In this paper we propose a novel back-propagation neural network power macromodeling technique that captures the dependence of power dissipation in CMOS circuits on its primary input signal statistics.Features of the input streams of the circuits are extracted to model average power based on BP neural network.In contrast to previously proposed macromodels

it can be used to do RTL power analysis for any given logic circuit without a large look-up table and it dose not require empirically construct specialized analytical equations for the power macromodel.In experiments with the ISCAS-85 circuits

the average absolute relative error of our power macromodel is below 10%.The root-mean-square error is about less than 5% for most circuits.