Gate Level Statistical Simulation Based on Parameterized Models for Process and Signal Variations

Bao Liu
UC San Diego


Abstract

We propose gate level statistical simulation to bridge the gap between the most accurate Monte Carlo SPICE simulation and the most efficient circuit level statistical static timing analysis (SSTA) for a new level of efficiency-accuracy tradeoff. Our method is based on (1) a multi-point waveform characterization by signal arrival times at multiple voltage thresholds, (2) a parameterized current source gate model for process variations, (3) a parameterized gate performance model for process and signal waveform variations, and (4) Monte Carlo simulation. Our experimental results show that our proposed gate level statistical simulation achieves orders of magnitude of efficiency improvement based on the constructed gate models, while achieving within 3.91% (9.19%) accuracy in average for the means (standard deviations) of signal arrival times at multiple voltage thresholds.