Statistical estimation of average power dissipation in CMOS VLSI circuits using nonparametric techniques

We present a new statistical technique for estimation of average power dissipation in digital circuits. Present statistical techniques estimate the average power based on the assumption that the power distribution can be characterized by a preassumed function. A large error can result if the assumption is not met. To overcome this problem, we propose a nonparametric technique in which no distribution function needs to be assumed. A set of distribution-independent upper and lower bounds of the average power are developed using the properties derived from the order statistics. A stopping criterion is designed based on the bounds for a desired percentage error with a specified confidence level. Since it does not resort to assuming any particular distribution function, the technique can be applied to all circuits irrespective of their power distributions. Comparison is made against the present statistical technique based on the central limit theorem. Experimental results show that the proposed technique is much more accurate and robust, yet the efficiency characteristic of statistical techniques is still preserved.