Machine Learning-Assisted Numerical Investigation on Multiphysics Coupling for Copper Electroplating in High-Aspect-Ratio Through-Glass Vias

Abstract

Through-glass vias (TGVs) are key structures for high-density vertical interconnection in three-dimensional advanced packaging. The quality of copper electroplating filling of TGVs directly affects the electrical performance, thermal stability, and service reliability of interconnect structures. Conventional finite element simulation requires separate modeling and transient solution for different geometries and process parameters, which leads to high computational cost and limits rapid screening over a wide process space. To address this issue, this work establishes a multiphysics finite element model for copper electroplating filling in high-aspect-ratio TGVs based on the Langmuir adsorption kinetics model. Representative electroplating filling behaviors, including defect modes such as premature sealing, seam formation, and void formation, as well as butterfly-shaped deposition morphology, are reproduced by adjusting additive concentrations. On this basis, a machine learning surrogate model is constructed to learn the nonlinear mapping among TGV geometry, electroplating process parameters, and filling displacement curves, enabling rapid prediction of electroplating filling curves under different process conditions. The results show that the artificial neural network (ANN) model can effectively learn the influence of TGV geometry and electroplating process parameters on filling morphology evolution. The curve-level coefficient of determination on the test set reaches 0.9206, and the predicted curves agree well with the finite element results. The proposed FEM-machine learning framework improves the efficiency of multi-parameter process screening while retaining physical consistency, providing a data-driven tool for process optimization of copper electroplating filling in high-aspect-ratio TGVs.