Half-metallic ferromagnets are promising for the electrode materials of magnetic tunnel junctions (MTJ) since they possess a highly spin-polarized electronic structure at the Fermi level. The tunneling magnetoresistance (TMR) ratio of MTJs,however,steeply decreases as temperatures increase. Thus,half-metallic ferromagnets having a high Curie temperature are highly desirable for improving the TMR ratio at room temperature. We explored candidates among about 100,000 sorts of quaternary Heusler alloys with the aid of machine learning. First,we constructed a database that stored physical quantities evaluated by first-principles calculations for more than 4,400 sorts of randomly selected quaternary Heusler alloys. We successfully developed a predictor with relatively high precision for the formation energy,magnetization,and Curie temperature of quaternary Heusler alloys. The predictor revealed about 500 candidates that are energetically stable and possess a Curie temperature higher than 800 K. After verifying the candidates with the use of first-principles calculations,we successfully discovered several half-metallic Heusler alloys,such as CoIrMnSi and CoIrMnGe,that are suitable for the electrodes of MgO-based MTJs. In conclusion,we demonstrated that machine learning is a quite useful technique for efficient exploration during the screening of desirable materials in spintronics.
展开▼