Logo image
Use Artificial Neural Network to fit stress-strain curve of Chaboche model on lead-free solder SAC305
Conference paper

Use Artificial Neural Network to fit stress-strain curve of Chaboche model on lead-free solder SAC305

C.E. Lee, Cadmus Yuan and K.N. Chiang
Proceedings of Technical Papers - International Microsystems, Packaging, Assembly, and Circuits Technology Conference, IMPACT, pp.137-142
2023

Abstract

Artificial Neural Network Chaboche Kinematic Hardening Model curve fitting Finite Element Method Machine Learning Wafer Level Chip Scale Packaging Hardware and Architecture Control and Systems Engineering Electrical and Electronic Engineering
People nowadays have an increasing demand for mobile devices, which causes electronic devices to become smaller and multifunctional. Therefore, the goal of electronic packaging is to minimize the size while maintaining or increasing its I/O number. To reach this goal and also follow Moore's law, the electronic packaging technique has gone through a long-term evolution, from traditional pin-through hole (PTH), to surface mount technology (SMT), then to area array type (AAT). Stats from the presence of chip scale packaging (CSP) and multi-stacked substrate, electronic packaging steps into the era of 3D IC. This research focuses on Wafer Level Chip Scale Packaging (WLCSP). Reliability testing is a crucial step in the development of electronic packaging before be introduced to the market. One popular test used for this purpose is the Accelerated Thermal Cycle Test (ATCT), which significantly reduces the time required to obtain reliability results. Additionally, finite element analysis is commonly used as a numerical method to construct simulation models. In this study, ANSYS is utilized to simulate the behavior of WLCSP under thermal cycle testing, enabling the estimation of solder ball's failure cycles through empirical formula. The mesh size for largest distance from neutral point (DNP) is fixed for more accurate simulation result. However, the accuracy of the material properties is an important factor that affects simulation result. The Chaboche Kinematic Hardening model is widely used to describe the plastic strain behavior of materials under cyclic loading. To acquire accurate Chaboche model coefficient, this research utilizes artificial intelligence to fit the curve from validated experimental equation. Artificial Neural Network (ANN) are chosen as the algorithm we use for curve fitting. By using different numbers of neuron and layer, different activation function and different optimizer, we can get curves that fit the experimental data. The optimized coefficients and the stress-strain curves we get are expected to improve the reliability of finite element analysis, providing accurate simulation for engineering applications.

Metrics

1 Record Views

Details

Logo image