A fast estimation of SRAM failure rate using probability collectives

Abstract

Importance sampling is a popular approach to estimate rare event failures of SRAM cells. We propose to improve importance sampling by probability collectives. First, we use "Kullback-Leibler (KL) distance" to measure the distance between the optimal sampling distribution and the original sampling distribution of variable process parameters. Further, the probability collectives (PC) technique using immediate sampling is adapted to analytically minimize the KL distance and to obtain a sampling distribution as close to the optimal as possible. The proposed algorithm significantly accelerates the convergence of importance sampling. Experiments demonstrate that proposed algorithm is 5200X faster than the Monte Carlo approach and achieves more than $40X$ speedup over other existing state-of-the-art techniques without compromising estimation accuracy.