Artificial intelligence is transforming the world, but it also comes with a growing energy problem.
Modern AI systems require enormous computing power, and the powerful chips that perform these calculations generate huge amounts of heat.
Keeping these chips cool has become one of the biggest challenges facing the technology industry.
Researchers at the Korea Advanced Institute of Science and Technology (KAIST) have now developed a new cooling technology that could significantly improve the energy efficiency of future AI systems.
Their new approach uses tiny channels embedded directly inside semiconductor chips and cools them using ordinary room-temperature water.
As AI chips become faster and more powerful, they produce increasingly large amounts of heat. Traditional cooling methods, such as fans and metal heat spreaders, are approaching their practical limits.
Cooling systems themselves also consume a great deal of electricity, especially in large data centers that house thousands of servers.
To address this problem, the KAIST team created an ultra-efficient liquid-cooling system that works from inside the chip.
The researchers built microscopic channels inside a silicon semiconductor chip. These channels are thinner than a human hair and allow water to flow directly through the chip, carrying heat away much more effectively than air cooling systems.
The team used a design known as a manifold microchannel structure. In many liquid-cooling systems, coolant enters from one side of the chip and must travel through long, narrow channels before exiting on the other side. This long journey creates resistance and requires powerful pumps to move the liquid.
The new design works differently. It uses multiple entry and exit points so that water only travels a short distance within each channel. The researchers compare it to a delivery network with many local distribution centers instead of a single warehouse serving everyone.
Because the water has shorter distances to travel, the system experiences much less resistance and requires far less pumping power. The coolant is also distributed more evenly across the chip, helping maintain a stable temperature throughout the device.
The researchers carefully optimized every aspect of the system, including the size, shape, number, and arrangement of the channels as well as the flow rate of the water. They combined simplified computer models with highly detailed simulations to identify the most effective design among countless possibilities.
After manufacturing the optimized design on a real silicon chip, the team tested its performance under extremely demanding conditions. The cooling system successfully kept chip temperatures below 100 degrees Celsius even when the chip generated exceptionally high levels of heat.
The most impressive result was its energy efficiency. The system achieved a performance level about ten times higher than the previous world-leading result reported in 2020. In practical terms, this means it needs only about one-tenth of the pumping power to remove the same amount of heat.
Remarkably, the system does not require expensive materials or complicated technologies. It uses ordinary room-temperature water and can be produced using manufacturing processes that are already compatible with today’s semiconductor industry.
The researchers believe the technology could help cool not only AI processors but also high-performance computers, advanced chip packaging systems, power electronics, and defense equipment.
As AI continues to expand, innovations like this could play an important role in reducing energy consumption and overcoming one of the biggest barriers to future computing performance.
