All Tests for a Fault are Not Eyually Valuable for Defect Detection

Abstract

The standard approach to generating a test set for a logic circuit is to select a set of target fault,s and generate one test for each fault (or at least most of the faults). By far, the most common target fault set is composed of single stuck-at faults. However, many potential manufacturing (non-target) defects are not included in the target fault set, so that their detection is just a fortuitous coincidence. In most cases, many different tests may exist which detect a given single stuck-at fault. However, each of these tests (for the same stuck-at fault) may perform very differently in terms of their non-target defect detection. We investigate this phenomenon empirically using a subset of the ISCXS 85 benchmark circuits, and our results show that extremely wide variations in defective part level are possible over sets of tests which all guarantee 100% single non-redundant stuck-at fault coverage. Our objective in this work is the investigation of non-target defects and their impact on defective part level. Here, our thrust is not to develop new test gene rut i o n methods .