Optimum probability model selection using Akaike's information criterion for low power applications

Abstract

Optimal probability model selection for power estimation in low power VLSI applications is studied. Akaike's information criterion is used to estimate the optimal number of components in a mixture density model for the simulated power data. Theory behind the proposed algorithm is discussed followed by experimental results for ISCAS '85 benchmark circuits and a large industrial circuit. The method is shown to perform well for both large and small circuits even when the number of observed samples is small. The algorithm is promising as a pre-processing step to automatically compute the optimal probability model before any other power estimation procedure is applied. We also note that the method is applicable to other problems in VLSI for model selection.