A Fast and Accurate Temperature Prediction Method for Microfluidic Cooling with Multiple Distributed Hotspots
Yudan Pi,Jing Chen,Min Miao,Yufeng Jin,Wei Wang
DOI: https://doi.org/10.1016/j.ijheatmasstransfer.2018.07.127
IF: 5.2
2018-01-01
International Journal of Heat and Mass Transfer
Abstract:As the feature size decreases and integration degree increases, integrated circuits face a serious thermal challenge. Microfluidic cooling (i.e., forced convection of coolant inside micrometer-sized channels) has been regarded as a promising solution for the growing challenge of heat dissipation inside electronic devices. However, to the best of authors' knowledge, the cooling performance of fluids inside microchannels is usually formulated under a uniform heat source, which is not the case in real applications where hotspots are distributed. The effect of the size of multiple distributed hotspots on microfluidic cooling has not previously been investigated. By implementing micro/nanofabrication technologies, hotspots with feature sizes of 50 mu m, 100 mu m, and 200 mu m, were designed and fabricated on an 80-mu m-thick Si substrate to mimic multiple distributed hotspots. Parallel microchannels, which were 60 mu m deep, 50 mu m wide, and 3440 mu m long, were prepared for microfluidic cooling. In each single-hotspot heating experiment, a fitting parameter of the normalized heat transfer coefficient (h) was extracted from the power input and the local temperature excess based on the thermal spreading resistance model. The results indicated that h, i.e. the cooling performance on the hotspot, decreased with hotspot size. Therefore, in cases with varied hotspot sizes on a single chip, separate thermal analysis for each hotspot may be required to correctly predict microfluidic cooling performance. A simplified model based on the superposition principle was proposed to predict the temperature excess of multiple distributed hotspots with microfluidic cooling. The method was experimentally verified to attain a deviation less than 15%. Computation time was noticeably reduced from more than four hours (for a full-scale numerical simulation) to several minutes after implementing the simplified model. (C) 2018 Elsevier Ltd. All rights reserved.