Modeling the impact of back-end process variation on circuit performance

Abstract

We present a stochastic approach to account for on-chip interconnect process variation. A Monte Carlo approach is taken using actual process distributions to generate realistic 3-D performance corners. Accurate analytical models are used to provide a >3 order of magnitude speedup over simulation techniques. Resulting delay and noise performance spreads are 33 to 63% tighter than those found using a conventional technique. We apply this method to a clock distribution network to more precisely determine clock skew.