High-Throughput Screening of Rattling-Induced Ultralow Lattice Thermal Conductivity in Semiconductors

Thermoelectric (TE) materials with rattling model show ultralow lattice thermal conductivity for high-efficient energy conversion between heat and electricity. In this work, by analysis of the key spirit of the rattling model, we propose an efficient empirical descriptor to realize the high-throughput screening of ultralow thermal conductivity in a series of semiconductors. This descriptor extracts the structural information of rattling atoms whose bond lengths with all the nearest neighboring atoms are larger than the sum of corresponding covalent radiuses. We obtain 1171 candidates from the Materials Project (MP) Database that contains more than 100 000 materials. Combining the empirical equation of high-throughput computation with a machine learning algorithm, we compute the approximate lattice thermal conductivities (κ_L) and find the κ_L values of 532 materials are less than 2.0 W m^-1 K^-1 at 300 K, which can be regarded as the criteria of ultralow κ_L in general. In particular, we demonstrate that halide double perovskites structures show ultralow κ_L, which provides valuable references for promising low κ_L materials in future experiments. In order to further verify our computational results, we calculate accurate κ_L for Rb₂SnBr₆ and CsCu₃O₂ as candidates with the low lattice thermal conductivity by solving the phonon Boltzmann transport equation. In particular, we demonstrate that Rb₂SnBr₆ has the lowest κ_L value of 0.1 W m^-1 K^-1 at 300 K of all known thermal conductivity materials with the rattling model so far.