ON THE MASKING PROBABILITY WITH ONE'S COUNT AND TRANSITION COUNT.

The masking probability of two data compression techniques, one's count and transition count, is considered. It is shown that if the inputs are randomly applied, the masking probability of a fault asymptotically approaches ( pi N)-** one-half , where N is the length of the input sequence. This indicates that for long input sequences, the masking effect can be ignored. For three most common data compression techniques, namely, signature analysis, one's count, and transition count, the masking probability is practically negligible if the test length is already of modest size. Random pattern testable design for self-test should concentrate on assuring that all faults are detected at the primary outputs, and pay no attention to which of the three compression techniques is actually used, provided the test length is sufficiently long.