Gradual Diagnostic Test Generation Based on the Structural Distance Between Indistinguished Fault Pairs Irith Pomeranz 1 School of Electrical & Computer Eng. Purdue University W. Lafayette, IN 47907, U.S.A. pomeranz@ecn.purdue.edu Sudhakar M. Reddy 2 Electrical & Computer Eng. Dept. University of Iowa Iowa City, IA 52242, U.S.A. reddy@engineering.uiowa.edu Abstract - The size of a diagnostic test set is significantly larger than the size of a fault detection test set. As a result, fault detection test sets may be used for initial defect diagnosis, and diagnostic tests may be added only when needed to narrow down a set of candidate defect sites. Between a fault detection test set and a full diagnostic test set there is a large range of test sets that can be used for improved (initial) diagnosis. We describe a diagnostic test generation process that produces such a range of test sets. As part of the process we rank the indistinguished fault pairs of the fault detection test set according to the importance of distinguishing them, and perform diagnostic test generation based on the ranked list. The ranking of fault pairs is based on the structural distance between the two faults of a pair. This allows failure analysis to explore fewer and more localized areas of the circuit as the size of the diagnostic test set is increased. Diagnostic test generation can stop with a test set of any size between a fault detection and a full diagnostic test set, having targeted the fault pairs that are most important to distinguish. Keywords - defect diagnosis, diagnostic test generation, failure analysis, test generation. I. INTRODUCTION Defect diagnosis procedures [1]-[25] use the results of test application to determine the sites where defects are likely to be present in a circuit that failed under the application of a test set. These sites are referred to as candidate defect sites. In general, candidate defect sites are identified based on a comparison between the observed response of the failing circuit and the responses of modeled faults to the applied test set. For a candidate defect site, the presence of a modeled fault at the site produces a faulty response that matches the observed failing response of the applied test set. The definition of a match takes into account that the behavior of a defect may be different from the behavior of any modeled fault. For example, in [22], [24] and [25], stuck-at faults are used as a basis for diagnosis. Deviations in defect behavior from the behavior of stuck-at faults are accommodated by using scoring algorithms that estimate the likelihood of a defect being associated with the site of a stuck-at fault given the response of the failing circuit. The accuracy of defect diagnosis depends directly on the test set used for diagnosis. Diagnostic test generation [26]-[31] is a process whose goal is to produce a test set T where, for every pair of modeled faults f 1 and f 2 , there is a test t ∈ T that distinguishes f 1 and f 2 . A test t is said to distinguish f 1 and f 2 if the circuit produces a different output response depending on whether f 1 or f 2 is present in the circuit. To explain the importance of distinguishing fault pairs to defect diagnosis, we consider two faults f 1 and f 2 that are not distinguished by the test set T . If f 1 is identified by a defect diagnosis procedure as a match for an observed response of a failing circuit, a defect at the site of f 2 is equally likely to be present in the circuit, and both faults will contribute to the set of candidate defect sites. In contrast, if f 1 and f 2 are distinguished by T , a defect diagnosis procedure may identify that one of the faults is a better match than the other for the observed response of a failing circuit, and include only one of the faults in the set of candidate defect sites. Thus, a diagnostic test set that distinguishes as many fault pairs as possible contributes to more accurate defect diagnosis. The target faults considered by the diagnostic test generation procedures in [26]-[31] are single stuck-at faults. Although defects are not likely to behave exactly like stuck-at faults, diagnostic test generation based on single stuck-at faults is justified by the following observations. 1. Research supported in part by SRC Grant No. 2007-TJ-1643. 2. Research supported in part by SRC Grant No. 2007-TJ-1642. 2010 25th International Symposium on Defect and Fault Tolerance in VLSI Systems 1550-5774/10 $26.00 © 2010 IEEE DOI 10.1109/DFT.2010.49 349