Abstract
This paper presents an efficient transition-fault diagnosis pattern generation procedure to identify equivalent-fault pairs and generate diagnosis patterns for nonequivalent-fault pairs. Two major techniques are proposed. The first one is a fault-inactivation method that can quickly distinguish most fault pairs by inactivating one fault while detecting the other in each fault pair. The second one is a fault-transformation method that can transform the problem of distinguishing two transition faults into the problem of detecting a transition fault. Both methods involve only one copy of the original circuit and require only an ordinary ATPG tool for transition faults. Furthermore, both methods can deal with multiple fault pairs at the same time and thus not only the total CPU time can be significantly reduced but also the dynamic test compaction capability of the ATPG tool can be utilized. Experimental results on all possible transition-fault pairs of both ISCAS'89 and IWLS'05 benchmark circuits show that the fault-inactivation method can distinguish about 95.6% of distinguishable fault pairs quickly and the fault-transformation method can deal with almost all the remaining indistinguished fault pairs. The average ratio of the number of diagnosis patterns over that of original test patterns is only 0.41.