2D MAGNETIC SEMICONDUCTORS

A crystal graph multilayer descriptor

Adv. Mater., 32, 2002658 (2020)



?



2D ferromagnetic (FM) materials are crucial for next-generation spintronic devices owing to their atomic thickness and controllable electron/spin degree of freedom. However, due to the diversity of 2D structures and the complexity of magnetism, massive search for 2D FM materials is still a tough task. Recent development of machine learning technique has shown great potential in rapid searching for material with target property in large chemical space. Nevertheless, due to the lack of material data and proper descriptor, searching for 2D FM materials remains a challenge.



Recently, the research team led by Prof. Jinlan Wang from School of Physics, Southeast University, China has reported 89 intrinsic ferromagnetic 2D materials through rapid screening framework powered by advanced machine learning techniques and high-throughput calculations. Some of the selected FM materials have considerable high Curie temperature, for example, the Curie temperature of CrCuTe2 monolayer is predicted to be 898 K. Additionally, a sizeable database of 2D magnetic materials, including 542 FM and 917 AFM materials, has been built for further research. Specifically, a new descriptor named CGMD (Crystal Graph Multilayer Descriptor) is developed to describe complex 2D structures and various properties by combining crystal graph and elemental properties in form of multiple feature layers. CGMD shows good performance (over 90% accuracy) on predicting thermodynamic stability, magnetic ground state and bandgap of 2D materials even with a small-scale training dataset. The comprehensive data-driven framework with flexible descriptor might pave a feasible route for machine learning based rapid screening of diverse structures and/or complex properties.