Low-Cost Thermal Solution for AI Processors

The competition asked teams to tackle a real problem plaguing the chip industry: as AI and datacenter processors push higher power densities, thermal interface materials (TIMs) and lid geometry increasingly become the bottleneck. I had to first get up to speed on techniques like chip stacking and how different TIMs behave under thermal and mechanical load before I could productively iterate on a solution.

I ran thermal and static deformation analysis in SolidWorks, optimizing lid geometry to reduce thermal warpage and limit the pump-out effect that degrades TIM performance over time. In parallel, I researched the CTE, thermal conductivity, and other properties of popular TIMs and built a comparison matrix to select the best material for the use case. To search the geometry space efficiently, I built a parametric model of the lid and ran a design of experiments study to automate testing across different lid geometries and converge on optimal measurements.

After submitting my results in early February, I was selected as a finalist and invited to present at the IEEE ECTC 76th conference in Orlando — a trip fully funded by ECTC. I distilled months of analysis into a 10-minute technical presentation for an audience of industry professionals, and won first place in the BS/MS student track among the finalist teams.