Triple cooling boost with Microsoft’s new microfluidic chip design

As processors designed for data centers heat up more, companies are finding increasingly innovative methods to cool them. Nvidia and its collaborators are reportedly testing new cold plates and immersion cooling techniques for upcoming AI GPUs, while Microsoft is suggesting etching microfluidic channels directly onto the backs of chips to lower peak temperatures by as much as 65%, making this approach three times more efficient than cold plates.

Rather than using a conventional cooler on their processors, Microsoft has embedded tiny channels directly into the silicon of its chips. This innovation enables coolant to flow through these canals and provide direct cooling to the chips. Removing heat closer to its origin enhances cooling efficiency significantly in comparison to traditional cold plates.

Although the effectiveness of this approach may differ among various chips, it has the potential to significantly enhance power usage efficiency, ultimately reducing operating expenses for data centers. The management of data center operations involves more than just supplying power to hardware; effective cooling solutions, which can be costly, are crucial as well.

Microsoft anticipates that thermal constraints could restrict the development of AI and other chips within "as soon as five years." The efficiency of cold plates might become a limiting factor, prompting Microsoft's investment in microfluidic cooling technology.

Thanks to its superior effectiveness, Microsoft is confident that microfluidics technology will facilitate the creation of more compact servers in the future. Additionally, this improved cooling capacity could extend the lifespan of CPUs, GPUs, and various accelerators. However, utilizing coolant directly on silicon could also introduce its own set of challenges.

Microsoft has employed AI algorithms to design the channels within its chips to optimize their cooling efficiency. It raises questions about how well this will function outside controlled environments. After all, coolant will eventually degrade, leading to concerns about its impact on the silicon over time.

At present, Microsoft aims to implement this technology primarily in data centers. There are no current intentions to introduce this technology to consumer-level CPUs. It’s unlikely any CPU or GPU producer would feel comfortable letting consumers use processors that require direct cooling methods on silicon.

As always, for the latest news on hardware launches and industry developments, be sure to follow our dedicated hardware coverage.