Researchers at the Ningbo Institute of Industrial Technology in China have turned to diamond-copper coatings to improve the cooling efficiency of artificial intelligence (AI) data centers. The new material increased cooling efficiency by up to 80 percent, helping break the ‘thermal wall’ that tech companies face. 

The recent rise in AI technology has also increased demand for cooling technologies to dissipate heat generated by high-performance chips. The superior performance of the specialized AI chips that power applications like ChatGPT and Claude also generates significant heat as a byproduct. 

This needs to be removed quickly to ensure the data centers operate efficiently.  Tech companies call this a ‘thermal wall’ that must be breached if AI technology is to mature further.

China needs to import high-end heat-dissipation materials at high cost to keep up with global technologies and is looking for indigenous technologies that can offer it cost competitiveness and self-sufficiency as it builds its computing infrastructure. 

Diamond-copper composite

Researchers at the Ningbo Institute of Industrial Technology of the Chinese Academy of Sciences (CAS) developed a diamond-copper composite material and deployed it on an AI computing node in Zhengzhou, Henan province. 

When tested, the composite material delivered a thermal conductivity exceeding 1,000 watts per meter-Kelvin (W/mK). The materials chosen for the composite have high thermal conductivity of their own. Copper, widely used as a conductor, has a thermal conductivity of 400 W/m · K, while diamond has a thermal conductivity of 2,000 W/m · K. 

Using a composite material delivered a thermal conductivity between those of the two materials. However, it was good enough to increase chip performance by another 10 percent, the South China Morning Post said in its report. 

Scalable manufacturing process

The researchers have also developed a scalable process to help overcome manufacturing hurdles in producing the composite material. This proprietary approach overcomes challenges of surface treatment and dispersion associated with the material. 

To work at an industrial scale, the researchers have roped in Jiangxi Copper Company, a major copper producer in the country, which can deliver the composite in larger quantities.

The cooling technology is much in need of the hour for AI, as traditional heat-dissipation and power-supply architectures are rapidly approaching their physical limits. Data centers are already under fire for their high power consumption, largely met by burning fossil fuels. Now, their cooling demands are consuming tons of water every day.

The researchers also set up a laboratory for testing interfaces and cooling environments, which is open to partners such as chip makers, server manufacturers, integrators, and other operators. 

By bringing all these stakeholders together in one place, the researchers aim to address compatibility challenges and accelerate the development of industry standards for server-supporting infrastructure. 

While chipmaking is a beast China would want to tame someday, it is currently content to provide the support infrastructure for building large-scale intelligent computing clusters.

By focusing on developing an ecosystem of advanced computing support infrastructure, China’s participation remains critical as the new era of AI computing begins. 

Recommended Articles