In operation, the MOSFET transistors are switched as
diagonal pairs: Q I & Q2 alternating each half cycle
with Q3 & Q4 to provide a square wave voltage output
at the AC terminals of the inverter. The waveform of
the output current depends on the inverter output frequency which is the switching rate of thc MOSFET
transistors. Driving the series resonant load off
resonance - i.e. at a MOSFET switching frequency
differing from the natural resonant frequency of Ct
and Lt - results in low output current, while driving
it at resonance results in maximum power to the load
coil. In fact the output current is controlled in a
closed loop by varying the driving frequency.
Diode
FET
Diode
FET
1,2
1,2
3,4
3,4
Inverter Waveforms for Light Load
Figure 3
The output current takes the form of an "inductive
ramp" with positive current from the DC source
flowing through MOSFET transistors and a nearly
equal negative current being returned to the source by
the MOSFET diodes. The net DC current is nearly
zero and very little power is delivered to the highly
reactive load (PF = 0). Figure 4 shows the waveforms
Q1,D2
The differences in switching speeds between the
MOSFET transistor and its relatively slow reverse diode
imposes an operational limitation on the inverter. Had MOSFET diodes 3 & 4 been conducting
at the time MOSFET transistors QI and Q2 were
switched on, they would have failed to recover in time
to prevent short circuit currents from flowing between
the positive bus and negative bus through the oncoming transistors and offgoing diodes. Commutation of
current from a diode to a FET in adjacent legs of the
inverter occurs whenever the output current leads the
output voltage. One simple solution to the problem is
to avoid leading currents and assure that the output
current always lags the output voltage (lagging power
factor). This is accomplished by always gating the
MOSFETs at a frequency greater than the resonant
frequency of Ct and Lt. With such a lagging power
factor, current is always commutated from "its own
diode" to the MOSFET transistor being gated on.
EGS
Output
E&I
oj
oj/,/!'>b
I
FETl's
F
I
o
I
1
I
I
;:
I I
I I
: /
V
I
-.;;:::::;;>
/
I
..... ,1
1
I
Diode "s
I
Input I
Conductmg
DeViceS
Inverter Waveforms for Heavy Load
Figure 4
6-73
are in the MOSFET "tum on" charging path being
dimensioned for a circuit Q = 1. The "tum off"
discharge path includes only R2 & R3 (DI shorts Rl)
increasing the circuit Q = 2 resulting in a faster discharge than charge of the input capacitance. This provides an asymmetrical gate drive signal with relatively
slower tum on time and relatively faster tum off time.
The result as shown in Figure 6 is that the "off going
MOSFET" reaches its gate threshold voltage before
the "on coming MOSFET" attains its threshold, providing a gating dead time of about 50 ns for the pair.
Figure 7 shows G-S tum on tracking waveforms of
diagonally fired MOSFETs.
for the same series load being driven at nearly resonance by decreasing the inverter output frequency.
The output current is a sinusoid with a high amplitude.
In this situation a mostly positive current flows from
the DC source through the MOSFET transistors with
very little negative current being returned by the
MOSFET diodes. The net DC current is large and
"heavy power" is delivered to the highly resistive
load (PF = I). In both extremes, the commutation of
current from a MOSFET diode to "its own MOSFET
transistor" is illustrated.
B. Gate Drive Circuits
The gate drive circuitry shown in Figure 5 is designed
to perform the following functions:
Since the gate drive is an alternating source, a significant negative potential (see Figure 6) is on the gate
of the "off-going MOSFET" when the "on coming
MOSFET" begins to switch. Thus, transient D-G
currents caused by switching of the adjacent MOSFET
should not cause the "off going" MOSFET to tum
back on spuriously.
I. Charge and discharge the input capacitances
(Ciss) of the inverter MOSFETs for drain-to-source
switching times of 100 ns or less.
2. Provide a dead time between tum on and tum off
of alternate firing MOSFETs serially connected in
one pole of the inverter to prevent short circuit
"shoot through" currents.
3. Provide a low impedance to the gate and source of
each MOSFET during steady state inverting operation to prevent possible misfiring by the switching
of the adjacent device. [6]
4. Prevent misfiring of the MOSFETs during shutdown.
-
-
--
1'\.
\
03
-
-
. . . .v
-
--
-
I POLE OF
r---------~
02
--
ir"""'
Lx'
--
-
-- [,(
V V X'
:=f ./
--
V
03
~ -- -
-
-\ - -I=-
'- I--f-"
1
10 Vldlv
50 nsld, ...
Photograph IS double
exposed to show on
and oNgOing dead
times for the deVices
\
02
INVERTER
OTHER PARALLELED Q1
GATE CIRCUITS
-
-
" r--
+ DC
MOSFET G-S Turn On-TUrn Off Waveforms
for Alternate Firing Devices
Figure 6
DRIVER
01
04
-
~
AC
o
-
OTHER PARALLEL!::D 04
GATE CIRCUITS
, _________
-DC
L
03
-
--
~
04
-~
-
--
04
I
I
--
~/
/-
-
-
-
~
?""
-
-
-
~ ;;::::::;
~
10 Vidlv
50 nsidlv
~
....::~.\
-f~ - - --
03
--
Photograph IS double
exposed to show on and
offgorng dead limes for
the deVices
'-.:==---
_
MOSFET G·S Turn On-Turn Off Waveforms
for Series Fired Devices
Figure 7
Basic Gate Drive Circuits
Figure 5
MOSFETs Q9 and QIO and transformer DT form a
push-pull driver with four isolated secondaries for
gating the inverter power devices. The turns ratio from
'h primary to each secondary is I: 1 being specially
wound for low leakage inductance in the order of 0.4
fLH. This unavoidable transformer leakage inductance
plus stray inductance in the same order of magnitUde
as the leakage forms a RLC circuit to charge the input
capacitance of the MOSFET. Resistors RI, R2 & R3
If the driver were stopped by removing pulses from
driver MOSFET transistors Q9 and QIO, the current
established in the primary of DT would freewheel
through the other 'h primary and reverse diode of the
device that was off. This would result in possible
unscheduled gating of the inverter MOSFETs until
the energy in DT transformer was dissipated and then
result in an open circuited G-S drive connection thereafter. If the tuned load on the inverter was still ringing
6-74
3,',,60HZ
DCFU
II
DCCT
i--1::::I--
::~Bi1~Eo~
RFBC
G1
GATING PULSES
G2
G3
G4
-----"I
r-----,
F==l
I
1-,
I "I
04
L __ ---1r---1L-________ ----l
I
I
Q2
1==1
ciiJE ~ G2
RBO:RD~
L _____ ____ ---l
S2
L __ -.l
t~----+-----<-~tt4---:=======~===1!::===~--1
-::-
'------------0- 11 RFCT
CONTROL CARD
RFCC
Basic RF Generator
Figure 8
down through the inverter MOSFET diodes to RF
bypass capacitor, the switching of the diodes - Le_,
rapid charge of D-S voltage - might now gate on
the open circuited adjacent MOSFET causing a "shoot
through" and failure. To alleviate this condition,
driver MOSFETs Q9 and QIO are both turned on at
the STOP command while series MOSFET Qs is
turned off. Thus, the driver transformer is shorted at
turn off and secondary drive voltage is reduced to
nearly zero immediately. This presents a low impedance to the G-S drive circuits diverting any transient
D-G currents from the gate of inverter MOSFET and
preventing abnormal tum on.
inverter powered from 3 phase than from single phase
given the same voltage stress on the transistors.
Radio frequency filtering by RFL choke and RFFC
feed through capacitor minimizes RF feedback (at
twice the output frequency) to the AC power lines.
The DC surge suppressor holds down any surge voltage across bypass capacitor RFBC during the following situations:
1. Turn on of the main circuit breaker and LC resonant charging of RFBC.
2. Turn off of the inverter gating under heavy load
and transfer of energy from the RFL choke to
RFBC.
3. Tum off of the inverter gating under light load and
transfer of energy from the high Q output circuit to
RFBC.
C. KW RF Generator Circuits
The basic schematic of the overall 3 KW RF generator
is shown on Figure 8. A 3 phase full wave diode
rectifier is used to effectively utilize the voltage rating
of the MOSFET transistors in the inverter. This is so
because a single phase full wave rectifier would stress
the inverter transistors with a 1.57/1 peak to average
voltage (about 393 V peak for a 250 V DC supply)
unless an elaborate DC filtering and lightload bleeder
were used. The peak to average stress with the 250
V DC 3 phase supply is 1.049/1 or only 263 V peak.
Since induction heating loads rarely have waveform
modulation restrictions (especially when 3 phase diode
rectifiers are used) filtering is quite unnecessary.
Thus, about 50% more power can be handled by an
The inverter bridge has four paralleled MOSFET
transistors per leg mounted on a common water cooled
heat sink. The gate circuits of each transistor is as
shown in Figure 5. Four such circuits are mounted on
a FET drive board and assembled to the heat sink for
each leg. Paralleling of the 'four gate drive circuits is
done by one set of leads from the appropriate driver
transformer secondary. Each inverter leg assembly
contains a snubber network (RN, CN), tranzorb voltage suppressor network (VZ, DZ) and DC equalizing
resistor (RO).
6-75
-
I
D. Control
The four inverter leg assemblies (QI-Q4) are interconnected in a symmetrical layout by multiple sandwiched, flat bus, such that equal and opposite
currents are flowing to minimize inductances. This
low inductance technique is used from RFBC bypass
capacitor, through each drain-source tie, to the RF
output. The total inductance from DC input to RF
output is in the order of 0.3 /-tH.
Basic control requirements for the RF generator are:
1. Control the RF output current (and necessarily the
output power) by a closed loop system which
effects changes in output frequency to excite the
series load nearer to or further from its natural
resonance.
2. Detect the phase angle (rp) of output current relative to output voltage and "switch in" limiting
phase control to maintain a worst case lagging
phase nearly zero. This is requisite for two situations: (1) light load where the "asked for current"
by the reference is higher than can be achieved at
resonance. This would result in an out-of-control
situation with the frequency sliding past resonance
to the low frequency limit. (2) to prevent improper
commutation of current from the slow reverse connected MOSFET diodes to the fast transistors.
3. Detect voltage on the series tuning capacitor (Ct>
and "switch in" limiting voltage control where
excessive potential would occur as with higher than
expected circuit Q.
4. Provide high frequency and low frequency limiting.
Three RF current transformers provide control and
protection signals to the controller:
• RFCT provides a closed loop feedback signal for
control of the RF output current and for load overcurrent trip.
• rp CT provides a signal for output phase measurement and limiting.
• DCCT provides a signal for instantaneous trip of the
gating for an inverter fault.
The output rating of the 3 KW RF generator is nominally a 250 V peak square wave at 18.7 Arms
(sinusoid). The output voltage remains nearly constant, except for 60 Hz line variations, whereas the
current will vary in amplitude as required. The fundamental of the square wave (E 1 = .9 x 250 = 225
Vrms) is the exciting voltage for the series resonant
load circuit. RF induction load coils exhibit Q's in the
order of 20-60. Thus quite high potentials (4500
V-13,500 V ideally) would be developed across a
coil and tuning capacitor connected directly to the RF
output terminals when driven at the resonant frequency for maximum power. In general, RF load coils
. at 3 kW or so, can have much less impedance than
4500 V/18.7 A = 240 ohm and generally require
impedance matching.
Figure 9 shows the basic control scheme which implements the aforementioned requirements. It is a
straightforward closed loop system with the operator's
control potentiometer PTI providing the current reference Ic and RFCT/RECI providing the current feedback signal Ifb to OAI error integrator. The output of
OAI decreases the frequency of pulses from the VCO
(which is twice the output frequency) until zero error
is achieved; i.e. the magnitude of output current is
satisfied. Pulses from the VCO clock the timing flipflop whose Q and Q outputs drive OR gates ICI2A,B
and ICI2C,D alternately at lh the VCO frequency.
The OR gates drive emitter follower EFTI and EFT2
which, in turn, provide low impedance gating signals
to the driver N channel MOSFETs Q9 and QIO.
The usual RF matching transformer is air core with
considerable exciting current and poor primary to secondary coupling. Such a transformer is not suitable for
use with an inverter because of the reactive current and
voltage requirements.
Rise time and fall time at the gates of Q9 and QIO are
about 10 ns with drain-source switching times in the
order of 20 ns. The ICI2A,B and ICI2C,D paralleled
pairs of OR gates are paralleled within one quad IC to
increase current sinking and sourcing to enhance their
switching times. All logic elements are CMOS operated at 15 volts. This allows direct interfacing with
bipolar and FET transistors in the controller without
need for level conversion.
A universal, core type transformer (RFT) was developed to suit a 54:1 load impedance matching range. 7
Split primary and secondary windings on a 2 mil strip
wound core are seriesed or paralleled to provide output
voltages and currents from 5.1 V/823 A to 37.5/112 A
in approximately 15 steps. Thus induction coils with
impedances ranging from 102 V/823 A to 2250 V/112
A (20 :;;; Q :;;; 60) can be matched at the 3 kW level.
The transformer is designed to operate from 100kHz
to 500 kHz with a maximum primary exciting current
of 1 amp and maximum loss of 94 watts.
A high STOP or FAULT signal from the sequencing
and protective logic circuits simultaneously make the
6-76
C33
q.
PHASE SHIFT
CONTROL
REC2
R42
DT
+15 V
(INTERRupn
GATE DRIVE
PT'
REC 1
Control System
Figure 9
outputs of OR gates IC12A,B and IC12C,D a high.
Driver MOSFETs Q9 and QlO are both turned on
shorting the primary of the driver transformer (DT)
for low impedance stopping of the inverter gate drive
pulses. At the same time the -15 V gate drive signal
to the P channel series clamp MOSFET Qs is
removed, disconnecting the drive power supply from
the shorted driver.
Output
Input
VI
Vdl
Vep
0
0
0
1
1
0
0
I
1
0
1
Series capacitor voltage limiting occurs when the voltage sensed by transformer CPT and rectified by REC2
tends to exceed the maximum level desired. Capacitor
voltage limit pot PT3 sets the maximum level at Vc* .
When rectified voltage signal Vc is less than Vc* ,
integrator OA2 has a negative output back biasing
DI5 and disconnecting OA2 from summing node SI
to OAI. Should Vc tend to exceed Vc" , the output of
OA2 goes positive and is connected to OA1 summing
node. This constitutes additional negative feedback to
OAI which "makes up" the existing difference between Ic* and 1tb limiting the VCO frequency from
further reduction and holding the capacitor voltage
at the limit level.
1
Thus, an output occurs from the EXCLUSIVE NOR
gate whenever two 1's or two O's overlap, resulting in
an average output voltage versus phase of the two
square waves as shown in Figure 11. This "phase
output" after processing by OA5 integration, OA6
inverter and diode D16 provides negative feedback
to SI summing node of OAI whenever it exceeds the
minimum desired phase determined by the setting of
PT2 potentiometer. The VCO frequency is limited
from further reduction holding the phase angle between voltage and current at the limit level.
The interesting part of the phase detector lies in the
development of the VI and Vdl input signals to the
EXCLUSIVE NOR gate IC3C.
The phase detector circuitry is shown in Figure 10.
Basic phase detection occurs with the EXCLUSIVE
NOR gate IC3C. The inputs are square waves representative of the RF output current (VI) and RF
output voltage (Vdl). The truth table for EXCLUSIVE
NOR gate IC3C is as follows.
First it should be recognized that the phase characteristic shown in Figure 11 is really not suitable for
stable closed loop control. This is because the phase
6-77
-
I
C2
+15V
"31
OCT
I~II
01.
PHASE SHIFT
CONTROL
Vd
Phase Detector
Figure 10
(15 VOL.TS)
---:-,'---+-*-1-'-''-7----1--(75 VOLTS)
90
180
~4---f~-~~~~~--_r--:~~
36.
270'
90
90
270
Average Output of EXCLUSIVE NOR Phase Detector
Figure 11
Average Output of EXCLUSIVE NOR Phase Detector
with 90· Lag
Figure 12
detector average output voltage does not monotonically increase with variation of cp from 90° lag to
90° lead. At cp = 0°, the slope changes and the
average output decreases for further increasing leading
phase. Therefore, detection at cp = 0° is hazardous
since the phase control loop would become unstable
with each incremental increase in cp incrementally
decreasing the error signal to OA5. This results in
increasing error to OAI which necessarily drives the
frequency of the VCO to its lower limit giving nil
current output at cp = 90° lead. The solution to this
problem is to phase shift the current 90° lagging and
change the characteristic to that shown in Figure 12.
Thus, at cp = 0°, should the phase tend to go leading,
the phase error would tend to increase resulting in
more (not less) feedback to hold the phase at the
desired level. This is accomplished by wideband amplifier OA3 connected as an integrator. OA3 integrates
the current signal from cpCT which is squared by ZD
clamp network into a ramp (see Figure 13). The
integrated current waveform from OA3 is AC coupled
to comparator OA4 which generates a square wave
at 90° lagging (almost) from the original current signal. The lagging square wave is further "squared
up" by EXCLUSIVE NOR gate IC3A into VI.
The second problem is in the "almost" 90° lag generated by OA4 comparator. It happens that there is a
constant, added delay time (t) of about 200 ns between
the actual RF current and output of OA4 - mostly
caused by OA4 (an LM211 high speed comparator).
This fixed time delay is formidable since it represents
a variable phase angle delay of 7.2°_36° when the
frequency is varied from 100 kHz to 500 kHz. The
problem is further compounded by a constant lead
time (t') of Vd voltage signal from IC12A,B which is
a digital representation of the RF output voltage. Vd
leads the actual RF output voltage by about 300 ns.
6·78
OUTPUT
CURRENT
SQUARED
CURRENT
SIGNAL
INTEGRATED
CURRENT
SIGNAL
=1~VZD~~I
H--I'----+--+-++"'+--+-f-+---+-I--H
==-J
01
50 Vldlv
2 AJdlv
~r-f+-+-t--H"-+-"-IIIr+--H
EOC '" 242 V
IOC=134A
HII---+-+--+-++-+--+-f--HIII+--H
--
,:...............
f '" 247 kHz
...............
vI TPl
OELAYED
CURRENT
SIGNAL
MOSFET Voltage and Current Waveforms
Figure 14
OUTPUT
VOLTAGE
SOOns
Vd
LOGIC GATE
VOLTAGE
SIGNAL
vdl
OELAYED
VOLTAGE
SIGNAL
,l
,,
IrTP:!2_ _~
"'",
,
GIn
: TP8
PHASE
DETECTOR
OUTPUT
-
~
,,
V,I, ~jVERAIGE VALUE)
o
I
..
'\
\
Phase Detector Waveforms
Figure 13
IOC=137A
f= 226 kHz
~
~~
I
The solution to this problem is to make the voltage
signal Vdl lag the actual RF output voltage by the
same fixed time as the current signal VI lags the
output current. Since the EXCLUSIVE NOR phase
detector measures the relative phase of VI and Vdl, the
absolute phase of each is unimportant as long as they
are equal. To accomplish this, delay line DLl and
adjustable vernier capacitor C50 on the output of buffer gate IC24D provide most of the required time delay
(t' + t). The remaining is provided by the propagation
times of the squaring gate IC3B and paralleled delay
line buffers IC24A,B,C.
100V/dlV
10A/d1Y
EOC = 225 V
/
RF Output Voltage and Current Waveforms
Figure 15
E. Test Results
The RF generator was tested from a 1 kW output
level using 4 HPWR-6501 MOSFETs to over 3 kW
using 16 MOSFETs, 4 paralleled per leg. Some pertinent data and statements about those results follow:
Solid State 3 KW, 100 kHz-450 kHz RF Generator
Figure 16
1. Waveforms of voltage and current (see Figures 14
and 15) were as expected, except that MHz ringing
at current commutations were significant. Worst
case D-S ringing occurred off resonance with
maximum EDS = 370 V at 250 V DC.
2. Current sharing among the paralleled MOSFETs
was excellent with a worst case unbalance of
13.5% from the mean and a typical unbalance of
less than 10% measured. The only precaution taken
for paralleling was symmetry in circuit layouts and
matching of gate threshold voltages (Vt) as follows:
3. Over 3 KW of output power was achieved with a _
typical set of recorded data tabulated below:
Edc
Ide Eo1
101
Volts Amps Volts Amps
DC
DC ons
rms
236
16.6
198
kHz
18.74 387.2
Degrees
Lt
=
30 p,H Ct
=
3466
20.9
(1) Above data for RLC load -
Vt within 1,2 V for devices in one inverter leg.
Vt within 1,2 V for groups in alternate firing legs.
Vt within 1 V for groups in diagonal firing legs.
DC·RF
Po
Efficiency
Watts
%
(Calculated) (Calculated)
88.5
no transformer
0.0055 uf R
=
9.87 ohm
(2) MOSFET T03 case temperature rise above
inlet water < 5°C
6-79
I
Conclusions
References
The objective of this project was to investigate the
feasibility of using power MOSFET transistors for RF
power conversion and to achieve at least 1 KW output
at 450 kHz. A resonant inverter approach was chosen
using the highest rated available MOSFETs. We were
learning how to use the high powered MOSFET at
about the sarne time that semiconductor manufacturers
were gathering data for their applications. Our overall
objective was to develop basic techniques to eventually achieve tens of kilowatts as higher nited devices
became available.
[1]
[2]
[3]
[ 4]
A 1 KW model and a 3 KW production unit were built
that met the goals of power, frequency and efficiency.
Significant ancillary developments were megahertz
CMOS controls, nanosecond risetime drivers, ironcore RF load matching transformers and circuit
measuring techniques for nanosecond transitions.
[5]
[6]
More work needs to be done to "harden" the design
for rugged field applications. The low surge current
capability of the MOSFET (like the bipolar transistor,
but unlike the SCR) makes it difficult to protect even
with high speed gate suppression. We would recommend that compatible protective devices and/or techniques be developed to match the speed and surge
capability of the MOSFET. We recommend also, that
semiconductor manufacturers develop higher rated
MOSFETs in more suitable packages for higher power
designs.
[7]
Acknowledgments
The authors wish to acknowledge the assistance of
Dave Hoffman, Siliconix; Victor Li, Hewlett-Packard;
and Brian Pelly, International Rectifier, in application of the power MOSFET to this project. We
are indebted to Reuben Lee, retired from Westinghouse, for design of the RF matching transformer and
for consultative help in design of other magnetics.
6-80
B. R. Pelly, "Applying International Rectifier's
Power MOSFETs," IR Application Note AN930.
D. Hoffman, "VMOS - Key to the Advancement of SMPS Technology," Power Conversion
International - March/April, 1980, Volume 6,
Number 2, PP 37-42.
S. Davis, "Switching-Supply Frequency to Rise;
Power FETs Challenge Bipolars," EDN, January 20, 1979, PP 44-50.
R. Severns, "The Power MOSFET, A Breakthrough in Power Device Technology," Intersil
Application Bulletin A033.
W. Fragle, B.R. Pelly, B. Smith; "The HEXFET's Integral Reverse Rectifier - A Hidden
Bonus for the Circuit Designer," Power Conversion International - March/April, 1980, Volume
6, Number 2, PP 17-36.
S. Clemente, "Gate Drive Characteristics and
Requirements for Power HEXFET's," IR Application Note AN-937.
W.E. Frank, Reuben Lee; "New Induction
Heating Transformers," IEEE Transactions on
Magnetics presented at the 3rd Joint IntermagMagnetism and Magnetic Materials Conference;
Montreal, Canada, July 20-23, 1982.
6.3 Using Power MOSFET Transistors to Interface
from Ie Logic to High Power Loads
(AN79-6)
Topics considered are (1) the nature of the load, (2)
general driving requirements of MOSFETs, and (3)
the output characteristics of the logic element.
on tum-off, whereas capacitive-like loads cause power
surges on tum-on. The power waveform should be
obtained, a suitable rectangular model derived, and
peak junction temperature computed using techniques
discussed elsewhere [1,2]. This calculation is especially important for incandescent lamp and motor
loads as their current surges last from tens to
thousands of milliseconds which causes a significant
surge in the temperature of the MOSFET power
driver.
Load Considerations
Inductive Loads
Freedom from second breakdown limitations makes
driving highly inductive or capacitive loads a natural
application for MOSFETs. Inductive loads include
transformers, solenoids or relays. High current inrush
loads such as incandescent lamps, pulse forming networks, and motors also are generally handled easily.
Some attention must be given to the load characteristics, however.
Usually with inductive loads the peak voltage spike
should be limited to a value below the breakdown
rating of the transistor. Three techniques are commonly employed: free-wheeling diodes, peak clipping
and snubbing. Typical circuits are shown in Figure 1.
Power MOSFETS are an ideal element to interface
power loads to integrated circuit logic. Although circuit design is simple, there are a few rules and precautions to observe in order to minimize power
dissipation and to have reliable operation, which are
not obvious at first glance.
The spikes caused by most electromechanical inductive loads such as solenoids or relays are effectively
handled by the free-wheeling diode in part (a). The ~
low impedance of the diode usually causes the current ~
to have a long decay time, however, which may be
I
intolerable in some applications. Speed may be traded
for overshoot voltage, by using a resistance, R, in
series with the diode [3].
In common with bipolar semiconductor devices,
Power MOSFETS can be damaged if their voltage
ratings are exceeded. Although their avalanche energy
capability is much better than that of bipolar transIstors, it is not good design practice to have the MOSFET absorb inductive energy unless the part is rated
for this type of service. The spikes generated from
inductive loads may have tremendous energy content
and usually some means of limiting their amplitude
must be provided.
The free-wheeling diode may be an inexpensive rectifier such as the IN4002. However, junction rectifiers
do exhibit a turn-on transient which may allow excessive overshoot if the MOSFET is being driven off
rapidly. For high speed switching, a Schottky or low
voltage ion-implant rectifier is required; the ordinary
fast recovery rectifier does not have a fast tum-on
time.
In addition the transient power generated during the
tum-on and tum-off intervals must be determined in
order to check for excessive channel temperatures.
Highly inductive loads may generate significant power
6-81
v·
r
.,
I
I
I
LOAD:
:
I
I
I
I
I
L __ J
RIOPT)
BVoss > Vz > v+
(a) Free-Wheeling
Diode Clipper
(b) Zener Diode Clipper
(c) R-C Snubbing
Methods of Limiting Inductive Spikes
Figure 1
Often the safest and least expensive limiting technique
is to use a zener diode as shown in part (b). The zener
responds in picoseconds and can protect the MOSFET
from supply transients as well as the inductive spike;
consequently, zener limiting is particularly attractive
on raw power buses. In a manner similar to using a
resistor in series with a free-wheeling diode, faster
decay of load current is achieved by clipping at a level
above the supply voltage.
Capacitive and High Inrush Loads
Usually no auxiliary circuitry is required with
capacitive-like loads. Although the MOSFET has no
failure mode akin to secondary breakdown, it is necessary to observe the safe area curves of the MOSFET in
order to avoid excessive temperature excursion during
current inrush. When inrush power is excessive, usually increasing the gate drive will reduce it and may
hold it within bounds. A lamp circuit and waveforms
are shown in Figure 2.
The R-C snubber is commonly used in power conversion circuits to limit spikes caused by transformer
leakage inductance and wiring inductance. It also reduces power dissipation by shaping the load line to
appear more resistive. Resistor R, in series with the
capacitor, is required to limit the current surge on
tum-on (a good idea even when MOSFETs are used)
and to ensure that the circuit is adequately damped.
Since the circuit is basically a resonant tank, it will
exhibit a damped oscillation unless the circuit Q is 1/2
or less. Values are usually empirically determined.
The peak voltage across the network will not exceed
that calculated using the energy relationship: 1/2
U 2 = 1/2 CV2. Solving for the voltage, it is found
that
V= I
Figure 2.1 illustrates the typical transient current,
voltage and transistor dissipation of an incandescent
lamp load being driven by a MOSFET circuit that is
not drive limited - it can supply all the current that
the load demands while maintaining operation in the
ohmic region. Under these conditions, the output voltage quickly swings from its off state (VI) to on state
in a time dictated by the transient gate drive current
and the MOSFET's capacitances. The peak current
is a function of the lamp's cold resistance and
decays quasi-exponentially as the lamp filament
heats and resistance increases.
The transistor dissipation waveform is similar to the
power waveform of Figure 2.1(c); this can be equated
to the rectangular power pulse of 2.1(d) to simplify
peak and average power calculations.
VUc
The resonant frequency can be calculated from the
usual relationship and R selected so that Q ~ 1/2 by
using
The case where the power transistor is drive limited
(where it cannot supply all of the peak current that the
load demands) usually but not always - results
in greater transistor dissipation. This is illustrated in
the transient waveforms of Figure 2.2 where the peak
current is much less than in the previous example, but
also where transistor operation does not enter the
ohmic region initially during the switching transition.
The resulting power dissipation pulse is greater and
may be destructive.
R = 47TfL
The equations and experience indicate that larger values of C lower the peak voltage and resonant frequency and consequently the resistor R also must be
reduced. An optimum value exists for a given L-C
combination which results in minimum overshoot.
Another consideration is to minimize the power dissipation in the transistor; various techniques are discussed elsewhere [ 4,5] .
MOSFETs make ideal drivers for incandescent lamps
because they can handle high current surges without
6·82
+V,
v
oJ
f
'PK
PEAK
CURRENT
'PK
I.)
'a
+V'
~, ~.o
OUTPUT
VOLTAGE
v,
v,
PPK
PPK
PPK
PPK
Ib)
PEAK
DISSIPATION
I,)
EaUIVALENT
RECTANGULAR
POWER
PULSE
(d)
2.1) Current Determined
by Load
2.2) Current Limited
by Gate Drive
Waveforms When Driving an Incandescent Lamp, High Inrush
Loads Develop Similar Waveforms
Figure 2
A bootstrap circuit is shown in Figure 3. Operation is
as follows: when the driver bipolar is on, the gate
potential is near ground and the MOSFET is off.
Capacitor C is charged to VDD through the load and
diode Dl. When the driver goes off, the gate voltage
rises, turning on the MOSFET, which raises the source
potential. If C is many times larger than the input
capacitance of the MOSFET, it acts as a voltage
source in series with the MOSFET source terminal
potential, thereby providing a gate-source voltage
close in value to VDD. The capacitor C will lose
charge with time through the reverse resistance of
diode D I, so that it becomes impractically large if the
load must be held on more than a few seconds, unless
a strobing technique is used as discussed in the following paragraph. However, many loads, such as
hammer drivers in high speed printers, are actuated for
under a millisecond so that capacitors in the range of
0.1 /-LF are adequate. In cases when the load is a solenoid, it may be permissable for the MOSFET to come
out of the ohmic region as a result of a partial loss of
gate drive after the solenoid has pulled in, because the
hold-in current is very low. However, the power dissipation of the MOSFET may increase significantly un- _
less the drop in gate drive reduces drain current to a
rather low value.
failure caused by secondary breakdown. Lamp drivers
may need to handle two types of surges, cold resistance inrush and flashover.
Cold resistance inrush occurs on all lamps during
turn-on; the peak is between 12 and 18 times the
steady state current. Furthermore, the inrush current
may be from 2 to 5 times rated current 5 milliseconds
after power application. The inrush depends somewhat
upon the lamp's design and the cold temperature of the
bulb, which in turn is dependent upon the lamp's
operating duty cycle and ambient temperature. Since
the turn-on surge is a repetitive transient, for
maximum MOSFET reliability, it is usually desirable
to have sufficient gate drive to place the operating
point in the ohmic region during the surge in order to
minimize power dissipation.
Driving Grounded Loads
In many cases, the load is connected to ground and
cannot be arranged as shown in Figures 1 and 2.
Driving a grounded load forces the MOSFET to be
used in a source follower circuit. The difficulty with a
follower is that, to keep the MOSFET in the ohmic
region with a large drain current flow, the gate must be
about 10 volts above the source potential, which is
only slightly lower than the supply. Therefore, the
gate drive voltage must come from a voltage source
which is about 10 volts above the supply voltage. If
such a voltage is available, no significant problem
exists other than ensuring that the driver circuit has
a sufficient voltage rating. When no fixed voltage
source is available, it may be generated using the
bootstrap technique.
VaD
Driving Loads Connected to Ground by Using Bootstrapping
Figure 3
6·83
When the load must remain actuated at full power for a
long period of time, a pulsed signal may be applied to
the drive circuit instead of a DC level. When load
current is to flow, the driver is off, except for occasional brief on periods during which the MOSFET's
source voltage drops to ground allowing C to recharge
to VDD. The source for this strobe signal could be a
system clock signal or the AC line. The higher the
pulse repetition frequency, the smaller C may become,
but C should usually be at least ten times the Ciss of
the MOSFET to avoid transferring more than 10% of
the charge on C to the gate during tum-on.
current from the drain terminal to the gate terminal.
Damage to the device from excessive current may
result if the impedance of the current loop is very low,
but this is usually not a problem unless the drainsource voltage is high enough to cause the bipolar to
operate in a BVCEO(SUS) mode. Since the BVDSS
rating of the FET is essentially the BVCBO rating of
the bipolar, BVCEO(SUS) is approximately one half
of BVDSS. Consequently, a second breakdown failure of the zener bipolar may occur when the drainsource terminal voltage exceeds one half of B VDSS.
The carrier transit time through the channel is under
Ins for most MOSFET structures, resulting in cut-off
frequencies on the order of a gigahertz. Consequently,
very fast switching is readily achieved, but parasitic
oscillations can be troublesome if certain precautions
are not observed.
General Driving Considerations
Regardless of the type of logic or network used to
drive a MOSFET, consideration must be given to its
properties such as:
1. The input protection zener diode
2. The high frequency response
3. The capacitive input impedance
Usually oscillations are prevented by observing one or
more of the following guidelines:
I. Keep lead and trace lengths short.
2. Place ferrite beads on the gate lead close to the
gate terminal or use a resistor of 10 to 1000
ohms in series with the gate.
3. Avoid a layout which may couple output signal to the input.
4. Surround the MOSFET with a ground plane
and shield output from input.
The input protection zener diode integral with some
devices places restrictions on the drive levels. The
positive voltage on the gate with respect to the source
should not exceed the maximum voltage rating of the
zener diode nor should the zener become forward
biased by allowing the gate-to-source voltage to become negative.
Since the MOSFET gate input resistance is essentially
infinite, many can be driven from a CMOS or TTL
output. However the input impedance of a MOSFET is
capacitive and the drain current essentially follows the
voltage on the gate. Although switching speed, per se,
is not important when driving a lamp or electromechanicalload, the limited transient current available from the logic element may result in switching
slow enough to cause significant transient power dissipation, particularly when a number of MOSFET
stages are being driven in parallel. Accordingly, a
transient analysis of some sort is usually required.
The reasons for these restrictions are evident from the
circuit of Figure 4 where it is seen that the zener is
actually the base-emitter junction of a bipolar transistor whose collector is connected to the drain. Consequently, the zener exhibits a negative resistance
characteristic similar to the BVCEO of a transistor.
Above a few hundred microamperes, the voltage may
switch from, for example, 20 volts to 8 volts. The
lower voltage will limit the amount of current available from the MOSFET and excessive dissipation in
the zener and drive circuitry may occur.
o
A fairly simple, yet quite accurate, analysis is to use
a charge control approach as described by Evans
and Hoffman [6]. For any particular time interval,
A. _ (A. VGS)(Cin)
tIG
where
A. VGS is the gate-source voltage change
Cin is the effective input (gate-source)
capacitance
IG is the average gate current during switching
A Parasitic npn Transistor in Zener Protected MOSFETs
Figure 4
If the drive circuit can cause a negative gate-source
voltage - a common situation with a source follower
driving a capacitive load - the bipolar becomes
turned on in the forward direction causing a flow of
(all values must be determined for the time interval of
interest).
6·84
Table 1
Pertinent Switching Relationships
Interval
Symbol
Gate Voltage Change
Tum-On Delay
td(on)
VG(TH)- VG(ofO
Rise Time
tr
VG @ IDl(on) - VG(TH)
Tum-Off Delay
td(oft)
VG(on) - VG @ ID2(on)
Fall Time
tf
VG @ ID2(on) - VG(TH)
Ciss
Ciss
8VDS
+ - - Crss
8VGS
Ciss
Ciss
8VDS
+ - - Crss
8VGS
consists of a p-channel MOSFET connected in series
with an n-channel MOSFET (drain-to-drain) with the
gates tied together and driven from a common
Table I shows the appropriate quantities to use in the
equation. For completeness, turn-on and turn-off
delay relations are included, but these intervals are
rarely of interest in power circuits. However, dissipation may be a problem during rise and fall time.
J~. ,
{
-V~~o
Capacitance values used are the average values as V G
varies over the ranges shown for the time interval of
interest. Appropriate VDS values must also be used to
detennine the capacitance. Key VGS points are:
VG(off)
Capacitance
+Vl
~
+Vl
,-CHANNEL
Off state gate voltage prior to
turn-on
= Threshold gate voltage
V G(fH)
VG@IDI(on) = VG corresponding to the peak
value of drain current for
capacitive or resistive loads or
the value of drain current when
the drain voltage enters the
ohmic region for inductive
loads
On-state gate voltage prior to
VG(on)
turn-off
VG@ID2(on) = VG corresponding to the value
of drain current flowing prior to
turn-off
=
CMOS Inverter
Figure 5
signal - hence, the name CMOS (complementary
MOS). When the input signal goes positive (+ V), the
p-channel MOSFET is essentially off and conducts
only IDSS (picoamperes). The n-channel unit is forward biased but since only IDSS is available from the
p-channel, VDS is very low. Conversely, when the
input goes low (zero), the p-channel device is turned
full on, the n-channel device is off, and the output will
be very near + V. Since the current (without a load) is
extremely small, the inverter dissipates almost no
power in either stable state; the only dissipated power
of consequence occurs during the switching transitions
as capacitances are charged. Due to the extremely high
input impedance of a MOSFET, the CMOS gate has
the capability of interfacing with many MOSFETs
when only static conditions are considered.
Driving from CMOS Logic
The widely used CMOS logic elements are ideal for
direct coupling to power MOSFETs because CMOS
can operate with supplies up to 15 volts, a level which
provides ample drive for the MOSFET. Since switching speed and transient load handling capability is
related to the output impedance of the drive source, a
brief examination of the CMOS circuit follows.
The DC resistance between drain and source when the
device is turned on is generally labeled "ON resistance" RON or RDS(ON). However, the CMOS gate
has a limited output capability detennined by the gain
of the n- and p-channel devices. Equivalent circuits
of the CMOS output are shown in Figure 6. '
All CMOS circuits usually have an output configuration as shown by the inverter of Figure 5. The inverter
6-85
approximately as the square of increasing Vas. The
p-channel transistor exhibits similar, but complemented, characteristics with less- gain and a more
gradual transition from a current source to a resistor.
+V2
V2
+V,
ISlnk .. RL2 + Ronln)
VO"V2-~
RL2 + Ronln)
Is~
When driving a capacitive load the initial voltage
change across the load will be a ramp due to the
current source characteristic, followed by a rounding
off due to the resistive characteristic dominating as
VDS approaches zero. For fastest turn-on and therefore lower dissipation in the MOSFET, the peak output current should be achieved while the CMOS inverter is operating in the constant current mode. To
accomplish this. the maximum current from ~he
MOSFET is that which corresponds to a Vas that IS a
few volts below the CMOS supply voltage, VI. From
Figure 7, note that operating the CMOS dr~ver. at
higher VCC will have a profound effect on ~wltchmg
speed because the CMOS output current mcreases
roughly as the square of Vas and the voltage where
rounding occurs has been pushed to a higher level.
,d.,o"ll
"~HANJ
n·CHANNEL
(b) Sinking Current
(a) Sourcing Current
CMOS Source/Sink Capabilities
Figure 6
A look at the characteristic curves of CMOS transistors will provide insight into the dynamic driving
source impedance presented to the MOSFET gate.
Figure 7 shows the characteristic curves of n-channel
and p-channel enhancement mode transistors typically
found in CMOS circuits. Referring to the curve of
Vas = 15V (gate-to-source voltage) for. the nchannel transistor, note that for a constant drIve voltage Vas. the transistor behaves like a current source
for drain-to-source voltages greater than Vas - VT
(VT is the threshold voltage of an MOS transistor - on the order of a volt or two). For a VDS
below Vas - VT, the transistor behaves essentially
like a resistor. Similar curves are obtained for lower
values of Vas except that the magnitude of the current is significantly smaller and, in fact, IDS increases
Therefore, the optimum interface from CMOS to a
MOSFET is shown in Figure 8. In this configuration,
the turn-on current is supplied from the p-channel FET
which has the poorest characteristics of the CMOS
pair, but when operating at 15 volts, serious rounding
of the MOSFET's gate waveform can usually be made
to occur at a level above that required to handle the
load current. The turn-off current is supplied from
the n-channel FET; it maintains good drive capability
down to the threshold voltage of the MOSFET's gate
which minimizes tailing of the drain current.
45
vG~"~e~·I'5IV
0
".~~
VGS@VCC- 1OV
5
11
VGS" vee l•
L
OR 74COO
0
'0
'5
Vos - OUTPUT VOLTAGE (V)
-15
- 10
Driving VMOS with a CMOS Gate
Figure 8
VOS@VCCc 5V
VGS" Ve~ = 1'0 ~
,
Driving from TTL Logic
The lower logic levels used in TTL make it much less
satisfactory than CMOS for direct coupling to power
MOSFETs; however, TTL can be directly coupled
when lower output currents are required, or some
additional circuitry can be added to make the match
more universally applicable. The coupling problems
are readily appreciated by analyzing the output circuit
of TTL.
VGS@VCC"15~
i..-'r
30
Transistor Output Characteristics of a CMOS Inverter
Figure 7
6-86
pull-up resistor. To maintain a reasonably fast rise
time, it is necessary to limit the peak drain current to
the value obtained for a particular MOSFET type
when the gate voltage is at 3.5 volts; the additional
1.5 volts are used as overdrive to place the operating point in the ohmic region.
Figure 9 shows the totem-pole output circuit configurations commonly used in TTL. When the driver is
off, the output is high, however, the output is slightly
over two diode drops below the supply voltage for
either configuration. Since the nominal supply is 5
volts, the output is approximately 3.5 volts, a level too
low to fully utilize a power MOSFET.
Full utilization of power MOSFETs is achieved with
open collector TTL as shown in Figure 11. The open
collector circuits do not have the top transistors and
the lower transistors are designed to be used with
supplies up to 15 volts. The MOSFET rise time is now
mainly dependent upon the external resistor used. For
fast rise time, the lower resistance values required may
cause objectionable dissipation when the TTL output
is low; use of the circuit of Figure 12 will provide high
speed and low dissipation. It essentially restores the
totem pole output to the TTL circuit using an external
high voltage transistor. Since the bipolar transistor
does not saturate, a general purpose transistor with a
high fT will provide fast drive signal to the gate.
v'
v+
(a) Medium-Speed TTL
Output Configuration
(540017400 Families)
J:
1
+V
(b) High-Speed TTL
Output Configuration
<5V
+'5V
R,
Basic TTL Output Configurations
Figure 9
RL
Va
tOKn
When the driver is on, the lower output transistor is
also on; the output is low, on the order of a few tenths
of a volt. The low level is satisfactory to ensure that
the power MOSFET is cutoff, in most cases, but the
high level needs to be raised.
Open Collector TTL Is Used to
Provide Greater Enhancement Voltage
Figure 11
A method of boosting the TTL output level is shown in
Figure 10. The external resistor allows the full-supply
voltage to be applied to the MOSFET gate, but in
doing so, the TTL output transistor becomes cut-off as
the level increases above 3.5 volts. Consequently, the
high drive capability and low output impedance of
TTL is not effectively utilized, as the drive to the final
value of gate voltage must come solely from the
'K!l
Va
+5V
+v
'5V
I.L
,o.:}ova
V,n
1
A "Totem Pole" Driver Increases Switching
Speed and Reduces Dissipation
Figure 12
Driving the VMOS with Standard TTL
Figure 10
6-87
-
[2] Bill Roehr, et al. "Silicon Rectifier Handbook,"
2nd Edition, Chapter 2, pp. 30-39, Motorola,
Inc., Phoenix, AZ.
[3] loco cit. "Silicon Rectifier Handbook," Chapter
8, pp. 117-121.
[4] E.T. Calkin and B.H. Hamilton, "Circuit Techniques for Improving the Switching Loci of Transistor Switches in Switching Regulators," IEEE
Conference Record, 1972 lAS Annual Meeting,
pp. 477-484.
[5] Rollie J. Walker, "Circuit Techniques for Optimizing High Power Transistor Switching Efficiency," Proceedings of Powercon 5, May 1978,
Power Concepts, Inc., Oxnard, CA.
[6] Arthur Evans and Dave Hoffman, "Dynamic
Input Characteristics of YMOS Power Switch,"
Siliconix Application Note, AN79-3.
Summary
MOSPOWER PETs can easily interface power loads
to integrated circuit logic. Spikes from inductive loads
usually must be limited to a level below the breakdown voltage of the MOSFET. High inrush,
capacitive-like loads usually require high gate drive to
place operation in the ohmic region during inrush.
To avoid deleterious operation and ensure fast switching, consideration must be given to the input protection zener diode - if present - and capacitive input
impedance. Direct coupling to CMOS is usually very
satisfactory if the IC supply is 10 to 15 volts. With
T2L, open collector elements are usually required to
obtain sufficient gate voltage for the power MOSFET.
References
[1] Bill Roehr and Bryce Shiner, "Transient Thermal
Resistance - General Data and Its Use"; Motorola
Application Note, AN569 , Motorola Semiconductor Products Inc., Phoenix, AZ.
6·88
6.4 :MPP500: The First Single Package
Complementary MOSPOWER Device
Introduction
The new MPP500 series of complementary power
MOSFET pairs provide increased convenience for
designers of small motor drives, inverters, and similar
devices with load ratings of up to 150 W. Mechanical
design is simplified because only half as many packages are needed to construct either a single-phase or
a three-phase bridge. Electrical design is simplified
because device matching by the designer is not required and because complicated isolated drives for
the upper switches in the bridge are no longer necessary. This last simplification alone can result in a
dramatic savings in system cost.
In the MPP500, both FETs have equal die size and
are mounted on a common header. Thus, the power
handling capability of the two devices is, essentially,
equal; for most applications, the difference in RDS(on)
can be ignored. Three different drain-source voltage
ratings are available: 100 V (MPP500), 80 V (MPP50l),
and 60 V (MPP502); all three devices will handle 2.3
A continuous or 7.7 A pulsed.
Input Drive Requirements
The real advantage of the MPP500 lies in the ways it
may be driven. Because the upper FET is a p-channel,
it is only necessary to lower its gate terminal 10 V
below the positive power rail to fully turn ON the
device. This can be accomplished with any opencollector or open-drain logic output. It is not even
necessary for the drive transistor to withstand the full
power supply voltage. Because, if the power is stable,
a suitable zener may be interposed between the
current sink and the p-channel gate. Shut off can
easily be accomplished with one pull-up resistor.
Furthermore, the same current sink that turns on the
p-channel FET ON can be used to simultaneously
turn the complementary n-channel FET OFF. This is
one of the benefits of using FETs with equal input
capacitance. In fact, in Figure la the CMOS gate used
to drive the MPP500 does not even have an open
drain output. Further, because the PLUS 40 process
with a 40 V gate breakdown is used in the MPP500,
no zener is required for power input voltages less than
40 V. Some overlap conduction will occur under these
circumstances during the switching transition as it
does in CMOS logic, and for the same reason. A zener
between the gates minimizes this.
MPP500 Device Characteristics
An MPP500 power MOSFET pair consists of one nchannel and one p-channel MOSFET which are
matched for die size, breakdown voltage, and power
handling capability. Die size matching means that the
p-channel member of the pair has higher ON-resistance (RDS(on» than the n-channel member and that
both FETs have equal input capacitance.
The reason for the difference in RDS( on) is that holes
(the majority carriers in p-channel devices) have less
mobility in a semiconductor structure than do electrons (the majority carriers in n-channel devices). To
construct devices with equal RDS( on), the p-channel
device would require approximately 2.4 times the
active area of the equivalent n-channel device and
would, consequently, have 2.4 times as much input
capacitance. This would mean that for equal gate
drive impedance the p-channel device would switch
more slowly than the n-channel device, and in certain
applications, this would complicate system design
considerably.
Reprinted courtesy of PCI July, August 1984
6·89
--
Switching speed is a matter to which some attention
must be paid. For resistive loads or situations where
no zener is desired between the nand p channel gates,
it makes sense to switch as rapidly as possible. Rapid
switching minimizes overlap currents and maximizes
efficiency. As with all other power MOSFETs, the
limit to how fast you may switch depends on how
much time and money can be put into the gate drive
circuit .
The MPP500 as a Power CMOS Driver
The circuits shown in Figure 1 are useful for a variety
of solenoid, lamp, and relay driver applications.
Inverse diodes are inherent in all power MOSFETs,
so additional diodes for inductive loads are not re9..uired unless very fast switching speeds are necessary
II 1. Such fast switching speeds are seldom required
for mechanical loads like motors and relays.
• BOV
o---~-.,
UK
,----,
1I4W
I
+lS V
0---.,
•
:
J+
1
i)
>-_.'_.7.......-+':
LOGIC
For reactive loads, the situation changes and a slower
gate drive circuit may be a better choice. This situation
exists because the reverse recovery time of the MOSFET inverse diode is typically between 200 and 400
ns. When switching speeds much faster than the diode
reverse recovery time are desired, additional components will be needed. Otherwise, reverse conduction
through the intrinsic diode or dV/dt effects during
switching will damage the MOSFETs. At input voltages below 40 V, Schottky diodes in parallel with the
MOSFET diodes are sufficient, since the forward
drop of the Schottkys is less than the MOSFET diode
drop. They, thus, will prevent the MOSFET diodes
from conducting. At higher voltages, more complex
arrays, using blocking diodes and conducting diodes
or more complex drive schemes involving active
commutation are required. For most inductive load
circuits, a slower drive which trades switching losses
for conduction losses in the extra diodes is simpler,
less expensive, and dissipates less power. Generally,
all that is required to switch at less than super-fast
rates is a less expensive drive circuit. A CMOS logic
gate is generally sufficient. A 4000 series CMOS gate,
for example, will drive the MPP500 (both gates) with
a current rise/voltage fall time of 300 to 700 ns. A
series gate resistor can slow this further if desired.
SIGNAL
L ___ ,
'50 WATT
POWER OUTPUT
MPP
500
COMo----+-------I---=
a:
.05
/v
0-
,,,\
\
~
/
-
"I'--- i""- ~WrTS
~V
1 WATT
50
100
L
Pou,=2j WAT/
200
500
1K
2K
5K
10K
20K
FREOUENCY (Hz)
Distortion vs Frequency for 25 Watt Amplifier
Figure 7
The output devices are Siliconix type VN64GA, which typically have a transconductance of 3 mhos, an rds(on) of 0.3
Ohms, and an input capacitance of 700 pF. These characteristics are needed in order to have sufficient gain in
the current bias loop, good power efficiency, and wide
open loop bandwidth.
The usually published frequency response at 1 watt power output is not shown as it conveys little information. Response is
down 112 dB at 19 Hz and 100 kHz when driven from a 1 Kr!
source with the input filter removed.
Some listening tests have been run using a variety of associated
equipment. Most listeners notice improved reproduction of
high frequency transients and have difficulty in detecting
overloads and clipping unless excessive.
The power efficiency calculates at 53% at the 25 watt output
level. The power dissipation is essentially independent of frequency and varies little with power output. It is about 17 watts
at idle and increases to about 22 watts at output levels from 10
to 25 watts.
A 50 Watt Design
An amplifier designed to produce 50 W into 8 !l is shown
on Figure 8. For this application low distortion is only
required over a range from 50 Hz to 10 kHz. It was
adapted to use an available power supply which produces
+ 78 V and - 52 V under no load and nominal line
conditions.
Most assemblies show no evidence of slew rate distortion until
at least 30 kHz when a slight crossover glitch appears on the
waveform at a level of 15 watts or more. The glitch is level sensitive due to imperfect bias tracking as a function of power
output. Bias tracking also becomes worse as the frequency increases, probably because of stored charge problems in the rectifier diodes used for 02 and 03. This tracking error would
normally not be encountered on speech or musical signals.
Figure 9 shows total harmonic distortion at selected frequencies as a function of power output and Figure 10 shows distortion at 1, 10, and 50 walls as a function of frequency. In the
important mid range, THO is under 0.1 %.
Choice of suitable transistors for the diff-amp is limited. Good
results have been obtained by using matched pairs of type
2N2484. The gain typically is about 400 at 2 mA and although
the 60 volt VCEO rating seems marginal, no problems have
been experienced.
The narrower power bandwidth of the 50 watt amplifier as
compared to the 25 watt amplifier is a direct result of the
higher input capacitance of the higher power output devices.
6-102
To avoid serious high frequency distortion the diff-amp current was increased from 2 rnA to 4 rnA; the higher current
necessitated a 2: I reduction in base circuit resistance in order
to avoid an excessively tight match on the current gain of the
diff-amp pair with the result that the bypassing action of C5,
kept at I I'F, is less effective at low frequencies. The narrower
power bandwidth of the 50 watt amplifier does not reflect a
significant difference in the frequency response at I watt as
compared to the 25 watt design.
1.0
lz
as
~
0.2
~
0.1
I
f~50
o
C
f-'"'
0.05
~
0.02
o
f-"
0.01
::;
~
:
0.005 ~
~
/
~
~
I
....,.. r r-::d
~·HZ
/ 10 kHz V ......
0 == ..... 1 kHz
Hz
...-foOH;-
a kHz
g
0.002
I-
0.001
0.1 0.20.5 1
2 5 10 20 50 100
POWER OUTPUT (WATTS)
Although not shown, data at 5 watts output is similar to that at
10 watts. The low frequency distortion is higher at these levels
than at 50 watts because of the imperfect filtering action of
Cs. Because of the clipper circuit, the voltage on Cs is a larger
percentage of the input signal as the output drops from the 50
watt level.
Distortion vs Power Output for 50 Watt Amplifier
Figure 9
l
1.0
z
o
~
The harmonic distortion spectrum is shown on Figure 11. The
dynamic range of the instrument used is 90 dB. Note that the
third harmonic is prominent at all power levels. The even harmonics quickly disappear, but at the 50 watt level, odd harmonics up to the 13th were detectable. Probably none of these
harmonics is discernable by ear.
~ 0.21-~~~...-+--I--+-+--+-*--A
is
~
o
0.1
~ 0.05 ~-.J--I~-+......:~""",--t:~+--+-,:A--l
""
..J
:: 0.02
e
Power data is shown on Figure 12. The power efficiency
calculates at 63 0J0 at the 50 watt output level. The power
dissipation is essentially independent of frequency. The heat
sink must handle about 40 watts of power.
I--+--I'--+---'"",,;:;::--+-+--t;;;-f--j
0012~0-~5~0-1~0~0~2~00~~5~0~0-'1~K~~2K~-'~~~~
FREQUENCY (Hz)
Distortion vs Frequency for 50 Watt Amplifier
Figure 10
r------~------------------~---1_-o+80V
5100
7.5K
3.6K
1/2W
VN1200A
7.5K
.22
..::::;~_+-l_O+-t--l,j~
~O
LJ......f
L-._ _......
150K
3.6K
1I2W
15K
27011
UNLESS NOTED.
ALL RESISTORS: 1/4W.5%
ALL CAPACITORS: 100V. VALUES IN ,.F
+SDV
Practical 50 Watt Autobias
Figure 8
6-103
-=-
60
...
;!
zw
65
:IE 70
...--.,.---.---,----r---,-"71r----,1.3
\ 'V-000HARMONICS 150 WATTS
g
\
,....
\
1\ \
1\
I ,
\
z
:> 75
9
.....
80
w
...........,
1----+--±:=+-~~'-l-----1I---l1.1
\>'EVEN HARMONICS
,
,
\
85
'
10',
WATTS
,
'
\
\
, 1 WATT
"
,
1=2500 Hz
901~2~3~4~5~6~7~8~9~1~0~1~IJ-l~3~~15'
HARMONIC
Locus of Harmonic Spectrum
Figure 11
Power Requirements
FIgure 12
Finding suitable driver transistors for the diff-amp was not
easy. The best transistor discovered to date is a D38WI4
manufactured by General Electric. The transistor has
BVCEO > lOOV, which allows ample voltage margin, and
hFE > 400 which places base current < 10 pA at 4 rnA of collector current. Consequently, base current matching to within
I pA is not too difficult and this match will produce a maximum error of 20 mV in the diff-amp which translates into
a 20 rnA error in the MOSFET idle current.
-20
...J
;!
.J.
R(1l)
.1
Po=10W __
'- 47
1=2.5 kHz
2
213
0
-30
20
ffi
-40
...,.
-60 r--
"~ -50 rOO ~
:IE
213
::I
9
..
~
-70
"0
-80
-90
The MOSPOWER output transistors are VN1200As. They
have 120 V breakdown ratings, an "on" resistance under
0.2 ohms, transconductance of 5 Siemens, and an input
capacitance of 1200 pF. This excellent combination of
characteristics plays a major part in achieving the excellent
results displayed by the simple autobias circuit.
-~~
,
1\'" ~
~
47r
THO(%)
4.1
2.1
1.5 _
4.9
-
213il"'== ........
00
i
o
~
8
10 12
HARMONIC
14
15
18
20
Open Loop Output Spectrum Even Harmonics
FIgure 13
-20
..... -30
;!
Finally, Figures 13 and 14 illustrate the tradeoffs in open loop
harmonic content as the diode series resistance is varied in the
dynamic bypassing scheme. With R =470, the bypassing effect
is negligible, and no harmonics past the 11th are discernible. As R is reduced, high order harmonics are introduced
which increase in level. However, harmonics below the 6th
minimize at various values of R. In the final amplifier
design, 2/30 was chosen as it minimized the large 3rd
harmonic resulting in lowest overall total harmonic distortion. By better matching of the push-pull configuration, it
should be possible to reduce the even harmonics below the
levels shown.
ffi
-40
::E
~ -50
z
::I
LL
-60
~
m
-70
..
~ -80
2/30
~
,-
,:-...
l\" ~
I
PO =10J _
1=2.5 kHz
-
R(1l)
47
2
213
0
" '" -
THO(%)
4.1 2.1
1.5 4.9
~
""r--.,.
R_ 47n
~
r--...
~ t- ;::::: on
2130
1
-90
9
11
13
HARMONIC
15
r-..
17
20
19
21
Open Loop Output Spectrum Odd Harmonics
FIgure 14
Conclusions
The circuit scheme presented illustrates that using MOSPOWER transistors as power output devices produces an
amplifier of extraordinary performance considering the circuit simplicity. It offers the following advantages over
bipolar amplifier counterparts:
1. Only one driver stage
2. Simple overload protection
3. Stable bias point
4. Power efficiency independent of frequency.
6·104
6.6.2 A Simple Direct-Coupled
Power MOSFET
Audio Amplifier Topology Featuring
Bias Stabilization
©
1982 IEEE. Reprinted, with pennission, from INTERNATIONAL CONFERENCE ON CONSUMER
ELECTRONICS, June 9-11, 1982, Arlington Heights, IL, pp. 136-137.
Abstract
Utilizing the high gain and high input impedance of
short channel power MOSFETs, a simple circuit has
been devised to provide sufficient drive for amplifiers
up to 200 watts output. The circuit described features
automatic control of the quiescent bias level and offers
performance which meets criteria for high quality
audio reproduction.
Introduction
Short-channel power MOSFETs offer several characteristics attractive for audio power amplifier applications namely:
1. Low distortion at all frequencies because of a linear
transfer characteristic and a very high cut-off frequency.
2. A low drive power requirement because of a nearly
infinite input resistance, relatively low capacitance, and high transconductance.
3. Failure immunity because second breakdown occurs well above the power rating.
4. Freedom from minority carrier storage delay time
following signal overloads because FETs are
strictly majority carrier devices.
5. High power efficiency because of low "on" resistance.
rent with temperature at the desired idle level. Furthennore, as a result of production line variations a
threshold voltage range on the order of 2 volts must be
accommodated. Consequently, these FETs must be
used within a closed loop bias system. The puzzler has
been to provide a means of sensing the idle level
despite the presence of the high current half-wave
pulses which occur under Class-AB operation. A successful solution to the bias problem was the subject of
an earlier paper by the author [ 1], however, the simple
circuit in that paper used AC coupling of the audio
signal. The circuit describeq in this paper applies this
bias technique to a direct-coupled design. The resulting circuit is simple and offers outstanding perfonnance.
The Autobias Principle
The scheme previously reported extracts the idle
current from the total current by means of a clipper
circuit and filter which produces a voltage related to
the idle level. A small resistor in series with the source _ _
of one of the power FETs is used to sample output ~
current. The voltage at idle is one-half of the clipping
I
level so that when the signal is large enough to cause
current cut-off on negative half cycles (thereby producing an asymmetrical wavefonn) the peak clipper
and filter produce an average voltage which is substantially the same as the zero signal level, regardless
of the amplitude of the peak current, as illustrated by
the wavefonns of Figure 1. The filtered voltage is used
One of the major design problems has been providing
a means of stabilizing the idle current level because
short-channel FETs exhibit an increase of drain cur-
6·105
V+
J
Rl1
T
C6
----
VOl
01
IIDLE
1
R13
CUPPING
LEVEL
R16
~ ~----~----~--~~------
O'
C7
Clipper Circuit Waveforms
a) Clipper Input
b) Clipper Output
Figure 1
O.
RB
Op-Amp Bias Control for Class AB
Figure 2
in a feedback loop to control the FET gate voltage and
thereby maintain the bias current within close limits.
Dl and D2 under large signal conditions. In addition,
R16 should be over twenty times R20 to prevent a
significant portion of the DC level on C7 from biasing
D2. Capacitors C6 and C7 filter the clipped audio
signal.
A simple technique for achieving a stable automatic
bias current is shown in Figure 2. The op-amp serves
as a differential amplifier in a voltage comparator loop
which serves to maintain the levels at its inputs equal.
Thus the idle current is simply VDl/2RB, since R12
and R13 are equal. The loop gain is so high that
variations of FET threshold voltage or other circuit
constants have no measurable effect upon the idle
current level. It is important to use a two stage clipper
and also match the diodes Dl and D2 to avoid shifts in
idle level under large signal conditions; a dual
matched signal diode is an ideal choice. Circuit resistors should be chosen to force equal currents through
Circuit Topology
The general circuit topology is shown in Figure 3. It
consists of an input differential amplifier using n-p-n
transistors directly coupled to a p-n-p push-pull stage
which drives the output power FETs in a totem pole
arrangement. A similar topology has been previously
used by Sampei, et al [2], and Harvey [3], however,
the implementation is different and neither circuit has
automatic bias.
(+)
RC
02.
R20
RC
FEEDBACK
RB
Amplifier Topology for Power MOSFETs
Figure 3
6-\06
+v
+v
~---+---
51K
os~t
R2
51K
'70 pF
....
a~!
R23
~:~~ to Z,
Rl0
UK
as
V'
Rl.
51 K
2N3904
..
--t'"CR3
~~~~ to Z2
R19
62011
Ra
R20
06811
04
8211
jD2
R2.
RB
211
R2S
12V,04W
-4~V
+
Complete Amplifier Schematic
Figure 5
07
The loads for the p-n-p drivers are provided by current
sources (CR2 & CR3) and high-valued load resistors
(RG) R2l, R22. This technique permits high gain to
be achieved while allowing all the driver current to
charge the input capacitance of the output devices.
The drivers may work class AB when the input is a
fast rising signal thereby providing a high peak current
in phase with the drain current to provide a high slew
rate.
\
.OS
'" ""
~ 05
L4
e
~
I
a
03
....
02
:;!
01
The drivers and the diff-amp stages are designed so
that they cannot be driven into saturation. To do this
requires that the stages load resistances and quiescent
currents be chosen to limit the available voltage swing;
to achieve reasonable gain from the diff-amp requires
a driver supply voltage which at fIrst seems unnecessarily high (32 V) but is readily available from a
standard 24 volt transformer. The current drain from
this supply is fairly low, however, requiring only 8
rnA for each driver plus 2 rnA for the diff-amp.
............
P
....I=2W ',._
QS, as. VN120D
f= 10 kHz
..........
'-.....
t-
f", 1 kHz
o
°
.S
MOSFET IDLE CURRENT (AJ
Effect of Idle Level on Harmonic Distortion
Figure 6
The - 9 V' line which rides on the - 40 V rail also
supplies about 20 rnA and must be well fIltered to
prevent AC ripple from being amplifIed by CR3 and
injected into the gate of Q6. A zener diode is an
effective fIlter and also provides a reference level for
the current source transistors CRl and CR2. Neither
the absolute value nor the regulation of the other
supplies is critical. Any desired power output may be
obtained from the circuit by simply altering the value
of the main rails. Automatic switching between two
supplies could also be used to improve power efficiency. No circuit changes are necessary other than
altering the divider to the supply terminal, pin 7, of the
op-amp to avoid exceeding its voltage rating (30 V)
and using transistors with suffIcient voltage breakdown.
In addition, the driver supply is non-critical with regard to hum or absolute value. It is possible to obtain
this supply by bootstrapping from the amplifIer output, but a very large capacitor would be required to
avoid a serious increase in low frequency distortion.
The result is that bootstrapping works out to be less
satisfactory and more expensive than the approach
shown. The supply need not ride on the + 40 V rail. It
is just as satisfactory to have a + 72 V supply with
respect to ground; it could serve both channels of a
stereo pair.
6-108
Power dissipation and efficiency as a function of
power output are shown on Figures 8 and 9. Power
efficiency is about 60% at full output and is essentially
independent of'frequency.
The diode D4 serves not only as part of the bias clipper
network but also provides dynamic bypassing of the
2 ohm sense resistor. The diode reduces third harmonic distortion and increases power output at the expense
of introducing a number of low-level high order
harmonics into the output. This action is illustrated
in detail in reference [1]. A similar network must be
used in the source circuit Q5 to balance both halves
of the push-pull output stage.
5.
/'
/
V
The zener diodes Z 1 and Z2 provide overload protection with a shorted output. They are chosen to limit the
gate drive to a level such that ID< PD(rated)!V+. In
addition, the zeners prevent large voltages from
breaking down the FET gate-oxide layer. The zeners
also conduct in the normal diode forward direction
when the gate signal would normally swing the gate
signal below the source. This action not only protects
the gate but prevents the signal from saturating CR3
which would render the bias control loop inoperative.
./
-----
•
01
02
05
~ t:-
.
1
5.
..
,
V
IV
20k,V
~kH'
~~
--
-V
•• ,
20
12
•
01
....- P>- C
V- f"""'..
.
,
:--
------
~
,.
02
05
10
20
50
10
20
50
100
POWER OUTPUT (WATTS)
Current from Main Supplies versus Audio Power Output
Figure 9
Figure 10 shows a 100 kHz square wave output when
the input filter Rl Cl is removed. With the filter in
place, the slight 1 MHz ringing disappears. Rise time
and frequency response can be varied by altering the
value of C5.
,.
/'
10
14
Curves of distortion as a function of frequency and
power output are shown in Figure 7. In summary, total
harmonic distortion is generally under 0.1 % throughout the audio range. The unusual rise in 20 Hz distortion at 16 watt is caused by the imperfect filtering of
the output of the clipper circuit in the bias control
loop. Because of the clipper circuit, the voltage on C5
becomes a larger percentage of the input signal as the
output drops from the 50 watt level.
L
50
Power Dissipation in Both Output Transistors versus
Audio Power Output
Figure 8
Performance
V
20
POWER OUTPUT (WATTS)
Since most power MOS devices have a cut-off frequency in the gigacycle range, parasitic RF oscillations can be troublesome. Most assemblies require
bypasses on the order of 0.22 MF, connected from
supply lines at the FET location. In addition, leads
running to the gate terminal should be either shielded
or used as a twisted pair with a ground lead.
1
10
20 kHz
10 kHz
20 Hz
30Hz
1 kHz
1\
.v
100
POWER OUTPUT (WATTS)
Output Voltage with 100 kHz Square Wave Input. Input Filter
Removed (5 V/Division)
Figure 10
Distortion Characteristics
Figure 7
6·109
The output coupling filter permits capacitive loads to
be driven without introducing oscillation. With a 1 p,F
load, distortion rises slightly at high frequencies.
References
[1] Bill Roehr, "The Autobias Amplifier, A New To-
pology for Automatically Biased Audio Amplifiers
Using Power MOSFETs," Journal of the Audio
Engineering Society, Vol. 30, No.4, April 1982,
pp. 208-216.
Some listening tests have been conducted. Listeners
are usually impressed and use terms like "smooth,"
"musical," and "excellent transient reproduction" to
describe the sound. In addition, it is very difficult to
detect when the output stages are driven into clipping.
It appears that the circuit is suitable for the highest
quality audio reproduction.
[2] Tohru Sampei, Shin-ichi Ohashi, and Shikayaki
Ochi, "100 Watt Super Audio Amplifier Using
New MOS Devices," IEEE Transactions on Consumer Electronics, Vol. CE-23 , No.3, August
1977, pp. 409-416.
Summary
[3] Barry Harvey, "Power-FET Amplifier Designs
Boost Fidelity, Reduce Complexity," EDN, Vol.
25, No. 17, Sept. 20, 1980, pp. 137-142.
A simple circuit using short-channel power MOSFETs
has been devised. It features high power efficiency
and low distortion over the audio band, high slew rate,
bias stability, fast overload recovery, and short-circuit
protection. Listening tests have rated it very high in
audio accuracy. Since it is potentially low cost and
usable for power levels up to 200 watts, it should
have wide application in the audio industry.
[4] Edward M. Cherry and Gregory K. Cambrell,
"Output Resistance and lntermodulation Distortion of Feedback Amplifiers," Journal of the
Audio Engineering Society, Vol. 30, No.4, April
1982, pp. 178-191.
6·110
6.6.3 A MOSFET Power Amplifier with Error
Correction
Abstract
higher turn-on voltage drive requirements and smaller
transconductance at low current levels. The former
tends to contradict generalizations which have been
made to the effect that drive circuits for power MOSFETs are less expensive, at least for the reliable
source-follower configuration. The latter results in
transconductance droop in the crossover region if bias
currents are not fairly high. Such transconductance
droop can result in crossover distortion.
Power MOSFETs are emerging as the device of choice
for high-quality power amplifiers because of their
speed, reduced needfor protection and falling cost. A
low-distortion power amplifier design is illustrated
which includes output stage error correction to reduce
the effect of transconductance droop in the crossover
region and thus allow operation at more efficient bias
levels.
Introduction
In this paper we will illustrate a high-performance
amplifier design which utilizes the advantages of the
power MOSFET while dealing with the drawbacks of
the device. Although not taken to an extreme, the
underlying philosophy of the design is that smallsignal silicon is inexpensive, i.e., that the overwhelming portion of expense in a power amplifier is in
items like the power transformer, filter capacitors,
power transistors, heat sinks, chassis and related
hardware. Thus, in order to take full advantage of the
performance achievable with the MOSFET output ~
stage, a very high quality front-end and driver are - . .
provided. The driver, operating from regulated
boosted supplies, is capable of providing high voltage
and current swings to the power MOSFETs with good
headroom. Output stage transconductance droop is
dealt with by employing a simple but very effective
output stage error correction technique proposed by
Hawksford. [1] The resulting design achieves a 20kHz THD figure of less than 0.0015 percent at an idle
bias of only 150 mAo
The rapid evolution of power MOSFETs during the
last few years has brought them to the point where
they are now very attractive for use in audio amplifier
power output stages. Important improvements which
have been made include increased voltage, current and
dissipation ratings, reduced "on" resistance, availability of complementary pairs and greatly reduced
cost. Although a 75-watt MOSFET is still more expensive that a ISO-watt bipolar transistor, the premium
is small when considered relative to total amplifier
cost and improved performance.
The purpose of this paper is to demonstrate the level of
performance achievable with current technology and
illustrate practical circuit techniques for achieving this
performance.
Power MOSFETs have several fundamental advantages over bipolar power transistors, most notably
speed and freedom from secondary breakdown. The
latter provides higher "usable" power dissipation,
improved reliability, and freedom from safe-area limiter circuits which can misbehave and cause audible
degradation. MOSFETs also have some disadvantages
in comparison with bipolar transistors. These include
MOSFET vs. Bipolar Output Stages
Design of MOSFET power amplifiers is quite
straightforward and conventional as long as differ6·111
(a)
2'
16
(b)
80 ta pULLE TelT
10Wv
)V_ ==
/
2.
I--J
p.s
/1
PJSE JST
rt
I
I
VOS'" 25V
16
J
12
YGS-6'i_~
V
r-
V
l v10
20
30
40
=
/I
(1/
I--
TJ
J,
-2'
-16
I:i
V
II
v
80 !S PULlE yeJT
-12
E
-
(d)
l-
I
I
I
i
-30
-40
10
Ili lL
1/
lJ /
j
TJ= 12.t
1-""" ]'-......
TJI =25"C,
TJ=I -...Ic .......I'-..... rA
G
b& 'L
Iffl
~
V
z -.
E
1
-20
6
LPULl. TElT
~ -12
I
-4
-Jv- t-Jv- t - -10
-2'
-16
-Jv- t - -
-.
•
I- ..
VDI=:Z5~
Vo =Jv- t - -.
;j V
V/,
~
[6 V
VGS. GATE TO SOURCE VOLTAGE (VOLTS)
J
G
~
V
55' C
2
50
-r- r-----
i
z
-i
= +125"C
TJ=2S"C
r-(J=i
r-----
YOS. DRAIN TO SOURCE VOLTAGE (VOLTS)
(c)
lL
'L
1/
L
-2
-50
VOS. DRAIN TO SOURCE VOLTAGE (VOLTS)
t?
-4
-6
-.
-1.
VGS. GATE TO SOURCE VOLTAGE (VOLTS)
Drain and Gate Characteristics for Power MOSFET Types IRF-132 (a&b) and IRF-9130 (c&d)
Figure 1
(Drawings courtesy of International Rectifier. Inc.)
ences between MOSFETs and bipolar transistors are
understood. Figure 1 shows the drain and gate transfer
characteristics typical of the vertical DMOS devices
used in this project (International Rectifier types IRF132 and IRF-9130). [2] The important point to see
here is that these enhancement devices require about 3
volts of forward gate bias to begin to tum on (e.g.,
gate threshold voltage, Vt) and may require as much as
10 volts to conduct high currents (12 amperes). While
the required bias voltage is thus higher than for bipolar
transistors, it can still be generated by the traditional
Vbe multiplier circuit. Thermal bias stability for the
power MOSFETs is much better than that for bipolar
transistors, even though the vertical DMOS devices
have a Vgs temperature coefficient of about - 5.0
mVrC at a typical bias current of 150 rnA. This is
partly due to the MOSFET's lower transconductance
at the bias point. The MOSFETs negative temperature
coefficient of transconductance also tends to improve
thermal bias stability. As a result, in some cases, a
Vbe multiplier transistor or associated reference diode
need not be mounted on the heat sink.
If the popular source-follower output stage configura-
tion is used, the substantial gate drive voltage required
for high currents means that the driver stage should be
provided with a "boosted" power supply voltage
greater than that of the main high-current supply in
order to take full advantage of the voltage swing available from the latter. The current requirements for the
boosted supply are small, and it can be regulated at
little additional expense, thus reducing hum, crosstalk
and modulation distortion. Several high-quality bipolar power amplifiers also use boosted driver supplies,
some regulated.
While bipolar transistors are regularly placed in
parallel with small individual emitter ballast resistors,
the paralleling issue is not as straightforward for
power MOSFETs, at least in linear applications. It has
been said that the negative temperature coefficients of
transconductance and "on" resistance of MOSFETs
act to suppress current hogging by one transistor, thus
permitting easy paralleling of MOSFETs without ballast resistors. This appears to be true for hard-
6-112
switching applications where the paralleled devices
are all fully turned-on together (Le., channels fully
enhanced by forward gate voltage) so that current and
dissipation imbalances are only a result of mismatched
"on" resistance.
Bipolar
IRF.1J2
30
IRF·9130
25
20
However, the issue is more complex for linear, and
especially low-distortion, applications because the
operating region of interest is not the fully turned-on
region, but rather the linear region wherein drain current at a specified gate voltage is important. Specifically, recognizing that the gate threshold voltage
specification for these devices is 2-4 volts, an examination of the gate transfer characteristics of Figure I
indicates that a very serious current imbalance can
exist unless gate threshold voltages among paralleled
devices are reasonably matched. It is also apparent
that reasonable temperature differentials will not
adequately reduce the imbalance. Because of the size
of the worst-case threshold voltage differentials possible, source ballast resistors are not a reasonable approach to achieving balance. It thus appears that for
high quality audio applications where paralleled
devices are necessary, both threshold voltage and
transconductance of paralleled devices should be
matched. Threshold matching that guarantees that all
devices are "on" to some extent in the quiescent bias
state, and transconductance matching to within 20% is
probably adequate.
15
10
05
14
10. AMPERES
Power MOSFET Transconductance versus Drain Current. For
comparison, note much greater ratio of transconductance to
operating current typical of bipolar transistors
Figure 2
This is done because both halves are "on" and contribute transconductance in the crossover region while
·only one half contributes transconductance at currents
well outside the crossover region. This usually results
in bias currents of less than 100 rnA per output transistor, sometimes as low as 20 mAo This small amount
of bias current compared to several amperes of signal
current being handled can sometimes result in unexpected temporary bias inadequacy, resulting in crossover distortion, because a small change in circuit parameters (about 50 mY) can cause the bias current to
vary considerably. This can happen as a result of the
time delay in the intentional thermal feedback loop
created by placing the bias generator sensing junction
on the heat sink; the temperature of this junction will
differ from that of the power transistors as a result of
thermal delay, low-pass filtering and attenuation.
After a high dissipation interval ends, the amplifier
may find itself temporarily underbiased because the
power transistor junctions cooled down faster than the
bias transistor.
Modem complementary MOSFETs, with maximum
"on" resistances of only about 0.3 ohms, are just
about as efficient as bipolar transistors in terms of
voltage dropped from supply rail to load in output
stages. However, they typically require a higher
operating current to achieve a given transconductance.
This characteristic is illustrated in Figure 2. The
device transconductance in a source-follower or
emitter-follower output stage is important because it
determines the small signal voltage drop through the
stage as a function of current. This is especially important in Class-AB stages where it is desirable that
the sum of the effective transconductances of both
halves be high and be constant with current so as to
avoid crossover distortion. It can be seen from Figure
2 that approaching this condition with power MOSFETs requires fairly substantial bias current (as a
rough starting point, that current where transconductance is one-half its high-current asymptotic value), on
the order of a few hundred milliamperes.
Bipolar output stages can be operated in an overbiased
mode, but the penalty can be dangerously reduced
thermal stability if larger heat sinks are not used, or
increased crossover distortion if larger ballast resistors
are used. Much less thermal feedback is necessary for
thermal stability of MOSFET output stages and their
higher bias is less likely to disappear under transient
thermal conditions. Compared to bipolar designs,
Class-AB MOSFET power output stages also tend to
have a wider Class-A region of operation and a
smoother transition to the Class-B region of operation.
In contrast, bipolar power transistors are typically
biased at a much lower current, but this is not entirely
advantageous. A typical bipolar output stage will often
be biased approximately where the dynamic emitter
resistance of the output devices (118m) at crossover is
equal to the associated ballast resistance as a compromise in achieving approximately constant total
output stage transconductance as a function of current.
Although power MOSFETs require virtually no drive
current at low frequencies, their substantial input
capacitance means that drive circuits with forward and
reverse drive capabilities similar to those employed
6-113
with bipolar transistors should be used for wideband,
high slew-rate circuits for demanding audio applications. Although the gate-source capacitance can be on
the order of 700 pF, this capacitance is effectively
"bootstrapped" in a source follower output stage,
typically reducing its effect by about an order of magnitude. The smaller gate-drain capacitance, about 100
pF, is also present. A 100 VI/J-s slope with an effective
capacitance of 170 pF thus requires a 17 rnA current
capability from each driver.
BiPolar
10
MOSFET
~
,~
K\
05
Power MOSFETs tend to be inherently faster than
bipolars, partly because there are no minority carrier
effects. Their speed is primarily limited by the ability
of the drive circuitry to charge the internal gate electrode capacitance through the effective gate resistance.
The wider bandwidth, reduced excess phase and reduced variation of device speed with voltage and current tend to allow greater high-frequency negative
feedback with greater stability. The higher switching
speed also tends to reduce dynamic crossover distortion. Furthermore, the MOSFET's higher switching
speed greatly reduces the flow of Class-AB
common-mode current at high frequencies which
poses such a destructive threat to many bipolar designs.
02
1
10
20
50
VeE or YOS. VOLTS
100
200
Safe Operating Area (SOA) Comparison of Bipolar Ring-Emitter
Transistor (2SA-1072) and a Power MOSFET (IRF-9130). Rated
power dissipations are 120 Wand 75 W, respectively
Figure 3
much energy (product of power and time) is required
to raise the temperature of the hottest point on the chip
(which will not be much different than the average
temperature of the whole chip) to a dangerous point.
However, power MOSFETs do have a tendency to
very high frequency parasitic oscillations in real-world
circuits. This appears to be a result of the natural high
speed of the device combined with the substantial
drain-source capacitance (300 pF) typical of these
devices, making formation of a Colpitts oscillator easy
if inductance is present in the gate circuit. This often
necessitates the use of a small resistor in series with
the gate. In combination with the device input capacitance, this resistor (typically about 100 ohms) can
create an additional pole which tends to reduce the
high-frequency improvement over bipolar transistors.
Figure 3 shows a comparison of safe operating area for
a power MOSFET and a typical bipolar power transistor. Notice that there are no steep secondary breakdown SOA slopes at high voltages for the
MOSFET - it is essentially limited by simple power
dissipation over its full voltage range. This is also true
for short-term dissipation well in excess of rated continuous dissipation, where ,thermal time constants
govern the allowable excess dissipation. For example,
a 25-ampere peak with 100 volts across the MOSFET
can be handled for 10 microseconds. Figure 3 illustrates that "usable" dissipation (SOA at higher voltages) for a MOSFET may be equal to that of a bipolar
power transistor of substantially higher rated power
dissipation. Safe operating area at high voltages is
particularly important when difficult reactive loads are
being driven. In many power amplifiers the use of
multiple paralleled output devices is for reasons of
increased safe operating area rather than simple thermal considerations. Finally, freedom from secondary
breakdown means freedom from complex safe-area
limiter circuits, some of which are notorious for their
misbehavior. [3]
Perhaps one of the most important advantages of
power MOSFETs for audio use is their freedom from
secondary breakdown and large safe operating area
(SOA). A highly simplified explanation of secondary
breakdown in bipolar transistors is that it results from
localized current hogging which in turn results from
localized thermal "hot spots." Transistor current at a
given base-emitter voltage has a very strong positive
temperature coefficient. Thus, a "hot spot" carries
more current and gets even hotter as a result. This
regenerative process, once started, can be very rapid
and unforgiving. It can persist even after the external
voltage and current conditions re-enter the safe
operating area, leading to destruction. The relationships in a power MOSFET are in contrast degenerative
in nature because hotter regions exhibit reduced transconductance and thus tend to conduct less of the total
current. This tends to equalize the temperature across
the chip. The safe area of a MOSFET is thus primarily
governed by simple thermal considerations of how
Figure 4 illustrates a simple 50-watt MOSFET power
amplifier design. It is notably similar to what a simple
bipolar power amplifier design would look like. Transistors Q1 and Q2 comprise the input differential
amplifier whose output is converted to a single-ended
current by current mirror Q3, 4. This current feeds the
common-emitter pre-driver Q5 which is provided with
a constant-current load. Capacitor C1 provides
Miller-effect feedback compensation and establishes a
6-114
B++ SOY
R6
"0
R7
"0
R8
"
25mA
a3
'-o..J
r"'"
~
~.
L,
~.
JT
I~
V a7
......
it-
R'3
'00
3.
B+ 35V
a.
r
":j~D'
R12
'70
;.~'
Va,
IN
.......
R5
.3 K
"
R'
lOOK
R'
'00
OUT
a'.........
R•
•8K
R3
'00
-=
":j~3
R'
"0
R'O
, K
a.'"
-=
~
R"
'DO
R11
560
VQ8
.......
O'mA
<),omA
,.~.
t-
I~
a,o
L
B- 35Y
B-- 50 V
A Simple MOSFET Power Ampillier. Note use 01 boosted supplies lor driver circuitry to satisly power MOSFET
gate drive requirements
Figure 4
stable gain-crossover frequency of approximately 2
MHz. Transistor Q6 is connected in a conventional
Vbe multiplier circuit to provide adjustable bias
(nominally about 8 V) for the output stage. If thermal
feedback is required, a sensing diode placed directly in
series with the emitter of Q6 (with appropriate modification of associated resistor values) and mounted on
the heat sink will provide approximately the correct
degree of compensation. Emitter-follower drivers Q7
and Q8 provide a low-impedance drive for the gates of
power MOSFETs Q9 and QlO. The drivers isolate the
high-impedance pre-driver collector circuit from the
nonlinear input capacitance of the MOSFETs and provide adequate charge and discharge current for the
MOSFET gate circuits. The boosted supplies for all
circuits prior to the output stage enable the drive circuitry to provide adequate gate voltage to fully turn-on
the MOSFETs while maintaining margin against saturation. Zener diodes DI-D4 protect the MOSFETs
from excessive gate-source voltages of either polarity.
provide higher performance and take full advantage of
the capability of the MOSFET output stage. Although
substantially adding to the complexity of the schematic in appearance, such improvements primarily involve only small-signal, low-voltage transistors and
inexpensive passive components, and thus contribute
only a small percentage increase to total amplifier
cost.
The front-end for the amplifier to be discussed here is
shown in Figure 5. The input stage is a differential
JFET-bipolar cascode with a constant current bias _
supply. The cascode allows the use of a low-noise dual
JFET, achieving a referred input noise of less than
6 nV/-fHZ. It also provides good common-mode and
power supply rejection, necessary because negative
feedback is not very effective in reducing power supply and common-mode impairments introduced at the
input stage. The degenerated JFET input stage can
handle fairly large open-loop input signals with relatively low distortion, making the amplifier relatively
immune to transient intermodulation distortion (TIM)
and RFI effects.
High-Performance Input and Driver Circuits
As would be the case with a bipolar design as well,
many improvements can be made to the "front-end"
of the simplified amplifier of Figure 4 in order to
The input stage is loaded by current sources (Q6, Q7)
to provide high open-loop gain at low frequencies.
6-115
I
+BO.
.15
A17
A'.
470
410
A23
A'8
, K
'K
33
Q7'o.J
08.......
,....,
,....,
D'
D.
• 19
'K
I
A••
A21
'K
~
"a'.
~
...'"
Q1';;:t-:~,~.
i¥08
't~
'00
Va.
1- a'
l
1
I
c~b:
"I
IN
..
A' •
NPD5564
,
A4
R3
C4
,
R2
'00
0"
a' ......
D4
--*-
. ...
..
••
o-t
'\
.....
'00
C,
••F
~Q9
Vas
.....
+1SV
'I
C3
A"
A"
.~
...
::
"
I
.,.
o.LJ
rl
AS
'00
,,,.
'~F~ ,...22...
A12
...... FEEDBACK
FROM
OUTPUT
':"
'00
tl,· .
o,."-.J
, K
-=-
0.'o.J
n
"..,
A7
UK
STAGE
-t+-
A••
.
,7
TO
OUTPUT
22K
A'.r""
A11
-=-'7K
A'
03'
"K
'00
'7
D7
~
)015
-: ,...D'.
4.7 V
A'
•.3.
",~D'
,.K
A25
RO •
'00
470
A27
'00
-=-
- so.
-1BV
MOSFET Power Amplifier Front End. Differential cascode circuitry minimizes distortion
Figure 5
Transistors QI8 and QI9 provide regulated bias for
the cascode bases and emitter-follower collectors.
Adequate current is available so that these voltages
remain stable even under clipping conditions. Diodes
D4-D7 prevent the cascodes from saturating when the
amplifier clips. Zener diode D8 provides for two identical drive signals offset by 22 volts to allow for
biasing and error correction in the output stage.
Emitter followers Q8 and Q9 isolate the input stage
from second-stage (pre-driver) loading effects and
produce a combined common-mode feedback to properly bias Q6 and Q7. This also provides additional
common-mode rejection by reducing the commonmode impedance seen by the collectors of the input
stage. Limiter diodes D2 and D3 prevent excessive
signal swings at the collectors of Q4 and Q5 when the
amplifier is clipPing.
Overall negative feedback connections and frequency
compensation are also shown in Figure 5. Rll and
R12 set the closed-loop gain at approximately 20. The
resistance of this divider was chosen to be fairly low to
avoid noise and maintain good high-frequency
characteristics. As a result, current flow and dissipation is not insignificant (100 mW in R12 at 50-watt
operating level). To avoid thermally-induced distortion at low frequencies, these resistors should be
over-sized metal-film types, I-watt and 2-watt respectively.
The complementary pre-driver stage consists of a differential cascode (QIO-13) loaded with a Darlingtoncascode current mirror (Q14-I7) to provide a singleended drive for the output stage. The cascode achieves
high speed by eliminating Miller effect and allowing
the use of fast low voltage transistors in the commonemitter differential amplifier. Elimination of Miller
effect is also important in reducing high-frequency
distortion resulting from nonlinear collector-base
junction capacitance. [4] The cascode configuration
also improves low-frequency linearity and power supply rejection by reducing Early effect. The complementary pre-driver structure, made possible by the
current mirror, greatly reduces second-order distortion.
Feedback compensation is provided by C4 and R13
which implement rolloff feedback from the output of
the pre-driver to the inverting amplifier input, estab6-116
lishing a stable gain crossover frequency of about 2
MHz. Providing compensation by feedback to the
input stage tends to allow improved slew rate and
reduced power supply coupling; the latter because
both ends of the network are ground-referenced (in
contrast to the Miller-effect compensation of Figure
4). Elements C3 and Rl4 act to stabilize the loop
formed by C4 and R13. This front-end design enables
the amplifier to achieve a slew rate in excess of 300
Vlp-s.
35
30
25
IRF·9530
20
IRF-132
Output Stage and Error Correction
In virtually any well-designed power amplifier the
output stage ultimately limits performance. It is here
where both high voltages and large current swings are
present, necessitating larger, more rugged devices
which tend to be slower and less linear over their
required operating range. The performance-limiting
nature of the output stage is especially true in Class-B
and -AB designs, where the signals being handled by
each "half" have highly nonlinear "half-waverectified" waveforms and where crossover distortion
is easily generated. In contrast, it is not difficult or
prohibitively expensive to design front-end circuitry of
exceptional linearity.
Output Stage Transconductance versus Output Current
150mA). Reduced Total Transconductance In Central
Region can Cause Crossover Distortion.
Figure 6
<'bias
=
and less consistent, making it a significant parameter
in source-follower stages.
Figure 8 illustrates an error correction technique described by Hawksford which is well-suited to this
application. [1] Here the output stage, being a source
follower, is modeled as having exactly unity gain with
an error voltage e(x) added. This error represents any
departure from unity gain, whether it is a linear departure due to less than unity gain, a distortion due to
transconductance nonlinearity, or injected errors like
power supply ripple. A differential amplifier, represented by summer S2, merely subtracts the output
from the input of the power stage to arrive at e(x). This
error signal is then added to the input of the power
stage by summer S 1 to provide that distorted input
which is required for an undistorted output. Note that
this is an error cancellation technique like feedforward
as opposed to an error reduction technique like negative feedback. This technique is in a sense like the dual
of feedforward. It is less expensive because the point
of summation is a low-power internal amplifier node.
It is less critical of component tolerances and frequency response matching because less circuitry is
enclosed and that circuitry is simple. Feedforward
tends to become less effective at very high frequencies
because the required phase and amplitude matching
for error cancellation becomes progressively more
difficult to maintain. The technique of Figure 8 also
tends to become less effective at very high frequencies
because, being a feedback loop (albeit not a traditional
negative feedback loop), it requires some amount of
compensation for stability, detracting from the phase
and amplitude matching.
Overall negative feedback greatly improves amplifier
performance (including dynamic distortions such as
transient intermodulation distortion [4]), but it becomes
progressively less effective as the frequency or speed
of the errors being corrected increases. Highfrequency crossover notch distortion is a good example. For this reason, several high-performance
amplifier designs now employ feedforward error correction in addition to conventional negative feedback.
However, some of these designs can be complex and
expensive. The philosophy of this design is based on
the observation that only the output stage needs extra
error correction and that such local error correction
can be less complex and more effective.
While the power MOSFET has many advantages, it
was pointed out that the lower transconductance of the
MOSFET will result in considerable crossover distortion unless rather high bias currents are chosen. Figure
6 illustrates this effect by showing the individual and
summed transconductances of both halves of a
Class-AB MOSFET output stage as a function of net
output current. At a bias current of 150 rnA and a load
of 8 ohms, this transconductance variation can result
in open-loop output stage harmonic distortion on the
order of one percent, as pictured in Figures 7a and 7b.
Mismatches in the transconductance characteristics of
the top and bottom output devices also contribute to
the distortion of Figure 7. Again, while bipolar transistor transconductance is high enough and consistent
enough that it is relatively unimportant in an emitterfollower stage, MOSFET transconductance is smaller
A schematic of the MOSFET power amplifier's output
stage and error correction circuit is shown in Figure 9.
The error correction circuit is a slightly modified ver-
6-117
(a)
(b)
(\
f\
J
}
\
1\
~
:A
-A r-..., \
y
1\v
""-...: ~
\
!'\/: 'T "-
\v r
"V
--\ 'J
~ ..........
\
(c)
\
l- f\:-.
'- ~ 7
~
\
(d)
,...- .............
1---'"
"'-.
--
/
- '"
......
. . . .'--v
- -...........
L
I--
............
Output Stage Open-Loop distortion (THO); a) 1 kHz, No Error Correction; b) 20 kHz, No Error Correction; c) 1 kHz, with Error
Correction; d) 20 kHz, with Error Correction. Vertical Distortion Scale 0.5 percent/div. All Measurements at Full Power (50 W).
Figure 7
adequate to allow for the maximum bias plus VGS
signal swing required by the MOSFETs. Transistors
Q22 and Q23, in conjunction with R32 and R33,
control the DC voltage drop across R34 and R3S.
They thus set the bias for the MOSFETs by means of a
Vbe-referenced feedback loop which also includes
Q24, Q2S, Q26 and Q27. Resistors R38 and R39
control the loop gain of the bias loops and improve
stability. Overall frequency compensation of the error
correction and bias loops is provided by R36, R37,
C6, C7 and CI0.
82
)( + f:(X)
81
Output Stage Error Correction 1
Figure 8
Figures 7c and 7d show open-loop distortion of the
output stage with error correction to be less than 0.1 %,
illustrating an improvement of better than an order of
magnitude, even at 20 kHz. This was achieved with
5% tolerance resistors. While use of closer-tolerance
resistors would improve the correction at lower frequencies, where it is unnecessary, their use would
make a smaller improvement at 20 kHz because performance there is beginning to be limited by the speed
of the error correction loop. Sensitivity of 20-kHz
THD to tolerance in the error correction circuit has
been measured to be approximately 0.0002% per percent in the closed-loop amplifier. For ultimate performance, a pot can be placed between the junctions of
R38, R39 and R44, R4S.
ilion of one illustrated in reference 1. Emitter followers Q20 and Q21 isolate the high-impedance predriver output nodes from the output stage and provide
a low-impedance signal for the error correction summation process. Double emitter followers Q24, Q26
and Q2S, Q27 provide a high-current drive capability
for the MOSFET gates and isolate the error correction
summing nodes from the MOSFET gate loads. Note
that Q24 and Q25 can be fast, inexpensive smallsignal transistors. Transistors Q22 and Q23 and resistors R38-R45 comprise the differential amplifier for
summer S2 of Figure 8. The output current of these
transistors is ~ummed with the input signal (summer
S 1 of Figure 8) by means of R34 and R35. For error
correction, the top and bottom halves (Q22 , Q23)
work independently to produce identical correction
signals. Input signals offset by ±1l volts, as supplied
by the pre-driver circuit, provide DC operating voltage for these circuits. These offset voltages must be
The output stage is completed by C8, C9 and RSO-S3
for control of parasitic oscillations and Dll-D14 for
protection of the MOSFET gates from excessive drive
voltages. As mentioned in Section 2, power MOSFETs are considerably more prone to high-frequency
6-118
+5DV
0'.
R4.
I.
...
Jr-...
R34
-K
a26
~
R3.
I ••
47
ROO
47
R40
'70
022.......
R44
330
I"""
Q
ili
~
It
ill
Il'
BIAS
~: /.
~
.,
R3B
'7
Cl0~
R33
391
R3,
27
l~
R43
270
R46
1.8K
.......
T;
R45
33.
R41
27.
R51
47
C7
ii-R35
••0
~027
V
i:
100
R47
180
R4'
'7.
0231""'1
R37
180
r
O'4
O"
~.
IIf-~
+35 V
02B
':4 """011
15V
~ ~ '~'~
- - - - 0 OUT
... :"'013
15V
,.
~
014
15V
I~
029
L
-35 V
M,
10
...... 021
-SOY
MOSFET Power Amplifier Output Stage. Q22 and Q23 Provide Error Correction Signals
Figure 9
modulation) would be ineffective in detecting them.
Some of these beliefs have been shown to be
unfounded. 4-10 Nevertheless, it was decided to include some of these newer measures of distortion in
the performance evaluation.
parasitic oscillations than bipolar power transistors
because of their inherent high-speed nature and because of their substantial drain-source capacitance,
making it easy to form an efficient Colpitts oscillator
structure with inductance in the gate circuit. The
amount of series gate resistance required for suppression of parasitic oscillations grows in proportion to
the amount of inductance in the gate circuit. For highspeed output stage operation it is therefore important
to minimize this inductance. Although not employed
in Figure 9, this can be done especially well by
shielding the gate leads back to the driver transistors,
grounding the shield to the local bypass ground at each
end. Then only a lO-ohm series resistor at the driver
end and a ferrite bead at the gate end are necessary.
In spite of the error correction, which improves performance by more than an order of magnitude, transconductance variation in the output stage is still the
dominant source of distortion in this amplifier. For
this reason, output stage bias current continues to
influence performance and tradeoffs can be made. The
measurements presented here were made at a bias
current of 150 rnA, resulting in a quiescent output
stage power dissipation of 11 watts for the 50-watt
amplifier. It should also be noted that the transconductance characteristics of the N- and P-channel output devices were not matched.
Performance
This amplifier employs substantial amounts of negative feedback (40 dB at 20 kHz), and 20-kHz total
harmonic distortion (THD) was the primary performance metric used in the design process. In recent years
several new forms of distortion have been described,
sometimes in the belief that they were caused by large
amounts of negative feedback and that traditional
measures of distortion (e.g., harmonic and inter-
A word about measurement technique is in order. In
many cases the distortions being measured were below
those levels measurable by conventional equipment
and techniques. In order to add dynamic range to that
provided by the equipment employed, a distortion
magnifier circuit was utilized. This circuit scales the
output level of the noninverting amplifier under test
6·119
-
(a)
THD;,s FREQUENCY
001
0003
~
c:f 0001
i!:
00003
00001
(b)
20
50
100
200
SOD
1K
2K
FREQUENCY, Hz
SK
10K
20K
SOK
/'
Total Harmonic Distortion (THO) as a Function of Frequency
Figure 10
n
IY\
down to that of the input, subtracts the two, reintroduces 11 percent of the scaled-down amplifier
output signal, and finally multiplies the result by 9 for
presentation to the measuring equipment. The net
effect is to provide unity gain for the fundamental and
a gain of ten to distortion products generated by the
amplifier under test. Amplitude and phase balance
adjustments were incorporated into the output signal
path prior to the subtraction to achieve a fundamental
null of greater than 60 dB to frequencies beyond 20
kHz. The excellent noise and distortion performance
of the 5534-type operational amplifiers employed
make this approach effective.
001
THD
va
\
~
\
['J
/
Y.
'--
\ \.v/
l/-...
'\
ft
'-'
IY\
\
\
20 kHz Total Harmonic Distortion (THO) Products at Full Power
(50 W) a) without Error Correction (THO Analyzer Reads .02%);
b) with Error Correction (THO Analyzer Reads .0006%).
Figure 12
The latter further improves the measurement floor in
most cases. Most of the other distortion tests
employed a similar arrangement. Due to an oscilloscope calibration error, all vertical deflections in the
photographs are 6.4 percent low.
Figures 10 and 11 show total harmonic distortion as a
function of frequency and power. Dashed portions of
the curves indicate that distortion is below the residual
of the measuring system. Figure 12 shows the
appearance of the 20-kHz full-power harmonic distortion products without and with output stage error correction. Figure 13 illustrates a virtually unmeasurable
level of SMPTE* intermodulation distortion (60 &
6000 Hz, 4:1).
LEVEL
0003
~
c:f 0001
________ 20 kHz
i!'
j
I
1\
00003
1 kHz
Dynamic intermodulation distortion (DIM), a test for
measurement of transient intermodulation distortion
(TIM), is shown in Figure 14. [12] In this test a
3.18-kHz square wave and a 15-kHz sinewave are
mixed 4:1 and passed through the amplifier. A spectrum analyzer is used to measure the in-band intermodulation components. Performance is shown for
both 30-kHz and 100-kHz first-order low-pass filtering of the square wave source (DIM-30 and DIM100). As predicted by the good 20-kHz THD performance and high slew rate of this amplifier, both DIM-30
and DIM-100 distortion levels are very low; in fact,
the former is unmeasurable.
00001
10
lEVEL, WATTS
Total Harmonic
20
50
100
~istortion
(THO) as a Function of Level
Figure 11
To measure harmonic distortion, for example, a sensitive THD analyzer [11] with a 20-kHz measurement
floor of about 0.001 percent was employed in combination with this distortion magnifier to achieve a residual of about 0.0003 percent at 20 kHz, primarily
limited by noise of the power amplifier under test. The
distortion output of the analyzer was then observed
with both an oscilloscope and a spectrum analyzer.
Interface intermodulation distorion (lIM) [9] is measured by applying 1000 Hz to the amplifier under test
• Society of Motion Picture and Television Engineers.
6·120
0.01
SMPTE 1M va LEVEL
10
0....
03
0001
01
00003
003
PIM vs LEVEL
if.
i
0.0001
-------10
20
001
100
50
10
LEVEL, WATTS
SMPTE Intermodulation Distortion as a Function of Level
Figure 13
Phase Intermodulation Distortion (PIM) as a Function of Level.
Note that Phase Modulation Is Expressed in rms nanoseconds.
Figure 15
and 60 Hz to a test amplifier, each of which drives
opposite ends of an 8-ohm load resistor. A spectrum
analyzer is then used to measure distortion products at
the output of the amplifier under test. Both amplifiers
are operated at half the rated power of the amplifier
under test and distortion products are referred to the
I-kHz level at the output of the amplifier under test.
For this test the spectrum analyzer was preceded by a
modified version of the distortion magnifier to produce a magnification of 100. lIM was unmeasurable,
at less than 0.0001 percent.
0..
100
50
20
LEVEL, WATTS
Damping factor (DF) as a function of frequency is
shown in Figure 16, and is extremely high. It is high
for three reasons: 1) the power MOSFETS present
very light loading to the drivers, producing a low
open-loop output impedance essentially equal to the
inverse of their transconductance; 2) the error correction circuit tends to drive this open-loop output impedance to zero; 3) substantial overall negative feedback
further reduces the output impedance by an amount
approximating the feedback factor (40 dB at 20 kHz).
Inclusion of a parallel R-L network (0.5 n, 0.5 JLH) at
the output in series with the load for complete capacitive load stability will reduce the high-frequency
damping factor to 125 at 20 kHz.
DIM va. LEVEL
DAMPING FACTOR V$ FREQUENCV
0.1
10.0001---------------_.....
;l-
f
~
3.000
0...
~
'",.000
/D'M-l00
____1_
0.01
I
DIM-3D
300
0.003
100
10
LEVEL, WATTS
""
&0
100
20
50
100
200
500
1K
2K
5K
10K
20K
50K
100K
FREQUENCV, Hz
Dynamic Intermodulatlon Distortion (DIM-30 and DIM-100) as a
Function of Level
Figure 14
Damping Factor as a Function of Frequency
Figure 16
Phase intermodulation distortion (PIM) is shown in
Figure 15. [10] PIM is measured in the same way as
SMPTE-IM, except that phase modulation of the carrier is measured instead of amplitude modulation. The
phase modulation is then expressed in time (e.g., rms
nanoseconds) .
Although the need for this much damping factor is
doubtful, the importance of DF on frequency response
and coloration has sometimes been underestimated.
This is explained by the fact that most speaker systems
are designed assuming they will be driven by a pure
voltage source (sometimes, it seems, with limitless
6·121
(a)
(a)
(b)
(b)
I
I
t-hr
V
.....J
r--- - I f
~
j\,
- II
V
20 kHz Square Wave into a 1-0hm Resistor In Series with a
1-Microfarad Capacitor; a) Smail-Signal (1 V/dlv); b) Full-Power
(Input Bandlimited to 200 kHz). Top Trace 20 V/div. Bottom
Trace Is Output Current at 20 A/div. Timebase is 10 /Ls/div.
Figure 18
20 kHz Square Wave into an 8 Ohm Load;
a) Small-Signal (1 V/div); b) Full Power (20 V/div)
Figure 17
current capability as well!). For example, the impedance of a nominal4-ohm system may dip to 2.5 ohms
and rise to over 50 ohms at various points across the
frequency band due to driver and crossover resonances. A typical bipolar amplifier may have a damping factor of 100 (perhaps less at high frequencies),
resulting in frequency response deviations on the order
of 0.3 dB with such a load. Coloration due to low DF
may also partly explain audible differences among
vacuum-tube and low-feedback designs.
Table I summarizes overall performance of the
amplifier.
Table 1
Summary of MOSFET Power Amplifier Performance
POWER OUTPUT (RI = SO)
Figure 17 illustrates small-signal and full-power 20kHz square waves into an 8-ohm load. Figure 18
shows small-signal and full-power 20-kHz square
waves into a reactive load consisting of I-ohm and 1
p..F in series. In the full-power case the square wave
has been band-limited to 200 kHz by a first-order
low-pass filter. Figure 19 shows 500 kHz small-signal
square waves into an 8-ohm resistive load and a reactive load consisting of I-ohm and I-p..F in series. It
also shows a full-power 500 kHz square wave into an
8-ohm load. Few bipolar amplifiers would survive this
test.
TOTAL HARMONIC DISTORTION (20-20 kHz)
<0.001%
SMPTE IM DISTORTION
0.00013%
DYNAMIC lNTERMODULATION DISTORTION
(DIM·30)
(DIM· 100)
<0.006%'
0.014%
INTERFACE lNTERMODULATION DISTORTION (IIM)
0.0001%
PHASE INTERMODULATION DISTORTION (PIM)
<0.1 os
SLEW RATE
>300 VI/Ls
RISE TIME
100 os
DAMPING FACTOR (20-20 kHz)
>sooo
SIN ("A" WTD,
108 dB
re I WAIT)
* Below Measurement Floor
6-122
sow
(b)
(a)
r r---,
I"t--
".-
r r--
--,
I
r- r---.,
r- h
\
-r--
......... f---J
' - r--
I
' - t--'
(e)
,...
r- 1----\
r,---
I---,
1\
\\.. V--
\.... ---.J
500 kHz Square Wave Response; a) Smail-Signal, B-Ohm Load
(1 V/div); b) Small-SignaI1-0hm and 1-p.F Series Load (1 V/div);
e) Full-Power, B-Ohm Load (20 V/dlv.). Tlmebase is 0.5 p.s/div.
Figure 19
effectively by means of a simple error correction circuit. Although a MOSFET power amplifier can still be
expected to cost a little more, the improved characteristics seem to justify the small premium in applications where performance is important.
Conclusion
Power MOSFETs are capable of exceptional performance when used in combination with good drive circuitry and simple error correction circuitry. Their
ability to operate without complex and unreliable
safe-area limiting circuitry makes them especially
useful for demanding audio applications. Compared
with bipolar transistors, the major disadvantage of
MOSFETs (and source of distortion) seems to be the
lower transconductance, but this can be dealt with
Editors Note:
This paper is intended to illustrate design techniques
and performance achievable, and not as a construction
project.
6·123
References
[1]
M. J. Hawksford, "Distortion Correction in
Audio Power Amplifiers," 65th Convention of
the Audio Eng. Soc., preprint # 1574, February
1980.
[2]
International Rectifier HEXFET Databook, International Rectifier Corporation, Copyright
1981.
[3]
T. Holman, "New Factors in Power Amplifier
Design," J. Audio Eng. Soc., Vol. 29, No. 7/8,
pp. 517-522, July/August 1981.
[4]
R. R. Cordell, "Another View of TIM," Audio,
Vol. 64, Nos. 2 & 3, February and March 1980.
[5]
W. G. Jung, M. L. Stephens, C. C. Todd, "An
Overview of SID and TIM," Audio, Vol. 63,
Nos. 6-8, June-August 1979.
[6]
P. Garde, "Transient Distortion in Feedback
Amplifiers," J. Audio Eng. Soc., Vol. 26, No.
5, pp. 314-321, May 1978.
[7]
E. M. Cherry, "Transient Intermodulation Distortion; Part I-Hard Nonlinearity," IEEE
Trans. Aconst., Speech, Sig. Proc., Vol.
ASSP-29, pp. 137-146, April 1981.
[8]
R. R. Cordell, "A Fully In-band Multitone Test
for Transient Intermodulation Distortion," J.
Audio Eng. Soc., Vol. 29, No.9, pp. 578-586.
[9]
R. R. Cordell, "Open-Loop Output Impedance
and Interface Intermodulation Distortion in
Audio Power Amplifiers," 64th Convention of
the Audio Eng. Soc., preprint # 1537, November
1979.
[10] R. R. Cordell, "Phase Intermodulation
Distortion-Instrumentation and Measurements," J. Audio Eng. Soc., Vol. 31, No.3,
pp. 114-124, March 1983.
[11] R. R. Cordell, "Build a High-Performance THD
Analyzer," Audio, Vol. 65, Nos. 7-9, Ju1ySeptember 1981.
[12] E. Leinonen, M. Otala, J. Curl, "A Methodfor
Measuring Transient Intermodulation Distortion
(TIM)," J. Audio Eng. Soc., Vol. 25, No.4, pp.
170-177, April 1977.
6·124
6.6.4 Boost OP-AMP Output Power With
Complementary Power MOSFETs
(AN83 .. 5)
Introduction
A Current-Boosted DC Power Amplifier
Many high-quality, monolithic op-amps are available today
at low cost. They offer low noise and distortion, high slewrate and wide bandwidth. However, due to the limited
chip-size, their small output transistors cannot deliver any
appreciable power to a load.
Circuit Topology
The schematic for this simple circuit is shown in Figure 1.
It is configured as a non-inverting DC amplifier with a
closed-loop gain of 11. Drive for the power MOSFET output
stage is taken from the op-amp's power supply pins, rather
than its output pin. This may seem somewhat unconventional, but there are very good reasons for doing this. The
output pin could be used to drive a pair of MOSFETs configured as a complementary source-follower. However, the
source-follower output-stage has two major disadvantages:
a voltage gain of less than unity, and a substantial DC
threshold offset-voltage. This results in peak gate-drive
voltages, from the op-amp, several volts greater in magnitude than the output at the MOSFET sources. Also, the
output of most op-amps can only swing to within about 2
volts of the supply rails. Thus, the peak voltage swing at the
output of the source-follower is restricted to significantly
less than the power-supply rail voltages.
Higher power monolithic and hybrid op-amps are available,
but their cost is inevitably much higher than their lowpower counterparts. Also, their performance tends to
be inferior in the categories of noise, distortion and slewrate.
Clearly, a circuit topology that combines a low-power, high
quality op-amp, a pair of inexpensive complementary
power MOSFETs and a few additional components, is
desirable. This combination should be expected to have
very good performance, since power MOSFETs are inherently linear devices.
This Applications Note presents a power-boosted op-amp
circuit using Siliconix Nand P channel power MOSFETs. It
is configured as a current-boosted DC amplifier operating
from standard op-amp power supply voltages. The design
takes advantage of a new thermally-stable biasing technique for the MOSFETs. Emphasis is placed on the ease of
interfacing complementary power MOSFETs to an op-amp,
and the high performance obtainable.
The amplifier shown in Figure 1 does not have these limitations. Both the op-amp's output stage and the
MOSFET current-booster stage provide voltage gain. The
op-amp's output stage is used as an inverting commonemitter phase-splitter, while the MOSFETs are used as an
inverting common-source output-stage. Since the collectors
6-125
+20V
1.3K
+20V
20K
R10
R11
.-----.-----T---~~~
100
01
2N3904
J
+20V
/
/
03
IRF 9520
/'
02/
2N3904
R6
R7
IC 1
Vln
-------_---------''4
R1
300
OUT
R9
"':-MOSFET
BIAS
ADJUST
RS
6
TL071
100K
---
~ IBIAS = 50 rnA
1K
1K
240
RL=sn,50W
1
100K
R2
10K
04
IRF 520
100
':'
R5
NOTE:
ALL RESISTORS 1/4W, 5% (EXCEPT RLI.
VALUES IN n
UK
-20V
-20V
A Current Boosted DC Power Amplifier Using an Op-Amp, a Pair of Complementary Power MOSFETs
and a Few Additional Components
Figure 1
V+
V+
03 P-CHANNEL
DEVICE
V'out
6
TL071
OP-AMP
~
Voul
RS
/
R9
RL
':'
':'
1
04 N-CHANNEL
DEVICE
R5
V-
V-
A Simplified View of the Current Boosted Amplifier, Showing the Secondary Feedback Loop. This Is Formed
by the Op-Amp's Output Stage, the MOSFETs, and the Feedback Network RS and Rg.
Figure 2
6-126
of the output transistors in almost all monolithic op-amps
are connected to their power supply pins, they can be used
to drive the MOSFETs.
where:
Load resistors R4 and R5 are connected from the op-amp's
supply pins to the power supply rails. They generate the
necessary bias and drive voltages that the MOSFETs
require. Since these voltages are referenced to the power
supply rails rather than the output voltage, a significantly
larger peak output swing is possible. This circuit easily
generates peak gate-to-source enhancement voltages for
each MOSFET in excess of 10 volts. Thus, the output voltage swing is limited only by the peak output currer,t multiplied by the RDS(on) of the MOSFETs.
and
RSRg
«
RS +Rg
then,
Thus:
It should be noted that small resistors (1000) are included
in series with the gates of each MOSFET. These are to suppress any potential parasitic oscillation that might occur
in the circuit. The resistors should be placed as close to the
MOSFET packages as possible, for maximum suppression.
Care should be taken in the layout of the circuit to avoid
excessively long lead lengths as well as any ground-loops.
and so,
and
reA
)
reA + REI + RE3
(
reA
)
reA + REI + RE3
40
26
64
I2S
2.5
1.3
1.0
240
rnA/V
0
0
0
A/V
KO
KO
0
8 0
{J
Vout
Vin
Vout
Vin
gm3 R4RL
~
(3)
RS+Rg
124.0
1 + {J(124.0)
l/{J
RS + Rg
Rg
(4)
This analysis has been done for the upper half of the
output stage only. Since the upper and lower halves are
reasonably well-matched, this analysis will suffice for both.
Looking at Figure 2, it can be seen that a secondary feedback loop is formed by the op-amp's output stage, the
power MOSFETs and the feedback network RS and Rg. The
effect of this feedback loop is to stabilize the voltage gain
seen, going from Vin to Vout. It can be shown through
linear circuit analysis, that at low frequencies:
(
l/grnA
If, as in this case;
Quiescent bias current for the op-amp also flows through
resistors R4 and R5. Since this current is regulated internally by the op-amp, it does not affect circuit performance.
In fact, it proves beneficial because some DC bias voltage
for the MOSFETs is generated across R4 and R5. More will
be said about DC biasing for the MOSFETs later.
Vout
Vin
grnA(QA)
reA
REI
RE2
gm3(Q3)
R4
RS
Rg
RL
(1)
Vi out
Rg
RSRg (
1
)
{J = - - = - - + - (2)
V out
RS + Rg RS + Rg gm3R4RL
6-127
Equation (4) shows that the low-frequency voltage-gain of
the combined output-stage is set independently of the opamp. This additional gain reduces the voltage swing at the
output of the op-amp, and improves the bandwidth, slewrate and distortion of the amplifier. Theoretically this
improvement should be equal to 1/ {J or about a factor of 5
in this case. However, this value is only observed in the
reduction of distortion over the source-follower outputstage. The rather large input capacitance of the MOSFETs
(several hundred pF), degrades the frequency response
somewhat and slows the amplifier down. The actual
improvement in slew-rate was observed to be closer to a
factor of 3, which is still more than acceptable. The full
power bandwidth of this amplifier was well over 100KHz.
A detailed frequency response analysis of this amplifier
has not been presented here, because it is somewhat
lengthy. Suffice to say, however, that the additional gain
provided by the combined output-stage increases the
overall gain-bandwidth-product of the circuit. As well, the
MOSFET stage adds an extra pole to the open-loop transfer
function. Thus the amplifier is unstable for closed loop
gains less than about 1/ {J, or 5 in this case. If unity or very
-
low values of gain are desired, an externally compensated
op-amp such as the TL080 should be used in place of the
TL071. An external compensation capacitor can then be
chosen so that the circuit will not oscillate at the particular closed-loop gain used.
Ql and Q2 form a variable, temperature-compensated
current-sink, connected to the positive supply pin of the
op-amp. AP. trimpot RS is reduced from its maximum
value, the additional current drawn by Ql, through R4,
increases. At some point, the voltage across R4 will reach
the threshold voltage of the P-channel MOSFET and it will
begin to conduct current.
Biasing the MOSFET Pair - A New Approach
This current will tend to force the output of the amplifier
positive, if the N-channel device is conducting a lesser current. But the output of the amplifier is DC coupled back to
the inverting terminal of the op-amp. Thus, the op-amp
adjusts the voltage across R5 to force the N-channel device
to conduct the same current as the P-channel device. AP.
trimpot RS is varied, the voltage at the op-amp's output
pin can be seen to go more negative as the bias is
increased. This makes sense because more current is then
flowing out of the op-amp's negative supply pin, which
increases the gate-to-source voltage of the N-channel
MOSFET.
AP. mentioned previously, the op-amp's DC operating current generates some bias voltage for the MOSFETs, across
R4 and R5. This voltage is intentionally made less than
either MOSFETs threshold voltage. Consequently the biasing circuitry comprised of Ql and Q2. has full control over
the MOSFET idle current. The MOSFET threshold voltages
are typically 3 volts for the N-channel device and 4 volts for
the P-channel.
Texas Instruments' TL071 BIFET op-amp was chosen for
its low supply current of 2.5roA (max) as well as its high
speed and low noise.
The low supply current ensures that the voltage across R4
and R5 can be reduced to less than 3 volts. With the bias
adjust trimpot (RS) set for minimum bias, the MOSFETs
should not conduct any current. In some cases though, the
values of R4 and R5 may have to be changed to ensure that
the MOSFETs do turn off with RS set for minimum bias.
It can be seen that the op-amp is used not only as the main
gain element in this amplifier, but also as the bias controller for the N-channel MOSFET. Any variations in the
idle current ofthe P-channel MOSFET are mirrored, under
control of the op-amp, by the N-channel device. This keeps
the output centered at zero.
Table 1
Current Boosted DC Amplifier Performance
Conditions:
TA = 25°C. SUPPLIES
MOSFET BIAS CURRENT
sn
±20V. RL =
50mA. CLOSED-LOOP GAIN (AvcLl
Output Voltage Swing:
+1S.5V. -19.0V
Output Sink/Source Capability:
2.5A DC
Rise Time:
1J.LS
Slew Rate:
35 V/J.LS
Overshoot:
+ve: 15%. -ve: 20%
Output Offset:
(Equal to AVCL x VIO (op-amp))
75mV
Noise Floor (at 10Hz bandwith):
-90dB at 100mW into
THO:
0.0075% at 100Hz and 20W
0.0162% at 1 KHz and 20W
0.1350% at 10KHz and 20W
Bias Stability:
After 20W output for Y, hour the bias
decreased to 43mA.
6·128
sn
= 11
Some provision must be made to compensate for the variation in MOSFET idle-current with temperature, at fixed
gate-to-source voltages. The temperature coeffIcient of
drain-current at low bias levels is positive, which is due to
the decrease in threshold voltage with increasing temperature. This variation in threshold voltage is approximately
-5mV/aC. If the voltage across 14 is changed by this
amount as the output devices heat up, then the idle current will be thermally stable.
the threshold voltage with temperature. Consequently the
MOSFET idle current will decrease slightly as the temperature increases. This is desirable because the output stage
power dissipation at idle will then decrease as the MOSFETs
heat up. In this amplifier, the idle current decreased by
approximately 15% for a 60a C temperature rise at the
heatsink surface.
This is easily achieved by thermally bonding (i.e. glueing)
Q2 of the bias circuitry to the heatsink on which the
MOSFETs are mounted. As Q2 heats up its base-emitter
voltage decreases by about 2.2mVI" C. The effective change
in voltage across R4 with temperature is thus:
Table I lists the amplifier performance measured on a
breadboard version. Of particular note is the large output
voltage-swing possible, with a low impedance load. This
makes it suitable for use in applications such as Winchester head-actuator drive amplifIers. Here, the larger the
output voltage swing, the faster the head accelerates
across the surface of the disc.
dV (14)
dT
R4
--- x
R6
+ R7
Amplifier Performance
(-2.2mVI"C)
The excellent noise, distortion and slew-rate specifications
make this amplifier suitable for audio use as well. Supplied
by ±20V generated from a 12V battery, through a DC-to-DC
converter, two of these amplifiers would be more than
adequate as a car-stero booster-amplifier.
In this amplifier, the sum ofR6 + R7 was set to about 4200
for a MOSFET bias current of 50mA. From equation 5, this
gives an effective decrease in voltage across R4 of
- 6.9mV/ a C. This decrease is somewhat larger than that of
Small robots too, could benefit from servo power-amplifiers based on the design presented here.
6-129
6.7 Bipolar- FET Combinational Devices
Introduction
Because of the fundamental difference in operating
mechanisms, FETs and BITs have different operating characteristics. From the user's point of view,
FETs have the advantages of drive simplicity, fast
switching speed, and second breakdown ruggedness,
but have the disadvantage of higher conduction resistance. To reduce FET conduction resistance and
minimize the power loss, a larger chip area is required. For a typical 400 V device with the same
conduction drop, FETs require between 2.5 and 3
times the chip area of an equivalent BIT.
Darlington Configuration [4, 6J
In this configuration, a FE T and a BIT are connected
in Darlington fashion with the F