摘要
As Moore's Law approaches its limit, advanced packaging technologies are continually being introduced to enhance efficiency and reliability, aiming to overcome or surpass the current limitations in semiconductor technology. Design-on-simulation technology has been widely used for designing advanced packaging structures. It can reduce the time and cost of obtaining reliability results; however, different researchers' models may yield varying results. To reduce the discrepancy, this research combines simulation and artificial neural networks (ANNs) to develop a prediction model that minimizes the differences between various research results. In this study, an ANN is employed to fit Chaboche material models across varying strain rates, enabling the precise determination of stress-strain curves. To assess the reliability of solder joints, two fatigue life models are utilized: the Coffin-Manson strain-based approach and the energy density method. For the Coffin-Manson model, two approaches are investigated: one involving strain-rate-dependent parameters and another using fixed parameters while adjusting the critical mesh size. The results indicate that both approaches produce stable and accurate predictions, highlighting the effectiveness of integrating ANN-based modeling with finite element analysis for reliability assessment in wafer-level packaging.