COOL: A Cooling-Aware Point Transformer Framework for Thermal Prediction in Advanced 3D/3.5D IC Packaging

2026-08-16

AI Summary

This study addresses the inefficiency of traditional solvers and the inability of existing models to capture cross-layer coupling and cooling dynamics in advanced packaging thermal prediction by proposing the Cool framework. This approach models heterogeneous components as attributed 3D point clouds and introduces a cooling-structure-aware Point Transformer, incorporating a physics-informed boundary condition loss to enforce interfacial thermal continuity. Experimental evaluations on a custom benchmark demonstrate that the framework achieves a normalized mean absolute error of 2.4%, significantly outperforming state-of-the-art learning-based methods. Furthermore, it accelerates computation by over 15.7 times compared to commercial finite element method solvers, thereby enabling efficient and accurate thermal prediction for advanced packaging applications.

Abstract

Advanced 3D and 3.5D IC packaging significantly improves integration density but elevates thermal management challenges due to cross-layer heat coupling and complex cooling structures. Traditional solvers deliver high fidelity but are too slow for iterative design flows, while existing learning-based methods either fail to capture inter-die thermal coupling or treat cooling structures as static components, limiting their applicability in real packaging co-design scenarios. In this work, we introduce COOL, a cooling-aware point transformer framework that represents heterogeneous assemblies (dies, interposers, TIMs, heat spreaders) as annotated 3D point clouds embedding geometric, material and power attributes. COOL explicitly encodes geometric boundaries and cooling structures, and introduces a physics-informed boundary condition (PI-BC) loss to enforce thermal consistency at material interfaces and cooling boundaries. Extensive experiments demonstrate that COOL achieves a remarkable 2.4\% NMAE on our constructed benchmark of multi-package thermal designs, substantially outperforming existing learning-based approaches while providing over 15.7x speedup compared to commercial FEM solvers.

Problem

• Thermal Prediction
• 3D/3.5D IC Packaging
• Cross-layer Heat Coupling
• Cooling Structures

Innovation

• Cooling-Aware Point Transformer
• Physics-Informed Boundary Condition Loss
• 3D Point Cloud Representation
• Thermal Prediction
• 3D/3.5D IC Packaging