Diagnosis of CMOS op-amps with gate oxide short faults using multilayer perceptrons

Abstract

CMOS operational amplifier transistors containing a single gate oxide short (GOS) fault between the source and drain were diagnosed from SPICE simulations of the supply current responses to ramp and sinusoidal test stimuli. Multilayer perceptron (MLP) artificial neural networks were trained to classify the faulty transistors from the responses. Functional testing did not always reveal the GOSs so this method offers reliability testing against future failure since the GOSs can deteriorate during operation of the circuit. The GOSs were modeled by a diode and series resistance at various distances from the source. The breakdown voltages of the model diode significantly affected the responses and diagnostic accuracies. If they are in the expected practical range (/spl les/2 V) and are uniform in value, then by combining test results from both stimuli, accuracies of 100% are obtainable. If their values are variable or higher, the accuracies decrease, and the test reduces to a go/no go test. No test pins are required so the method is applicable to any circuit.