Diamonds, lasers and oil may not be the first things you think of when considering ways to keep chips and computers cool. But as modern chip designs pack and stack more transistors into smaller spaces, heat has emerged as a serious problem.
The semiconductor industry is risking everything to solve this. What Stick can enable not only the scaling of AI data centers, but also many applications in consumer electronics, communications, and military equipment.
As senior editor Samuel K. Moore explained to me between cold tongue sandwiches at 2nd Avenue Deli. ieee spectrumIn the office, better thermal management is essential for the next generation of nodes.
“As we start making more 3D chips, the heat problem gets worse,” said Moore, who has been making semiconductors on and off for a quarter century.
For the special report in this issue, Moore teamed up with Associate Editor Dina Jenkina, who oversees our computing coverage. He spoke to engineers at IEEE conferences such as IEDM and Supercomputing about how technologists are removing heat in new and surprising ways.
“As we start making more 3D chips, the heat problem gets worse.” -Samuel K. moore
The first step toward solving an engineering problem is to describe it accurately. James Myers of Imec in Cambridge, England, wrote “Will Heat Cause a Moore’s Law Meltdown?” described how transistors entering commercial production in the 2030s will have a power density that will increase temperatures by 9 degrees Celsius. In data centers where millions of hot chips are crammed together, this surge can force the hardware to shut down or risk permanent damage.
In “Next-Gen AI Needs Liquid Cooling,” Jenkina takes readers in-depth about four contenders to beat this heat with liquids: plates cooled with a water-glycol mixture attached directly to the hottest chips; A version of that technique in which a special dielectric liquid boils into vapor; submerging entire servers in tanks filled with dielectric oil; And doing the same in a tank of boiling dielectric fluid.
Although liquid cooling works well, “it is also more expensive and introduces additional points of failure,” Moore warned. “But when you’re consuming kilowatts and kilowatts in such a small space, you do what you have to do.”
As surprising as servers in boiling oil may sound, two other articles in this issue focus on even more radical cooling technologies. One of these involves using a laser to cool the chips. The technology, outlined by Jacob Balma and Alejandro Rodriguez of Minnesota-based startup Maxwell Labs, involves converting phonons (vibrations in a crystal lattice that carry heat) into photons that can be fired away from the pipe. The authors argue that their technology can “target hot spots with laser precision.”
Meanwhile, Stanford’s Shrabanti Chaudhary takes a broader approach to the heat problem, wrapping transistors in polycrystalline diamond films. His team’s technology has advanced remarkably quickly, reducing the diamond-film growth temperature from 1,000 °C to less than 400 °C, making it compatible with standard CMOS manufacturing.
None of these solutions are cheap, and so the future of chips is going to be expensive as well as hot. This probably doesn’t bother big AI companies sitting on huge piles of investors’ cash. As Moore explains in Polishing the Pickle, “The demand for AI chips is limitless, so you have to do things you might not have thought of doing before and swallow the expense.,
From articles on your site
Related articles on the web