Effective sampling method for statistical leakage analysis: Hybrid sampling method

Abstract

This paper proposes a new sampling method for Monte-Carlo (MC)-based statistical leakage analysis. To address the slow convergence rate of traditional MC, variance reduction techniques were introduced. Among them, Quasi Monte-Carlo (QMC) has the fastest convergence rate. However the convergence improvement decreases as the number of parameter dimensions increases. To address this problem, the proposed method uses not only QMC but also another variance reduction technique called Latin hypercube sampling. Depending on the influence of parameters on leakage currents, the sampling method to be applied is determined. Experimental results for the same number of samples, 10k, using six circuits of ISCAS-85 benchmarks showed that the proposed method and QMC has average leakage mean error of 3.39% and 4.86% when compared with the golden MC results.