Thermal management of high heat loads using design optimized heat sinks

Abstract

The rapid evolution of high-speed GPUs, quantum computing, and AI drives a global trend toward increasing computational power, which in turn generates extreme local heat fluxes. Conventional single-phase cooling strategies are no longer sufficient to manage these thermal loads due to their inherent limitations; they require excessively high flow rates, large pumping systems, and complex secondary cooling infrastructure. Flow boiling in micro-channel evaporators presents a well-documented and robust alternative. Unlike single-phase flow, which relies on sensible heating, flow boiling exploits the latent heat of vaporization to dissipate heat almost isothermally. This mechanism allows for high heat removal rates at relatively low mass flow rates, making it a highly efficient solution for the thermal management of high-flux electronics. However, harnessing two-phase flow heat transfer requires more complex heat sink designs, which can be challenging to fabricate using conventional methods. This is where additive manufacturing plays a crucial role, enabling the fabrication of advanced micro-channel evaporator configurations, offering greater design flexibility and the ability to produce compactdesigns. As advanced micro-channel configurations become more sophisticated, conventional thermal management models struggle to capture the intricate heat transfer mechanisms involved. This necessitates the development of more complex thermal management models that can accurately predict performance and are suitable for optimization. To bridge the gap between complex additively manufactured geometries and predictive modeling, I present a novel, computationally efficient, and analytically differentiated modeling framework. This model accurately captures the conjugate heat transfer between the heat sink and the refrigerant flowing through embedded micro-channels. Specifically, I couple a one-dimensional homogeneous equilibrium model for the two-phase flow with a steady-state heat conduction model that uses singular line-sinks to represent the evaporators. I further establish the reliability of this proposed model through rigorous verification and validation studies. Leveraging the analytical differentiability of this model, I employ a coupled adjoint method to compute the sensitivity of the system’s performance to high-dimensional design inputs. This enables scalable gradient-based optimization, allowing me to discover complex, additively manufacturable micro-channel heat sink geometries. Ultimately, these optimized designs demonstrate the capability to dissipate extreme heat fluxes that far exceed the physical limits of conventional single-phase cooling. Finally transient two-phase flow in micro-channel evaporators exhibits deterministic chaos under slug, and churn flow regimes. Sudden exposure to high-heat fluxes, or extreme inlet sub-cooling act as a trigger to the onset of chaotic transient behavior. To address the optimization of such chaotic dynamical systems, I present a gradient-based optimization framework for systems governed by chaotic dynamics using discontinuous periodic orbits (DPO’s).