A Precise Block-Based Statistical Timing Analysis with MAX Approximation Using Multivariate Adaptive Regression Splines

Abstract

The impact of process variations on timing has become significant in advanced technology nodes. In this paper, a multivariate adaptive regression splines (MARS) delay model is proposed that considers both global and local process variations to characterize this impact more accurately. In order to obtain MAX operation results, the first three moments of MARS gate delay distribution are calculated, converting the timing distribution to a skew-normal representation. Eventually, based on an approximation MAX operation, the block-based SSTA propagates the arrival time through the timing diagram. Tested with 10 ISCAS85 benchmark circuits, the average mean squared error and standard deviation error of the path delay calculation are 0.52% and 0.88%.