Researchers at the Korea Advanced Institute of Science and Technology (KAIST) have developed a technology that can reduce the cooling power required for artificial intelligence (AI) data centers to one-tenth of current levels.
On June 16, KAIST announced that a team led by Professor Kim Sung-jin from the Department of Mechanical Engineering and Professor Lee Ik-jin from the Department of AI has created a new liquid cooling structure that combines manifolds and microchannels within semiconductor chips. The manifold distributes and collects coolant through multiple pathways, while the microchannels are tiny water pathways thinner than a human hair.
Previous technologies faced challenges with coolant concentration in certain channels, but the researchers optimized the structure to ensure uniform coolant flow across all channels. This enhancement improves cooling performance while reducing energy loss.
Tests conducted using silicon wafers showed a coefficient of performance (COP) of 106,000, which is more than ten times higher than the previous record reported in the journal Nature in 2020. This indicates that the energy required to remove the same amount of heat has decreased to one-tenth of that needed by existing technologies.
Notably, this technology can be implemented using only room-temperature water, without the need for complex cooling methods involving nanostructures, diamond materials, or boiling liquids. It is also expected to be easily integrated into existing semiconductor manufacturing processes without additional investment in equipment, enhancing its commercialization potential.
The researchers believe this technology could be applied to large AI semiconductors, such as graphics processing units (GPUs) for AI data centers. NVIDIA is developing its Vera Rubin platform with power supply and cooling systems as key components, anticipating that cooling technology will become a critical competitive advantage in the commercialization of next-generation AI semiconductors. The researchers foresee the potential application of their technology in ultra-high-performance chips comparable to NVIDIA's upcoming AI platform, Vera Rubin.
Professor Kim Sung-jin stated, "In the AI era, not only semiconductor performance but also how effectively heat is managed will be a key competitive factor. We hope this technology will serve as a crucial tool in reducing power consumption and operational costs for AI data centers."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.