Improving Performance of Effect-Cause Diagnosis with Minimal Memory Overhead

Abstract

Effect-cause diagnosis procedures are the most commonly used in industry to diagnose VLSI circuits that fail manufacturing test or field applications. Fast and effective diagnosis procedures are essential to diagnose large numbers of failing dies for yield ramp-up. We have recently proposed a method to speed up effect-cause diagnosis procedures by using a dictionary of small size [26]. In this paper we propose methods to further reduce the dictionary size and still achieve higher performance. Experiments on several industrial designs demonstrate that, on average, effect-cause diagnosis procedures can be speeded up by 3.5X while requiring minimal memory overhead for a very small dictionary.