Microsoft unveiled a new method of cooling chips it claims can significantly boost both efficiency and energy consumption.

Using microfluidics, where liquid coolant is brought directly inside the silicon, Microsoft engineers developed a system capable of effectively cooling a server from inside the chip itself.

The technique sees small channels the size of human hair etched directly on the back of a silicon chip, creating grooves that allow cooling liquid to flow directly onto the chip and remove heat.

“Microfluidics would allow for more power-dense designs that will enable more features that customers care about and give better performance in a smaller amount of space,” Judy Priest, corporate VP and CTO of cloud operations and innovation at Microsoft, explained.

Lab tests saw microfluidic chips perform up to three-times better at removing heat compared to cold plate cooling currently employed in Microsoft data centers. The approach also reduced the maximum temperature rise of the silicon inside a GPU by 65%.

Microsoft engineers enlisted the help of AI to further improve their in-chip cooling approach. Teaming up with Swiss startup Corintis, the scientists used AI to optimize cooling channel designs, resulting in a pattern that mimicked the veins of a leaf or butterfly wing.

Microsoft engineers suggest that microfluidics could allow data centers to increase the density of servers – further increasing compute capacity while significantly reducing latency.

“We want microfluidics to become something everybody does, not just something we do,” Jim Kleewein, a technical fellow at Microsoft 365 Core Management, noted. “The more people that adopt it, the better, the faster the technology is going to develop, the better it’s going to be for us, for our customers, for everybody.”

While innovative, racks would still require evaporative cooling and heat exchangers, as they wouldn’t eliminate the need for downstream systems to dissipate the extracted heat.

Microfluidic history

The concept of microfluidic chip cooling is not new, with Stanford researchers David Tuckerman and R F Pease proposing the idea of using “microchannels” etched into a heatsink back in 1981. And in the early 2000s, fellow Stanford researchers would launch Cooligy, a startup attempting to develop microchannels on a heatsink built directly onto a chip,

Microsoft has been exploring microfluidics for some time now, teasing such efforts in July 2024 upon unveiling its initiative to create direct-to-chip liquid-cooled server solutions to decrease water consumption.

Microsoft's microfluidics efforts mark its second breakthrough in the past month aimed at taking data center workloads to the next level.

The hyperscaler's engineers developed an innovative interconnect system called Mosaic that utilizes directly modulated microLEDs, commonly found in screen displays, to maintain high network speeds while consuming significantly less power than traditional lasers.