Efficient Statistical Analysis in MMSIM DCMatch & ACMatch

Efficient Statistical Analysis in MMSIM DCMatch & ACMatch YuHua Huang RDC-CAD Jun 26, 2015

Outline Variability in Integrated Circuit Mismatch Analysis DCMatch Sensitivity Analysis ACMatch Sensitivity Analysis 2

Variability in Integrated Circuit Yield Loss Defect Density Yield Lithography and New Material Based Yield Design Based Yield Technology Nodes (nm) 4

Need for Statistical Analysis Process variations from wafer to wafer, and run to run, result in a statistical distribution for several key model parameters which can affect your yield. Identical devices may exhibit a mismatch in several key parameters due to diffusion gradients across a die. This leads to circuit problems such as input offset in op-amps and comparators which need to be evaluated. Statistical Analysis Device Variation Circuit Performance 5

Device Mismatch Two devices with the same physical layout never have quite the same electrical properties. Variations between devices are called mismatches. Mismatches are produced by random processes and produce the following effects: Offset errors in comparators and amplifiers Voltage differences in band-gap reference DNL and INL in DACs and ADCs.. In simulation, devices are often modeled with identical numerical values and expressions for the electrical behavior: Simulations often fail to include the effects of mismatch. Precise values of mismatch can not be predicted. 7

Effects of Mismatch on OP Amplifier XOP V DD M7 CMFB M8 Absolute node Variation V O M5 M6 V Z2 M3 M4 V Z1 V1 M1 M2 V2 M9 M10 IC Relative node Mismatch GND 8

Mismatch Analysis DCMatch To calculate effects of variations on a circuit s DC operation point. The DCMatch analysis performs DC device mismatching analysis for a given output. It computes the deviation in the DC operating point of the circuit caused by mismatch in the devices. ACMatch To calculate effects of variations on a circuit s AC response. The ACMatch analysis linearizes the circuit about the DC operating point and computes the variations of AC responses due to statistical parameters defined in statistical blocks. 9

The Advantage of DCMatch Use foundry s Monte Carlo model. Results match with Monte Carlo Analysis. Report variation contributions of parameters and devices. At least one order faster than Monte Carlo Analysis. You need to run 250 s of simulations to develop good statistics which means this takes 250 s of times longer than DCMatch (which is 1 DC simulation). 11

DCMatch Analysis To calculate effects of variations on a circuit s DC operating point. DCMatch is applicable to linear system or system that can be linear approximated at the DC operation point. 12

Circuit Candidates for DCMatch Operational Amplifiers Bias Circuits DACs Bandgap Charge Pumps Circuit Structure VIP VIN + - OUT 13

DCMatch Model in Spectre Using Foundry Monte Carlo Model Global variation - Statistical equivalent of corner files Local variation - Random variation between adjacent devices with identical layout PMOS fast typical slow fast NMOS 14

DCMatch Analysis Flow 15

DCMatch Analysis in ADE 1. 2. Monte Carlo Model 3. 16

DCMatch Analysis Command Spectre netlist dcmm (VBG) dcmatch method=statistics mismatch=yes Spice netlist simulator lang=spectre dcmm (VBG) dcmatch method=statistics mismatch=yes simulator lang=spice Workshop $MMSIM_INSTALL/tools.lnx86/spectre/examples/DCMatch 17

DCMatch Simulation Result Contributions of parameters and devices Sigma of total output variations 18

Monte Carlo versus DCMatch Feature Monte Carlo * DCMatch Variation Define Foundry MC Model Foundry MC model Accuracy V(VREF,INP) -1.199mV +/- 6.281mV V(VREF,INP) -1.199mV +/- 6.313mV Run Time 254s 0.731s Contributions Yes Yes *Monte Carlo use Low-Discrepancy Sequence Sampling method, run DC simulation 250 times Development Tools Schematic /ADE-XL : Cadence Virtuoso 616 ISR12 Simulation : Cadence MMSIM 141 ISR9 Process : 0.18u Circuit : OPA 19

The Advantage of ACMatch Use foundry s Monte Carlo model. Results match with Monte Carlo Analysis. Real Imaginary Mag Phase variations of each parameter contributing to the output variation of the AC signal defined. Total variation of the output AC signal. Significantly faster than Monte Carlo Analysis. 21

Circuit Candidates for ACMatch Operational Amplifiers Fully differential circuits Continuous-time integrated filters Crystal oscillators Circuit Structure VIP VIN + - OUT 22

ACMatch Analysis in ADE 1. 2. Monte Carlo Model 3. 23

ACMatch Analysis Command Spectre netlist Acmm (VBG) acmatch start=1k stop=10g dec=1 Spice netlist simulator lang=spectre Acmm (VBG) acmatch start=1k stop=10g dec=1 simulator lang=spice Supported Options Allowed noed numbers:1,2 1 node : AC response on the node is the output. 2 nodes : AC response difference of the two nodes is taken as output start, stop, step, dec :same as in linear AC analysis. where :screen logfile file rawfile Where the results should be sent; Note that if where=rawfile,the results are also printed on screen and logfile. 24

ACMatch Simulation Result Contribution of each mismatch paarameter Header Total Contribution Next Frequency Next Frequency 25

Monte Carlo versus ACMatch Feature Monte Carlo * ACMatch Variation Define Foundry MC Model Foundry MC model Accuracy Mag(V) Std 1K : 119m 10K : 119m 100K : 117.5m 1M : 327m 10M : 11m 100M : 121.8u Mag(V) Std 1K : 123m 10K : 123m 100K : 117m 1M : 314m 10M : 10.4m 100M :112u Run Time 418s 74.4s Contributions Yes Yes *Monte Carlo use Low-Discrepancy Sequence Sampling method, run AC simulation 250 times Development Tools Schematic /ADE-XL : Cadence Virtuoso 616 ISR12 Simulation : Cadence MMSIM 141 ISR9 Process : 0.18u Circuit : OPA 26

DCMatch / ACMatch Analysis Summary Pros: They are useful tools for getting insight into sources of mismatch. Avoid yield loss due to variations. Simply include the Monte Carlo model. Fast and easy to run. Pointer to major contributions of variation. Results match with Monte Carlo Analysis. Significantly faster than Monte Carlo Analysis. Cons: Rely on linear approximation to get accurate results. 27