Test Point Insertion Using Artificial Neural Networks

Abstract

A method of data collecting, training, and using artificial neural networks (ANNs) for evaluating test point (TP) quality for TP insertion (TPI) is presented in this study. The TPI method analyzes a digital circuit and determines where to insert TPs to improve fault coverage under pseudo-random stimulus, but in contrast to conventional TPI algorithms using heuristically-calculated testability measures, the proposed method uses an ANN trained through fault simulation to evaluate a TP's quality. The time of feature extraction is demonstrated to be significantly faster compared to heuristic-based TP evaluation, and the impact of inserted TPs is shown to provide superior stuck-at fault coverage compared to conventional heuristic-based testability analysis.