Analysis of Accelerated Temperature Cycle Test Data Containing Different Failure Modes

This paper addresses the problem of separating out the failure distribution of each kind of failure observed in accelerated temperature-cycle aging of one assembly lot of a 16-lead molded DIP package. A failure distribution is established for each of five kinds of failure (i.e., failure modes) and at each of two different temperature-cycle aging conditions. From these separate and independent distributions, the distribution parameters and an acceleration factor are determined for each failure mode. The data are analyzed by use of the hazard plotting technique developed by Nelson.1 The necessary algorithms for hazard plotting are developed to handle data of the type obtained in this experiment. Two characteristics of the data that the algorithms take into consideration are first, it is grouped data (i.e., the number of failures within a time interval is the measured quantity) and second, the failures are effectively "non-interfering" (i.e., the occurrence of a failure for one failure mode does not interfere with either the aging or observation of other failure modes). In addition, it is further developed so that the results can be plotted on any of several standard probability papers (normal, log-normal, Weibull, etc.). The analysis of these aging data shows that there is wide variation in the acceleration factors, median lives and dispersions (sigmas) for the different failure modes. It further shows that these parameter differences account for much of the bimodal structure of the total or composite failure distribution.