A method to diagnose faults in analog integrated circuits using artificial neural networks with pseudorandom noise as stimulus

Abstract

This paper describes a new, fast and economical strategy for fault diagnosis of analog integrated circuits. The methodology is based on a technique of using a pseudo random noise generator as the test pattern generator and a model-based observer, which is implemented through a feed forward artificial neural network in the form of a single hidden-layer perceptron. The proposed methodology can be implemented in any personal computer with a data acquisition card for on-line operation. Its main advantages are the low time requirement for learning and diagnosing. The method is quite robust and is able to detect small component variations without problems. This technique has been successfully applied to diagnose both hard and soft faults in a bipolar junction transistor based operational amplifier and a MOS operational amplifier.