講演情報

[7p-P03-36]Machine Learning and DFT Prediction for Tailored Clean Crystallization for Lattice-Mismatched MgO/Fe

〇(D)Andi MuhNurFitrah Syamsul1, Kohji Nakamura1 (1.Mie Univ.)

キーワード:

MgOFe、Interfacial defects、Machine learning

Magnetic Tunnel Junctions (MTJs) with Perpendicular Magnetic Anisotropy (PMA) are central to spin-transfer torque MRAM, with the Fe/MgO interface offering strong PMA and voltage-controlled magnetic anisotropy (VCMA) [1,2]. However, lattice mismatch between Fe and MgO can introduce interfacial strain, leading to structural defects that compromise crystal quality and stability. These effects are further influenced by synthesis methods and oxidation states, as seen in related systems like Fe4N/MgO [3,4]. While DFT is commonly used to study such interfaces, most studies are limited to small unit cells, restricting access to a wider configuration space. To address this, we apply the Global Optimization with First-principles Energy Expressions (GOFEE) method to explore low-energy interfacial structures in Fe/MgO (001) using a 3x3x1 supercell with two monolayers of Fe and MgO [5]. Calculations are performed using the Linear Combination of Atomic Orbitals (LCAO) method with the Perdew-Burke-Ernzerhof (PBE) functional, a (2,2,1) k-point grid, and a double-zeta polarized basis set. Two initial conditions are considered: a clean interface and an oxidized one where one oxygen atom is shifted toward the Fe layer without altering stoichiometry. For each condition, the GOFEE method is applied with a 1.1 A rattle to Mg and O atoms, while Fe atoms remain fixed. After 50 GOFEE iterations, the clean interface reliably produces crystalline MgO, while the oxidized configuration does not, indicating that initial oxygen placement strongly influences interface quality.