[18a-D511-1~10] 22.1 Joint Session M "Phonon Engineering"

▼ [18a-D511-9] Machine learning model for interfacial thermoelectric properties of bulk-2D-bulk heterostructures

〇(D)Yifei Li¹, Junichiro Shiomi¹ (1.UTokyo)

Keywords:thermoelectric materials, interfacial heterostructures, machine learning

Interfacial engineering of grain boundaries in polycrystalline materials is a commonly used strategy to improve their thermoelectric figure or merit. The main approach has been to introduce significant interfacial phonon scattering and increase the interfacial thermal resistance. In addition, through the energy filtering effect caused by the band offset at the interface, the Seebeck coefficient can be enhanced without affecting the electrical conductivity, thereby realizing the decoupling of thermoelectric parameters. Adding two-dimensional materials at the interface can further increase the temperature drop and thus increase the interfacial thermoelectric voltage. However, the current theoretical studies mainly used the homogeneous theory, which cannot accurately reflect the contribution of the interfacial thermoelectric properties to the overall ZT. In this work, we develop a reliable machine learning model for interfacial thermoelectric properties. A variety of bulk-2D-bulk interfacial heterostructures are constructed, covering the most common bulk materials (Bi₂Te₃, PbTe, Mg₂Si, etc.) and 2D materials (graphene, TMDCs, h-BN, etc.). The interfacial thermoelectric parameters are calculated using high-throughput calculations based on the atomistic Green's function method. The obtained values are used for the training of the machine learning model, with the atomic types and structural characteristics as descriptors. The model can evaluate the energy filtering effect on the interfacial thermoelectric properties in different systems, and rapidly screen structural combinations with high interfacial ZT values, providing potential applications for the experimental design of interfacial engineering.