Around 70% of the world’s energy is lost as waste heat. Improving predictions of heat movement through semiconductors and insulators could lead to more efficient power generation systems. However, modeling the thermal properties of materials, particularly the phonon dispersion relation, is challenging.
MIT researchers and collaborators developed a machine-learning framework that predicts phonon dispersion relations up to 1,000 times faster than existing AI methods and up to 1 million times faster than traditional approaches. This breakthrough could lead to more efficient energy systems and microelectronics by better managing heat.
The team includes Ryotaro Okabe and Abhijatmedhi Chotrattanapituk (MIT graduate students), Tommi Jaakkola (MIT professor), and researchers from MIT, Argonne National Laboratory, Harvard University, the University of South Carolina, Emory University, the University of California at Santa Barbara, and Oak Ridge National Laboratory.
“Phonons are the culprit for the thermal loss, yet obtaining their properties is notoriously challenging, either computationally or experimentally,” said Mingda Li, associate professor of nuclear science and engineering and senior author of a paper on this technique.
The research was published in Nature Computational Science.
