Thermal Management

Multi-Objective Optimization of Flow Channels for Uniform and Non-Uniform Heat Sources

For high-performance chips with escalating power densities and localized hot spots, liquid-cooled microchannel heat sinks face a critical trade-off between thermal performance and flow resistance. This paper presents a multi-objective topology optimization framework that minimizes average temperature and fluid dissipation while incorporating a high-temperature penalty to suppress extreme local overheating. The model is evaluated under uniform heating and a realistic non-uniform source (AMD EPYC 9965 MCM Chiplet CPU). Results show that the penalty term effectively drives channel evolution toward hot spots, reducing peak temperatures with modest pressure-drop increases. Under uniform heating, optimized channels achieve cooling comparable to serpentine designs but with flow resistance near that of parallel or pin-fin arrays. Under non-uniform heating, the topology adaptively narrows inlet-side channels to redirect coolant to distal CCDs, attaining a peak temperature of 139 °C and a pressure drop of 519 Pa—while conventional parallel and array channels exceed 200 °C due to distal cooling deficiency. This study confirms that topology optimization with localized hot-spot penalties effectively mitigates extreme temperatures, offering a viable path for engineering liquid cold plates in next-generation heterogeneous processors.