Searching High Spin Polarization Ferromagnet In Heusler Alloy Via Machine Learning

Spintronics uses spin as an information carrier,manipulating spin to realize related devices containing functions such as data storage,logical operations,and quantum computing,it is the most potential application device in the post-Moore era.In order to achieve effective spin injection,such materials should be highly spin-polarized magnetic materials,and in order to ensure normal use at room temperature and industrial preparation possibility,should have a Curie temperature higher than room temperature.In order to screen out stable ferromagnetic materials with high spin polarization in Heusler alloys composed of transition metals,which can be applied to spintronics devices,this paper designs a screening workflow based on deep neural network combined with density functional theory calculations.The strong function fitting ability of deep neural network algorithm makes the efficiency of processing and analysis massive material data significantly improved.About 65,000 materials were collected from the Open Quantum Materials database,and density functional theory calculations considered double spin channels were used to calculate the spin polarizability of 3450materials.These data were used as model training data sets.Three deep neural network models were obtained through training,and the lattice constants,formation energy and spin polarizability of 10577 candidate materials were predicted respectively.According to the conditions that the spin polarizability is greater than 0.87 and the formation energy is less than 80 meV/atom,this workflow has screened out 192 high spin polarizability materials that may be synthesized experimentally and exist stably.According to many previous studies,57 of these materials have been reported as half-metals and18 of them have been reported as semiconductors.In particular,six unreported Heusler alloys have been identified as promising half-metallic ferromagnets,and the Curie temperature estimated by the mean field approximation method of them is higher than or near room temperature.Among them,Fe₂CrGe with antiferromagnetic as the ground state,considering the lattice matching degree in the experimental synthesis,can be transformed from the antiferromagnetic ground state to the ferromagnetic ground state under the influence of the quadrangular deformation.This workflow,with the help of rich material database resources and excellent data fitting capabilities of machine learning,uses only simple descriptors to complete the prediction of the properties of tens of thousands of candidate materials.It is an efficient method for discovering high-spin electronic materials,and open the door for exploring other functional materials.