Machine‐Learning‐Based Exploration of Bending Flexoelectricity in Novel 2D Van der Waals Bilayers

Abstract Accurate examination of electricity generation stemming from higher‐order deformation (flexoelectricity) in 2D layered materials is a highly challenging task to be investigated with either conventional computational or experimental tools. To address this challenge herein an innovative and computationally efficient approach on the basis of density functional theory (DFT) and machine‐learning interatomic potentials (MLIPs) with incorporated long‐range interactions to accurately investigate the flexoelectric energy conversion in 2D van der Waals (vdW) bilayers is proposed. In this approach, short‐range interactions are accurately defined using the moment tensor potentials trained over computationally inexpensive DFT‐based datasets. The long‐range electrostatic (charge and dipole) and vdW interaction parameters are calibrated from DFT simulations. Elaborated comparison of mechanical and piezoelectric properties extracted from the herein proposed approach with available data confirms the accuracy of the devised computational strategy. It is shown that the bilayer transition metal dichalcogenides can show a flexoelectric coefficient 2–7 times larger than their monolayer counterparts. Noticeably, this enhancement reaches up to 20 times for Janus diamane and fluorinated boron‐nitrogen derivatives of diamane bilayers. The presented results improve the understanding of the flexoelectric effect in vdW heterostructures and moreover the proposed MLIP‐based methodology offers a robust tool to improve the design of novel energy harvesting devices.