Here is the abstract you requested from the Thermal_2018 technical program page. This is the original abstract submitted by the author. Any changes to the technical content of the final manuscript published by IMAPS or the presentation that is given during the event is done by the author, not IMAPS.
|Comprehensive numerical analysis of an impinging cold-plate using 6SigmaET|
|Keywords: Liquid Cooling, Cold-plate, CFD|
|With increasing computing power densities and rapid advancements in micro-fabrication techniques, previously tested and efficient conventional air cooling has pushed the operating temperatures of the electronic packages closer to, and sometimes over, its limit. One alternate solution that has been widely proposed is the placement of a cold-plate directly on the main processing chip. While direct use of the cold-plate can keep the operating temperature of high-power packages within the necessary limits, maximizing the efficiency of the cold- plate is an ongoing challenge. In this paper, a novel solution is proposed, where flow is separated through the cold-plate into two distinct paths resulting in lower pressure drop and thermal resistance, therefore increasing the efficiency of the cold- plate. The results are corroborated with lab experiments using 6SigmaET, a commercially available Computational Fluid Dynamics (CFD) modelling tool. The thermal performance of an impinging cold-plate can be dramatically increased by effectively separating the flow into two distinct branches. The result is a significantly lower pressure and thermal resistance, making the use of impinged cold-plates preferable when there is no strict space limitation; such as when flow entering the internal cold plate is perpendicular to the electronic board. The impinging cold-plate consists of an array of micro-channels, typically made of copper, connected by a longitudinal grove with a trapezoidal cross- section. The coolant enters the channels through a curved rectangular diffuser (90o deg. bend). The collector includes two ducts and a miniature reservoir for collecting the coolant from the micro-channels that are connected to an exit pipe. In the experimental setup, the geometry of the cold-plate needed to be measured. Accurate determination of the thickness of the fins and the width of the channel was critical in measuring the pressure-drop. This was achieved using microscopic images of the cross-section of the cold-plate submerged in an epoxy, which helped when building the cold-plate computationally to run the optimization. Deriving from the experimental setup, a complex impinging cold-plate was computationally modeled in 6SigmaET. To accurately capture the pressure- drop and hear transfer at this scale, 6SigmaET’s Multi-Level Unstructured Solver (MLUS) was employed. The mathematical formulation of this method employs a hierarchy of Cartesian grids and a face-to-cell connectivity graph to discretize the differential equations. The method uses a finite volume scheme with staggered variable arrangement and a pressure-based segregated iteration procedure to solve a discrete algebraic analogue of the Navier- Stokes equations. The method provides accuracy and robustness similar to the structured procedure and is more flexible in resolving arbitrary geometries and different solution scales. With the unstructured grid providing the necessary resolution of geometrical details, the model exhibits a substantial reduction in the number of computational cells. Results obtained from the simulation model compared well against the measured data with an acceptable error. As originally proposed, the impinging cold-plate with two distinct flow paths resulted in a lower pressure drop. There was a significant improvement in efficiency, resulting in pump-power savings. The integrity of the computational model means that further optimization studies can be conducted utilizing the state-of-the- art mathematical formulations of 6SigmaET.|
|Kourosh Nemati, Applications Engineer
San Jose, CA