Solid-state micro-cooling could help data centers reduce hotspots in SSDs and other subsystems that liquid cooling and system fans don't fully reach.
Managing heat in AI infrastructure is getting more complicated as rising power density turns thermal design into a performance constraint. Liquid cooling is increasingly focused on high-power CPUs and GPUs, but not every heat-producing component in a server sits under a cold plate. SSDs, memory, retimers, and optical subsystems can still develop localized hotspots that affect throughput and reliability.
That is the gap xMEMS says its solid-state micro-cooling technology is designed to address. The company has begun mass production of its XMC-2400 and XMC-4800 chips, which use a silicon-based design to move air without rotating blades or bearings. xMEMS says the products mark the first large-scale commercialization of solid-state active cooling for edge AI systems.
“As AI infrastructure evolves, component-level cooling will become an increasingly important complement to traditional system-level thermal management,” Mike Housholder, vice president and general manager of the Thermal Management Business Unit at xMEMS, told Data Center Knowledge.
Why Liquid Cooling Leaves Some Hotspots Untouched #
Cold plates remove heat from the surface they touch directly. In liquid-cooled servers, that can leave some subsystems with uneven thermal coverage.
xMEMS sees enterprise SSDs as the clearest near-term example. Housholder said a cold plate attached to one side of an SSD can cool the controller and some NAND packages, while NAND on the opposite side may run hotter. That temperature gradient, he said, can contribute to thermal throttling and limit I/O performance during sustained workloads.
“We see enterprise SSDs as the strongest near-term use case,” Housholder said. “As AI workloads drive faster storage and more sustained data transfers, thermal throttling can limit SSD performance.”
The company argues that localized airflow inside the drive can reduce that imbalance. “Placing uCooling inside the SSD to cool the opposite-side NAND can reduce or remove this temperature gradient to maximize performance,” Housholder said. He added that similar thermal challenges are beginning to emerge in other dense AI infrastructure components.
How Solid-State Micro-Cooling Works #
Unlike a conventional fan, solid-state active cooling uses a silicon chip to generate directed airflow. xMEMS builds its uCooling chips on a piezoMEMS semiconductor platform and says the all-silicon design can deliver high-pressure airflow through narrow micro-channels without the mechanical wear associated with small rotating fans.
According to xMEMS, the XMC-2400 delivers up to 28 cubic centimeters of airflow per second at up to 1,300 Pa of back pressure and consumes about 150 milliwatts (mW). The XMC-4800, which the company is targeting for enterprise SSDs, delivers up to 48 cubic centimeters per second at up to 1,100 Pa and consumes about 240 mW.
“At approximately the size of a thumbnail, the XMC-2400 can be embedded into highly constrained environments to deliver highly targeted airflow to heat-generating components where conventional approaches are difficult to implement or scale,” Housholder said.
(Image: xMEMS)
Why Micro-Cooling Is a Complement to Existing Cooling Systems #
The company is not positioning the technology as an alternative to rack- or server-level cooling. Housholder said its airflow is only a fraction of what a conventional system fan can provide, making it better suited to localized hotspots inside tightly packaged subsystems than to whole-system thermal management.
“Solid-state active cooling is best viewed as a complementary technology that helps address localized hotspots within a broader cooling architecture, rather than a replacement for conventional cooling methods,” Housholder said. “Comparatively, uCooling’s airflow is a small fraction of that from a conventional system fan, but its ability to be embedded inside hard-to-reach subsystems brings value to the overall system thermal architecture.”
In practice, that means the technology is more likely to arrive as part of another component than as a field-installed data center retrofit. xMEMS says uCooling can be mounted on flex or rigid PCBs and pre-assembled inside subsystems such as SSDs and laser modules, where airflow is difficult to deliver with chassis fans or cold plates alone.
Where Solid-State Micro-Cooling Could Expand Next #
Longer term, xMEMS sees a path toward co-packaging thermal solutions that bring cooling closer to the processor package itself. For mainstream data center processors, however, Housholder said air-based micro-cooling may give way to liquid micro-channel designs because of the power density involved. He said air-based uCooling is likely to be more viable in edge AI systems, where thermal envelopes are lower and space constraints are tighter.
“Given the power density of data center processors, liquid (not air) micro-channels in the co-packaging architecture may be the next step,” Housholder said. “There are likely applications in co-packaging at the edge that are more viable for our air-based uCooling.”
The future of cooling is likely to incorporate a combination of technologies.
“Longer term, we expect future thermal architectures to combine liquid cooling, passive thermal solutions and localized active cooling technologies working together to optimize overall system performance,” Housholder said.