Stimuli Redundancy Reduction for Nonlinear Functional Verification Coverage Models Using Artificial Neural Networks

Abstract

As functional verification persists in being one of the most demanding and tedious tasks of SoC development, the research community continues to explore expert systems that reduce the time cost for reaching coverage closure. Some typical coverage items that are difficult to fill using Machine Learning inference are the coverpoints with nonlinear probability distributions, such as power-of-two values or "min & max" values. This paper presents an efficient solution based on Artificial Neural Networks that efficiently reaches coverage closure for such coverpoints. This article highlights the solution implementation, underlines the experimental results, and states suggestions for further research.