Yield fluctuations and defect models

Abstract

This paper establishes a general statistical framework for analyzing wafer fallout data in order to obtain information about process induced defects. The statistical behavior of fallout during testing is characterized. A general model is presented that relates the yield to the test coverage and the defect distribution. This model incorporates arbitrary variations in the occurrence probabilities of different defects and arbitrary dependencies between the defects, although it still ignores correlations between the occurrence probability of a defect and whether or not this defect is detected. A statistical acceptance criterion for this model is proposed, based on a convexity property of the model. It is applied to a published set of yield data and shows that some striking features in these data cannot be due to mere statistical fluctuations. Some explanations of these features will be discussed. © 1995 Kluwer Academic Publishers.