講演情報
[18p-S2_201-1]Surrogate Machine Learning Investigation of Flat versus Island Growth in Fe/MgO(001)
〇(D)Andi MuhNurFitrah Syamsul1, Kohji Nakamura1 (1.Mie Univ.)
キーワード：
Fe/MgO/Fe、islandgrowth、machinelearning
High-performance Fe/MgO/Fe magnetic tunnel junctions require stable, flat interfaces for coherent tunneling, but lattice mismatch and surface energy differences between Fe and MgO promote Fe island formation during early growth. To address this, we applied a surrogate machine learning approach to an Fe/MgO(001) heterostructure to study stability during initial monolayer formation. Two formation schemes were considered: MgO on an Fe substrate and Fe on an MgO substrate. In these schemes, the top layer was relaxed while the substrate layer was fixed as a flat surface. A Gaussian process regression (GPR) model was trained on density functional theory (DFT) data from a reference MgO/Fe(001)-(3x3) heterostructure (one monolayer each, nine atoms per species) and expanded using random atomic displacements. Active learning, using the global optimization with first-principles energy expression (GOFEE) algorithm, was utilized to generate structural configurations corresponding to the global energy minimum and configurations near the global minimum (local minima). Following 50 GOFEE iterations, approximately 100 structures were generated for each scheme. The MgO on Fe substrate produced a relatively flat monolayer configuration for its global minimum, whereas the Fe on MgO substrate scheme produced an island-like configuration, consistent with experimental observations. For the local minimum states within the range of 30.04 to 47.15 meV/atom above the global minimum, the Fe on MgO substrate exhibited flattened configurations, indicating that flat Fe interfaces can exist as metastable configurations within a specific energy range. Controlled energy input, such as annealing that equates to 348.60–547.15 K during early monolayer growth, can be a practical method to overcome Fe islanding energy barriers and stabilize initial flat Fe interface growth.