cool findings
Houston researcher develops efficient method to cool AI data centers
A University of Houston professor has developed a new cooling method that can remove heat at least three times more effectively from AI data centers than current technologies.
Hadi Ghasemi, a distinguished professor of Mechanical & Aerospace Engineering at UH, published his findings in two articles in the International Journal of Heat and Mass Transfer. The findings solve a critical issue in the growing AI sector, according to UH.
High-powered AI data centers generate huge amounts of heat due to the GPU and operating systems they use with extreme power densities, which introduce complex thermal challenges. Traditionally, cooling methods, like microchannels, which use flow and spray cooling, have had limitations when exposed to extreme heat flux, according to UH.
Ghasemi’s research, however, found a more effective way to design thin-film evaporation structures to release heat from data centers and electronics at record performance.
Ghasem’s solution coupled topology optimization and AI modeling to determine the best shapes for thin film efficiency, ultimately landing on a branch-like structure—resembling a tree.
The model found that the “branches” needed to be about 50 percent solid and 50 percent empty space for optimum efficiency, and that they could sustain high heat fluxes with minimal thermal resistance.
“These structures could achieve high critical heat flux at much lower superheat compared to traditionally studied structures,” Ghasemi said in a news release. “The new structures can remove heat without having to get as hot as previous removal systems.
Ghasemi’s doctoral candidates, Amirmohammad Jahanbakhsh and Saber Badkoobeh Hezave, also worked on the project. The team believes their results show the impact of a physics-aware, AI design and can help ensure reliability, longevity and stability of AI data centers.
“Beyond achieving record performance, these new findings provide fundamental insight into the governing heat-transfer physics and establishes a rational pathway toward even higher thermal dissipation capacities,” Ghasemi added in the release