Modeling the unknown! Towards model-independent fault and error diagnosis

In this paper, we provide techniques for fault and error diagnosis based on capturing unmodeled faulty behavior. We present a technique for capturing the effects of all possible faulty behaviors that can be generated from specific sets of nodes (called X-lists) in the circuit. Since all possible erroneous behaviors are captured, this provides a way for drawing powerful diagnostic inferences about the presence of faults at these sets of nodes when analyzing the observed faulty responses. We also present an efficient diagnosis algorithm that exploits the modeling of all possible behaviors and can be built in a framework of conventional test and simulation tools. Experimental results with numerous diagnosis experiments are then used to demonstrate that the techniques developed can indeed be used to achieve significant improvements in the accuracy of diagnosis