Moment-based estimation of switching activity in non-linear architectures

The significant power optimization possibilities in the early stage of the design flow demand the use of energy estimation techniques at high levels of abstraction. In this work, we address the modeling of the switching activity in non-linear architectures, where the classical techniques based on Gaussian models cannot be applied. Using the characterization of the probability density function (PDF) with a projection in an orthogonal polynomial base, a technique is presented to calculate the switching activity from the moments of the signal. The approach has been validated with circuits and excitations from realistic applications. Comparisons with reference bit level simulations are reported to assess the accuracy of the technique.