A Temperature-modified Semi-empirical Quantitative Physical Model for Predicting Interfacial Thermal Resistance for Chiplet Packaging

With the rise of large language models such as Deepseek, ChatGPT, and Grok-1, the demand for artificial intelligence acceleration chips based on chiplet-based advanced packaging has been increasing significantly, which also brings remarkable thermal management challenges. Herein, we conduct thermal modeling of the chiplet-based packaging with a 2.5D interposer using computational fluid dynamics (CFD). A multi-point contact model is used to simulate the real contact interfaces among thermal interface materials (TIMs), silicon chips, and metal vapor chambers. The heat dissipation mechanisms of silicone grease-based and carbon fiber-based TIMs with distinct surface/interfacial properties and viscoelastic behaviors at varying temperatures are systematically investigated. A temperature-modified semi-empirical quantitative physical model is proposed to predict the interfacial thermal resistance of TIMs. This study elaborates the mechanisms of TIMs in thermal management of advanced packaging under thermal physical fields, while establishing quantitative predictive models to advance interface science between soft/hard materials.