The investigation of modified Norris-Landzberg acceleration models for reliability assessment of Ball Grid Array packages

Norris-Landzberg acceleration model, only applying the conditions of the temperature cycling tests such as temperature range, maximum of temperature and cycle times, has been widely using to predict the lifetimes of electronic packages since 1969, due to its simplicity not considering complex damage information such as stress-strain relationships. During the last few years, a number of researches have shown that the N-L model exponents are not constant depending on the type of packages and solder materials (ex. the transition from SnPb solders to lead free solders like SnAgCu alloy) resulting in the variation of acceleration factors. It will increase the average errors of prediction if the original model is continually using without any modification especially for Ball Grid Array (BGA) packages since they are nowadays in the majority in this industry with many types for different applications, moreover, commonly stress-sensitive. The lifetime data of several BGAs presented in this study show that these exponents and acceleration factors depends on the different solder materials, solder pastes, design parameters (die size, package size, ball pitch, and ball size), and even prebaking conditions. Therefore, the modification of N-L model is necessary for the BGAs in order to precisely predict their lifetimes in the field level. According to our survey, not many modified N-L models are presented for all packages. In this paper, two modified N-L models developed by Salmela and Dauksher were introduced respectively since they are likely to easy use. Compared to the original model developed by Norris and Landzberg, Salmela demonstrated that his stress-dependent model is able to reduce the average errors of prediction from 52.4% to 17.8% using the data with respect to different package types and solder materials. Our data were applied to validate Salmela's model, and the results show that the average errors of prediction decrease from 41% to 32%. In case of Dauksher's model, the average errors of prediction also decrease from 41% to 36%. As the results, our data have demonstrated that these two models are capable of providing the better accuracy in predicting the lifetimes of BGAs compared to the original N-L model, though they have improved yet the average errors so much. Furthermore, the Norris-Landzberg-based acceleration models may keep its validity on the reliability assessment of Ball Grid Array packages.