High-Throughput Screening of Two-Dimensional Multifunctional Janus M2X2 via Machine Learning Force Fields

Abstract

Two-dimensional (2D) Janus materials possess unique physical properties due to their broken mirror symmetry, yet their large compositional space makes systematic discovery challenging. Here, we perform a high-throughput, data-driven screening of Janus M₂X₂ monolayers to identify stable materials with multifunctional optoelectronic and electromechanical properties. From 15,428 designed candidates, a transfer-learning-based ensemble machine-learning force field enables efficient stability evaluation. Stepwise thermodynamic, dynamical, and mechanical filtering reduces the set to 7 stable monolayers. Hybrid-functional calculations show that 6 are semiconductors with band gaps of 1.78 ~ 3.49 eV. Notably, Al₂TeSe exhibits a large in-plane piezoelectric coefficient (d₁₁ = 8.95 pmV^-1), low-barrier sliding ferroelectricity (∼22 meV), and strong second-order nonlinear optical response (up to 1064 pmV^-1). In addition, the intrinsic out-of-plane electric field supports charge separation and suitable band alignment for photocatalytic water splitting. This work demonstrates an efficient strategy for chemical space exploration and identifies Janus M₂X₂ monolayers as promising multifunctional 2D materials.