講演情報
[11p-E214-2]Machine Learning-Based Sampling for Designing Amorphous Heavy Metals
〇(D)Andi MuhNurFitrah Syamsul1, Kohji Nakamura1 (1.Mie Univ.)
キーワード：
amorphous、machine learning、partition function
Amorphous materials are promising for SOT-MRAM because their disordered structures eliminate grain boundaries, reducing electron scattering to yield uniform magnetic properties. However, standard ab initio techniques evaluate stability at 0 K, failing to predict ambient thermodynamic stability driven by configurational entropy. To address this, we use the active learning framework with biased exploration to sample local minima and map the structural state density. Amorphous phases are generated by progressively introducing boron into crystalline heavy metals (Ta, W, Pt). While pure metals exhibit sharp state densities at the ground state, boron addition broadens this distribution across higher energy levels. This broadening begins at ~5% boron concentration, which suggests the amorphous transition.