Estimation of power sensitivity in sequential circuits with power macromodeling application

We propose a novel technique based on Markov chains to accurately estimate power sensitivities to primary inputs in CMOS sequential circuits. The power sensitivity defines the change in average power dissipation due to changes in the input signal specification. Such sensitivities are estimated as by-products of the average power estimation, leading to an efficient implementation. A key application of power sensitivities is to construct a power surface in the specification space so that power dissipation under any distribution of primary inputs can easily be obtained, thereby providing an effective power macromodel for high level power estimation. We demonstrate that such a power surface can be approximated by only a limited number of representative points. This will dramatically reduce the CPU and memory requirements. Results on a large number of benchmark circuits have verified the feasibility and accuracy of this technique.