Statistical reliability prediction

Abstract

An analysis technique is presented which allows the construction of a statistical model of device lifetime given data obtained from traditional reliability life testing. Unlike techniques presently used, this method does not require that each life test be completed to a high percentage failure, produces a model whose accuracy is quantifiable, and can predict expected times to failure at percentages other than 50%. Confidence and prediction intervals can be calculated around all model parameters (such as activation energy), all performance parameters (such as estimated median time to failure, MTF), and individual performance predictions (estimated times-to-failure, TTF). This procedure is demonstrated using the electromigration failure mechanism, but can be applied to any failure model.