As data centers ramp up to meet the demands of AI workloads, developing innovative ways to keep systems cool is critical.

When it comes to cooling a chip, there are two sources of heat that must be managed, and available solutions may focus on cooling down the system or removing heat from the system.

One source of heat is the processor itself, which is the compute engine, Athar Zaidi, GM of power ICs and connectivity systems for Infineon Technologies’ power and sensor systems division, told EE Times in an interview. A large amount of heat is also generated through the power management solution around the processor, which are not 100% efficient. “Every time you do a voltage or current conversion, you lose efficiency, and that efficiency is manifested as heat,” he said.

Power management and cooling are intertwined, Zaidi added. “To take the heat out, you have to come up with the cooling solution, which requires energy too, so it’s a double-edged sword,” he said. “You’re wasting power in compute and in powering that compute. Then you need to consume power to take that heat out from the system.”

Because there is a strong relationship between cooling and power management, cooling is critical to managing a data center’s energy footprint, which can impact where it gets built as it requires permits that outline the facility’s allowable energy budget, Zaidi said. What a data center owner does within the allowable power envelope is up to them, but that envelope defines its total cost of ownership (TCO) and return on investment.

In February, Infineon launched its TDM2254xD series dual-phase power modules that address both electrical and thermal performance to help data centers operate at higher efficiency to meet the high-power demands of AI GPU platforms—while also significantly reducing TCO.

As computing gets more dense, cooling and thermal management solutions must get smaller, which is why Infineon’s TDM2254xD series dual-phase power modules have a footprint that are roughly 10 mm squared, and no higher than 8 mm. “When you are talking about power, it’s not only about efficiency—it’s about thermal management,” Zaidi said.

When it comes to AI workloads, power is increasingly becoming the bottleneck, he added. “Compute was the bottleneck because everything was under one kilowatt, and that power can be handled on a board.”

But with platforms like Nvidia’s Grace Blackwell, consumption is hitting 3.3 kilowatt and air cooling is not enough, Zaidi said. “It has to be all liquid cooled. That defines the topology on the power management.”

There is, however, more than one way to cool a chip.

Lenovo recently announced its sixth generation Neptune liquid cooing for use across its ThinkSystem V3 and V4 portfolios. The company introduced its proprietary cooling technology more than a decade ago, which uses liquid to remove heat while enabling customers to realize up to a 40% reduction in power consumption compared to traditional air-cooled systems.

Neptune is a direct water-cooling solution that recycles loops of warm water to cool data center systems and keep all the server components cool. The latest iteration includes a new GPU cold plate design, as well as an enhanced cold plate cooling fan and flow design.

ZutaCore, meanwhile, has opted for a waterless, electric liquid approach for its HyperCool technology, rather than using a single-phase liquid, direct-to-chip method.

In an interview with EE Times, Shahar Belkin, ZutaCore executive VP of products, said using liquid can be tricky. “You can never trust it to stay where you want it.” In a single gaming PC, it can be controlled, but in a data center with a rack full of hundreds of CPUs and thousands of connectors and tubing, it is statistically impossible, he said, and a single phase leak is extremely damaging to IT systems.

ZutaCore employs electric liquid because it does not have any power conduction running through it, Belkin said. It is also a two-phase boiling and condensation process because it brings liquid to the cold plate, which turns it into vapor—this is how large amounts of heat are removed off the processors and away from servers. Because no water is used in the cooling system, equipment is protected from corrosion and other water-related threats, he said.

Belkin added that ZutaCore’s patented approach is technically not cooling and takes advantage of physics. “We’re not actually cooling; we are just moving the heat away from the CPU. You can reuse that heat somewhere else.” In addition, the liquid is not proprietary—ZutaCore can buy liquids from a variety of vendors.

The liquids are part of the reason the solution is sustainable, Belkin added, as they are non-flammable, non-toxic and environmentally friendly. “These are liquids that are very low on global warming potential.”

ZutaCore is not the only company that thinks it has a unique approach to cooling the data center. Phononic is using solid-state technology to cool semiconductors, and it can be used for many other different use cases—not just computing environments.

In an interview with EE Times, Phononic CEO Tony Atti said the next frontier of sustainability is emissions associated with refrigerants, which is why he sees so much potential for the company’s solid-state technology across so many industries. “We have no refrigerants whatsoever.”

Phononic’s solid-state thermoelectric device—a thermoelectric cooler (TEC)—exploits proven thermoelectric principles and what is known as the Peltier Effect, Atti said. When an electric current is passed through a TEC, it transfers heat from one side to the other. Each TEC is comprised of “legs” made of semiconductor materials, which move heat using carriers.

The legs are arranged in pairs, consisting of an n-type semiconductor leg (in which the carriers are electrons) and a p-type semiconductor leg (in which the carriers are holes). The legs are electrically connected in series in such a way that when current is applied to the TEC, the carriers all move in the same direction, transferring the heat from one side of the device to the other.

“At the transceiver level, it’s very much a component solution,” Atti said. “We custom design based upon the specs provided to us by our partners, but once they understand what we can do, then the iterative design process starts.”

Phononic’s active, responsive approach to heat transfer can be used for more than just chips, Atti added. The company’s solid-state technology can be used to heat, cool, and even freeze small and large spaces for broad range of applications because it scales well, he said, while also being climate friendly as it makes a real impact on greenhouse gas emissions.

By: DocMemory
Copyright © 2023 CST, Inc. All Rights Reserved