Microsoft's AI-designed microfluidic chip cooling

Microsoft is developing microfluidic cooling that flows liquid coolant directly through channels etched into a chip’s silicon, a technique it says removed heat up to three times better than the cold plates used in data centers. In lab tests the company reported a 65% reduction in the maximum temperature rise of a GPU’s silicon, and it built a microfluidic cooling system for a server running core services for a simulated Microsoft Teams meeting. The channel patterns — about the width of a human hair — were designed using AI and inspired by natural vein networks to direct coolant where it chills most efficiently.

Bringing coolant straight to the silicon removes insulating layers used with cold plates, so the liquid doesn’t need to be chilled as much, which could lower energy used for cooling. Microsoft says microfluidics could enable more overclocking, denser server packing, fewer servers for the same capacity, and potentially 3D chip architectures by allowing fluid to flow through stacked silicon. But the company has no timeline: real-world validation, hardware and supply-chain changes (for example, when grooves would be etched in manufacturing), and durability concerns remain. Microsoft notes the coolant can be the same water/propylene-glycol mix cold plates use. Other efforts are underway too — HP received $3.25M from the DOE for related work — and Microsoft warns efficiency gains can produce rebound effects (the Jevons paradox), a point even CEO Satya Nadella has commented on.

Source