Using Neural Networks for Novelty-based Test Selection to Accelerate Functional Coverage Closure

TL;DR

This paper introduces a neural network-based framework for test selection in simulation-based verification, significantly reducing simulation time to achieve high coverage efficiently.

Contribution

It presents a configurable, automated neural network framework for test selection that outperforms random methods in coverage closure speed and efficiency.

Findings

Up to 49.37% simulation reduction to reach 99.5% coverage

Neural network configurations outperform random selection

Computational overhead is negligible compared to simulation savings

Abstract

Novel test selectors used in simulation-based verification have been shown to significantly accelerate coverage closure regardless of the number of coverage holes. This paper presents a configurable and highly-automated framework for novel test selection based on neural networks. Three configurations of this framework are tested with a commercial signal processing unit. All three convincingly outperform random test selection with the largest saving of simulation being 49.37% to reach 99.5% coverage. The computational expense of the configurations is negligible compared to the simulation reduction. We compare the experimental results and discuss important characteristics related to the performance of the configurations.