Deep Submicron CMOS and the New Era of Creativity in Analog Design John A. McNeill Worcester Polytechnic Institute (WPI), Worcester, MA mcneill@ece.wpi.edu McNEILL: CREATIVITY IN DSM CMOS MAY 3, 2006
Overview Analog / Mixed Signal IC Design Role of Creativity DSM CMOS Effects on Analog Design Short L, Thin t ox Matching Issues Self-Calibrating ADC Overview Design Details Results Conclusion McNEILL: CREATIVITY IN DSM CMOS MAY 3, 2006 2
Overview Analog / Mixed Signal IC Design Role of Creativity DSM CMOS Effects on Analog Design Short L, Thin t ox Matching Issues Self-Calibrating ADC Overview Design Details Results Conclusion 3
Career Classification CREATIVE USEFUL ARTIST POET ADVERTISING PROFESSOR ENGINEER DOCTOR TEACHER NURSE LAWYER STOCKBROKER GOOD PAY 4
Why be creative? Need Easy problems solved already Tough problems need creative solution Dealing with environment of change Coping, thriving Human nature Fun! 5
Creativity Resources 6
Creativity Framework Explorer Artist Judge Warrior 7
Creativity Framework Explorer Artist Judge Seek out new information Survey the landscape Get off the beaten path Poke around in unrelated areas Gather lots of ideas Shift your mindset Don't overlook the obvious Look for unusual patterns Warrior 8
Creativity Framework Explorer Artist Judge Warrior Create something original Multiply options Use your imagination Ask what-if questions Play with ideas Look for hidden analogies Break the rules Look at things backward Change contexts Play the fool 9
Creativity Framework Explorer Artist Judge Evaluate options Ask what's wrong Weigh the risk Embrace failure Question assumptions Look for hidden bias Balance reason and hunches Make a decision! Warrior 10
Creativity Framework Explorer Artist Judge Warrior Put decision into practice Commit to a realistic plan Get help Find your real motivation See difficulty as challenge Avoid excuses Persist through criticism Sell benefits not features Make it happen Learn from every outcome 11
Example: Time (Stages of project) Explorer Background Research Artist Brainstorm Options Judge Choose Solution Warrior Implement Design 12
Why a Creativity Model? Education Standardized-test-numbed students Paralysis in face of open-ended problem Designer Awareness of strengths, weaknesses Recognize preferences Not Right or Wrong! One way of looking at process Orchard analogy 13
Example: Modes of Thinking Explorer Artist Divergent Soft Qualitative Judge Warrior Convergent Hard Quantitative 14
Example: Preferred Problem Solution Explorer Artist Add Complexity Judge Warrior Eliminate Complexity 15
Overview Analog / Mixed Signal IC Design Role of Creativity DSM CMOS Effects on Analog Design Short L, Thin t ox Matching Issues Self-Calibrating ADC Overview Design Details Results Conclusion McNEILL: CREATIVITY IN DSM CMOS APRIL 28, 2006 16
Good Old Days W/L I D Large strong inversion region Square law, easy hand analysis Op 't Eynde and Sansen, "Design and Optimization of CMOS Wideband Amplifiers," CICC 1989 17
W [µm] 10 4 10 3 TSMC L=0.25µm process 10 2 10 1 10 0 10-6 10-5 10-4 10-3 10-2 Moderate inversion Graphical / numerical analysis I D [µa] 18
DSM CMOS Thin t ox : Gate Leakage µa Gate Currents! Tunneling current through thin t ox R. Van Langevelde et. al., "Gate current: Modeling, L extraction and impact on RF performance, IEDM 2001 19
DSM CMOS: MOSFET Current Gain Bipolar-like current gain for longer L A.-J. Annema et. al., Analog Circuits in Ultra-Deep-Submicron CMOS, IEEE J. Solid-State Circuits, Jan. 2005, pp. 132-143 20
DSM CMOS: Gate Leakage Long L devices unsuitable R. Van Langevelde et. al., "Gate current: Modeling, L extraction and impact on RF performance, IEDM 2001 21
Overview Analog / Mixed Signal IC Design Role of Creativity DSM CMOS Effects on Analog Design Short L, Thin t ox Matching Issues Self-Calibrating ADC Overview Design Details Results Conclusion McNEILL: CREATIVITY IN DSM CMOS APRIL 28, 2006 22
Matching Classical: Matching improves with Spend area to match Power penalty to drive C OX W L WL Pelgrom et.al., "Matching properties of MOS transistors," IEEE J. Solid-State Circuits, Oct. 1989, pp. 1433-1440 23
Technology Dependence As V DD scales down with L min Some improvement in matching A Vth K. Bult, "Analog Design in Deep Sub-Micron CMOS," ESSCIRC2000, Sept. 2000. 24
Technology Dependence Dynamic Range limited by matching K. Bult, "Analog Design in Deep Sub-Micron CMOS," ESSCIRC2000, Sept. 2000. 25
Speed / Accuracy / Power Tradeoff Limited by matching, not noise Some improvement with technology Kinget, " Device mismatch and tradeoffs in the design of analog circuits," JSSC, June, 2005 26
Matching / Gate Leakage Issues Spend area: Gate leakage mismatch increases with Limit to attainable matching WL A.-J. Annema et. Al., Analog Circuits in Ultra-Deep-Submicron CMOS, IEEE J. Solid-State Circuits, Jan. 2005, pp. 132-143 27
Matching / Gate Leakage Issues Break limit: Spend area (same L): But extra power penalty A.-J. Annema et. Al., Analog Circuits in Ultra-Deep-Submicron CMOS, IEEE J. Solid-State Circuits, Jan. 2005, pp. 132-143 28
Or: Abandon Matching! Options: Fix with analog complexity: Autozero, Enz and Temes, "Circuit Techniques for Reducing the Effects of Op-Amp Imperfections: Autozeroing, Correlated Double Sampling, and Chopper Stabilization," Proceedings of the IEEE, November 1996, pp. 1584-1614 or Fix with digital complexity 29
Overview Analog / Mixed Signal IC Design Role of Creativity DSM CMOS Effects on Analog Design Short L, Thin t ox Matching Issues Self-Calibrating ADC Overview Design Details Results Conclusion McNEILL: CREATIVITY IN DSM CMOS APRIL 28, 2006 30
Self-Calibrating ADC Goals General: Take advantage of CMOS scaling Digital Relax requirements on analog precision All calibration / complexity in digital domain Background Calibration continuous in background Deterministic Short time constant for adaptation No requirements on input signal behavior Specific Implementation: 16b 1MS/s Cyclic ADC in 0.25µm CMOS J. McNeill, et. al., "'Split-ADC' Architecture for Deterministic Digital Background Calibration of a 16b 1MS/s ADC," ISSCC2005 31
Cyclic ADC RESIDUE AMPLIFIER v IN S/H G +! DAC COMP +/-V REF v RES d k DIGITAL x TIMING 1) Sample input, compare to threshold digital decision d 2) Amplify input by factor G 3) Subtract d. V REF residue voltage v RES 4) Repeat cycle with v RES as input Result: sequence of decisions d k 32
v RES(I) Residue amplifier: G + DAC! Cyclic ADC v RES(O) SLOPE = G Residue plot: v RES(O) COMP +/-V REF d k v RES(I) Input-Output Relationship: d = -1 d = +1 v " d! V RES ( O) = G! vres ( I ) REF Multiply input by cyclic gain G, subtract d. V REF 33
Example: 3-Cycle ADC Follow residues; start Cycle 1 residue: Cycle 2: Cycle 3: Rearrange: v RES (3) v RES (2) v RES (1) = Gv [! d V ] IN v IN IN! d V " vres (1) 644 7448 = G Gv 1 REF REF! d " vres (2) 644447 44448 = G G Gv " REF [ [ IN! d1v REF ]! d2vref ]! d3vref 1 2 V [ ] 2 1 0 G d1 + G d2 G d V REF 3 v RES ( 3) = G v IN! + 3 34
1: Cyclic ADC as Negative Feedback Loop [ ] 2 1 0 G d1 + G d2 G d V REF 3 v RES ( 3) = G v IN! + 3 Cyclic amplifier trying to "blow up" v IN DAC trying to drive residue to zero Residue voltages bounded if G isn't "too big" Safety margin: Choose G < 2 Bonus: Redundancy 35
Redundancy Key: Multiple valid decision paths to output code -1 or +1 OK d = -1 d = +1 d = -1 d = +1 G = 2 G < 2 36
2: Digital Correction Divide both sides by G 3 V REF and rearrange v IN 1 1 1 1 RES (3) = d1 + d2 + d 2 3 3! 3 REF G G G G VREF V v Output code x (radix G) Quantization error Digital reconstruction from comparator decisions d k : Use estimated gain G (EST) to calculate output code x : x & = $ 1 #! d & + $ 2 #! d ( EST ) 1 2 G( ) G( ) G! % EST " % EST " % ( EST ) " 3 # d Only G needed to digitally correct ADC linearity Calibration: G (EST) = G to within converter accuracy 1 & + $ 1 3 37
Output Code v IN = 1 V REF G d 1 + 1 G 2 d 2 + 1 G 3 d 3 + L Analog: G = 2 to within converter accuracy Calibration: trim, match x Digital: Use estimated gain G (EST) to calculate output code x : & = $ 1 #! d & + $ 2 #! d ( EST ) 1 2 G( ) G( ) G! % EST " % EST " % ( EST ) " 1 & + $ 3 # d Calibration: How to determine G (EST) = G to within converter accuracy? 1 3 38
Previous Calibration Techniques Deterministic? (All) Digital? Background? [1] [2] [3] [4] [5] [6] [7] [8] [9] No previous technique has all desired features 1. Galton, "Digital cancellation of D/A converter noise in pipelined ADCs," TCAS-II, March 2000 2. Murmann..., "A 12b 75MS/s Pipelined ADC using open-loop residue amplification," ISSCC2003 3. Liu.., "A 15b 20MS/s CMOS Pipelined ADC with Digital Background Calibration," ISSCC2004 4. Nair..., "A 96dB SFDR 50MS/s Digitally Enhanced CMOS Pipelined A/D Converter," ISSCC2004 5. Ryu..., "A 14b-Linear Capacitor Self-Trimming Pipelined ADC," ISSCC2004 6. Erdogan..., "A 12-b Digital-Background-Calibrated Algorithmic ADC with -90-dB THD," ISSC1999 7. Chiu..., "Least mean square adaptive digital background calibration of pipelined ADCs," TCAS-I, Jan. 2004 8. Lee, "A 12-b 600 ks/s digitally self-calibrated pipelined algorithmic ADC," JSSC, Apr. 1994 9. Karanicolas, "A 15-b 1-MS/s digitally self-calibrated pipeline ADC," JSSC, Dec. 1993 39
Previous Digital Background Calibration CONVERSIONS REQUIRED FOR CALIBRATION 2 10 9 5 10 8 10 7 10 6 10 5 2 4 3 1 2N [1] Galton2000 [2] Murmann2003 [3] Liu2004 [4] Nair2004 [5] Ryu2004 10 4 12 14 16 N BITS RESOLUTION 40
Statistical Techniques Problem How long to calibrate with 2 2N samples? 12 bits, 75 MS/s [2] 2 2 " 12! 75Msps 200ms 16 bits, 1 MS/s 2 2 " 16! 1Msps 1 hour Deterministic approach needed The problem: How to do a deterministic calibration procedure in background without a known input? 41
Split ADC Architecture ADC OUTPUT CODE ADC "A" x A + + x = x A + x B 2 v IN ADC "B" x B + - "x = x B # x A ERROR ESTIMATION DIFFERENCE Average of A, B results is ADC output code Calibration signal developed from difference 42
Intuitive View of Split ADC x v IN ADC "A" x A ADC "B" xb + + + - x t RESIDUE MODES ERROR ESTIMATION "x = x B # x A Different paths to (ideally) same answer Estimate errors from "disagreements" Only way for A, B to always agree is for both to be correctly calibrated 43
Robert Frost: New Hampshire... a figure of the way the strong of mind and strong of arm should fit together, One thick where one is thin and vice versa. V T N H 44
Robert Frost: New Hampshire... a figure of the way the strong of mind and strong of arm should fit together, One thick where one is thin and vice versa. V T N H Key idea: two partners trying to do the same thing in different ways 45
Same Area, Noise, Speed, Power ANALOG DIGITAL ANALOG DIGITAL v IN C g m x SPLIT v IN C 2 C 2 g m 2 g m 2 A B x A x B x x A + x B 2 Speed f T = b g m C b g m 2 C 2 = " b g m $ # C % ' & Power P = p " g m p " g m 2 + p " g m 2 = p " g m Noise " x = n kt " 1 C 2 n kt % $ ' # C 2 & 2 " + 1 2 n kt % $ ' # C 2 & 2 = n kt C Negligible impact on analog complexity 46
Overview Analog / Mixed Signal IC Design Role of Creativity DSM CMOS Effects on Analog Design Short L, Thin t ox Matching Issues Self-Calibrating ADC Overview Design Details Results Conclusion McNEILL: CREATIVITY IN DSM CMOS APRIL 28, 2006 47
Evaluation Block Diagram REF EVALUATION BOARD FPGA V IN INPUT SIGNAL COND TEST CHIP CYCLIC ANALOG CYCLIC DIGITAL DATA FORMATTING DSP INTERFACE OTHER FPGA FUNCTIONS TO RAM / DSP "PRODUCT" CYCLIC TIMING CNVST EXT TIMING Test chip mostly analog Digital on FPGA (code "synthesis-ready" for product) 48
ADC Block Diagram S/H G A + " " $ # 1 ˆ G A % ' & k Σ x A v IN S/H COMPS COMPS G B DAC PATH A PATH B DAC + " d ka d kb A L.U.T. ˆ G A ERROR COEFF L.U.T. ˆ G B B L.U.T. " $ # 1 ˆ G B % ' & k " k 1 % $ ' # G& k Σ Σ Σ Σ Σ " + SDK A µ µ SDK B x B ˆ " A "x ˆ " B ERROR EST. + + x CYCLIC RESIDUE AMPLIFIERS OFF-CHIP DIGITAL PROCESSOR (FPGA) 49
ADC Digital Correction S/H G A + " " $ # 1 ˆ G A % ' & k Σ x A v IN S/H COMPS COMPS G B DAC PATH A PATH B DAC + " d ka d kb A L.U.T. ˆ G A ERROR COEFF L.U.T. ˆ G B B L.U.T. " $ # 1 ˆ G B % ' & k " k 1 % $ ' # G& k Σ Σ Σ Σ Σ " + SDK A µ µ SDK B x B ˆ " A "x ˆ " B ERROR EST. + + x CYCLIC RESIDUE AMPLIFIERS OFF-CHIP DIGITAL PROCESSOR (FPGA) 50
ADC Digital Correction COMPARATOR DECISIONS [ -1, 0, +1 ] d ka " 1 $ # ˆ G A % ' & DECISION WEIGHT L.U.T. k Σ x A ACCUMULATE OUTPUT CODE Decision weight L.U.T. Periodically recalculated in background Separate L.U.T.s for A, B output codes 51
Error Estimation S/H G A + " " $ # 1 ˆ G A % ' & k Σ x A v IN S/H COMPS COMPS G B DAC PATH A PATH B DAC + " d ka d kb A L.U.T. ˆ G A ERROR COEFF L.U.T. ˆ G B B L.U.T. " $ # 1 ˆ G B % ' & k " k 1 % $ ' # G& k Σ Σ Σ Σ Σ " + SDK A µ µ SDK B x B ˆ " A "x ˆ " B ERROR EST. + + x CYCLIC RESIDUE AMPLIFIERS OFF-CHIP DIGITAL PROCESSOR (FPGA) 52
x x A B A, B Outputs = x + = x + Error Estimation [ SDK A]! A [ SDK B ]! B Difference [ SDK ] [ ] B! B " SDK A A # x =! IDEAL ERROR Ideal x cancelled from estimation signal path No need for long decorrelation times Deterministic: solve for ε A, ε B from a few Δ x observations SDK error coefficients can be determined from comparator decisions 53
Error Estimation Difference: [ SDK ] [ ] B! B " SDK A A # x =! SDK A, SDK B Error coefficients ε A, ε B Fractional errors in G A, G B estimates d ka Σ SDK A ERROR COEFF L.U.T. " k 1 % $ ' # G& k Need different d ka, d kb for visibility to errors 54
Multiple Residue Mode Amplifier S/H G A + " " $ # 1 ˆ G A % ' & k Σ x A v IN S/H COMPS COMPS G B DAC PATH A PATH B DAC + " d ka d kb A L.U.T. ˆ G A ERROR COEFF L.U.T. ˆ G B B L.U.T. " $ # 1 ˆ G B % ' & k " k 1 % $ ' # G& k Σ Σ Σ Σ Σ " + SDK A µ µ SDK B x B ˆ " A "x ˆ " B ERROR EST. + + x CYCLIC RESIDUE AMPLIFIERS OFF-CHIP DIGITAL PROCESSOR (FPGA) 55
Multiple Residue Mode Amplifier v IN S/H +V TH 0 G SEL + " +V REF 0 "V REF DAC d CYCLE DECISION -1 / 0 / +1 "V TH PATH PATH: 00 CYCLIC 01 HIGH 10 LOW 11 WIDE DECISION d: -1 +1-1 0 +1-1 0 +1-1 0 +1 2b PATH sets residue mode entirely in digital domain 56
S/H, 1.5b DAC, G=1.92 Cyclic Amplifier v IN S/H +V TH 0 G SEL + " +V REF 0 "V REF DAC d CYCLE DECISION -1 / 0 / +1 "V TH PATH PATH: 00 CYCLIC 01 HIGH 10 LOW 11 WIDE DECISION d: -1 +1-1 0 +1-1 0 +1-1 0 +1 2b PATH sets residue mode entirely in digital domain 57
INL Shapes Vary by Residue Mode [ SDK ] [ ] B! B " SDK A A # x =! INL shape same as SDK A, SDK B error coefficients 58
Cyclic Amplifier: 3-Capacitor? Advantages Easier to do signal-independent reference current Decouple reference, cyclic gain paths (CM!) Disadvantages Extra capacitor area Extra noise gain (killer!) Output only valid on one phase (1/2 cycle) Less time for comparator DAC cap Feedback cap Signal cap P. Ferguson, Practical Aspects of Delta-Sigma Data Converter Design, MEAD Microelectronics 59
Cyclic Amplifier: 2-Capacitor Advantages Less cap area Lower noise gain Output valid both phases Easier on comparator SDBVOUT DB C D DT VOUT STPA VCM SFBVIN CF SCF VIN SFBVIN SFBVOUT SDTA SDBP SDBZ SDBM SDTVCM A VREFP VCM VREFM VCM Disadvantages Signal-dependent reference current Reference, cyclic gain paths constrained (CM!) 60
Cyclic Amplifier: 2-Capacitor 2-Cap chosen: Lower total capacitance for a given noise performance Different feedback β in DAC, sample modes Changes effect of amplifier noise "DAC mode" β ~ 1/2 "Sample mode" β ~ 1 C D CF CF C D VCM VCM 61
S/H, 1.5b DAC, G=1.92 Cyclic Amplifier V CM V INP V OP 13.5pF 15pF V REFP V CM V CM V OP V REFM V OM 13.5pF 15pF V OM kt/c noise limited large C V CM V INM 62
Op-Amp V CM V INP V OP 13.5pF 15pF V REFP V CM V CM V OP V REFM V OM 13.5pF 15pF V OM kt/c noise limited large C V CM V INM 63
Op-Amp Requirements Param C L I DD f T A OL V OUT SR Spec 20-30pF 33mA 150 MHz 100 db +/- 1.8V 500 V/us Comments Required by kt/c noise limit 80% of total IC power goal (100mW) 16 bit settling, 30 ns, 1st half cycle Maintain over full signal range Trade SNR, linearity Trade with ft, settling time 64
Op-Amp VB5 VB1 VB5 V IP V OP V IM V OM VB2 VB3 VB4 SNR ±2Vpp swing Output not cascoded 16b linear ~100dB A OL 2-stage First stage Gain boosted cascode Bult & Geelen, "A fast-settling CMOS op amp for SC circuits with 90-dB DC gain," JSSC, Dec 1990 Pan et. al., "A 3.3-V 12-b 50-MS/s A/D converter in 0.6-µm CMOS with over 80-dB SFDR," JSSC, Dec. 2000 65
Op-Amp: Design for 90dB SNR Noise contributors: VOUT VCM VIN Sample cap kt/c SFBVIN SFBVIN Op-amp g m SDBVOUT STPA CF SFBVOUT C D DB DT SCF SDTA SDBP SDBZ SDBM SDTVCM A VREFP VCM VREFM VCM Plot SNR, total current I BIAS as function of C F, I BIAS g m 66
Op-Amp: I BIAS C F Optimization DIFF PAIR I BIAS [A] SNR [db] TOTAL OP-AMP CURRENT [ma] SAMPLE CAPACITANCE C F [F] 67
Op-Amp: I BIAS C F Optimization C F limited g m limited SNR 90dB Bias current, sample cap tradeoff 68
Overview Analog / Mixed Signal IC Design Role of Creativity DSM CMOS Effects on Analog Design Short L, Thin t ox Matching Issues Self-Calibrating ADC Overview Design Details Results Conclusion McNEILL: CREATIVITY IN DSM CMOS APRIL 28, 2006 69
Die Photo ADC "A" ADC "B" SWITCHED CAP NETWORK OP-AMP COMPARATORS 70
Measured INL 71
Temperature Performance 72
Calibration Convergence 73
Comparison with Previous Work 10 9 10 8 10 7 10 6 CONVERSIONS REQUIRED FOR CALIBRATION 2 5 4 3 1 2 2N [1] Galton2000 [2] Murmann2003 [3] Liu2004 [4] Nair2004 [5] Ryu2004 10 5 10 4 THIS WORK 12 14 16 Long decorrelation times not necessary N BITS RESOLUTION 74
Performance Summary Technology Supply Voltage Resolution Conversion Rate SNR INL DNL Power Consumption * Die Area * 0.25µm 1P4M CMOS 2.5 V 16 b 1 MS/s 89 db +2.1 / -4.8 LSB +0.66 / 0.47 LSB 105mW 1.16mm x 1.38mm * Excludes digital on FPGA 75
"Split ADC" architecture Average: Output code Difference: Drive to zero to correct errors Deterministic: Rapid self-calibration Suitable for high resolution ADCs 16b 1MSps Cyclic ADC Self-calibration in ~ 10,000 conversions Complexity moved into digital domain 76
Overview Analog / Mixed Signal IC Design Role of Creativity DSM CMOS Effects on Analog Design Short L, Thin t ox Matching Issues Self-Calibrating ADC Overview Design Details Results Conclusion McNEILL: CREATIVITY IN DSM CMOS APRIL 28, 2006 77
DSM CMOS Conclusions Performance challenges Change in role of analog techniques Opportunities Digital complexity enabled Need for designer creativity Choose best from both worlds 78
Acknowledgments Analog Devices Precision Nyquist Converters group Bob Adams Bob Brewer Larry DeVito Paul Ferguson Colin Lyden Katsu Nakamura Richard Schreier Larry Singer Stanford University Boris Murmann 79
80
Creativity References Self-Calibrating ADCs R. Von Oech, "A Whack on the Side of the Head" New York: Warner, 1998. ISBN 0446674559 R. Von Oech, "A Kick in the Seat of the Pants" New York: HarperCollins, 1986. ISBN 0060960248 CMOS Design Op 't Eynde and Sansen, "Design and Optimization of CMOS Wideband Amplifiers," Proc. CICC, 1989. R. van Langevelde, A. J. Scholten, R. Duffy, F. N. Cubaynes, M. J. Knitel, and D. B. M. Klaassen, "Gate current: Modeling, L extraction and impact on RF performance, Proc. IEDM, 2001. A.-J. Annema, B. Nauta, R. van Langevelde, and H. Tuinhout, Analog Circuits in Ultra-Deep-Submicron CMOS, IEEE J. Solid-State Circuits, Jan. 2005. C. Enz and G. Temes, "Circuit Techniques for Reducing the Effects of Op-Amp Imperfections: Autozeroing, Correlated Double Sampling, and Chopper Stabilization," Proceedings of the IEEE, Nov. 1996. J. McNeill, M. Coln, and B. Larivee, "'Split-ADC' Architecture for Deterministic Digital Background Calibration of a 16b 1MS/s ADC," ISSCC2005 B. Murmann and B. Boser, "A 12-bit 75-MS/s Pipelined ADC Using Open-Loop Residue Amplification," IEEE J.Solid-State Circuits, Dec. 2003. Matching M. Pelgrom, A. Duinmaijer, and A. Welbers, "Matching properties of MOS transistors," IEEE J. Solid-State Circuits, Oct. 1989. P. R. Kinget, " Device mismatch and tradeoffs in the design of analog circuits," JSSC, June, 2005. K. Bult, "Analog Design in Deep Sub-Micron CMOS," ESSCIRC2000, Sept. 2000. 81