High-Throughput Screening of Aperiodic Superlattice for Minimum Thermal Conductivity Based on Atomistic Simulation-Informed Effective Medium Theory and Genetic Algorithm

Abstract

Superlattices with suppressed thermal conductivity are of great significance in the field of thermoelectricity and can improve the thermoelectric conversion efficiency of materials. Due to Anderson localization of coherent phonons, aperiodic superlattices have lower thermal conductivity than their periodic counterparts. At present, the thermal conductivity of superlattices is mostly predicted through ab initio or molecular dynamics simulations, which is computationally expensive and limits the size of the system. Meanwhile, there are many layered structural combinations for aperiodic superlattices, making it difficult to efficiently screen through all the combinations to search structures with the minimum thermal conductivity. In this work, based on a modified series thermal resistance model (STRM), a new effective medium theory (EMT) is established to predict the thermal conductivity of periodic and aperiodic superlattices. An adjacency factor near the maximum-resistance layers and a correction function, respectively, are introduced to account for the phonon coherence effect and the degree of randomization in the layer thickness. Combined with the genetic algorithm, EMT enables high-throughput screening of millions of aperiodic superlattice structures. This work demonstrates that the thermal conductivities of aperiodic superlattices at a wide range of system size can be constantly reduced to 1.4∼1.8 W/(m·K), which occurs at averaged periodic thicknesses in a stable range of 2.0∼2.5 nm.