Reliability-Based Design Optimization of BGA Assembly Using Hybrid FEM-ANNs Approach

Abstract

In this study, we propose a hybrid framework combining finite element method (FEM) simulations, machine learning and metaheuristic algorithms to optimize the reliability of Ball Grid Array (BGA) assemblies subjected to thermomechanical loads. The framework aims to simulate the thermomechanical behavior of the component and predict its fatigue life using the Coffin-Manson model. The main geometric and material parameters of the electronic component are selected to build a dataset from the FEM outputs. Then, artificial neural networks (ANNs) are trained to predict the FEM results, improving computational efficiency. Next, the search and rescue (SAR) algorithm is applied to optimize component parameters, both to maximize predicted reliability and the lifetime of solder joints. The results demonstrate that the hybrid framework reduces the computational cost associated with running FEM simulations, and allows more effective design optimization through prediction and maximization of component reliability. This framework offers a powerful tool for reliability-based design optimization in electronic packaging, with potential applications in a variety of electronic components.