Guardbands in Random Testing

The fault coverage of a random test can be estimated by fault simulation. If the simulation is performed by another random sequence than those used under test or a fault sample is used, a random difference between the simulation result and the fault coverage has to be considered. The simulation result must be larger than the fault coverage that has to be guaranteed. The difference is called guardband. In the paper the distribution of the fault coverage and the distribution of the difference has been derived by the mathematical model of independently detectable faults. Afterwards it is corrected using experimental data. The comparison between theory and experiment unveils a feature of random test, to which no attention has been paid in the past. The correlations in the fault detection process can not be ignored in determining guardbands. As the final result relations for guardband calculation are given.