Evolution of flat bands in MoSe$_2$/WSe$_2$ moiré lattices: A study combining machine learning and band unfolding methods

Abstract

Moir\'e lattices have served as the ideal quantum simulation platform for exploring novel physics due to the flat electronic bands resulting from the long wavelength moir\'e potentials. However, the large sizes of this type of system challenge the first-principles methods for full calculations of their electronic structures, thus bringing difficulties in understanding the nature and evolution of the flat bands. In this study, we investigate the electronic structures of moir\'e patterns of MoSe$_2$/WSe$_2$ by combining ab initio and machine learning methods. We find that a flat band with a bandwidth of about 5 meV emerges below the valence band edge at the K point for the H-stacking at a twist angle of 3.89$^{\circ}$. Then, it shifts dramatically as the twist angle decreases and becomes about 20 meV higher than the valence band maximum for the twist angle of 3.15$^{\circ}$. Multiple ultra-flat bands emerge as the twist angle is reduced to 1.7$^{\circ}$. Band unfolding reveals that the ultra-flat bands formed by the $\Gamma$ and K valleys show distinct behaviors. The $\Gamma$-valley flat bands are sensitive to the interlayer coupling, thus experiencing dramatic changes as the twist angle decreases. In contrast, the K-valley flat band, which shows a weak dependence on the interlayer coupling, is mainly modulated by the long wavelength moir\'e potential. Therefore, a relatively small angle (2.13$^{\circ}$) is required to generate the K-valley flat band, which experiences a transition from the honeycomb to the triangular lattice as the twist angle decreases.