Computation of lower bounds for switching activity using decisiontheory

摘要

Accurate switching-activity estimation is crucial for powernbudgeting. It is impractical to obtain an accurate estimate bynsimulating the circuit for all possible inputs. An alternate approachnwould be to compute tight bounds for the switching activity. In thisnpaper, we propose a nonsimulative decision theoretic method to computenthe lower bound for switching activity. First, we show that thenswitching activity can be modeled as the decision error of an abstractntwo-class problem. It is shown that the Bayes error L* is a lower boundnfor the switching activity. Further, we improve L* to obtain a tighternbound L1, which is based on the one-nearest neighbornclassification error. The proposed lower bounds are used fornswitching-activity characterization at the register transfer (RT) level.nExperimental results for the RT-level switching-activity estimates fornISCAS'85 circuits are presented. This technique is simple and fast andnproduces accurate estimates