Statistical timing analysis via modern optimization lens

Abstract

We formulate statistical static timing analysis (SSTA) as a mixedinteger program and as a geometric program, utilizing histogram approximations of the random variables involved. The geometric-programming approach scales linearly with the number of gates and quadratically with the number of bins in the histogram. This translates, for example, to solving the SSTA for a circuit of 400 gates with 30 bins per each histogram approximation of a random variable in 440 seconds.