CCS Timing Library Characterization Guidelines Char Guide V3.4
User Manual:
Open the PDF directly: View PDF 
.
Page Count: 51
| Download | |
| Open PDF In Browser | View PDF |
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
CCS Timing Library Characterization
Guidelines
Version 3.4, October 2016
Abstract
This document describes CCS Timing library requirements and key items to
consider during library characterization. The intended audience of this
application note is library developers who characterize libraries that will be
used with post-layout RC delay calculation in the Synopsys Galaxy Sign-Off
Platform.
The information described in this document is valid for Library Compiler
version X-2005.09-SP3 and after.
Version 3.4
10/20/16
1
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
Copyright Notice and Proprietary Information
© [2006 - 2016] Synopsys, Inc. This Synopsys software and all associated documentation are proprietary
to Synopsys, Inc. and may only be used pursuant to the terms and conditions of a written license
agreement with Synopsys, Inc. All other use, reproduction, modification, or distribution of the Synopsys
software or the associated documentation is strictly prohibited.
Destination Control Statement
All technical data contained in this publication is subject to the export control laws of the United States of
America. Disclosure to nationals of other countries contrary to United States law is prohibited. It is the
reader's responsibility to determine the applicable regulations and to comply with them.
Disclaimer
SYNOPSYS, INC., AND ITS LICENSORS MAKE NO WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, WITH REGARD TO THIS MATERIAL, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
Trademarks
Synopsys company and certain product names are trademarks of Synopsys, as set forth at
http://www.synopsys.com/Company/Pages/Trademarks.aspx.
All other product or company names may be trademarks of their respective owners.
Third-Party Links
Any links to third-party websites included in this document are for your convenience only. Synopsys does
not endorse and is not responsible for such websites and their practices, including privacy practices,
availability, and content.
Synopsys, Inc.
690 E. Middlefield Road
Mountain View, CA 94043
www.synopsys.com
Version 3.4
10/20/16
2
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
Revision
Comments
1.2
Original CCS Timing release.
2.0
2.1
2.2
3.1
4.0
Version 3.4
Updated cover page, Section 2.1, Simulation Autostop and Full
Transition, and Section 7, Library Compiler Checks for CCS Timing
Libraries, on tolerance of full rail voltage check from 2% to 5% for Library
Compiler X-2005.09-SP3.
Updated Introduction section to provide background
Added new Section 3.1, Receiver Model Library Requirements, detailing
library transition time index requirements for receiver models. Sections
3.1-3.4 from previous revision incremented by 1.
Updated Section 6.2, Input Characterization Waveforms, on
recommendations for input transition time to minimize Miller effect at
startup.
Updated Section 6.7, Deriving Capacitances for the Receiver Model, with
information on truncating pre-transition data points for the Synopsys
script used to derive C1 and C2. Some material moved to Section 3.1.
Added new Section 3.2, C1 and C2 Calculation for Receiver Models,
describing requirements for calculating C1 and C2 capacitance
acquisition. Sections 3.2-3.5 from previous revision incremented by 1.
Added details about C1 and C2 characterization methods to section 6.7,
which describes characterization methods to ensure good results for C1
and C2 values.
Added guidelines for characterization index value selection
Changed segmentation_tolerance in HSPICE options to 0.005, from
0.01
Added table of contents
Added detailed explanation of driver and receiver algorithms
Added details on implementation of CCS script algorithms
Added new section on “CCS Timing Liberty Extension (June 2016)”
10/20/16
3
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
Copyright Notice and Proprietary Information
1 Introduction
1.1
THE CCS SOLUTION FOR TODAY AND TOMORROW
1.2
CCS TIMING MODEL
1.3
CHARACTERIZATION FOR CCS TIMING
2 CCS Timing Driver Model Requirements
2.1
SIMULATION AUTOSTOP AND FULL TRANSITION
2.2
ZERO CURRENT VALUES
2.3
CURRENT POLARITY
2.4
PEAK CURRENT
2.5
CURRENT VECTOR VARIABLES
2.6
NEGATIVE REFERENCE_TIME
2.7
IDENTICAL REFERENCE_TIME FOR INPUT TRANSITION TIME WITH VARYING LOADS
3 CCS Timing Receiver Model Requirements
3.1
RECEIVER MODEL LIBRARY REQUIREMENTS
3.2
C1 AND C2 CALCULATION FOR RECEIVER MODELS
3.3
RECEIVER MODELS WITH RISE AND FALL TABLES
3.4
RECEIVER MODELS WITH SINGLE-EDGE TABLES
3.5
ARC-BASED RECEIVER MODELS WITHOUT LOAD DEPENDENCY FOR A SINGLE EDGE
3.6
ARC-BASED OR PIN-BASED RECEIVER MODEL
4 Requirements for CCS Timing Voltage and Temperature Scaling Libraries
4.1
STRUCTURAL REQUIREMENTS BETWEEN CCS TIMING LIBRARIES
4.2
CHARACTERIZATION LOADS BETWEEN CCS TIMING LIBRARIES
4.3
LIBRARY VOLTAGE
5 CCS Timing General Library Requirements
5.1
DELAY AND TRANSITION TIME TRIP POINTS
5.2
ACCURACY OF LIBRARY
5.3
RANGE OF INDICES
5.4
NUMBER OF SIGNIFICANT DIGITS
6 Library Characterization for Timing Analysis
6.1
CIRCUIT SIMULATOR SETTINGS
6.1.1
Control of Simulation Time Step
6.1.2
Simulation Speed
6.1.3
Simulation Algorithm
6.2
INPUT CHARACTERIZATION WAVEFORMS
6.3
SYNOPSYS PREDRIVER WAVEFORM ALGORITHM
6.4
DELAY AND TRANSITION TIME TRIP POINTS
6.5
OPERATING RANGE OF CELLS
6.6
SELECTING CHARACTERIZATION POINTS
6.6.1
Geometric Spacing and Over-Sampling
6.6.2
Plotting of Interpolation Errors
6.7
SEGMENTING THE CURRENT WAVEFORM
6.8
DERIVING CAPACITANCES FOR THE RECEIVER MODEL
6.9
OTHER CONSIDERATIONS
6.9.1
Simultaneous Switching
6.9.2
Merging Timing Arcs
6.9.3
Three-Dimensional Timing Arcs
6.9.4
Level Shifters
6.9.5
Cells with Unbuffered Inputs
7 Library Compiler Checks for CCS Timing Libraries
8 CCS Timing Liberty Extension (June 2016)
9 References
Version 3.4
10/20/16
2
5
5
6
6
8
8
13
14
16
17
18
19
20
21
23
24
25
26
27
29
29
29
29
30
30
31
31
31
33
33
34
34
35
35
38
39
39
41
42
42
42
46
47
47
48
48
49
49
49
50
51
4
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
1 Introduction
The Composite Current Source (CCS) modeling technology is the first in the
Electronic Design Automation industry to deliver a complete open-source
current based modeling solution for timing, noise and power. Along with the
available parsers, characterization/validation tools, and guidelines, this opensource Liberty™ modeling format enables efficient characterization for cell
library creators. For IC designers, the CCS modeling technology enables
comprehensive nanometer design analysis and optimization for the first time.
Designers can reduce design margins and speed design closure by eliminating
iterations.
The widely available Non-linear delay and power models (NLDM/NLPMs) have
long been the method used to abstract delay and output transition time
characteristics of cells for usage in timing flows. At process geometries of 90nm and below, many new effects cannot be accurately modeled using this
approach. Some of these modeling challenges include:
High Impedance Interconnect
Miller effect
Dynamic IR-drop
Multi-voltage, and Dynamic Voltage and Frequency Scaling (DVFS)
Driver weakening
Temperature inversion
Increasing variations
To make matters worse, some of these effects are inter-dependent between
timing, noise and power. For example timing and slew rates affect power which
impacts IR-drop which in turn changes timing. Also signal integrity can impact
power which in-turn impacts IR-drop and timing.
1.1
The CCS Solution for today and tomorrow
With the advent of smaller nanometer leading to the issues discussed above,
the composite current source (CCS) approach of modeling cell behavior has
been developed to address the new effects of very deep submicron processes.
The CCS modeling technology consists of a single Open-Source Liberty™
library that covers current-based modeling for timing, noise and power along
with all the necessary tools and guidelines for accurate library generation and
validation.
CCS models include data specifically targeted at characterizing a cell’s timing,
noise, and power behavior. Because CCS is current-based, it enables both
temperature and voltage scaling of all cell behavior. This scaling capability can
Version 3.4
10/20/16
5
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
be applied to timing, noise and power and is more sophisticated than
conventional interpolation, thus improving accuracy. The ability to scale
accurately for voltage and temperature reduces the number of library corners
that must be characterized and greatly simplifies low power design (e.g. multivoltage and DVFS designs) flows in addition to enabling new capabilities such
as IR-Drop delay analysis.
The Composite Current Source modeling technology addresses today’s existing
and emerging design requirements—the physical effects of nanometer designs
as well as the needs of design strategies such as multiple-voltage domains. In
one model, this open-source Liberty format combines the cell data needed to
support timing, noise, and power analyses that are efficient yet accurate
because they begin with current values that are characterized for the relevant
nanometer dependencies. Because the open CCS format is extensible, these
models constitute a foundation that can be enhanced as needed to meet future
requirements such as variation-aware or statistical timing analysis.
1.2
CCS Timing Model
Timing modeling with CCS is composed of a driver model and a receiver model.
The driver model is a time- and voltage-dependent current source. Because this
current source’s drive resistance is essentially infinite, the model provides high
accuracy even when the drive resistance is much lower than the interconnect
impedance.
The CCS receiver model consists of two capacitances C1 and C2, allowing
dynamic adjustment of capacitance during the transition. The capacitance
values can also be dependent on input slew, output load and state of the cell.
The cell response (output voltage waveform) for a characterized input slew and
output load can be easily determined by integrating the current waveform.
When the cell is driving a detailed RC network, data from multiple characterized
current waveforms (for different output load values) are used to determine the
current dynamically during the transition. Sophisticated scaling of the current
waveform enables high accuracy calculation between characterized values of
input slew, output load, Vdd and Temperature.
Using these capabilities, CCS timing models achieve accuracy within 2% of
SPICE. At the same time, the models are simple enough to enable very fast
delay calculations for highly complex nanometer designs.
1.3
Characterization for CCS Timing
Characterizing a cell for CCS models is similar to characterizing nonlinear delay
models (NLDMs). The overall characterization process for timing, noise, and
power is faster, primarily because CCS enables two orders of magnitude (100X)
faster noise characterization runtime than NLDM.
Version 3.4
10/20/16
6
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
For CSS characterization, an input stimulus is chosen to produce a specific
input slew time (Sinp) ; a load capacitance (Cout) is connected to the output pin;
and a circuit simulation is run in the same way as for NLDM. But instead of
measuring voltage thresholds at the output pin, current is measured through the
load capacitor and into the input pin. The current through C out is used for the
driver model, and the current into the input pin is used to determine the receiver
model.
iinp
Sinp
iCout
Cout
These characterization experiments are repeated for a table of different S inp and
Cout combinations. The current through Cout is saved for every circuit simulation
timestep and then reduced to a much smaller set of current and time (i, t)
points. The points are chosen such that Vout(t) can be accurately reproduced for
every timestep during the transition
The current and voltage at the input pin are saved and then used to determine
C1 and C2 values such that gate-level delay calculation can closely match times
to the delay threshold and to the second slew threshold at the input pin.
So to summarize, the composite current source (CCS) approach of modeling
cell behavior has been developed to address the new effects of very deep
submicron processes. Characterization for NLDMs and CCS Timing can be
done simultaneously, allowing easy CCS Timing library data extraction from the
simulation results. In this application note, we will focus on characterization of
the CCS driver and receiver models for accurate post-layout RC delay
calculation in the Synopsys Galaxy Sign-Off Platform. Each of the following
topics will be discussed:
CCS Timing library requirements, including driver model, receiver
model, scaling, and general library requirements
Key items to consider in library characterization for meeting CCS
Timing requirements, including current waveform segmentation and
deriving receiver model capacitances
CCS Timing library screening checks that are in X-2005.09 Library
Compiler
Version 3.4
10/20/16
7
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
The intended audience of this application note is library developers who
characterize libraries that will be used with post-layout RC delay calculation in
the Synopsys Galaxy Sign-Off Platform. It is recommended that the reader be
familiar with the following CCS Timing collateral:
CCS Timing White Paper
CCS Timing Liberty Syntax
CCS Timing Library Validation Guidelines
CCS Timing Library Correlation
2 CCS Timing Driver Model Requirements
In this section, we will describe the CCS Timing driver model requirements for
current vectors in the library. For all cases, Library Compiler X-2005.09 screens
for these conditions and issues an error, warning, or informational message.
2.1
Simulation Autostop and Full Transition
For CCS Timing, the recommended characterization autostop value is when the
output pin voltage is within 1mV of rail. Speeding characterization by using
setting simulation autostop to the second transition time threshold will cut off the
current measurement before it is completed. Figure 1 shows simulation results
for the output voltage of a cell that a) has an autostop value at the second
transition time threshold (70% of 1.08V or 0.756V) and b) has an autostop value
within 1mV of the final rail voltage of 1.08 V.
Figure 1: Simulation Result with Autostop Feature
a) BAD: Autostop set to slew_upper_threshold_pct_rise value of 70%
of Vdd
Version 3.4
10/20/16
8
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
Cell Output Voltage
1.20
rail voltage = 1.08 V
1.00
slew_upper_threshold_pct_rise = 0.756 V
Voltage (V)
0.80
0.60
Simulation stops at
slew_upper_threshold_
pct_rise voltage
0.40
0.20
0.00
0
1
2
3
4
5
Time (nS)
Version 3.4
10/20/16
9
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
b) GOOD: Autostop set to Vdd – 1mV
Cell Output Voltage
1.20
rail voltage = 1.08 V
1.00
Simulation stops
within 1mV of
final rail voltage
Voltage (V)
0.80
0.60
0.40
0.20
0.00
0
1
2
3
4
5
Time (nS)
To sanity check current vector data, the current can be integrated using
the characterization capacitance to obtain the voltage. The voltage
should reach within 5% of the final rail voltage. Specifically, the formula
below should be used to integrate the current.
0.5
Vout 2 Vout1
* Iout 2 Iout1 * Time2 Time1
Cchar
The final voltage should meet the criteria below.
For a rising transition, absVout Vdd Vmargin
For a falling transition, absVout Vss Vmargin
where Vmargin = 0.05 * Vdd Vss
Beginning with the X-2005.09-SP3 version of Library Compiler, the check for
final rail voltage will be changed from 2% to 5%. See Example 1.
Example 1: Library Compiler Warning Message for Final Rail Voltage
Version 3.4
10/20/16
10
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
Warning: Line 358, The final signal voltage of the vector
is 1.08329, which does not reach within the 5 percent of
the rail voltage 1.02. (LBDB-668)
In some cases, the cell output may not be able to achieve within 5% of the final
rail voltage. As a minimal requirement, the cell should reach the voltage
specified by the second transition time threshold. Cells that do not meet this
criteria should be reviewed by the library developer. The X-2005.09 version of
Library Compiler will issue an error message for this condition. See Example 2.
Figure 2 shows a cell that does not reach full rail voltage on its output.
Example 2: Library Compiler Error Message for Second Transition Time
Threshold
Error: Line 358, The final signal voltage of the vector
is 0.700329, which does not reach beyond the 2nd slew
threshold voltage, which is 0.714000. (LBDB-669)
Version 3.4
10/20/16
11
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
Figure 2: Cell Output Voltage Does Not Reach Full Rail
Cell Output Voltage
1.20
rail voltage = 1.20 V
1.00
Voltage (V)
0.80
0.60
0.40
0.20
0.00
0
1
2
3
4
5
Time (nS)
Since the voltage waveform should at least cross the voltage specified by the
second transition time threshold, current vectors with zero current should not be
included in the library. See Example 3.
Example 3: Current Vector with Zero Current
vector ( CCS_fall ) {
reference_time : 2.172803e+00 ;
index_1 ( "0.879602" ) ;
index_2 ( "0.102702" ) ;
index_3
2.411707e+00,
3.320412e+00,
3.662858e+00,
( "1.795857e+00, 2.147012e+00,
2.641051e+00, 3.069397e+00, 3.223209e+00,
3.351926e+00, 3.403433e+00, 3.494074e+00,
3.763292e+00, 3.899033e+00" ) ;
values ( "0.000000e+00, 0.000000e+00, 0.000000e+00,
0.000000e+00, 0.000000e+00, 0.000000e+00, 0.000000e+00,
Version 3.4
10/20/16
12
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
0.000000e+00, 0.000000e+00, 0.000000e+00, 0.000000e+00,
0.000000e+00, 0.000000e+00" ) ;
2.2
Zero Current Values
The current vector should not contain zero current values. PrimeTime assumes
that the initial current is zero before the first time value in the current vector and
the final current is assumed to be the last value stored in the current vector (a
small value). Example 4 shows a current vector that stores an initial and final
current value of zero. Example 5 shows a current vector that contains a zero
value as one of its current values.
Example 4: Current Vector with Initial and Final Zero Current
vector ( CCS_fall ) {
reference_time : 2.172803e+00 ;
index_1 ( "0.879602" ) ;
index_2 ( "0.102702" ) ;
index_3
2.411707e+00,
3.320412e+00,
3.662858e+00,
( "1.795857e+00, 2.147012e+00,
2.641051e+00, 3.069397e+00, 3.223209e+00,
3.351926e+00, 3.403433e+00, 3.494074e+00,
3.763292e+00, 3.899033e+00" ) ;
values ( "0.000000e+00, -5.749690e-03, -1.797790e02, -4.025600e-02, -1.030730e-01, -1.222390e-01, 1.298390e-01, -1.305350e-01, -1.280700e-01, -1.078560e01, -4.595000e-02, -2.170950e-02, 0.000000e+00" ) ;
Example 5: Current Vector with Zero Current
vector ( CCS_fall ) {
reference_time : 2.172803e+00 ;
index_1 ( "0.879602" ) ;
index_2 ( "0.102702" ) ;
index_3
2.411707e+00,
3.320412e+00,
3.662858e+00,
Version 3.4
10/20/16
( "1.795857e+00, 2.147012e+00,
2.641051e+00, 3.069397e+00, 3.223209e+00,
3.351926e+00, 3.403433e+00, 3.494074e+00,
3.763292e+00, 3.899033e+00" ) ;
13
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
values ( "-6.393840e-03, -5.749690e-03, -1.797790e02, -4.025600e-02, -1.030730e-01, -1.222390e-01, 1.298390e-01, -1.305350e-01, -1.280700e-01, -1.078560e01, -4.595000e-02, 0.000000e+00, -2.170950e-02" ) ;
Library Compiler X-2005.09 will issue an error message if there is a zero value
in the current vector. See Example 6.
Example 6: Library Compiler Error Message for Zero Value in Current
Vector
Error: Line 381, The 'values' attribute of the 'vector'
has a value '0.000000', which is equal to '0.000000'.
(LBDB-664)
2.3
Current Polarity
For the output_current_rise group, the current values must be positive in
the values section of the current vector. Similarly, for the
output_current_fall group, the current values must be negative.
The PMOS devices in a cell with rising output will cause the output current of the
cell to be positive. Similarly, NMOS devices sink current and will cause the
output current of a cell to be negative. In Library Compiler X-2005.09, an error
will be issued for current vectors of this type and the library will not compile. See
Example 7.
Version 3.4
10/20/16
14
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
Example 7: Library Compiler Error Message for Mixed Polarities in
Current Vector
Error: Line 75, The 'values' attribute of the 'vector'
has a value '-0.196476', which should be positive for the
output_current_rise vector. (LBDB-664)
Example 8 shows that the second current vector in the
output_current_rise section has negative current values, which does not
meet the criteria for current polarity.
Example 8: Incorrect Current Polarity in Current Vector
output_current_rise () {
vector ( CCS_rise ) {
reference_time : 1.018468e+00 ;
index_1 ( "0.0100" ) ;
index_2 ( "0.104" ) ;
index_3
1.030994e+00,
1.447324e+00,
2.044892e+00,
( "1.026000e+00, 1.030780e+00,
1.034013e+00, 1.213171e+00, 1.316816e+00,
1.584940e+00, 1.715004e+00, 1.868917e+00,
2.377380e+00" ) ;
values ( "1.277399e-01, 2.053170e-01, 2.381160e-01,
2.377790e-01, 2.142830e-01, 1.894730e-01, 1.310250e-01,
7.119820e-02, 3.582050e-02, 1.474390e-02, 5.060350e-03,
7.094180e-04" ) ;
}
vector ( CCS_rise ) {
reference_time : 1.018468e+00 ;
index_1 ( "0.0100" ) ;
index_2 ( "0.465" ) ;
index_3
1.035269e+00,
3.435500e+00,
6.505548e+00"
( "1.026178e+00, 1.030994e+00,
1.800915e+00, 2.275368e+00, 2.837299e+00,
4.019846e+00, 4.659704e+00, 5.573401e+00,
) ;
values ( "1.341709e-01, 2.441400e-01, 2.440120e-01,
2.213040e-01, -1.964760e-01, -1.396400e-01, 7.716020e-02,
Version 3.4
10/20/16
15
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
3.794630e-02, 1.600170e-02, 5.013170e-03, 1.570870e-03" )
;
}
2.4
Peak Current
The first current value in the current vector should not be the peak of the current
waveform, nor should the last value in the current vector be the peak of the
current waveform. For rising outputs, the peak value is the most positive value
in the current vector, and for falling outputs, the peak value is the most negative
value in the current vector. Possible causes for the peak current as the first
point in the current vector are characterization transition times that are less than
the minimum transition time for the process (see Section 6.4) or simulation
issues, such as large simulation time steps or accuracy settings. Library
Compiler X-2005.09 will issue an error message for this condition. See Example
9. Example 10 and Figure 3 illustrate a current vector with these characteristics.
Example 9: Library Compiler Error Message for Peak Current as Initial
Point in Current Vector
Error: Line 95, The first value of the 'values' attribute
of the 'vector' is the peak value. (LBDB-665)
Example 10: Current Vector with Peak Current as Initial Point
output_current_rise () {
vector ("CCS_rise") {
reference_time : 4.015000e+00;
index_1("0.024000");
index_2("0.07900");
index_3(“4.760000e+00,4.760459e+00,4.801981e+00,5.302611e
+00,5.818013e+00,6.289269e+00,6.992352e+00,7.785586e+00")
;
values("0.963412e+00,0.062298e+00,0.060605e+00,0.051152e+
00,0.030261e+00,0.011958e+00,0.002145e+00,0.000454e+00");
}
Figure 3: Peak Current as the First Point in a Current Vector
Version 3.4
10/20/16
16
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
CCS Current Vector
0.250
Current (mA)
0.200
0.150
0.100
0.050
0.000
0.00
0.50
1.00
1.50
2.00
2.50
Time (nS)
2.5
Current Vector Variables
The template specifying composite current source driver models must always
have time as the value for variable_3. The input_net_transition and
total_output_net_capacitance values can be assigned to either
variable_1 or variable_2. See Example 11. If variable_3 does not
have the value time, Library Compiler X-2005.09 will issue an error message
for this condition. See Example 12.
Example 11: Correct CCS Driver Model Templates
output_current_template ( CCS_rise ) {
variable_1 : input_net_transition ;
variable_2 : total_output_net_capacitance ;
variable_3 : time ;
output_current_template ( CCS_fall ) {
Version 3.4
10/20/16
17
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
variable_1 : total_output_net_capacitance ;
variable_2 : input_net_transition ;
variable_3 : time ;
Example 12: Library Compiler Error Message for variable_3 with
input_net_transition Value
Error: Line 198, The index of 'input_net_transition' can
define only one value. (LBDB-653)
2.6
Negative reference_time
The reference_time attribute is the time at which the input voltage
characterization waveform crosses the input delay threshold. The start time for
a CCS Timing characterization is the time at which the voltage transition on the
input to the cell starts. This precedes the reference_time. The
reference_time is used to calculate the cell delay in CCS delay calculation.
See Figure 4 for an example of reference_time.
Version 3.4
10/20/16
18
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
Figure 4: reference_time Definition
Input Voltage Waveform
tsim = tdelay_threshold =
reference_time
tsim = 0
tsim =
tstart
Since reference_time is related to physical circuit behavior, Library Compiler
X-2005.09 will issue an error if a negative reference_time is used in the
library. See Example 13.
Example 13: Library Compiler Error Message for Negative
reference_time
Error : Line 491, The 'reference_time' of the 'vector' is
-0.419265, which should be positive. (LBDB-667)
2.7
Identical reference_time for Input Transition Time with Varying
Loads
The reference_time attribute must be identical for the same input transition
time and edge (either rise or fall) across all capacitive loads. This ensures that
the input characterization waveform is consistent across all characterization
loads. Library Compiler will flag this as a compilation error if this condition is not
met. Example 14 shows that for an input transition time of 20pS, the
reference_time for a load of 0.104pF is 1.018468nS and is 1.018469nS for a
load of 0.465pF. Example 15 shows the error message in Library Compiler for
this condition.
Version 3.4
10/20/16
19
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
Example 14: Inconsistent reference_times for Varying Loads with
Identical Input Transition Time
output_current_rise () {
vector ( CCS_rise ) {
reference_time : 1.018468e+00 ;
index_1 ( "0.0200" ) ;
index_2 ( "0.104" ) ;
index_3 ( "1.026000e+00, 1.030780e+00,
1.030994e+00, 1.034013e+00, 1.213171e+00, 1.316816e+00,
1.447324e+00, 1.584940e+00, 1.715004e+00, 1.868917e+00,
2.044892e+00, 2.377380e+00" ) ;
values ( "1.277399e-01, 2.053170e-01, 2.381160e-01,
2.377790e-01, 2.142830e-01, 1.894730e-01, 1.310250e-01,
7.119820e-02, 3.582050e-02, 1.474390e-02, 5.060350e-03,
7.094180e-04" ) ;
}
vector ( CCS_rise ) {
reference_time : 1.018469e+00 ;
index_1 ( "0.0200" ) ;
index_2 ( "0.465" ) ;
index_3 ( "1.026178e+00, 1.030994e+00,
1.035269e+00, 1.800915e+00, 2.275368e+00, 2.837299e+00,
3.435500e+00, 4.019846e+00, 4.659704e+00, 5.573401e+00,
6.505548e+00" ) ;
values ( "1.341709e-01, 2.441400e-01, 2.440120e-01,
2.213040e-01, 1.964760e-01, 1.396400e-01, 7.716020e-02,
3.794630e-02, 1.600170e-02, 5.013170e-03, 1.570870e-03" )
;
}
Example 15: Inconsistent reference_time Error Message in Library
Compiler
Error: Line 115, The values of 'reference_time' and
'input_net_transition'are not consistent in group
'output_current_rise'. (LBDB-654)
3 CCS Timing Receiver Model Requirements
There are multiple ways to create composite current source receiver models, for
example, pin-based and arc-based. It is recommended that a full receiver
model with rise and fall information should always be generated. The following
sections will discuss the requirements for receiver model library tables as well
as receiver model characterization considerations. These are discussed
Version 3.4
10/20/16
20
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
because characterizing the full receiver model for some cells may be
problematic. For these situations, additional syntax for the receiver model is
supported. The method chosen can impact the quality of results.
3.1
Receiver Model Library Requirements
The CCS Timing receiver model uses two capacitance values to model the
variation of cell input capacitance during the input signal transition. For
Primetime delay calculation, the C1 value is used before the delay trip point has
been reached in the transition and the C2 value is used after the delay trip point
has been reached. This is true for both rise and fall transitions. See Figure 5.
Figure 5: Receiver Model Capacitances
C2
Delay Trip Point
C1
C1
Delay Trip Point
a) Rising Input
C2
b) Falling Input
The receiver_capacitance*_rise and receiver_capacitance*_fall
library syntax always follows the transition occurring on the input of the cell. The
C1 capacitance value is recommended to be chosen such that the current
produces a voltage that matches the value at the input delay trip point of the
cell. Similarly, the C2 capacitance value is recommended to be chosen such
that the current produces a voltage that matches the second slew trip point of
the cell.
Since the C1 and C2 values are dependent on a portion of the input waveform
that is smaller than that covered by the library transition time, the receiver model
tables should reflect this. Receiver model transition time indexes should use the
partial transition time of the input waveform and extrapolate these values to the
range of the library transition time. Figure 6 shows an input characterization
waveform that has different input transition time values for C1 and C2.
Version 3.4
10/20/16
21
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
Figure 6: Characterization Waveform with Different Transition Time
Values for C1 and C2
C2, s2
Upper Transition
Time Rise Trip Point
Delay Trip Point
C1, s1
Lower Transition
Time Rise Trip Point
To derive the final library transition time index for the C1 and C2 tables, do the
following.
1. Add measure statements to the characterization simulation deck to
measure s1 and s2, the transition times corresponding to the part of the
waveform used for C1 and C2. See Examples 16 and 17.
Example 16: Measuring s1 and s2 for Rising Input Waveform
.meas tran s1r trig v(in) val=’slew_lower_thresh_voltage_rise’ rise=1
targ v(in) val=’input_delay_thresh_voltage_rise’
rise=1
.meas tran s2r trig v(in) val=’input_delay_thresh_voltage_rise’ rise=1
targ v(in) val=’slew_upper_thresh_voltage_rise’
rise=1
Example 17: Measuring s1 and s2 for Falling Input Waveform
.meas tran s1f trig v(in) val=’slew_upper_thresh_voltage_fall’ fall=1
targ v(in) val=’input_delay_thresh_voltage_fall’
fall=1
.meas tran s2f trig v(in) val=’input_delay_thresh_voltage_fall’ fall=1
targ v(in) val=’slew_lower_thresh_voltage_fall’
fall=1
2. Extrapolate the measured s1 and s2 values to the full library transition
time range using the formulas below. This ensures that all receiver model
transition time indexes are across the defined library transition time trip points.
Version 3.4
10/20/16
22
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
receiver_capacitance1_rise transition time index =
s1r *
slew _ upper _ threshold _ pct _ rise slew _ lower _ threshold _ pct _ rise
lib _ slew _ derate * input _ threshold _ pct _ rise slew _ lower _ threshold _ pct _ rise
receiver_capacitance2_rise transition time index =
s 2r *
slew _ upper _ threshold _ pct _ rise slew _ lower _ threshold _ pct _ rise
lib _ slew _ derate * slew _ upper _ threshold _ pct _ rise input _ threshold _ pct _ rise
receiver_capacitance1_fall transition time index =
s1 f *
slew _ upper _ threshold _ pct _ fall slew _ lower _ threshold _ pct _ fall
lib _ slew _ derate * slew _ upper _ threshold _ pct _ fall input _ threshold _ pct _ fall
receiver_capacitance2_fall transition time index =
s2 f *
slew _ upper _ threshold _ pct _ fall slew _ lower _ threshold _ pct _ fall
lib _ slew _ derate * input _ threshold _ pct _ fall slew _ lower _ threshold _ pct _ fall
Note that if ramps are used for input characterization waveforms, the receiver
model transition time indexes will be identical to the driver model transition time
indexes.
3.2
C1 and C2 Calculation for Receiver Models
For optimal CCS receiver accuracy, The value of C1 in Figure 7 must be chosen
to ensure that the point (V2,t2) is exactly matched. The value of C2 must be
chosen to ensure that the point (V3,t3) is exactly matched. The recommended
characterization methods for C1 and C2 are described in Section 6.7. The
Synopsys CCS characterization “jumpstart” kit characterization methods are
detailed there.
Figure 7: Input Waveforms for rising and falling slews
Version 3.4
10/20/16
23
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
Second Slew Threshold (V3, t3)
C2
Input Delay Trip Point (V2, t2)
C1
First Slew Threshold (V1, t1)
First Slew Threshold (V1, t1)
C1
Input Delay Trip Point (V2, t2)
C2
Second Slew Threshold (V3, t3)
3.3
Receiver Models with Rise and Fall Tables
Cells with library attributes of timing_type : rising_edge or
timing_type : falling_edge (rising- or falling-edge triggered flip-flops)
are recommended to have receiver models with both
receiver_capacitance*_rise and receiver_capacitance*_fall
information. This is typically the CLK pin of a register cell. The CLK pin of a
register can be a side load on a clock network with rising and falling edge
triggered flip-flops, see Figure 8 below.
Figure 8: Rising- and Falling-Edge DFFs on a Clock Network
Version 3.4
10/20/16
24
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
F1
CLK
F2
If a cell is a side load on a net and is missing the rise or fall capacitance tables,
PrimeTime will use the NLDM lumped pin capacitance instead of the CCS
Timing receiver model. This is why you should provide both rise and fall
receiver model tables.
The same consideration should be given to half unate arcs such as
combinational_rise and combinational_fall arcs, as well as preset
and clear timing arcs.
3.4
Receiver Models with Single-Edge Tables
In some cases, a net may only have one transition of interest, or the receiver
model edge for when the cell is a side load may not be an important design
consideration. For example, all sequential cells in a design may have positive
unate clear arcs. In this case, receiver model tables for rise will not be of
interest. For half unate timing arcs, receiver models of this type will be compiled
and used in PrimeTime. If a pin has missing receiver model information for an
edge, the NLDM lumped pin capacitance will used instead. Library Compiler will
issue a warning for single edge receiver models of half unate timing arcs. See
Example 18. A receiver model with rise and fall information is required for full
unate timing arcs. Library Compiler will issue an error message if an edge is
missing. See Example 19.
Example 18: Library Compiler Warning Message for Single Edge
Receiver Model of Half Unate Timing Arc
Warning: Line 923, The timing arc does not have full
receiver modeling information. (LBDB-673)
Example 19: Library Compiler Error Message for Single Edge Receiver
Model of Full Unate Timing Arc
Error: Line 148, The 'receiver_capacitance1_rise' is
missing for this timing arc. (LBDB-267)
Version 3.4
10/20/16
25
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
3.5
Arc-Based Receiver Models Without Load Dependency for a
Single Edge
Some cells may have load dependency for one edge of the receiver model, but
not the other. For example, the fall receiver model for the CLK pin of a rising
edge flip-flop will have little load dependency. In this case, one capacitance
index can be used in the arc-based receiver model. See Example 20.
Example 20: Arc-Based Receiver Model without Fall Load Dependency
receiver_capacitance1_rise ( rcvr_rise_1 ) {
values ( \
"1.947000e-03, 1.959000e-03, 1.988000e-03,
2.016000e-03, 2.034000e-03",\
"1.960000e-03, 1.969000e-03, 1.994000e-03,
2.020000e-03, 2.037000e-03",\
"1.992000e-03, 1.997000e-03, 2.012000e-03,
2.030000e-03, 2.043000e-03",\
"2.031000e-03, 2.032000e-03, 2.035000e-03,
2.043000e-03, 2.051000e-03",\
"2.185000e-03, 2.184000e-03, 2.179000e-03,
2.171000e-03, 2.161000e-03" \
) ;
}
receiver_capacitance2_rise ( rcvr_rise_2 ) {
values ( \
"2.251000e-03, 2.223000e-03, 2.169000e-03,
2.129000e-03, 2.107000e-03",\
"2.287000e-03, 2.255000e-03, 2.190000e-03,
2.141000e-03, 2.114000e-03",\
"2.381000e-03, 2.340000e-03, 2.253000e-03,
2.182000e-03, 2.138000e-03",\
"2.536000e-03, 2.474000e-03, 2.349000e-03,
2.247000e-03, 2.182000e-03",\
"2.782000e-03, 2.682000e-03, 2.488000e-03,
2.337000e-03, 2.243000e-03" \
) ;
receiver_capacitance1_fall ( rcvr_fall_1 ) {
index_1 ( "0.065, 0.110, 0.400, 0.900, 1.50” ) ;
index_2 ( "0.000" ) ;
values ( \
"1.882000e-03”,\
“1.887000e-03”,\
“1.901000e-03”,\
Version 3.4
10/20/16
26
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
“1.916000e-03”,\
“1.927000e-03" \
) ;
}
receiver_capacitance2_fall ( rcvr_fall_2 ) {
index_1 ( "0.065, 0.110, 0.400, 0.900, 1.50” ) ;
index_2 ( "0.000" ) ;
values ( \
"2.189000e-03”,\
“2.169000e-03”,\
“2.129000e-03”,\
“2.100000e-03”,\
“2.084000e-03" \
) ;
}
3.6
Arc-Based or Pin-Based Receiver Model
If both rise and fall tables of an arc-based receiver model show little load
dependency, a pin-based receiver model should be used. This type of behavior
would be expected for multistage cells and the CLK pin of sequential cells. See
Examples 21 and 22 for an arc-based receiver model that has been converted to
a pin-based receiver model.
Example 21: Arc-Based Receiver Model with No Load Dependency
lu_table_template ( rcvr_rise_1 ) {
variable_1 : input_net_transition ;
variable_2 : total_output_net_capacitance ;
index_1 ( "0.065, 0.110, 0.400, 0.900, 1.50” ) ;
index_2 ( "0.002, 0.010, 0.100, 0.500, 0.700" ) ;
}
pin ( Z ) {
timing ( ) {
related_pin : "A" ;
timing_sense : negative_unate ;
receiver_capacitance1_rise ( rcvr_rise_1 ) {
Version 3.4
10/20/16
27
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
values ( \
"1.947000e-03, 1.947000e-03, 1.947000e-03,
1.947000e-03, 1.947000e-03",\
"1.960000e-03, 1.960000e-03, 1.960000e-03,
1.960000e-03, 1.960000e-03",\
"1.992000e-03, 1.992000e-03, 1.992000e-03,
1.992000e-03, 1.992000e-03",\
"2.031000e-03, 2.031000e-03, 2.031000e-03,
2.031000e-03, 2.031000e-03",\
"2.185000e-03, 2.184000e-03, 2.184000e-03,
2.185000e-03, 2.185000e-03" \
) ;
} /* end of arc-based receiver_capacitance1_rise */
Example 22: Corresponding Pin-Based Receiver Model
lu_table_template ( rcvr_rise_1 ) {
variable_1 : input_net_transition ;
index_1 ( "0.065, 0.110, 0.400, 0.900, 1.50” ) ;
}
pin ( A ) {
receiver_capacitance() {
receiver_capacitance1_rise ( rcvr_rise_1 ) {
values ( \
"1.947000e-03",\
"1.960000e-03",\
"1.992000e-03",\
"2.031000e-03",\
"2.185000e-03" \
) ;
} /* end of pin-based receiver_capacitance1_rise */
receiver_capacitance2_rise() {
. . . .
} /* end of pin-based receiver_capacitance2_rise */
receiver_capacitance1_fall() {
. . . .
} /* end of pin-based receiver_capacitance1_fall */
receiver_capacitance2_fall() {
. . . .
} /* end of pin-based receiver_capacitance2_fall */
} /* end of pin-based receiver model */
Version 3.4
10/20/16
28
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
4 Requirements for CCS Timing Voltage and
Temperature Scaling Libraries
During the process of creating the scaling relationships, PrimeTime runs
consistency checking between the libraries in each scaling group. There are
some requirements that the libraries need to satisfy, in order for scaling to work
correctly
4.1
Structural Requirements Between CCS Timing Libraries
The libraries must be structurally identical. For example, the cell names, pin
names, and order of timing arcs should be the same. Ensure that the set of arcs
and pins is identical across the members of the scaling library groups. The
positions of the arcs and pins within the libraries must be the same as well. The
number of output-capacitance indices for all timing arcs should be identical.
If libraries are not structurally identical, scaling relationships will not be created
and warning message SLG-202 is issued by Primetime:
Warning: Completion of scaling library groups failed.
(SLG-202)
4.2
Characterization Loads Between CCS Timing Libraries
The output-capacitance indices used for Composite Current-Source (CCS)
driver data must have identical values across the libraries as well. This is not a
requirement for CCS timing receiver models or constraints. If this requirement is
not satisifed, the following warning will be issued by Primetime:
Warning: The members of a scaling library group have
significantly different output-capacitance indices; this
can adversely affect scaling accuracy. (SLG-203)
4.3
Library Voltage
The nominal voltage and the voltage specified in the default operating condition
of the library must be identical. See Example 23.
Example 23: Nominal Voltage and Operating Condition Voltage Identical
in the Library
library (EXAMPLE) {
nom_voltage : 1.050000;
operating_conditions (MAX) {
process : 1.000000;
Version 3.4
10/20/16
29
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
temperature : 125.000000;
voltage : 1.050000;
}
default_operating_conditions : MAX ;
If an operating condition is not defined in the library, Library Compiler X2005.09 will issue a warning message, create an operating condition, and specify
it as the default operating condition that matches the nominal voltage in the
library, as shown in Example 24.
Example 24: Missing Default Operating Condition Warning Message in
Library Compiler
Warning: Line 13, The default_operating_conditions is not
defined. operating_conditions "nom_pvt" is created and
set as the default_operating_conditions. (LBDB-662)
5 CCS Timing General Library Requirements
In this section, we will describe general guidelines for creating accurate CCS
Timing libraries.
5.1
Delay and Transition Time Trip Points
Unlike NLDM, CCS Timing accuracy is not impacted by output transition time
and delay trip point selection for a given input transition time. This is because
CCS can calculate the voltage waveform for the entire transition, making the
result independent of trip points. Thus, selecting transition time trip points by
correlating cell output transition time and delay is not required. However,
selection of transition time trip points should take into account accuracy for
transition time propagation down a timing path. Usually the best correlation
between input transition time and output behavior is obtained when the
transition time is measured in the linear region about the logic threshold. Unlike
NLDM, CCS will tolerate asymmetrical transition time trip points chosen within
5% of the delay trip points. A library with 25%-55% transition time trip points for
rise and 75%-45% transition time trip points for fall with rise and fall delay trip
points of 50% is not recommended for NLDM, but is acceptable for CCS-Timing
with adequate library validation.
The delay and transition time trip points used during characterization should
match those specified in the library. Table 1 shows the variables used to specify
trip points in the library header.
Version 3.4
10/20/16
30
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
Table 1: Library Variables for Delay and Transition Time Trip Points
Delay
input_threshold_pct_rise
output_threshold_pct_rise
input_threshold_pct_fall
output_threshold_pct_fall
Transition Time
slew_upper_threshold_pct_rise
slew_lower_threshold_pct_rise
slew_upper_threshold_pct_fall
slew_lower_threshold_pct_fall
If a library contains transition times extrapolated beyond the characterization
range, the slew_derate_from_library variable should be set in the header
of the library For a library characterized with 30%-70% trip points and
extrapolated to 10%-90% trip points, the slew derate is calculated using the
formula below.
slew _ derate _ from _ library
5.2
0.7 0.3 0.5
simulation _ trip _ pt _ range
extrapolat ed _ trip _ pt _ range 0.9 0.1
Accuracy of Library
The library should have sufficient indices, and the selection of the indices should
be such that the interpolation error compared to the reference circuit simulator is
within acceptable tolerance. For more details on this topic, please see the CCS
Timing Library Correlation document.
5.3
Range of Indices
The range of indices for input transition time and output capacitance must
expand across the entire range of allowable values when the cell is used in a
design. Make sure that the minimum input transition time and output
capacitances extend to minimum values allowed in designs. If the library will be
used for signal integrity analysis, some consideration should be given to
extending the characterization range. Cells should not be used outside the
characterization range, as this may cause large differences in delay and output
transition time compared to the reference circuit simulator. Extrapolations
outside the table should be avoided.
5.4
Number of Significant Digits
Current vector data can be represented with scientific notation in the
reference_time attribute, index_3 variable, and values attributes. This is
not essential, but will preserve digits in the current waveform with a small load.
It is recommended to use at least 6 significant digits. Example 25 shows a
current vector using scientific notation.
Version 3.4
10/20/16
31
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
Example 25: Current Vector with Scientific Notation and 6 Significant
Digits
vector ( CCS_fall ) {
reference_time : 1.012923e+00 ;
index_1 ( "0.009692" ) ;
index_2 ( "0.102702" ) ;
index_3 ( "1.028487e+00, 1.032531e+00,
1.034390e+00, 1.156875e+00, 1.269003e+00, 1.353348e+00,
1.529358e+00, 1.627232e+00, 1.750080e+00, 2.095409e+00" )
;
values ( "-1.710798e-01, -2.676370e-01, -2.663900e01, -2.564110e-01, -2.358110e-01, -1.983590e-01, 7.114770e-02, -2.992160e-02, -8.843480e-03, -5.037530e05" ) ;
}
For the receiver model, the capacitance values in the table should have a
minimum of four non-zero significant digits. Example 26 illustrates a receiver
model table that meets this criteria. If the library does not meet the guideline for
significant digits, Library Compiler X-2005.09 will output an informational
message as shown in Example 27.
Example 26: Receiver Model with 4 Non-Zero Significant Digits
receiver_capacitance1_fall
values ( \
"0.003689, 0.003810,
"0.004035, 0.003851,
"0.005041, 0.003921,
) ;
}
( CCS_fall_1 ) {
0.003819", \
0.003829", \
0.003855" \
Example 27: Library Compiler Informational Message for Significant Digits
Information: Line 10934, The 'vector' has 'values'
0.000958, which has less than 4 significant digits.
(LBDB-670)
Version 3.4
10/20/16
32
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
6 Library Characterization for Timing Analysis
During library characterization and generation, there are many factors that must
be considered to ensure accurate timing information in the library. This section
discusses the following key items.
•
Circuit simulator settings
•
Input characterization waveforms
•
Delay and transition time trip points
•
Operating range of cells
•
Selecting characterization points
•
Segmentation tolerance
•
Receiver modeling
•
Other considerations
6.1
Circuit Simulator Settings
During the characterization process, a reference or “golden” circuit simulator
such as HSPICE is used to simulate the cell. It is important to simulate with
appropriate settings to achieve the best accuracy. In general, the accuracy
settings for the spice simulator need to be tighter than what is traditionally used
for NLDM library characterization.
Using the HSPICE option “accurate” does a good job of achieving the needed
accuracy but will significantly degrade the simulator performance. This should
be avoided unless other less costly options do not achieve the desired
accuracy. Use of the HSPICE setting, “.option runlvl=5” achieves a good
accuracy vs. performance tradeoff. Improper accuracy settings typically result in
too few sample points or simulation noise in the CCS current waveform.
Simulation noise usually appears as oscillations in either the first 10% of the
waveform or the last 30% of the waveform. This can be viewed by importing
the output of HSPICE simulation into a graphic post processing tool such as
CosmosScope. If tightening of the simulation accuracy does not correct this
situation, then the problem may be caused by the process model or voltage or
temperature settings. This will need to be corrected to obtain a valid CCS
waveform. Additional information on setting accuracy options for HSPICE can
be found in the HSPICE Simulation and Analysis User Guide. For simulators
other than Hspice, the same method can be used to obtain optimal settings for
accuracy and performance.
Version 3.4
10/20/16
33
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
6.1.1
Control of Simulation Time Step
For HSPICE, dynamic time step controls may be used. For more details, please
see reference [3]. This method ensures that enough points are captured when
voltages in the circuit are changing quickly and reduces the number of points
when voltages in the circuit change slowly. This can be achieved by setting a
large simulation time step in the .tran statement and setting the absvar and
relvar variables in the .option command to control the simulation time
steps. Example 28 shows the simulation commands that should be included in
an HSPICE deck to enable dynamic time stepping. Segmentation tolerance will
be discussed in Section 6.6.
Example 28: Using Adaptive Time Stepping in HSPICE
.param segmentation_tolerance=0.005
.param vswing=’vdd-vss’
.param relative=’segmentation_tolerance/vswing’
.option accurate absvar=segmentation_tolerance
relvar=relative
.option runlvl=5
.tran 10ps 1.0u
If the simulator used for library characterization does not have adaptive time
step control settings like absvar and relvar, small enough fixed time steps
should be used such that the maximum voltage change per time step is less
than the segmentation tolerance.
6.1.2
Simulation Speed
If the .tran simulation setting in Example 28 is used, autostop should be
included in the Spice deck to minimize simulation runtime, otherwise a total time
of 1 microsecond will be simulated. The autostop feature terminates the
simulation after all .meas statements have been completed. If characterizing
for CCS, .meas statements within 1mV of the final rail voltage on the input and
output of the cell should be added to the Spice deck. Example 29 shows the
autostop and .meas statements for an inverter with a rising output. If the cell
has previously been characterized for NLDM, autostop may already be set at
the second transition time threshold and should be changed when
characterizing for CCS timing. See Section 2.1 for more details.
Example 29: Using autostop to Speed Simulation
Version 3.4
10/20/16
34
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
.option autostop
.meas tran tstopi
when v( input)=’vss+0.001’
.meas tran tstopo
when v( output)=’vdd-0.001’
6.1.3
fall=1
rise=1
Simulation Algorithm
Simulator transient analysis has several methods for numerical integration. The
trapezoidal algorithm should be used for library characterization. It provides
the highest accuracy and fastest simulation time. The gear method should be
avoided, since it is targeted for inductive or lossy circuits and is not as accurate.
The trapezoidal method is the default numerical integration algorithm in
HSPICE.
6.2
Input Characterization Waveforms
In an actual design, the waveform seen at a cell’s input pin arrives from any cell
in the library through a complex RC network. All of the input waveforms along a
design path will differ from the pre-set stimulus waveform used for
characterization. It is necessary to create a stimulus waveform that minimizes
the error introduced by this difference.
In order to reduce potential optimism introduced by characterization waveforms,
min libraries used for min analysis should be characterized using the fastest
waveform shape or the linear ramp method. Max libraries used for max analysis
should be characterized using waveform shapes that resemble those in an
actual design. Traditional library characterization has typically used the output of
another cell, known as an “active driver” to generate the waveform shapes. The
active driver method is not recommended for several reasons.
The first reason is that there are many different driver/load cell combinations in a
design. Choosing a specific driver improves accuracy for receiver cells
connected to that cell at the expense of accuracy for receiver cells connected to
other driver cells. Empirical results with many libraries has shown that overall
correlation accuracy is greater with the Synopsys predriver waveform shown
below than with active driver cells, especially for worst-case results.
The second reason is that the use of an active driver can introduce errors
depending upon the method chosen to generate a desired output slew. These
methods include:
Version 3.4
Sweep output capacitance for fixed input slew
Sweep input slew for fixed output capacitance
Sweep both
10/20/16
35
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
The shape of the output waveform is different in the above cases even if the
slew values are exactly the same. Picking one method may minimize the error
for some specific cell instances in characterization and static timing analysis, but
it may worsen the error for other instances. For a design testcase with
homogeneous cells and/or networks, using an active driver specific to the
testcase can give better results, but that approach may be meet sign-off quality
for static timing analysis under all conditions.
The best choice for a pre-driver waveform that minimizes error for designs
should be based upon empirical results on many different technologies libraries
and designs. The Synopsys pre-driver waveform method does attempts to
place the introduced error between the two extremes of:
Fast input slew with no RC network effect
Slow input slew with significant RC network effect
Figure 9 compares the Synopsys pre-driver waveform against a fast input slew
and a slow input slew. The Synopsys pre-driver waveform method strives to
minimize the maximum error possible over all driver, load, and net scenarios.
The CCS Timing characterization scripts provided by Synopsys generate
characterization waveforms using the Synopsys pre-driver waveform method.
Figure 9: Synopsys Pre-Driver Waveform
Version 3.4
10/20/16
36
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
The input characterization waveforms should remain at a fixed voltage for a
short period of time before transitioning. This allows the circuit to stabilize in its
DC operating condition. The transition should not start at t=0.
If the active driver method is chosen for library characterization, the reasons
against using active drivers should be carefully considered. In addition, if the
method for active driver characterization sweeps output capacitance for a fixed
input slew, the input slew should be chosen to minimize Miller effect at
simulation startup. Figure 10 shows the output of an active driver with a Miller
effect of ~10% of the full rail voltage. This is an excessive amount and can
impact library accuracy because the effect is propagated to the cell being
characterized. When the cell’s output current is integrated, the voltage is
assumed to be the same polarity as the current. During the Miller glitch, current
and voltage have opposite polarities, and this information is not included in the
library and is a source of correlation error.
Figure 10: Output of Active Driver with Large Miller Effect
Version 3.4
10/20/16
37
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
Active Driver Miller Effect
1.20E+00
1.00E+00
Voltage (v)
8.00E-01
6.00E-01
4.00E-01
2.00E-01
0.00E+00
0.000000001
1.05E-09
1.1E-09
1.15E-09
1.2E-09
1.25E-09
1.3E-09
-2.00E-01
Time (s)
6.3
Synopsys predriver waveform algorithm
The Synopsys “CCS jumpstart scripts” include a predriver waveform
generator. It produces a SPICE syntax linear combination of PWL and
exponential waveform:
v_ramp(t) = Vdd * (t-tstart)
v_step(t) = Vdd * (1 - exp(-(t-tstart)/RC))
v_predriver(t) = 0.5 * (v_ramp(t) + v_step(t))
The predriver waveform may be demonstrated by using this HSPICE netlist:
* Get the pre-driver waveform for input slew = 1ns
.param inslew=1n
.param vdd = 1.0v, vlow=0.2, vhigh=0.8
.param tl=`1n+0.2n`, th=`1n+0.8n`
.param pct_low=0.2, pct_high=0.8
* ramp input
V1 1 0 PWL (1n 0, '1n + inslew' vdd)
* step-response
.param rc=`(th-tl)/log((1.0-pct_low)/(1.0-pct_high))`
.param tx=`tl+rc*(log(1.0-pct_low))`
V2 2 0 EXP ( 0 1 tx rc 20n rc)
* mix of ramp and step-response * 0.5 * (v1 + v2)
E3 3 0 poly(2) 2 0 1 0 0 0.5 0.5
.option post=1
.tran 1ps 20n
Version 3.4
10/20/16
38
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
.plot
.end
6.4
Delay and Transition Time Trip Points
For CCS Timing, selection of transition time trip points will impact the transition
time value that is reported and used for delay calculation in the next stage.
During library characterization, the current waveform is captured for the full railto-rail transition of the output, and during delay calculation PrimeTime calculates
the voltage transition at the driver and receiver(s). The input transition time to
the next stage is measured from the waveform at each receiver using the
transition time trip points specified in the library.
6.5
Operating Range of Cells
The operating range of cells should be considered when you determine the
minimum and maximum values of indices in the table. The methods described
here are an example of how to determine the operating range of a cell. Library
developers may use different methods.
Typically, the operating range is constrained by the electrical rules for the library
and process. This is how the operating range may be determined:
•
The minimum transition time value can be determined by the output
transition time of the inverter with the maximum size (small Rd) driving the
minimum size inverter (large Rd) with no wire capacitance. See Figure 11.
Figure 11: Determining Library Minimum Transition Time
inv16
inv1
•
The minimum output capacitance can be determined by the load of the
minimum size inverter. This typically entails evaluating the inverter with the
smallest transistor sizes (width and length), and therefore smallest value of gate
capacitance at the input port.
Version 3.4
10/20/16
39
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
•
The maximum transition time and maximum output capacitance are
interdependent. The maximum transition time is dependent on process
technology, design rules, and design applications. A cell’s maximum
capacitance can be determined by its maximum fan-out load without exceeding
the maximum transition time. See Figure 12.
Version 3.4
10/20/16
40
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
Figure 12: Cell Driving Its Maximum Load, Cmax
Cmax = N*Cin
…
Important note: You should set max_transition and max_capacitance
attributes in the library to make sure that cells are not used outside of the
characterization range.
6.6
Selecting Characterization Points
The operating range sets the minimum and maximum indices in the table. Once
the operating range has been determined, the next step is to select the
intermediate characterization points. The following decisions must be made:
•
How many indices are needed in the table?
•
What are the values of the indices in the table?
As a general rule-of-thumb, Synopsys recommends starting with a 7 by 7 table
as the minimum table size. The decision on intermediate indices should be
made such that the interpolation errors are minimized. Library characterization
experts have developed different techniques to determine the indices. In
general, selecting indices that are not equally spaced can minimize interpolation
errors. Typically, in regions where the input transition time is small (fast) and
the output load is small (light), delay and output transition time tend to increase
non-linearly with increasing input transition time and output load. Thus, in these
regions, more characterization points are needed to minimize interpolation
Version 3.4
10/20/16
41
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
errors and to achieve better correlation with reference circuit simulators. We will
discuss two different methods for determining indices.
6.6.1
Geometric Spacing and Over-Sampling
One technique uses a number generator program to generate the indices.
Given the minimum and maximum operating range, a number generator
program is used to generate the indices using geometric spacing, such as a
distribution of equal size logarithmic steps. This approach ensures that in
regions with small values of input transition times and output capacitance
values, the indices are placed closer together in order to minimize interpolation
errors. The technique also over-samples by generating tables from 5 by 5
indices, up to 21 by 31 indices.
Data for each table (delay and transition time) are plotted and evaluated to
determine non-linear regions in the table. Based on this, a smaller number of
indices are selected; while maintaining the requirements for minimizing
interpolation errors.
For more information on the above technique, please see reference number [1].
6.6.2
Plotting of Interpolation Errors
Another technique also generates a set of indices that are not equally spaced
for minimizing interpolation errors. However, at the start of the cell
characterization, indices are equally spaced. For each quadrant of the table, a
matrix of 10 by 10 equally spaced points are simulated using the reference
simulator. The interpolation errors are measured and plotted. Based on these
plots, one may observe that in regions where input transition time and/or output
capacitances are small, the interpolation errors tend to be larger compared to
the other regions of the table. One may apply automation to re-position the
indices in the table while reducing interpolation errors.
For more information on the above technique, please see reference number [2].
6.7
Segmenting the Current Waveform
The output current waveform from the simulation will typically have hundreds of
points. To reduce library size, a minimal number of points from the current
waveform should be saved. However, there is a trade-off between CCS library
accuracy and number of points saved.
The Synopsys “CCS jumpstart scripts” include a module that reduces the
detailed waveform to a shortened piecewise-linear version. This process is
called segmentation. The script uses a segmentation tolerance to determine the
Version 3.4
10/20/16
42
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
accuracy of the segmentation. The number of points saved from the current
waveform increases as the required accuracy is increased.
Here is a brief summary of how the “CCS jumpstart scripts” segmentation
algorithm in “CCS_Segment.pm” operates:
For each timing arc:
1. Obtain i(t), V(t) waveforms from HSPICE simulation
2. Traverse from end time to start time to select starting time for current
measurement
• Skip all initial voltage glitches of reverse polarity
3. Traverse from start time to end time to select stop time for current
measurement
• Use HSPICE autostop to end simulation when voltage reaches to rail
• To support cases where output does not reach rail, select a simulation
time that is at least double the second slew threshold time
4. Find peak current value
5. Modify first current value to allow Voltage to exactly match at first saved
time point
• Calculate extrapolated first (time, current) sample point
• Ensure that time is positive, and integrated voltage matches actual
6. Modify first current value to ensure that it is not the peak current
• Accurate peak current sampling is critical for scaling in Primetime
• Library Compiler will flag an error if waveform is monotonic
7. Traverse all remaining current values, and select only the remaining i(t)
values necessary to assure that
| V(actual) – V(ccs) | < ccs_segment_tolerance,
Use trapezoidal integration to calculate the CCS voltage
V (tn )
[ I (tn ) I (tn 1)] (tn tn 1)
2
C
tstart
tn
8. Write the optimized i(t) vector into the characterization database
9. Perform Quality of Results test on the segmented waveform
Figure 13 shows the characterization schematic, the resulting simulation
waveforms, and the segmented current that is included in the library.
Figure 13: Segmenting the Output Current
a) Characterization Schematic
Version 3.4
10/20/16
43
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
iinp
iCout
Sinp
Cchar
b) b) Simulation Results
Cell Voltages
1.20
Vin
1.00
Vout
Voltage (V)
0.80
0.60
0.40
0.20
0.00
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
-0.20
Time (ns)
Version 3.4
10/20/16
44
5
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
Output Current
0.12
0.10
Iout
Current (ma)
0.08
0.06
0.04
0.02
0.00
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
3.50
4.00
4.50
-0.02
Time (ns)
c) Segmented waveform
results
Segmented Output Current
0.12
0.10
Current (ma)
0.08
0.06
0.04
0.02
0.00
0.00
0.50
1.00
1.50
2.00
2.50
3.00
Time (ns)
The quality of the segmented current is checked by integrating it, and the
resulting voltage waveform is compared to the simulation voltage waveform.
Points are selected from the current waveform such that the voltage difference
is within the specified segmentation tolerance. The simulation time step must be
Version 3.4
10/20/16
45
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
set such that the maximum voltage change per time step is less than the
segmentation tolerance. Circuit simulators such as HSPICE allow adaptive time
step control such that voltage change does not exceed the segmentation
tolerance. If the simulator used for library characterization does not have
adaptive time step control, small enough fixed time steps should be used to
meet this criterion. For more information, refer to section 6.1.1.
6.8
Deriving Capacitances for the Receiver Model
The CCS receiver model uses two capacitance values to model the cell loading
during the transition. For optimal CCS receiver accuracy, the value calculated
as C1 should be chosen to ensure that the point (V2,t2) is exactly matched. The
value calculated as C2 should be chosen to ensure that the point (V3,t3) is
exactly matched.
The Synopsys “CCS jumpstart scripts” includes a module that calculates
optimal values for each receiver capacitance. Here is how the module
“CCS_Solve.pm” operates:
For each rising timing arc:
1. Find C1 “seed value” below delay threshold
t2
C1seed =
I (t )dt
t1
t1 = starting time , t2 =time at delay threshold V2
V2
2. Find C2 “seed value” above delay threshold
t3
C2seed =
I (t )dt
t2
t2 = time at delay threshold V2, t3 = ending time
(Vdd V 2)
3. Calculate optimal C1 and C2 values.
• Brent algorithm is used to minimize curve fit error over a specific
waveform range
• The default waveform range is a single point (V2, t2) for C1
C1 = CCS_Solve::brent (C1seed , …, (V2, t2));
• The default waveform range is a two points (V2, t2) and (V3, t3,) for C2
C2 = CCS_Solve::brent (C2seed , …, (V2, t2), (V3, t3,));
4. Test receiver values by calculating delay and slew values and
comparing to actual delay and slew values.
Version 3.4
10/20/16
46
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
5. Store results into characterization database in Liberty syntax
Figure 14: Characterization Waveform for rising and falling slews
Second Slew Threshold (V3, t3)
C2
Input Delay Trip Point (V2, t2)
C1
First Slew Threshold (V1, t1)
First Slew Threshold (V1, t1)
C1
Input Delay Trip Point (V2, t2)
C2
Second Slew Threshold (V3, t3)
.
6.9
Other Considerations
6.9.1
Simultaneous Switching
Cells with simultaneously switching inputs are not recommended for gate-level
timing analysis. Libraries characterized using simultaneously switching inputs
should not be used for static timing analysis.
Version 3.4
10/20/16
47
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
6.9.2
Merging Timing Arcs
Merging related timing arcs to produce a single worst-case arc may be
interpreted by CCS as not physically possible data and is not recommended.
Merged CCS current data may generate RC-104 warning messages. A sample
message is shown below in Example 30.
Example 30: PrimeTime RC-104 Warning Message
Warning: Failed to compute the cell timing arc
(design/INV) INV/A-->Z (max-falling negative_unate)
because there is a problem with the driver's CCS data.
[r/f inp_slew = 0.034/0.034, out_cap = 0.107
units)]
6.9.3
(lib
Three-Dimensional Timing Arcs
Timing arcs with delay and transition time dependent on input transition time,
output load of the arc, and additional external loading on a related cell output
are not recommended for RC delay calculation. CCS Timing does not support
three-dimensional timing arcs. In Figure 15, the timing arc from A to Y is
dependent on the load at YB. Similarly, the timing arc from A to YB is dependent
on the load at Y.
Version 3.4
10/20/16
48
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
Figure 15: Cell with Inter-Dependent Three-Dimensional Timing Arcs
Y
A
YB
6.9.4
Level Shifters
A cell can be characterized with the input rails different from the output rails.
This is supported for CCS Timing.
6.9.5
Cells with Unbuffered Inputs
If the cell is multistage, a pin-based receiver model is recommended for CCS to
capture the pass gate behavior for each input transition. An arc-based receiver
model can be used for a single stage. A fully indexed receiver model table can
be used to capture the load dependency of the transmission gate when it is
open. When it is closed, a receiver model with a single capacitance index can
be used, since there is no load dependency. See Section 3.4 for more
information.
7 Library Compiler Checks for CCS Timing
Libraries
Below is a summary of the CCS Timing checks implemented in the
2005.09 version of Library Compiler.
X-
CCS Library Requirement
Check
Section
LC Message
Transition Check
Integrate the current waveform and check
that the voltage is within 5% of the final rail
voltage. Warn if this criteria is not met, and
error out if voltage fails to reach second
transition time threshold.
2.1
LBDB-668
Version 3.4
10/20/16
LBDB-669
49
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
Current Polarity
Check that the current polarity is correct for
each timing arc.
2.3
LBDB-664
Zero Current
Check that the current vector does not
contain zero current for the entire current
vector.
2.1
LBDB-664
Check that the current vector does not
contain zero current values.
2.2
Peak Current
Check that the first or last current in the
current vector is not the peak current.
2.4
LBDB-665
Library Voltage
Check that the nom_voltage and the default
operating condition voltage in the library are
identical.
4.3
LBDB-662
Identical reference_time
Check that the reference_time attribute is
identical for a fixed transition time and
varying loads.
2.7
LBDB-654
Negative reference_time
Check that the reference_time attribute is
non-negative for each current vector.
2.6
LBDB-667
Number of significant
digits
Check whether receiver model table values
and current vectors have enough significant
digits whether using scientific or decimal
notation.
5.4
LBDB-670
Current Vector Variables
Check that variable_3 in an
output_current_template is always
time.
2.5
LBDB-653
8 CCS Timing Liberty Extension (June 2016)
To further improve accuracy of timing analysis, the receiver capacitance modeling
syntax was extended to support multiple segments. In the multisegment receiver
capacitance model, the voltage rise or fall at a cell input node is divided into multiple
(typically, more than two) mutually exclusive segments, as shown by the dotted lines in
the following figure. The receiver capacitance values are characterized for each voltage
segment. Using multiple segments better represents the nonlinearity of the net receiver
capacitance at the node.
Version 3.4
10/20/16
50
Synopsys, Inc.
690 East Middlefield Road
Mountain View, CA 94043 USA
Figure 16: Multisegment Receiver Capacitance Model
For each segment, use the new receiver_capacitance_rise and
receiver_capacitance_fall groups to store the receiver capacitance values as
lookup tables. To specify the segment, use the new segment attribute in these groups.
You can define these groups in the pin-level receiver_capacitance and timing
groups. This model uses the same input slew in the input_net_transition
(index_1) values of the lookup tables across all the segments, irrespective of different
waveform shapes in each segment. The input slew is defined using the library-level
slew_derate_from_library attribute.
9 References
[1]
“Using PathMill® to Characterize Library Cells”, Hai Vo-Ba, System VLSI
Technology Operation, Hewlett Packard, Synopsys Users’ Group Conference,
2001.
[2]
“Constructing Better Non-Linear Delay Models (NLDM) to Improve Timing
Closure”, Steven E. Start, Erik N. Comparini, American Microsystems, Inc.,
Synopsys User s’ Group Conference, 2001.
[3]
“Simulation Speed and Accuracy”, Pages 327-330, HSPICE Simulation
and User Guide, Version W-2005.03, March 2005, from Synopsys, Inc.
Version 3.4
10/20/16
51
Source Exif Data:
File Type : PDF File Type Extension : pdf MIME Type : application/pdf PDF Version : 1.5 Linearized : Yes Author : Karen Linser Create Date : 2016:10:20 18:06:36-07:00 Modify Date : 2017:01:16 14:43:38-08:00 Language : en-US Tagged PDF : Yes XMP Toolkit : Adobe XMP Core 5.4-c006 80.159825, 2016/09/16-03:31:08 Format : application/pdf Creator : Karen Linser Title : CCS Timing Library Characterization Guidelines Creator Tool : Microsoft® Word 2016 Metadata Date : 2017:01:16 14:43:38-08:00 Producer : Microsoft® Word 2016 Document ID : uuid:5d308014-d0ba-4836-ba6d-16b85b2ecc89 Instance ID : uuid:0bec9c50-bce6-4da7-a44c-93f14d7a22ec Page Count : 51EXIF Metadata provided by EXIF.tools