Discovery of the Layered Thermoelectric Compound GeBi2Se4 and Accelerating Its Performance Optimization by Machine Learning

Abstract

Searching for new materials with intrinsically low lattice thermal conductivity is crucial for the exploration of high‐performance thermoelectric materials. Herein, the layered compound GeBi2Se4 with intrinsically low lattice thermal conductivity is discovered, and its thermoelectric performance optimization is accelerated by machine learning. The ultralow lattice thermal conductivity of 0.53 W m−1 K−1 at room temperature for the GeBi2Se4 sample can be ascribed to the large anharmonicity and miscellaneous crystal defects. By alloying tellurium (Te) at the selenium (Se) site, the lattice thermal conductivity is further reduced due to the alloy scattering effect and chemical bond softening while the density‐of‐states effective mass of electrons is significantly increased. Finally, the best n‐type thermoelectric GeBi2Se1.9Te2.1 sample with a dimensionless figure of merit zT of 0.56 at 460 K is screened out by machine learning and verified by experiments, which increases by 140% in comparison with the pristine GeBi2Se4.