Robust Estimation of Timing Yield with Partial Statistical Information on Process Variations

Abstract

This paper illustrates the application of distributional robustness theory to compute the worst-case timing yield of a circuit. Our assumption is that the probability distribution of process variables are unknown and only the intervals of the process variables and their class of distributions are available. We consider two practical classes to group potential distributions. We then derive conditions that allow applying the results of the distributional robustness theory to efficiently and accurately estimate the worst-case timing yield for each class. Compared to other recent works, our approach can model correlations among process variables and does not require knowledge of exact function form of the joint distribution function of process variables. While our emphasis is on robust timing yield estimation, our approach is also applicable to other types of parametric yield.