Statistically Enhanced Analogue and Mixed-Signal Design and Test

Abstract

The design, manufacture and operational characteristics (e.g., yield, performance, and reliability) of modern electronic integrated systems exhibit extreme levels of complexity that cannot be easily modelled or predicted. Different mathematical methodologies have been explored to address this issue. Monte Carlo simulation is the most widely employed and straightforward approach to evaluate the circuits' performance statistics. However, the high number of trial cases and the long simulations times required to obtain results for complex circuits with a ppm resolution, lead to very long analysis times. The present work addresses the evaluation of alternative statistical inference methodologies which allow obtaining similar results departing from a smaller dimension data set of Monte Carlo simulations from which the overall population is estimated. These methodologies include the use of Bayesian inference, Expectation-inimization, and Kolmogorov-Smirnov tests. Results are presented which show the validity of these approaches.