Failure probability analysis and critical node determination for approximate circuits

Abstract

Approximate circuits (ACs) relax the constraint of 100% functional correctness, and can be utilized in many emerging applications, which increases the importance of their reliability analysis. Here, a model and an algorithm are proposed for computing the failure probability of ACs and for determining the critical nodes in these circuits. The presented method is not limited to the type and scale of ACs. Comparison of the results on three series of ACs shows that, the proposed analytical method is more than 98% accurate and four orders of magnitude faster than the random fault injection method. Experiments on circuits in the EvoApprox8b library demonstrate the plausibility of the calculated failure probability and distribution of critical nodes.