Abstract
A credible statistical algorithm is required to determine the parameters of the bimodal Weibull mixture distribution exhibited by the gate oxide breakdown phenomenon, consisting of extrinsic early-life defect-induced failures and intrinsic wear-out failure mechanisms. We use a global maximization algorithm called simulated annealing (SA) in conjunction with the expectation-maximization (EM) algorithm to maximize the log-likelihood function of multi-censored gate oxide failure data. The results show that the proposed statistical algorithm provides a good fit to the stochastic nature of the test failure data. The Akaike information criterion (AIC) is used to verify the number of failure mechanisms in the given set of data and the Bayes' posterior probability theory is utilized to determine the probability of each failure data belonging to different failure mechanisms. (C) 2007 Elsevier Ltd. All rights reserved.