A Numerical Study of Multiphase Dielectric Fluid Immersion Cooling of Multichip Modules Including Effects of Nucleation Site Density and Bubble Departure Diameter Functions

Abstract

In this study, the computational fluid dynamics programs ANSYS Icepak along with ANSYS Fluent were used to model and simulate the multiphase dielectric fluid immersion cooling of a multichip model on a printed circuit board using a variety of functions for nucleation site density and bubble departure diameter. Numerical results were validated against experimental data for the same geometry and working fluid. The numerical model in this study is a reduced form of an experimental setup, which consisted of four equally spaced, 24 mm by 24 mm die, arranged in a square pattern and centered on a printed circuit board. The numerical model consisted of the die and printed circuit board vertically suspended in a large, quiescent 15°C subcooled pool of Novec 649. Heat was uniformly dissipated from the die at fluxes ranging from 3 to 12 W/cm2. Pool boiling curves were generated from the numerical results and compared with experimental data. This study involved the comparison of several functions for two boiling parameters: nucleation site density and bubble departure diameter. Four combinations of the standard functions available in Fluent were simulated. Numerical results for the worst combination over predicted the wall superheat by an average of 35°C for heat fluxes ranging from 3 to 12 W/cm2. In addition, the vapor fraction was under predicted by an order of magnitude and the increasing slope of heat flux versus wall temperature seen in the nucleate boiling regime of typical boiling curves was not consistently reproduced. Parametric studies were performed in which six bubble departure diameter functions and twelve nucleation site density functions were simulated with the remaining boiling parameters set to default Fluent functions. Next, combinations of nucleation site density and user-defined bubble departure diameter functions were simulated together. The worst of these over predicted the wall superheat by an average of 3.4°C for heat fluxes in the range of 3 to 12 W/cm2. The best agreement was seen when experimental data for bubble departure diameter were implemented as a user-defined function along with user-defined nucleation site density; these simulations matched experimentally measured wall superheat within 1.3°C in the same heat flux range.