Microsoft is experimenting with a totally different way to cool AI chips by using microfluidics to deliver liquid directly inside of the silicon. Check it out:
Unlike using regular cold plates that are positioned on top of the chip and are blocked by packaging layers, Microsoft's new approach cuts tiny channels into the silicon itself, where coolant flows through the grooves and drags heat away from the source.
Microsoft says that in its tests, the use of microfluidics removed heat up to 3x more effectively than cold plates, and that it also sliced the maximum GPU temperature rise by 65%, depending on the workload and the chip type. AI models were also used to map hotspots on the chip, and to direct the coolant with increased precision.
The company adds that it would make new servers smaller, reduce datacenter power costs, and extend the lifespan of GPU and AI accelerators. As next-gen semiconductors continue to get hotter and hotter with each generation, Microsoft warns that cold plates might not be sufficient in the next 5 years, which is why there's work being done now with partners to refine this new technology, from packaging to coolant chemistry to full server integration.

Judy Priest, corporate vice president and chief technical officer of Cloud Operations and Innovation at Microsoft, explains: "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. But we needed to prove the technology and the design worked, and then the very next thing I wanted to do was test reliability".




