A method for the evaluation of behavioral fault models

Many fault models have been proposed which attempt to capture design errors in behavioral descriptions, but these fault models have never been quantitatively evaluated. The essential question which must be answered about any fault model is, "If all faults in this model are detected, is the design guaranteed to be correct?" In this paper we present a method to examine the degree to which an arbitrary fault model can ensure the detection of all design errors. The method involves comparing fault coverage to error coverage as defined by a practical design error model which we describe. We have employed our method to perform a limited analysis of the statement and branch coverage fault models.