Combined local microchannel-scale CFD modeling and global chip scale network modeling for electronics cooling design

Abstract

Microchannel cold plates enjoy increasing interest in liquid cooling of high-performance computing systems. Fast and reliable design tools are required to comply with the fluid mechanics and thermal specifications of such complex devices. In this paper, a methodology accounting for the local as well as the device length scales of the involved physics is introduced and applied to determine the performance of a microchannel cooler. A unit cell of the heat transfer microchannel system is modeled and implemented in conjugate CFD simulations. The fluidic and thermal characteristics of three different cold plate mesh designs are evaluated. Periodic boundary conditions and an iteration procedure are used to reach developed flow and thermal conditions. Subsequently, two network-like models are introduced to predict the overall pressure drop and thermal resistance of the device based on the results of the unit cell evaluations. Finally, the performance figures from the model predictions are compared to experimental data. We illustrate the cooling potential for different channel mesh porosities and compare it to the required pumping power. The agreement between simulations and experiments is within 2%. It was found that for a typical flow rate of 250 ml/min, the thermal resistance of the finest microchannel network examined is reduced by 7% and the heat transfer coefficient is increased by 25% compared to the coarsest channel network. On the other hand, an increase in pressure drop by 100% in the case of densest channel network was found. © 2009 Elsevier Ltd. All rights reserved.