Automotive Power CAD Infineon Statistical Simulation Methodology for a Smart Power Technology 2007, Oct 19 th AK Bipolar 2007 / Munich Dr. Elmar Gondro Page 1
Outline Motivation Modeling Flow Model Quality Conclusion Page 2
Motivation / Wake-Up Zero Zero Defect Defect TD Design Design 4 Manufacturability Yield Yield Enhancemen tt CAD Automotive Excellence PD Flow Quality Model Quality Nominal & statistical Representation of all future Fab Outputs Page 3
Five Steps Extraction Process of Electrical Parameters Extraction Centering µ Nominal Parameters PCM Spec Limits Deviations σ Correlations Statistical Parameters Mismatch Testchip Matching Consts Page 4
Step 1: Nominal Parameters 1e6 10000 I C meas simu I B meas simu 100 PCM Limits Mismatch I [µa] 1 0.01 0.0001 1e 6 PCM parameters: Vbe Ic β Ib Vbd none 1e 8 0.2 0.4 0.6 0.8 1 1.2 V BE [V] source: /EP/-Docu (Bipolar / Tempsense) Page 5
Step 1: Nominal Parameters PCM Limits Mismatch Re-Simulation Schematics SMART5 PCM: 41 of 75 device types covered 143 of 194 PCM setups can be re-simulated simultaneously source: /PCM/-Docu (Bipolar) Page 6
Step 2: Parameter Centering past (T7) PCM Limits Mismatch propability density µ Page 7
Step 2: Parameter Centering PCM Limits Mismatch propability density Definition!!! µ=target Page 8
Step 2: Parameter Centering PCM Limits Mismatch propability density Nom µ=target Page 9
Step 2: Parameter Centering PCM Limits Mismatch propability density µ=target=nom Page 10
Step 3: Process Deviations PCM Limits Mismatch propability density 6 σ 6 σ Centering and ±6σ cpk=2 Definition!!! µ=target=nom Page 11
Step 3: Process Deviations today past (T7) PCM Limits Mismatch propability density µ=target=nom Page 12
Step 3: Process Deviations today past (T7) future PCM Limits Mismatch propability density µ=target=nom Page 13
Step 3: Process Deviations PCM Limits Mismatch propability density µ=target=nom Page 14
Step 3: Process Deviations PCM Limits Mismatch propability density µ=target=nom Page 15
Step 3: Process Deviations PCM Limits Mismatch propability density Centering and ±3σ cpk=1 Definition!!! 3 σ 3 σ µ=target=nom Page 16
Step 3: Process Deviations Normalized PCM Window with 3 σ Process Variations Parameter Classification Vt MNLE2 YY simu PCM Limits G MNLE2 MM Mismatch Ron MNLE2 MM Isat MNLE2 YY MOSFET Target source: /MQ/-Docu (Low Volt NMOS) Page 17
Step 3: Process Deviations Normalized PCM Window with 3 σ Process Variations Parameter B QNBH 200n Classification MM simu B QNBH 20u YM PCM Limits B QNBH 200u MM VBE QNBH 1m MM Mismatch VBE QNBH 50u MM VCEO QNBH no simulation possible RM VCB QNBH no simulation possible RM VEB QNBH no simulation possible RM Bipolar Target source: /MQ/-Docu (qnbh Bipolar) Page 18
Step 3: Process Deviations PCM Limits I [µa] Gummel Poon and beta Plot 10000 100 1 0.01 0.0001 1e 6 I C meas simu I B meas simu 90 80 meas simu beta @ Ic=200n, 20u, 200u Mismatch 1e 8 0.2 0.4 0.6 Vbe @ Ic=50u, 1m 70 60 0.8 1 1.2 V BE [V] β [ ] 50 40 30 20 Bipolar 10 0 1e 6 0.0001 0.01 1 100 I C [µa] source: /EP/-Docu (qnbh Bipolar) Page 19 10000
Step 4: Correlation Table PCM Limits Mismatch Page 20
Step 4: Correlations Nom=0.838 =0.943 0.95 PCM: c = 0.818 Sim: c = 0.751 PCM Limits Mismatch Target=0.852 Vt MNNE2 [V] 0.9 0.85 Nom=0.863 0.8 =0.761 0.75 0.75 0.8 0.85 =0.738 Target=0.834 Vt MNLE2 [V] 0.9 =0.93 source: /PCM/-Docu Page 21
Step 5: Mismatch Parameters Special Device Pair Measurements on Testchip required PCM Limits Mismatch σ mismatch = const mismatch 2 Area Threshold voltages of MOS transistors Current gains of bipolar Sheet resistances of poly resistors Page 22
Monte Carlo Sections Comment Distribution nom Gaussian distribution of samples µ = Target = (+)/2 σ = (-µ)/3 = (-)/6 unif uniform distribution of samples µ = Target = (+)/2 σ = (- µ)/sqrt(3) = σ[nom]*sqrt(3) Page 23
Monte Carlo Sections nom unif Comment Gaussian distribution of samples µ = Target = (+)/2 σ = (-µ)/3 = (-)/6 uniform distribution of samples µ = Target = (+)/2 σ = (-µ)/sqrt(3) = σ[nom]*sqrt(3) Distribution unif2s3s uniform distribution of samples within 2σ and 3σ µ = Target = (+)/2 σ (-µ)*0.84 = σ[nom]*2.5 speclimits parameters shifted to their Spec Limits ( or ) µ = Target = (+)/2 σ = -µ = σ[nom]*3 Page 24
Monte Carlo Sections w/ Parasitics w/o Parasitics Comment Distribution nom nom_nopar Gaussian distribution of samples µ = Target = (+)/2 σ = (-µ)/3 = (-)/6 unif unif_nopar uniform distribution of samples µ = Target = (+)/2 σ = (-µ)/sqrt(3) = σ[nom]*sqrt(3) unif2s3s unif2s3s_nopar uniform distribution of samples within 2σ and 3σ µ = Target = (+)/2 σ (-µ)*0.84 = σ[nom]*2.5 speclimits speclimits_nopar parameters shifted to their Spec Limits ( or ) µ = Target = (+)/2 σ = -µ = σ[nom]*3 Page 25
Example 1: PCM Par Vt_MNLE2 MC Section nom 4000 3500 Nom=0.838 PCM µ=0.836 σ=0.0121 N =1677 Sim µ=0.84 σ=0.032 N =1000 Spec µ=0.834 σ=0.032 propability density [%] 3000 2500 2000 1500 1000 500 0 0.75 =0.738 0.8 0.85 Target=0.834 0.9 =0.93 V th [V] source: /PCM/-Docu Page 26
Example 2: PCM Par Vt_MNLE2 MC Section unif 4000 3500 Nom=0.838 PCM µ=0.836 σ=0.0121 N =1677 Sim µ=0.838 σ=0.0551 N =1000 Spec µ=0.834 σ=0.055 3000 propability density [%] 2500 2000 1500 1000 500 0 0.75 =0.738 0.8 0.85 Target=0.834 0.9 =0.93 V th [V] Page 27
Example 3: PCM Par Vt_MNLE2 MC Section unif2s3s 4000 3500 Nom=0.838 PCM µ=0.836 σ=0.0121 N =1677 Sim µ=0.838 σ=0.0807 N =1000 Spec µ=0.834 σ=0.081 propability density [%] 3000 2500 2000 1500 1000 500 0 0.75 =0.738 0.8 0.85 Target=0.834 0.9 =0.93 V th [V] Page 28
Example 4: PCM Par Vt_MNLE2 MC Section speclimits 8000 7000 Nom=0.838 PCM µ=0.836 σ=0.0121 N =1677 Sim µ=0.838 σ=0.0962 N =1000 Spec µ=0.834 σ=0.096 propability density [%] 6000 5000 4000 3000 2000 1000 0 0.75 =0.738 0.8 0.85 Target=0.834 0.9 =0.93 V th [V] Page 29
Corner Sections Comment depfast_dmosfast depslow_dmosfast depfast: dmosfast: δv th (MNND)=-0.139V δv th (MNND2)=-0.154V δl cap (MNTE, MNSE2)=-0.25µm depfast_dmosslow depslow_dmosslow depslow: dmosslow: δv th (MNND, MNND2)=0.243V δl cap (MNTE, MNSE2)=0.32µm There are no universally valid worst cases in BCD Technologies Corners have to be defined by Device Team according to the needs of Product Development! Page 30
Corner Sections Comment depfast_dmosfast depslow_dmosfast depfast: dmosfast: δv th (MNND)=-0.139V δv th (MNND2)=-0.154V δl cap (MNTE, MNSE2)=-0.25µm depfast_dmosslow depslow_dmosslow depslow: dmosslow: δv th (MNND, MNND2)=0.243V δl cap (MNTE, MNSE2)=0.32µm FastFast SlowFast FastFast: SlowFast: δv th (NMOS)=-3σ δt ox =-3σ δl int =-3σ δc j =-3σ δv th (NMOS)=3σ δv th (PMOS)=3σ δw int =+3σ δv th (PMOS)=3σ FastSlow SlowSlow SlowSlow: FastSlow: δv th (NMOS)=3σ δt ox =+3σ δl int =+3σ δc j =+3σ δv th (NMOS)=-3σ δv th (PMOS)=-3σ δ w int =-3σ δv th (PMOS)=-3σ Page 31
Model Sections 8 Monte Carlo Sections 4 sections with different distributions 4 no Parasitics sections (nopar) with neglect of: - (substrate) parasitic devices - voltage/current/power warnings - paramtests (geometry checks) Corresponding nopar sections speed up simulation by 25%. No section toggle required for nominal and Gaussian MC simulation 8 Corner Sections 4 corners for NMOS/PMOS speed 4 corners for leakage current of depletion MOS and slew rate of DMOS Performing Monte Carlo analysis issues an error. Page 32
Model Quality The SMART5 PCM comprises 194 measurements (ASM52). 143 of them can be re-simulated. PCM evaluations may serve as a testbench to quantify the model quality (w/o correlation and mismatch). 1. Does the nominal simulation reproduce the PCM Target? nom = Target 2. Does the mean value of the Monte Carlo simulation reproduce the nominal value? µ = nom 3. Does the standard deviation of the Monte Carlo simulation reproduce one third of the distance between the PCM Upper Spec Limit and the PCM Target (cpk=1)? 3 σ = -Target Page 33
Model Quality source: /MQ/-Docu Page 34
Model Quality Parameter Classification Vt MNLE2 G MNLE2 YY MM simu meas Ron MNLE2 MM Isat MNLE2 YY Target source: /MQ/-Docu Page 35
Conclusion PCM: not only Process Monitoring (TD), but also Device Monitoring (CAD) Recommended Usage of Monte Carlo Section nom with 100 runs for small circuits Monte Carlo Section unif2s3s_nopar with 100 runs for large circuits Corner Sections only if you know what you are doing (worst-worst case) Reproduction of Correlations Monte Carlo Section nom Monte Carlo Section unif, unif2s3s, speclimits Corner Sections To be discussed Benefits of Quarterly Monitoring? Device type coverage (SMART5 PCM: 41 of 75 types)? PCM measuring regions? PCM AC and Temp measurements needed? Page 36