Weighted random patterns with multiple distributions

Abstract

It is well known that random test lenths can be reduced by orders of magnitude using biased random patterns. But there are also some circuits resistant to optimising. In this paper it is shown that this problem can be solved using several distributions instead of a single one. Firstly we compute bounds of the error caused by the assumption that fault detection consists of completely independent events. Secondly we prove a sharp estimation of the error caused caused by assuming the random property instead of the pseudo-random property of shift register sequences. Finally a heuristic is presented in order to compute an optimal number of random pattern sets, where each set has its specific distribution and its specific size.