Automatic selection of test frequencies for the diagnosis of soft faults in analog circuits

In this paper a fault diagnosis technique for electronic analog circuits is described. Diagnosis is obtained by comparing input-output measurements with examples contained in a fault dictionary, by means of a neural classifier. A harmonic analysis is used, i.e. the test input stimuli are sinusoidal waves. A novel method for optimizing the fault dictionary construction is proposed. In particular, the stimuli selection is optimized by means of a sensitivity analysis of the Circuit Under Test-CUT-relying on a probabilistic approach based on randomized algorithms. This technique allows removing all the hypothesis assumed by the related literature. In fact it allows to remove the small perturbation assumption and presents a poly-time solution independently from the dimension of the perturbation