Why Cooling the Components that Cold Plates Can’t Reach Is Becoming the Next Infrastructure Challenge
By Mike Housholder, VP & GM, Thermal Management Business Unit, xMEMS Labs
Liquid cooling in compute and switch trays has become one of the most important innovations in scaling modern data center infrastructure. As processors continue to consume more power and system densities increase, direct liquid cooling has emerged as the most effective way to remove heat from high-performance computing systems. Cold plates and liquid loops are enabling levels of power density and sustained performance that would be difficult to achieve with conventional system fan cooling alone.
The industry’s investment in liquid cooling is both necessary and justified. However, a growing misconception has emerged alongside its adoption: that liquid cooling solves the entire thermal problem. In reality, it solves only part of it. Cold plates are exceptionally effective at cooling the surfaces they’re mounted to. But modern systems contain far more than CPU, GPU, HBM, SOCAMM memory, NICs and VRMs. SSD storage, optical and laser modules, and other critical electronics continue to generate heat while often residing outside the liquid-cooling loop.
As system power densities continue to rise, the next generation of thermal bottlenecks are increasingly found in the places liquid cooling cannot directly reach.
The Limits of Cold Plates
Liquid cooling works because it relies on direct thermal contact. Attach a cold plate to a processor, accelerator, or switch ASIC, and substantial amounts of heat can be removed efficiently. The limitation is equally straightforward. A cold plate requires both good physical contact to the heat generating component and maximum surface area to perform efficiently.
Modern computing systems contain far more than processors. SSD storage, pluggable optical transceivers and lasers, and numerous supporting electronics all generate heat and all contribute to overall system performance.
Many of these devices cannot easily be integrated into a liquid-cooling loop. They may require serviceability, occupy space-constrained locations, or simply lack a practical surface for cold-plate attachment, such as spring-attach pluggable with limited conduction paths.
As a result, an increasing number of critical components remain thermally underserved despite the overall system being considered “liquid cooled.”
The Rise of Hyperlocalized Thermal Challenges
This shift to liquid-cooled servers is creating a new category of problem: hyper-localized hot spots. Not every thermal challenge requires cooling an entire rack or server. Increasingly, the challenge is cooling a specific hotspot within a subsystem whether that be a storage controller or an external laser source. Conventional system fans were designed to cool systems broadly. They move large volumes of air through an enclosure and distribute cooling across many components simultaneously.
That approach remains valuable, but with the shift to liquid cooling it becomes increasingly challenging to accommodate fan arrays to deliver much needed airflow to hard-to-reach components. Modern AI data center systems require a more precise approach.
When One Side Is Cool and the Other Isn’t
Storage devices provide one of the clearest examples of the thermal gap. As the industry prepares for Gen7 NVMe SSDs, liquid cold plates on the controller side of the drive will likely be required to achieve sustained read/write performance.
But it often leaves the NAND devices on the opposite side operating at significantly higher temperatures than the NAND on the controller side. The result is a temperature gradient across the NAND that throttles I/O throughput based on the hotter-side memory.
As storage bandwidth continues to increase, thermal imbalances across a drive can affect sustained read and write performance, consistency, and long-term reliability. The goal should not be cooling one side of the SSD but to cool the entire SSD for maximum read/write performance.
Why Liquid Cooling Needs a Partner
Historically, thermal solutions have often been viewed as competing alternatives: liquid cooling versus air cooling, active cooling versus passive cooling. The future is neither but a combination of both.
Liquid cooling excels at removing large heat loads from concentrated sources. What it does not provide is localized airflow precisely where it is needed. This is where hyper-localized onboard cooling becomes increasingly valuable.
In an SSD, for example, a cold plate may efficiently cool the controller side of the drive while localized active airflow cools the non-cold-plated side containing NAND devices. One solution cools through conduction. The other extends cooling through active convection. Together, they create a more thermally balanced system.
At xMEMS Labs, we’ve focused on addressing exactly this challenge. Our solid-state µCooling® technology delivers hyper-localized active convection in an ultra-compact silicon form factor, enabling targeted airflow precisely where traditional cooling approaches struggle to reach: buried components or inside a sub-system enclosure. Designed to complement liquid-cooled architectures, µCooling helps reduce hotspots and temperature gradients in SSDs, optical modules, and other critical infrastructure components that operate outside the cold-plate footprint. As compute, storage, and interconnect densities continue to increase, these previously overlooked thermal gaps are becoming increasingly important to overall system performance.
Closing the Thermal Gap
As system densities continue to increase, overall performance will depend less on cooling the hottest device and more on eliminating hotspots and temperature gradients throughout the system. Liquid cooling remains an essential foundation for the future of computing infrastructure.
But it is not the complete answer. A growing number of performance-limiting components will remain beyond the reach of cold plates. Closing that thermal gap requires localized, onboard cooling capable of reaching the areas liquid cooling cannot.
The next phase of thermal innovation may be µCooling, complementing liquid cooling by addressing hotspots that cold plates can’t reach. Because ultimately, system performance is determined not only by the component you cool best, but also by the components you don’t cool well enough. And increasingly, that’s where the next bottleneck is found.
Want to see what closing the thermal gap looks like in practice? Visit xMEMS at CES for a private demonstration of µCooling technology and a discussion on how active localized cooling can enhance next-generation AI and high-performance computing systems. Request a meeting today.
Want to see more content like this? Select xmems.com as a preferred source.