Statistical Timing Macromodels of Digital IP Libraries

In this paper we present an innovative statistical methodology that allows one to get an accurate, hierarchical macro-model of the probability density function (pdf) of digital IP block timings. Every timing measurement is expanded in series of orthogonal polynomials, and the coefficients of the expansion are directly obtained from the (known) moments of the pdf of SPICE model parameters. Good agreement is demonstrated by comparing the macro-model with traditional Monte-Carlo simulations on a 0.25 μm CMOS standard cell library.