With the increased availability of approximate computing circuit approaches,evaluating their reliability becomes highly necessary.This paper proposes an accurate and efficient approximate circuit reliability assessment method based on the probabilistic gate model(PGM).This method conducts targeted reliability analysis on the input vector.Firstly,the circuit is transformed into a probabilistic polynomial using the probabilistic gate model.It reduces the circuit variables causing correlation issues at the fan-out end gates to eliminate such problems.The method evaluates the reliability of the approximate circuit by considering the impact of result accuracy.Experimental results demonstrate that this approach is significantly faster than the Monte Carlo method by five orders of magnitude and 50%faster than the other method.It ensures high accuracy at a comparable speed in large-scale circuits.Furthermore,a genetic algorithm is employed based on the proposed method to search for the reliability bounds of approximate circuits,assisting designers in evaluating the applicability of these circuits.In response to the issues addressed in this article,the genetic algorithm has been enhanced.Experiments results show that the improved genetic algorithm yields better results,demonstrating the effectiveness of the method proposed in this paper for approximate circuit reliability calculation and reliability boundary search.
NETL
NSTL
万方数据
