The Effect of Filling the Unspecified Values of a Test Set on the Test Set Quality

Test generation and test data compression processes create test vectors with unspecified input values that need to be filled. We study the extent to which filling the unspecified input values affects the untargeted fault coverage of a test set. To make the study independent of any particular test generation or test data compression scheme, we consider test sets for stuck-at faults that are obtained by first unspecifying as many values as possible without losing stuck-at fault coverage, and then filling the unspecified values randomly. The results indicate that there are significant differences in the untargeted fault coverage between different test sets. The differences in the average number of detections of stuck-at faults are less noticeable. We also show that adding a small fraction of untargeted faults to the set of faults considered during the unspecifying process improves significantly the untargeted fault coverage after filling of unspecified values.