Abstract
The electromigration (EM) phenomenon often exhibits two failure mechanisms. The early failures are due to voids developed in the via or at the via-line interface while the wear-out failures are predominantly attributed to voids developed within the interconnect line itself Presence of these two failure mechanisms necessitates the use of a bimodal mixture distribution for analyzing EM data. Also, due to the presence of incubation time in EM, the three-parameter lognormal distribution should be applied to account for the initial failure-free duration. An efficient global optimization technique called "simulated annealing" is introduced in this work to globally optimize the log-likelihood function of the mixture distribution and determine the values of the lognormal parameters that accurately fit the experimental EM failure data. (c) 2008 Elsevier B.V. All rights reserved.