Improving Diagnosis Resolution with Population Level Statistical Diagnosis

Abstract

In this paper, we present a diagnosis resolution improvement methodology for scan-based tests. We achieve 89% reduction in the number of suspect diagnosis locations and a 2.4X increase in the number of highly resolved diagnosis results. We suffer a loss in accuracy of 1.5%. These results were obtained from an extensive silicon study. We use data from pilot wafers and 11 other wafers at the leading-edge technology node and check against failure analysis results from 203 cases. This resolution improvement is achieved by considering the diagnosis problem at the level of a population (e.g. a wafer) of failing die instead of analyzing each failing die completely independently as has been done traditionally. Higher diagnosis resolution is critical for speeding up the yield learning from manufacturing test and failure analysis flows.