Neural-fuzzy machine learning approach for the fatigue-creep reliability modeling of SAC305 solder joints

Abstract

The accuracy of reliability models is one of the most problematic issues that must be considered for the life of electronic assemblies, particularly those used for critical applications. The reliability of electronics is limited by the fatigue life of interconnected solder materials, which is influenced by many factors. This paper provides a method to build a robust machine-learning reliability model to predict the life of solder joints in common applications. The impacts of combined fatigue and creep stresses on solder joints are also investigated in this paper. The common alloy used in solder joint fabrication is SAC305 (Sn–Ag–Cu). The test vehicle includes individual solder joints of SAC305 alloy assembled on a printed circuit board. The effects of testing temperature, stress amplitude, and creep dwell time on the life of solder joints were considered. A two-parameter Weibull distribution was utilized to analyze the fatigue life. Inelastic work and plastic strain were extracted from the stress–strain curves. Then, Artificial Neural Networks (ANNs) were used to build a machine learning model to predict characteristic life obtained from the Weibull analysis. The inelastic work and plastic stains were also considered in the ANN model. Fuzzy logic was used to combine the process parameters and fatigue properties and to construct the final life prediction model. Then a relationship equation between the comprehensive output measure obtained from the fuzzy system and the life was determined using a nonlinear optimizer. The results indicated that increasing the stress level, testing temperature, and creep dwell time decreases reliability. The case of long creep dwell time at elevated temperatures is worst in terms of impact on reliability. Finally, a single robust reliability model was computed as a function of the fatigue properties and process parameters. A significant enhancement of the prediction model was achieved compared to the stress–life equations.

Low-cycle fatigue life assessment of SAC solder alloy through a FEM-data driven machine learning approach

“…The predictions are evaluated using ML-processed experimental data, effectively capturing the high sensitivity of thermal aging behavior to solder joint thickness, geometric dimensions and other parameters. Bani Hani et al (2023b) presented a robust ML reliability model for predicting the life of solder joints in electronic assemblies, considering the combined effects of fatigue and creep stresses. The model incorporated features such as inelastic work and plastic strain, and fuzzy logic was used to combine process parameters and fatigue properties.…”