1993_IDT_High Speed_CMOS_Logic_Design_Guide 1993 IDT High Speed CMOS Logic Design Guide
User Manual: 1993_IDT_High-Speed_CMOS_Logic_Design_Guide
Open the PDF directly: View PDF .
Page Count: 148
Download | |
Open PDF In Browser | View PDF |
u c 1993 IDT High·Speed CMOS Logic Design Guide Includes Double-Density and 3.3V Application Notes Q) '1:J ·S c:J c m .& II) .-m () • .. .9 en 0 :E () 1a. en• .c .-::cm ... Q C") 0) 0) .... ® dt Integrated Device Technology, Inc. 1993 High Speed CMOS Logic Design Guide TABLE OF CONTENTS Introduction .. ... ..... ...... ... ... ........ ...... ... ........ ... ... ... ........ ......... ... ... ..... ...... ........... ... ... ... ... ... ..... ... ... ...... ....... Product Selector Guide .......................................................................................................................... 1 2 High Speed Design AN-117 AN-102 AN-103 AN-116 AN-118 AN-47 AN-48 AN-49 AN-50 AN-51 AN-52 AN-53 AN-54 AN-82 Double-Density Logic Characteristics/and Applications ...................................................... Power Off Disable in lOT's Double-Density Family ............................................................. 3.3V Logic ........................................................................................................................... Decoupling Double-Density Components ........................................................................... Maximum Frequency Considerations for the IDT49FCT805/6 ............................................ Simultaneous Switching Noise ............................................................................................ Using High Speed Logic. .... ..... ...... ... ... ........ ... ......... ..... ...... ...... ..... ...... ... ...... ........ ......... ... ... Characteristics of PCB Traces ............................................................................................ Series Termination .............................................................................................................. Power Dissipation in Clock Drivers .. ........... ... ... ...... ..... ...... ...... ..... ...... ... ...... ........... ... ...... ... FCT Output Structures and Characteristics ................ ...... ...... ..... ...... ... ..... ...... ........... ........ Power-Down Operation ....................................................................................................... FCT-T Logic Family ............................................................................................................. Clock Distribution with IDT Guaranteed-Skew Clock Drivers .............................................. 6 25 27 35 39 41 49 50 54 56 58 61 63 71 Error Detection and Correction Designing with the IDT49C460 and IDT39C60 Error Detection and Correction Units ........ Protecting your Data with IDT49C465 32-Bit FlowThru EDC Unit ...................................... Error Detection and Correction with 49C466 ..... ........ ... ... ...... ..... ... ... ... ...... ..... .............. ... .... Cascading 49C460 and 49C465 ........ ........... ... ........ ...... ...... ..... ...... ... ... ..... ... ... ... ... ... ..... ..... 82 96 112 123 Conference Papers CP-10 Integrated Line Termination Solutions for High Speed Logic .............................................. CP-11 New Output Structure Designs Reduce System Noise Problems ....................................... CP-12 Double-Density Logic: the Fast. Flexible Family ................................................................ CP-13 Mixed Supply Design Considerations for Notebook PCs .................................................... 127 130 135 139 AN-24 AN-64 AN-94 AN-96 Integrated Device Technology, I nco reserves the rightto make changes to its products or specifications at any time, without notice, in order to improve design or performance and to supply the best possible product. lOT does not assume any responsibility for use of any circuitry described other than the circuitry embodied in an lOT product. The Company makes no representations that circuitry described herein is free from patent infringement or other rights of third parties which may result from its use. No license is granted by implication or otherwise under any patent, patent rights or other rights, of Integrated Device Technology, Inc. LIFE SUPPORT AGREEMENT Integrated Device Technology's products are not authorized for use as critical components in life support devices or systems unless a specific written agreement pertaining to such intended use is executed between the manufacturer and an officer of lOT. 1. Life fupport devices or systems are devices or sytems which (1) are intended for surgical implant into the body or (b) support or sustain life and whose failure to perform, when properly used in accordance with instructions for use provided in the labeling, can be reasonably expected to result in a significant injury to the user. 2. A critical component is any component of a life support device or system whose failure to perform can be reasonably expected to cause the failure of the life support device or system, or to affect its safety or effectiveness. The lOT logo is a registered trademark, and BiCameral, BurstRAM, BUSMUX, CacheRAM. DEenat, Double~Density, FASTX, FourFort, FLEX1-CACHE, FLEX)"PAK, Flow-thruEDC, 10Tlc, JOTenvY. IDT/sae, IDTfsim, IDT/ux, MacStation; MICROSUCE, PaietteDAC,REALB, R3051 , R3081 , RISCompiler, RISController, RISCors, RISC SubSystem, RISCwindows, SARAM, SmartLogic, SyncFIFO, SyncBiFIFO, SPC, TargetSystem and WideBus are trademarks of Integrated Device Technology. Inc. All others are trademarks of their respective companies. ii INTRODUCTION This collection of application notes and conference papers is presented to provide information which has been proven useful to the designers who use or plan to use high-speed TIL-compatible, CMOS logic in high-performance systems. These notes cover a broad range of topics; some discuss aspects of high-speed design with new application notes specifically related to the new Double-Density family. Also included are application notes on error detection and correction. Double-Density Family Defined The demand for higher integration and high speed continues to push the need for high-performance, high-density logic families. IDT has recently introduced a new series of extraquiet, high-performance, 16-, 18-, and 20-bit logic functions. This exciting new family offers users significant board area savings, power savings, higher speeds, excellent guaranteed low noise characteristics, and guaranteed low output skew. The Double-Density family is the premiere octal upgrade and is considered the high-speed, low-power replacement for all existing CMOS and SiCMOS wide bus width products. High Speed Design With the push for higher clock rates to increase system performance in a predominantly TIL world, the hardware designer is now required to deal with problems and issues which were largely academic only a few years ago. The introduction of high-speed, high-drive CMOS logic, combined with the increased packing density of multi-layer boards has resulted in the awareness and the need to deal with problems such as ground bounce, transmission line effects, and dynamic power dissipation. Error Detection and Correction Today's high-performance systems are becoming increasingly DRAM intensive. IDT has developed a range of highperformance EDC devices that eliminate the performance penalties once associated with these circuits while assuring the designer continuous, error-free operation necessary in high-reliability systems. ABOUT THE PRODUCT SELECTOR GUIDE Pages 2-5 contain the 1993 IDT Logic Product Selector Guide. The data book page numbers are preceded by the alpha character "G". This signifies the IDT 1992/1993 HighPerformance Logic data book. Following the character will be the section number and the page number, consecutively. "CALL" might appear in the page number area as well. This means the data sheet was not included in the data book. A copy of the data sheet may be obtained by contacting IDT's Literature Department at (800) 345-7015 [( 408) 492-8674], or from your local sales representative. Shaded areas indicate new products introduced since the last publication of the product selector guide. Boldface type indicates an improved IDT product feature. The availability column shows the date when limited production quantities will be available. The symbol ..... after the data book page number indicates further updated information has become available since the publication of the 1992/1993 High-Performance Logic data book. This information should be available from your 10caiiDT sales office. High-Speed CMOS Logic Products - Bus Interface Devices o FCT and FCT-T CMOS families are the fastest in the industry with maximum propagation delays as low as 3.6 ns. 2. 24mA balanced output drive with on-chip resistors for internal bus and point to point driving. o The FCT-T family offers the lowest-power solution. No other logic family uses less dynamic power, standby power, or static high or low power. 3. 3.3V bus interface logic for systems with 3.3V regulated supplies. o o The clock driver family provides guaranteed low-skew clock distribution in a variety of configurations. The new Double-Density bus interface family offers users significant board area savings, power savings, higher speeds, guaranteed low noise, and a choice of output drive characteristics. Three configurations are available: 1. Standard 64mA high drive device for bus and backplane interface. o The EDC (Error Detection and Correction) devices can detect multiple errors as fast as 10ns and correct as fast as 14ns. o IDTfeatures a series of read-write buffers with 8& 18 bit bidirectional registers and 16-bit pipeline registers. o The DSP Building Blocks are composed of 16-bit ALUs, Multipliers, and Multiplier-Accumulators with speeds as fast as20ns. o Parity generation and checking circuits designed to minimize part count and maximize parity checking time. Max. Speed (ns) Max. Power Description Mil. Com'l. (mW) Avail. Page IDT29FCT52A1B/C IDT29FCT52AT/BT/CT/DT Non-inverting Octal Registered Transceiver 7.3 4.5 1.0 NOW G 6.26 G 6.1 IDT29FCT53A1B/C IDT29FCT53AT/BT/CT Inverting Octal Registered Transceiver 7.3 6.3 1.0 NOW G 6.26 G6.1 IDT29FCT520A/BIC IDT29FCT520AT/BT/CT/DT Multi-level Pipeline Register 7.0 5.2 1.0 NOW G6.27 G6.2 IDT29FCT521 AT/BT/CT/DT Octal Multi-level Pipeline Register (level 1 replaced) 7.0 5.2 1.0 NOW G 6.2 IDT54/74FCT1381A1C IDT54/74FCT138T/AT/CT 1-of-8 Decoder 6.0 5.1 1.0 NOW G6.29 G 6.3 IDT54174FCT139/A1C IDT54174FCT139T/AT/CT Duall-of-4 Decoder 6.2 5.0 1.0 NOW G6.30 G 6.4 IDT54175FCT151 T/AT/CT 8-lnput Multiplexer 6.2 5.6 1.0 NOW G 6.5 3.9 1.0 NOW CALL Part Number Data Book FCT/FCT-T FAMILY IDT54174FCT157T/AT/CT/DT Quad 2-input Multiplexer 5.0 IDT54174FCT161/A IDT54174FCT161T/AT/CT Synchronous Binary Counter w/Synchronous Reset 6.3 5.8 1.0 NOW G 6.31 G6.7 IDT54/74FCT163/A IDT54/74FCT163T/AT/CT Synchronous Binary Counter w/Asynchronous Master Reset 6.3 5.8 1.0 NOW G 6.31 G6.7 IDT54/74FCT1821A Carry Lookahead Generator 10.7 7.0 1.0 NOW G 6.32 IDT54/74FCT191/A Up/Down Binary Counter 10.5 7.8 1.0 NOW G 6.33 IDT54/74FCT193/A Up/Down Binary Counter 6.9 6.5 1.0 NOW G 6.34 IDT54/74FCT240/A/C IDT54/74FCT240TIAT/CT/DT Inverting Octal Buffer Line Driver 4.7 3.6 1.0 NOW G 6.35 G 6.10 IDT54/74FCT241/A1C IDT54/74FCT241T/AT/CT/DT Inverting Octal Buffer Line Driver 4.6 3.6 1.0 NOW G 6.35 G 6.10 IDT54/74FCT244/A1C IDT54/74FCT244T/AT/CT/DT IDT54/74FCT3244/A Inverting Octal Buffer Line Driver 4.6 3.6 1.0 NOW 4.8 1.0 NOW G 6.35 G 6.10 G7.7 4.5 3.8 1.0 NOW 3.3V Inverting Octal Buffer Line Driver IDT54/74FCT2451A1C IDT54174FCT245T/AT/CTIDT IDT54/74FCT3245/A Inverting Octal Buffer Transceiver 3.3V Inverting Octal Buffer Transceiver 4.6 1.0 NOW IDT54/74FCT257T/AT/CT/DT Quad 2-input Multiplexer w/OE 5.0 3.9 1.0 NOW CALL IDT54174 FCT273/A1C IDT54174FCT273T/AT/CT/DT Octal D Flip-Flop with Reset 6.5 5.8 1.0 NOW G 6.37 IDT54/74FCT2991AlC IDT54/74FCT299/AT/CT Octal Universal Shift Register w/Common Parallel 110 Pins 7.5 6.5 1.0 NOW G 6.38 G 6.13 IDT54174FCT3731AlC IDT54174FCT373T/AT/CT/DT Octal Transparent Latch 5.1 3.8 1.0 NOW G 6.39 G 6.14 2 • G 6.36 G 6.11 G7.8 • Hiah-Seeed CMOS Loaic Products Max. Speed (ns) Mil. Com'l. Max. Power !mWI Avail. Data Book Paae Part Number Descrietion IDT54174FCT374/A/C IDT54/74FCT374T/AT/CT/DT Octal D Register 6.2 4.2 1.0 NOW G 6.40 CALL IDT54/74FCT377IA1C IDT54/74FCT377T/AT/CT/DT Octal D Flip-Flop w/Clock Enable 5.5 4.4 1.0 NOW G 6.41 G 6.16 IDT54/74FCT399/A IDT54/74FCT399T/AT/CT Quad Dual-Port Register 6.6 6.1 7.0 NOW G 6.42 G 6.17 IDT54175FCT521/A/B/C IDT54174FCT521T/AT/BT/CT 8-Bit Identity Comparator 5.1 4.5 1.0 NOW G 6.43 G 6.18 IDT54174FCT533/A/C IDT54174FCT533T/AT Inverting Octal Transparent Latch w/3-State 5.1 4.7 1.0 NOW G 6.39 G 6.14 IDT54/74FCT534/A/C IDT54/74FCT534T/AT/CT Inverting Octal D Register w/3State 6.2 5.2 1.0 NOW G 6.40 G 6.15 IDT54/74FCT540/AlC IDT54/74FCT540T/AT/CT Inverting Octal Buffer/Line Driver 4.7 4.3 1.0 NOW G6.35 G 6.10 IDT54/74FCT541/A/C IDT54/74FCT541T/AT/CT Non-Inverting Octal Buffer/Line Driver 4.7 4.3 1.0 NOW G 6.35 G 6.10 IDT54/74FCT543/A1C IDT54/74FCT543T/AT/CT/DT Non-Inverting Octal Latched Transceiver 6.1 4.4 1.0 NOW G 6.44 G 6.19 IDT54174FCT573/A1C IDT54/74FCT573T/AT/CT/DT Octal Transparent Latch 5.1 3.8 1.0 NOW G 6.39 G 6.19 IDT54/74FCT574/A1C IDT54/75FCT574T/AT/CT/DT Octal D Register wi 3-State 6.2 4.2 1.0 NOW G 6.40 G 6.15 IDT54/74FCT620T/AT/CT Inverting Octal Bus Transceiver w/3-State 5.1 4.5 1.0 NOW G 6.21 IDT54174FCT621T/AT Non-Inverting Octal Bus Transceiver w/Open Drain 12.5 12.0 1.0 NOW G6.22 IDT54/74FCT623T/AT/CT Non-Inverting Octal Bus Transceiver w/3-State 5.4 4.8 1.0 NOW G 6.21 IDT54174FCT640/AlC IDT54/74FCT640T/AT/CT Inverting Octal Transceiver 4.7 4.4 1.0 NOW G 6.36 G 6.11 IDT54174FCT645/A1C IDT54174FCT645T/AT/CT/DT Non-Inverting Bidirectional Transceiver 4.5 3.8 1.0 NOW G 6.36 G 6.11 IDT54/74FCT646/A1C IDT54/74FCT646T/AT/CT/DT Octal Transceiver/Register 6.0 4.4 1.0 NOW G 6.45 G6.20 IDT54/74FCT648T/AT/CT Octal Transceiver/Register 6.0 5.4 1.0 NOW G 6.20 IDT54/74FCT651T/AT/CT Inverting Octal Registered Transceiver 6.0 5.4 1.0 NOW G 6.20 IDT54/74FCT652T/AT/CTIDT Non-Inverting Octal Registered Transceiver 6.0 4.4 1.0 NOW G 6.20 IDT54174FCT821 AlBIC IDT54174FCT821AT/BT/CT 10-Bit Non-Inverting Register 7.0 6.0 1.0 NOW G 6.46 CALL IDT54/74FCT823A1B1C 9-Bit Non-inverting Register IDT54174 FCT823AT/BT/CT/DT 7.0 5.0 1.0 NOW G 6.46 G 6.23 IDT54/74FCT825A/B/C IDT54/74FCT825AT/BT/CT 7.0 6.0 1.0 NOW G 6.46 G 6.23 5.0 3.8 1.0 NOW G 6.47 G 6.24 IDT54174FCT828AT/BT/CT/DT 10-Bit Inverting Register 5.0 3.8 1.0 NOW G6.24 IDT54174FCT833A1B 8-Bit Transceiver w/Parity 10.0 7.0 1.0 NOW G 6.48 IDT54/74FCT841 AlBIC IDT5417 4FCT841 ATlBT/CTIDT 10-Bit Non-Inverting Latch 6.3 4.2 1.0 NOW G 6.49 G 6.25 IDT54/74FCT843A1B1C IDT54174FCT843AT/BT/CT 9-Bit Non-Inverting Latch 6.3 5.5 1.0 NOW G 6.49 G 6.25 8-Bit Inverting Register 10-Bit Non-Inverting Buffer IDT54/74FCT827AlBIC IDT54/7 4FCT827AT/BTlCT/DT 3 • • High-Speed CMOS Logic Products Max. Speed (ns) Com'l. Mil. Max. Power (mW) Avail. Data Book Page Part Number Description IDT54/74FCT845A1B1C IDT54/74FCT845AT/BT/CT 8-Bit Non-Inverting Latch 6.3 5.5 1.0 NOW G 6.49 G 6.25 IDT54/74FCT861A1B1C 1O-Bit Non-Inverting Transceiver 6.5 6.0 1.0 NOW G 6.50 IDT54/74FCT863A1B1C 9-Bit Non-Inverting Transceiver 6.5 6.0 1.0 NOW G 6.50 Clock Driver w/Guaranteed Skew 6.8 5.8 1.0 NOW G 6.28 CLOCK DRIVER FAMILY IDT49FCT805lA iQW~ff«too$l.lltl¢m:::II,I::I:}I' ,: : : : : : : : : : : : : : .: :.: • ':.::.,,:::• ;,/", rrIm:rm ::::::::::::::::I:'\j.;.::.:';'."I~II41§ mr:I:rr1i9.I IDT49FCT806/A Inverting Clock Driver . 6.8 5.8 1.0 rIIMAi!!!~~:~:rJti~MI NOW :. ::' ~ ~~ ~~ :: :; =:' :; =; :; ::' ::: ::: :: :; ~: ;: :..:.::::::::::::::::: ::',::.:' . . . , , ' , ' , ' , ' , . ' , " : : ",:: :': ..:.:.: ; ,;. . .: " .: :: :: :; :: :: :: .;. , ,~.: ,i.~· ....... ,,',"',.::::: ::.:::; :: :: :: ::: . ,: :: :: :: '. " : :: .: " . .;: ~~ ~~ ~~ ::~ ~; :: :~ =; :: ::: " . . . " . " .; .; :. :. ,,' ,'; ;: :. . . , , ' , ' , ',' ::::; .;::::;.:". Low Skew PLL-Based Clock Driver N/A N/A 1.0 Q2'93 CALL 16-Bit Buffer/Line Driver 4.7 4.3 0.006 NOW G 5.1 G 5.1 IDT54/74FCT16244T/AT/CT IDT54/74FCT162244T/AT/CT 16-Bit Buffer/Line Driver 4.6 4.1 0.006 NOW G 5.2 G 5.2 IDT54/74FCT16245T/AT/CT IDT54/74FCT162245T/AT/CT 16-Bit Bidirectional Transceivers 4.5 4.1 0.006 NOW G 5.3 G 5.3 IDT54174FCTI6373T/AT/CT IDT54/74FCT162373T/AT/CT 16-BitTransparent Latches 5.1 4.2 0.006 NOW G 5.4 G 5.4 IDT54/74FCT16374T/AT/CT 16-Bit Register (3-State) 6.2 5.2 0.006 NOW G 5.5 G5.5 16-Bit Bus Transceiver/Registers (3-State) 6.0 5.4 0.006 NOW IDT54174FCT162646T/AT/CT G5.9 G 5.9 IDT5417 4FCT16500AT/CT IDT5417 4FCT162500AT/CT 18-Bit Registered Bus Transceiver (3-State) 5.1 4.6 0.006 NOW G 5.6 G 5.6 IDT54/74FCT16501AT/CT IDT54/7 4FCT162501ATlCT 18-bit Registered Bus Transceiver (3-State) 5.1 4.6 0.006 NOW G5.7 G5.7 IDT54/74FCT16543T/AT/CT/DT 16-Bit Latched Transceiver IDT54/74FCT162543T/AT/CT/DT 6.1 4.4 0.006 NOW G 5.8 G 5.8 IDT54/74FCT16952T/AT/CT/DT 16-Bit Registered Transceivers IDT54/74FCT162952T/AT/CT/DT 7.3 4.5 0.006 NOW G 5.11 IDT54/74FCT88915 5V DOUBLE-DENSITY FAMILY IDT54/74FCT16240T/AT/CT IDT54174FCTI62240T/ATICT IDT54174FCT162374T/AT/CT IDT54174FCT16646T/AT/CT IDT54/74FCT16652T/AT/CT IDT54/74FCT162652T/AT/CT 16-Bit Transceiver/Registers 6.0 5.4 0.006 NOW G 5.10 IDT54174FCTI6823AT/BT/CT 18-Bit Bus Interface Registers 7.0 6.0 0.006 NOW G 5.12 20-Bit Buffers 5.0 4.4 0.006 NOW G 5.13 IDT54174FCT162823AT/BT/CT IDT54174FCTI6827AT/BT/CT :]ii'r G6.28 IDT54174FCTI62827AT/BT/CT 3.3V DOUBLE-DENSITY FAMILY IDT54/74FCT163244/A 3.3V CMOS 16-Bit Buffer/Line Driver 4.8 0.006 NOW G7.1 IDT54174FCT1632451A 3.3V CMOS 16-Bit Bidirectional Tranceivers 4.6 0.006 NOW G7.2 4 • High-Speed CMOS Logic Products Max. Power (mW) Avail. 6.3 0.006 NOW Max. Speed (ns) Com'l. Mil. Data Book Page Part Number Description IDT54/74FCT164245T CMOS 16-Bit Bidirectional3.3V to 5V Translator IDT54/74FCT163373/A 3.3V CMOS 16-Bit Bidirectional Latches 5.2 0.006 NOW G 7.3 IDT54/74FCT163374/A 3.3V CMOS 16-Bit Register 5.2 0.006 NOW G7.4 IDT54/74FCT163646/A 3.3V CMOS 16-bit Bus Transceiverl Registers 6.3 0.006 NOW G 7.6 IDT54/74FCT163501/A 3.3V 18·Bit Registered Bus Transceiver 5.1 0.006 NOW G7.5 ERROR DETECTION AND CORRECTION G 7.9 Detect Time IDT39C60 IDT39C60-1 IDT39C60A IDT39C60B 16·bit Cascadable EDC Replaces Am2960, -1, A; N2960, MC74F2960, -1, A 36 28 24 22 32 25 20 18 300 NOW NOW NOW NOW G 8.10 IDT49C460 IDT49C460A IDT49C460B IDT49C460C IDT49C460D 32-bit Cascadable EDC Replaces Am29C660. Functional equivalent to DP8402; ASJALS632 44 33 28 21 16 40 30 25 16 12 350 NOW NOW NOW NOW NOW G 8.11 G 8.11 G 8.11 G 8.11 G 8.11 IDT49C465/A 32-bit Flow-thruEDC'" two separate bidirectional 32-bit buses; expandable to 64-bit, 144-pin PGA. 15 20 400 NOW G 8.12 IDT49C466 64-bit Flow-thruEDC-two separate bidirectional 64-bit buses; 208-pin PGA 25 20 500 NOW G 8.13 73200 16-bit 8-level-deep pipeline Register; replaces four Am29520s 12 10 150 NOW G 8.7 73201 16-bit 7-level-deep pipeline Register with pass-through mode 12 10 150 NOW G 8.7 73210/AlB 8-bit bidirectional registers with parity; two registers from B to A 7.5 6.0 50 NOW G 8.8 73211/A1B 8-bit bidirectional registers with parity; one register from B to A 7.5 6.0 50 NOW G8.8 IDT7381 16-bit cascadable ALU (replaces Logic Devices' L4C381) 20 16 150 NOW G 8.6 IDT7383 16-bit cascadable ALU (32 instructions) 20 16 150 NOW G 8.6 IDT7210L 16 x 16-bit with 35-bit output, replaces TDC1010J 20 16 225 NOW G8.4 IDT7216L 16 x 16-bit, replaces Am29516 20 16 200 NOW G 8.5 IDT7217L 16 x 16-bit with single-clock architecture, replaces Am29516 20 16 200 NOW G 8.5 350 NOW NOW NOW G 8.2 READ-WRITE BUFFERS DSP BUILDING BLOCKS MICROSLICETM PRODUCTS (Com'I.) = 47ns = 37ns = 28ns IDT49C402 IDT49C402A IDT49C402B 16-Bit Zf!P Slice, quad 2901 with 8 additional destination functions and 64 x 16 register file capacity-superset of Am29Cl 01, CY7C91 01 , WS159016 IDT39Cl0B IDT39Cl0C 12-bit Sequencer with 33-deep stack -replaces AM291 O/A, CY7C910 o to Y = 20ns o to Y = 12ns 150 NOW NOW G 8.2 IDT49C410 IDT49C410A 16-bit Sequencer with 33-deep stack address up to 64K microcode Dto Y = 20ns to Y = 12ns 150 NOW NOW G 8.3 A,B addr to Y A,B addr to Y A,B addr to Y o 5 - ---- - ----- ~ W [.7 APPLICATION DOUBLE-DENSITYTM LOGIC NOTE CHARACTERISTICS AND APPLICATIONS AN-117 Integrated Device Technology, Inc. By Anupama Hegde and Stanley Hronlk TABLE OF CONTENTS INTRODUCTION The increasing demand for integration in computer systems has lead to the evolution of logic components with wider buses and highly integrated functionality. Shorter delays and fewer components are welcome features when board space and performance are at a premium. With such increased integration and increased ciock speeds, power dissipation and simultaneous switching noise become important issues. In the light of such developments, IDT has introduced a new 16-bit wide logic family - Double-Density - to address the system designer's need for increased performance, reduced noise and low power dissipation. The family not only doubles the number of bits in a single package but also offers users other features such as a small package size, extremely low power dissipation, low leakage, reduced ground bounce and a very friendly user interface. The Double-Density logic family is an extension of IDT's Fast CMOS TTL-compatible (FCT-T) line of logic products. These devices are built using IDT's submicron CMOS 5E technology. The Double-Density logic family consists of three subfamilies, each of which seeks to address specific design needs. The first is the High-Drive FCT16xxxT family, designed for backplane and other applications that require a high static and dynamic output drive capability. The second is the Balanced-Drive FCT162xxxT family. This family features a balanced output drive and has excellent noise characteristics, which is a central issue to most high-speed applications. The last is the Low Voltage (3.3V), FCT163xxx family which is aimed at low-power, high-performance design applications. For a detailed discussion of the FCT163xxx family, please refer to Logic Application Note, AN-103 ( "3.3V Logic" ). All three Double-Density sub-families offer lower power solutions than their industry counter parts. This application note discusses IDT's Double-Density HighDrive and Balanced-Drive families which are designed to give the best cost/performance ratio in a variety of applications. It provides useful information needed to evaluate and/or design in these devices. The focus is on the families' features and benefits, product characteristics and design guidelines. The application note is divided into several sections, each dealing with a specific characteristic or feature of the family. Design information and actual data on the Double-Density FCT16xxxT and FCT162xxxT families are provided in the form of graphs and tables making them easier to interpret and access. In cases where the characteristics are common to both the FCT16xxxT and FCT162xxxT families, only one graph is given for both families. PAGE DOUBLE-DENSITY LOGIC FAMILIES Description .............................................................. 7 DOUBLE-DENSITY FAMILY FEATURES Functions ................................................................ Speed grades ......................................................... Operating range ..... ..... ... ............ ..... ... ... ... ........ ... .... 7 7 7 DC ELECTRICAL CHARACTERISTICS Input characteristics .. ........ ...... ... ...... ..... ... ...... ... ...... Output characteristics............................................. Power dissipation.. ... ..... ... ... ... ... ... ..... ......... ........ ..... Static and dynamic drive ......................................... 8 9 11 12 AC PERFORMANCE AC test conditions................................................... Propagation delay.. ... ........... ...... ........ ... ...... ... ..... .... Rise and fall time .. ........ ... ... ... ... ... ........... ................ Enable and disable times ........................................ Skew ....................................................................... Simultaneous switching noise ................................. 13 14 15 16 17 17 MISCELLANEOUS ISSUES ESD ........................................................................ Decoupling ........ ..... ... ......... ... ..... ...... ... ... ..... ...... ...... Maximum input rise and fall times .......................... Unused inputs .. ...... ..... ... ... ... ........ ...... ......... ..... ....... 19 19 19 19 SUMMARY ................................................................ 19 APPENDIX A Balanced-Drive capability........ ........ ...... ... ..... ... ... ... APPENDIXB Standard device specifications ............................... APPENDIXC Package information .. ... ... .............. ........ ... ... ... ... ..... The lOT logo is a registered trademark and Double-Density is a trademark of Integrated Device Technology, Inc. 6 20 21 24 DOUBLE-DENSITY LOGIC CHARACTERISTICS AND APPLICATIONS APPLICATION NOTE AN-117 DOUBLE-DENSITY LOGIC FAMILIES DOUBLE-DENSITY FAMILY FEATURES lOT's Double-Density FCT16xxxT and FCT162xxxT device fam ilies are an extension olthe lOT FCT-T (Fast Cm as TILcompatible - TIL level output swings) family. Many of the features discussed in this application note are common to all three Double-Density families, but since this document deals exclusively with the FCT16xxxT High-Drive and the FCT162xxxT Balanced-Drive devices, all further references to lOT Double-Density are used in connection with these last two families. Description lOT's Double-Density product line is comprised of 16, 18 and 20-bit fast CMOS interface parts available in 48 and 56 pin dual-in-line surface mount packages. Three broad families are available to serve different market segments. FCT16xxxT High-Drive Family The key features of the FCT16xxxT family are high static and dynamic drive. This family includes drivers which a suitable for driving low-impedance busses, back planes or heavily terminated lines while retaining excellent noise immunity. Certain standards such as the VME bus electrical specification require a high static output drive for which FCT16xxxT devices are a good choice. Low-impedance and high-capacitance driving ability are the hallmarks of the HighDrive family. Functions Double-Density devices, as the name indicates, are comprised of two byte-wide functions on a single die, in a single package. These components offer 16,18 or 20-bit bus widths with control signals for each group of 8, 9 or 10 bits (for standard functions) that correspond to their octal counterparts. This allows designers to use these new devices without changing their system architecture. Most of the popular octal functions are available in the Double-Density family. In addition to the dual byte-wide functions, there are several new functions available in Double-Density which broaden the product line offerings beyond the octal capabilities. These products take advantage of the larger pin counts of the packages by offering increased functionality while providing single control pins for both bytes. FCT162xxxT Balanced-Drive Family The FCT162xxxT family offers controlled edge rates and Balanced-Drive resulting in reduced undershoot and overshoot. Balanced-Drive components significantly reduce board level noise and solve many of the transmission line problems that develop with high-speed logic. The Balanced-Drive output structure has been designed to match the impedance of a typical transmission line for effective line driving. This is accomplished through a symmetric pull-up and pull-down current drive capability during transitions, and through integrated source-termination. The source termination emulates external series termination without severely limiting the drive capability of the device. The Balanced-Drive characteristics provide transmission line matching for signals undergoing both HIGH and LOW transitions. In addition, controlled output transitions reduce EMI and RFI effects. The main features of the Balanced-Drive family are reduced ground bounce, fewer line reflections and smaller over/undershoot making it very effective for point-to-point line driving. For further details on the Balanced-Drive implementation refer to Appendix A. Speed Grades The Double-Density components are specified at the same speed asthe industry standard FCT/FCT-Tfamily. This means that existing deSigns can be upgraded to reduce space and improve performance with very little effort. Operating Range lOT Double-Density family components have been characterized and specified over the extended commercial temperature range (industrial range) of -40 °C to +85°C and the power supply(Vcc) range of ±10% to be consistent with the trend to offer system deSigners a wider operating range and greater operating margin. FCT163xxx Low-Voltage(3.3V) Family lOT's LOW Vcc family is intended for use with a regulated 3.3V power supply and has been developed to maintain system speed at lower voltages. A benefit of using the reduced power supply is the significant reduction in the overall system power consumption. These devices will interface with the future and existing 3.3V devices. The family is designed to be TTL compatible and meets the J EDEC standard 8-1 A for lowvoltage interfacing. Another benefit of the low-voltage family is the decreased switching noise due to a reduction in overall output voltage swing. These devices are ideal for use in laptop and notebook computer designs. Space applications will also benefit from the lower noise and power levels from using a lower Vcc. lOT's low-voltage 163xxx family uses the same process technology as the other 5V Double-Density families, meaning that at 3.3V supply, the 163xxx family provides an excellent reliability margin due to minimal part stress. 7 DOUBLE-DENSITY LOGIC CHARACTERISTICS AND APPLICATIONS APPLICATION NOTE AN-117 DC ELECTRICAL CHARACTERISTICS on voltage below ground, the input clamp diode turns on as shown by the increase in current around -0.6V. The diode clamps negative voltage spikes at the input. Balanced-Drive and High-Drive devices have no clamp diode to Vee at the input. Consequently, voltages much higher than Vee can be tolerated with input breakdown occurring at approximately 14V as shown in Figure 3. These characteristics apply to pure inputs only, whereas I/O{inpuV output) pins will follow the characteristics shown in Figure 8. INPUT CHARACTERISTICS The input structure of the FCT16xxxT and FCT162xxxT Double-Density families is shown in Figure 1. The equivalent circuit for the input stage includes the ESD protection circuit, input clamp diode, TTL-CMOS translator and the hysteresis circuit. IIN(mA) CLAMP .... DIODE -40 -60 3020 drw 03 Figure 3. Double-Density Input Characteristics Figure 1. Input Circuit for Double-Density FCT16xxxT and FCT162xxxT Although breakdown occurs at approximately 14V, the absolute maximum steady state input voltage rating for reliable device operation is limited to a lower level. FCT16xxxT can tolerate a steady state voltage of 7V at inputs and I/O pins while FCT162xxxT can tolerate 7V at inputs and Vee + 0.5V at I/O pins. Beyond this, long term reliability of the device may be affected. The absolute maximum voltage specification of GND -Q.5V for all pins on both the FCT16xxxT and FCT162xxxT, and Vee +0.5V for I/O and output pins on the Balanced-Drive part refers to a steady-state condition. The specification does not imply that the device cannot withstand transient overshoots! undershoots greater than the rated values. Under transient conditions, the device is expected to withstand an overshoot of 7V or an undershoot of -3V for 20ns regardless of clamp diodes. While the part will not sustain damage from these overshoots/undershoots, it is possible that false switching may occur if there are device inputs sitting at a marginal voltage level or if other marginal conditions exist. In addition to the above, Double-Density devices have the advantage of extremely-low input capacitance where typical values range from 3 to 5pF. This makes it possible for a highimpedance source elsewhere on the board to drive many Double-Density inputs. All of the inputs for the Double-Density family are designed with 1OOmVof hysteresis in the input stage as shown in Figure 2. Due to the hysteresis, the effective LOW-to-H IGH and HIGH-to-LOW switching thresholds are separated. This feature improves both static and dynamic noise margins. It is also useful in providing immunity against overriding noise associated with slowly ramping inputs. High Low _ _ ...,~_ 1.45V 1.55V VIN (Volts) 3020drw02 Figure 2. Double-Density Transfer Characteristic showing hysteresis The typical input V-I characteristics for the High-Drive and Balanced-Drive Double-Density devices are shown in Figure 3. In the normal input operating voltage range of 0 to 5V, practically zero current is drawn by the input stage. The input stage offers very high-impedance and; therefore, low leakage currents «+/-51JA). When the input voltage is one diode turn- 8 DOUBLE-DENSITY LOGIC CHARACTERISTICS AND APPLICATIONS APPLICATION NOTE AN-117 VOH (V) OUTPUT CHARACTERISTICS The output stage of FCT16xxxT and FCT162xxxT devices has a "totem pole" configuration with N-channel pulldown and N-channel pullup transistors. The N-channel pull-up in the output stage reduces the output swing to TTL levels by limiting the output voltage in a logic HIGH state to around 3.5V (TTL logic high level). A simplified output circuit for the High-Drive family is shown in Figure 4a and a circuit for the BalancedDrive family is shown in Figure 4b. o 2 4 3 5 VOH SPEC· -25 -50 -75 ~ 3' Vcc -100~ FROM PREVIOUS STAGE LOGIC HIGH STATE -125 ...........-0 OUTPUT -150 Ta = 25°C Vee = 5V OESIGNAL ~=::jt==~ 3020 drw 05 3020 drw04 Figure 5. Logic High Output Characteristics Figure 4a. Output Circuit for FCT16xxxT 350 Vee 300 FROM PREVIOUS STAGE 250 ;;( .s OUTPUT OE SIGNAL 200 .J LOGIC LOW STATE .Q ~=:::t=~ 150 100 3020 drw 4b Ta = 25°C 50 Vee = 5V Figure4b. Output Circuit Schematic for FCT162xxxT 2 3 VOL (V) The FCT16xxxT and FCT162xxxTfamilies both have resistors in the drain of the device pull up structure giving the two families similar pull-up capabilities. The clamp diode to Vee in the pull up of the FCT162xxxT does not affect the output characteristics of the device if the output voltage remains between Vcc and ground. The typical logic high output impedance characteristics are shown in Figure 5 for both FCT16xxxT and FCT162xxxT devices. In the logic HIGH state, the output impedance, as obtained from the linear portion of the V-I characteristic, is 28!l. 4 5 3020 drw 06 Figure 6. FCT16xxxT Logic LOW Output Characteristics The typical output characteristics are different for the two families for the logic LOW state. This is because of a different pull down structure for the Balanced-Drive (FCT162xxxT) family which includes a resistor in the source of the pull down transistor. Because of this, the Balanced-Drive part has a higher output impedance when pulling down than the HighDrive part. The typical V-I curve for the High-Drive logic low state is shown in Figure 6. From the chart it can be seen that the output impedance for the logic LOW state is about 6!l which is sufficiently LOW to allow the part to drive low-impedance loads such as backplanes operating at high speeds. 9 DOUBLE-DENSITY LOGIC CHARACTERISTICS AND APPLICATIONS APPLICATION NOTE AN-117 FCT16xxxT output breakdown LOGIC LOW STATE 200 150 loul(mA) ;;( .s 100 Ta = 25°C Vcc= 5V -' .Q FCT16xxxT, FCT162xxxT parasitic clamp diode 50 234 VOL (VOLTS) 5 3020 drw 07 -.. -40 -60 Vcc YOUl(Y) 3020 drw 08 Figure 8. Output High-Z Characteristics for FCT16xxxT and FCT162xxxT Figure 7. FCT162xxxT Logic LOW Output Characteristics The Power-Off Disable feature is useful in interface applications where one side of the interface may intentionally or inadvertently become unpowered. Some of the applications where this is possible include Hot Insertion where a board is installed into an active, powered system or where a cable may accidently become disconnected or attached to a live system. The power off disable feature of the FCT16xxxT family ensures that the device outputs will remain in the high-impedance state when Vee=OV and that there is no path through which voltage applied to an output can raise the supply rail of the powered down device. Applications which may require the Power-Off Disable feature include power conserving systems, fault tolerant systems, telecommunications, networking systems and backplanes. These applications may use Power-Off Disable to power off unused portions of the system to conserve power, allow the replacement of faulty boards in an active system, or switch portions of a network in and out without disrupting network communications. Refer to AN-1 02 [6] for a detailed discussion on the Power-Off Disable feature. Any signal pin of an FCT16xxxT device may be at any voltage within the ABSOLUTE MAXIMUM VOLTAGE rating forthe input while the device is powered off. During the poweroff state, the device supply voltage is at OV and the device outputs remain in high-impedance, hence there is no contention when plugging such a device into an active bus. Devices having this feature specify an "IOFF" current in the datasheet. This is a measure of the leakage current at the outputs when Vee=OV. During power ramping, control (OE) signals must be suitably conditioned to ensure that the outputs remain in the highZ state. For negative assertion output enable pins, the output enable should be tied to a logic HIGH (>Vee/2). When the outputs of these devices are so conditioned, all 3-state outputs will offer high-impedance independent of the power supply (within ABSOLUTE MAXIMUM Vee limits). At Vee levels of less than 1.5V, the device outputs will not turn on, regardless of the state of the output enable. The FCT162xxxT family does not have the Power-Off Disable feature. The presence of a diode to Vee at the outputs of these devices creates a DC path to the supply pin during power-off conditions precluding use of these devices in poweroff situations. The V-I curve forthe Balanced-Drive part isshown in Figure 7 and shows an output impedance of about 12.50. This gives the Balanced-Drive family enhanced output edge rate control which improves the signal characteristics during transitions and improves the overshoot and undershoot characteristics of the device. Due to differences in the output structure of the Balanced-Drive from that of the High-Drive part, the Balanced-Drive will exhibiit about 0.5ns less propagation delay for the HIGH-to-LOW transition than the High-Drive part for loads of less than 200pf. A discussion of this can be seen in Appendix A. As the graphs show, the actual device static drive current under typical conditions is larger than that indicated by the datasheet specifications. The typical dynamic drive current during switching will be much larger than the data sheet static specification and can also be seen from the charts. For example, a device driving a capacitive load, upon switching, will easily drive over 100mA dynamic current until the load acquires the new state. The dynamic switching current for the FCT162xxxT family is specified in the data book at 115mA (typical for both directions), but the FCT16xxxT switching current will be much higher for the transition to a logic LOW. The output V-I characteristics (pure outputs and 1/0) for FCT16xxxT and FCT162xxxT in the high-impedance state are shown in Figure 8. Undernormal operating voltages, in the high-Z state, the pull-up and pull-down transistors are off and the output current is small (50~), as revealed by an almost flat line, close to zero on the current axis of the V-I graph. For output voltages one diode voltage drop below ground, the parasitic diode olthe N-channel pull-down transistor turns on, clamping the voltage at around 0.6V. For output voltages above Vce, since the FCT16xxxT devices have no parasitic diode to Vec, breakdown occurs at approximately 14V which gives the Power-Off Disable feature. FCT162xxxT devices have a parasitic diode to Vee at their outputs and hence voltages above Vce+0.5V are clamped. The user must maintain static output voltages within the maximum ratings stated in the datasheets in order to maintain reliable operation. 10 DOUBLE-DENSITY LOGIC CHARACTERISTICS AND APPLICATIONS APPLICATION NOTE AN-117 POWER DISSIPATION . Power calculations are done to determine power supply sizing, cooling/heat sink requirements and criteria for device selection. Power calculations can also determine maximum reliable operating frequency. Power dissipation calculations for High-Drive and Balanced-Drive parts are discussed below. 5 <' 4 §. 'S 3 .. 2 c.5 CD Q. g Power dissipation in an unloaded device is PD(unloaded} = Vcc x Ic where Ic= Total power supply current Vcc - VTP, only one of the transistors of the input translator is on and the other is off. Hence, the current drawn from the supply is very small. For VTN < VIN < Vcc - VTP, both the pull-up and pull-down transistors are partially on and current is drawn from the supply. The ~Icc is maximum at the switching threshold (nominally 1.SV for FCT/ FCT-T at Vcc=SV) and progressively tapers off for input voltages greater or less than this. Icco (loaded) = Icco (unloaded) + fCLVOH where CL = Load Capacitance 11 DOUBLE-DENSITY LOGIC CHARACTERISTICS AND APPLICATIONS 10 ACT 8 / / J o ~ ~ o -- / ABT ............... ~ ./'" ~ ..--- -- L~ 2 / I / ,-~ FCT-T 40 20 APPLICATION NOTE AN-117 a very low-static power, at higher frequencies they consume more total power than corresponding Bipolar/BiCMOS parts, due to a larger dynamic power supply component. This may have been true for older technologies. The Double-Density family, however, has dynamiC power dissipation characteristics that are much superior to older CMOS technologies and existing advanced BiCMOS logic families. As an example of a power comparison between ABT and FCT-Twith no load (Figure 10), an FCT-T part will consume the same current level at 34MHz as ABT at BMHz. It should also be noted that when the devices are not switching, ABT consumes a significant static current level while FCT will approach zero in current consumption. STATIC AND DYNAMIC DRIVE Drive is one of the key features that differentiates the FCT16xxxT and FCT162xxxT Double-Density families. HighDrive devices are specified at 64/-32mA static drive while the Balanced-Drive devices are specified at 24/-24mA static. The static drive is the current that the device is continuously capable of sourcing or sinking in a stable logic state. It is typically specified at the minimum VOH or the maximum VOL. Dynamic drive is the transient output current that the device can sink or source during a H IGH-to-LOW or a LOW-to-H IGH transition. Dynamic drive is not always specified in datasheets, but it can be determined from the device V-I graphs. Static drive indicates the ability of the device to maintain static noise margins in the presence of a DC load. It is the current at which VOH and VOL are specified. Dynamic drive, on the other hand, indicates the ability of the device to charge and discharge capacitive loads while meeting the AC performance limits and incident wave switching requirements. Unless there is a DC load on the line or a low-impedance line (e.g. a loaded backplane) is being driven, a large static drive is typically not required to drive CMOS devices to valid logic levels. Due to their large input impedance and low-input current requirements, CMOS devices do not require a driver with a large static drive capability except in the case of lowresistive termination. Dynamic drive, however, plays a more important role since it determines the speed derating with capacitive loading. FCT162xxxT devices have a modest static drive compared to FCT16xxxT devices; but, as indicated by datasheet specifications, they have ample dynamiC drive (typically> 100 mA at 1.5V). Due to the particular implementation of the source termination, unlike some other series terminated parts, the FCT162xxxT devices do not sufferfrom lackof drive duetothe additional impedance. 60 Frequency, MHz 3020 drw 10 Figure 10. IcC vs Frequency (no output loading) 30 '~r-~-~- ./ -~ ACY 24 :g Ci: 18 -1----- ./ -~~- E / o .!:! 12 V /' /" ...... V ./ V / V',/ " . / 6 -fo----- ~ o ~ o ~ 20 V./" A~ FCT-T "., 40 Frequency, MHz 60 3020drw11 Figure 11. Icc vs Frequency (with 50pF output load ing) Figure 10 shows the total power supply as a function of frequency for various 16-bit logic families. Figure 11 shows a similar graph for the loaded case. The total supply current in these graphs includes the static and dynamic power supply components but not dice as the inputs are at CMOS levels. inputs. Because of their internal structure, bipolar technologies have a larger static or quiescent power component than CMOS. CMOS has almost zero power dissipation under static conditions due to the fact that one transistor of the CMOS pair is off in a valid logic state leading to a high-impedance path between Vec and ground. It is seen that power dissipation characteristics of both FCT16xxxTand FCT162xxxTfamiliesaresimilar. Total power dissipation, under typical operating conditions, however, is an important Figure of merit. A common argument against CMOS families in the past has been that although they have 12 DOUBLE-DENSITY LOGIC CHARACTERISTICS AND APPLICATIONS 0.40 APPLICATION NOTE AN-l17 _. . ."..- _. "., .,- '-r- _. . -,- .VCC, = f?V ,~ 3'8t~Ta~.~.;5~'bd~·I+~·I"~~"I'·III~ 3.6 « E 1 > -£ o > 0.20 -60 -20 20 60 Temperature, °C 100 3.4 « 2.8.!OO 4.4 4.6 3020 drw 12a "" N . . 0.30 > .,; 0 > , 0.25 .- -~ . , .. ~ .... f,· , ,...1- "'F +"1'. 4.6 4.8 :~. • .. - ,, , j ':.t ft:- ~ ." ... 4.8 5.4 5.6 3020 drw 13b AC PERFORMANCE AC TEST CONDITIONS Double-Density is offered in the same speed grades as the octal FCT and FCT-T devices. With different speed grades available, the designer can choose cost/performance tradeoffs for the overall system implementation. The following chart shows the commercial speed grades for four of the most popular Double-Density functions. . , ""t' ..J.... . . ~......... 5.0 5.2 Vce, V Figure 13b. VOH vs VCC at 25°C ·1'a _25'oC ! 0.35 E @J 1 '''~- :.:=~-;t'''';FI·al": Figure 12a. VOL vs Temperature at VCC = 5V 0.40 . .. 0.20 4.4 5.0 Vee, V 5.2 5.4 5.6 3020 drw 12b Function 244 245 373 374 NOTE: Figure 12b. VOL vs VCC at 25°C Because of the excellent noise characteristics of BalancedDrive devices, it is in the best interest of the designer, to use Balanced-Drive parts unless high static drive is essential. Figures 12a and 12b show the variation of VOL with temperature and Vee at a drive current of 24mA static drive. Figures 13a and 13b show similiar characteristics for VOH at -24mAstatic drive. Referto appendix B forstandard datasheet specifications. 3.8 .~ ,.... 3.6 « E 3.4 @J 3.2 "" "t >-£ -'.- ,- -1-'·- "·"-H- ,+, .. Vcc ... .... ... . .. .-t. ~ ~ ~ m II- _.... .+ ............................. ,- ." , , ~ ......... .- .~ ,..... ,- Aspeed(l) C Speed(l) 6.5 7.0 8.0 10.0 4.8 4.6 5.2 6.5 4.1 4.1 4.2 5.2 3020 1b1 01 1. All speeds in ns. AC limits of the Double-Density family are specified over extended commercial, industrial and military temperature ranges. Vee tolerance is ±10 % for both commercial and military devices. The limits are based on the worst case operating conditions with Vee = 4.5V and TA = 85°C for commercial grade limits and TA = 125°C for military grade limits. All AC parameters are measured using the industry standard load shown in Figure 14. AC tests are done with a single bit switching and timings may need to be derated for the worst case of 16/18/20-bits switching simultaneously. -J..- ._,.....' ,- .., -+- ,...- , " 5V Standard Speed(l) - 1"- ,......, ." _. I- ~ 7.0V Vcc 0 > 3.0 2.8 -60 500n -20 20 Temperature,oC 60 VIN 100 VOUT I-f<+-......-I D.u.T. I-f<+-......- 3020 drw 13a Figure 13a. VOH vs Temperature at VCC = 5V ... RT 500n 3020 drw 14 Figure 14. Standard Logic Test Load 13 DOUBLE-DENSITY LOGIC CHARACTERISTICS AND APPLICATIONS APPLICATION NOTE AN-117 1.2 The 50pF capacitance in the standard load allows for stray and parasitic capacitances due to the ATE test fixture. In most measurements the switch is in the open position. The 500n resistors offer a nominal DC load and help establish "artificial" logic levels for testing enable and disable time parameters. In many cases, the test load may not be a true representation of a customer's load. When the loading is lighter, the limits may be used as a worst case, but when the loading is heavier they may need to be derated. Also the system under consideration may normally operate in a narrower temperature and/ or Vee range. In such circumstances, it is useful to have the option to derate for the different variables that affect the AC characteristics. In order to serve this purpose a variety of derating graphs are provided in the following sections. 1.1 1;c Cii 1.0 "tI Gl .!:! i6 Derating factor E 0 : -H#+ = .0125 x tpo 110°C : N 0.9 z 0.8 -100 o 100 200 Temperature,OC 3020 drw 16 Figure 16. FCT162xxxT Delay vs Temperature PROPAGATION DELAY DELAY AS A FUNTION OF TEMPERATURE Figures 15 and 16 show the normalized graphs of propagation delay as a function of temperature for FCT16xxxT and FCT162xxxT. The FCT16244T and FCT162244T are used as a basis for these graphs. 1.2 tPLH i tPHL i ---GI- -+-- >- .!!! 1.1 l fit : f-Vcc = 5V Note: Due to the similarity in output structure, the graphs are representative of the characteristics of all devices belonging to the same family. Derating factors are useful in accounting for the effect of a variable on a specific parameter. For example given a temperature derating factor of 0.0157tPD/1 O°C for FCT16xxxT, the FCT16245CT (delay spec = 4.1 ns) when operating at 25°C can be assumed to have a delay of: Gl C "tI Gl tPD = [4.1 - 0.015 x 4.1 (85-25)/10] = 3.73ns max Delay as a Function of Supply Voltage Figures 17 and 18 show the normalized graphs of propagation delay as a function of supply voltage. Again the data is based on the FCT16244T and FCT162244T, but is representative of all devices within each of these families. With an increase in Vee the current available for charging and discharging goes up and hence the propagation delays go down. For reduced Vee, the current goes down and there is an increase in the propagation delay. .!:! iii E 0 1.0 z 0.9 -55 ~ Derating factor = N.015? x tpo/l O°C ! -5 45 Temperature,OC 95 145 3020 drw 15 1.1 Figure 15. FCT16xxxT Delay vs Temperature 'l, 1;- Propagation delay is related to the current available for charging and discharging the internal and external capacitances. The current itself depends on a number of process parameters, which vary with temperature. As a result, propagation delay has a temperature gradient as shown. ;!l l5 'iii J i i 1.0 .- ~-t~! ~- tpLH tpHL ----G~~-. " , I ._- ~ ~ D.. f..-..- ~~ ~ -t~ 'i Z~ 0.9 4.4 4.6 4.8 5.0 5.2 Vee Figure 17. FCT16xxxT Delay vs VCC 14 5.4 5.6 3020 drw 17 DOUBLE-DENSITY LOGIC CHARACTERISTICS AND APPLICATIONS 1.2 ie' ~ '~."'.""''''~ iii ............... Q c 0 ~ 1.1 -- ,......_- ------ ............. --- ................. ~ .,D.. Cl ~ D.. tPHL tPLH ......... 6 -- ! .- E ............ 1.0 --- --=t-~~t=~- ,.......-- _._i--- - c ~ 5 4 - !',C, ........... ..,- ":...- ~ ...... .... 3 l - ~ ~ I ........... ... --- .................... -- --+-' / ~ f----- --I I ~ 0.9 4.5 4.75 5 50 5.5 5.25 Vee, V 150 CL,pF 3020 drw 18 Delay as a Function of Number of Outputs Switching 8 The AC characteristics are tested and specified for single bit switching. In reality, multiple bits can switch simultaneously generating internal noise which causes the observed propagation delays to change. Figure 19 shows the change in propagation delay as a function of number of outputs switching. The data forthis graph is based on the FCT16245T and FCT162245T. ._- -- --- -- 1---6 ~ ! tPHL ! 5 & 4 .--- 3 .....- ..,.r- tPLH 2 2 ~ -- -- ._- ,- - 50 c 16xxx .- ,....~ o 150 CL,pF - 10 # of outputs switching ~>- - ! -- -- "- -- ,......- - -- --I 250 3020 drw 21 Figure 21. FCT162xxxT Delay vs Load - 162xxx I/) 3020 drw 20 ! 7 3 ._f--. 250 Figure 20. FCT16xxxT Delay vs Load Figure 18. FCT162xxxT Delay vs VCC o -- ---=r-":-P" .-- 2 z E -+--+--i o 0.4 l 4.4 ~r 1.0 iii E ~l 0.8 50 100 150 Load,pF 200 250 5.2 5.4 3020 drw 26 Figure 26. Enable Time vs VCC for FCT16xxxT 300 3020drw23 Figure 23. FCT162xxxT Trlse and Tfall vs Load ENABLE AND DISABLE TIMES Like propagation delay and rise and fall times, enable and disable times also vary with temperature and Vee. Figures 24, 25, 26 and 27 show the variation of enable time with temperature and Vee and are based on data taken on the FCT16245T and FCT162245T. Variation of disable time is more a function of the load than the device itself. Therefore derating of disable time is independent of the device but depends on the temperature and Vee coefficients of the load. 16 5.6 DOUBLE-DENSITY LOGIC CHARACTERISTICS AND APPLICATIONS 2.0 -----I!I- CD E i= CD jj 1.6 1.2 ~iU 1 0.8 0.4 z 1 , , E Is tPZL tPZH j as I: W APPLICATION NOTE AN-117 ~t- , ! 0.0 4.4 ! +i . ~ ... - .. - 4.6 + j ! 4.8 5.0 Vee, V 5.2 5.4 up. Due to this correlation between speed and ground bounce, high-speed logic families arouse increased concern over noise due to the simultaneous switching of outputs. The standard method of measuring ground bounce involves keeping one output in a logic LOW state and switching all other outputs at a given frequency (1 MHz typical). The quiet output will reflect the bounce on the internal device ground. Similarly Vee bounce can be observed by keeping one output in the logic HIGH state. The naming convention for ground bounce characterization is as follows: VOLP = Peak positive ground bounce on a LOW output VOLV = Peak negative ground bounce on a LOW output VOHP = Peak positive Vee bounce on a HIGH output VOHV = Peak negative Vee bounce on a HIGH output 5.6 3020 drw 27 Figure 27. Enable Time vs VCC for FCT162xxxT SKEW Skew reflects the worst case propagation delay difference between the different outputs. Skew results from design (routing/placement), process and packaging specific to the device. For the user, as the clock frequency goes up, timing margins become tighter. Signal skew makes the timing margin even tighter. Reduction in the spread of the output propagation delays results in reduced skew and improved timing margins. There is a tendency among designers to compute the difference between the maximum and minimum specifications and use this Figure as the worst case skew; however, this is not a realistic measure of skew because the maximum (worst case) and minimum (best case) delays occur under mutually exclusive conditions of temperature, power supply voltage and process. In order to provide the user with the required information, Double-Density devices have an output skew specification of O.Sns maximum between outputs on the same device. For further discussion on types of skew, the reader is referred to Application Note AN-82[8]. V "5 c.. "5 0 ~ Cl 0 -1----!--:C....~-~~-.......,_======I -....;;:::......- VOLV o Time 3020 drw 28 Figure 28. Ground Bounce for HIGH-to-LOW Transitions V SIMULTANEOUS SWITCHING NOISE In a digital circuit when multiple outputs switch, the current through the ground or Vee lead changes rapidly. As this current flows through the ground (or Vee) return path, it develops a voltage across the parasitic inductance of the bond wire and package pin. This phenomenon is called simultaneous switching noise. The noise is seen as ground bounce or Vee bounce and can cause problems such as data loss and false triggering in a system. For a detailed discussion of ground bounce the reader is directed to Application Note AN-47 [4]. Ground and Vee bounce cannot be entirely eliminated, but they can be minimized by controlling edge rates, reducing output swing and providing multiple power, ground and Vee pins. Double-Density has 8 ground pins and 4 Vee pins. This is a marked improvement over the ground distribution in the octal devices and manifests itself in the noise characteristics of the device. Ground bounce depends on many factors with device speed being one of the important influences. As device speed goes up, the rate of change of current in the parasitic inductances increases and the related switching noise goes 0 -1-_ _ _1-_.::..._ _ _ _ _ _ _ _ _ _ _ _-1 "[ "S 0 o Time 3020 drw 29 Figure 29. Vee Bounce for LOW-to-HIGH Transitions The "ground bounce" phenomenon is primarily dependent on the output edge rate, number of outputs switching and package lead inductance. Other factors that can affect ground bounce are the output swing and drive. Most current technology consists of TTL -compatible logic where the output drivers can be made to pull the outputs from rail-to-rail with a p.. channel pullup transistor ("CMOS level outputs") or they may 17 DOUBLE-DENSITY lOGIC CHARACTERISTICS AND APPLICATIONS APPLICATION NOTE AN-117 Ground Bounce vs Load drive the output to approximately 3V with an n-channel pullup (''TTL-level outputs"). The equations below illustrate why the ground bounce is higher with CMOS level output swings. i = CldVo Simultaneous switching noise is affected by the output loading. As the value of the load capacitance increases, the charge stored in the capacitance is increased. The larger charge implies a slower rate of discharge through the package parasitics and hence lower ground bounce as shown in Figures 31 and 32. Another way to look at it is, the load slows down the signal transition increasing the transition time and reducing ground bounce. ... (i) ill where, Vo = Output voltage CL = load capacitance Vg = voltage induced on ground pin Lg = inductance at the ground pin Vg = dt using (i) in (iia), we get Vg = LgCL d2 Vo > a: ... (iib) ;g dt2 ~ 0.8 -g ..5Cl _ 111 "0. iii E o z VOLV // 0.6 // I? -- 300 400 3020 drw31 FCT16xxxT FCT162xxxT 1.2 > ;! ~ 0.8 0.4 0.0 0 100 200 CL,pF Figure 32. VOlV vs load VOLP 10 # of outputs switching 200 1.6 'l o 100 2.0 .-- !:::="'" r- i-o -~ .~ Figure 31. VOlP vs Load ij 0.4 0.2 ./ "/ t-- : -..- _. -- .~ CL,pF 1.2 --r-~ 0.8 0.0 Every logic family has a characteristic edge rate (for a given load condition) and every package has a fixed lead inductance. Hence given a specific family, package, and loading, a distinct correlation can be observed between the number of outputs switching and the amount of undershoot/overshoot due to ground bounce. Figure 30 illustrates this relationship for the Double-Density devices in an SSOP package. 1.0 TA= 25°C .Yr:.9. ..?y. .... ...: 0.4 Ground Bounce vs Number of Outputs Switching 2l 1.2 FCT16xxxT FCT162xxxT -~ Because of the lower voltage swing, TTL level outputs give better noise immunity than CMOS, rail swing, output levels. The other factor affecting ground bounce is output drive. The higher the output current, the greater the switching noise. This is evident from the equation (iia) above. FCT16xxxT/162xxxT devices are available in SSOP and Cerpack packages. These packages have multiple ground and Vee pins, giving multiple parallel paths and low inductance to the high-speed switching currents. The surface mount package avoids the inductance of through hole mounting and is the package configuration for high-speed logic. The combination of these package characteristics results in a significant reduction in the magnitude of simultaneous switching noise. Package dimensions and pin inductance values are shown in Appendix C. c ~ 1.6 ... (iia) Lg di - 2.0 20 3020 drw 30 Figure 30. Double-Density ground bounce vs number of outputs switching (16-bit function) 18 300 400 3020 drw32 DOUBLE-DENSITY LOGIC CHARACTERISTICS AND APPLICATIONS APPLICATION NOTE AN-117 MISCELLANEOUS ISSUES MAXIMUM INPUT RISE AND FALL TIMES When an input rises or falls slowly, the noise overriding the input can cause the device to oscillate for some time before the output becomes steady. Such oscillations are not desirable because they cause the effective propagation delay to go up by a large amount, and affect system reliability. There is a restriction on the maximum input rise and fall time for reliable circuit operation. For FCT16xxxT, the slowest recommended rise and fall times on the device inputs are SOns. The hysteresis circuit in the input stage (see Figure 2) helps in reducing the effect of noise on slowly changing inputs. ESD There are two standard methods to characterize ESD, the machine model and the human body model. Double-Density devices meet MIL-STD-883 (Method 3015) Class 2 ESD level requirements. This means that they can withstand voltage levels greater than 2KV (4KV typical) under the human body model test methodology. With the machine model test methodology, Double-Density ESD tolerance is greater than 200V. Storage of devices in protective foam and the use of a ground strap during handling are standard precautions. UNUSED INPUTS DECOUPLING Good decoupling techniques are required to achieve optimum noise and performance levels from Double-Density components. The multiple ground and Vee pins on the SSOP and Cerpack packages reduce the effect of the lead inductance of the devices, but do not provide a single point for the capacitor. The best style and location of the decoupling capacitor is the use of a surface mount chip capacitor placed directly over the component, on the reverse side of the board. An alternative location for boards with components on one side only is to place the capacitor at either end of the component. The capacitor should have a LOW series inductance and resistance in order to maximize the effect of the capacitance. The function of the capacitor is to provide a lowimpedance path to high-frequency components, while also satisfying the load's initial high-current requirements. O.3 J The unused inputs of logic devices should not be left floating because these high-impedance inputs act like antennae and pick up noise. To prevent the noise on such inputs from switching the device and consuming extra power, all unused inputs must be connected to Vee or ground, preferably via a current limiting series resistor to limit unwanted current surges. SUMMARY lOT has introduced the Double-Density logic family to meet the demands of high-performance systems. These highintegration components save board space while maintaining the compatibility with the older octal devices in terms of speed and functionality. Three different varieties are offered to meet to the users' diverse application needs. The standard HighDrive family is ideal for bus/backplane applications. The Balanced-Drive family has user friendly outputs for low noise, reliable systems. The 3.3V family is ideal for power-sensitive ~pPlications such as laptops and notebooks. The various characteristics, graphs, and information provided herein are intended to aid the designer in achieving desired cost/performance goals. 3.3 ~ ---. ) ._ Z = 30n ) ~------------3020 drw 33 Figure 33. Buffer Driving a Loaded Trace REFERENCES [1] "Series Termination"(AN-SO), High-Speed CMOS Logic Design Guide, lOT, Inc., November 1991. [2] "Power Dissipation in Clock Drivers"(AN-S1), HighSpeed CMOS Logic Design Guide, IDT, Inc., November 1991. [3] "Characteristics of PCB Traces"(AN-49), High-Speed CMOS Logic Design Guide, IDT, Inc., November 1991. [4] "Simultaneous Switching Noise"(AN-47), High-Speed CMOS Logic Design Guide, IDT, Inc., November 1991. [5] "3.3V Logic"(AN-1 03), IDT 3.3V Logic Product Information, IDT, Inc., February 1992. [6] "Power off Disable in lOT Double-Density Devices"(AN102), lOT, Inc., 1992-93. [7] "Decou piing Double-Density Components"(AN-116), IDT, Inc.,1992. [8] "Clock Distribution Simplified With lOT Guaranteed Skew Clock Drivers"(AN-82), Logic Databook, lOT, Inc., 199091. As an example of a decoupling situation, Figure 33 shows a driver driving a 30n transmission line as could be encountered when driving a heavily loaded backplane. When the output of the buffer switches LOW-to-HIGH, about 100mA (dynamic) drive is required [(3.3 - 0.3)/30 = 100]. For 16 bits switching simultaneously 1.6A dynamic drive is required. For a Vcc drop of 0.1 V and a 3ns rise time on the signal, the value of the decoupling capacitor can be decided as shown below. i = C(dV/dt) approximating this with deltas 1.6 = C(0.1/3x10**-9) Hence, C = 0.0481lF With this calculation, the value of the capacitor can then be selected from the standard capacitor values. A capacitor equal to the value in the equation or larger should be selected to achieve the noise value decided upon. The value of 0.11lF is a good choice. The application note AN-116, Decoupling Double-Density Components, contains a complete discussion on capacitor selection and placement. 19 DOUBLE·DENSITY LOGIC CHARACTERISTICS AND APPLICATIONS APPLICATION NOTE AN·117 APPENDIX A BALANCED-DRIVE CAPABILITY The output structure of the Balanced·Drive part (Figure 6) has been specifically designed to give similar transition times with matching absolute edge rates in both the high going and low going directions. This means the lOT Balanced·Drive Family has a matched dynamic drive capability for driving transmission lines, but still has sufficiently low static output impedance to maintain state during steady state conditions. As shown in Figure A 1, during transitions the Balanced-Drive output structure will current limit at about 115mA typical forboth low-to-high and high-to-Iow transitions, limiting surge current into the transmission line, but still providing sufficient drive current to drive all except the most demanding loads. Limiting the pull down drive capability of the device during transitions also reduces problems with undershoot, ground bounce and general noise generation. In almost all applications, the Balanced-Drive part will generate less noise than BiGMOS components and the associated ringing that BiGMOS generates. 3020 drw35 Figure A2, Static Balanced·Drive Output Adding External Resistors to Balanced-Drive With the Balanced-Drive components, the effective series termination impedance of 2sn is slightly lower than the characteristic impedance of most transmission lines. Backplanes may sometimes approach levels as low as 30n, but most lines will have a higher impedance. Despite this difference, the current surge driven into the transmission line will be limited by the output impedance and controlled edge rate. If it is desirable to further enhance the quiet operation and to achieve a close match between the source impedance and the line impedance, the user may add external resistors to the output of his Balanced-Drive parts. Because of the integrated resistor in the drain olthe pull up FET, and the output structure of the Balanced-Drive part, the addition of an external series resistor will not cause significant attenuation of the output signal in the pull-up structure of the device as happens with standard TTL output components. This means that when series resistors are added to a Balanced-Drive part, the LOWto-HIGH edge rate will retain similar .characteristics to the HIGH-to-LOW edge rate. With similar edge rates, the performance and noise levels will be superior to situations where an external resistor has been added to a standard TIL output part. 3020 drw 34 Figure A1, Dynam ic Balanced·Drive Output During steady state DG conditions, lOT specifies the static drive capability of the devices at 24mA worst case in both directions as shown in Figure A2. Balanced-Drive components will give better line matching than equivalent drivers with lower output impedance eliminating the need for external resistors in most cases. Driving Capacitive Loads Standard parts which have a low pull-down impedance and a high pull-up impedance have problems with RG time delays when driving capacitive loads. This is due to the inability to match the line impedance for both directions. Usually the addition of a large enough resistor to perform line matching in the pull down direction will cause significant degradation of the pull up capability of the part. IDT's Balanced-Drive parts have overcome this problem by providing equal line driving capability in both directions and will drive capacitive loads equallywell in both directions with or without external resistors. This means that adding a series resistor will not degrade the speed of the part significantly because the effective line matching resistor will have a lower resistance value than one that would be used with a standard part. 20 DOUBLE-DENSITY LOGIC CHARACTERISTICS AND APPLICATIONS APPLICATION NOTE AN-117 IDT's Balanced-Drive and High-Drive components have very similar response times for capacitive loads of less than 300pf. Because of its quicker response, the Balanced-Drive part will switch faster than a High-Drive part for loads of less than 200pf. As the load increases above 200pf the High-Drive part will respond more quickly for a high-to-Iow because of its lower output impedance. Despite these minor differences the differences in the effects of capacitance on propagation delay between the High-Drive and Balanced-Drive parts should be less than 0.5ns for any load of less than 300pf. The capacitive derating curves are shown in Figure A3. When the capacitive load is larger than 400pf at high switching speeds, the load will begin looking like a direct short upon switching and may begin to overdrive the output of both the Balanced-Drive and High-Drive part. The power dissipation equation is P = fCV 2 and should be used to calculate the power consumption for each pin to assure that the total power dissipation in the device does not exceed the maximum rated limits. To avoid exceeding the limit for heavy capacitive loads, a 100n series resistor can be used to limit the current to acceptable levels. APPENDIXB STANDARD DEVICE SPECIFICATIONS This section includes specifications and ratings for lOT's High-Drive and Balanced-Drive families. The standard tables for DC Electrical characteristics, output drive, capacitance and absolute maximum ratings are common to all DoubleDensity device types and are the same as in any characteristics are similiar if not identical for all devices. CAPACITANCE (TA = +25°C, F = 1.0MHz) Symbol CIN Parameter(1) Conditions Input Capacitance Typ. Max. Unit VIN = OV 4.5 6.0 pF pF 1/0 ClIO Capacitance VOUT= OV 5.5 8.0 COUT Output Capacitance VOUT= OV 5.5 8.0 NOTE: 1. This parameter is measured at characterization but not tested. pF 3020 tbJ 02 8~------------------------------~ ABSOLUTE MAXIMUM RATINGS 7 Symbol Rating VTERM(2) Terminal Voltage with Resect to GND -0.5 to +7.0 -0.5 to +7.0 V VTERM(3) Terminal Voltage with Resect to GND -0.5 to VCC -0.5 to VCC V TA Operating Temperature -40 to +85 -55 to +125 °c TBIAS Temperature Under Bias -55 to +125 -65 to +135 °C TSTG Storage Temperature -55 to +125 -65 to +150 6 5 ~ 4 :J' e.3 2 o 50 100 150 200 250 300 Capacitive Load (pf) Figure 350 PT 400 3020 drw 36 lOUT A3, Capacitive Derating for Double Density Power DiSSipation DC Output Current Commercial 1.0 Military 1.0 -BOto +120 -60 to +120 Unit °C W mA NOTES: 3020 tbl 03 1. Stresses greater than those listed under ABSOLUTE MAXIMUM RATINGS may cause permanent damage to the device. This is a stress rating only and functional operation of the device at these or any other cond itions above those indicated in the operational sections ofthis specification is not implied. Exposure to absolute maximum rating conditions for extended periods may affect reliability. 2. All device terminals except FCT162XXXT Output and 1/0 terminals. 3. Output and 1/0 terminals for FCT162XXXT. Balanced-Drive and High-Drive SUMMARY Utilizing the Balanced-Drive family in high speed bus driving applications will solve most of the problems with noise and line termination experienced with traditional high-speed, low-output impedance line drivers. The Balanced-Drive logic family provides the benefits of series termination on signal lines while avoiding many of the drawbacks of adding external series termination resistors. The Balanced-Drive family of devices from lOT will generate lower levels of noise than other components in the same speed grade, solve many of the termination line problems encountered by the designer, reduce power consumption and give superior performance to other high speed logic. Balanced-Drive is the logic family of choice for most applications. 21 DOUBLE-DENSITY LOGIC CHARACTERISTICS AND APPLICATIONS APPLICATION NOTE AN-117 POWER SUPPLY CHARACTERISTICS Symbol Parameter Typ,C2) Max. Unit - 0.5 1.5 mA VIN=VCC VIN=GND - 60 100 Vcc = Max. Outputs Open 1i = 10MHz VIN= VCC VIN= GND - 0.7 2.5 50% Duty Cycle xOE= xDIR = GND One Bit Toggling VIN = 3.4V VIN= GND - 0.9 3.3 Vcc = Max. Outputs Open 1i = 2.5MHz VIN= VCC VIN = GND - 2.5 5.5(5) 50% Duty Cycle xOE = xDIR = GND Sixteen Bit Toggling VIN = 3.4V VIN = GND - 6.5 17.5(5) Test Conditions(1) .:llcC Quiescent Power Supply Current TTL Inputs HIGH VCC= Max. VIN = 3.4V(3) ICCD Dynamic Power Supply Current(4) Vcc = Max. Outputs Open xOE = xDIR = GND One Input Toggling 50% Duty Cycle Ic Total Power Supply Current(S) Min. NOTES. 1. For conditions shown as Max. or Min., use appropriate value specified under Electrical Characteristics for the applicable device type. 2. Typical values are at Vce = 5.0V, +25°C ambient. 3. Per TTL driven input (V IN = 3.4V); all other inputs at Vcc or GND. 4. This parameter is not directly testable, but is derived for use in Total Power Supply Calculations. 5. Values for these conditions are examples of the Icc formula. These limits are guaranteed but not tested. 6. Ie = IQUIESCENT + IINPUTS + IDYNAMIC Ic = Icc + Alcc DHNT + ICCD (fCPNCPf2 + fiNi) Icc = Quiescent Current (ICCL, ICCH and Iccz) Alcc = Power Supply Current for a TTL High Input (VIN = 3AV) DH = Duty Cycle for TTL Inputs HIGH NT = Number of TTL Inputs at DH Icco = Dynamic Current Caused by an Input Transition Pair (HLH or LHL) fcp = Clock Frequency for Register Devices (Zero for Non-Register Devices) NCP = Number of Clock Inputs at fcp fi = Input Frequency Ni = Number of Inputs at fi 22 flA/ MHz mA 3020 tbl 04 DOUBLE-DENSITY LOGIC CHARACTERISTICS AND APPLICATIONS APPLICATION NOTE AN-117 DC ELECTRICAL CHARACTERISTICS OVER OPERATING RANGE Symbol Parameter Test Condltions(1) VIH Input HIGH Level Guaranteed Logic HIGH Level 2.0 VIL Input LOW Level Guaranteed Logic LOW Level Input HIGH Current (Input Pins) Vee= Max. VI = vee - VI = GND - IIH Input HIGH Current (I/O Pins) Input LOW Current (Input Pins) ilL - Vo= 0.5V - (3-State Output Pins) VIK Clamp Diode Voltage Vee + Min., liN = -18mA loS Short Circuit Current Vee = Max., va = GND 3) -80 10 Output Drive Current Vee = Max., Va = 2.5V(3} -50 VH Input Hysteresis Quiescent Power Supply Current Vee= Max. Va = 2.7V High-Impedance Output Current 10ZL - Vee = Max., VIN = GND or Vee Max. - - Input LOW Current (I/O Pins) 10ZH leeL Typ'!;':) Min. Unit -V 0.8 V ±5 J.tA ±15 ±5 ±15 J.tA ±10 ±10 --{).7 -1.2 V -140 -200 mA -180 mA - - 100 - mV 0.05 1.5 mA leeH leez 3020 tbl 05 OUTPUT DRIVE CHARACTERISTICS FOR FCT16XXXT Symbol VOH Parameter Output HIGH Voltage Min. Typ.(2) Max. Vee = Min. 10H =-3mA 2.5 3.5 V VIN = VIH or VIL 10H =-12mA MIL. 2.4 3.5 - 2.09 3.0 - V - 0.2 0.55 - - ±100 Test Conditions(1) Unit V 10H = -15mA COM'L. 10H = -24mA MIL. 10H = -32mA COM'L(4) VOL 10FF Output LOW Voltage InpuVOutput Power-Off Leakage Vee= Min. 10L = 48mA MIL. VIN = VIH or VIL 10L = 64mA COM'L. Vee = OV, VIN or Va :'>4.5V V J.tA 3020 tbl 06 OUTPUT DRIVE CHARACERISTICS FOR FCT162XXXT Symbol Parameter Test Conditions(1) Min. Typ.(2) Max. Unit 10DL Output LOW Current Vee = 5V, VIN = VIH or VIL, VOUT = 1.5V(3) 60 115 150 mA 10DH Output HIGH Current vce = 5V, VIN = VIH or VIL, VOUT = 1.5V(3) -60 -115 -150 mA VOH Output HIGH Voltage vce= Min. 10H = -16mA MIL. 2.4 3.3 - V VIN = VIH or VIL IOH = -24mA COM'L - 0.3 VOL Output LOW Voltage Vcc= Min. IOL = 16mA MIL. VIN = VIH or VIL 10L = 24mA COM'L NOTES: 1.For conditions shown as Max. or Min., use appropriate value specified under Electrical Characteristics for the applicable device type. 2.Typical values are at Vcc = 5.0V, +25'C ambient. 3.Not more than one output should be tested at one time. Duration of the test should not exceed one second. 4.Duration of the condition can not exceed one second. 23 0.55 V 3020 tbl 07 DOUBLE-DENSITY LOGIC CHARACTERISTICS AND APPLICATIONS APPLICATION NOTE AN-117 APPENDIXC PACKAGE INFORMATION AREA COMPARISON FOR CERPAK PACKAGES Double-Density devices are offered in the space saving 48- and 56-pin SSOP (shrink small outline package) for commercial and the 48- and 56-pin CERPACK for military. Figure C1 shows the dimension details for the two SSOP packages, and figure C2 shows the dimensions of the two CERPACK packages . 48 CERPAK ................r-----."..:;::;;;:::::;----.- 390 mil 390mil Body Length 626 mil 727 mil Body Height 86 mil 86 mil Body Weight .6g .6g 8JC 8JA Pin Inductance 25 mil Pitch 56 CERPAK Body Width 5 5 70 75 1.5nH (ground) 1.8nH (ground) 3.1nH (I/O) 3.1nH (I/O) 3020 drw 09 0.625/ 0.725 -I 1 • The thermal characteristics are used for estimating the junction temperature for given loading and frequency conditions. The devices should be operated below the maximum junction temperature for reliable operation over an extended period of time. For the FCT16xxxT family the maximum junction temperature allowed is about 150°C. 3020 drw37 0.406 Figure C1. The Physical Dimensions of SSOP 25mil Pitch Both SSOP packages have the same width and pin-to-pin separation but different body lengths. Due to the fine pitch (25 mil), significant area savings are achieved over multiple octal packages. Table C1 shows the key dimensions and characteristics of the SSOP packages. 48SS0P 56 SSOP 300 mil 300mil Body Length 625 mil 725 mil Body Height 2.74mm 2.74mm Body Weight 0.6grams - 8JC 8JA Pin Inductance 17 17 70 70 1.5nH (ground) 1.8nH (ground) 2.9nH (VO) 3.1nH (1/0) 0.727 3020 drw 38 AREA COMPARISON FOR SSOP PACKAGES Body Width 0.6261 Figure C2, The Physical Dimensions of CERPACK 3020 drw 08 24 ~ ~ POWER-OFF DISABLE IN lOT'S FCT16xxxT DOUBLE-DENSITYTM DEVICES APPLICATION NOTE AN-102 Integrated Devlc:e Technology, Inc:. By Stanley Hronik INTRODUCTION Power-off Disable is a feature of IDT's FCT16xxxT family which allows a user to shut down portions of his system, or to remove and install boards or peripherals during full power conditions, in a properly designed system. In order to have the ability to remove and install during live power conditions, the devices must be able to withstand the maximum rated voltages on their inputs and outputs when Vcc is between 5 volts and zero. The devices must not source or sink current that would affect the remaining fully functional system, or drive an active data bus in the power off state. The purpose of the Power-off Disable feature is to maintain data integrity, to eliminate bus contention, and to avoid component damage during power off while tied to or while being inserted into or removed from an active system. This application note describes the Power-off Disable feature built into IDT's FCT16xxxT High-drive Double-Density family of products and portions of the eight bit FCT6xxTfamily, both of which are designed for standard 5V supply operation. Design techniques and applications for partially powered systems are discussed and the circuit configurations used in the FCT16xxxT Double-Density family of components to achieve the Power-off Disable feature are described. the computer without powering off the system. Also, frequently a cable will become inadvertently disconnected during system installation, or if an activity adjacent to a system causes a disruption of the setup. Under these conditions the components interfacing the disconnection point should sustain no damage from the powered disconnection and subsequent powered reconnection. COMPONENT SELECTION AIiIDT logic components will support Power-off Disable on their inputs. The IDT logic components that will support Power-off Disable on their outputs, and liD ports include all of the FCT16xxxT High Drive family, the FCT164245T 3.3 volt translator, and FCT620T, FCT621T, FCT622T, FCT623T, FCT646T, FCT648T, FCT651T, and FCT652Tfrom the eightbit family. Component families which will not support Power-off Disable on their outputs have a parasitic clamp diode from the device output or liD to Vcc. Some of these families are the FCT162xxxT Balanced Drive parts, the FCT 3.3 volt devices (except the translator), the FCT CMOS output devices, and the FCT -T octals which have not specifically been mentioned as having Power-off Disable. APPLICATIONS OF POWER OFF DISABLE Vee There are many applications for the Power-off Disable feature in systems that either intentionally or accidentally become partially powered. A few of them are: a) Fault Tolerant/Fault Resilient Systems Fault-tolerant/fault-resilient systems often employ multiple redundancy. Critical system elements and functions are duplicated with the provision to transfer operation from one module or board to its "identical twin" if a fault is detected. The faulty board can then be removed from the rack for repair while the system is in operation, and replaced by a substitute board to maintain redundancy. The active data bus on the backplane must not be affected during the "live" insertion of the board. b) Power Conserving Systems Certain parts of a system may be powered off or operated at a lower Vcc to conserve power in redundant, backup or unused circuits. The maximum amount of power is conserved when the Vcc rail of the powered down section is at GND potential. Under these conditions, all signal pins of the powered down circuits must offer high impedance in order to prevent loading of the surrounding circuits. c) Attaching Peripherals It is a fairly common practice for people who are using computers and peripherals to connect the peripheral to Cb1992 Integrated Device Technology, Inc, 2546 drw 01 Figure 1. FCT16xxxT Input Stage SYSTEM DESIGN CONSIDERATIONS Input Considerations Figure 1 shows the equivalent schematic of the input stage of an IDT FCT16xxxT Double-Density device. A signal on the input that is between 0 volts and the Absolute Maximum Rating Voltage will see the gate to the MOS-FET transistors, the cathode of the clamp diode to ground, and the ESD protection circuit. There is no clamp diode from the input pin to Vcc and therefore the input will maintain a high impedance when the voltage on the input rises above the Vcc level. With 25 4/92 POWER-OFF DISABLE IN lOTS FCT16xxxT DOUBLE-DENSITY DEVICES APPLICATION NOTE AN-102 Situations to Be AVOided Vcc at any voltage level between 0 and the maximum rated voltage, the input voltage could be as high as the device's Absolute Maximum Rating without damaging the device or causing significant current drain into or out of the input. The connection of Vcc only without GN D should be avoided. Under this condition the devices will "look for" a ground connection on their inputs or outputs through the clamp diode and ESD protection circuit as the pins are connected. With all of the components on the powered down board using a common ground, the entire current draw of the board may pass into the component ground pin on a single device and out the grounded input or output of that component. If the driving component on the other side of the connection is a bus driver or similar device, currents high enough to cause device damage to the connected components may occur. Tying device inputs directly to Vcc and GND on a backplane or cable without series resistors may also produce a damaging current flow. This situation usually arises with board configuration pins on a backplane (e.g. for card slot identification, parity sense, pipeline register depth, etc.) All configuration pins should be current limited with series resistors if they are to tie directly to device inputs. This same situation is experienced if unused inputs are tied to Vcc or GND across a cable without series resistors. If the above situations are avoided while attaching Vcc first, there is still the probability that bus contention or bus interference will occur which will cause system disruption during the board installation or removal. Designers should make provisions to assure that during installation into, and removal of components from powered systems, the GND connection is made first. Some methods of guaranteeing that the GND pins will make first contact are as follows. The ground pins on the card edge connector can be made longer than the other pins so they make contact first. The ground pins can be placed towards both ends of the connector with Vcc near the center so that when the board goes in at a slight angle, the ground will make contact first. A ground wire can be connected between the board ground and the connection point ground prior to, and during insertion, to assure ground contact before Vcc. In the case of an electrically isolated connector, the connector shroud can be grounded. Output Considerations Figure 2 shows the equivalent circuit schematic of an output buffer with 3-state control as used in the FCT16xxxT family. The output stage contains a totem-pole structure which consists of N-Channel pull up/down transistors. These transistors have a clamp diode from their drains to GND, but have no clamp to Vcc and therefore no low impedance path to Vcc through a forward biased diode. EN OUT DATA~----------~ Figure 2. FCT16xxxT Output Buffer Combination Input/Output Port Considerations The I/O ports on IDT's FCT16xxxT High Drive and other families that support Power-off Disable contain combinations of the input and output structures shown in figures 1 and 2. These parts will handle powered installation/removal and other Power-off Disable functions with no problems. I/O ports on families that do not support Power-off Disable on the device outputs have a diode from the I/O port to Vcc causing the same restrictions as the device outputs for those families. Using Power-off Disable In the system design, the designer must disable the outputs on all drivers which interface the pOint at which connection is being made, through control of the output enable pins, to assure that the bus interface components are not driving the bus despite their low power condition. On devices that require a logic high on the output enable pin (EN) to disable the output, maintaining the output enable pins near Vec will assure that the output is disabled despite the voltage on Vcc approaching zero. As the Vcc level drops, the voltage required for a logic high will drop proportionately and the high impedance is maintained for power supply voltages from OV to Vccmax (maximum operating Supply Voltage). If properly done, the power down condition will be transparent to active boards on the system bus despite other active components in the system driving the bus to logic high conditions. CONCLUSION The need for Power-off Disable is becoming prevalent in many of the high reliability and power conserving systems currently being designed. To utilized Power-off Disable, certain design criteria must be followed. The first item is to assure that the ground pin is connected across the interface before the Vcc pin, the ground pin remains connected while Vcc is connected, and that Vcc is disconnected before the ground pin. The second item is to condition the output enables of the powered down component so that the device outputs will remain disabled. The third item is to assure that there are no inputs hardwired to GND or Vcc across the interface without series resistors. The IDT parts which support Poweroff disable will function in partially powered situations and will 26 ~ g APPLICATION NOTE AN-103 3.3 VOLT LOGIC Integrated Device Technology, Inc. By Stanley Hronik and Suren Kodical INTRODUCTION The demand for an increase in performance and portability of digital systems has created a need for integrated circuits of ever increasing performance and density with reduced power consumption. To meet this need, the semiconductor industry is moving rapidly towards a lower supply voltage standard based on 3 .3Vs nominal supply. This new standard allows the retention of TTL threshold levels while providing improved reliability, noise levels and a significant reduction in power consumption. The lower power consumption internal to the circuitry will allow component densities to reach new levels of integration utilizing fine line technologies. Eventually it is expected that the sophistication level of 3.3V circuitry will dwarf 5V technologies. Currently in the marketplace there are processors, memories and ASICs available which are either designed for or characterized for 3.3V operation. To support and accelerate the transition to the new 3.3V environment, Integrated Device Technology has developed families of 3.3V logic and bus interface circuits which will enable the designer to develop complete 3.3V systems. The 3.3V components now available are approaching the speeds of their 5V counterparts and in the near future will exceed those capabilities. These speed breakthroughs will be obtained by the further miniaturization of components as the lowering of breakdown levels allows the reduction in critical dimensions, path sizes, and device capacitances which will increase the device speed. Additional benefits include an increase in reliability due to lower power consumption and a reduction in noise generation due to lower voltage swings. Systems that will find use for 3.3V logic include battery operated portable systems such as notebook/laptop computers, hand-held electronic field instruments and portable communications gear. Space and military systems that require low power with high reliability should find the 3.3V operation an attractive alternative. Thermally sensitive systems such as any electronic system that operates in harsh environments or has extremely high density will benefit from 3.3V logic. Eventually the uses of 3.3V logic will exceed those of the 5V systems. These families are offered in a plastic Shrunk Small Outline Package (SSOP) for commercial applications and a CERPAK package for military applications. These packages offer significant space savings. This Application Note on IDT's 3 .3V families will discuss the following topics: • Input and Output structures • Performance characteristics • Power, noise and reliability issues • System considerations • Solutions for "mixed supply" operation KEY FEATURES OF THE 3.3V LOGIC FAMILY The FCT3xxx and FCT163xxx families of circuits are optimized for high performance 3.3V systems. Important features of these families include: • True TIL compatibility with a minimum of 400 mV noise margin. • Rail-to-rail output swing provides additional noise immunity. • Quiescent power supply current of 80 f.1A maximum. • Typical Ground Bounce is underO.3 Vforthe FCT163xxx family and under 1.0V for the FCT3xxx family. • Input and Output leakages guaranteed to be under 500 nanoamps. • Propagation delays match those of 5 V FCT logic for like functions. • All inputs (except I/O) can be driven from either 3.3V or 5V components. • Extended Commercial temperature range of -40°C to +85°C. • ESD >2000 V per MIL-STD-883, Method 3015. The low leakage and quescent currents offer a distinct advantage in power-critical systems through longer battery life and improved reliability. True TIL compatibility offers the system designer the freedom to mix 3.3V components and 5V components on the same board in most situations without sacrificing noise margin. By offering the same speed grades as the 5V logic for the same function, reduced power consumption is achieved without performance compromise. 3.3V LOGIC FAMILY FROM lOT 3.3V LOGIC FAMILY CHARACTERISTICS IDT is responding to the need for products that operate from a 3.3 ± 0.3 volt supply by introducing new logic families which have been named FCT3xxx and FCT163xxx. These families contain 8-bit and 16-bit bus interface functions, key "glue" logic elements and special interface components which solve interfacing problems in mixed 3.3V/5V supply systems. The IDT parts have all of the benefits inherent in 3.3V logiC as well as, excellent drive capability and performance characteristics. ©1993 Integrated Device Technology, Inc. DC Specifications AppendixA shows the complete set of DC specifications for the FCT3xxx and FCT163xxx families. These DC specifications meet or exceed the proposed JEDEC standard for 3.3V ± 0.3V TTL-compatible circuits. 27 4192 3.3 VOLT LOGIC APPLICATION NOTE AN-t03 Input Characteristics The input to the FCT3xxx and FCT163xxx families consists of an ESD protection circuit, a clamp diode to GND, and the gates of two MOSFET transistors as is shown in Figure 1. Vee HYSTERESIS ~ ESD-P-ROTECTION- - - - - I - AND INPUT CLAMP TO NEXT STAGE INPUT ---- ----i--- 2728 drw 01 Figure t. Input Stage of FCT 3.3V Components The Input protection circuit provides excellent immunity (>2000 volts) to the effects of electrostatic discharge (ESD). It also provides an effective clamp for input voltage undershoots by forward-biasing the p-njunction during input voltage excursions below device ground. Note that there is no similar clamp to vee. Therefore the input can withstand a maximum voltage rating of 7 volts, thus allowing the inputs to be driven by either3.3V circuits or 5V circuits without exceeding maximum ratings. Typical input V-I characteristic is shown in Figure 2. The input translator consists of two CMOS inverters with a feedback circuit which provides Hysteresis in the input transfer characteristic by means of a change in the ratio of Pchannel to N-channel transistor areas in the input translator. Typical transfer characteristic with 150 m V of hysteresis is shown in Figure 3. Hysteresis increases static noise immunity in both logic states and also offers immunity to noise superimposed on slow edge-rate input signals if the amplitude of the superimposed noise is less than the hysteresis margin. VOH 40 Ii (ma) 30 20 VOUT (VOLTS) Breakdown 10 VOL -20 Clamp -30 -40 1.40V 1.55V VIN (VOLTS) 2728 drw 02 2728 drw 03 Figure 3. Typical Input Hysteresis Characteristics Figure 2. Typical Input V-I Characteristics 28 ---- .......... -~ 3.3 VOLT LOGIC APPLICATION NOTE AN-103 Output Characteristics The outputs on the FCT163xxx and FCT3xxx parts switch rail to rail in order to provide a good noise margin and fu1l3.3V CMOS output capability at low current levels and TIL-compatible guaranteed output levels at high drive currents. In order to achieve the CMOS output levels there is a P channel 'source' between the device output and Vcc acting as a pullup on the output. Inherent in this transistor is a parasitic clamp diode between the output and Vcc which will drain excess current if the voltage on the output rises a diode drop level above the Vcc rail. Similarly, there is a parasitic diode to GND associated with the N channel 'sink' transistor. Because of the presence of the clamp diode to Vcc there is no ability to shut off the power on a component that has an output directly tied to an active data bus. In other words, the FCT 3.3V components have no power-down disable capability on device outputs. Also, the 3.3V output cannot be readily connected to a 5V output on a bus because of the possibility of drawing large amounts of current through the parasitic diode under certain conditions. When connecting a 3.3V output to a 5V bus or output, the user needs to utilize the techniques for overcoming this problem as described later in this document or use the 5V to 3.3V interface components which lOT has developed to solve this problem. The output structure of a typical FCT 3.3V logic circuit is shown in Figure 4. Along with this the typical Output V-I characteristic for a high-impedance (Hi-Z) state @ Vcc = 3.3Vs is shown in Figure 5. Vee Figure 6 shows the output drive characteristics of the FCT 3.3V families for a logiC LOW state. The low output impedance allows incident wave switching when driving transmission lines of >45n (typ.) characteristic impedance. The output also offers excellent dynamic current capability to drive large capacitive loads without significant speed degradation. The typical propagation delay derating is 1ns per 50pF of output load. 140 120 100 lOUT 80 (ma) 60 Logic Low State 40 20 1.0 2.0 4.0 VOUT(V) 2728 drw 06 Figure 6. Typical Output Low Characteristics o o 1.0 2.0 3.0 4.0 VOUT (V) -20 -40 -60 lOUT (ma) -80 -100 -120 -140 Logic High State 2728 drw 07 Figure 7. Typical Output High Characteristics 2728 drw 04 Figure 4. Output Structure of the FCT 3.3V Logic Families Figure 7 shows the output drive characteristics for a logic HIGH state. The V-I characteristic is somewhat linear due to the parallel combination of the P channel 'source' transistor and the N channel transistor shown in Figure 4. Again, the low output impedance in the logic HIGH state allows incident wave switching when driving transmission lines of >35n (typ) characteristic impedance. The output 'pull-up' circuit also offers excellent dynamic current drive to charge large capacitive loads without significant speed degradation. Typical propagation delay derating is 1 ns per 50pF for the low to high transition. 8 Clamp 6 loz (ma) 4 2 2 Clamp 4 6 8 2 Voz (VOLTS) 4 Vcc = 3.3 volts POWER SPECIFICATIONS There are three components of power supply current in CMOS circuits. The total power supply current is given by: 2728 drw 05 Figure 5. Typical Three-State Output V-I Characteristics Ic 29 = IQUIESCENT + IINPUTS + IDYNAMIC 3.3 VOLT LOGIC APPLICATION NOTE AN-103 The first one is the Quiescent Power Supply current (Icc). This is a static current through the power pins of the device when all inputs are held at either supply rails. This current is typically in the nanoampere range. The second component of supply current is the sum of the static component through the input stages (8Icc) when the input voltage is between VCC and GND. Under this condition both N channel and P channel transistors of the input translator shown in Figure 1 can conduct current. The magnitude of this current is dependent on the voltage on the input pin. Typical input translator current 81cc as a function of input voltage is shown in Figure 8. 1.0 (nom) represents a 56% reduction in the dynamic power dissipation. The equation for dynamic power dissipation can be rewritten as follows: ICCD= fCV IcCD is expressed in mA or ~ per MHz of switching frequency which will be constantfor a set voltage and capacitance. In bus interface circuits with several identical paths, it is customary to specify ICCD per bit. Over a wide frequency range, IcCD is a linear function of frequency. Figure 9 shows ICCD for the FCT163245 16-bit transceiver in comparison with other popular 5V 16-bit transceivers. 1.3V 25 ACT a15.5V ACT a15.0V ""Icc O.B Per Input (ma) 0.6 20 0.4 Icclbit 15 (ma) 0.2 ABT at S.SV 10 FCT-T a15.5V FCT-Tal 5.0V FCT3.3V 5 Figure 8. Typical alcc characteristics 20 40 60 BO 100 Frequency MHz The 81cc component is specified in the data sheets for the FCT3xxx and FCT163xxx families at VIN = 2.4V and at VIN = VCC - 0.6V. The first test condition where VIN = 2.4V is useful when a worst-case power supply mismatch is expected between two circuits interfacing with each other. When the circuits are operating from substantially the same supply voltage, the condition of VIN = Vcc - 0.6V is more realistic. A detailed explanation of the use of the power supply components is provided in the data sheets forthe 3.3V logic products. The 81cc component can be significant when a 5V circuit is driven from a 3.3V circuit, particularly if the 3.3V supply is at its minimum of 3.0 volts and the 5V supply is at 5.5Vs. Under these conditions, both the P channel and the N channel transistors olthe input translator are operating in the saturation region. The third component of power supply cu rrent is the Dynam ic Power Supply Current or ICCD. This component represents the power consumed by the device in charging and discharging the internal node capacitances of the circuit through the transistors. Since each internal node switches between the supply rails, the power consumption is given by the formula: P = Vicco = fCV2 where f = equivalent frequency of the voltage transitions, C is the net capacitance of the circuit ( sum of the switching node capacitances) and V is the supply voltage. Notice the squared term in the equation. This equation shows the benefits of a lower operating voltage since a drop from 5Vs (nom) to 3.3Vs 2728 drw 09 Figure 9. ICCD vs Frequency for the 16245 A comparison of dynamic power supply current per MHz per bit for various 16-bit transceivers is shown in Table 1. Note that this current represents the power consumption in the device with the outputs unloaded. Device IceD/MHz/Bit Vee 240!JA 5.5V ABT16245 90!JA 5.5V FCT16245T 4B!JA 5.0V FCT163245 27!JA 3.3V ACT16245 2728 tbl 01 Table 1. Dynamic Power Supply Current Comparison Power Consumption Due To Device Output Loading From the preceding discussion on power supply current components, one can see that the dynamic current will become dominant at higher frequencies. When the outputs of a circuit drive capacitive loads, the dynamic power supply current component due to the load capacitance increases the device power dissipation. This component is given by the equation: 30 3.3 VOLT LOGIC APPLICATION NOTE AN-l03 PL = fCLV2 Since the voltage swing on the output pin is the same as that on the internal nodes, the total power dissipation of the device with loaded outputs is: PTOTAL = f(CPD + CL)V2 where Cpo represents an equivalent total capacitance for internal device nodes, CPO can be obtained from the following equation: CPD = IceD I V (~MHz). RELIABILITY The 3.3V family will inherently dissipate less power and run at cooler temperatures than corresponding 5V parts. The reduced supply voltage level means that internal dielectrics are not stressed at the levels they would be in higher voltage applications and therefore device breakdowns will be less prevalent than in higher voltage families. The lOT 3.3V components retain the 7V breakdown voltage on device inputs, equivalent to that of 5V parts, giving a wide margin of protection against potentially damaging input voltage levels. The devices also have ESO protection on the inputs and outputs helping to avoid handling problems. In addition to the device reliability, system reliability is improved by using 3.3V components which generate less system noise, reduce device power dissipation and possibly reduce system size and cooling requirements. NOISE CONSIDERATIONS Accompanying the increase in system speed and bus widths, there is a concern about the increase in system noise. Several types of device-generated noise are associated with high speed integrated circuits. The most commonly recognized noise component is the Simultaneous Switching noise, often referred to as Ground Bounce. This transient noise is a result of voltage developed across the parasitic inductance associated with the ground return path of the circuit during simultaneous HIGH-to-LOW switching of several outputs. At its worst, ground bounce can cause false switching and data integrity problems in storage elements such as latches and registers. Since the ground lead inductance plays a major role in the amount of ground bounce generated, Octal bus interface circuits with a single ground generate more noise than 16-bit bus interface circuits with 8 ground return paths. The 3.3V logic families from lOT feature edge-rate control to contain ground bounce to 1.0V typical in the octal interface components and to O.3V typical in the 16-bit interface components. These noise levels are considerably lower than those in 5V circuits of the same performance characteristics. lOT's Application note AN-47 offers a detailed discussion on this subject. The second component of noise is that associated with output edge rates. For fast edge rates a PCB trace behaves like a transmission line. Signal overshoots and undershoots are generated when driving such traces without termination 31 which matches the characteristic impedance of the traces. Under the worst-case scenario, these transients can degrade system reliability. The FCT3xxx and FCT163xxx families are designed to minimize these noise components. For a more complete treatment on this subject, the reader is referred to lOT Application note AN-49 'Characteristics of PCB Traces.' Lastly, by avoiding abrupt voltage changes during the output voltage transitions, transmission of EMI and RFI noise is minimized. This helps with the board layout by reducing the need for special considerations in component placement, decoupling, line termination, and shielding and leads to quicker, easier designs that meet FCC guidelines. Noise Immunity To improve immunity to system noise, the FCT3xxx and FCT163xxx families of components have been optimized for an inputtrip point at 1.4 Vtypical, at the center of the guaranteed input voltage range. Addition of the hysteresis feature further increases immunity to noise on signal lines. Since the outputs are designed to swing rail to rail, any 3.3V component interfacing with the 3.3V logic from lOT will experience the benefit of an additional noise margin. PERFORMANCE The device performance for the FCT3xxx and FCT163xxx families is the same as for the 5V FCT16xxxT family and FCT xxxT families. These speeds can be seen in the lOT Logic Data Book. Propagation delay limits of key parameters for commonly used functions in the FCT 3.3V families are compared with industry-accepted 74AC series devices operating at 3.3Vs. See Table 2. Function Path 74FCT163xxxA 74ACxxx 245 Input to Output 5.2n5 12.4n5 373 Input to Output 5.2ns 15.0ns 374 Clock to Output 6.5ns 18.0ns 2728 tbl 02 Table 2. Speed Comparison of FCT 3.3V with AC at 3.3Vs INTERFACING 3.3V COMPONENTS TO 5V FAMILIES During the transition from 5V to 3.3V, there will be a need to interface between 3.3V parts and 5V parts in a system or between a 3.3V system and a 5V peripheral. When operating 3.3V parts and 5V parts in the same system, the designer needs to be aware of the compatibility issues between the families and provide protection against component damage in some cases where excess current or voltage may develop. The 3.3V family was designed to be TTL-compatible with 3.3V or 5V components, but there are many situations inherent in the interface that will cause problems if the interface is not properly constructed. 3.3 VOLT LOGIC APPLICATION NOTE AN-103 3.3 Volt 5 Volt 5 Volt TIL TIL 7.0 3.3 Volt TIL TIL 7.0 Input and Output Breakdown 7.0 Volts 6.0 Input Breakdown 7.0 Volts 6.0 VOH 5.5 Volts Max. VOHMax. 3.6 Volts 5.0 5.0 4.0 4.0 3.0 3.0 VOHMin. 2.4 Volts VOH 2.4 Volts Min. 2.0 2.0 VIH 2.0 Volts 1.0 1.0 VIL 0.8 Volts 0 GND VIH 2.0 Volts VIL 0.8 Volts VOL 0.5 Volts VOL 0.4 Volts GND Output Breakdown Vcc+ 0.3V GND 2728 drw 10 0 GND 2728 drw 11 Figure 10. 3.3 Volt Device Driving a 5 Volt System Figure 11. 5 Volt Device Driving a 3.3 Volt Device 3.3V Logic Driving a 5V System As described earlier in this application note, the FCT 3.3V logic components have CMOS outputs which offer rail to rail output swing and guarantee TTL compatible output logic levels at high drive currents. Because of this, there is no problem with 3.3V logic driving standard 5V TTL logic. The logic input thresholds on the 5V TTL parts are VIL = O.BV, VIH = 2.0V. The 3.3V logic is guaranteed to drive a LOW to less than O.4V and to drive a HIGH to greater than 2.4V at rated output load currents. This provides very good driving voltages forthe 5Vinputswith noise immunity making 3.3V outputs fully compatible with 5V TTL inputs. Despite the logic compatibility of the 3.3V outputs with the 5V inputs, the designer needs to be aware of the slight increase in power consumption that may occur when interfacing 3.3V parts to 5V parts as discussed in the section on power consumption. True 5V CMOS devices that utilize CMOS thresholds (not TTL) require higher input voltages to reach the guaranteed minimum logic high. (The 3.3V logic is not guaranteed to reach the output voltage levels necessary to drive CMOS level inputs on true CMOS devices operating at a 5V supply level.) rating on the 3.3V parts. Problems may develop from the output clamp diode when 5V TTL outputs are connected to 3.3V outputs or I/O ports on a bus. The logic HIGH on 3.3V device outputs and I/O ports is limited to Vcc + O.5V. Driving a 3.3V liD port directly from a 5V part may exceed this absolute maximum rating and damage the 3.3V device. Other mixed supply situations include interfaCing a 3.3V module or sub-system with a 5V system and interfacing a 3.3V system such as a laptop/notebook computer with an external 5V peripheral such as a printer. The 3.3V parts have no power down disable on the device outputs and therefore they cannot be unpowered and receiving voltage on their outputs without sustaining damage. It is likely in a dual power supply system that one supply will become active or shut down prior to the other. To avoid damage to the 3.3V part, the current from the 5V outputs to the 3.3V outputs must be limited. Members of the 5V FCT16xxxT and FCT6xxT family do have power-down disable capability and therefore will sustain no damage in the partially powered situation regardless of whether only the 5V supply is active or only the 3.3V supply is active. To avoid excessive current flow from an active 5V output into a 3.3V output, the user may place a resistor between each 5Voutput and the corresponding 3.3V output or I/O. When the 3.3V 110 is in a high-impedance state, a current will flow from the 5V part to the 3.3V part through the resistor and clamp diode causing a voltage drop between the devices and preventing damage. When calculating the value of the resistor, the designer should consider the worst case situation of the 5V 5V Logic Driving a 3.3V System As stated earlier, 5V logic will drive 3.3V inputs (not I/O ports or three-state outputs) directly and will meet all voltage levels and input requirements of the 3.3V logic without component damage on either side while retaining full functionality. The 5V TTL component may output up to 5.5Vs worst case which will not exceed the 7V absolute maximum input voltage 32 3.3 VOLT LOGIC APPLICATION NOTE AN-103 supply at the maximum operating rated voltage and the 3.3V supply at the minimum operating voltage. If the 3.3V system may power down when the 5V system is active the calculation should consider the 3.3V supply at zero. See Figure 12. +5VDC or +3.3VDC Output Stage of a 5 volt 74FCT621T I R .....-_.....-'--''--_ _ _ To any 3.3Vor 5V Input or 1/0 -l Vcc = a volts -----------, 500 ohm 5.5 volts R must be sufficiently large to current limit the output to less than 20 ma worst case if R is tied to +5V DC. 2728 drw 13 Figure 13. Using Ope Drain Components as a 3.3V Interface 5 Volt Part Series Resistor I -----------.1 I 3.3 Volt Part ._--------2728 drw 12 Using Open-Drain Components as an Interface Figure 12. A 5V Part Driving a 3.3V 1/0 Port Placing a series resistor between 5V and 3.3V parts will cause some speed degradation due to the RC time constant that will develop between the series resistor and the input capacitance of the 3.3V part and other parts that may be similarly attached. The designer must be aware of the following situation when connecting 5V parts to 3.3V parts with a series resistor. If the 3.3V power supply is off and the 5V power supply is active, current may flow from the 5V output into the 3.3V output and through the clamp diode to the 3.3V power supply plane. When this happens, the voltage on the 3.3V power plane will rise if there is no low impedance path from the power plane to ground. Undercertain conditions, it may rise sufficiently to cause unstable operation of the 3.3V parts or failure of the power on reset when the 3.3V supply is activated. This condition may be avoided by shorting the 3 .3V power plane to ground when the power is off. This could be done through the use of a pull down transistor in the power on reset circuit. Another solution is to use the 3.3V to 5V interface components developed by lOT which avoid this and other problems. If the user can guarantee power sequencing and tracking between the 3.3V and 5V power supplies, the value of the current-limiting resistor between the 3.3V inputs and 5V outputs can be significantly reduced. 33 Another means of interfacing 5V outputs to 3.3V lID ports is to use the FCT621 T and FCT622T 5V open drain components as interface devices. If the pull-up resistor from the open-drain output is returned to the 3.3V supply as shown in Figure 13, the voltage on the bus will never exceed the 3.3V supply voltage and will assure safe operation. If the series resistor is tied from the device output to the 5V power supply instead of the 3.3V power supply, the output current would be limited by the resistor and also provide safe operation. 5V TO 3.3V INTERFACE COMPONENTS To combat the problems associated with mixed supply operation, any of IDT's 3.3V unidirectional parts can be used as translators when connected as described earlier. This would include the FCT163244, FCT163373, FCT163374 and others. When connecting 3.3V outputs to a 5V bus, it is necessary to use special interface components which will handle the higher voltages on the device outputs. One of the devices that IDT has developed for this purpose is the FCT164245Twhich will directly tie a bidirectional 3.3V bus to a bidirectional 5V bus. This part is also useful in situations where there is no power sequencing or where one of the power supplies may remain off while the other is active. 3.3 VOLT LOGIC APPLICATION NOTE AN-103 DEVICE CONFIGURATION AND NOMENCLATURE CONCLUSION Using the FCT3xxx and FCT163xxx families of logic parts in 3.3V, high speed designs will help the designer overcome problems with noise and power consumption, and improve reliability. With the high density and lower power consumption, the user may be able to reduce his overall system size and cost without sacrificing performance. Prior to the availability of a complete 3.3V family, there are techniques for interfacing the 3.3V parts with 5V systems including special translator functions for mixed supply environment. The FCT3xxx and FCT163xxx families of devices are ideal for transmission line driving, point-to-point driving, bus interface and memory interface applications. A1I3.3V parts are pin compatible with the industry standard for their 5V counterparts. This includes the FCT163xxx parts that are configured as 16 bit wide DOUBLE-DENSITY parts for wide bus widths and the FCT3xxx octal and glue parts. The DOUBLE-DENSITY device types within the family retain the familiar naming conventions of the eight bit families (e.g. 245, 373, 374 etc.) but are arranged with two octal functions with independent control per package giving the devices dual eight bit capability or full sixteen bit operation. An example of the naming convention would be an FCT163245 which is a sixteen bit 3.3V device, with an FCT245T type function. The pins on the DOUBLE-DENSITY devices are configured so that all power and ground pins are in identical locations for all functions. In addition data inputs have a flowthru architecture for ease of board layout. APPENDIX A DC ELECTRICAL CHARACTERISTICS OVER OPERATING RANGE Symbol VIH Min. Tvp. Max. Unit Guaranteed Logic HIGH Level 2.0 5.5 V 2.0 -0.5 - Vcc +0.5 O.S V VI = 5.5V - - ±0.5 J.LA VI = Vcc - +0.5 VI =GND - - Parameter Input HIGH Level (Input pins) Test Conditions Input HIGH Level (I/O pins) VIL Input Low Level Guaranteed Logic LOW Level IIH Input HIGH Current (Input pins) Vcc= Max. Input HIGH Current (I/O pins) IlL Input LOW Current lozH High Impedance Output Current IOZL (3·State Output pins) VIK IODH Clamp Diode Voltage Vec=Min.,IIN=-1SmA Output HIGH Current Vce = 3.3V, VIN = VIH or VIL, Vo = 1.5V IODL Output LOW Current Vce = 3.3V, VIN = VIH or VIL, Vo = 1.5V VOH Output HIGH Voltage Vce= Min. IOH = -o.1mA VIN = VIH or VIL IOH = -6mA Mil. 10H = -8mA Com'l. Vce = Min. VIN = VIH or VIL IOL = 0.1mA IOL = 16mA IOL = 24mA VOL Output LOW Voltage los Short Circuit Current VH Input Hysteresis leeL lecH lecz Quiescent Power Supply Current Vce= Max. Vo=VCC Vo=GND Vec = Max., Vo = GND Vee = Max., VIN = GND or Vee ±D.5 +0.5 J.LA ±D.5 -0.7 -60 -1.2 V - mA 90 - mA Vee-0.2 - 2.4 3.0 V - - - - 0.2 0.3 - 135 - mA 150 - 0.02 SO J.LA 0.2 0.4 0.5 V mV 27281b103 34 ~ :J ~ APPLICATION NOTE AN-116 DECOUPLING DOUBLE-DENSITYTM COMPONENTS Integrated Device Technology,lnc. By Stanley Hronik INTRODUCTION of simultaneous switching noise. This also reduces the effects of inductance in the board metalization by decentralizing the current path. When laying out a circuit board for use with DoubleDensity, the designer should use full ground and power planes with all ground and power pins on all devices connected to the proper plane. No power or ground pins on any Double-Density device should be left floating. The board should have a large capacitor near the power entry point on the board to stabilize any power surges from the power distribution system. If the board has an effective power distribution system, decoupling Double-Density should be easier than decoupling corner Vcc and ground packages. Double-Density is IDT's FCT-T family of 16, 18, and 20 bit bus interface components. The family is functionally compatible with the Wide Bus product line, but offers users power savings, higher speeds, and excellent guaranteed low noise operation with a choice of output drive characteristics. The components are available in a 64ma drive version for use in backplane driving where line termination exists, a balanced drive 24ma version with internal series line termination for quiet operation, and a 3.3V version for use in applications requiring the lower supply voltage. Although IDT's Double-Density family is more forgiving than the older eight bit families, the decoupling must be properly executed to achieve optimum results. The three versions of the Double-Density components (64ma, 24ma, and 3.3V) have similar switching speeds and therefore have similar decoupling needs despite having different device parameters and output specifications. Because of this, the techniques for decoupling which are addressed herein apply uniformly across all versions of the Double-Density family. When setting goals for decoupling a circuit, a designer should focus his attention on reducing the radiated emissions to meet the FCC limitations for the geographical area in which his circuit will be used. When this has been accomplished, the low noise levels needed to prevent cross talk and false switching will probably have been achieved. If the circuit contains low-level analog signals, additional decoupling will probably be necessary. GND Vee GND GND 3 4 5 6 7 8 9 10 11 12 8048-1 13 14 15 16 17 SELECTING THE CAPACITOR PACKAGE Vee The effectiveness of decoupling capacitors is often degraded by the series resistance and inductance inherent in the capacitor. The use of chip capacitors reduces the inductance due to the capacitor leads and through hole placement, making chip capacitors the optimum choice where that package style can be used. If the use of chip capacitors is not possible, the package should allow the capacitor to fit close to the board with very short lead lengths. With through hole capacitors the excess lead should be trimmed as close to the board surface as possible. GND 18 19 20 21 22 23 24 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32 31 30 29 28 27 26 25 GND Vee GND GND Vee GND 3011 drw01 Figure 1. A 4811in Double Density package Capacitors have a circle of influence around them meaning that pins and components that are in the immediate proximity of the capacitor will be effectively decoupled while those components that are further away will be less affected. This characteristic is modeled in Figure 2. While the number of capacitors per component will vary depending upon the decoupling needs of the circuit and the cost sensitivity of the design, it is suggested that the designer use at least one ceramic capacitor per IC on the board. In cost sensitive applications it may be possible to reduce the overall number of capacitors to less than one per component. In noise PC BOARD POWER AND GROUND PLANE IDT's Double-Density devices are packaged in a 48- or 56pin SSOP with multiple power and ground pins as shown in Figure 1. Unlike older logic families with corner Vcc and ground pins, Double-Density packages have eight ground pins and 4 Vcc pins which are equally spaced on both sides of the package. By providing multiple, short, parallel paths for current, the lead inductance is lowered, reducing the effects The lOT logo is a registered trademark and Double-Density is a trademark of Integrated Device Technology, Inc, ©1992 Integrated Device Technology, Inc. 35 DECOUPLING DOUBLE·DENSITY COMPONENTS sensitive circuits (e.g. ND applications) more than one capacitor type per component may be necessary. Tantalum capacitors work well in high frequency analog situations which already include a ceramic capacitor. If the board has components on only oneside, the capacitor can be placed at either end of the device (considering the direction to the power source) with a solid connection to the power and ground planes as shown in Figure 4. This will give I I 3011 drw 04 Figure 4. Capacitor macement for boards with one component side the board designer direct access to the sides of the package to run traces and avoid placement interference. If more than one capacitor is used, the capacitors can be placed at either end of the package or if that causes placement conflicts the capacitors can be located at adjacent corners of the package as shown in Figure 5. Figure 2. All capacitors have a "circle of influence" around them CAPACITOR PLACEMENT RELATIVE TO THE COMPONENT + The decoupling capacitor should be placed as close as possible to the component being decoupled. It is not necessary to provide a separate capacitor for each Vcc on the device, but the decoupling capacitor should be solidly connected to both planes providing a short conductive path to all power and ground pins on the device. Relative to the component, the optimum placement would be on the reverse side of the board, centered over the device as shown in Figure 3. The power and ground pins on the SSOP package that are adjacent to the device outputs will generate more switching noise than pins near the device inputs or enable pins. Since these pins are usually near the center of the package, placing the decoupling capacitor on the reverse side of the board centered over the component will provide the shortest path from the capacitor to the pins giving optimum decoupling. + 3011 drw 05 Figure 5. Placement of dual capacitors on the component side for special applications POSITION RELATIVE TO POWER ENTRY The decoupling capacitor should be placed between the power entry point on the board and the component that is being decoupled as is shown in Figure 6. Since the ultimate source of all charge is the power entry point, placing the capacitor between the component and the entry point positions the capacitor to receive and stabilize the voltage prior to the component. If the capacitor and component are reversed, the charge will flow directly from the entry point to the component and not pass the capacitor. This will make it easier for noise in the power system to reach the component, and for noise generated by the component to propagate into the 3011 drw 03 Figure 3. Capacitor placementfor a board with components on both sides 36 DECOUPLING DOUBLE-DENSITY COMPONENTS power system. When the capacitor is not in the path of the charge flow, the capacitor can only react to noise after it has passed into the component or power system rather than prevent it. I ToVCC model can be considered as a single capacitive element with no inductance or resistance. As the noise frequency increases to levels above 50MHz the influence of the inductance and resistance increases to the point that the inductance becomes the dominating component of the capacitor's impedance. If noise issues are critical, the designer may be required to use two capacitors to cover both the high and low frequency components of the spectrum. Capacitors that are operating near their characteristic frequency will resonate at that frequency and provide good noise filtration with no inductive kick. Using this characteristic, designers frequently use a 0.01 uf or a 0.001 uf capacitor to filter out high frequency noise. The characteristic frequency of the capacitor can be obtained by contacting the capacitor manufacturer. Typically ceramic capacitors have a characteristic frequency in the 20-40MHz range. Tantalum capacitors typically work best at frequencies above BOMHz and have decreasing effectiveness at lower frequencies. While picking capacitors at the resonant frequency of the circuit will prove beneficial to noise suppression, the selection of smaller valued capacitors will not. The value of the capacitor should always be at the resonant frequency or have a larger capacitance value. If small valued capacitors are used, it will be necessary to have good low frequency decoupling through additional larger valued capacitors located near the component. 3011 drw 06 Figure 6. The decoupling capacitor works best if placed between the power source and the pin being decoupled FREQUENCY CONSIDERATIONS USING CAPACITORS WITH DIFFERING INTERNAL STRUCTURES lOT's Double-Density components have very high internal switching speeds, but the lead inductance, packaging, and board placement will absorb most of the high frequency components above 120MHz before the noise can enter the power distribution system or radiate. The frequencies most likely to cause problems by reaching the power distribution system are in the 60-BOMHz range (40-120MHz on a broader scale) and therefore the decoupling effort should be focused on this range. Capacitors can be modeled as a capacitor in series with an inductor and a resistor as shown in Figure 7. With typical ceramic capacitors at low noise frequencies «10MHz), the When using capacitors of different values and different types on the same board in close proximity to one another, it is possible for the two capacitors to begin interacting with one another. With different frequency responses, the two capacitors will respond with different timings to voltage spikes with CAPACITOR 1 SWITCHING' TRANSIENT Jl GROUND LEAD -,---JI r - - - - - - - - - - - - -. NOISE 1 SPIKE R L 1 CAPACITOR 2 1 :j: C 3011 drw 08 1 _____________ I L 1 Figure 8. If using two capacitors with different characteristics, oscillations may develop between them 1 ,----------------~ 3011 drw 07 the result being an instantaneous difference in the internal voltages between the two capacitors. The two capacitors can then begin oscillating by passing charge between themselves through their characteristic series inductors as modeled in Figure B. The end result will be a reduction in the effective decoupling that takes place and the possible addition of a noise source on the board. Figure 7. Model of the Influence of a noise source and a decoupling capacitor on a node 37 DECOUPLING DOUBLE-DENSITY COMPONENTS CONCLUSIONS Additionally, two capacitors with highly different dielectric constants in close proximity on the same board may affect each other because the high dielectric constant part will dampen the high frequency resonance of the low dielectric constant capacitor and the end result will be reduced noise suppression. To effectively handle the problem of capacitors with differing internal structures, be sure the two types are not placed in close proximity to one another as shown in Figure 9. This can be done by placing the different capacitors at opposite ends + To properly decouple lOT's Double-Density parts do the following: 1) Select a capacitor that will meet the needs of the component. A 0.1 uf ceramic is a good starting point. A 0.01 uf ceramic may be better in situations which have good low frequency decoupling elsewhere on the board but are experiencing high frequency noise. 2) For best results use at least one capacitor per IC. More may be necessary in noisy situations, fewer may be possible with careful component and capacitor placement. 3) Place the capacitor as close to the component as possible. On the reverse side of the board is best. At either end of the component is also good. 4) Make sure that there is a good connection between all ground pins on the component and the capacitor. This should be in the form of a solid ground plane under the device. The same is true for the Vcc pins. 5) Place the capacitor between the component and the power entry point on the board if possible. 6) In critical situations two or more capacitors of different values may be used. In this case select a larger capacitor to provide low frequency stability and a smaller capacitor to give a series resonance to higher frequencies. Examples may be a 0.1 uf or a 1uf ceramic along with a 0.001 uf or a 0.01 ufo 7) If caps of different internal structures are used, they should not be placed in close proximity to one another. Capacitors of different types that are close to one another may resonate between themselves and negate any positive effect of using two capacitors. + &SSSSSj 3011 drw 09 Figure 9. Exercise care when positioning two capacitors with differing frequency response in close proximity to one another of the device or equally spaced on the backside of the board. If these choices are notfeasible, the capacitors can be spaced in opposing corners of the device. The key to placement is to avoid having the Vcc pins of two capacitors adjacent or having the Gnd pins of two capacitors adjacent. 38 APPLICATION NOTE AN-118 MAXIMUM FREQUENCY CONSIDERATIONS FOR THE IDT49FCT80S/6 Integrated Device Technology, Inc. By Anupama Hegde INTRODUCTION Power dissipation in an unloaded CMOS device can be calculated using the following equation: An important factor in determining maximum operating frequency of a device is total power dissipated in the device. In CMOS circuits, the total power dissipation is dominated by the frequency-dependant dynamic component, since the static component is typically very small. The maximum permissible power dissipation is determined by the thermal characteristics of the package and the environment, which, in turn, sets the upper limit forthe operating frequency. Although this is true for all CMOS circuits, the implications are particularly important in the case of clock drivers, where multiple outputs are continuously switching at the system clock rate and are driving significant capacitive loads. Factors that influence the maximum operating frequency of a device based on power dissipation include: • Maximum allowable junction temperature for reliable operation • Junction-to-ambient thermal resistance of the device • Ambient temperature • Inherent device power dissipation characteristics • Number of TTL level inputs • Number of outputs switching • Output loading The following equations demonstrate how these factors determine operating frequency limits. Po (unloaded) = (ICCQ + ~Icc Nr Dr + Nof Icco)Vcc 1 1 static power 1 1_'1--' ... dynamic power ...... (iia) where, f = output frequency ICCQ = static or quiescent current Nr = Number of TTL level inputs ~Icc = static current due to TTL level inputs Dr = Duty cycle of TTL level inputs Icco = dynamic power supply current per MHz per bit No = Number of output bits switching at frequency f Some manufacturers express the dynamic power dissipation component in terms of another parameter - Cpo, rather than Iceo, as shown above. Cpo = Iceo/Vcc = equivalent device power dissipation capacitance Equation (ii) may be expressed in terms of Cpo as follows: Po (unloaded) = (lcCQ + ~Icc Nr Dr +NofCpoVcc)Vcc .......... (iib) MAXIMUM FREQUENCY CALCULATIONS In order to determine the maximum operating frequency, one must first calculate the maximum allowable power dissipation based on thermal limitations and maximum allowable junction temperature for the device. Starting with the thermal equation; 9JA = (TJ - TA)/PO the maximum allowable power dissipation for a device is given by: POMAX= (TJMAX - TA)/9JA .......... (i) where, 9JA = Junction to Ambient thermal resistance of the device TJ = Junction Temperature TJMAX = Maximum allowable TJ (TJMAX = 150°C for IDT logic devices) TA = Ambient Temperature Po = Total Power Dissipation POMAX = Maximum allowable Power Dissipation based on TJMAX Device loading can dramatically alter the dynamic power dissipation. Additional power is dissipated due to switching of the load. Consequently, equations (iia) and (iib). change as follows with the addition of a load: Po (loaded) = Po (unloaded) + NofCLVOH2 = (Iceo + ~Icc Nr Dr + ND f Icco) Vcc + NofCLVOH 2 1 1 1 static Ipower I 1 dynamic power .......... (iiia) = (ICCQ + ~Icc Nr Dr + NofCpoVcc) Vcc + NofCLVOH 2 .......... (iiib) where, CL = Load Capacitance VOH = Logic High Output Voltage level (which is the voltage across the load) The IDT logo is a registered trademark of Integrated Device Technology, Inc. 2912 ©1993 Integrated Device Technology, Inc. 39 12/9' MAXIMUM FREQUENCY CONSIOERATIONS FOR THE IOT49FCT805/6 APPLICATION NOTE AN-118 Since CMOS devices have a negligible static power dissipation component, and VOH equals VeeforCMOSoutputswing FCT devices, equation (iiib) is sometimes approximated as; Assumed conditions: TA = 25°C Vee = 5V 11 bits switching CL = 40pF (on each output) No airflow/cooling } Package = 20-pin SOIC } ==> (9JA = 90°C /Watt) f = operating frequency in MHz TIL level inputs (for CMOS level input signals ~Iee = 0) PD (loaded) = PD (unloaded) x [1 + CllCPD] .......... (iiic) From eqn(iiia), PDMAX = (leeQ + ~Iee NT OT + NDfMAX IceD) Vee + NDfMAXCLVOH2 .......... (iv) This gives: PDMAX =(TJMAX - TA) / 9JA = (150- 25) / 90 ... using equation (i) = 1.38W Substituting the PDMAX value obtained from eqn(i) in equation (iv) gives: Power dissipation of the device, at frequency f, with no load is: (TJMAX- TA)/9JA = (leeQ+~lee NTOT + NDfMAX IceD) Vcc + NDfMAXCLVOH 2 PD (unloaded) = (1.5+2.5x 2 x 0.5+0.2 x f x 11)mA x Vee V ... using equation (ii) = (4 + 2.2f) mA x Vee V = 5(4 + 2.2f) x 10-3 W For a given set of operating conditions, the only variable in this equation is the frequency. So, transposing all other factors to the righthand side gives the operating frequency limit. With the assumed 40pF loading on each output: = fMAX {(TJMAX-TA}/9JA {NolccoVcc + NOCLVOH2} (lcCQ+~lccNTDT}VCC}/ PD (loaded) = PD (unloaded) + [11 f x 40 X 52 X 10-6]W ... using equation (iii-a) = PD (unloaded) + [111 x 10.3] W .......... (v) The application of these equations is illustrated in the following example for the lOT 49FCT805/6 clock drivers. Using eqn (iv) gives: 1.38W = PD (unloaded) + 11fMAX x 10.3 1.38W = 5(4 + 2.2fMAX) x 10-3 + 11 fMAX x 10-3 1.38W = 5(4 + 4.4fMAX) x 10-3 => fMAX = 62.2MHz This is the figure obtained assuming TTL level inputs. With CMOS level inputs, since ~Iee = 0, fMAxgoes up to 62.7MHz. Similar calculations can be done to obtain the operating frequency limits for various other cases. The table below gives fMAX values for these different cases. Example (for the 49FCT80S/6) 49FCT805/6 specifications: leeQ = 1.5mA ~Iee =2.5mA IceD = 0.2mAlMHzloutput #of Bits TA 11 25°C 6 70°C 11 25°C 6 70°C 11 25°C 11 70°C NOTES: 1. LFM = Linear Feet per Minute. 2. Graphs for 9JA are given in Section Load 40pF 40pF 40pF 40pF 40pF 40pF Package 20-pin SOIC 20-pin SOIC 20-pin SSOP 20-pin SSOP 20-pin SSOP 20-pin SOIC Airflow(1) o LFM OLFM OLFM OLFM 250LFM 250LFM 6JA(2) 90°CIW 90°CIW 10rCIW 107°CIW 82°CIW 82°CIW fMAX 62.2MHz 72.5MHz 52.2MHz 60.6MHz 68.4MHz 43.4MHz 2913 tbl 01 4.1 of 1992 High-Performance Logic data book. SUMMARY This application note discusses frequency limits based on thermal considerations. The theory discussed here is applicable to all CMOS devices, but the issue is particularly relevant to clock driver operation. This is due to the large number of simultaneously switching outputs and higher aver- 40 age power dissipation in clock drivers, as compared to other logic devices such as simple data/address buffers. Other considerations such as timing margins may fix an even lower maximum operating frequency than one calculated on the basis of thermal limits, in which case, power considerations do not limit the operating frequency. d ~ APPLICATION NOTE AN-47 SIMULTANEOUS SWITCHING NOISE Integrated DevIce Technology, In~. By Suren Kodlcal INTRODUCTION ence between the device power trace (ground or Vcc trace) and the external power plane. The term "lead" used here refers to the combination of bonding wire and package pin. A specific area of interest is the simultaneous switching of several "sink" transistors during the logic HIGH to LOW transition and the resultant transient potential difference between the chip ground and the external ground plane. This phenomenon is the simultaneous switching noise on the device (chip) ground plane and is commonly referred to as "GROUND BOUNCE". Second, the high dynamic drive currents will cause very fast voltage edges at the switching outputs of the device subjected to predominantly capacitive loads. For example, a load capacitance of 50pF (equivalent of 6 to 7 typical CMOS inputs) will be discharged at a rate of 2V/ns during the high-ta-Iow output transition if the dynamic drive current of the output sink transistor is 100mA. Such rapid edge rates will make relatively short PCB traces look like transmission lines. For example, a 2V/ns edge rate will make a typical trace of 6 inches or more look transmissive. These fast edges will contribute to system noise due to ringing, overshoots and undershoots on the signals, EMI and RFI due to sharp output voltage transitions and cross-talk between two adjacent signal lines on a PCB surface. In this application note we will discuss the phenomenon of GROUND BOUNCE, its contributing factors and some design and application guidelines for minimizing the effects of ground bounce. The ne~ for in~reasing levels of throughput and improved performance '" today s systems has placed certain demands on the logic. and interface ~evices used in these systems. Two of the key reqUirements are high speed and high dynamic drive. Often the traditional glue logic and interface parts are in the critical timing paths and playa key role in determining system performance. Better s~eed (shorter propagation delays) lead to improved timing margl~s and offer opportu~ities for performance upgrading. The techniques used for Improving the speed also result in faster edge ra~es at ~he .outputs of these devices. As edge rates get faster, p~lnted circuit boa.rd traces and back plane wiring appeartransmisslve at shorter distances. More and more interconnections between circuits now have to be treated as transmission lines. This scenario leads to a requirement for higher dynamic drive at the outputs of. m~st ~igh-speed circuits in order to drive low impedance transmission lines and to sustain high levels of DC current nthe traces or backplane wiring are terminated at the far end. Thi~ ~imultaneo.us r~qu!remen.t for high speed and high drive has certain Important Implications. First, the high speed in most CMOS integrated circuits is achieved by improved device processing and topology. Internal nodes slew faster and transistors reach their saturation c~rren.t more r~pi~ly, result~ng in a higher rate of change of curre~t (dI/t!t) '" all sWltc~,"g transistors. Since most outputs of glue logiC. and Interface deVices are designed to handle high levels of dy~amlC current, the ~ate of build-up of current is particularly severe In the output transistors. When several outputs switch simultaneousl~, the total build-up of current in the common ground or Vcc lead Inductance can be substantial (of the order of 200mAlns to 300mAlns) and can develop a large transient potential differ- THE "GROUND BOUNCE" PHENOMENON Figure 1 shows the equivalent circuit of a typical CMOS output buffer stage with the package parasitics and the external load. PCB Vee PLANE : ~ Vee Lead CHIP Vee SWITCHING CURRENTS FROM OTHER CIRCUITS ON CHIP Output Lead Cl LOAD Ground Lead -+ ,: Figure 1. Output Buffer with External Parasltlcs © 1988lnllegratod Device Technology, Inc. 41 111811 APPLICATION NOTE A~7 SIMULTANEOUS SWITCHING NOISE The parasitic components which influence ground bounce are; (a) inductance and resistance of the ground bond wire and pin, (b) inductance and resistance of the output bond wire and pin, and (c) load impedance. Forfirst order analysis, the parasitics associated with the Vee terminal can be ignored. Also the external ground plane is assumed to be ideal. During the output high to low transition, the sum of output load current and all switching current through the internal gates of the device flows through the ground lead. The rate of change of this current (di/dt) develops a voltage drop across the ground lead inductance (Lg) and causes a positive ground bounce or an overshoot in an otherwise quiet ground. This positive bounce is normally followed by an undershoot coincident with the voltage waveform on the output terminal. The amplitudes of both positive and negative ground bounce are a function Lgdi/dt and of the number of outputs switching simultaneously. The ground bounce phenomenon can be clearly observed at an unswitched "LOW" output of a device by switching several other outputs simultaneously from logic HIGH to LOW. Figure 2 shows a typical output voltage transition and the corresponding ground bounce as observed at the unswitched LOW output. The positive ground bounce is primarily the result of the rate of change of current (dig/dt) through the ground lead inductance. This rate is determined by the rate at which the gate to source voltage (Vgs) of the sink transistor changes. During the early part of the output fall time, the ground voltage rises while the output voltage (at the drain of the transistor) falls, forcing the sink transistor into the linear region. The transistor then behaves like a resistor Ron (the "on" resistance of the transistor in the linear region). For the remainder of the output voltage excursion, the equivalent circuit at the output can be treated like a resonant L-C-R circuit formed by the ground and output lead inductance, load capacitance and the total resistance in the loop which includes Ron. The oscillation frequency is determined by the net values of Land C while the damping is determined by L and the total resistance in the loop. VOUT . . . - - (SWITCHED OUTPUT H ~ L TRANSITION) 1 • _ _ _ _ _ POSITIVE GROUND BOUNCE V NEGATIVE GROUND BOUNCE TIME Figure 2. Ground Bounce Waveform allel with a soon resistor to ground. Two outputs are connected to the oscilloscope; one for observing the HIGH to LOWtransition of a "switched" output and the other for observing the ground bounce on the "quiet" output. At these outputs, the soon load is split into a 450n resistor in series with the son input impedance of the scope probe. Alternatively, a soon load resistor can be returned to ground and a high impedance probe connected to the device output pins. With careful layout, proper bypassing to filter out high frequency noise and with a good oscilloscope and probes (bandwidth of at least 400 MHz), it is possible to observe the ground bounce on the internal ground of the chip by observing the voltage at the unswitched "LOW· output whose sink transistor operates in the linear region and provides a "Kelvin connection" to the chip ground. Ground bounce is also generated during the output LOW to HIGH transition. However, the magnitude of this ground bounce is much smaller because of the absence of load current in the ground lead. GROUND BOUNCE MEASUREMENT There is no industry standard per se for measuring ground bounce. However, the method most commonly used by IC vendors and customers alike is based on observing the disturbance of the logic LOW level of an unswitched output of a multiple output device while switching all other (or several other) outputs from HIGH to LOW state. Figure 3 shows the schematic for measuring ground bounce on a device such as the FCT244 octal buffer. One output is in the LOW state while 7 outputs are switched simultaneously. The load on each output consists of a SOpF capacitor to ground in par- 42 APPLICATION NOTE A~7 SIMULTANEOUS SWITCHING NOISE @ @ oselLLoseop PULSE GEN PRR = 1 MHZ 50 OHM OUT INPUTS LOAD CAPACITORS ARE PLACED AS CLOSE TO THE OUT AS POSSIBLE Figure 3. Ground Bounce Teat Circuit A scheme similar to the one shown in Figure 3 can be adapted for any multiple-output circuit. It can also be modified (by changing the load on the switched outputs) to observe and measure ground bounce during HIGH Z to LOW transitions in devices with 3-state control. rameter also determines the saturation current (dynamic drive current) in the output sink transistor. A shorter Leff results in a faster device, but at the same time gives a larger di/dt in the sink transis· tor. Therefore, there is a direct correlation between ground bounce (caused by di/dt) and speed. This relationship is shown in Figure 4 where the positive ground bounce is plotted as a function of tpHL for an FCT244 device in PDIP, SOIC and LCC. The ground bounce is measured at room temperature and Vee = 5V using the test method shown in Figure 3. THE RELATIONSHIP BETWEEN GROUND BOUNCE AND SPEED In CMOS circuits, the effective channel length (Leff) is the primary determ inant of speed. However, for a given topology, this pa· 3 2 (voHs) Vg 3 4 5 tPHL 6 (ns) Figure 4. Figure 4 illustrates two important points. First, because faster devices show a higher amplitude of ground bounce, one must exercise caution when comparing different logic families or different vendors for ground bounce. The samples to be compared should be in the same type of package and their propagation delays should be in close proximity. Second, package parasitics (ground lead inductance in particular) have a significant impact on the mag· nitude of ground bounce. In a standard DIP package, the corner pin ground lead inductance (pin #10 in a 20 pin package) is around 12 nH. In an SOIC this inductance is only about 7 nH and shrinks to around 4 nH in an LCC. The difference in ground bounce ampli· tude for different packages is clear from the above graph. 43 SIMULTANEOUS SWITCHING NOISE APPLICATION NOTE AN-47 rent in the transistor. As more outputs switch simultaneously, the peak current in each transistor actually decreases, although the total current in the ground lead increases. This "diminishing returns" effect results in a non-linear relationship between ground bounce and the number of outputs switching simultaneously. Figure 5 shows the ground bounce for an FCT244 octal buffer in a PDIP package measured on pin #18 under nominal operating conditions. EFFECT OF NUMBER OF OUTPUTS SWITCHING Ground bounce increases as more outputs switch HIGH to LOW simultaneously. Actual measurements indicate that the relationship is not linear. The reason is as follows. When the chip ground voltage rises due to the Ldi/dt effect, it modulates the gate-tosource voltage (Vgs) of the sink transistor and limits the peak cur- Vg (V) EXPECTED RANGE o~---+----r---+----+--~~--~---- 2345678 NUMBER OF OUTPUTS SWITCHING Figure 5. Simultaneous Switching Effect on Ground Bounce skew for an FCT244 device used as a clock driver. The actual measurements were made under worst case commercial temperature and Vee conditions for speed. In critical clock driving applications, the absolute magnitude of output skew can be reduced by using devices in SOIC or LCC packages. Switching fewer outputs per device will further reduce the skew. However, this approach has to be weighed against the device to device skew if more packages have to be used as a result. 4. The most serious effect of ground bounce is associated with the loss of dynamic noise margin which results in the loss of stored data in latches and registers. This loss of noise margin is often caused by the negative ground bounce which follows the high to Io.w transiti~n of several simultaneously switching outputs (see Figure 2). Simply stated, the undershoots on the chip ground lower the input threshold, or trip point, of the device. This has the same effect as an input making a LOW to HIGH transition relative to the chip ground. Hthe undershoot is large enough to bring the input threshold near the logic LOW level of any of the inputs held LOW (with reference to the external ground plane), all dynamic noise immunity in the input stage of the device is destroyed. The apparent LOW to HIGH transition of clock (or latch enable) and any logic "LOW" data inputs of registers (or latches) will have the effect of losing stored "LOW" data which is now replaced by "HIGH" data. EFFECT ON DEVICE PERFORMANCE Ground bounce causes a variety of effects in the application environment as described below: 1: The most commonly. observed effect is the noise on a "quiet" logic LOW output level In a device when several other outputs ~witch from high to low simultaneously. Ifthe amplitude ofthe positive ground bounce exceeds the input threshold of the device driven by it, then all noise margin disappears and the driven device may recognize the noise as a legitimate input transition. A worka~und for this problem is to allow some settling time before the signals atthe output of the device are treated as valid. This solution g~.nerally appl!es to combinatorial paths only. A large positive transition on certain control signals such as CLOCK, LATCH ENABLE, RESET, etc. can cause loss of data. Such a problem can only be solved by taking steps to reduce the magnitude of the positive g~ound bounce below the recognition level (threshold) of the device. 2. Changes in the chip ground voltage disturb the thresholds or trip points, of internal gates. This can cause non-monotonic output transitions that look similar to the effect of short unterminated transmission lines. Often this is not a serious issue. 3. The phenomenon described in (2) also causes a skew or a separatio~ betwe?n edges.of several outputs switching simultaneo~sly. ThiS skew IS a function of the number of outputs switching. Figure 6 shows the effect of simultaneous switching on output 44 SIMULTANEOUS SWITCHING NOISE APPLICATION NOTE AN-47 SKEW: SEPARATION BETWEEN THE FASTEST AND THE SLOWEST EDGE WHEN SEVERAL GATES ARE SWITCHED SIMULTANEOUSLY IN THE SAME DIRECTION 1000 800 600 Tskew (ps) 400 EXPECTED BOUNDS 200 0 2 3 4 5 6 8 7 NUMBER OF OUTPUTS SWITCHING Figure 6. Effect of Simultaneous Switching on Output Sk_ Using Smaller Packages Since ground bounce is the voltage induced in the ground lead inductance by the rapid rate of change of current through it, there is a direct correlation between the amount of inductance and the magnitude of ground bounce. Ground lead inductance can be reduced by using packages with smaller internal cavities and lead dimensions. For example, for corner Vee and GND configuration, the typical ground lead inductance for a 20 or 24 pin Plastic DIP package is of the order of 12 nH to 15 nH. This inductance drops to about 7 nH in an SOIC package and to about 4 nH in an LCC package. Figure 7 shows the effect of package lead inductance on the positive ground bounce for an FCT244 device. Another method of reducing ground lead inductance is to arrange the pad layout such that the power pins (particularly the GND pins) are at the center of the package for the shortest lengths. Mhough this is an acceptable solution, it raises standardization and compatibility issues on industry-standard functions. This choice does exist for new functions and as an addition to existing standard functions. SOLUTIONS TO THE GROUND BOUNCE PROBLEM Ground bounce is a pervasive phenomenon. It can be minimized or circumvented, but rarely eliminated since the parasitic inductance cannot be totally removed from the package. The solutions to the ground bounce problem take essentially two forms; (a) minimizing the effects of ground bounce and (b) minimizing the magnitude of ground bounce. To minimize the effects of ground bounce, the design should be made "ground bounce tolerant". This can only be done at the expense of system throughput, since additional time must be allowed for the ground bounce to settle. As a result, the benefit of using high speed logic is partially negated. The techniques for minimizing the magnHude of ground bounce take many forms. They are generally aimed at reducing eHher the parasHic inductance, or the amount of dildt or both. These techniques are discussed in some detail below. MEASURED ON OUTPUT: PIN 18 3 2 Vg (volts) 1 MEASURED RANGE o o 4 8 12 16 GROUND LEAD INDUCTANCE - nH Figure 7. Effect of Package Lead Inductance 45 20 SIMULTANEOUS SWITCHING NOISE APPLICATION NOTE AN-47 Series Damping di/dt can be reduced either by limiting the magnitude of the peak current (Imax) through the ground leador by slowing down the buildup of the total ground lead current during the output transition. The value of Imax depends on the size of the output sink transistor as well as on the load. Since this is a dynamic phenomenon, the peak current depends more on the amount of energy stored in the load capacitance. One effective method of limiting the magnitude of Imax is to use a series damping resistor at the output. During the output transition, this resistor comes in series with the "on" resistance of the output buffer and limits the peak current, and hence the di/dt. Figure 8 shows the effect of series damping resistance on ground bounce for an FCT244 device. 3 NO SERIES DAMPING 2 (voHs) Vg 3 4 5 tpHL 6 (ns) Figure 8. Effect of Series Damping on Positive Ground Bounce As seen from the figure above, the series damping resistor causes a significant reduction in ground bounce. However, it is important to understand the implications of using series damping. Since the total output impedance is significantly higher, the transmitted signal is attenuated at the driving end; the attenuation being determined by the source impedance (Ro + Rs) and the loaded transmission line impedance (ZL). Ro is the "on" impedance of the output circuit. This attenuation limits the amplitude of the first incident wave. Therefore, the series damping technique must be used with caution and is not recommended if first incident wave switching is desired. This subject is covered more thoroughly in the Application Note entitled "SERIES TERMINATION'. The series damping resistor, if properly chosen, does have the advantage of limiting overshoots and undershoots on the transmitted signal without increasing system power dissipation. The series damping resistor also decreases the magnitude of negative ground bounce and the undershoot on HIGH to LOW transitions. Therefore, series damping is effective in driving CMOS memories, particularly DRAMs where undershoots on input signals are undesirable. swing at the output is limited, less energy stored in the load. For example, if the output swing is limited to 3.3 volts nominal (similar to most bipolar or BiCMOS totem-pole outputs) instead of rail-torail, the energy stored in the output load can be decreased by a factor of 2.5:1 for a given load capacitance. This results in a smaller positive as well as negative ground bounce. There is a beneficial side effect of this method. Since the high to low transition starts from a lower voltage level, the fall time component (the time taken for the output to switch from the logic HIGH level to the 1.5V measurement level) of tPHL is smaller than that for a rail-to-rail transition. This translates into an improvement in tPHL. However, since speed improvement is not the primary objective, some or all of this speed improvement can be sacrificed in order to further reduce the simultaneous switching noise. This can be achieved by means of a circuit configuration which provides a smaller initial dildt during the logic HIGH to LOW transition. The resulting degradation in the output fall time cancels the speed improvement. In practice, about 40 improvement in ground bounce (relative to rail to rail swing output) can be obtained for the same speed. The FCT- T family of products introduced by lOT has been designed using this approach. Figure 9 shows the ground bounce characteristics for IDT74FCT244T/AT devices in PDIP and SOIC packages in comparison with the IDT74FCT244/A. Reduced Output Swing Another technique for reducing ground bounce relies on limiting the energy stored in the load on the device output(s). If the voltage 46 SIMULTANEOUS SWITCHING NOISE APPLICATION NOTE AN-47 3 244 BUFFER 2 Vg (volts) + SIMULATION DATA C 3 CHARACTERIZATION DATA 4 5 tpHL 6 (ns) Figure 9. Improved Ground Bounce In FCT-T Family duced by using smaller packages with lower ground lead induc· tance, by switching fewer outputs of a device simultaneously, or by using a series damping resistor in the rail-ta-rail swing FCT logic. The new FCT-T logic family is designed to offer much smaller ground bounce at the same speed by reducing the output voltage swing and by controlling the dildt in the ground lead during the output transitions. SUMMARY The requirement for high speed and high drive results in the phenomenon of ground bounce or simultaneous switching noise in high speed logic and interlace circuits. The effects of ground bounce range from a noise spike on a quiet output to data loss in registers and latches. The magnitude of ground bounce can be re- 47 USING HIGH-SPEED LOGIC APPLICATION NOTE AN-48 Integrated DevIce Technology. Inc. By Suren Kodlcal This application note gives some general guidelines and recommendations for using high speed logic such as FCT, FCT A and FCT B family of products. when compared to driving with devices with bipolar outputs which have a higher logic low level due to the offset voltage of the Schottky-damped NPN sink transistor. b. Reducing DC loading, Le. reduce the fanput, on the outputs of devices that drive the data and control inputs of latches and registers. c. Using devices with "hysterisis" on the inputs. This will further improve dynamic as well as static noise margin. d. Use of series damping resistors (25 to 35 ohms) at the output of latches and registers will reduce the undershoot on the device intenal ground due to simultaneous switching of mu~iple outputs of the device. This undershoot normally follows the overshoot (also referred to as the ground bounce). Series damping will overdamp the series L-C-R circuit formed by the parasitic ground path inductance, load capacitance and the low impedance of the sink transistor. 2. Several measures can be taken to reduce the power supply noise - both ground bounce and Vee bounce: a. Contention should be avoided on devices connected to a bus. Although bus contention is not detrimental to the device in a normal application, it causes very large positive and negative di/dt in the ground and Vee paths. Such contention has the same effect as charging a very large load capacitance. b. Series termination (i.e. series damping resistors) will also reduce the magnitude of the ground bounce by limiting the maximum transient current and thereby decreasing the total energy transferred to the parasitic inductances. c. Use of local high frequency filtering will minimize the propagation of noise on ground and Vee traces. 3. Avoid running control lines through a device that drives datal address buses. 4. Since the magnitude of ground and Vee noise is a direct function of parasitic inductance, if all other conditions are unchanged, much benefit is gained by using packages with lower bond-wire and lead inductance. Thus, surface mount packages (SO, LCC, PLCC, etc.) will offer lower levels of ground and Vee noise than standard DIP packages 5. Output drive "overkill" should be avoided. In non-critical paths, use of low-drive circuits will generally reduce the overall supply noise. 6. In very high speed circuits, minimize the loading per device to reduce total load capacitance. POWER DISTRIBUTION 1. Use Ground and Vee planes on mu~i-Iayer boards. 2. On two-layer boards with no Vee and Ground plane, use a "grid" type ground distribution system to equalize ground potential at different points on the P C board. 3. Use Power Distribution Elements - PDEs (conductor-dielectric-conductor) to reduce characteristic impedance. Use separate PDEs for devices that sw~ch large amounts of sink and source currents. 4. Do not use jumper wires for ground connections. 5. Provide a separate "noisy" ground distribution for high current drivers, particularly those driving backplanes. 6. Place high current driving circuits near their loads. For example, place backplane drivers at the edge of the board. 7. Make adequate provision for supplying transient energy to handle PC trace impedance and load capacitance. This is done by connecting individual bypass capacitors across the power pins of high current switching circuits. 8. Use low-inductance, ceramic disk capacitors (4700 pFto 0.1 uF) for high frequency fi~ering. These can be used in parallel with normal bypass capac~ors. SIGNAL TRACES 1. Treat the PC board traces as transmission lines. A conservative rule of thumb is to consider a trace as a transmission line if the unloaded signal transition time at the driving end equals the round-trip propagation delay for the trace in question. Typically, the transmission line delay is 0.15 ns per inch, or 0.3 ns for one inch round-trip. That means, for a transition time of 2 ns, a trace longer than 7 inches should be considered as a transmission line. 2. To minimize cross-talk between signal traces, avoid running sensitive signal lines close to traces connected to high current drivers. 3. Any signal lines that cross each other should be placed at right angles to further reduce cross-talk. DEVICE SELECTION 1. Select devices which offer the largest amount of "real" noise margin. Ground noise due to simu~aneous switching of multiple outputs causes a loss of dynamic noise immunity in the logic low state. Therefore, it is important to improve "low level" noise immun~y. This can be achieved by: a. Using CMOS outputs to drive inputs of "storage" devices such as latches and registers. This will offer better noise margin SUMMARY A combination of high speed (particularly fast edge rates) and high drive contribute to increased noise in power supply path as well as in signal paths. Much care is needed to minimize such noise so that maximum performance benefit is derived from the FCT family of high speed logic products. © 111881ntog_ DevIce T_agy, Inc. 11/88 48 APPLICATION NOTE AN--49 CHARACTERISTICS OF PCB TRACES fJ Integrated DevIce Technology. Inc. By Suren Kodlcal Printed circuit board traces carrying high speed digital signals can behave like transmission lines for fast edge rates of the driving signal. The transmission line effects can cause signal distortion. ~vershoots •. undershoots and crosstalk between adjacent lines. It IS therefore Important to understand this behavior for trouble--free board design. In this application note we discuss the transmissive effects of PCB traces. the relevant electrical parameters and a simple technique for measuring unloaded or loaded characteristic impedance. PCB TRACE AS A TRANSMISSION LINE A PCB trace is normally regarded as a very low impedance medium which carries electrical signals from one point to another. This is true for most signals with relatively slow edge rates (long rise and fall times). However. when a trace is subjected to fast edge rates. its behavior changes completely. It behaves like a transmission line with a certain characteristic impedance Z. This impedance now presents a load to the driving circuit. In addition. the transmissive trace introduces a finite signal delay from source (driving end) to destination (receiving end). The equivalent circuit for a transmission line. represented by distributed l+R and C is shown in Figure 1. Lo ---------~ T Figure 1. Tranamlsslon Une - In this schematic. La = Inductance of the trace per unit length. and Co = Capacitance per unit length. For the purpose of this discussion the series resistance can be ignored. thus treating the PCB trace as a "Iossless" transmission line. The subscript "0" implies that the inductance and capacitance pertain to the "unloaded" condition. i.e. the inductance is the "self inductance" of the trace and the capacitance is that offered by the dielectric separating the trace in question from the adjacent conducting media. Two important parameters can be derived from lo and Co. The first one is Zo. the characteristic impedance of the trace. The second parameter is To. the propagation delay per unit length of the trace. It should be noted that the parameter Zo is independent of the length of the trace. Equivalent Circuit Zo = (lo/Co )1/2. normally defined in ohms ......... (l) To = ( loCo )1/2 per unit length. normally defined in nanoseconds ..... (2) Example A typical MICROS TRIP PCB trace (a dielectric separating the trace from the ground plane on one side and free air on the other 3 sides of a rectangular trace cross-section) which is 10 mils wide and 1.5 mil thick separated from the ground plane by 15 mil glassfilled epoxy has a typical Co =2 pF/inch and La =10 nHlinch. Using the above equations. we get Zo = 70 n and To = 0.15 ns/inch. The table in Figure 2 gives various transmission line geometries and their parameters. © 1989ln1egratod Device Technology. Inc. 11/89 49 CHARACTERISTICS OF PCB TRACES APPLICATION NOTE AN-49 TYPE CO-AX WIRE OVER GROUND Zo ohms To ns / inch @ 50-125 0.13 0 70 -170 0.14 30-150 0.15 15 -100 0.19 50-200 0.16 GEOMETRY GND .1211;i~~6.I¥ MICROSTRIP LINES GN12-I. STRIP LINE GN~I: :;~:;::::;I PC BOARD TRACES I -I I .",..., E:ll DIELECTRIC I Figure 2. Transmlsalon Une Geometries input capacitance, most CMOS inputs offer a capacitive load (ignoring the effects of input lead inductance). As a result, the transmission line parameters are modified under the loaded condition, because the additional distributed load capacitance must be taken into account in addition to the unloaded distributed capacitance Co. To simplify the discussion,let us assume that the distributed load capacitance is represented by Cl per unit length, as shown in Figure 3. EFFECT OF LOADING The concept of an unloaded transmission line applies to pointto-point connections which consist of a driver and a receiver at the two ends of a trace, with no connections to the trace in-between. Most often, a PCB trace is tapped at several points and connected to inputs of several ICs. Clock, RIW, Chip Select lines and Data and Address buses are examples of this. These IC inputs represent a quasi-<:listributed load to the driving circuit. Whereas bipolar TTL inputs present a DC leakage path to Vee in addition to the .. ---_.- --------_._ .. _.---_ .. UNIT LENGTH Figure 3. Loaded Transmlaslon Une 10 pF per inch .. Using the example for the PCB trace given earlier, the loaded parameters can be calculated using equations (3) and Taking the effect of CL into account, the loaded trace impedance ZL and the loaded transmission delay TL per unit length are given by: ZL = Zo [ Col (Co + Cl) ]1/2 ohms............................... (3) andTl - To [(Co+ CL)/CO p/2 per unit length ................... (4) (4): Zl =70 [2/ ( 2+ 10 ) )1/2 =70[ 1/6 p/2 =7012.45 TL = 0.15 x 2.45 = 0.37 nslinch. =29 Q and Example This example illustrates the need for a significantly higher drive as the trace impedance drops from 70 Q to 29 Q. It also shows the impact of such loading on clock skew caused by the increase in transmission delay. Consider a clock driver driving a bank of registers in DIP package, mounted 0.5 inches apart. Assuming a typical input capacitance of the clock pin of 5 pF, the PCB trace is loaded with 5 pF capacitance every 112 inch. This is equivalent to a distributed Cl of 50 APPLICATION NOTE AN-49 CHARACTERlsncs OF PCB TRACES ful. Described below is one such method which is particularly applicable to traces with well-{jistributed loading. Equipment required: 1. Pulse generator with known source impedance (typ. 50 0) and capable of rise and fall times faster than 2.5 ns. 2. Oscilloscope: >350 MHz bandwidth. 3. High impedance scope probes. WHEN IS THE PCB TRACE TRANSMISSIVE? As a general statement, a PCB trace looks like a transmission line for fast edge rates of the transmitted signal. To quantify this, a commonly used rule of thumb is: TREAT A PC BOARD TRACE AS A TRANSMISSION LINE IF Ts...2Lx h. Method a. Hthe source impedance of the pulse generator is~, it can be easily obtained by observing the unloaded output waveform of the pulse generator, and then loading the pulse generator output with a resistance that will halve the amplitude of the original signal. The value of this load resistance is the source impedance (Rs) of the pulse generator. b. Connect the pulse generator and the oscilloscope to one end of the PCB trace. Use a minimum of 9 inches of PCB trace. Insert the devices that will form the distributed load on the PCB trace. See Figure 5. In this equation, T = output transition time (rise or fall time) TL= loaded transmission delay per unit length. T l = To for the poinHo-point case l = length of the PCB trace Example Consider the loaded transmission line in the example cited above. If the clock driver has an output transition time of 5 ns, the length at which the PCB trace should be treated as transmissive is given by: l = T 12Tl - 51 (2 x 0.37) = 6.8 inches. I! is clear that, with slower edge rates a driver can drive longer traces without transmission line effects. The table in Figure 4 shows the limiting signal line length for different logic families based on typical edge rates for each of the families and lYRical unloaded signal traces. lOGIC FAMILY PCB TRACE SIGNAL LINE LENGTH· (INCHES) lS 25 S,AS 11 F,ACT 8 AS,ECl 6 FCT, FCTA 5 'SCOPE Figure 5. Tesl Sel-Up for Measuring PCB Trace Paramelers c. Set up the pulse generator to obtain a 5V amplitude square wave of 1 MHz frequency. Adjust the rise and fall times to get 30 ns (10% to 90%). These slow edges ensure that the PCB trace behaves like a lumped load and not like a transmission line. Overshoots and undershoots on the waveform are avoided. Record the amplitude (Vs) observed on the oscilloscope under these conditions as shown in Figure 6A. d. Now change the setting on the pulse generator to get the fastest rise and fall time. Observe the high to low transition on the oscilloscope. Note that there is a step in the output transition as shown in Figure 6B. Record the amplitude of the first segment of the output transition (VI) and time interval between the start of the first transition and the start of the second transition (2l x T l). e. Determine the characteristic impedance of the loaded PCB "Length above which the signal trace looks like a transmission line. Figure 4. Signal Line Length vs Logic Family This table shows that, as we go to faster logic devices, it becomes more critical to understand the transmission-line effects. Note that the signal lengths given in the table are not guarantees for any logic family. The actual limiting signal length is a function of trace and board characteristics, trace loading and the edge rates of individual devices in any logic family. MEASUREMENT OF PCB TRACE PARAMETERS trace (Zl) by the formula: Zl = Rs x [VII (Vs - VI) 1 Since both Zl and Tl depend on board layout and loading, a simple practical method of determining these two parameters is use- 51 CHARACTERISTICS OF PCB TRACES v APPLICATION NOTE AN-49 Vs V J -- -- Q 2L·T, ~ Time (ns) Measuring Vt and TL with Fast Edge Rate Time (ns) Measuring Vs w~h Slow Edge Rate Flgure6A. Flgure6B. device may not respond to the signal at the driving end (see Figure SB) until the reflected wave reinforces the signal after a turnaround delay along the PCB trace. This implies that if a driver is driving a PCB trace from one end, the receiver nearer to the driver will respond afterthe receiver at the far end of the PCB trace. Such skew may be unacceptable in certain conditions. The relative values of trace impedance and the load impedance at the far end also determine the amount of reflection and hence the overshoots and undershoots on the waveform. By understanding the implications of transmissive traces, a designer can choose the right termination and drive capability of the driving circu~ to derive the maximum benefit. The line propagation delay to the end of the trace is given by TL. Example = In a test performed on a 9 inch trace, we get As - 45 n, VI 2.8 V and 2L x TL = S ns. Then, ZL = 45 x [ 2.8 / (5 - 2.8) I =57n. TL= S/(2x 9) = 0.33 ns per inch THE IMPORTANCE OF PCB TRACE CHARACTERISTICS SUMMARY The relative values of source impedance (As) of the driving circuit and the loaded characteristic impedance of the transmission line (ZL) determine the effectiveness of driving transmission lines. As Rs increases for a given ZL, the amplitude of the transmitted component of the waveform (VI) decreases. If the transmitted wave does not cross the threshold of a receiving device, the receiving This application note describes the effect of fast edge-rates on the behavior of PCB traces. A simplified method for measuring the trace parameters in a given application environment is shown. The procedure discussed here can be extended to fully loaded backplanes. 52 ~ g APPLICATION NOTE AN-50 SERIES TERMINATION Integrated DevIce Technology. Inc. By Suren Kodlcal Series termination is one of several forms of terminating transmissive lines. In this bulletin we discuss the pros and cons of series termination and the effect of termination impedance on simultaneous switching noise (a.k.a. Ground Bounce). SERIES TERMINATION Figure 1 shows a typical case of a series terminated driver connected to a load via a PCB trace. ~ Flo, • WHY TERMINATE? With the constant push for higher speeds, particularly in the area of standard logic and bus interface products, system designers have to deal with devices with fast edge rates. At the same time, high packing density of multl-layer boards results in PC board traces with low loaded impedance and long transmission delays. This combination of fast edge rates and low transmission line impedance requires the system designer to pay careful attention to PCB design in order to maximize the benefits of today's highspeed logic. As more and more devices on the boards go to CMOS technology, typical nets consist of outputs with fast edges looking into transmission lines with some distributed capacitance along the line or lumped capacitance of CMOS IC inputs at the end of the line or a combination of both. In the absence of some form oftermination, overshoots and undershoots on the signal can impose bandwidth limitation on the system, or subsystem due to settling time requirements or, even worse, can cause false triggering and data loss. Termination oftransmission lines is the time-honored method of improving signal quality. There are several forms of termination: a. Parallel or shunt termination: a single resistor terminated to either Vee or GND at the end of the PCB trace. For backplanes, termination is provided at each end. b. Series termination: a single resistor is connected between the output node of the driver and the PCB trace or any other transmission line being driven. c. Thevenln termination: two resistors form a potential divider at the far end of the transmission line. The junction of the two resistors goes to the transmission line and the two ends typically go to GND and Vee. This type oftermination is commonly used on backplanes at both ends. d. RC termination: an R-C series combination is connected between the transmission line and GND at the far end. Each of the termination schemes listed above has certain advantages and disadvantages. A detailed discussion of the relative merits of these schemes will be part of a separate application note. In this issue, we will focus on SERIES TERMINATION. ~\ DRIVER PCB TRACE Zo +-TD--' € LOADS " - - SERIES TERMINATION Rs Figure 1. The effective output impedance of the driver is now the sum of device source impedance (Ro) and the series terminating resistance (Rs). This modified output impedance of the driver comes in series with the characteristic impedance (Zo) of the PCB trace and forms a potential divider for the incident signal. Therefore, the signal that propagates down the trace is a fraction of the "open-circuit" signal at the driving end. The magnitude of the transmitted wave is given by the following equation: VI = Vs [ Zo I (Ro + Rs + Zo) I ................ (1) This equation shows that, as the total source impedance approaches the characteristic impedance of the line, approximately half of the incident wave will be transmitted to the other end of the trace. Since the load impedance is much larger than Zo due to the high input impedance of the CMOS devices, most of the transmitted wave is reflected. As a result, overshoots and undershoots on the signal are minimized at the receiving end. If Ro + As is much smaller than Zo, a larger portion of the incident wave is transmitted down the trace. Since most of it is reflected, such a condition will cause overshoots and undershoots at the receiving end. Figure 2 shows the effect of series termination under prefect matching (total source impedance equals trace impedance). UiT"' G) OPEN-CIRC SIGNAL AT NO'" SENDING-E SIGNAL AT Wl To To To RECEIVING END SIGNAL AT - ,.--- To J @ Figure 2. © 1988 Integrated Device Technology, Inc. 11i8t 53 SERIES TERMINATION APPLICATION NOTE AN-SO The example shown highlights some important points: 1. For the best signal quality, the series terminating resistor should be chosen such that the total source impedance (Ro + Rs) is close to the characteristic impedance of the PCB trace. It is not essential that the two quantities match exactly. It is, however, important to ensure that the total source impedance is not greater than the trace impedance. Otherwise, multiple reflections may be needed to obtain the entire signal transition at the receiving end. 2. The waveforms in Figure 2 clearly show the effect of series termination on the waveform at both the sending end A and receiving end B. For a perfectly matched condition, the signal will attain the final value at the sending end after a round-trip delay ( 2To ), although it will attain the final value after one transmission delay (To) at the receiving end. H loads (several IC inputs) are distributed along the PCB trace and are driven by the driver at one end, this condition results in signal skew which may be unacceptable. Therefore, series termination may not be the most suitable form of termination for distributed loads. Of course, the ratio of source to trace impedance can be adjusted to ensure that the threshold of the receiver is crossed on the first Incidence of the transmitted wave, but this is normally at the expense of some undershoot and overshoot during signal transitions. 3. There is no limit to the number of lumped loads (as shown in Figure 1) that can be used, provided that the total DC loading does not reduce the static noise margin because of a voltage drop across the series terminating resistor. This implies that series termination is well-suited to drive inputs of CMOS devices because of their very low input current requirements. The primary limitation to the number of CMOS loads is the additional delay due to the total input capacitance being driven. 4. When series termination is used with lumped loads, the distance between the indvidualloads should be kept to a minimum. If the loads being driven are spread apart, a preferred method of driving them from one source is to make several groups of loads and drive each group from the driving source via individual transmission lines with their own series termination resistors. The driver should of course be capable of handling this additional transmission line loading. 5. It is clear that the series termination does not add any power dissipation to the system. it is, therefore, the preferred form of termination if power dissipation is a key consideration. 6. Series termination adds flexibility to the design in that the termination values can be tailored to suit a variety of trace characteristics and timing requirements. Reworking equation #1 , we get: Zo= (Ro+ Rs)[ (VOH-V1l)/VIL) Therefore, Rs = Zo [ VIL I (VOH - VIL) j- Ro ............... ( 2 ) Using the values for Vs and VI and for a device source impedance of 6 ohms, the maximum value of series termination resistance which will assure incident wave switching is given by Rs = Zo [0.81 (4.8 - O. ) j- 6 = (0.2Zo - 6) ohms For example, ff the PCB trace impedance is 70 ohms, the maximum value of series termination resistance is 8 ohms to assure incident wave switching. A similar consideration for the low to high transition yields the expression: Rs = Zo [ ( VOH - VIH) I VIH )- Ro .............. ( 3 ) For a VOH = 4.8 V, VIH = 2.0 V and a source impedance of 25 ohms in the logic high state, the maximum value of series termination resistance to assure incident wave switching is, Rs = (1.4Zo - 25) ohms Again, if the PCB trace impedance is 70 ohms, the maximum value of Rs is 73 ohms. This indicates thatthe high to low transition is the worst case. The above example shows that a requirement for incident wave switching will impose severe restrictions on the series termination resistance due to the high to low switching case. Since the termination value is much less than the trace impedance, a certain amount of overshoots and undershoots are to be expected on the output waveform at the far end of the PCB trace. In effect, the incident wave switching requirement is in conflict with signal integrity for FCT logic devices with rail to rail output switching when using series termination. If signal integrity is the primary consideration, then the series termination has to be chosen to match the trace impedance. However, signal skew has to be tolerated when driving a transmission line with distributed loading. Alternatively, series termination should be limited to driving lumped loads at the far end of the transmission line (PCB trace). In high-speed switching circuits, series termination offers another advantage. When driving predominantly capacitive loads, the series resistor serves to limit the peak current in the output pull-down transistor and therefore the resultant di/dt in the parasitic lead and bond wire inductance. This has the beneficial effect of limiting the amount of ground bounce (induced by the L.di/dt effect) as a result of simultaneous switching of high drive outputs. SERIES TERMINATION WITH FCT DRIVERS Like most TTL-<:ompatible drivers designed to meet the standard DC specffications, the FCT output buffers offer different output impedance in the logic LOW and HIGH states. Typically, the output impedance is 6 ohms in the LOW state and 25 ohms in the HIGH state. Since the internal thresholds of all TTL-<:ompatible devices (independent of technology) are with reference to GND and the noise immunity is normally worse in the logic LOW state, it is important to consider the logic LOW state and the high-to-Iow transitions when evaluating the effect of terminations. First, let us consider the requirements for first incident wave switching. The aim is to cause enough voltage swing on the first part of the transmitted wave to cross the threshold of a receiving device close to the driver. For a typical VOH of 4.8V with CMOS P-channel pull-up transistors and specified VIL = 0.8 V for the receiver, the required amplitude of the incident signal is VI. VOHVIL = 4.8 - 0.8 - 4 V. The open circuit swing is Vs = 4.8 V. SUMMARY Series termination is an effective method for minimizing overshoots and undershoots on signals with fast edges and for reducing the amount of ground bounce caused by simultaneous switching. An understanding of the device output characteristics, particularly the output impedance values, is required to properly determine the value of series termination in order to assure incident wave switching. 54 -.. ~. APPLICATION NOTE POWER DISSIPATION IN CLOCK DRIVERS AN-51 Integrated DevIce Technology, Inc. By Suren Kodlcal Power dissipation in switching circuits is discussed in this bulletin, particularly with reference to CMOS clock driver circuits. The IDT54/74FCT244 octal buffer is used as an example to compare the power supply current in CMOS, bipolar and bipolar-based BiCMOS technologies over a wide range of operating frequencies. ECI2M. First, we need to determine the Cpo for the device. Since Cpo = leeoN in pF if IceD is expressed in IJAIMHz, we can determine Cpo using the max. limit specified for lecD in the data sheet. Therefore, Cpo = 250 15.5 = 45 pF POWER DISSIPATION COMPONENTS When the device is loaded with 30 pF capacitance per output, the dynam ic dissipation component increases due to load. The loaded value is given by, There are two components of power dissipation in integrated circuits. One is the steady-state component. This is the dissipation when all inputs are held at some fixed voltage level. The other component is frequency dependent and is generally referred to as the dynamic component. In CMOS and CMOS-based BiCMOS circuits, the steady-state component is further divided into two sutrcomponents; the quiescent power supply current (Icc) primarily due to device leakage and the quescent power supply current when inputs are at TTL high level (6Icc). This latter component applies to circuits with TTL compatible inputs. In bipolar and bipolar-based BiCMOS circuits, no such distinction is made and it is customary to specify power supply current for a given logic state on the output(s). The dynamic component of power dissipation (ICCD) is dominant in CMOS circuits because most of the power is dissipated in moving charge in the parasitic capacitors of CMOS gates. Therefore, the simplified model of a CMOS circuit consisting of several gates looks like one large capacitor which is charged and discharged between power supply rails. For this reason, a parameter called Cpo (power dissipation capacitance) is often specified as a measure of this equivalent capacitance and is used by the designers to estimate the dynamic power supply component. In the bipolar technology, the dynamic component is generally very small in comparison with the steady-state component because internal voltage swings are small. Since power supply parameters are traditionally specified under "unloaded" condition, a comparison of power dissipation for a given device type (FCT244 with F244, for example) based on data sheet numbers alone can be misleading. For a true "apples-to-apples" comparison, the effect of capacitive load on the device should be taken into consideration. This is particularly true in the case of clock drivers which drive heavy capacitive loads and operate at high frequency. Under these conditions, the dynamic power dissipation component due to output loading could be significant in both bipolar and CMOS circuits. This is illustrated in the following section by using the '244 Octal Buffer as an example. leGD ( loaded) = { (Cpo + Cl) I Cpo} ICGD = { (45 + 30) 145 } 250 IJA I MHz I bit = 0.42 mA I MHz I bit. When all eight outputs are switching simultaneously, the total IceD (loaded) is 3.3 mAlMHz. If quiescent power dissipation is ignored, the above equation can be used to determine the total power dissipation at any frequency when the input levels are CMOS compatible. For the case where the inputs are driven from a bipolar TTL device, the 6lcc component needs to be added in order to obtain the total power dissipation. Assuming a 50 duty cycle, for 61ec (max.) of 2 rnA, this static Icc component is 8 rnA. Figure 1 shows the power dissipation versus frequency for both conditions. F244 The specified power dissipation is lecl R 90 rnA and IccH = 60 rnA. For a 50 duty cycle, the steady-state dissipation is 75 rnA. In addition, the dynamic dissipation component appears due to the external load capacitance and the output pin capacitance of the device. For Cl = 30 pF and COUT 10 pF (assumed), the dynamic component can be derived: = IceD (loaded) = 40 pF x (4.3 V - 0.3 V) in IJAIMHz. = 160 IJA I MHz I bit where the 4.3 V - 0.3 V represents the voltage swing (for Vce = 5.5V) on the total load capacitance. For 8 outputs switching simultaneously, lecD is 1.28 mAl MHz. The total dissipation as a function of frequency is also shown in Figure 1. BCJ244 The BCT family from TI is developed with a bipolar-based BiCMOS process. Therefore, the power dissipation characteristics are similar to F244. The steady-state dissipation is 57.5 rnA for a 50 duty cycle. The dynamic component of the dissipation is 1.28 mAlMHz. The total power dissipation versus frequency is again shown in Figure 1. '244 Example Considerthe '244 as a clock driver with 30 pF load on each of the 8 outputs, operating at room temperature and Vee ~ max. Power dissipation of IDTs FCT244 is compared with F244 (FAST"') and Tl's BCT244. Data sheet numbers are used where applicable. © 1989ln1ogratod Device Technology, Inc. 11/88 55 POWER DISSIPATION IN CLOCK DRIVERS APPLICATION NOTE AN-51 The graph in Figure 1 shows that over a wide range of frequencies the power dissipation of FCT family of circuits is much less than that of BCT and F families, even under heavy capacitive loading. 120 100 80 TOTAL Icc (mA) 60 ! FCT244! (TIL LEVEL INPUTS) 40 20 o 5 10 15 FREQUENCY (MHz) 20 25 Figure 1. Total Icc va Frequency method can be extended to any other product and can be used to determine realistic power consumption if the loading and effective operating frequency can be estimated for each device. SUMMARY A simple method for calculating "real" power dissipation in an operating environment is shown by using '244 as an example. This 56 APPLICATION NOTE AN-52 FCT OUTPUT STRUCTURES AND CHARACTERISTICS ~ Integrated DevIce Technology, Inc. By Sure" Kodlcal ration is designed to give a resistive characteristic during the LOW to HIGH transition at the output. INTRODUCTION The FCT family of products has gone through an evolution in terms of die size, process technology (critical dimensions) and circuit implementation. Originally, the family of products was derived from the Z-step gate arrays ("4004" gate array for small gate count and "8000" array for large gate count). Later, a "shrunk" version of the smaller array was developed to obtain performance improvement. This array is called the V-step. Recently, some of the high volume runners have been "customized", i.e. redesigned to minimize the die size and get some performance improvement with a more efficient topology. These customized versions are called the W-step devices. The current FCT portfolio consists of a mix of Z, V and W step devices. This bulletin describes the output structures used in different steppings and the corresponding output characteristics for the logic HIGH and LOW states. "4004" Z- STEP OUTPUT BUFFER The schematicforthe buffer used in the "4004" Z step devices is shown in Figure 1. This output consists of an N-channel "sink" transistor which turns on in the logic low state at the output and maintains a logic low voltage close to GROUND for normalloading. The pull-up or "source" circuit consists of a combination of a Pchannel transistor, an N-channel transistor and an NPN bipolar transistor with a series current limiting resistor. This circuit configu- Figure 1. Stap Output Structure The output V-I characteristics for the Z step output structure in the logic HIGH and LOW states are shown in Figures 2A and 28, respectively. (VOUT) o 2 3 4 5 250 200 -50 150 (mA) -100 (mA) 100 -150 50 -200 o -250 2 3 4 5 (VOUT) Rgure 2B. "4004" Z Step Logic "Low" Characteristics Figure 2A. "4004" Z Step Logic "High" Characteristics The output characteristic in the logic HIGH state is dominated by the current limiting resistor in series with the NPN pull-up transistor. As the output reaches the Vee rail, the output characteristic is primarily influenced by the P-channel transistor. In the logic LOW state, the output characteristic is that of a large N-channel pulldown transistor. Note that the characteristics shown in Figure 2 represent typical process parameters at 25°C. © 111891ntegra1td Device Todmalogy, Inc. 11/88 57 FCT OUTPUT STRUCTURES AND CHARACTERISTICS APPLICATION NOTE AN-52 Y-5TEP OUTPUT BUFFER The circuit schematic for the Y step output buffer is shown in Figure 3. Figure 3. Y-Step Output Structure This output structure is designed to get shorter propagation delays. The output characteristic in both HIGH and LOW states is non-linear as shown in Figures 4A and 48, below. (VOUT) 3 4 5 250 200 -50 150 -100 (rnA) (rnA) 100 -150 50 -200 o -250 2 3 (Vour) 4 5 Figure 4A. Y Step logic "High" Characteristics Figure 4B. Y Step Logic "Low" Characteristics W STEP & "8000" Z STEP OUTPUT BUFFER structure consists of a parallel combination of P--channel and Nchannel transistors in the pull-up circuit and a large N--channel transistor in the pull-down circuit. The schematic of the output buffer used in the "8000· Z step gate array as well as the W step devices is shown in Figure 5. The 58 FCT OUTPUT STRUCTURES AND CHARACTERISTICS APPLICATION NOTE AN-52 Figure 5. W Step and "8000" Z Slep Output Structure The pull-up structure yields an almost resistive characteristic in the logic HIGH state. The characteristic in the logic LOW state is again non-linear due to the N~hannel transistor. These output characteristics are shown in Figures SA and S8. (Vour) o 2 3 4 300 5 250 -50 200 -100 150 (mA) (mA) -150 100 -200 50 -250 o 2 3 4 5 (Vour) Figure 6A. W & "8000" Z Step Logic "High" Characterlsliea . Figure 68. W & "8000" Z Step Logic "Low" Characterlstlea The characteristics are intended to aid the designer in developing nominal circuit simulation models so that the effect of driving different types of lumped and transmissive loads can be evaluated. In order to develop suitable models. the customer should first determine the stepping for the subject device. This information can be obtained by contacting IDTs LOGIC Marketing group. SUMMARY The output V-I characteristics are determined by the circuit implementation and transistor geometries. Output buffer schemes and the corresponding typical characteristics for FCT devices manufactured in the Z. Y and W stepping are shown in this bulletin. 59 ~ g APPLICATION NOTE AN-53 POWER-DOWN OPERATION Integrated DevIce Technology, Inc. By SureR Kodlcal power-down operation because the logic high noise immunity is not compromised. In addition. these circuits have the following design features: • The inputs (except for 110 ports) do not have a clamp diode to Vee but do have a clamp diode to ground to prevent excessive undershoot on the inputs. • The outputs have P-channel pull-up transistors to raise output high level close to Vee. The P-channel devices have a junction diode as an integral part of the geometry. To prevent this junction from floating. the cathode of this diode is tied to Vee. the most positive potential. Similarly. the N-channel transistors used in the output circuit have an integral junction diode whose anode is tied to GNO. the most negative potential. Figure 1 shows the diodes associated with inputs and outputs. INTRODUCTION In a POWER-DOWN mode. a device operates with a supply voltage that is lower than the normal operating range of 5V ± 5 for commercial grade and 5V ± 10 for military grade. This should not be confused with the "/ow-powerdissipation standby mode" of CMOS static RAMs where part of internal circuitry is shut off to reduce standby power. The power-down mode is used to either conserve power in a part of a system or to provide a battery back-up in faulttolerant systems. The devices operating in the power down mode are expected to co-exist with other devices which are connected to normal power supply rails in the same system. This bulletin discusses the use of our FCT devices in the power---. tPLHA1 OUTPUT A2 OUTPUTA1 tPLHA2 OUTPUT A2 OUTPUT An MIN DELAY tPLHAn INPUT DEVICEB OUTPUT An OUTPUT B1 \ .•• L---{S>--. tPLHB1 OUTPUT B1 OUTPUT B2 MAX DELAY tPLHB2 OUTPUTB2 OUTPUT Bm tPLHBm Package Skew, t SK(T) = tP(max Device B) - tP(min Device A) = OUTPUTBm MAX-MIN DELAY tSK(T) t PLHB2 - tPLHAn (same transition, temperature, supply voltage, loading and package type) 2521 drw 07 Package skew is the difference in propagation delay between the fastest and the slowest outputs of two or more devices for the same input and output transition. Figure 7. Package Skew t8K(T) Schematic and Timing Diagram 74 CLOCK DISTRIBUTION SIMPLIFIED WITH IDT GUARANTEED SKEW CLOCK DRIVERS APPLICATION NOTE-B2 THE IDT49FCT805 & IDT49FCT806 CLOCK DRIVERS where two generalized clock drivers are driven by a common input. The result is "n" outputs from device A and "m" outputs from device B making the same transition. The package skew is the difference in propagation delay between the slowest output of one device and the fastest output of the other device for the same transition. In this case the output An is the fastest output and the output B2 is the slowest If tPLHAn is 4.0ns and tPLHB2 is 6.0ns the package skew is 2.0ns. Certain conditions must be satisfied for the package skew specification to apply. The devices must have the same Vce, ambient temperature and be assembled in the same package type. Also each device must have equivalent loading and be of the same speed grade. Part-to-part skew is difficult to specify because it implies that the characteristics every part ever sold will operate within a window of operation. The window of operation ensures that parts that run too fast or too slow do not get sold. Figure 8 shows the measured values of package skew on several IDT49FCT80SAs under typical conditions. The maximum measured package skew from this sample is S2S picoseconds. In the IDT49FCT80S/806 data sheet this value is guaranteed to be less than 1.S nanoseconds over the commercial operating range. The IDT49FCT80S and 49FCT806 are high-speed guaranteed skew clock driver chips specifically designed to meet the clocking requirements of today's high-performance systems. The logic diagram and pin configuration of the IDT49FCT80S are given in Figure 9. The IDT49FCT806 is the inverting option of IDT49FCT80S. Skew inthe IDT49FCT80S/806 is minimized throughout the design process. Careful circuit design and layout in silicon have resulted in a pin configuration that is specifically designed for very low output and pulse skew. Independent power and ground pins reduce the amount of ground bounce and dynamic threshold shift caused by multiple outputs switching. The 1:S input to output ratio reduces the amount of capacitive loading on the previous stage which simplifies termination and reduces component count when compared to conventional solutions. The devices are optimized for both PDIP and SOIC packages. The IDT49FCT80S/806 clock drivers consist of two independent banks of drivers. Each bank drives five output buffers from a single standard TIL compatible CMOS input. The input has 200mV of hysteresis for increased immunity to system noise. Independent active low output enable pins (OEA and OE8) control each of the banks, allowing for independent control ofthe outputs. This feature may be used in applications where clock bussing or a power savings mode is required. The input INB drives the B bank as well as an output called MON (Monitor). The MON output is not controlled by OEB and therefore runs continuously. The MON signal can be used for priming phase locked loops or driving diagnostic hardware. Each IDT49FCT80S/806 has 3 ground pins and 2 VCC pins. The ground pins, GNDA and GNDB, are located in the middle of the package to minimize inductance in the ground return path. The two grounds are returns for the A and B bank output buffer and pre-driver currents. The third ground pin, GNDa (Quiet Ground), provides a ground return for the remaining circuitry. The ground pin arrangement reduces ground bounce on the outputs and noise on the thresholds of the internal logic. Since GNDA and GNDB are completely 49FCT805A PACKAGE (PART-TO-PART) SKEW VERSUS TEMPERATURE Vee = S.OV 2.0 - . - - - - - - - - - - - - - - , 1.S ~---- tSK(T) Package • Skew 1.0": (ns) SPEC----~ . O.S ···.·.·.w.·.·.·...... ,·,·,·,,·,·,·,···········,·,·,·,··,·," ..., ...... -........... ..... .... .......:... :.:.:-~:-:.:-: ~ ~ :.:~.: MEAS.SKEW/ 0.0 ~-r-"--... .-"r-r--."T.-r-r---y,-r-.,-" o 20 40 60 80 TEMP (C) 2521 drwOB Figure B. Measured Package Skew tSK(T) for SeveraIIDT49FCT805As 20· PIN DIPORSOIC OA1-0As OB1-0Bs VCCA VCGS OA1 OA2 OA3 081 OB2 OB3 GNDs GNDA OBs OB6 OA4 OAs GNDa MON7 OEA OEB INA INB PIN CONFIGURATION LOGIC DIAGRAM 2521 drw 09 Figure 9. Logic Diagram and Pinout of the IDT49FCTB05 75 CLOCK DISTRIBUTION SIMPLIFIED WITH IDT GUARANTEED SKEW CLOCK DRIVERS APPLICATION NOTE-82 isolated from each other on the die, switching effects on one bank will have minimal effects on the other bank. The independent Vec pins, VeeA and VeeB, supply power to the two banks. Each output of the lOT49FCTB05!B06 clock driver features a high current drive output buffer. These outputs can be used to drive both TTL and CMOS loads. With a typical Vol of 0.3 volts the buffer can sink64mA. For a typical Voh of 3.B volts the output buffer can source 24mA. These output buffers are optimized around the 1.5 volts switching threshold which is the standard for TTL compatible logic. These output buffers can easily meet Ihe edge rate requirements of loday's microprocessors and peripheral components. Typical edge rates for Ihe IOT49FCTB05A are 1.0 volVnanosecond for risetime and 2.0 vall/nanosecond for falltime The Switching Characteristics (Table 1) for the FCTB05N B06A show the maximum propagation delay (tPLH/HL) to be 5.B ns and the minimum propagation delay to be 1.5ns. If the skew is calculated by subtracting the minimum delay from the maximum delay the result is a number much larger than the tSK(O) spec of 700ps. How can lOT guarantee a 700ps output skew number and still have such a wide range of minimum and maximum propagation delay values? The range of values between the minimum and maximum propagation delay reflects the wide range of conditions under which the part must operate and the range of manufacturing process parameters. Consider a part that under typical conditions has a median propagation delay of 5.0ns. According to the tSK(O) specification of 700ps, each output of that driver will switch within a 5.0±0.35 ns window. If the median propagation delay drops to 4.0ns, due to variations in Vee or temperature, the specification guarantees that each output will then switch within a 4.0±D.35ns window. In the unlikely event that the operating conditions cause the median delay to drop to 1.B5ns, then all outputs will switch within a 1.B5±0.35ns window. It is important to recognize that all the devices are assumed to be operating under the same conditions. If one part is running fast because of cold temperature and high Vcc, all the other parts will be running fast as well. The following data is provided to show that the lOT49FCTB05A does indeed meet its skew specifications over the commercial operating range. Figure 10 shows the range of output skew measurements for low-to-high and high-to-Iow transitions with Vees of 4.75 and 5.25 volts. For both low-to-high and high-to-Iowtransitions, WORKING WITH THE DATA SHEET In the past, designers have used the minimum and maximum limits of a clock driver's propagation delay specifications to determine skew in their designs. With the I0T74FCT244A (tPHL min= 1.5ns and tPHL max=4.B ns) the difference between the two limits results in a3.3nswindow. With the lOT 49FCTB05! B06, subtracting the minimum from the maximum limit is no longer necessary because the skew is specified in the data sheet. However, because the IOT49FCTB05!B06 data sheet still specifies a 1.5ns minimum for propagation delay there may be some confusion as to whether or not the skew specifications are real. In the following discussion it will be shown that meeting the skew specifications is not a problem for IOT49FCTB05. SWITCHING CHARACTERISTICS OVER OPERATING RANGE IDT49FCT805/806 IDT49FCT805A/806A Com'l. Com'l. Symbol Conditions (1) Parameter Min.<2) Max. Max. Unit 1.5 6.5 1.5 5.8 ns Output Enable Time OEA to OAn, OEsto 08n Output Disable Time OEA to OAn, OEsto 08n 1.5 8.0 1.5 8.0 ns 1.5 7.0 1.5 7.0 ns tSK(O)(3) Skew between two outputs of same package (same transition) - 0.7 - 0.7 ns tSK(p)(3) Skew between opposite transitions (tPHL-tPLH) of same output Skew between two outputs of different package at same power supply voltage and temperature (same transition) - 1.0 - 1.0 ns - 1.5 - 1.5 ns tPLH Propagation Delay CL= 50pF tPHL INA to OAn, INs to 08n RL= 500n tPZL tPZH tPLZ tPHZ tSK(T)(3) NOTES: 1. 2. 3. Mln.(2) See test circuit and waveforms. Minimum limilS are guaranteed but not tested on Propagation Delays Skew guaranteed across temperature range but measured at maximum temparature only. Skew parameters apply to propagation delays only. Table 1. IDT49FCT8051806 Switching Characteristics 76 CLOCK DISTRIBUTION SIMPLIFIED WITH IDT GUARANTEED SKEW CLOCK DRIVERS APPLICATION NOTE·82 49FCT805A OUTPUT SKEW VERSUS TEMPERATURE LOW-TO-HIGH, Vcc= 4.75V LOW-TO-HIGH, Vcc= 5.25V 0.8 0.8 0.7 tSK(O) SPE 0.7 Output 0.6 Skew (ns) 0.5 Output 06 Skew . (ns) 0.5 0.4 0.4 0.3 tSK(O) SPE 0.3 0 20 40 TEMP (C) 60 80 0 40 TEMP (C) 60 80 HIGH-TO-LOW, Vcc= 5.25V HIGH-TO-LOW, Vcc= 4.75V 0.8 0.8 tSK(O) 0.7 tSK(O) 0.7 20 Output 06 Skew . (ns) 0.5 Output 06 Skew . (ns) 0.5 0.4 0.4 0.3 0.3 0 20 40 TEMP (C) 60 0 80 20 40 TEMP (C) 60 80 2521 drw 10 Figure 10. Measured Output Skew tSK(O) for SeverallDT49FCT805As over the Operating Range the graphs show that output skew is maximum at hot temperature (70°C). In each case the skew is well below the data sheet specification of 700ps. Figure 11 shows the range of pulse skew measurements with Vees of 4.75 and 5.25 volts. The measured pulse skew peaks at hot temperature and is slightly greater for a Vee of 4.75V. The measured performance is safely within the data sheet specification of 1 .Ons. Figure 12 shows the range of package skew measurements for low-to-high and high-to-Iow transitions with Vees of 4.75 and 5.25 volts. For both low-to-high and high-to-Iow transitions the skew peaks at hot temperature with minimal differences between low and high Vee. Again the measured performance easily meets the data sheet specification of 1.5ns. 49FCT805A PULSE SKEW VERSUS TEMPERATURE Vcc = 5.25V 1.5.....-------------, Vcc =4.75V 1.5.....--------------, Pulse 1.0 - t + - - - - tSK(P) SPEC----~ .................................................:.:...:.:-:..:.:.:.:.:... Pulse 1.0 - f + - - - - tSK(P) SPEC ----~ Skew .................................-.. :.:.:.:.:.:~.:.:.'I.•••: ..... ••••: Skew (ns) (ns) 0.5 MEAS.SKEW/ :.:~.:.:.:.:-: 0.5 0.0 +--,-..-..--.--.--.--.--.--.--.-~ o 20 40 60 80 TEMP (C) MEAS.SKEW/ 0.0 -!-"-"-'---'-"-'---'-"-'--'-"'--; 60 80 o 20 40 TEMP (C) 2521 drw 11 Figure 11. Measured Pulse Skew tSK(P) for SeverallDT49FCT805As over the Operating Range 77 CLOCK DISTRIBUTION SIMPLIFIED WITH lOT GUARANTEED SKEW CLOCK DRIVERS APPLICATION NOTE-82 CLOCK DISTRIBUTION SIMPLIFIED (1), the skew between the ends of the trace approaches 15ns (38" X 0.37nS/inch). Given a 20ns cycle time (40 MHz), 14ns of clock skew implies that 70% of the clock cycle is given to clock distribution. A second approach is the clock tree shown in Figure 14. By adding a level of buffers between the clock source and the 75 loads, the capacitive loading on the buffer outputs is reduced from 750pF to 50pF and the amount of PCB trace associated with each driver is reduced to 2.5". If IDT74FCT244As are used at least three packages (8 drivers per package) will be required. Since the 244's do not specify skew the designer might assume that each device output will switch within a 3.3ns window (tPHLmax-tPHLmin= 4.8ns -1.5ns). If the output transitions at points B, C, and D occur within a 3.3ns window, then the outputs of the second level (point E) may occur within a 6.6ns window. Assuming a 20ns cycle time (50Mhz) the designer has lost 33% of the cycle time to clock distribution without even considering transmission line effects. A third approach is to use IDT49FCT805As as shown in Figure 15. Since each group of six buffers in Figure 14 can be replaced by 1/2 of an IDT49FCT805, only two devices are To show how easy it is to design with the IDT49FCT805, consider a hypothetical clock distribution system. The system has a 50Mhz clock source and must drive 75 loads. Each load is a CMOS input connected by 700 micro-strip trace at a densityof 1 load every 0.5 inches. Assume that all the inputs are positive edge triggered and the objective is to minimize skew. One approach would be to drive all 75 inputs with a single clock driver output (Figure 13). There are many problems with this approach. The first problem is the large amount capacitance associated with 75 CMOS inputs. Assuming 10pF maximum of capacitance per CMOS input, the total capacitive load is 750pF. A standard clock driver such as the IDT74FCT244A has .itPLH of 2ns/1 OOpF for loads above 50pF. If the IDT74FCT244A is used the capacitance alone adds up to 14ns of additional propagation delay. If 75 loads are distributed along a single trace, the trace length is 38 inches (75 X 0.5 inputs/inch). If the PCB trace has an intrinsic delay of 0.15ns/inch (1), the delay from point B to point C is 5.7ns (38" X 0.15 nS/inch). Using a loaded trace delay ofO.37ns/inch 49FCT805A PACKAGE (PART-TO-PART) SKEW VERSUS TEMPERATURE Vcc= 5.25V 2.0....._-------------. VCC= 4.75V 2.0....._-------------. 1.5 " M - - - - - tSK(T) SPEC----~ 1.5-k---- tSK(T) SPEC----~ Package Skew 1.0 (ns) Package Skew 1.0~ (ns) 0.5 0.0 +-_._.-r~---.-_r_.....__._r__r__.___.___1 o 20 40 60 80 TEMP (C) ."...-...:-:-:.:-.~:.:.:.:.:::~::::.::::::....;.;.::.::. ::::::::.;~::~::~::::.::::~:=:;::~:::;~::::: ····~·EAS. SKEW/ o 20 40 TEMP (C) 60 80 2521 drw 12 Figure 12. Measured Output Skew tSK(T) for SeverallDT49FCT805As over the Operating Range CLOCK DISTRIBUTION SYSTEM FOR 75 LOADS - SINGLE DRIVER CONFIGURATION- .. 40" OF 70n MICRO-STRIP TRACE CLOCK SOURCE 2521 drw 13 Figure 13. Single Driver Clock Distribution System 78 CLOCK DISTRIBUTION SIMPLIFIED WITH lOT GUARANTEED SKEW CLOCK DRIVERS APPLICATION NOTE-82 CLOCK DISTRIBUTION SYSTEM FOR 75 LOADS "CLOCK DRIVER TREE CONFIGURATION" CLOCK SOURCE rvl-----1H I ® 2521 drw 14 Figure 14. Clock Tree Distribution System Using 244'5 79 CLOCK DISTRIBUTION SIMPLIFIED WITH IDT GUARANTEED SKEW CLOCK DRIVERS APPLICATION NOTE-S2 CLOCK DISTRIBUTION SYSTEM FOR 75 LOADS - CLOCK DRIVER TREE CONFIGURATION - 2.5" OF 700. MICRO-STRIP TRACE CLOCK SOURCE rv 2521 drw 15 Figure 15. Clock Tree Distribution System Using IDT49FCTS05's so CLOCK DISTRIBUTION SIMPLIFIED WITH IDT GUARANTEED SKEW CLOCK DRIVERS APPLICATION NOTE-82 required to implement the design. Using the 1.Sns package skew specification reduces the 6.6ns skew window to a 1.5ns skew window. If 0.92Sns of loaded trace delay (2.5" X 0.37 nsf inch) is considered the maximum skew is 2.425ns including a 1st order treatment of transmission line effects. For a 20ns cycle time, the penalty imposed by the clock distribution system is reduced to 12% of the cycle time including transmission line effects. Besides reducing the size of the skewwindow of the second approach by 77%, the IDT49FCT805 increases the level of integration associated with the clock distribution tree. A significant benefit is the reduced loading on previous stages. Reduced loading helps minimize skew and makes the termination of clock lines clean and simple. The reduced chip count also saves valuable board space and simplifies the layout of the board. RECOMMENDATIONS To realize the performance benefits offered by the lOT 49FCT80S/806 clock drivers, lOT recommends the following high speed design practices: • Use low impedance power and ground planes. • Keep loading balanced and light. • Keep trace lengths short, avoiding sharp bends and discontinuities (eg. use two 45° bends vs one 90° bend). • Decouple both Vec pins with a combination of capacitors (0.1IlF and 0.011lF or 0.005IlF) for effective high frequency filtering. • Use termination for signal lines longer than 3 inches. • Only use parts of same speed grade (non-A or A speed). CONCLUSIONS SUMMARY Clock skew is an important design consideration in today's high-speed systems. For successful and reliable operation, the clock skew must be kept within an acceptably small fraction of the system clock period. The IDT49FCT80S and IDT49FCT806 simplify the design of minimum skew clock distribution networks by specifying guaranteed low-skew performance. The skew specifications allow system designers to control the clock skew at each stage of the design which simplifies the problem of meeting global system requirements. With the IDT49FCT80S/806 clock driver and a design methodologythat pays close attention to high-speed design issues, maximum system performance can be achieved without risking reliability. The following features of the IDT49FCT805/806 address clock driver skew and clock distribution problems: • Circuit design, chip layout, and pin configuration specifically designed for very low output, pulse and package skew. • Independent power and ground pins for reduced ground bounce and dynamic threshold shift. • High current drive capability for driving heavily loaded/ terminated PCB traces. • 1:5 input/output ratio for reduced loading on previous stages. • 11 outputs reduce the need for additional drivers-saves board space and simplifies PCB layout. • Multiple grounds and Vecs to minimize ground bounce effects on propagation delay and skew. • Input Hysteresis for increased immunity to system noise. • Available in SOICs for increased packing density and reduced lead inductance. REFERENCES (1) "Application Note AN-49", High-Speed CMOS Logic Design Guide, Integrated Device Technology Corp., November 1989. 81 ~ ~ Integrated Device Technology. Inc. DESIGNING WITH THE IDT49C460 AND IDT39C60 ERROR DETECTION AND CORRECTION UNITS APPLICATION NOTE AN-24 By Roben Stodleck INTRODUCTION buffers to be a latched type such as an lOT? 4FCT646 instead ofthe IDT?4FCT245shown. A family of waveforms appropriate for the bus format shown in Figure 1 is shown in Figure 2. The waveform diagrams do not include precise timing considerations which are left to the designer. In any given system. any of the buffers separating the EDC from the memory IC's may be eliminated if bus capacitance and speed considerations allow. The Error Detection and Correction (EDC) chip itself is one element of an EDC system. How it is connected to the surrounding system and controlled is left to the system designer. ~ecause there are so many design variations possible, it is Important for the designer to develop a clear idea of the target design before beginning the design process. Basic design ap~roaches and perturbations are enumerated in this application note. The details of the EDC control logic depend on the configuration of the EDC system, EDC bus topology, the nature ofthe CPU orsystem bus involved, and the nature of diagnostic hardware used. The data bus topology is highly dependent on the individual target system. This application note approaches the bus topology issue first. The advantages and disadvantages of using EDC word widths that are different from the system bus are discussed. The next topic to be covered is the use of E DC in a system with a cache. Then the operational configuration of the EDC system is discussed. This implies answering questions about how the EDC unit handles errors in a particular system is discussed. How an operating system deals with the EDC function is discussed, fOllowed by a practical discussion of some non obvious hardware topics. The final topic is memory system diagnostics and verification. An appendix includes tables and software that are useful in debugging and in writing diagnostic software for an EDC board. EDC Bus Width vs. System Bus Width Thewidth of the EDC bus and the System Bus are normally matched. However, there are valid reasons for making the EDC bus both wider or narrower than the system bus. Wide EDC words are significantly more efficient than narrower EDC word widths in terms of the amount of checkbit memory used for a given amount of data memory. The amount of check-bit memory required for 64 data bits is 8 bits if the 64 data bits are organized as one word and 14 bits if it is handles as two 32-bit words. Twenty-four bits of check-bit memory would be required for 64 data bits organized as four 16-bit EDC data words. For the purposes of speed, it would be ideal to have 8-bit EDC words for systems that do byte write operations. This woul? make it unn~~essary to ever have to read a memory locatIOn before writing a partial word on these systems. Unfortunately, eight-bit EDC words are grossly inefficient in terms of check-bit memory usage. Therefore, The EDC word widths are normally 16-bits or more. Since the EDC word widths must generally be 16-bits or more for check-bit memory efficiency, and since generalpurpose computers generally use byte or partial-word-write operations, general-purpose computers force the EDC unit to be able to process partial EDC word-width write operations. Partial word-width write operations require the E DC subsystem to execute a read-modify-write type memory cycle. Thus, the EDC controller must take over control of the memory system and execute a read before completing a partial word write. For some applications, where EDC is in use, it may be desirable to speed up processing by prohibiting partial word operations either at the hardware level or software level. Speed critical sections of code should be executed without partial-wordwrite operations. The read-modify-write EDC cycle executed during a partial-word write is identical to the EDC correction cycle executed during a read cycle when an error has occurred. The readmodify-write EDC cycle should not be confused with the readmodify write cycle executed by some CPU's. Verification of a memory system using an E DC word wider than the system word is complicated by the fact that all me~ory write ~ycles become read-modify-write cycles (Le. partlal-word-wrlte EDC cycles). Careful consideration of diagnostic procedures needs to be made during the design to avoid unnecessarily complex debugging procedures. Data Bus Topology Most contemporary CPUs execute write operations of a byte or other sub-word width types. These cause special problems for all EDC units since EDC transactions with the memory are carried out on whole width EDC words. To facilitate partial word write operations with the IDT39C60 or IDT49C460 type E DC u nits, a set of tri-state transceivers are normally required between the system bus and the EDC unit. These buffers are required to prevent bus contention between the CPU or system bus drivers and the EDC units data outputs during partial word write operations. Figure 3 shows a bus arrangement appropriate for large DRAM arrays. The need for isolation of the EDC data bus and the system bus is shown by examining the data paths, shown with white arrows. These are used by the final write operation of a partial-word write cycle. In this case, only data bits 0-7 are being written from the processor to memory. If the processor or system bus drivers can be tri-stated on byte boundaries then this set of buffers could be removed, but this is not a common Situation. Depending on the memory size, additional buffering may be required between the EDC and the memory bus proper. The buffer configuration must be determined before beginning the EDC and memory controller deSign. An appropriate general purpose bus topology is shown in Figure 1. It is common for one or the other sets ofbi-directional 1/93 ©1993 Integrated Device Technology, Inc. 82 DESIGNING WITH THE IDT49C460 AND IDT39C60 ERROR DETECT10N AND CORRECTION UNITS rr.;;8 CPU ct f- APPLlCAT10N NOTE AN-24 Ll) - - - - "" DIAG MODEo, 1 = 00 LEOIAG = X CODE IDa, 1 = 00 2917 drw 10 Figure 7. Memory correct and check-bit regenerate. Identical for the second phase of a read operation in which an error has occurred, and for a partial-word-write operation except for the state of the individual byte output enables. 91 DESIGNING WITH THE IDT49C460 AND IDT39C60 ERROR DETECTION AND CORRECTION UNITS ERROR DECIMAL SYNDROME HEX 0 53 52 0 0 51 0 APPLICATION NOTE AN-24 HEX 0 1 2 3 4 5 6 7 56 55 54 50 0 0 0 0 0 0 1 0 1 0 0 1 1 1 0 0 1 0 1 1 1 0 1 1 1 NE C4 C5 T T 30 16 32 48 C6 64 T 0 80 96 112 DECI I.1AL EQUIVALENT» 1 2 3 4 5 6 7 8 9 A B C D E F 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 0 0 0 1 1 1 1 0 0 0 0 1 1 1 1 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 CO T T 14 T M M T 1 17 33 49 65 81 97 113 C1 T T M T 2 24 T 2 18 34 50 66 82 98 114 T 18 8 T M T T M 3 19 35 51 67 83 99 115 C2 T T 15 T 3 25 T 4 20 36 52 68 84 100 116 T 19 9 T M T T 31 5 21 37 53 69 85 101 117 T 20 10 T M T T M 6 22 38 70 86 102 118 T 4 26 T 71 87 103 119 M T T 54 M 7 23 39 55 C3 T T M T 5 27 T 8 24 40 56 72 88 104 120 T 21 11 T M T T M 9 25 41 57 73 89 105 121 T 22 12 T 1 T T M 10 26 42 58 74 90 106 122 17 T T M T 6 28 T 11 27 43 59 75 91 107 123 T 23 13 T M T T M 12 28 44 60 76 92 108 124 M T T M T 7 29 T 13 29 45 61 77 93 109 125 16 T T M T M M T 14 30 46 62 78 94 110 126 M M T T 15 31 47 63 T = Two errors NE= NO ERROR Cn = check-bit error bit n M = Multiple errors n = data-bit error bit n n = decimal equivalent of the syndrome 0 T T M 79 95 111 127 2917tbi01 Table 1. 32-blt Syndrome Tables with Hex, Binary and Decimal Equivalents. 92 DESIGNING WITH THE IDT49C460 AND IDT39C60 ERROR DETECTION AND CORRECTION UNITS ERROR HEX 0 1 2 3 4 5 HEX 57 56 55 54 53 52 51 50 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 1 1 0 0 1 0 0 0 0 0 0 T 0 3 C2 4 T 5 6 0 1 1 0 T 6 7 8 9 0 1 1 1 0 0 1 0 0 1 0 1 C 1 1 0 1 1 0 E F 1 1 1 1 1 1 0 1 1 5 0 1 0 1 C4 16 C5 32 T T 17 33 T 48 14 49 C6 64 B 1 0 1 0 1 0 1 1 C 1 1 0 0 D 1 1 0 1 M M T M T 161 40 162 177 193 209 M T 66 T M 51 15 52 67 T 68 T 83 35 84 T 210 2 211 T M M T T M T 99 57 100 115 131 147 51 148 163 41 164 179 195 T T 116 63 117 132 T T 54 M T T M M T T M T 70 86 102 58 103 59 104 118 134 150 52 151 53 152 166 42 167 43 168 182 198 M T 56 72 T 165 T 212 3 213 4 214 101 T T 149 M 196 T 40 11 41 12 42 M 133 T 180 47 181 85 T T T 119 135 T T 120 136 T 197 E 1 1 1 0 F 1 1 1 1 T T M 215 231 247 T M T M T 184 200 216 5 217 6 218 T T M M T T M T 89 105 121 137 49 138 153 169 185 201 T T M T 11 27 23 28 43 13 44 59 75 T T M 106 60 107 122 T T 123 139 T T M T 154 54 155 170 186 202 1 203 44 T 171 187 T M T T M M T T M T 60 76 108 61 109 124 140 156 55 157 172 45 173 188 204 T 77 T 92 39 93 1 M T T M 29 45 61 0 13 16 14 T T M 30 46 62 T 63 78 32 79 T T 125 141 T 219 7 220 T M T 189 205 0 206 221 M M T T M M T 94 110 126 142 158 174 190 T 222 T T M 48 T T M T M 95 111 127 143 159 175 191 207 223 NE = NO ERROR T = Two errors Cn = check-bit error bit n M = Multiple errors n = data-bit error bit n n = decimal equivalent of the syndrome M M M 90 38 91 T 226 242 24 T 227 243 31 T 228 244 25 T 229 245 26 T 230 246 199 73 33 74 T T T T 58 M 224 240 30 T 225 241 183 57 17 M M 194 M 24 21 25 22 26 47 T 178 69 0 M T 130 T 87 37 88 31 T 114 53 36 T M 208 145 50 146 50 15 T 192 T T T 46 176 129 71 T T 160 T M T T 144 113 T T C7 128 M 34 8 35 36 9 37 10 38 62 112 97 56 98 55 1 A M T 0 9 1 0 0 1 81 34 82 39 1 8 1 0 0 0 T T T T 96 T 0 7 0 1 1 1 65 18 18 19 20 19 21 20 22 T 23 1 6 0 1 1 0 80 7 C3 8 12 D 4 0 1 0 0 M 10 B 3 0 0 1 1 1 9 A 2 0 0 1 0 --===-==--=-=-===-=-=====--------=-=-=-----=---=-=-=-=-----=-=--------------NE 0 CO 1 C1 2 1 1 1 0 0 0 1 APPLICATION NOTE AN-24 232 248 27 T 233 249 28 T 234 250 T M 235 251 29 T 236 252 M T 237 253 M T 238 254 M T 239 255 2917 tbl 02 Table 2. 64-bit Syndrome Tables with Hex, Binary and Decimal Equivalents. 93 DESIGNING WITH THE IDT49C460 AND IDT39C60 ERROR DETECnON AND CORRECTION UNITS APPLICATION NOTE AN-24 CB DATA CB DATA CB DATA CB DATA 0 1 2 3 4 5 6 7 8 9 A 8 C 0 E F 10 11 12 13 14 15 16 17 18 19 1A 1B 1C 10 1E 1F 28 1000F 1000 27 1000C 2B 24 10003 10024 3 C 10028 0 10027 10028 F 10022 5 A 10020 6 10021 1002E 9 2E 10009 10006 21 1000A 20 22 10005 20 21 22 23 24 25 26 27 28 29 127 10100 1010F 128 10103 124 12B 1010C 1012B 10C 103 10124 10F 10128 10127 100 10120 10A 105 10122 109 1012E 10121 106 121 10106 10109 12E 10105 122 12D 1010A 40 41 42 43 44 45 46 47 48 49 4A 48 4C 40 4E 4F 50 51 52 E 10029 10026 60 61 62 63 1002A D 2 10025 1002 25 64 101 10126 10129 10E 10125 102 10D 1012A 10100 12A 125 10102 129 1010E 10101 126 10108 12C 123 10104 12F 10108 10107 120 107 10120 1012F 108 10123 104 108 1012C 2A 2B 2C 20 2E 2F 30 31 32 33 34 35 36 37 38 39 3A 3B 3C 3D 3E 3F 53 54 55 56 57 58 59 5A 58 5C 50 5E 5F 2A 10000 26 10001 1000E 29 10004 23 2C 10008 20 10007 10008 2F 8 1002F 10020 7 1002C 8 4 10023 65 66 67 68 69 6A 68 6C 60 6E 6F 70 71 72 73 74 75 76 77 78 79 7A 78 7C 70 7E 7F 2917tbl03 Table 3. Minimal 32-check·bit to data tables tor diagnostic use. One data value is listed to generate every possible check·bit pattern. 94 -- ....- - - - - - DESIGNING WITH THE IDT49C460 AND IDT39C60 ERROR DETECTION AND CORRECTION UNITS DATA 0 1 2 3 4 5 6 7 8 9 A 8 C D E F 20 21 22 23 24 25 26 27 28 29 2A 28 2C 2D 2E 2F CB C 43 46 9 5E 11 14 58 58 17 12 5D A 45 40 F 54 18 1E 51 6 49 4C 3 0 4F 4A 5 52 10 18 57 APPLICATION NOTE AN-24 DATA 100 101 102 103 104 105 106 107 108 109 10A 108 10C 10D 10E 10F 120 121 122 123 124 125 126 127 128 129 12A 128 12C 12D 12E 12F DATA 10000 10001 10002 10003 10004 10005 10006 10007 10008 10009 1000A 10008 1000C 1000D 1000E 1000F 10020 10021 10022 10023 10024 10025 10026 10027 10028 10029 1002A 10028 1002C 1002D 1002E 1002F CB 2F 60 65 2A 7D 32 37 78 78 34 31 7E 29 66 63 2C 77 38 3D 72 25 6A 6F 20 23 6C 69 26 71 3E 38 74 CB 2 4D 48 7 50 1F 1A 55 56 19 1C 53 4 48 E 1 5A 15 10 5F 8 47 42 D E 41 44 8 5C 13 16 59 DATA 10100 10101 10102 10103 10104 10105 10106 10107 10108 10109 1010A 10108 1010C 1010D 1010E 1010F 10120 10121 10122 10123 10124 10125 10126 10127 10128 10129 1012A 10128 1012C 1012D 1012E 1012F CB 21 6E 68 24 73 3C 39 76 75 3A 3F 70 27 68 6D 1F 79 36 33 7C 28 64 61 2E 2D 62 67 28 7F 30 35 7A 2917tbl04 Table 4. Minimal 32-bit data to check-bit tables for diagnostic use. At least one data value Is listed for every possible check-bit pattern. This table is identical to Table 3 except in sequence of presentation. 95 ~ ~ PROTECTING YOUR DATA WITH THE IDT49C465 32-BIT Flow-thruEDCTM UNIT APPLICATION NOTE AN-64 Integrated Devlc:e Tec:hnology, Inc:. by Tao Lin, Gerard Lyons and Frank Schapfel INTRODUCTION high-density dynamic RAMs, typically with access times of 100 nanoseconds or more, but with four times the density of static RAMs. So, when the state-of-the-art static RAMs are 1 Megabit large, the newest density dynamic RAM is 4 Megabits. Therefore, dynamic RAMs will always provide the most cost-effective implementation for system memory. Dynamic RAMs, though, are very prone to externally induced errors. These externally induced errors are called soft errors, since they do not cause permanent damage to the memory cell. Soft errors can be induced by system noise, alpha particle and power supply surges, and will cause random data bits to be flipped from "1" to "0", or vice versa. Although these soft error occurrences may be rare and inconsequential when using small amounts of DRAMs, large DRAM arrays are much more error prone. Also, as seen in Figure 1, larger DRAM components are much more susceptible to soft errors by virtue of their smaller memory cell size. Hardware errors may also occur on system memory boards. These hard errors occur if one RAM component or RAM cell fails and is stuck at "0" or stuck at "1 n. Although less frequent, hard errors may cause a complete system shut down. A TIME FOR ERROR-FREE MEMORIES With the advent of high-performance 32-bit RISC and CISC microprocessors, general purpose computing across a wide spectrum of applications software is now easily accessible on a desktop. We can now draw on computer resources which are very sophisticated, multi-task systems with distributed processing power, and we no longer must rely on the centralized mini-computers and mainframes for processing horsepower. Both the technical and the commercial computing environments demand the insatiable hunger for processing power. This increasing demand for sophisticated applications software requires more system memory on a local level. Tightly coupled microprocessors and cache memory are designed for optimized processing throughput, but the cache memory is no substitute for system memory. Cache memory is typically composed of very high-speed static RAMs, with access times of 35 nanoseconds or less. System or main memory is almost always comprised of slower but very Typical Error Rate % Per 1,000 Hours • SOFT ERRORS DUE TO ALPHA PARTICLES ONLY o HARD ERRORS 0.1 0.01 0.001 0.0001 Density Bits/Chip Figure 1. Typical Error Rates The lOT logo is a registered trademark and F'ow~thruEDC is a trademark of Integrated Device Technology, Inc. ©1993 Integrated Device Technology, Inc. 96 "'"II ~~ coPROCESSOR Data 14 I~~ ~:::! ~I .lo~ :r ~< Co me: 0:0 Data 1..- ~ I ~O '--(, e:~ ii!:l> -I::E ::::j Buffer 1 CPU ::J: -I ::J: m Address Tag is I I I Write Buffer .... U) I ~I I-Cache & D-Cache r- I L,J I Write I Buffer I Data Main Memory i! U) Address ~ '" ~11111!rl~-~1'llll'!lil 1!'!I!;I:!I ~;I !:rl:l il'l !i lfi"Jil~l!il - ~Data Check Bit Memory Address ~I DRAM Controller l> Timing Control Logic ~ C Crystal Oscillator Figure 2. A Typical Architecture of High-performance RiSe or else Systems ~ z ~ iii ~ PROTECTING YOUR DATA WITH THE IDT49C465 32-BIT Flow-thruE[)CTM UNIT APPLICAnON NOTE AN-64 ERROR CORRECTION TO THE RESCUE The operation of an EDC device can be generally split into:(1) generation of a coded word based on the data-word being written to memory. This coded word is called The CheckBit Word. This operation is called Generate; (2) detection of errors in a data-word read from memory by comparing the corresponding check-bit word read from memory and a newly generated check-bit word (based on the data-word read from memory) and if possible correcting this error. The comparison of these two check-bit words (an exclusive-or (XOR) function) produces the so-called Syndrome Word. This operation is called Detect/Correct. The coding scheme employed in lOT's EDC devices is a modified Hamming Code. For each data-word written to memory, a coded pattern, or check-bit word, is appended to the date-word. The new word (the data-word plus the checkbit word) can be termed a valid code. The modified Hamming Code establishes a Distance-of-4 between one valid code and another. This means that to go from one valid code to another, 4-bits have to change. It can be shown that a Distance-of-4 code enables you to detect all Single and Double-Bit errors and correct all Single-Bit errors. To implement a Distance-of-4 code on a 32-bit data-word, a 7-bit check-bit word must be appended. For a 64-bit word, a 8-bit check-bit word must be appended. The Hamming Code algorithm to generate a check-bit word from a 32-bit data-word or a 64-bit data-word can be found in either IDT49C460 data sheet or IDT49C465 data sheet. A scheme exists that not only is able to detect soft and hard errors, but is capable of correcting the erroneous bits. This scheme is implemented by a family of error detection and correction chips from Integrated Device Technology. Using a modified Hamming code, developed at AT&T Bell Labs, all single-bit errors may be detected and corrected, while all two-bit and mostthree-bit errors can be detected.IDT pioneered EDC chips, using CMOS technology in 1986, after recognizing the importance of large DRAM memory arrays in distributed computing. TYPICAL ARCHITECTURE OF HIGH-PERFORMANCE RISC/CISC SYSTEMS Figure 2 shows a typical architecture of high-performance RISC or CISC systems which have the following features: (1) high-speed cache memory (separate or common, Instructioncache and Data-cache) for fast access to frequently used instructions and data, (2) write and read buffers to handle the mismatch between the high-speed CPU and the slow-speed main memory and (3) high-speed flow-thru EDC unit to insure data integrity. While most high-performance computer systems in current market have the first and second features, the third feature is becoming more attractive and important when the main memory space grows and the memory word-length increases. Certainly, using an EDC unit is an effective way to improve the system reliability. EDC ARCHITECTURES AND WORD-LENGTH There are two basic architectures for EDC operation: flowthru and bus-watch. IDT provides a full line of EDC devices to support 16-bit and 32-bit bus-watch architectures and 32-bit and 64-bit flow-thru architectures, as shown in Table 1. GENERAL EDC OPERATION The basic function of an EDC device is to check the integrity of data being read from a memory system, flag an error if one has been detected and if possible correct that error. The IDT family of EDC devices implements this function using the same general principles, with some variations from device to device. Part Number Architecture Word-length Comment IDT39C60 Bus-watch 16-bit Cascadable up to 64·bit using 4 devices IDT49C460 Bus-watch 32-bit Cascadable up to 64-bit using 2 devices IDT49C465 Flow·thru 32-bit Cascadable up to 64-bit using 2 devices IDT49C466 Flow-thru 64-bit Table 1.IDT EDC Product Line 98 PROTECTING YOUR DATA WITH THE IDT49C465 32-BIT Flow-thruEDCTM UNIT APPLICATION NOTE AN-64 ------------------------------~ WRITE (GENERATE) READ (DETECT and CORRECT) Main Memory Main Memory CPU DRAM DRAM , Error IDT49C460 Check Bit Check Bit DRAM DRAM EDC L ______________________________ , ------------------------------- ,, (a) Common 1/0 Memory System ,------------------------------,, WRITE (GENERATE) READ (DETECT and CORRECT) - r-- • CD CPU ;t'\r > .~ JL ~ I- "- 'L~' .~ CPU ")I '§ "- CD ..... CD DRAM ~ to - Main Memory ~ -'~" ... c ~ ~ L-- r ~i--,111''- Main Memory DRAM j Error Data " - Error ;.. IDT49C4'e "EDC "I Data ... SC CB ... '" IDT49C460 Check Bit '--- EDC "- tsc '" \.. DRAM CB " Check Bit ..... DRAM 1- . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . , (b) Separate 1/0 Memory System Figure 3. Basic Configurations Using a Bus-watch EDC Architecture 99 PROTECTING YOUR DATA WITH THE IDT49C465 32-BIT Flow·thruE[)CTM UNIT APPLICATION NOTE AN-64 WRITE (GENERATE) READ (DETECT and CORRECT) Main CPU Main 1/1----''' Memory 1/'--.....1'''-1 Memory CPU DRAM DRAM Error Error Check Bit Check Bit DRAM DRAM (a) Common 1/0 Memory System READ (DETECT and CORRECT) WRITE (GENERATE) Main Memory Main Memory DRAM DRAM Error Error Check Bit Check Bit DRAM DRAM (b) Separate 1/0 Memory System Figure 4. Basic Configurations Using a Flow·thru EDC Architecture 100 PROTECTING YOUR DATA WITH THE IDT49C465 32·BIT Flow·thruEDCTM UNIT APPLICATION NOTE AN·64 System Data Bus 32- It 64·bit Memory Data Bus 32- It SD SD IDT49C465 Upper 32·bit 32·blt 8-bit Partial Checkblt (Generate Only) CBO Check Bit In CODE ID SYO =10 8-blt MD IDT49C465 Lower CBI PCBI Check Bit Out CODE ID = 11 CBI SYO PCBI IERR IMERR CBO 8- it (a) Cascading Flow·thru EDC IDT49C465 Data Bus ( 32-bit 64-bit I~ ~ ~ ----...s-lI Data IDT49C460 Upper IDT49C460 Lower 8-bit "" > /~ ... ~ ., CODE ID =10 CB ~ 32·bit 32·blt Check Bit In ~/E RR Data .1.. 8-bit SC ., CB 3-State Buffer 0-01 (b) Cascading Bus-watch EDC IDT49C460 Figure 5. 64-bit Configurations by Cascading Two 32·bit EDC Units 101 Check Bit Out CODE ID = 11 SC t-- . rv PROTECTING YOUR DATA WITH THE IDT49C465 32-BIT Flow-thruEDcm UNIT APPLICATION NOTE AN-64 BUS-WATCH ARCHITECTURE A bus-watch EDC such as the IDT49C460 has a single data bus. The basic configurations, using the IDT49C460 for common lID memory and separate lID memory, are illustrated in Figure 3. During a write (store) operation, the CPU sends data to the main memory. At the same time the data goes to the EDC unit, which then generates the check bits and stores them in the check-bit memory. On the other hand, during a read (load) operation, the data from the main memory and the check bits from the check-bit memory first go to the EDC unit. Based on the information carried by the check bits, the EDC unit can detect all single-bit and some multiple-bit errors, and correct all single-bit errors. The corrected data is then sent to the CPU. FLOW-THRU ARCHITECTURE In contrast to a bus-watch EDC, a flow-thru EDC such as the IDT49C465 provides two data buses: a system data (SD) bus and a memory data (MD) bus. The dual-bus architecture improves the throughput of the EDC operation and simplifies the interface between the CPU system bus and the memory bus. The basic configurations using the IDT49C465 for common lID memory and separate lID memory are illustrated in Figure 4. In the common lID configuration, during a write (store) operation, the data from CPU flows through the EDC unit and is written to the main memory. When the data flows through the EDC, the check bits are generated and stored into the check-bit memory. During a read (load) operation, the data from the main memory enters the EDC unit through the MD bus while the check bits enter the EDC unit through the CSI bus. The EDC unit then detects any errors and loads the corrected data to the CPU through the SD bus. In the separate lID configuration, during a write (store) operation, the data from CPU are directly sent to the main memory. At the same time, the data is sent to the EDC unit through the SD bus. The EDC unit then generates the check bits and stores them into the check-bit memory. During a read (load) operation, the data from the main memory enters the EDC unit through the MD bus while the check bits enter the EDC unit through the CBI bus. The EDC unit then detects any errors and loads the corrected data to the CPU through the SD bus. ......._ _ _ _• • = 32-bit wide Busses Syst. Mux Byte Irr==t===~rliii~~ L..JL-"''------' =rnI==L..--4K::J ~ !\mE MDO-31 MLE L-----..... ~~~SYOO_7 ERROR MERR SYNCLK ~,]!,)S' CBO O_7 C8OE" >------------==:t11-11lliJltlRJill••----tstt:.::::J PCBI0_7 SCLKENr-~------------------------~ CLEAR CB1 0 _7 CODE ID 1,0 MODE 2-0 Figure 6. Block Diagram of IDT49C465 102 PROTECTING YOUR DATA WITH THE IDT49C465 32·BIT Flow·thruEI)C1'M UNIT APPLICATION NOTE AN·64 CASCADING 32·BIT EDC DEVICES FOR 64·BIT MEMORY SYSTEMS As mentioned in the previous section, for a 32-bit data word, 7 check bits are necessary, while for a 64-bit data word, check bits are needed. Although the IDT49C460 and the IDT49C465 are both 32-bit EDC units, they have an 8·bit output·bus for output of generated check-bits to memory and an 8·bit input bus to read back check-bit from memory. In this way, they can be cascaded to support 64-bit applications. In the 32-bit mode, only 7 bits of the check-bit input and output buses are used, while in the 64-bit mode, all a bits are used. Figure 5 shows how two I DT 49C465s (or two IDT 49C460s) can be cascaded to build a complete 64-bit EDC unit. In the cascaded 64-bit mode, the EDC operation can be broken into two stages; a lower 32·bit stage and an upper 32·bit stage. For the IDT49C465 (see Figure 5a), a general description of the EDC operation is discussed below. 1. Generation starts by generating a Partially Generated Check·bit Word in the lower slice, based on the lower 32-bit of the 64-bit data-word, and sending this to the upper slice. The upper slice combines the Partially Generated Check-bit word from the lower slice, with its generated check-bit word (based on the upper 32-bit of the 64-bit data-word), to form a final check-bit word. Thus, the source of check-bit in a cascaded system is the upper-slice device. 2. Detection/Correction starts in the lower·slice where the check-bits from memory are input, as well as the lower 32-bit of the 64-bit data-word. Here the inputted checkbits are compared with the newly generated check-bits (based on the lower 32-bit of DATA-word) using an XOR function to produce a Partial Syndrome Word, which is passed onto the upper-slice device. At the same time, in the upper slice the upper 32-bits of 64-bit data-word is used to generated a so-called Partial Check·Bit Word which is sent to the lower slice. So now we have in both the upper and lower slice devices almost simultaneously two pieces of data; the Partial Syndrome Word (generated in the lower-slice) and the Partial Check·Bit Word (generated in the upper-slice). In each slice these two pieces of data are XOR'd to produce a Final Syndrome Word which is used to detect and correct errors on the 64-bit data word. a The IDT49C460 carries out Detect!Correct slightly differently (see Figure 5b), namely, the Partial Syndrome generated in the lower-slice is sent to the upper slice, then the Final Syndrome is generated in the upper-slice and this FinalSyndrome is now fed back to the lower-slice. Thus Detect! Correct in the IDT49C460 employs a serial approach, whereas the IDT49C465 uses a faster paralleled approach. Moreover, in the IDT49C460 case, an additional tri-state buffer such as the IDT74FCT244 is needed while in the IDT49C465 case, no additional external logic is needed. OVERVIEW OF THE IDT49C465 ARCHITECTURE The IDT49C465 architecture is an evolutionary development on the IDT49C460 EDC device. The IDT49C460 is a single-bus 32-bit EDC cascadable to 64-bits. The IDT 49C465 draws on this basic architecture to provide a dual-bus or flow-Thru 32-bit EDC cascadable to 64-bits. Figure 6 shows a block diagram of the IOT49C465; the key difference between the lOT 49C460 and the lOT49C465 is the presence of a second 32-bit DATA bus to provide the flow-thru path for data through the device. DATA BUSES The System Data Bus, or SO Bus, is a 32-bit bi-directional bus. Data is written to the EDC using this bus for Check·Bit Generation, so that when a data-word is written to memory the corresponding check-bits are written simultaneously. Also when a data-word read from memory is corrected, the corrected data-word is read from the SO Bus by the system processor. The SD Bus has associated with parity-checking and generation and also separate byte enables on the SD Bus' output buffers so that Partial Byte operations can be supported. The Memory Data Bus, or MD Bus, is a 32-bit bi-directional bus. Data written from the system processor through the SD Bus can be written to memory using this bus. When the processor is reading a word from memory, the data word is read in through the MO Bus and the corrected data word (depending on the status of the data) is sent to the processor through the SO Bus. 103 PROTECTING YOUR DATA WITH THE IDT49C465 32-BIT Flow-thruEDC"" UNIT APPLICATION NOTE AN-64 EXPANSION BUSES The IDT49C465 has four a-bit buses that are an integral part in the Detect/Correct path for both a 32-bit EDC system and a 64-bit EDC system. CBI(7:0) 1. When the IDT49C465 is operating as a 32-bit EDC system or is the lower 32-bit slice in a 64-bit E DC system, this a-bit bus is the input port for Check-Bits read from Memory. 2. When the IDT49C465 is operating as the upper 32-bit slice in a 64-bit EDC system, this bus is the input port for partial Syndromes from the lower slice. PCBI(7:0) 1. When the IDT49C465 is operating as a 32-bit EDC system, this bus is unused. 2. When the lOT 49C465 is operating as the lower 32-bit slice in a 64-bit EDC system, this a-bit bus is the input port for Partial Check-Bits read from the upper slice. 3. When the IDT49C465 is operating as the upper 32-bit slice in a 64-bit EDC system, this bus is the input port for Partially Generated Check-Bits from the lower slice. CBO(7:0) 1. When the IDT49C465 is operating as a 32-bit EDC system or is the upper 32-bitslice in a 64-bit EDCsystem, this a-bit bus is the output port for Check-Bits being written to Memory 2. When the lOT 49C465 is operating as the lower 32-bit slice in a 64-bit EDC system, this bus is the output port for Partially Generated Check-bits being sent to the upper slice. SYO(7:0) 1. When the IDT49C465 is operating as a 32-bit EDC system, this a-bit bus outputs the Final Syndrome word associated with the Detect/Correct logic. 2. When the lOT49C465 is operating as the lower 32-bit slice in a 64-bit EDC system, this bus is the output port for Partial Syndrome word being sent to the upper slice. 3. When the IDT49C465 is operating as the upper 32-bit slice in a 64-bit EDC system, this bus is the output port for Partial Check-bit word being sent to the lower slice. Operating Modes The IDT49C465 has 3 mode control pins, MODEID(2:0), which enable the user to select which mode the part is operating in. These modes are summarized in Table 2. MODE DESCRIPTION 000 ERROR DATE MODE X01 DIAGNOSTIC OUTPUT MODE X10 GENERATE-DETECT MODE 100 CHECK-BIT INJECTION MODE X11 NORMAL OPERATING MODE Table 2. IDT49C465 Operating Modes 104 PROTECTING YOUR DATA WITH THE IDT49C465 32-81T Flow-thruEDCTM UNIT APPLICATION NOTE AN-64 ERROR DATA MODE (000): In this mode the contents of the Error-Data Register are output uncorrected on the SO Bus. The Error-Data Register is a 32-bit register which gets latched under the following conditions: 1. an error condition has been detected by the EDC-ERROR) and is asserted low, 2. the on-chip 4-bit Error-Counter reads zero 0000 (i.e. no error has occurred since the last clear operation), 3. the input signal, SCLKEN), is held low so that the diagnostic clock, SYNCLK, is enabled, 4. the diagnostic clock, SYNCLK, undergoes a LOW-to-HIGH transition. Data is latched into the Error-Data Register from the output of the Memory Data Latch when and only when these conditions are met. Thus, the Error Data Register contains the Memory Data word corresponding to the first error since a clear operation (assuming SYNCLK has been run continuously). If the Error Data Register has just been cleared, then output of the contents of this register will provide a source of zero-data if that is required. BIT16:23 The contents of the Syndrome Register, which is an 8-bit register within the Diagnostic Unit, is output on the SO Bus at these positions (LSB at bit 16, MSB at bit 23). The Syndrome Register gets latched at the same time as the Error Data Register and contains the Final Syndrome corresponding to the first error to occur since a clear operation. BIT 24:27 The contents of the on-chip 4-bit Error-Counter are output on the SO Bus at these positions (LSB at bit 23 and MSB at bit 27). The Error-Counter which gets clocked under the following conditions: 1 . An error condition has been detected by the EDC, i.e. xto(ERROR) is asserted low, 2. the on-chip 4-bit Error Counter does not read 1111 (or F HEX), therefore, not more than 16 errors have occurred since the last clear operation, 3. the input signal, xto(SCLKEN), is held low so that the diagnostic clock, SYNCLK is enabled and 4. the diagnostic clock, SYNCLK, undergoes a LOWto-HIGH transition. DIAGNOSTIC OUTPUT MODE (X01) : In this mode a 32-bit Diagnostic Word is output on the SO Bus. The structure of this word is outlined in Figure 7. BITS 0:7 BIT 8:15 Error Type The output of the check-bit multiplexer is output directly on the SO Bus at these positions (LSB at bit 0 and MSB at bit 7). Whatever is being forced on the PCBI(7:0) input pins is output on the SO Bus at these positions (LSB at bit 8 MSB at bit 15). Test Error Counter Syndrome Bits Register Contents The Error Counter will tell the number of errors that have occurred since the last clear operation. BIT28:29 Reserved BIT30:31 The contents of the Error Signal Register, which is a 2-bit register is output on the SO Bus at these positions (SB at bit 30 and MSB at bit 31 ). Bit 0 and bit 1 of the register are set if a Multiple Error has been flagged and bit 1 only is set if a Single Error has been flagged at the same time and under the same conditions as the Error Data and Syndrome Registers are latched. Partial Checkbits from Input Pins 3 Byte 0 Byte 2 Byte 1 2311121 12° 716151413121110 716151413121110 716151413121110 -128 27 24 23 16 15 8 7 Byte MiS 31 Checkbits Figure 7. Output Syndrome/Diagnostic Word 105 PROTECTING YOUR DATA WITH THE IDT49C465 32-BIT Flow·thruEDcn< UNIT APPLICATION NOTE AN-64 GENERATE·DETECT MODE (X10) : In this mode, detection and generation take place but no correction. Data whether correct or not, passes thru the device from the MD Bus to the SD Bus. NORMAL OPERATING MODE (X11): This is the mode where normal detection/correction and generation takes place for a single-slice device (32-bit EDC system) or for the upper and lower slices in a cascaded 64-bit EDCsystem. CHECK·BIT INJECTION MODE (100): In this mode the check-bit multiplexer enables bits cl:7 from the output olthe System Data Latch to be fed into the EDC as a check-bit input, normal correction is activated. This is a very useful capability for carrying out a diagnostic check on the detect/correct path of the EDC. PARITY FOR THE SYSTEM BUS The IDT49C465 supports byte parity on the SD Bus, with the polarity of the parity ( even or odd) selectable using the input pin PSEL. If PSEL is low, then parity (both checking and generation) will be even. If SPSEL is high, then parity will be odd. The part has 4 parity I/O lines one for each byte of the SD Bus and a parity error signal, PERR), which flags a parity error on in-coming data by being asserted low. PARTIAL BYTE WRITE AND READ·MODIFY·WRITE CAPABILITY The IDT49C465 supports, through a number of features, Partial Byte Writes and Read-Modify-Writes cycles. Firstly the SD Bus has 4 Byte Enable signals associated with it, BE(3:0), these input lines provide, in conjunction with SOE), separate output enable control on each byte of SD bus data. The BE bus is also the control input to the Sys-Byte-Mux, this mux enables mixing on a byte-by-byte basis of data from the SD latch (A input to mux) and from the Pipe-Line latch (B input to mux). So, for example, if the processor wanted to do a Partial Write or Partial Store of a byte (byte position 3) to a memory location byte position 3 the following sequence would occur: (1) read the memory location in question through the MD Bus and correct if possible or necessary. The corrected data-word will be latched into the Pipe-Line latch, (2) the byte to be written is latched into the SD Latch at byte position 3, all other byte are undetermined, (3) Now we have both pieces of data necessary to construct the 32-bit word to be written to memory and (4) BE(3) is held low and all other BEs are held high. Thus the output of the Sys-byte-Mux is the correctly constructed 32-bit word which is then written to memory through the MD Bus with it's corresponding check-bits. 64-BIT GENERATE A very useful and ultimately cost-saving measure associated with the IDT49C465, is its 64-bit generate mode. If the CODE ID of the IDT49C465 is set at 01, the part is configured as a single-slice 64-bit generate EDC. While operating in this mode, the lower 32-bit of the 64-bit data word is input on the MD bus pins and the upper 32-bit of the 64-bit data word is input on the SD bus. The a-bit generated check bits are output on the CBO bus. In 64-bit generate mode, the EDC is dedicated to check-bit generation, all other features are disabled. Because the 64-bit generate is executed in a single slice, very fast generate speed can be achieved (15ns as opposed to 30ns in a two-slice 64-bit cascaded system). This feature can also help reduce part count. In 64-bit memory systems, it is common to use 4 32-bit EDC devices; 2 for detect/correct and 2 for generate. With the 64-bit generate capability, this part count is reduced from four to three. WHY FLOW·THRU EDC To fully understand the advantages of the IDT49C465 flow-thru EDC over the IDT49C460 bus-watch EDC, it is necessary to first know the architectural differences between the IDT49C465 and the IDT49C460. Figure a compares the simplified internal architectures of the two chips. As compared with the IDT49C460, the IDT49C465 has the following unique features: • • • • • • • • Dual data buses Dual check-bit generators: one for SD Bus and the otherfor MDbus Independent check-bit generation path Independent error detection/correction path Dedicated syndrome output Dedicated check-bit output Output pipeline latch Parity check/generation These features greatly simplify the interface of the EDC unit with the system data bus and the memory data bus, and thus can considerably improve the system performance. Generally speaking, in a single bus EDC architecture like the IDT49C460, the data bus connects to both the processor and the memory system. Thus, in a normal correction cycle, data is read into the EDC from memory through the data bus, and the data is corrected. Then, the data bus is enabled as an output and the corrected data is sent to the processor. Therefore, during a correction cycle, the data bus must be turned around from being an input to being an output. Consequently, a single bus architecture has inherent delays associated with the enable/disable times of the data bus output buffer. On the other hand, separate data buses, as in the IDT49C465, allow us to dedicate buses to a specific direction of data flow and, as such, is a superior architecture. In a 32-bit system using common I/O memory, the dual bus architecture of the IDT49C465 allows direct interface of the flow-thru EDC unit with the system data bus and the memory data bus, as shown in Figure 4a. On the other hand, if the IDT49C460 is used, then two sets of transceivers are needed to buffer both system data bus and memory data bus to the single data bus of the IDT49C460, as shown in Figure 3a. Similarly, in a 32-bit system using separate I/O memory, the dual bus architecture of the IDT49C465 allows direct interface of the flow-thru EDC unit with the memory data bus. Only a single set of transceivers is used to connect the CPU 106 PROTECTING YOUR DATA WITH THE IDT49C465 32-BIT Flow-thruEDCfM UNIT APPLICATION NOTE AN-64 system data bus to the EDC unit, as shown in Figure 4b. On the other hand, if the IDT49C460 is used, then a set of transceivers and a set of buffers are needed to hook up both system data bus and memory data bus with the single data bus of the IDT49C460. In particular, the multi-bus architecture and the independent error generation and detection/correction paths of the IDT49C465 provide significant performance improvement in a 64-bit system using two cascaded EDC units. Figure 9 shows the internal data paths of cascaded IDT49C465s and cascaded IDT49C460s during a read (error detecVcorrect) operation. In the IDT49C465 case, the entire error detecVcorrect path can be divided into two steps. In the first step, the lower 32-bit unit generates the partial check bits from the lower 32-bit data, and then compares the partial check bits with the original check bits to generate the partial syndrome bits. At the same time, the upper 32-bit unit generates the partial check bits from the upper 32-bit data. Then, the partial syndrome bits from the lower unit and the partial check bits from the upper unit are exchanged between the two units. In the second step, both lower and upper units generate the final syndrome bits independently and then correct errors in the lower 32-bit data and the upper 32-bit data, respectively, in parallel. Therefore, the total delay time is the sum of MD-to-SYO plus CBI-to-SD. On the other hand, in the IDT49C460 case, the entire error detecVcorrect path can be divided into three steps. In the first 107 step, like in the IDT49C465 case, the lower 32-bit unit generates the partial check bits from the lower 32-bit data, and then compares the partial check bits with the original check bits to generate the partial syndrome bits. At the same time, the upper 32-bit unit generates the partial check bits from the upper 32-bit data. However, in contrast to the IDT49C465 case, only the partial syndrome bits from the lower unit are sent to the upper unit. In the second step, the upper unit compares the partial check bits from the upper 32-bit data with the partial syndrome bits from the lower unit to generate the final syndrome bits. Then, the final syndrome bits are sent back to the lower unit. Finally, in the third step, the lower unit and the upper unit correct the errors in the lower 32-bit data and the upper 32-bit data, respectively. Consequently, the total delay time is the sum of DATA-to-SC plus BC-to-SC plus CB-to-DATA, which is much longer than the delay in the IDT49C465 case. Moreover, since the IDT49C460 has only one check bit input bus, an external octal tri-state buffer is needed to multiplex the original check bits and the partial check bits. Based on the above discussion, Table 3 summarizes the performance comparison between the IDT49C465 and the IDT49C460D, the fastest version of the IDT49C460. It can be seen that in most situations, the IDT49C465 has significant speed advantage over the IDT49C460. PROTECTING YOUR DATA WITH THE IDT49C465 32-BIT Flow-thruEDCTM UNIT APPLICATION NOTE AN-64 ( • • • • • •IIIIqIlJlll~ syndrome Features Of IDT49C465 ) -Dual Data Bus Architecture generator -Dual CheckBit(CB) Generators -Independent CB Generate Path -Independent Error DetecVCorrect Path -Dedicated Syndrome Output -Dedicated CB Output -Output Pipeline Latch -Parity Check/Generate -1 Off-chip Feedback for 64-bit Error Correct -144-pin PGA - c:::J Path For Detect and Correct Only Path For Generate, Detect and Correct (a) Simplified Block Diagram of IDT49C465 EDC Unit ( syndrome generator ()) Features Of IDT49C460 ) -Single Data Bus Architecture E -Single Checkbit Generator "0 c>,:l - -Shared Syndrome and CB Output eSa. CI) -2 Off-chip Feedback for S4-bit Error Correct a I11III1IIIIIIIIliliiii1IIIIIIIIliliiiil1li -S8-pin PGA - Path For Detect and Correct Only Path For 1""""""',',',',1 Generate Only c:::J Path For Generate, Detect and Correct (b) Simplified Block Diagram of IDT49C460 EDC Unit Figure 8. Internal Architecture Differences Between IDT49C465 and IDT49C460 108 PROTECTING YOUR DATA WITH THE IDT49C465 32·BIT Flow·thruEDCTM UNIT APPLICATION NOTE AN·64 SD32-63 MD32·63 Generate Final Upper 32-Bit CODE ID = 11 CBIO-7 Lower 32-Bit CODE ID= 10 Syndrome Syndrome SDO-31 MDO-31 (a) 64·bit Error Detect/Correct Path of Cascading IDT49C465 DATA32-63 DATA32-63 - ----------------1IIj .... Upper 32-Bit CODE ID = 11 Bits Generate Partial Syndrome DATAO-31 Lower 32-B it CODE ID= 10 DATAO-31 (b) 64-bit Error Detect/Correct Path of Cascading IDT49C460 Figure 9. Comparison of 64-bit Error Detect/Correct Path Between IDT49C465 and IDT49C460 109 "'-V 'l"~ ~-l -lm 6'g :ilz "'1:) Common 1/0 :r< Co Separate 1/0 me c:a 64-bit Cascade 32-bit (1) Read (1) MD->SD Write 20ns SD->CBO Read 15ns Read(1) Write MD->SYO 15ns SD->CBO CBI->SD 20ns PCBI->CBO 15ns 15ns MD->SD FCT2~2) Read(1) Write 20ns FCT2452) 5ns SD->CBO 5ns MD->SYO 15ns CBI->SD FCT24!) IDT49C 465 20ns o IDT49C 4600 15ns 35ns 40% Faster 26% Faster 34% Faster 28ns 19ns 47ns FCT2~) D->D FCT2452) 5ns 18ns FCT2J2) D->SC 5ns 5ns 14ns FCT2J2) D->SC - _L- ____ NOTES: 1. The EDC units perform correction always. 2. FCT245 is high-speed bidirectional transceiver. 25ns 20ns 30ns 5ns 14ns 11ns CB->D 12ns FCT2J2) D->SC CB->SC FCT24~2) 14ns D->D 11ns FCT2J2) 19ns 5ns 18ns FCT2J2) D->SC 5ns 14ns 5ns FCT2452) FCT24S2) 5ns 20ns SD->CBO 15ns ! 5ns PCBI->CBO 15ns 35ns -_._-- 30ns ------- 5ns FCT2J2) 5ns D->SC 14ns D->SC 14ns CB->SC 11ns CB->SC 11ns CB->D 12ns FCT245 2) Table 3. Performance Comparison 15ns 47ns 5ns ------ 5! :i!m 18% Faster 28ns 5ns §~ =i::e Write 40ns 12% Faster CB->SC FCT2J2) -~~ 30ns ~c 64-bit Cascade 32-bit - 5ns , S o z ~ iil i PROTECTING YOUR DATA WITH THE IDT49C465 32-BIT Flow-thruEDCTM UNIT APPLICATION NOTE AN-64 CONCLUSIONS Whether designing a correct always (flow-thru) EDC or bus-watch EDC memory systems, lOT offers a high performance solution for keeping memories error free. The key system benefit for using EDC is the continuous system operation, even with hard or soft errors occur. The key benefit for using a flow-thru EDC is the reduced memory design time when performing the correct always function, and improved performance for 64-bit memory systems. 111 ~ i,;;) APPLICATION NOTE AN·94 ERROR DETECTION AND CORRECTION WITH IDT49C466 ~ Integrated Devic;e Tedtnology, Inc;. By Anupama Hegde INTRODUCTION It is widely accepted that system failures and down time claim a heavy toll in terms of cost and performance. When a system crashes, unrecoverable data may be lost. Even in the best case, the user suffers a great deal of inconvenience. Erroneous data or loss of data integrity is a major cause of system failure. Hence system designers are constantly trying to minimize the occurrence of errors in their systems. Data errors commonly occur in storage devices such as RAM, disks and magnetic tapes. Henceerror handling schemes are commonly aimed at protecting the data in these devices. Dynamic memory is, in particular, highly susceptible to errors. As we move towards systems with larger memories, data protection schemes become increasingly relevant because the error rate is found to increase with the memory size. Till recently parity has been the traditional method of data protection. Parity, however, has several drawbacks: it masks out even bit errors, it cannot locate the bit in error and beyond a certain buswidth it can become inefficient. These factors have lead to the development of a new breed of devices for data protection. Error Detection and Correction circuits (EDCs or EDACs) are today's response to the growing need for greater data reliability. They afford a greater level of protection than that offered by parity, by using a higher order error correction code (Modified Hamming code). Implementing error detection & correction in a system involves not just the hardware to perform the error checking but also extra memory for storage of "checkbits", which are the basis of error detection and correction. An EDC may add some overhead to a system in terms of speed and cost. This is, however, compensated by a significant improvement in system performance. The following Figures underline this pOint: Given a 4M x 64 DRAM memory with an FIT(Failure in time) specification of 252 1 : MTBF (mean time between failures) without EDC = 1.76 years MTBF with single-bit error correcting 64-bit EDC = 3935.4 years The Figures given here are merely a guideline for comparison of the two cases and will depend entirely on the DRAM used and the memory configuration. Thus EDC can be looked upon as a sophisticated parity system with the capacity to correct. Two primary functions are performed by an EDC unit within the framework olthe memory read and write cycles - during a memory read, errors are detected andfor corrected and during a memory write, checkbits are generated. ERROR DETECTION AND CORRECTION WITH THE 49C466 Choosing the Right EDC EDC considerably upgrades reliability but, as mentioned earlier, there is a certain amount of overhead associated with implementing EDC. This is because of the checkbit memory required and the nanoseconds it adds to the main memory cycle times. The checkbit overhead can be minimized by increasing the size of the data word used to generate the checkbits. The table below illustrates this point and shows how parity compares with EDC for various data word sizes. Assuming a distance-of 4 Hamming code for the EDC and byte parity: As an example, consider a 32-bit EDC and a 64-bit EDC. We see from the table above, the former requires 7 checkbits DATA WORD SIZE PARITY BITS REQUIRED EDC CHECKBITS REQUIRED 16 32 64 128 2 4 8 16 6 7 8 9 2596tbl01 for each 32-bit data word. Hence every 64 bits of data requires 7+7=14checkbits in memory. A 64-bit EDC on the other hand requires only 8 checkbits to provide single bit error correction and dual bit error detection. Thus, as far as checkbit efficiency goes, a 64-bit EDC is better than a 32-bit EDC. One other advantage of a wider EDC is that less checkbit memory implies a lower error rate in checkbit memory. Going after even wider EDCs seems tempting in this light, but a compromise has to be made because of two factors. Prevalent system bus widths impose a practical limitation on how wide a bus can get. It is fairly common nowadays, however, to see a 64-bit memory bus and hence a 64-bit EDC is often easily accomodated. An EDC that is wider than the memory bus needs additional control and logic circuitry to integrate the EDC with the memory system. The greater this difference in (EDC and memory) bus widths, the more complex is this interface. Also, the error handling capacity effectively decreases as the EDC bus gets wider because the same level of protection (1 bit correction,2 bit detection) is provided for a wider word. Two 32-bit EDCs provide this for each 32-bit word, which allows correction of some 2-bit errors and detection of some 4-bit errors for each 64-bitword. Thus, constraints for a given system determine the optimum trade-off. NOTE: See appendix for further details on calculation of MTBFs. The lOT logo is a registered trademark of Integrated Device Technology, Inc. ©1993 Integrated Device Technology, Inc. 9/91 112 .....,....-- ------ ....... - -- -ERROR DETECTION AND CORRECTION WITH IDT49C466 APPLICATION NOTE AN·94 49C466 Operation and Features Memory Write By the nature of its function, the EOC is concerned with data transfers between the processor and memory. The IDT 49C466 has a flowthrough architecture which permits transparent data flow through the EOC. The 49C466 also increases checkbit efficiency by providing two 64-bit wide bidirectional data buses. The error detection & correction operation in the 49C466 is similar to previous generation lOT devices. A modified Hamming code is used to generate the checkbits and the syndrome decoding is identical to cascaded mode operation in the 32-bit EOC, the 49C465. The modified Hamming code used allows for flagging of all single, dual and three bit errors in the 64-bit data word. Some three bit errors alias as single bit errors, however, and may hence be wrongly decoded by the syndrome decoding logic (MERR* may not be asserted). The code allows for correction of all single bit errors in the data word. The 49C466 has five modes of operation. In the normal mode, two kinds of operations are performed. During a "memory write", checkbits are generated based on the data that is written and during a "memory read" single-bit errors in the data are corrected. In the detect-only mode, the "memory write" operation remains the same but during a "memory read" any errors detected are flagged by ERR and MERR pins. The data is passed though unaltered. The other three modes are useful fortesting & diagnostic purposes. The EDC mode of operation is set by the user by loading the mode register through the system data bus. Read and write paths are independent of each other in the 49C466. In fact, the device has alternate paths for each of these (read and write cycles). One path includes the 8/16deep data buffer while the other provides a latch in place of this buffer. WBSEL(Write Buffer Select) and RBSEL(Read Buffer Select) pins select the output of either the buffer or the latch (for example - RBSEL=high selects output from the Read Buffer rather than the MD_OUT latch, WBSEL = 0 selects the output of the SO_IN latch rather than the Write Buffer). During a memory write, the EOC generates checkbits corresponding to the data being written to memory (data flow from SD bus to MO bus). These are output onto the CBSYN bus and written to checkbit memory. This is a necessary part of EDC operation. Unless all checkbits corresponding to the data in memory have been generated and stored in checkbit memory, no error checking is possible. To prevent contention on the SD bus, the SD output buffer must be disabled, during a memory write. The MD and checkbit output buffers must be enabled to pass data and checkbits to memory. There are several ways that the "write path" can be configured but the maximum "write time" with input and output latches transparent is 15ns. Figure 2 illustrates the data path during a write operation. The alternate write path, where the data is buffered, is shown in Figure 3. Memory Read A memory read involves checking data from memory for errors. Given below is a description of what actually takes place during a "read cycle". Data and checkbits from main memory are fed to the EDC. Checkbits corresponding to the read data are generated within the EOC. The two sets of checkbits are internally compared to produce the syndrome word. The EDC then decodes the syndrome word to check for errors. The syndrome is also clocked into the syndrome register on the rising edge of SYNCLK. The syndrome register contents can be output on the CBSYN bus. The CBSEL pin controls the output of the CBSYN bus. While CBSEL is low, checkbits generated during a write are output. When CBSEL is high, syndrome generated during a read are output. Figure 4 shows the data path during a normal read operation. A buffered read operation is illustrated in Figure 5. ...-------------1 CB_in ERROR CORRECT L--J-------------'---1 MD_in SDjn Write Path CB-out 2596 drw 01 Figure 1. Basic Memory Read and Write paths Through the 49C66 113 ERROR DETECTION AND CORRECTION WITH IDT49C466 APPLICATION NOTE AN-94 BEO-? -'i---1I....----------, SOE -.f---,--, read path SOO-63 write back path ~ :::2: wH------~ SO OUT LATCH >----l~ ~ MOO-63 '----.---114-- MOE __-----~--~--SOOLE .---~._~- CBSEL I-a: _0 !Ill- o:':::~ CIl@ W SO CHECKBIT LATCH Z >- IZ OW Cl POOl ~ o ~-------------------------_44_~WBSEL EOC466 2596 drw 02 Figure 2. Memory Write without Write Buffer SOE -"1-----, r---------------------------~~ WBSEL BEO-?~+--tI...... SOO-63 ......----I~ SO OUT LATCH MOO-63 ~-- ____44__-MOE ----------~~~--SOOLE .------f--,.-~- SO CHECKBIT LATCH Z ~ SO checkbits o 2596drw 03 Figure 3. Memory Write with Write Buffer 114 CBSEL ----- -- ..... ERROR DETECTION AND CORRECTION WITH IDT49C466 APPLICATION NOTE AN-94 Beside its primary task of error checking, the 49C466 integrates certain other useful functions on chip. It supports parity checking and generation, partial word writes and diagnostics. It provides the user with flexibility by providing 8/16 deep Read and Write buffers and the facility to latch all data flowing in and out of the part. The parity generation and checking capability in the 49C466 is very useful in checking the integrity of the data being written to memory. Parity checking is done on data from the system side. The parity bits are input on the EDC POol pins. Parity is then generated for the input system data and a comparison of this and the input parity bits is carried out internally. The result of this comparison is reflected in the PERR pin output (a discrepancy asserts PERR). Parity bits are also generated for data read from memory and output on the POol pins once again. The parity type (odd or even) is selected using the PSEL pin. The PERR signal does not affect any of the other circuitry in the device and hence the user may safely ignore the parity feature if his system does not support parity. A partial word write or byte merge is analogous to a readmodify-write cycle. The checkbit word generated by a partial data word would be incomplete and hence incorrect. Hence the need to differentiate the "partial word write" case from a normal write operation. The following steps must be followed to ensure correct generation of checkbits for the complete 64 bit word : 1. Read the contents of the location being written to. 2. Merge the partial word with these contents forming a composite 64 bit data word. 3. Generate the 8 checkbits for this composite word. On account of the dual bus feature of the 49C466, data may be written to the EDC (from processor/cache) while data is read from main memory to the EDC. This feature is useful during partial word writes (writing less than eight bytes) and can buy some time for the designer for whom each nanosecond counts. The time required to turn the external buses around must, however, be taken into consideration. PARTIAL WORD WRITES The 49C466 supports "partial word writes" or "byte merging". These refer to write operations involving words shorter than 64 bits such as a byte write. Partial word writes are handled in the 49C466 by the byte multiplexer. The 49C466 has eight BE (byte enable) control pins. These control the byte multiplexer and enable the system data output buffer. When a BE input is high, it selects a byte from the MD "write back path" shown in Figure 3 and Figure 4 rather than the normal EDC write path. Each BE input is AND-ed with the SOE signal and the result determines which byte is output on theSD bus. A BE pin that is low disables the SO output buffer for the particular byte referenced and selects a byte from the write bufferor SO_in latch rather than one from the MD write back path. DESIGNING WITH THE 49C466 As typical application examples, we consider 32-bit and 64bit processor designs using EDC to protect the slower dynamic main memory. On account of its 64-bit data buses, the 49C466 is easily interfaced to 64-bit systems. As mentioned earlier, providing a 64 bit memory subsystem bus is often advantageous even in 32 bit processor systems. Thus, even though the interface between 32- and 64-bit wide buses may be slightly more complex, it is often worthwhile. Consequently providing a 64-bit EDC may still be an attractive choice. ERR~+---------~~~~ MD i-------\CHECKBIT MERR:I-----iQ!~~ SOE~-~---. BEQ-? ...-+-.....-:1......0. . -______~~========~--------~~L~A~TC~H~~~__ MD CHECKBIT GENERATOR MD IN LATCH write back path >--.......---< MDO-63 '-----e~~~ MOE POol PARITY r----,.-+~ \.r-~~t---I GENERATOR SD checkbits EDC 466 2596 drw 04 Figure 4_ Memory Read without Read Buffer 115 CBSEL ERROR DETECTION AND CORRECTION WITH IDT49C466 APPLICATION NOTE AN·94 ERROR:~~I SYNDROME I. ERR MERR RBREN ... MCLK DECODE MD CHECKBIT LATCH GENERATOR j. - r:rnrn ~ CBIO-? " If' MDllE SCLK &il: BEQ-7 .A SDO-63 f- RBSEl f ~'IR~D ~ ~YiF,~~, ~ I ERROR CORRECT write back pam " MD CHECKBIT GENERATOR l I ... ~ w =-... POol ~ GENERATOR PARITY ~ .....--.,.x ~ CD < MD IN LATCH I 7~ i I I"MDO-63 ... MOE CBSEl ~:::J SD ,"''''''''' - EDC466 " CBSYNO-? 2596 drw 05 Figure 5 Memory Read with Read Buffer R3000 BASED SYSTEM The MIPS R3000 RISC architecture has attained much popularity in recent years. Therefore, we discuss here a design example showing the interface between the 32-bit R3000 and the 64-bit 49C466. This typical R3000 design employs 2 data buses. The high speed "processor" data bus is the interface between processor, caches & buffers while the slower 'memory' data bus links the buffers and I/O devices (peripherals & memory). Providing two separate buses for the processor and memory subsystems, in this manner, serves multiple purposes. It reduces the traffic on each bus and allows for different speeds and widths on the two buses. In the design shown, the 49C466 is positioned between the 64-bit memory bus and main memory as in Figure 6. Programmable logic is used to generate the control signals for the 49C466 64-bitflowthruEDC"'. Appropriate control signals from the R3000 processor provide input to the PAL. The example considered here does not use the on-chip read/write buffers of the 49C466 and the device is used in a non-pipelined mode. External read/write buffers are provided as the interface between the 32-bit processor bus and the 64bit memory bus. A 64-bit wide memory bus demands that an appropriate scheme for (32-bit) word gathering be devised. One method of implementing such a scheme is described here. On write cycles, two 32-bit words are gathered in the two external write buffers and the 64-bit word is output to the memory bus. Gathering 32-bits words, in this fashion, may not always be permissible because of requirements thatthe two 32-bitwords be written to specific, non-sequential locations. Such cases are treated as partial word writes by the EDC. Each 32-bit word write is hanoled as an individual partial word write cycle. A partial word write cycle requires thatthe 32-bitword from the adjoining memory location is read before the whole 64-bit word is written back to memory. The two categories of writes - "gathering permissible" and "gathering not permissible" - are differentiated by comparing the write addresses generated by the processor for the two words. This demands that the first write address must be latched until the next write request is received. If the two addresses are not consecutive, a partial word write signal is sent to the PAL. This initiates a partial word write cycle for writing the first word to the EDC. To understand this particular situation better, let us consider a case where two write addresses have been compared and found to be non-sequential. The first write data is buffered in WBufA and the second one in WBufB. Each of these write request are treated as "partial word writes" and handled in the order they were received. In order to support byte access, the memory must be byte addressable. The contents of WBufA are output to the SD bus of the 49C466. To perform the "partial word write" cycle contents of the location address output from WBufA are simultaneously read onto the MD bus of the 49C466. Merging of these two 64bit wide data words is performed inside the 49C466 and the merged 64-bit word is output on the 49C466 MD bus and written back to the same location. When this cycle is complete, the same thing is done for the second write data which is stored in WBufB. During read cycles, the issue of gathering "reads" and comparing read addresses can be avoided altogether by always reading a 64-bit word from memory.The required 32 bits are selected and the rest ignored. 116 ERROR DETECTION AND CORRECTION WITH IDT49C466 APPLICATION NOTE AN-94 STEP 1 : Main Memory EDC MD ,.----1 IXXXXEFGH I BEO-? = 00001111 STEP 2: EDC Main Memo ~ S IABCDEFGH 16~ ~'$<'ll:' - ,- , ___ I BYTE "~~IMUX 2596 drw 06 Figure 6. Byte Merge Details of this scheme, implemented with IDT parts, are shown in Figures 7 and B. The design shown, uses a bus mUltiplexer, the IDT49FCTB04 to transfer each 32-bit word to the appropriate SD bus lines (0-31 or 32-63). The write buffer used (IDT 79R3020) buffers both addresses and data. Two bus multiplexers are used to create a crossbar type of arrangement, so that the output of each write buffer can be directed to either the upper or lower SD lines of the 49C466. 64 BIT SYSTEMS The 49C466 interface to a 64-bit processor is quite straightforward. Intel's iB60 and MIPS' R4000 are two popular processors matching this buswidth. Figure 9 shows an iB60 based system with EDC. In the system shown all memory accesses go through the 49C466. This kind of a setup would ensure filtering of errors from all memory data accessed. The iB60 does not support parity but external parity support circuits (like the AMD2BO) enable the user to still take advantage of the parity feature of the 49C466. This is particularly useful when caches are employed and some monitoring of data written to main memory is required. As mentioned earlier, with the trend to segregate processor and memory subsystem designs, a 64 bit memory bus is not unlikely in a 32-bit processor design. In these cases too the 49C466 interface is similar to the one shown above. Also additional hardware is required to correct more than one bit in error. Taking these factors into account, little motivation to move towards multiple bit error correction can exist at this point. Since hard errors are non-random it is possible for hardware to differentiate them from soft errors. Once this is done it is a simple matter to correct the data provided there is not more than one soft error present. Given below is a brief description of how hard errors can be eliminated. When an error occurs, the error data is latched in the "error data register" of the 49C466. This data can be read by the processor in the Error Data Mode. In order to filter out hard errors, this data should be inverted and written back to the same location. A subsequent read will serve to identify hard errors, for due to the hard error, the affected bits remain at their original logic level. Consider a case where bits 3,6 and 7 (shown by bold letters) are affected by hard errors: Correct Data Error Data (ED) Inverted error data Inverted error data after subsequent read (SlED) By XOR-ing the Error Data with the data from the subsequent read, the bits in (hard) error are isolated (indicated by zeroes). (SlED) XOR (ED) MORE EFFECTIVE EDCS Most present generation EDC units are able to correct only single bit errors. Given the probability of occurrence of multiple bit errors, this is, in most cases, sufficient. The ratio of single bit errors to dual bit errors ranges from 5:1 to 10:1. The probability of multiple bit soft errors occurring is negligible. = 10101001 =01100001 = 10011110 = 01010110 = 00110111 Thus, all that is externally needed to perform this check is XOR logic. 117 ERROR DETECTION AND CORRECTION WITH IDT49C466 APPLICATION NOTE AN-94 l R3DDD D - Cache ~5 I I I 1- Cache ~~ Processor Data Bus I e I RIW Buffers -' U I I Memory Data Bus V Controller --'" EDC466 Checkbit Memory <:=) D Main Memory 2596 drw 07 Figure 7. R3000 based System with EDC AddrBus2 LE ~CE1 u Even Memory Bank MOo-63 C/) ~ C/) '" '" w ~ ~ C/) C/) a:: o o « EOC 466 a:« 0. + Even CBMemory ~ CB10-7 CBSYNO-7 « o Odd CBMemory WBUFB 79R3020 X4 SOBS1B OE WBUFA WBUFB BMUXA BMUXB control control control control WtMemB MemWr CONTROLLER (with Address comparator logic) MemRd RdBusy EOC control CE1 CE2 WrBusy Figure 8. R3000 based EDC System - Memory Write 118 2596drw 08 ERROR DETECll0N AND CORRECTION WITH IDT49C466 APPLICAll0N NOTE AN-94 AddrBus2 u Even Memory Bank 1-________-1 Data Memory Moo-63 Odd Memory Bank '" :oJ co (f) (f) w a: o oc( ~ 0:: c( 0. ~ CBIO-7 CBSYNO-7 o Even Memory Bank 1--_____________1Checkbit Memory Odd Memory Bank RBUFC control BMUXC control MemWr--------~.~r~~--~~--_. MemRd --------~.~I RdBusy .---------1 WrBusy +--------1 EDC control CONTROLLER (with Address comparator logic) 2596 drw 09 Figure 9. R3000 based EDC System - Memory Read MEMORY SCRUBBING CONCLUSIONS An alternative (or in addition) to putting the EDC directly in the read/write path, is putting it in the DRAM refresh path. This ensures that memory is periodically checked for data validity. This is a good practice as it prevents buildup of single bit errors at infrequently accessed memory locations. There are, however, chances of errors slipping through the net in this scheme when an error occurs between a memory refresh and the next memory read or write. The urge to provide greater value and reliability in today's competitive computer systems, is likely to make EDC a standard feature in the near future. With the continuing trend towards wider buses, it is only natural to move towards wider EDCs. This also makes the EDC implementation more efficient. IDT is at the forefront of this new generation of EDCs with the powerful new 64-bit flowthruEDC, the 49C466. The 49C466 has an edge over it's competition by providing several useful features such as byte merge capability, 16 deep buffering, latches, diagnostic registers, parity generation and checking and competitive detect and correct times. Currently the device is available in 208 pin PGA and PQFP packages. 119 ERROR DETECTION AND CORRECTION WITH IDT49C466 APPLICATION NOTE AN-94 MEMWR? ~ ____________________-.YES BMUXA: PORTA ->PORTC(SDO-31) BMUXA: PORTA ->PORTC(SDO-31) BMUXB: PORTB->PORTC(SD32-63) Output address on AddrBus2 2596 drw 10 Figure 10. Flowchart for Write Buffer and Bus MuHiplexer Control During Memory Write 120 ERROR DETECTION AND CORRECTION WITH IDT49C466 APPLICATION NOTE AN-94 APPENDIX MTBF Calculations DRAM manufacturers specify a soft error rates in terms of FITs (failures in time). Assume a 1M x 1 DRAM with soft error rate = 252 FITs Thus, 109 MTBF for each 1M x 1 DRAM chip = 252 Output address on AddrBus2 and assert CEI and CE2 MTBF for a 4M x 64 memory system (without EDC) = 453 years 453 = - - = 1.76 years 4 x 64 [A memory system without EDC assumes a system failure each time a single bit error occurs. This may not be the reality but this exercise is aimed at showing the difference between two analogous cases, hence such assumptions are in place. Failures due to all higher order (dual, three bit, etc.) errors can be safely ignored in this case due to the overwhelming dominance of single bit errors over other higher order errors. 1 With EDC, extra memory is required to store checkbits, hence the number of DRAM chips required for the same memory system goes up. Checkbit memory (with 64 bit EDC) required for the above system Assert RDBUSY Total memory system chip count 1 = -x 4 x 64 = 32 chips 8 = ( 4 x 64 ) + 32 = 288 So now, 453 = 1.57 years Memory system MTBF = 288 BMUXC: PORTB -> PORTA or PORTC -> PORTA Now if we assume that with EDC, all single bit errors are corrected, failures due to dual bit errors become dominant. Neglecting failures due to higher order errors, we get an approximation of the MTBF using the following formula: MTBF (with EDC) = (MTBF without EOC) x = 1.57 x y( 2596 drw 11 Figure 11. Flowchart for Read Buffer and Bus Multiplexer Control During Memory Read 1tx4x1012 2 y( 1t X # f 0 d memory wor s) 2 ) = 3935.4 years Thus we see that there is a significant improvement in the MTBF of a system with EDC. 121 ERROR DETECTION AND CORRECTION WITH IDT49C466 APPLICATION NOTE AN·94 CLK GENERATION CIRCUITRY --0 VCC SCLK~ MCLK r~ RBSEL WBSEL SDILE CLK ,!)4 D BE MAIN MEMORY DRAM ,8 , SD "' 49C466 BE MD ,64... i860 1. MDOLE* CBSYN - READY* W/R* ,8 ... . CHECKBIT MEMORY ~ EDC_CONTROL_IN EDC_CONTROL_OUT EDC& DRAM CONTROLLER NB: EDC_CONTROL_IN: SDOLE*,MDILE,wBEN*,WBREN*,RBEN*,WBREN*,CBSEL,RSO-1, EDC_CONTROL_OUT: WBEF*,wBFF*,RBEF*,RBFF*,RBHF*,ERR*,MERR* 2596 drw 12 Figure 10. Flowchart for Write Buffer and Bus Multiplexer Control During Memory Write 122 I 'II g APPLICATION NOTE AN-96 CASCADING THE IDT49C460 AND IDT49C465 Integrated Device Technology, Inc. By Anupama Hegde the partial checkbits in anyone of the slices rather than after the complete checkbit word is produced. Figure 1 illustrates the operations involved in a cascade. During a regular memory read with EDC, checkbits and data from memory are fed to the EDC. Checkbits generated from the data are compared with the input checkbits resulting in the syndrome. This is decoded to produce error signals and to correct data. When using a pair of cascaded EDCs, the same operation is done as follows: Partial checkbits are generated from the partial data input to that particular slice. These are XOR-ed to produce the final syndrome. The final syndrome is made available in all the slices. An alternate way of doing things is to simultaneously generate the final syndrome in all the slices. This parallel approach is faster. INTRODUCTION Error Detection and Correction units or EDCs are used to check and ensure data integrity in computer systems. Standard off-the-shelf devices do not always meet the needs of the user and need to be cascaded. Cascading EDCs allow the user to expand the EDC bus width. Wider bus operation is advantageous because wider EDCs call for less checkbit memory. On the other hand, an EDC bus wider than the existing system data bus may call for complicated control schemes and hardware overhead. Typically, it is desirable to have the EDC and system bus widths match. Cascaded EDC operation requires additional logic circuitry. This can be built into the EDC or provided externally by the user. The IDT49C460 and the IDT49C465 each have the required cascading circuitry built into them. This application note explains cascaded EDC operation using the IDT 49C460 and IDT49C465. THE CASCADING PRINCIPLE When several EDC slices are combined to create a wider data bus, none of the slices can individually produce the complete checkbits. This is because only a part of the complete data word is fed to each slice. Partial checkbits are generated in each slice. The partial checkbits from each slice are XOR-ed to generate the complete checkbit word. The syndrome is produced by an XOR of the complete checkbits and the input checkbits. The sequential order in which these XOR operations occur may vary depending on the implementation. For example, the input checkbits may be XOR-ed with Slice 1 ( PCB1 CASCADING 32-BIT EDCS FOR 64-BIT OPERATION Any number of EDCs may be cascaded depending on the available support logic and performance demands. A dual EDCcascade, assupportedbythe IDT49C460and IDT49C465 is discussed here. 32-bit checkbit generation logic differs from the 32-bit partial checkbit generation logic that is required for cascading two 32-bit EDCs. Eight checkbits are needed by a 64-bit EDC but the 32-bit checkbit generation logic produces only 7 checkbits. Hence, additional logic is required to generate partial checkbits. The 64-bit checkbit generation scheme (as indicated by 64-bit checkbit generate tables in data sheets) may be split into two 32-bit generation schemes - one Slice 3 Slice 2 ) ( PCB2 ) 2530 drw 01 Figure 1. Complete Checkbit Generation From Cascaded Slices The lOT logo is a registered trademark of Integrated Device Technology, Inc. ©1993 Integrated Device Technology, Inc. 12191 123 CASCADING THE IDT49C460 AND IDT49C465 APPLICATION NOTE AN-96 operating on data bits 0-31 and the other on data bits 32-63. This determines the 'partial checkbit generation logic'. Each cascaded slice then produces 8 checkbits which can be XORed to produce the complete checkbit word. Also, there mustbesome mechanism to determine whether the EOC is in cascaded mode or not. If it is in cascaded mode, each EOC unit must be identified as a unique slice (first, second, etc.) in the cascade. The lOT 49C460 and the lOT49C465 differ in their architectures. The IDT49C460 has an older 'bus-watch' architecture with one data bus, while the lOT49C465 has a dual data bus Flowthru™ architecture. Both parts have 32-bit wide data buses and use the same basic principle of operation. The lOT49C460 and lOT49C465 have CodeD, 1 lines that program the EOC units to operate in non-cascaded mode or as upper or lower slices in cascaded mode. CASCADING THE IDT49C460 The figure below shows the correct connections for cascading two 32-bit IOT49C460s. This makes them function as a 64-bit EOC for either generate (memory write) or correct (memory read) operations. Control signals not shown are common to both units and parallel connections to each slice may be used. Memory Write During this kind of operation you need to generate 8 checkbits for a 64-bit data word. In 32-bit mode, the lOT49C460 generates 7 checkbits for each 32-bit data word. In cascaded mode the IOT49C460 must generate 8 partial checkbits for each set of 32 data bits. If you compare the checkbit generate tables (Tables 6 and 11) of the lOT 49C460 data sheet, you will notice that they are different. For 64-bit cascaded mode operation, Table 11 is to be used. Eight partial checkbits are generated by a logic implementation of the 64-bit checkbit generate table. In the lower slice; partial checkbit word = 64-bit checkbit generate table applied to data bits 0-31. In the upper slice; partial checkbit word = 64-bit checkbit generate table applied to data bits 32-63. Example: Partial Checkbit 0 in lowerslice (PCBO-,s) = 01 EB 02 EB 03 EB 05 EB 08 $ 09 $ 011 $ 014 $ 017 $ 018 EB 019 $ 021 EB 024 EB 025 EB 027 EB 030 Partial Checkbit 0 in upper slice (PCBO_us)= 032 $ 036 $~$~EB~EB~EB~EB~EB~EB~EB 054 EB 055 EB 058 EB 060 EB 061 EB 063 XOR-ing the Partial checkbit 0 from the upper slice and the Partial checkbit #0 from the lower slices gives the final checkbit o. Thus, check bit 0 for the 64-bit data word = (PCBO_ls) EB (PCBO_us) 01 EB 02 EB 03 EB 05$ 08 $ 09 EB 011 EB 014 EB 017 $ 018 EB 019 EB 021 $ 024 $ 025 $ 027 EB 030 EB 032 EB 036 EB 038 $ 039 EB 042 EB 044 EB 045 EB 047 EB 048 EB 052 EB 054 E9 055 EB 058 E9 060 E9 061 E9 063, which is what we get from Table 11 for 64-bit data. The complete checkbit word for the 64-bit data word is generated by XOR-ing each of the partial checkbits from the upper and lower slices. A block diagram illustrating this operation is given in Figure 2. UPPER SLICE LOWER SLICE 49C460 49C460 32 bit EDC DATAinO-31 32 bit EDC DATAin32-63 Final Checkbits 2530 drw 02 Figure 2. IDT49C460 Cascaded Mode "Memory Write" 124 CASCADING THE IDT49C460 AND IDT49C465 APPLICATION NOTE AN-96 Memory Read A memory read or correct operation involves generation of a complete set of 8 checkbits for the 64-bit data, computation and decoding of the syndrome word, and correction of the data. A regular 32-bit correct operation involves 7 -bit checkbit words and a 7-bilsyndrome. Apart from this, the procedure is the same as above. In 64-bit cascaded mode, you have 8 input checkbits (CBO7), but each individual slice only generates a 'partial checkbit word'. This is why it is necessary to combine the 'partial This entire operation is illustrated in Figure 2. It can be summarized as follows: 1) In the first pass through the lower slice the partial syndrome is generated. 2) The final syndrome is generated in the upper slice. If an error occurs in Data bits 32-63, the appropriate bit in the lower slice is corrected. 3) The final syndrome is sent back to the lower slice. If an error occurs in the Data bits 0-31, the appropriate bit in the upper slice is corrected. Final Syndrome LOWER SLICE UPPER SLICE 49C460 49C460 32 bitEDC Buffer Enable 32 bit EDC seQ-? Partial S ndrome Final 1 - - - - - , / Syndrome DATAin 32-63 2530 drw 03 Figure 3. IDT49C460 Cascaded Mode "Memory Read" checkbit words' from each slice to obtain a complete checkbit word. Once this is done the syndrome can be computed by XOR-ing the 8 input checkbits with the 8 (complete) generated checkbits as usual. Thus, Final Syndrome = (CBO-7) E!:J (8 checkbits generated from 64-bit data) = (CBO-7) E!:J [(PCB_lsO-7) E!:J (PCB_usO-7)) = [(CBO-7) E!:J [(PCBJsO-7)) E!:J PCB_usO-7) 1 1 1 1 1 lower slice 1 1_ _ _ _ _ -1 1 upper slice The first two terms constitute the partial syndrome. In the IDT49C460, the first XOR is done in the lower slice and the result is output on the SCO-7 bus of the lower slice. The second XOR is done in the upper slice and the (final syndrome) result is output on the SCO-7 bus of the upper slice. This final syndrome is then fed back to the lower slice. It must be made available in both slices in order for a correction of either DataO-31 or Data32-63 to be possible. USING THE IDT49C465 IN 64-BIT OPERATIONS In the IDT49C460 the partial checkbits and partial syndrome are output multiplexed on the same bus. In the IDT49C465, however, these are available on separate buses. During a 'memory write', the partial checkbits from the lower slice are output on the CBO bus of this slice. They are fed to the PCBI bus of the upper slice. The final checkbits are available on the CBO bus of this upper slice. While doing a 'memory read' or detecVcorrect, the partial syndrome (XOR of input checkbits and checkbits generated in that particular slice) is output on the SYO bus. It is fed to the CBI bus of the upper slice. In this slice, this partial syndrome is XOR-ed with partial checkbits generated in this upper slice to produce the final syndrome. Simultaneously, the partial checkbits produced in the upper slice (from upper 32 data bits) are sent back to the lower slice. They are then XOR-ed with the partial syndrome already available in the lower slice producing the final syndrome in the lower slice as well. Thus, the final syndrome is simultaneously generated in both slices of the IDT49C465 cascade. 125 CASCADING THE IDT49C460 AND IDT49C465 APPLICATION NOTE AN-96 Partial Checkbits LOWER SLICE UPPER SLICE CBIQ-? ==>H-~~~~ ---I~ DATAinQ<=>.\4-......... _ _ _..J SCQ-? Syndrome Register Final Syndrome 49C465 Syndrome Register 32-bit EDC 49C465 32-bit EDC 2530 drw04 Figure 4. IDT49C465 Cascaded Mode EDC Operation SUMMARY As system bus widths continue to grow, cascading EDCs may be necessary. This application note clearly explains the basic principle behind cascading EDCs. It also describes the cascading support available with two 32-bit IDT EDC chipsthe IDT49C460 and the IDT49C465. 126 ~ f;y CONFERENCE PAPER CP-10 INTEGRATED LINE TERMINATION SOLUTIONS FOR HIGH-SPEED LOGIC Integrated Device Technology, IDC. By: Stanley Hromk As logic speeds increase, there is an increase in noise generated by faster edge rates. To reduce noise emissions and generate cleaner waveforms, it is often necessary to terminate signal lines. The need for termination becomes more pronounced when the signals are driven over longer lines, smaller width traces, across backplanes, or at higher frequencies. Properly terminated lines will also reduce noise on the power distribution system which may ease board layout and decoupling. Termination with external components increases part count, cost and the complexity of board designs. In addressing this problem, component manufacturers have integrated termination schemes into the output buffers of interface circuits which greatly eases the board level design and layout task. By utilizing these traditional and integrated line termination schemes, designers have the ability to increase system performance and reduce costs. THE NATURE OF THE PROBLEM With the new families of high speed logic, internal propagation delays have been reduced to the point that the external interface accounts for most of the component propagation delay. In order to increase device speed further, the edge rates and drive capabilities of the devices have been increased. These faster edge rates are adding higher frequency components to the frequency spectrum of the transmitted signal. If the rise/fall time of the signal is shorter than the round trip propagation delay in the signal path, the signal path will act like a transmission line and need termination. The edge rate on high speed logic can be faster than 1 voltlns. A commonly used equation to determine the maximum unterminated trace length is as follows: L TRADITIONAL TERMINATION SCHEMES The following cursory review of the traditional termination methods gives insight into the benefits and drawbacks of each scheme. Parallel or Shunt Termination: Parallel termination is the addition of a line matching impedance at the receiver as shown in Figure 2. When the charge from the signal reaches the receiving end of the line, the resistor will drain off excess charge preventing reflections. Figure 2. Transmission Line with Shunt Termination ---IL = To maintain good signal quality on a transmission line, the dynamiC impedance of the transmission line needs to be constant throughout the line. This means that stubs, unterminated lines and very low source impedances can all contribute to poor signal quality. When a signal passes down a transmission line, if it sees a continuous impedance, it will travel smoothly without reflections. If there are abrupt changes in the impedance at any point in the transmission line, reflections will bounce off of these points and may cause noise if uncontrolled. In designing the transmission line it is best to utilize a single line from source to load. It is acceptable to drive a distributed load by daisy chaining the loads and avoiding stubs. If there is a low impedance load to be driven, it should be at the end of the transmission line. 2Tpd Thevenin Termination: Tr = Rise/fall time of the signal (ns) Tpd = Propagation delay of the transmission line (ns/ft) L = Maximum Unterminated Cable Length (It) The Thevenin Termination balances the DC loading on the driver between the logic high state and the logic low state reducing Signal distortion. In this scheme, there is a path for DC current to pass directly from the Vcc to Gnd through the termination resistors which will dissipate power regardless of the state of the output driver including the high-impedance state. f,t>- I I I I I I R1 ~~' +i I I I Ii Figure 1. Transmission Line ;;;:- ;;;:- ;;;:- R2 ;;;:-. ;;;:- 3014 drw 03 Figure 3. Transmission Line with Thevenin Termination The lOT logo is a registered trademark of Integrated Device Technology, Ino. 9/92 ©1993 Integrated Device Technology, Inc. 127 INTEGRATED LINE TERMINAL SOLUTIONS FOR HIGH-SPEED LOGIC RC Termination: An effective method of termination is the RC termination using a 47pf to 200pf capacitor in series with a resistor to ground at the receiver. Larger capacitors give a cleaner waveform but consume more power at high frequencies than lower valued capacitors. The resistor value should equal the characteristic impedance of the transmission line. 3014 drw 04 Figure 4. Transmission Line with RC termination RC termination is very effective for point to pointtermination over long distances because the signal passes down the transmission line and does not reflect. This makes it possible to use RC termination in cases where the propagation delay of the transmission line exceeds the cycle time of the driving Signal. With the AC coupling at the terminating end, there is no DC power consumed by the termination making the scheme useful for low power applications. CONFERENCE PAPER CP-10 As the output driver undergoes a transition, the voltage is divided equally between the output impedance (driver and resistor) and the transmission line, as shown in Figure 6. Because ofthe matched source impedance, the voltage at the receiver builds up to a level equivalent to the driving voltage and will reflect the signal back from receiver to source and eventually allow a full voltage transition at the source as shown in Figure 6. Series termination is the simplest, most effective form of reducing signal noise in transmission lines. This method consumes no DC power and helps to reduce dynamic power consumption. Series termination is the only method that handles stubs well and reduces ground bounce. In situations in which the loads on the transmission line are highly capacitive, the series termination may cause time delays due to the RC time constants. Summary of Traditional Schemes A relative comparison of the attributes of the four traditional termination schemes is summarized in Table 1. FEATURES OF THE FOUR TERMINATION SCHEMES Characteristic Series Termination!l): ~ 1 I= I I= I I 3014 dlW 05 Figure 5. Transmission Line with Series Termination Series termination is the addition of a resistor between the component output and the transmission line as shown in Figure S. The combination of the series resistor and the source impedance of the driver act to match the characteristic " \\.._ _ _ _ _-11 I I 1 ~~ I I "I I I Signal at transmitting e n d : I I : : I I \ rI Signal at receiving end RC Series 4 1 1 Dynamic Power Dissipation 4 3 2 1 4 4 3 1 1 Works with Stubs 3 2 1 Distributed Load Driving 1 1 1 2 Parts Count for Termination 1 2 2 1 Works With Heavy Capacitive Loads 1 1 1 2 Good for Low·lmpedance Backplanes 1 1 1 Works with Tpd > Data Rate 1 1 1 2 2 Ibl01 INTEGRATED SERIES TERMINATION: impedance of the trace. Open Circuit \ Signal at Transmitter Thev 3 Signal Distortion Due to Unbalance Iii>-= - Shunt Static Power Dissipation I 1 ' - - - - - - . . . J , 3 0 1 4 drw06 Figure 6. Signals Produced When Using a Series Terminator Improvements to the series method can be gained if the series resistor is integrated into the driving component. By integrating the resistor, the output impedances of the driver can be matched to the typical load of a transmission line in both the high and low going cases. Properly designed integrated schemes can outperform external resistor terminations by reducing source follower problems and increasing output drive current. The removal of the external resistors will also reduce part count. Figure 8A shows a high drive, MOS output structure without any integrated termination. In most component designs with this configuration, the impedance of the pull-up is higher than that of the pull-down. Especially important is the fact that the pull-up loses drive as the output voltages approaches the SV supply on a low-to-high transition, whereas the pull-down transistor will not suffer this affect on a high to low transition. 128 INTEGRATED LINE TERMINAL SOLUTIONS FOR HIGH-SPEED LOGIC CONFERENCE PAPER CP·10 Figure BB shows the same structure with an integrated a series resistor on the output. The effective pull·up and pull· down impedance is now matched better to the line, but the pull-up impedance still remains higher than the pull-down impedance. More importantly, because this configuration creates a source follower, the series resistor exacerbates the relationship between the output voltage and the loss of pull-up drive. An integrated series resistor, directly connected to an output pin, if not properly designed into the component, can easily be damaged by system transients. An additional reason for line termination problems in high speed logic is the fact that some drivers have different drive capability between the high going and low going signals. This makes it difficult to impossible to make a perfectly matched transmission line when the signal has different driving impedances for the two states. To avoid this problem many of the new components that have internal resistors also have a balanced drive capability so that both high going and low going signals demonstrate the same output impedance. Limiting the pull down drive capability of the device also reduces problems with undershoot, ground bounce and general noise generation. Reducing these problems will make the whole task of termination easier. --l Benefits of Integrated Series Termination --l FIG BA FIG BB FIG BC 1) Captures all of the benefits of external series termination a) No static power consumption b) Reduces dynamic power consumption c) Moderately good for stubs d) Noise reduction over other termination methods e) Reduces component ground bounce 2) Balanced drive produces less signal distortion 3) No external components are needed for termination a) Reduced part count, inventory, and costs b) Easier board layout 4) Terminated parts are pin compatible with non terminated parts a) Plug in replacement. b) No new CAD models 5) Higher drive capability than when using external series resistors FIGBD Figure 8. Various Integrated Output Configurations Placing the series resistor in the drain of both transistors as shown in Figure BC alleviates the problems with the source follower created in configuration of BB. The configuration in BC will provide good balanced drive and good line termination for both high to low transitions and low to high transitions. The value of the resistors can be adapted to match the line impedance without having to make compromises for the opposite transitions. Despite this, the resistor in the drain of the pull down N channel transistor is still exposed to the output pin and may short if stressed. In Figure BD the resistors are sized and placed in the drain of the pull-up and the source of the pull-down transistors to provide good balance between the two paths. The resistors are isolated from the external pin and are now protected against shorting. An added advantage is that a lower value resistor can be used to generate the same drive current that can be generated with a higher value resistor used in configuration BC. Several IC manufacturers have introduced components with integrated series resistors. Each of the configurations shown in figure B is available from at least one source. Balanced Output Drive Capability (m. Disadvantages of Integrated Series Termination 1) Captures some of the disadvantages of external series termination a) May not drive very low impedance backplanes b) May not have first incident wave switching with distributed loads 2) Some components are susceptible to damage (Check with manufacturer). CONCLUSION Olthe methods of termination available, the series termination works best in most applications, especially those situations utilizing shorter distances. Other termination methods have applications in which they will work best, but their overall performance is not as good as series termination in board level applications. When integrating the series resistor into the line driving component there are improvements that can be seen over the use of an external resistor, but careful consideration needs to be given to how the resistor is integrated to achieve optimum results. (1) "Series Termination," Application Note 50, 1990/91 Logic Data Book, Integrated Device Technology. 3014 drw 08 Note: This paper was presented at Electro192, May 12, 1992 Figure 9. Balanced Drive Output 129 ~ GJ ~ NEW OUTPUT STRUCTURE DESIGNS REDUCE SYSTEM NOISE PROBLEMS CONFERENCE PAPER CP-11 Integrated DevIce Technology. Inc. By Stanley Hronlk INTRODUCTION The new high-speed logic families available in the market place are capable of much higher data rates than earlier families. This is due to a combination of shorter propagation delays in the parts. faster output edge rates, and reduced output switching voltages compared to the CMOS switching levels. These component changes are requiring strict adherance to the design rules to maintain signal integrity. To assist designers in the task of maintaining high speed, low noise operation, many component manufacturers are developing new output structure designs to assist in signal waveform control. The output structures of the new devices are designed to control internal chip noise and to treatthe attached circuitry as a transmission line. Three of the techniques to accomplish this are adjusting the output voltage levels, controlling the output edge rate, and reducing reflections in the transmission line. Each of these techniques will lower the drive current surge in the device output structure which will also assist in controlling ground bounce and Vcc bounce. Reducing the output voltage swing from CMOS levels to TTL levels allows the signal to acquire state more quickly. The lower voltage swing will reduce the charge dumped into the transmission line which will reduce the energy and noise in the signal transmission. The output edge rate control and the transmission line effects are interrelated and must be approached as a unit. TRANSMISSION LINE EFFECTS DUE TO FAST EDGE RATES To determine if a transmission line effect is present, the equation in Figure 1 may be used to determine the maximum unterminated trace length. The equation states that if the risel fall time of the signal is shorter than the round trip propagation delay in the signal path, the signal path will act like a transmission line and need termination. As an example, lOT's FCT-T family has an edge rate of less than WIns which will require termination on any trace that is over approximately 3 inches. Because of the speed of newer logic families, the need for proper termination is becoming universal. To simplify board designs a few manufacturers are integrating termination schemes into the output structure of their devices to overcome the need for adding additional termination to the board. By integrating termination schemes into the output structure of the components,other signal quality improvements can also be obtained. To establish the benefits of integrated termination schemes, a few transmission line characteristics need to be understood. V Tf = Rise or Fall Time Tpd = Delay or Transmission Line Per Unit Length L = Length of the Transmission Line L> Tf 2Tpd If the equation is true, the line is transmissive and may need termination Figure 1. Transmission Line TRANSMISSION LINE PROBLEMS To maintain good signal quality on a transmission line, the dynamic impedance of the transmission line needs to be constant throughout the line. This means that very low source impedances, stubs, and high-impedance receivers without line matching termination can all contribute to poor signal quality. If there are abrupt changes in the impedance at any point in the transmission line, reflections will bounce off these points and may cause noise. In designing the transmission line it is best to utilize a single line from the source to the load by daisy chaining distributed high impedance loads and avoiding stubs which cause impedance mismatches. If lowimpedance loads are to be driven, they should be at the end of the transmission line opposite the source. When a driver changes output state and sends a signal down the transmission line, the driver will sense the transmission line but does not have the capability to sense the load at the far end of the line until the signal has had a chance to travel down to the load and reflect back. Because of this the driver for high-speed signals which meet the requirements of the equation for a transmission line as shown in Figure 1 should be optimized to drive the transmission line and not the load. A transmission line has the same impedance for both highgoing and low-going signals. Contrasting this, most transmission line drivers have different drive capabilities in the high and low going directions making it difficult to maintain a high signal integrity for both transitions because of the impedance mismatches between the two directions. The lOT logo is a registered trademark of Integrated Device Technology, Inc. ©1993 Integrated Device Technology, Inc. 130 5/92 NEW OUTPUT STRUCTURE DESIGNS REDUCE SYSTEM NOISE PROBLEMS CONFERENCE PAPER CP-11 Series Termination One of the most successful methods of dealing with transmission line effects is the use of series termination. Series termination is the addition of a resistor between the component output and the transmission line, as shown in Figure 2. The combination of the series resistor and the source impedance of the driver act to match the characteristic impedance of the trace or transmission line. Open Circuit 1: \ ..._ _ _ _ _ _ _.... Signal at transmitting end Signal at receiving end I=- I-= I-= 1 ~ Signal at Transmitter \ I ,~------~, ! drw 02 I drw03 Figure 3. Signals Produced When Using a Series Terminator Figure 2. Transmission Line with Series Termination As the output driver undergoes a transition, the voltage is divided between the output impedance (driver plus the resistor) and the transmission line, as shown in Figure 3. If the output impedance matches the impedance of the transmission line, the output waveform will develop a step at 50% of the voltage transition. If the impedances are not matched, the voltage will divide proportionately to the impedances. To achieve first incident wave switching, the source impedance should be slightly lower than the impedance of the transmission line so that the step appears after crossing the logic threshold of the receiver. After the driver makes a transition, the signal will propagate down the transmission line and strike the far end. Upon reaching the far end, the signal will look for an impedance matching that of the transmission line. Since the receiver is usually a high-impedance input, the voltage will build up at this point from the current flow until the voltage level has reached a high enough level to counter the inductive nature of the transmission line. If the source impedance matches the line impedance and there were no losses, the voltage would build to twice the driving voltage. At this point, the matching impedance is seen back on the transmission line and the signal will propagate back down the transmission line to the source. Since the driving voltage was about half of the final state voltage, this will bring the voltage in the entire transmission line to the proper level, as shown in Figure 3. If a designer has a distributed load, and the step voltages shown in Figure 3 do not cross the logic threshold of the daisychained load, first incidentwaveswitching may not be achieved. This can be countered by slightly reducing the series resistance values to adjust the step voltage levels so that the steps occur beyond the logic thresholds and first incident wave switching can be guaranteed. This will tend to overdrive the transmission line, but countering this effect will be the "lossy" nature of the line. The series resistor significantly cuts the drive current into the line alleviating most noise problems at the source. Resistive and capacitive elements of the transmission line will absorb portions of the charge which will quickly quiet the signal. 131 Using the series resistor limits, the charge that the source passes into the transmission line reduces the drive current surge of the source. This also reduces the need to dissipate the energy from the charge after the transmission line has reached full voltage levels which will reduce noise spikes on the line, cross talk, power dissipation, noise problems in the power distribution system, and reflection problems. In most situations, series termination is the simplest, most effective form of reducing signal noise in transmission lines. Series termination consumes no DC power, helps to reduce dynamic power consumption, reduces ground bounce and handles stubs better than most end of the line termination schemes. INTEGRATED SERIES TERMINATION Benefits can be received by integrating the series resistor into the output structure of the driving component. By integrating the resistor, the output impedances of the driver can be matched to the typical load of a transmission line for both the high and low going signals, which is not possible with external termination. Properly designed integrated schemes can outperform external resistor terminations by reducing source follower problems and increasing the output drive current. The removal of the external resistors will also reduce part count. To demonstrate the effects of various output structures, a review of several output configurations will be given. NEW OUTPUT STRUCTURE DESIGNS REDUCE SYSTEM NOISE PROBLEMS CONFERENCE PAPER CP·11 Figure 4. Totem Pole output structure with an N channel pull up transistor. This structure is excellent for backplane driving and low impedance loads. Figure 6. Drain resistors overcome the gain problems with the external resistor. The configuration in Figure 4 shows a totem pole structure with N channel FETs in both the pull up and pull down positions. The pull up FET requires a Gate to Source turn on voltage of about 1.5V which will drop the typical loaded output voltage to less than 3.5V giving TIL level outputs. In this configuration, the impedance of the pullup is typically higher than that of the pulldown, giving unbalanced drive capabilities in the two directions. With lOT's High Drive components that contain this configuration, the impedance is usually about 60hms pull down and 220hms pull up. This structure gives excellent drive capability for low-impedance backplanes and heavy loads that contain external termination. PlaCing the series resistor in the drain of both transistors, as shown in Figure 6, alleviates the problems with the gain created in the configuration of Figure 5. The value of each resistor can be adapted to match the line impedance, thus giving similar drive currents in both directions without conflicting with the opposite transition. The configuration in Figure 6 will provide a good balanced drive capability and good line termination for both high-to-Iow transitions and low-to-high transitions. In Figure 7, the resistors are sized and placed in the drain of the pull-up and the source of the pull-down transistors. With this configuration the same results as shown in Figure 6 can be achieved by properly sizing the resistors. lOT uses this structure in the Balanced Drive IDTFCT162xxx family. Figure 5. Totem Pole with a series resistor which simulates the addition of an external series resistor. Figure 7. A drain pull up and source pull down allow better drive in the pull down (IDTFCT162xxx). Figure 5 shows the same structure with a series resistor on the output. The effective output impedance is now matched more closely to the line, but the pull-up impedance remains higher than the pull-down impedance. Because the series resistor degrades the gate to source voltage drop in the pull up FET, the pull up drive capability is reduced, significantly increasing the effective impedance beyond the value of the resistor. The pull down drive is not affected by gate to source degradations. This creates an unbalanced condition, where the pull up drive capability is much lower than the pull down drive strength. The pull up may degrade to an impedance of over 1OOohms when utilizing a 250hm resistor while the pull down will remain at about 31 ohms with the same resistor. This problem will be experienced either when an external series resistor is used or when the resistor is integrated into the output structure of the device. Using resistors in the output structure of Figure 7 provides a natural internal current limiting capability to the component which will then allow a restructuring of the driving stage. By doing this it is possible to speed up the internal logic of the component to where the output switching will begin more quickly than is possible with the other configurations shown. Driving Capacitive Loads In situations in which the loads on the transmission line are highly capacitive, the use of series termination may cause time delays due to the RC time constants. When using lOT's Balanced Drive components, this problem is reduced because of the lower output impedance, quicker response time, and controlled edge rate. lOT's Balanced Drive and High 132 NEW OUTPUT STRUCTURE DESIGNS REDUCE SYSTEM NOISE PROBLEMS Drive components have very similar response times for capacitive loads of less than 300pf. Because of its quicker response, the Balanced Drive part will switch faster than a High Drive part for loads of less than 200pf. As the load increases above 200pf, the High Drive part will respond more quickly because of its lower output impedance. Despite these minor differences, the differences in the effects of capacitance on propagation delay between the High Drive and Balanced Drive parts should be less than O.Sns for any load of less than 300pf. The capacitive derating curves are shown in Figure B. CONFERENCE PAPER CP-11 both directions, limiting surge current into the transmission line, butstill providing sufficient drive currentto drive all except the most demanding loads. Limiting the pull down drive capabilityofthe device during transitions also reduces problem s with undershoot, ground bounce and general noise generation. During steady state DC conditions, IDT specifies the static drive capability of the devices at 24ma worst case in both directions as shown in Figure 10. Delay vs Capacitive Load 8~r-----~----~----------------, 7 6 5 drw09 tpd (LH) 3 Figure 9, Dynamic Balanced Drive Output o 50 100 150 200 250 300 350 Capacitive Load (pi) 400 drw 08 Figure 8, Capacitive Derating for Double Density Balanced Drive and High Drive When the capacitive load is larger than 400pl at high switching speeds, the load will begin looking like a direct short upon switching and may begin to overdrive the output of both the balanced drive and high drive part. The power dissipation equation is P =fCV 2 and should be used to calculate the power consumption for each pin to assure that the total power dissipation in the device does not exceed the maximum rated limits. To avoid exceeding the limit for heavy capacitive loads, a 1000hm series resistor can be used to limit the current to acceptable levels. Balanced Drive Capability The output structure of the Balanced Drive part has been specifically designed to give similar transition times with matching absolute edge rates in both the high-going and lowgoing directions. This means the IDT Balanced Drive family has a matched dynamic drive capability for driving transmission lines, but still has sufficiently low static output impedance to maintain state during during steady state conditions. As shown in Figure 9, during transitions, the Balanced Drive output structure will current limit at about 11 Oma, typical for ~ ~ ~ ~ F drw 10 Figure 10. Static Balanced Drive Output Adding External Resistors to Balanced Drive With the Balanced Drive components, the effective series termination impedance of about 220hms is slightly lower than the characteristic impedance of most transmission lines. Backplanes may sometimes approach levels as low as 300hms, but most lines will have a higher impedance. Despite this difference, the current surge driven into the transmission line will be limited by the output impedance and controlled edge rate. This will usually be sufficient to provide quiet operation without the addition of additional termination. If it is desirable to further enhance the quiet operation and to achieve a close match between the source impedance and the line impedance, the user may add external resistors to the output of his balanced drive parts. Because of the integrated resistor in the drain of the pull up FET in the Balanced Drive part, the addition of an external resistor will not degrade the gate to source voltage to the extent that the gain degradation 133 NEW OUTPUT STRUCTURE DESIGNS REDUCE SYSTEM NOISE PROBLEMS would take place with the standard TTL output of figure 4 for equivalent total resistance values. The voltage drop across the drain resistor will increase the gate to source voltage, partially countering the effect of the external series resistor and retaining some of the balanced drive attributes. As a result, adding an external resistor to a Balance Drive part will not cause a significant impedance mismatch between the high-going and low-going signals, as happens with normal TTL output components. CONFERENCE PAPER CP-11 CONCLUSION Utilizing the Balanced-Drive family in high-speed bus driving applications will solve most of the problems with noise and line termination experienced with traditional high-speed, low output impedance line drivers. The Balanced Drive logic family provides the benefits of series termination on signal lines while avoiding many of the drawbacks of adding external series termination resistors. The Balanced-Drive family of devices from IDTwili generate lower levels of noise than other components in the same speed grade, solve many of the termination line problems encountered by the designer, reduce power consumption and give superior performance to other high-speed logic. Balanced Drive is the logic family of choice for most applications. 134 ~ CONFERENCE PAPER CP-12 DOUBLE-DENSITYTM LOGIC: THE FAST, FLEXIBLE LOGIC FAMILY GJ L/ Integnoted Device Technology,lnc. By Wayne T. Yoshimoto Double-Density is comprised of three distinct versions: ABSTRACT This paper discusses how the FCT-T Double-DensityTM Logic family addresses the problems faced by PC designers. With microprocessor frequencies rising both in desktops and portables, logic must also rise in performance, but still maintain good noise and power characteristics. High speed/low noise operation for speed-critical paths and high speed/low power operation for battery operated systems are ideal for the designer. Since space is a premium both in portable and desktop PCs, high speed, low power, and low noise must be combined with small packaging. INTRODUCTION 16XXX 162XXX 163XXX Balanced 64MA Output Drive 1-----1 Output Drive 3.3V on Chip Resistors drw01 Figure 1. Double-Density Versions The high-drive version (FCT16XXX) provides-32mA/64mA (Ioh/lol) drive with ground bounce typically less than 1V. Power-off disable is supported for "live" insertion in fault tolerant systems. This version is a drop-in upgrade for ABT and ACT WideBus™ meeting or exceeding all DC specifications. The balanced drive version (FCT162XXX) provides-24mN 24mA (Ioh/lol) drive with integrated series terminating resistors. This reduces ringing/undershoots and ground bounce is typically less than .6V. These are pin-compatible with the high-drive family providing easy upgrade. The 3.3V version (FCT163XXX) is deSigned to meet the JEDEC LVTTL standard for 3.3V operation, not a recharacterized or derated 5V device. It provides a drive of 8mA/24mA (Ioh/lol) and ground bounce typically less than .3V. A number of unidirectional and bidirectional translators are part of this group also to provide solutions for mixed 3.3VI 5V hybrid systems. All three versions of the family are well suited in the following categories and thus provide solutions for PC designers of different types of systems. The level of integration and the rise in frequencies of the microprocessor has had an enormous impact on the PC marketplace. The performance attained only by huge mainframes 15 years ago can now be realized on the desktop. Similarly, the performance and battery life attainable 10 years ago in a 25 pound monster is today far surpassed by a 5 pound, 386-class machine. Correspondingly, the interface logic (buffers, latches, transceivers, etc.) surrounding the microprocessor has gone through significant improvements. Designers are beginning to realize that the older families of interface logic no longer meet the needs of today's and tomorrow's PCs. Identified shortcomings are in the areas of speed, power consumption (both static and dynamic), packaging, and noise. For example, in a desktop PC running at 50MHz, designers were forced to look elsewhere instead of the popular, butslow, "F" or "AS/ALS" families which were adequate for a 25MHz design. In the case of the newest notebook PCs, designers need to replace the slow "HCT" and high power "ACT" families. Newer families of logic, such as FCT-T Double-Density Logic, can provide designers of both high-performance desktops and low-power notebooks with the best of both worlds, as well as new capabilities, such as 3.3V operation. In order to keep up with the ever-increaSing frequencies of the microprocessor, logic products are constantly pushed to attain faster and faster speeds. Today, with the systems running at 50MHz, logic products with delays of less than 5ns are desirable. FCT-T DOUBLE-DENSITY DEFINED PERFORMANCE OF LOGIC FAMILIES FCT-T Double-Density is one of the IDT families of highperformance CMOS logic. It is comprised of 16-, 18-, 20-bit wide, extra-quiet, very-low-power, high-performance bus interface logiC products. Double-Density is an extension of IDT's earlier family of octal FCT -T devices. Microprocessors' buses have moved beyond 8-bits, creating the need for wider bus interface devices. With the wider devices, partcount is minimized as well board area. PERFORMANCE 245 Transceiver ACT - - FCT-T Octal 3.8ns 4.1ns 4.6ns 7.0ns DD High Drive 3.8ns 4.1ns 4.6ns 7.0ns DD Balanced Drive 3.8ns 4.1ns 4.6ns 7.0ns 4.6ns 7.0ns FAST ABT is a trademark of Integrated Device Std-Speed - ASiALS Double~Oensity - - DD3.3V The lOT logo is a registered'trademark: and Technology, Inc. O-Speed C-Speed A-Speed HCT - - 29ns 9.0ns 7.0ns - - 7.5n5 - - 4.6n5 IbIOt 5/92 ©1993 Integrated Device Technology, Inc. 135 DOUBLE-DENSITY LOGIC: THE FAST, FLEXIBLE LOGIC FAMILY CONFERENCE PAPER CP-12 FCT-T Double-Density provides the performance necessary for today's high-performance systems with the power consumption one would expect from CMOS technology. With future systems in mind, lOT has recently announced D-speed logic. AD-speed '244 buffer specifies a maximum of 3.6ns propagation delay. This is ideal for high-performance, 75Mhz and 100Mhz, workstations and 67Mhz+ P5 designs. Logic has come a long way from the days of LS, TTL, and even HCT. A comparison of the old and new generation of logic is shown in Table 1. 60mA 90mA 90mA ~ ICCZ ~ ICCH IS:'! ICCl POWER Running at high frequencies is not without its drawbacks. Since the amount of power dissipated by CMOS is proportional to the frequency, high-frequency operation also means high power. Figure 2 compares the dynamic power consumptions versus frequency of several logic families. As illustrated by the graph, Double-Density provides better power characteristics at all frequency points benefiting both the desktop (high-drive family) and the notebook (balanced drive/3.3V family) designer. 20 0 ~ .s Iz w a: a: 10 ::> 0 0 ..J ~ 0 I- 0 20 30 40 FREQUENCY (MHZ) 50 drw 02 Figure 2. Dynamic Power Comparison Dynamic power is important when the system is active. During inactivity, static power consumption becomes equally important. The CMOS technology families on the graph, Double-Density and ACT, benefit from very low static power consumption on the order of a few microamps. On the other hand, the other families, FAST (bipolar) and AST (SiCMOS), suffer from high-static-power consumption. See Figure 3. FCT16245 Double Density ABT16245 Widebus™ FASfTM (two octals)* drw03 Figure 3. Static Power Comparison NOISE Noise issues came to the forefront with the introduction of the first generation of fast CMOS logic in the mid-eighties. Most of these issues were smoothed out by the second generation, with the inclusion of output edge rate controls and reduced output swings. In addition, designers were aware of high-speed design techniques such as series termination and transmission line effects. The Double-Density family makes ground bounce and noise non-issues. The 48-/56-pin SSOP package provides multiple Vcc (4) and Ground (8) pins forthe entire family of 16-bit wide devices. See Figure 4 for the pinouts of a 48-pin SSOP package. As a reSUlt, the high-drive version specifies typical ground bounce of less than 1V. The balanced-drive version takes advantage of series termination to reduce noise. They provide integrated series resistors, bringing the typical ground bounce specifications even lower, to less than .6V. The 3.3V version specifies a typical ground bounce of less than .5V due to a smaller output swing of about 3.0V. 10E 101 102 GND 103 104 vee 105 106 GND 107 108 201 202 GND 203 204 vee 205 206 GND 207 208 20E 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32 31 30 29 28 27 26 25 lLE 101 102 GND 103 104 vee 105 106 GND 107 108 201 202 GND 203 204 • Flow-thru Pinouts • 4 Vccs, 8 Grounds • 25-mil pin pitch vee 205 206 GND 207 208 2LE Figure 4. SSOP Package 136 drw04 DOUBLE-DENSITY LOGIC: THE FAST, FLEXIBLE LOGIC FAMILY CONFERENCE PAPER CP-12 Double-Density family specifies output skew at typically less than 2S0 ps, and guaranteed less than SOO ps. These characteristics allow the IDT74FCT16240/244 to be used as minimum skew clock drivers. SSOP SO-56 ill 0.504 ITl ~ []06~ Dl7~ ~ 14- 3.3V OPERATION 3.3V operation will provide the next step in portable computing. Lighter, smaller, and longer lasting systems will be possible. For example, a 7" x 9" x 2", 4 pound, eight-hour system will be possible in the near future. IDT is ready for the conversion to a 3.3V supply voltage with the 3.3V DoubleDensity family of bus interface functions, key "glue" logic elements, and 3.3V to SV interface components. This family will allow designers to develop complete 3.3V or mixed supply 3.3V/SV systems. These devices have been designed to operate from a regulated 3 .3V supply for optimal performance and are not simply "recharacterized" or "derated" SV CMOS parts. Thus, they are able to maintain sub-Sns propagation delays for high-performance portables. See Table 3 for a summary of IDT's 3.3V Double-Density offerings. ~ 0.406 0.406 0.409 drw 05 Figure 5. SOIC vs. SSOP Package Comparison SOIC VS. SSOP PACKAGE (2) 20-pin SOIC (1) 48'pin SSOP (1) 56-pin SSOP Area (sq. Inches) 0.474 0.254 0.295 Body Width (Mils) 0.300 0.300 0.300 Area (Normalized) DOUBLE-DENSITY 3.3V OFFERINGS 1.00 0.536 0.523 tbl02 PACKAGING Packaging is rapidly becoming an important consideration, due to the integration and size limitations of today's systems. Luckily, packaging technology has not been a limiting factor. Today's highly integrated boards in desktop and notebook PCs have been a result of new packaging technologies. The Double-Density family is packaged in a 2S-mil pin pitch 48-/ S6-pin SSOP package (refer to Figure 4 in the last section). A comparison of the SSOP package versus the standard SOIC package can be seen in Figure S. A 47% area-savings can be realized when using the 48-pin SSOP package versus a 20-pin SOIC package. Future packaging capabilities are currently being studied. After all, the trend is toward smaller and smaller packages. Some of the alternatives are: a 21 O-mil wide, 19.6-mil pin pitch package or a 1S0-mil wide, 19.6-mil pin pitch package or a 1SO-mil wide, 1S.7-mil pin pitch package. The area savings versus the current SSOP package is 41%, S4%, and 62% respectively. This decision can not be made in a vacuum, since manufacturability must be considered. As the pin pitch and overall package size shrinks, fewer board manufacturers are able to handle production without large investments in new machinery. These considerations will weigh heavily upon where the next step in packaging will be. SKEW Skew is defined as the absolute value of the difference between the actual propagation delays for any two separate outputs of the same device. With processor clock frequencies migrating to SOMHz and 67MHz+, any delay due to skew increasingly becomes a percentage of the clock period. The 137 Part # Function Avail FCT163244 Buffer/Line Driver NOW FCT163245 Bidirectional Transceiver NOW FCT163373 Transparent Latch NOW FCT163374 Register (3·State) NOW FCT163501 Registered Transceiver NOW FCT163646 Transceiver/Registers NOW tbl03 High-speed desktops and fileservers will also benefit from 3.3Voperation. Their large memory requirements will force the use of large density DRAMs, 16Mb and 64Mb. Due to the gate geometries of these dense memories, a 3.3V supply voltage will be necessary. Also, at greater than SOMHz, these high-speed systems will benefit from the reduced swings in terms of performance and noise. 3.3V/5V TRANSLATION Using mixed supply systems will be a necessary transitional phase between the current 5V systems and the upcoming full 3.3V systems. Technological and performance limitations will slow down the conversion of some functions to 3.3V. Therefore, designers must understand the compatibility issues of interfacing between 3.3V and SV components, as well as what translation functions are available today. IDT's 3.3V Double-Density family is designed such that any input pin (except I/O) can take an input voltage up to 7V. This feature allows these 3.3V devices to be directly driven by a SV device. This also allows the unidirectional parts, i.e. FCT163244, FCT163373, FCT163374, to act as one-way SVto-3.3V translators. The unidirectional translators accept TTL-or CMOS-compatible signals from SV components at the inputs, and provide TIL-compatible output levels. These translators avoid any problems associated with a direct interface between a SV component and a 3.3V bus. DOUBLE-DENSITY LOGIC: THE FAST, FLEXIBLE LOGIC FAMILY CONFERENCE PAPER CP-12 Since not all bus interface applications are unidirectional, there also needs to be bidirectional solutions to the 3.3V to 5V interface. A solution to this is the FCT164245T. This part interfaces between a 3.3V bus and a 5V bus in a mixed 3.3V/ 5V supply environment, handling all ofthe 3.3V/5Vbidirectional interfacing issues (Figure 6). The FCT164245T is configured as a 16-bit '245 transceiver, giving the designer tri-state capabilities and the ability to select either bidirectional or unidirectional modes. The translator has separate 3.3V and 5V supply and I/O pins both on separate sides of the device. The translator part also features power-off disable, which allows the device to withstand the maximum rated voltages on the inputs and outputs while it is unpowered or partially powered. This is a desirable feature for bus-interface logic in a portable system, because of power management schemes which power down certain subsystems when inactivity is detected. The bus interface device will then have to buffer the active system bus with a powered-off device without damaging itself. 3.3V BUS 3.3V 5V 5V BUS 164245T TRANSLATOR drw06 Figure 6. Bidirectional 3.3V15V Interface CONCLUSION As the complexity and frequency of the microprocessor increases, surrounding glue and bus interface logic must keep pace. lOT has done so with the introduction of the DoubleDensity FCT-T Logic family. This family excels in all categories, including performance, noise reduction, power consumption, skew, and packaging. It also provides innovative 3.3V and 5V/3.3V translation options. The Double-Density Logic family provides solutions for a variety of PC applications ranging from a 3.3V subnotebook to a high-performance fileserver. REFERENCES 1. Alba, Manuel; "Application of High Speed/Low Power Logic to PCs"; SVPC'92 Proceedings. 2. lOT; "High-Performance Logic Data Book"; 1992. 3. Yoshimoto, Wayne; "Mixed Supply Design Considerations in Notebook PCs"; SVPC'92 Proceedings. NOTE: This paper was presented at WesCon'92, November 17, 1992. 138 CONFERENCE PAPER CP-13 MIXED SUPPLY DESIGN CONSIDERATIONS IN NOTEBOOK PCs Integrated Device Technology, Inc. By Wayne T. Yoshimoto INTRODUCTION The explosive growth of the portable Personal Computer (PC) market has manufacturers clamoring for market share by differentiating themselves in battery life and weight from their competitors. The push for longer battery life and lighter weight has reached a point where improvements via traditional design techniques alone are not satisfying users. In order to make the larger, quantum leap to eight or more hours of battery life and weight less than 4 pounds, designers must move to a 3.3V supply voltage. Although most of the components are currently available at 3.3V (Le. processor, chipset, DRAM, glue logic), the next generation of portable PCs will have to be "hybrid" systems with both 3.3V and 5V supplies until some of the missing components, namely the disk drive and LCD driver, become available. Even after a full 3.3V system is available, the designer must consider the interface with external peripherals, i.e. printers, modems, etc., since these are not likely to move to 3.3V in the near future. The designer will be faced with new challenges pertaining to designing in the mixed supply environment. MARKET OVERVIEW 1992 marks an explosive point in the portable PC market. Portables have invaded the mainstream and are appearing everywhere from the street corner to the office. Dataquest estimates that by 1994, portables will account for over onethird of the PCs shipped. This shift to portable computing reflects the strides made in the basic technology; they are now powerful enough to replace desktop PCs. There is now no need to have a desktop for the office and a portable for the road. 6 5 • Notebooks c==J Sub notebooks mlml Palmtops 4 !l '2 :::l '0 c: 3 ~ :E 2 o 1991 1992 1993 1994 1995 drwOl Figure 1. PC Worldwide Market Forecast Source: Phoenix Technologies Inc. The lOT logo is a registered trademark of Integrated Device Teohnology, Inc. ©1993 Integrated Device Technology, Inc. 8/92 139 MIXED SUPPLY DESIGN CONSIDERATIONS IN NOTEBOOK PCS CONFERENCE PAPER CP-13 The market is broken up into a number of segments based on weight: desktops, transportables, laptops, notebooks, subnotebooks, and handhelds. As a whole, the PC market is fast growing with a Compounded Annual Growth Rate (CAGR) of almost 10%. A closer look at the cross section of the market reveals that notebooks and subnotebooks are fueling this growth with a CAGR of 34% and 79% respectively, while the other categories are flat (Figure 1). In order to continue the projected rapid growth, designers must push the battery life and drive the weight down to points way beyond today's limits. To do this, notebook and subnotebook PCs must move to a 3.3V supply voltage. BENEFITS OF A LOWER SUPPLY VOLTAGE Moving to a 3 .3V supply voltage provides extended battery life and contributes to lighter and smaller systems. Component reliability, and thus, overall reliability is also enhanced resulting from lower heat buildup and a reduction in boardlevel noise. The end user can expect one of two things: lighter and smaller systems or higher-performance systems with more features, such as built-in faxlmodem or color displays, for the same weight. Decreasing the supply voltage directly increases the battery life. It reduces the power consumption exponentially, leading to a 56% savings [1-(3.3 2/5 2 )] at 3.3Vversus 5V. The power savings also can amount to less weight because smaller and lighter batteries can be used to provide the same amount of battery life. For example, a 5V notebook typically utilizes five 1.2V NiCd batteries while a 3.3V notebook would use only three batteries, thereby reducing the weight by 40%. On the other hand, the same number of batteries can also be used to provide a system with more built-in features or an even longer battery life. Lower power consumption also results in less heat generation, which leads to fewer cooling requ irements and higher reliability of the overall system. Noise generation has become a problem due to ihe sharp rise and fall times required by today's high speed systems. By decreasing supply voltage, the voltage swing olthe 3.3V parts outputs (typically3V) is reduced, compared to CMOS (rail-torail) swings (typically 4.8V) and TIL-compatible swings (typically 3.3V). This will reduce noise generation, which makes board level design easier, and also will reduce EMI radiation, requiring less shielding to obtain FCC approval. Lighter weight systems will be possible with the reduced shielding requirements. lOT 3_3V LOGIC CHARACTERISTICS lOT provides families of optimized low voltage (Vee = 3.3V±.3V) logic functions which allow designers to develop complete 3.3V systems. These families have been designed to operate from a regulated 3.3V supply for optimal performance and are not simply "recharacterized" or "derated" 5V CMOS parts. These "true" 3.3V families contain 16-bit (IDT74FCT163XXX) and 8-bit (IDT74FCT3XXX) bus interface functions, key "glue" logic elements, and 3.3V to 5V interface components. lOT's leading-edge CEMOS"" 5E process achieves the combination of very low power can- sumption and high speed. See Table 1 for a comparison between lOT 3.3V FCT and other 3.3V products. 3_3V LOGIC COMPARISON MFR. Device TPD (max) ICCL(typ) ICCD(typ) lOT FCT3245A 4.6ns 50uA 60uAlMHz lOT FCT163245A 4.6ns 50uA 50uAlMHz TI LVT 4.6ns 15mA - 10.5ns 5uA 221uAlMHz NSC LVQ245 tbl 01 Although TI's LVT family provides high speed by going to a BiCMOS technology, this is at the expense of static power consumption. National's LVQ family, which is merely a recharacterized version of its FACTQ family, has very slow speeds and high dynamic power. No other 3.3V family provides the high speed and low power combination of lOT's 3.3V logic families. In addition to speed and low power consumption, lOT's 3.3V logic families provide the following features: • True TTL compatibility with 400mV noise margin. • 50% area savings versus standard SOIC packages from SSOP packaging • Input and Output leakages less than 500 nA. • Propagation delays match those of 5V FCT logic for like functions • All inputs (except 1/0) can be driven from either 3.3V or 5V components • Extended commercial temperature range of-40°C to +B5°C. • ESD >2000V per MIL-STD-883, Method 3015. All of lOT's 3.3V products operate functionally with Vce as low as 2.5V, allowing them to be used in unregulated battery driven systems. WHY MIXED SUPPLY VOLTAGES? Mixed supply systems will be a necessary transitional phase between the current 5V systems and the upcoming full 3.3V systems. Technological and performance limitations will slow down the conversion of some functions to 3.3V. For example, disk drive and LCD driver components will not run off of a 3.3V supply until the end of 1992. Therefore a full 3.3V system will not be available until mid-1993. A mixed supply system, on the other hand, can be realized today. Even a mixed supply system provides enough of a gain in battery life and reduced weight that any manufacturer who waits until a fu1l3.3V system is possible will be at a competitive disadvantage. 140 MIXED SUPPLY DESIGN CONSIDERATIONS IN NOTEBOOK PCS CONFERENCE PAPER CP·13 3.3V driving 5V 5V driving 3.3V 3.3V Device 5V Device Bidirectional dow 02 Figure 2. 3V/SV Interface The mixed supply design concept will not go away once a fu1l3.3V notebook is possible. The problem of interfacing with external 5V peripherals will still exist. Issues faced while designing the 3.3V /5V interfaces on the motherboard will also be faced in the external interfaces. INTERFACING 3.3V TO 5V COMPONENTS In the new design environment of mixed supplies, the designer needs to consider the compatibility issues of inter· facing between 3.3Vand 5V components. Figure 2 shows the different 3.3V/5V interface scenarios. Although the 3.3V fam· ily is TTL·compatible with 3.3V and 5V devices, there are situations where excess current or voltage can develop and damage the components. The designer must understand these situations and provide for them. 3.3 Volt TTL S Volt TTL 7.0 Fora 3.3V device driving a TTL compatible 5V device, there are no interface problems. The logic input thresholds for the 5V TTL part are VIL = O.BV and VIH = 2.0V. The 3.3V device is guaranteed to drive a LOW of less than 0.4 V and a HIGH of greater than 2.4V at the rated output load currents. This provides ample noise immunity on both ends making this interface problem-free (Figure 3). A 3.3V device will not drive a non-TTL compatible CMOS 5V device, i.e. AC or HC, since the input voltages required are higher than 3.3V, typically 3.B5V. For a 5V device driving a 3.3V device, there are some potential interface problems which must be understood and dealt with; 5V logic will drive 3.3V logic's inputs (except 110) directly without damage (Figure 4). Some manufacturers have an internal clamp diode from the input or output pin to Vcc in order to boost ESD protection. In these cases, directly driving from a 5V device will forward bias the diode and cause excessive current to flow into the 3.3V supply. lOT's 3.3V Logic family implements ESD protection elsewhere and does not suffer from this problem. Thus, driving from a 5V device is not a problem (Figure 5A). However, on outputs and lias, there is a clamp diode to Vcc limiting the voltage applied to outputs to Vcc+.5V. If a voltage greater than Vcc+.5V is applied, the clamp diode will forward bias and possibly damage the device (Figure 58). Here the designer must be careful of the output swings of driving devices. Directly driving the 3.3V part from a CMOS swing (rail-to-rail) 5V device could exceed the absolute maximum and damage the device. As long as the 3.3V and 5V supplies track with each other, reduced-swing, TTL-compatible devices can safely drive 3.3V devices. S Volt TTL Input and Output Breakdown 7.0V 3.3 Volt TTL 7.0 6.0 Input Breakdown 7.0 Volts 6.0 VOH (max) S.SV 5.0 5.0 4.0 4.0 3.0 3.0 VOH (max) 3.6V VOH (min) 2.4V VOH (min) 2.4V 2.0 2.0 VIH 2.0V 1.0 1.0 VIL O.BV VIH 2.0 V VIL O.BV VOL (max) O.SV VOL (max) 0.4V GND Output Breakdown VCC+ O.SV 0 GND GND drw03 Figure 3. 3.3V Device Driving a SV System 0 GND Figure 4. SV Device Driving a 3.3V Device 141 dow 04 MIXED SUPPLY DESIGN CONSIDERATIONS IN NOTEBOOK PCS CONFERENCE PAPER CP-13 UNIDIRECTIONAL TRANSLATORS Input clamp to Vee As described above, a number of lOT's 3.3V FCT and 5V 3.3V 5V FCT-T parts can be used as unidirectional 5V-to-3.3V translators. The input pins of the 3.3V FCT family can take an input voltage up to 7V allowing them to be driven by 5V components. Figure 6 contains a list of 3.3V unidirectional transla5V tors currently available today. The 5V FCT -T family's reduced Device output swings, VOH typically 3.3V, allows them to safely drive 3.3V components. The unidirectional translators accept TTL- or CMOS-compatible signals from 5V components at the inputs and provide drw 06 TTL-compatible output levels. These translators avoid any Figure 58. Simplified Output Structure problems associated with a direct interface between a 5V component and a 3.3V component or between a 5V compo3.3V/5V BIDIRECTIONAL TRANSLATOR nent and a 3.3V bus. Since not all bus interface applications are unidirectional, No Input clamp there needs to also be a bidirectional solution to the 3.3V to 5V to Vee 3.3V interface. A solution to this is the IOT74FCT164245T. It 5V interfaces between a 3.3V bus and a 5V bus in a mixed 3.3VI 5V supply environment handling all of the 3.3V/5V bidirectional interfacing issues (Figure 7). It is configured as a 16-bit '245 transceiver, giving the designer tri-state capabilities and 5V the ability to select either bidirectional or unidirectional modes. Device The translator has separate 3.3V and 5V supply and 1/0 pins both on separate sides of the device (Figure 8). All 1/0 pins and input pins can withstand a maximum input drw05 voltage of 7V. This allows both ports to be driven by 3.3V or Figure 5A. Simplified Input Structure 5V devices. 3.3V 3.3V BUS r 5V Components 3r I ~ z w IDT74FCT3244 IDT74FCT163244 IDT74FCT163373 IDT74FCT163374 Z o a.. ~ 8 > (') <'i drw 07 Figure 6. Mixed 5V/3.3V System 142 MIXED SUPPLY DESIGN CONSIDERATIONS IN NOTEBOOK PCs 3.3V BUS 3.3V Component ~ ..... 3.3V 5V BUS 5V t t 164245T TRANSLATOR CONFERENCE PAPER CP-13 f- f+ CONCLUSION 5V Component dtw08 Figure 7. 3.3V to 5V Bidirectional Translator The translator part also features power-down disable. This allows the device to withstand the maximum rated voltages on the inputs and outputs while it is unpowered or partially powered. This is a desirable feature for bus-interface logic in a portable system because of power management schemes which power down certain subsystems when inactivity is detected. The bus interface device will then have to buffer the active system bus with a powered off device without damaging itself. To avoid damage to a part which does not support powerdown disable, the current from the 5V outputs to the 3.3V outputs or I/O must be limited. The designer may place a resistor between each 5V output and corresponding 3.3V I/O. When the 3.3V I/O is in a high-impedance state, a current will flow from the 5V part to the 3.3V part through the resistor and clamp diode causing a voltage drop between the devices and preventing damage. However, placing a resistor between the 5V and 3.3V part will cause some speed degradation due to the RC time constant which develops between the series resistor and the input capacitance of the 3.3V part and other parts that may be similarly attached. 3.3V 5V 1 b:o b: oa. a. > <'l > LO <'i • • x16 j 1 FCT164245T drw 09 Figure 8. IDTFCT164245T Translator Block Diagram 143 The benefits of a 3.3V supply voltage have forced designers to an interim solution while not all of the components are available at 3.3V. "Hybrid" or mixed supply systems seem to be the interim solution until a full 3.3V system is possible. When the full 3.3V system is possible, the concepts learned must be applied to interfacing with external 5V peripherals. Designing in the mixed supply environment provides new challenges for the designer both in the "hybrid" system and in the external peripheral interface. The compatibility issues of the 3.3V/5V interface and the solutions available today must be understood in order to see notebook PCs with 8 -12 hour battery life and weight under 4 pounds. NOTE: This paper was presented at the Silicon Valley Personal Computer (SVPC) Conference on 8/4/92. REFERENCES 1. lOT; "3.3 Volt Logic"; Application Note #103 2. lOT; "High Performance Logic Databook"; 1992 3. Prince, Betty; "ICs Going on a 3-V Diet"; IEEE Spectrum; 3/92 4. Ristow, Ken; "Mixing It up with 3.3 Volts"; Electro'92 Proceedings Integrated Device Technology, Inc. U.S. HEADQUARTERS Integrated Device Technology, Inc. 2975 Stender Way P.O. Box 58015 Santa Clara , CA 95052-8015 Tel: (408) 727-6116 FAX: 408-492-8674 UNITED KINGDOM Integrated Device Technology, Ltd. 21 The Crescent Leatherhead Surrey, UK KT2280Y Tel: (44) 0372-363339 FAX: (44) 0372-378851 HONG KONG Integrated Device Technology, Asia, Ltd. Room 1505, 15F, The Centre Mark 287-299 Queen's Road Central Hong Kong Tel: 852-542-0067 FAX: 852-544-5603 / JAPAN Integrated Device Technology, KK U.S. Building 201 1-6-15 Hirakarasho, Chiyoda-Ku Tokyo 102, Japan Tel: 81-3-221-9821 FAX: 81-3-221-7372 © 1993 Integrated Device Technology, Inc. Printed in U.S.A. MAN-LOGIC-00023
Source Exif Data:
File Type : PDF File Type Extension : pdf MIME Type : application/pdf PDF Version : 1.3 Linearized : No XMP Toolkit : Adobe XMP Core 4.2.1-c041 52.342996, 2008/05/07-21:37:19 Create Date : 2017:08:11 17:31:20-08:00 Modify Date : 2017:08:11 18:32-07:00 Metadata Date : 2017:08:11 18:32-07:00 Producer : Adobe Acrobat 9.0 Paper Capture Plug-in Format : application/pdf Document ID : uuid:3889912d-8bd8-da42-89b0-df26daa8eb99 Instance ID : uuid:e3b5583b-025a-1240-bea1-29721bccfb25 Page Layout : SinglePage Page Mode : UseNone Page Count : 148EXIF Metadata provided by EXIF.tools