Researchers have developed a machine learning approach that accurately predicts phonon scattering rates and lattice thermal conductivity for a wide range of materials, including silicon, magnesium oxide, and lithium cobalt oxide. By using deep neural networks (DNNs), the team were able to mitigate the computational challenges associated with calculating scattering rates and accurately predict thermal conductivity. The approach is particularly noteworthy given the role that thermal conductivity plays in both the design of materials for high-temperature applications and improving energy efficiency. In comparison to first-principles calculations, the DNN-based models were shown to be orders of magnitude faster, and could enable large-scale thermal transport informatics, the researchers stated.
The study was conducted by researchers at MIT, Argonne National Laboratory, and Shenzhen University.
Lead author Wei Chen is a professor of mechanical engineering and materials science and engineering at MIT.
