Machine Learning-Assisted Optimization of Transfer Molding Parameters for Wire Sweep Reduction in PBGA and Advanced IC Packaging

Abstract

Wire sweep remains a critical reliability concern in plastic ball grid array (PBGA) and advanced wire-bonded integrated circuit packages because excessive bond-wire deformation during transfer molding can lead to short circuits, open circuits, and long-term electrical degradation. This study develops and evaluates a machine learning-assisted surrogate optimization framework for selecting transfer molding parameters that reduce wire sweep while controlling filling time and void formation. The framework integrates validated three-dimensional fluid–structure interaction (FSI) simulation using coupled FLUENT–ABAQUS–MpCCI analysis with central composite design (CCD), response surface modeling, and small-data surrogate modeling. Three process and design variables are considered: inlet pressure, wire diameter, and vent height. The output responses are maximum wire sweep, filling time, and void percentage. The surrogate model is evaluated using predicted-versus-actual response plots and interpreted through normalized ANOVA-based feature-importance analysis. The results indicate that wire diameter and inlet pressure are the dominant factors affecting maximum wire sweep, whereas inlet pressure and vent height strongly influence filling time and void formation. The optimized condition of 5.57 MPa inlet pressure, 0.05 mm wire diameter, and 0.36 mm vent height was confirmed by FSI simulation to provide low wire sweep, controlled filling time, and reduced void formation. Given the limited dataset of 20 CCD simulation runs, the ML component is treated as a local engineering surrogate rather than a large-scale predictive model.