Speaker
Description
Through-glass vias (TGVs) are critical for high-density semiconductor packaging; however, their internal waist geometry remains difficult to measure using non-destructive techniques. This study proposes an integrated 3D reconstruction framework combining NURBS-based parametric modeling, triangular mesh fitting, and Z-axis boundary constraints to ensure geometric accuracy. Experimentally acquired point cloud data are processed to generate high-fidelity 3D models, while a reinforcement learning (RL)-based digital twin simulation environment is developed to optimize inspection conditions and generate synthetic datasets. The proposed approach demonstrates high reconstruction accuracy and provides a robust solution for advanced metrology and inspection of TGV structures.