Your data center is under water and hotter than ever

Anyone who’s jumped in a cold pool on a hot day understands the effectiveness of liquid cooling. This is certainly not a new concept.  Automobile radiators, one of the first examples of liquid cooling in industry, have existed for nearly 125 years.  Using liquid cooling – among other temperature-reduction methods – is not a novel idea for the data center either, where uncontrolled heat from servers, high-powered processors and a myriad of electronics gear could cause serious issues.  While data center liquid cooling applications have been relatively limited to specific high-heat systems, compute-intensive and data-intensive workloads of new server configurations continue to drive temperatures higher, placing more interest on more broadly-applied liquid thermal management approaches. 

The physical design of the data center prioritizes heat reduction because, when it comes to electronic systems, high temperature is a performance-killer and a potential system-destroyer, too.  While the absolute environmental temperature limit of the data center is 82° F, according to some industry recommendations, the ideal temperature range should fall somewhere between 73° - 75° F. [1] As you can imagine, that makes for a big air conditioning bill, not to mention a significant drain on the power grid.  Ensuring server systems (and other electronics) are controlled as much as possible for temperature is, therefore, critical not only for the uninterrupted storage and processing of data, but for more energy-efficient, lower-cost operations as well. 

In addition to active air cooling (air conditioning, fans, etc.) and optimized structure designs such as raised floors that help maximize air flow opportunities, cooling electronic systems – particularly in server racks – from the inside out is equally vital.  The use of thermal interface materials (TIMs), heat sinks and liquid cooling systems within high-density server electronics are the primary methods for achieving in-application temperature reduction.  For all of these approaches, massive innovation is underway as data volumes and speeds driven by AI, data mining, and analytics are only getting more intense – and hotter.

Liquid cooling in the data center takes many forms – from cooling pipes attached to cooling plates or chassis between PCBs/modules to full immersion cooling systems that submerge entire racks.  The idea is relatively straightforward:  a liquid coolant (water or dielectric coolant) circulates through pipes or other structures cooling the metal interface that acts as a heat collection device for the high-performance computing chips.  In the case of immersion cooling, components are completely submerged [2] in tanks, where dielectric coolant that does not harm the components circulates to reduce system operational heat.   While cooling plates for server boards/racks and immersion cooling have been used only in select areas of the data center, this temperature control method is projected to see a 20% CAGR from 2022 to 2028 [3] as data volume and intensity continue to rise. 

Data center operators are driven not only to manage performance optimization through heat reduction, but also to ensure more sustainable, energy-efficient data factories. There is just so much air cooling can achieve when power densities are high.  Liquid cooling amplifies the temperature reduction equation and is doing so with recirculated (i.e. waste-reducing) water or coolants that are highly effective.  This lowers the drain on power significantly. The question is:  Can the positive effects of liquid cooling be made even more substantial? 

In the past, attempts have been made to further improve the heat dissipation of liquid cooling systems by using TIMs between the component and the metal liquid cooling pipes/plates/chassis.  This accelerates heat dissipation between what would otherwise be metal-to-metal contacts.  This idea has merit.  Unfortunately, the thermal interface materials available – in the form of pads, adhesives, gels and liquids – are either not suitable (gels and liquids) for the application or cannot withstand the friction (pads and adhesives) often induced by pluggable components into the housing and/or the PCB being inserted into partitioned, liquid-cooled plate structures. The materials are pushed or scraped off and essentially become ineffective. 

However, recent TIM innovation is showing potential as a liquid cooling enhancement.  Proven to deliver noteworthy heat reduction for transceiver pluggable optical modules (POMs), a durable micro-thermal interface material (mTIM) applied in an ultra-thin layer accelerates heat dissipation by providing a thermally conductive interface, which is superior to metal-to-metal heat transfer. Currently employed for OSFP 400 GbE POMs in data centers, this solution has demonstrated significant temperature reductions versus a metal-to-metal interface. The per POM heat reduction’s collective effect (one line card can contain up to 32 POMs) is even more substantial. While an investigation into mTIM’s performance with both pipe and immersion liquid cooling designs is in the early stages, the properties and attributes of the durable coating suggest that it may offer considerable cooling acceleration for liquid cooling structures.  The material is compatible with multiple metals, is ultra-thin at 25 µm (+/- 5 µm) and highly durable.  

As data center server racks get hotter and even the most effective liquid cooling solutions are pushed to their limits, thermal control innovation – like mTIM – may be a consideration for enabling an even more sustainable, high-performance operation.