Statistical methods for delay fault coverage analysis

We have developed new statistical techniques for delay fault analysis. True value simulation is performed using a multi-value logic system describing signal states of two consecutive vectors. Signal statistics are used to estimate transition probabilities and observabilities. These allow us to estimate detection probabilities and the coverage for transition faults. For path delay faults, recognizing that the total number of possible paths can be exponential in circuit size, we devise an implicit random path sampling procedure to obtain a linear-time estimate of the coverage for all faults. We further derive a longest path theorem to estimate coverages with respect to longest paths through primary inputs and a selected set of fanout branches. Coverages are estimated for a minimal set of longest paths such that each signal lead is included in at least one target path whose propagation delay is no less than the delay of any path containing the lead. The fault coverage estimates closely agree with those obtained from delay fault simulation while giving a significant speedup