University of Houston engineers say they have cooked up microscopic, tree-shaped cooling structures that can pull heat off semiconductors up to three times faster than conventional designs. The result is still at the lab stage, but if it scales, future AI data centers could run cooler while easing some of the heavy energy and water demands that come with today’s liquid-cooled racks.
Using a physics-aware machine-learning framework paired with topology optimization, the team identified branched, porous thin-film geometries that the authors report remove heat at least three times more effectively than benchmark pillar and channel layouts, as published in the International Journal of Heat and Mass Transfer. The paper reports that those hierarchical, tree-like patterns reach much higher critical heat fluxes at lower surface temperatures than many previously studied structures.
How the tree-inspired cooler works
Ghasemi’s group says the core trick is extreme thin-film evaporation. A whisper-thin layer of coolant spreads across the branched microstructure and evaporates, carrying heat away while tiny capillary channels continuously refill the film. According to the University of Houston, the optimized geometries are roughly half solid and half void and are compact enough that dozens can fit on a single semiconductor wafer. The work, led by Hadi Ghasemi with doctoral candidates Amirmohammad Jahanbakhsh and Saber Badkoobeh Hezaveh, began around 2020 and was validated with lab tests and AI-driven modeling before being submitted to the journal.
What it means for Houston and the power grid
Local and state energy officials say the timing matters. Rising AI demand is expected to push electrification and data-center loads higher in the coming decade, and those shifts could squeeze consumers unless operators lock in meaningful efficiency gains, the Houston Chronicle reports. Community groups and watchdogs are already tracking where new, thirsty facilities are being planned, and a crowdsourced Texas data center watchdog map highlights water and electricity concerns that make efficiency innovations especially relevant for the region.
"A $1 million grant from the U.S. Office of Naval Research is helping fund the project," the Houston Chronicle reported. Even with that backing, the authors and engineers caution that these are controlled laboratory results. The published experiments show very high critical heat flux under specific conditions, but scaling those porous, branched architectures into manufacturable, reliable hardware for server racks will require further work and industry-level testing, as the journal paper explains.
If those engineering and supply-chain hurdles can be cleared, the design could lower the energy and water footprint of direct-to-chip cooling systems now being adopted in AI-first builds. University researchers say the findings point to a practical path forward, but they emphasize that system-level integration, not just component performance, will ultimately determine whether data centers actually realize the projected savings.
