A direct search method for finding the worst performance measure in integrated circuit design

We propose a direct search algorithm for solving the worst performance measure problem in integrated circuits. The algorithm is based on the Hooke-Jeeves algorithm. We confirm its reliability by comparing its results to the results of a Monte-Carlo analysis. In terms of speed the algorithm takes the equivalent of 3 to 9 finite difference gradient evaluations for finding a single worst performance measure value. Finally we use the algorithm in the process of robust design and confirm the obtained yield with Monte-Carlo analysis.