Unveiling the layer-dependent electronic properties in transition-metal dichalcogenide heterostructures assisted by machine learning.

Abstract

The electronic properties of layered two-dimensional (2D) transition-metal dichalcogenide (TMD) van der Waals (vdW) heterostructures are strongly dependent on their layer number (N). However, extremely large computational resources are required to investigate the layer-dependent TMD vdW heterostructures for every possible combination if N varies in a large range. Fortunately, the machine learning (ML) technique provides a feasible way to probe this problem. In this work, based on the density functional theory (DFT) calculations combined with the ML technique, we effectively predict the layer-dependent electronic properties of TMD vdW heterostructures composed of MoS2, WS2, MoSe2, WSe2, MoTe2, or WTe2, in which the layer number varies from 2-10. The cross-validation scores of our trained ML models in predicting the bandgaps as well as the band edge positions exceed 90%, suggesting excellent performance. The predicted results show that in the case of a few-layer system, the number of layers has a significant effect on the electronic properties. The bandgap and band alignment could be dramatically changed from bilayer to triple-layer heterostructures. However, with the increase of the number of layers, the electronic properties change, and some general trends can be summarized. When the layer number is larger than 8, the properties of the TMD heterostructures tend to be stable, and the influence of the layer number decreases. Based on these results, our work not only sheds light on the understanding of the layer-dependent electronic properties of multi-layer TMD vdW heterostructures, but also provides an efficient way to accelerate the discovery of functional materials.