Data-driven studies of magnetic van der Waals heterostructures

We use data-driven approaches to investigate magnetic van der Waals (vdW) heterostructures of the form AB₂X₄/C₂Y₃ based on a well-known heterostructure MnBi₂Te₄/Bi₂Te₃. A very large number of candidate materials (~10⁷) are considered in the study, which are formed by making chemical substitutions at the atomic sites of MnBi₂Te₄/Bi₂Te₃. Exploring this huge number of candidates by density functional theory (DFT) calculations or experiments alone is prohibitive. The use of machine learning is a promising way to efficiently explore this huge chemical space, thereby accelerating materials discovery. A subset of nearly 300 heterostructures is considered for investigation using first-principles simulations. We investigate the thermodynamic, electronic, and magnetic properties of the structures. We also compare the properties of MnBi₂Te₄-type monolayers with their corresponding heterostructures. Our calculations show that we can enhance the stability of some MnBi₂Te₄-type monolayers as well as alter their magnetic properties by combining them with non-magnetic Bi₂Te₃-type monolayers. Finally, machine learning techniques are used to predict the properties of all AB₂X₄/C₂Y₃ heterostructures and thus identify promising candidates for novel heterostructures that are chemically stable. Our study aims to discover novel vdW materials that could have applications in spintronics, optoelectronics, and quantum computing.