Machine Learning-Enabled Design of Point Defects in 2D Materials

Engineered point defects in two-dimensional (2D) materials offer an attractive platform for solid-state devices that exploit tailored opto-electronic, quantum emission, and resistive properties. Here, we develop an approach based on deep transfer learning, machine learning, and first-principles calculations to rapidly predict key properties of point defects. Properties including band gap and bulk formation energy are predicted for over 4,000 2D materials using deep transfer learning. 10,000 dopant, vacancy, divacancy, and antisite defect structures are generated in 150 wide band-gap materials and more than 1,000 defect band structures are computed via first-principles methods. We use physics-informed featurization to generate a minimal description of defect structures and present a general picture of defects across materials systems. We identify over 100 promising, unexplored dopant defect structures in layered metal chalcogenides, hexagonal nitrides, and metal halides including GeS, h-AlN, and MgI₂. We also find ten optimal substitutional defects for nonvolatile resistive switching in atomically thin memristor devices. Au and Ag substitutions in WTe₂- and MoTe₂-based devices are predicted to have switching voltages as low as 110 meV.