Towards the on-line diagnosis of faults in mixed signal circuits: a delta sigma converter

Abstract

There is interest in the testing and diagnosis of mixed analogue and digital circuits. Analogue blocks are difficult to diagnose because of the wide ranges of voltages, currents, and impedances possible, and because of feedback effects. High component densities and the limited number of pads or pins mean that it is desirable to develop methods that utilise those available. An approach to the problem is illustrated in the paper with reference to a delta-sigma modulator. The possible consequences of short or open circuits are obvious, but other faults such as incorrectly inserted components or out of tolerance components caused by process parameter variations or faulty processes may be revealed. The idea was to develop a method for diagnosing mixed signal circuits that would be useful on the production line. A test signal would be applied at the input of the circuit and the corresponding output bitstream recorded. Hopefully faults in the circuit would modify the bitstream from that of a nominal circuit, thus permitting fault detection (the test operation). If the resulting bitstreams were characteristic of the different faults, then these could be identified (diagnosis), or even quantified. It was therefore necessary to extract characteristic features from the bitstreams and to identify them. The former was done by some basic signal processing, whilst it was proposed to achieve the latter using an artificial neural network (ANN). An important constraint for this application was that the ANN could be used alternatively for training and testing without necessitating a total retraining whenever new training vectors or new types of fault were encountered. For these reasons the simplified fuzzy ARTMAP is used as the fault classifier. (6 pages)