摘要
Driven by the continuous advancement of 3D packaging technologies, heterogeneous integration has led to increasingly non-uniform power distributions, positioning thermal management as a critical bottleneck that restricts both system performance and long-term reliability. As High- Performance Computing (HPC) demands escalation, chiplet architectures have emerged as the mainstream solution due to their high integration density and modularity; however, these designs introduce unprecedented thermal complexity. This study proposes a comprehensive thermal model validated through 3D Finite Element Analysis (FEA), specifically focusing on the characterization of temperature profiles and heat flux distribution within multi-chiplet configurations.To overcome the computational burdens associated with the complex internal structures and the massive element counts inherent in chiplet models, this research adopts a streamlined packaging modeling methodology based on equivalent material theory. By substituting complex multi-layer structures and high-density interconnects with homogenized equivalent materials, the approach successfully balances heat conduction accuracy with simulation throughput. The feasibility of this equivalent μbump modeling is first rigorously verified using a standard FCBGA platform before being scaled to a chiplet model. This methodology is intended to serve as a robust, time-efficient predictive tool during the early stages of package design, providing a solid theoretical foundation for optimizing thermal management and enhancing the structural reliability of heterogeneously integrated systems.