Real-Time Evaluation of Effective Thermal Conductivity Profile for the Redistribution Layer (RDL)

This paper presents a novel approach for rapid evaluation of effective thermal property distribution in Redistribution Layers (RDLs) used in advanced packaging. The methodology combines GDSii file processing, finite element analysis (FEM), and machine learning to overcome the limitations of existing methods. The process involves dividing the RDL into small patches, assembling them into a structured mesh, conducting FEM analysis using an optimized in-house code for data generation, and training a U-Net-based machine learning model to predict effective thermal property distributions. This approach significantly reduces computational time compared to traditional methods, enabling simultaneous evaluation of the entire RDL material property distribution. The method achieves a balance between accuracy and efficiency, reducing evaluation time from days to seconds while maintaining reasonable prediction accuracy. This innovation facilitates thermal characterization and design optimization in chiplet-based packaging, potentially accelerating the development of next-generation electronic devices through improved co-design practices.