Logic circuit diagnosis by using neural networks

Abstract

This paper presents a new method of logic diagnosis for combinatorial logic circuits. First, for each type of circuit gates, an equivalent neural network gate is constructed. Then, by replacing circuit gate elements with corresponding neural network gates, an equivalent neural network circuit is constructed to the fault-free sample circuit. The testing procedure is to feed random patterns to both the neural network circuit and the fault-prone test circuit at the same time, and comparing, analyzing both outputs, the former circuit generates diagnostic data for the test circuit. Thus, the neural network circuit behaves like a diagnostic engine, and needs basically no preparation of special test patterns nor fault dictionary before diagnosing.