A better Way to Predict Reliability from Lifetest Data Using a Regression Technique
John Scarpulla, David C. Eng, and Denise L. Leung
TRW Inc., One Space Park Drive, Redondo Beach, CA 90278
In order to predict reliability of GaAs Integrated circuits, the lognormal-Arrhenius model is often used. An extrapolation to ordinary usage temperatures is often justified with the model. However, lifetest data is often incomplete (censored), making a simple regression analysis difficult. A regression technique is introduced here for the analysis of lifetest data. It solves these difficulties, is rigorous and reproducible, and is easy to implement. It also handles the setting of confidence limits on lifetime at the desired probability level. Also addressed is the important problem of specifying the test conditions against which product wafers are to be screened.