Statistical estimation of average power dissipation in sequential circuits

Abstract

In this paper, we present a new statistical techniquefor estimating average power dissipation in sequential circuits.Due to the feedback mechanism, in sequential circuitspower dissipation in consecutive clock cycles are temporallycorrelated, which violates the basic requirement ofstatistical mean inference procedures. We overcome thisproblem by using a randomness test and a sequential procedureto select a proper independence interval, which in turnis used to generate random power samples. A distribution-independentstopping criterion is applied to analyze thesample data and terminate the simulation upon achievementof the accuracy specification. The technique is successfullyapplied to a set of benchmark circuits.