Temperature-Dependent Variances of Distributions of Oxide Breakdown and Electromigration Failure Times

The standard accelerated-life-test model assumes a lognormal distribution of failure times. While the dependence of the median time-to-failure on stress has been studied extensively, the dependence of the variance of 1n t (sigma sup 2) on stress has been largely neglected. A number of theoretical and experimental studies show sigma of 1n t increasing with decreasing stress. Schwarz points out that if sigma is assumed to be independent of temperature, then the reliability prediction at operating conditions would be too optimistic. This paper proposes a temperature-dependent sigma relation based on the Arrhenius-like relation publish by Schwarz.