Modified Norris-Landzberg Model for Reliability of Pb-free BGA Components

Abstract

One important task for reliability engineers in new product design phase is to evaluate the long-term reliability of Pb-free solder joints under the service environment conditions. It is a common practice, due to the practical restrictions such as cost and aggressive development cycle time, to map the results under accelerated testing conditions to field conditions via well-benchmarked transform models. One of the most widely adopted models is the modified version based on Norris-Landzberg (NL) Acceleration Factor (AF) model [1]. The popular NL AF model simultaneously makes it easy to apply but difficult to correlate with data from different sources. In fact, there is ongoing discussion that questions the validity of the model [2]. To overcome the shortcomings of the model, Salmela [3] has suggested a correction term to consider package type and solder material along with the temperature and the dwell time on the solder joint. Chuang, et al. [4] have compared several models and confirmed that the prediction based on Salmela’s model is closer to their testing results. Ahmad, et al. [5] have proposed the constants of NL AF model as functions of the characteristic life under the accelerated testing conditions. All of those aforementioned variations of NL AF models, while improving the model prediction, would still need different constants for data under different situations; therefore, it is not convenient to use especially in product design phase where some critical information such as the package detail and testing results may not be available.In this paper, a unified NL AF model has been proposed by introducing an add-in impact function to factor package-related contributors such as package size, die size, joint pitch, and delta CTE. It is demonstrated that such an impact function enables NL AF model to correlate very well with a wide range of experimental data using one unified set of constants that were derived by nonlinear fitting of 112 data points from internal source and public domain publications. The robustness of the derived constants was then validated using the generic algorithm to ensure the best of the goodness-of-fit while minimizing the standard residual error. The resulting model has $\mathrm{R}^{2}$ equals 0.845 which showed a good correlation with the testing results. Furthermore, the derived model has been benchmarked with the testing results of iNEMI Pb-free solder alloy alternative project ([6]–[9]), an in-depth industry wide collaboration. The testing results are representable as two types of components with different solder alloys have been tested under multiple temperature cycling profiles. The benchmarking demonstrates a really good correlation as the majority of the predicted life, from the derived model, fall into + /-50% band compared with the testing results. Hence the proposed NL AF model, with a unified constant set, can be very helpful and handy in product design phase for reliability assessment.