Corintis raises $24M to target the next AI bottleneck, and collaborates with microsoft for chip cooling breakthrough

The chips that datacenters use to run the latest AI breakthroughs generate much more heat than previous generations of silicon. Anybody whose phone or laptop has overheated knows that electronics don’t like to get hot. In the face of rising demand for AI and newer chip designs, the current cooling technology will put a ceiling on progress in just a few years.

To help address this problem, Microsoft has successfully tested a new cooling system that removed heat up to three times better than cold plates, an advanced cooling technology commonly used. It uses microfluidics, an approach that brings liquid coolant directly inside the silicon where the heat is. Tiny channels are etched directly on the back of the silicon chip, creating grooves that allow cooling liquid to flow directly onto the chip and more efficiently remove heat. The team also used AI to identify the unique heat signatures on a chip and direct the coolant with more precision. 

Researchers say microfluidics could boost efficiency and improve sustainability for next-generation AI chips. Most GPUs operating in datacenters are currently cooled with cold plates, which are separated from the heat source by several layers that limit the amount of heat they can remove.

As each new generation of AI chips becomes more powerful, they generate more heat. In as soon as five years, “if you’re still relying heavily on traditional cold plate technology, you’re stuck,” said Sashi Majety, senior technical program manager for Cloud Operations and Innovation at Microsoft.

Microsoft announced that it has successfully developed an in-chip microfluidic cooling system that can effectively cool a server running core services for a simulated Teams meeting.

“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,” said Judy Priest, corporate vice president and chief technical officer of Cloud Operations and Innovation at Microsoft.

“But we needed to prove the technology and the design worked, and then the very next thing I wanted to do was test reliability,” Priest said.

The company’s lab-scale tests showed microfluidics performed up to three times better than cold plates at removing heat, depending on workloads and configurations involved. Microfluidics also reduced the maximum temperature rise of the silicon inside a GPU by 65 percent, though this will vary by the type of chip. The team expects the advanced cooling technology would also improve power usage effectiveness, a key metric for measuring how energy efficient a datacenter is, and reduce operational costs.

Microfluidics is not a new concept, but getting it to work has been a challenge across the industry. “Systems thinking is crucial when developing a technology like microfluidics. You need to understand systems interactions across silicon, coolant, server and the datacenter to make the most of it,” said Husam Alissa, director of systems technology in Cloud Operations and Innovation at Microsoft.

Just getting the grooves right is hard. The microchannel dimensions are similar in size to human hair, meaning there’s margin for error. As part of the prototyping effort, Microsoft collaborated with Swiss startup Corintis to use AI to help optimize a bio-inspired design to cool chips’ hot spots more efficiently than straight up-and-down channels, which they also tested. The bio-design resembles the veins in a leaf or a butterfly wing – nature has proven adept at finding the most efficient routes to distribute what’s needed. 

Microfluidics requires more than innovative channel design. It is a complex engineering challenge.

It required ensuring that the channels are deep enough to circulate adequate cooling liquid without clogging while not being so deep as to weaken the silicon such that it risks breaking. The team produced four design iterations in the past year alone.

Microfluidics also required designing a leak-proof package for the chip, finding the best coolant formula, testing different etching methods and developing a step-by-step process for adding etching to manufacturing the chips.

The breakthrough is just one example of how Microsoft is investing and innovating in infrastructure to meet demand for AI services and capabilities. For example, the company plans to spend over $30 billion on capital expenditures in the current quarter.

Those investments include developing its own family of Cobalt and Maia chips designed specifically to run Microsoft and customer workloads more efficiently. Since Microsoft deployed its Cobalt 100 chip, for instance, Microsoft and its customers are benefiting from its energy-efficient compute power, scalability and performance.

Chips are just one piece of the puzzle, though, since the silicon works within a complex system of boards, racks and servers within a datacenter. Microsoft’s systems approach means fine tuning every part of this stack to work together and maximize performance and efficiency. An important part of that is developing next-generation cooling techniques like microfluidics.

As a next step, Microsoft continues to investigate how microfluidic cooling can be incorporated into future generations of its first-party chips. It will also continue to work with fabrication and silicon partners to bring microfluidics into production across its datacenters, the company said.

“Hardware is the foundation of our services,” said Jim Kleewein, technical fellow, Microsoft 365 Core Management. “We all have a vested interest in that foundation – how reliable it is, how cost effective, how fast, how consistent the behavior we can get from it, and how sustainable, to name just a few. Microfluidics improves each of those: cost, reliability, speed, consistency of behavior, sustainability.”