Computational Modeling of Extreme Heat Flux Microcooler for GaN-Based HEMT

Abstract

This study explores an extreme heat flux limit of microcooler for GaN-based HEMTs (high electron mobile transistors) which have local power densities exceeding 30 kW/cm2 using both solid conduction simulation and single-phase/two-phase conjugate simulations. Solid conduction simulation models are developed for full geometry of the microcooler to account for the overall thermal resistances from GaN HEMT to working fluid. This allows investigating the temperature distribution of the suggested microcooler. Parametric studies are also performed to investigate the impact of geometries and heat transfer coefficients on the junction temperature. The solid conduction simulation results using COMSOL Multiphysics agree well with single-cell ANSYS Fluent simulation results. Separately, fluid-solid conjugate CFD (Computational Fluid Dynamics) simulation models provide the detailed flow information in the microchannel using a single-channel geometry with symmetry boundary conditions. Single-phase CFD simulations obtain the lower bound of total pressure drop and heat transfer coefficient at the microchannel walls for a mass velocity range of G = 6000–24000 kg/m2-s. The local temperatures and velocity distributions are reported that can help with identifying the locations of the maximum velocity and recirculation regions that are susceptible to dryouts. Two additional alternative tapered inlet designs are proposed to alleviate the significant pressure loss at the entrance of the SiC channel. The impact of the tapered inlet designs on pressure drops and heat transfer coefficients is also investigated. Two-phase simulations in microchannel are conducted using Volume-of-Fluid (VOF) method embedded in ANSYS Fluent to investigate two-phase flow patterns, flow boiling, and temperature distributions within the GaN HEMT device and SiC etched mircochannels. A user-defined function (UDF) accounts for the phase change process due to boiling at the microchannel walls. The results show that the time relaxation factor, ri has a strongly influence on both numerical convergence and flow solutions.