Symbolic Power Analysis of Cell Libraries

Abstract

Cell libraries are collections of logic cores (cells) used to construct larger chip designs; hence, any reduction in their power consumption may have a major impact in the power consumption of larger designs. The power consumption of a cell is often determined by triggering it with all possible input values in all possible orders at each state. In this paper, we first present a technique to measure the power consumption of a cell more efficiently by reducing the number of input orders that have to be checked. This is based on symbolic techniques and analyzes the number of (weighted) wire chargings taking place. Additionally, we present a technique that computes for a cell all orders that lead to the same state, but differ in their power consumption. Such an analysis is used to select the orders that minimize the required power, without affecting functionality, by inserting sufficient delays. Both techniques have been evaluated on an industrial cell library and were able to efficiently reduce the number of orders needed for power characterization and to efficiently compute orders that consume less power for a given state and input-vector transition.