Fault diagnosis of mixed signal VLSI systems using artificial neural networks

This paper presents an approach to the diagnosis of linear and nonlinear analog circuits. The diagnosis methodology is focused on the soft faults in analog circuits. An on-chip white noise generator provides the test stimulus and an artificial neural net (ANN) is used as the response evaluator. Our analysis shows that the white noise relative to the pole zero locations of the circuit transfer function has a significant impact on the classification efficiency of ANN. White noise based stimulus method works for some nonlinear circuits as long as they are constrained to operate in their small signal region of operation. Circuits with strong nonlinearity are difficult to diagnose using the noise stimulus approach. Our results are demonstrated for a linear filter, Schmidt trigger and the phase lock loop (PLL).