Physics-informed AI-Driven Design and Fabrication of Hotspots-Targeted Microjet Cooling for High-Performance Computation Systems

Interdisciplinary Areas: Data and Engineering Applications, Autonomous and Connected Systems, Micro-, Nano-, and Quantum Engineering, Power, Energy, and the Environment

Project Description

The growing demand for powerful microprocessors with high data rates, interconnect density, and bandwidth has increased heat flux and power density. This, in turn, has created localized hotspots on the chips, with peak power densities reaching up to 1 kW/cm2. These hotspots generate uneven temperature gradients across the chip, impacting system performance and reliability. Consequently, there is a need for compact and energy-efficient thermal management cooling solutions that specifically target these hotspots, particularly when power dissipation patterns change dynamically.

 
This project aims to leverage physics-informed artificial intelligence (AI) to drive an innovative impingement jet cooling concept, combined with a fluid distribution manifold system featuring multiple inputs and outputs. The objective is to achieve efficient chip cooling, minimize temperature disparities, and reduce pressure losses within the cooling system. To achieve this, advanced design optimization algorithms will be utilized to provide detailed guidelines for achieving optimal impingement nozzle diameters, ensuring uniform temperature distribution for any power distribution scenario. Additionally, physics-informed AI will be employed to optimize the intricate internal geometry of the fluidic manifold delivery system. The project will also explore advanced AI-enabled manufacturing methods for performance characterization and benchmarking purposes.
 

Start Date

 
2023 (2 years)
 

Postdoc Qualifications

Ph.D. in, Mechanical Engineering, Mathematics, Computer Science, Physics, or relevant background in electronic cooling, thermal management, additive manufacturing and/or machine learning.
 

Co-Advisors

 
Tiwei Wei, wei427@purdue.edu, Assistant Professor of Mechanical Engineering,
https://alphalab-purdue.org/ and https://3d-tsv.weitiwei.me/

Guang Lin, guanglin@purdue.edu, Professor, Department of Mathematics (primary appointment), Professor, Department and Statistics, Professor, School of Mechanical Engineering, https://www.math.purdue.edu/~lin491/
 

Short Bibliography

 
[1] W. Deng, S. Liang, B. Hao, G. Lin, F. Liang. Interacting Contour Stochastic Gradient Langevin Dynamics. ICLR 2022
[2] W. Deng, X. Zhang, F. Liang, G. Lin. An Adaptive Empirical Bayesian Method for Sparse Deep Learning. NeurIPS 2019
[3] Wei, T. W.*, Oprins, H., Beyne, E., & Baelmans, M. (2019, May). First Demonstration of a Low Cost/Customizable Chip Level 3D Printed Microjet Hotspot-Targeted Cooler for High Power Applications. In 2019 IEEE 69th Electronic Components and Technology Conference (ECTC) (pp. 126-134). IEEE.
[4] Wei, T. W.*, Oprins, H., Cherman, V., Beyne, E., & Baelmans, M. (2019). Low-cost Energy Efficient On-chip Hotspot Targeted Microjet Cooling for High Power Electronics. IEEE Transactions on Components, Packaging and Manufacturing Technology. vol. 10, no. 4, pp. 577-589, April 2020.
[5] Wei, T. W.*, Oprins, H., Cherman, V., Yang, S., De Wolf, I., Beyne, E., & Baelmans, M. (2019). Experimental Characterization of a Chip-Level 3-D Printed Microjet Liquid Impingement Cooler for High-Performance Systems. IEEE Transactions on Components, Packaging and Manufacturing Technology, 9(9), 1815-1824.