TY - JOUR

T1 - Accurate prediction of structural and mechanical properties on amorphous materials enabled through machine-learning potentials

T2 - A case study of silicon nitride

AU - Nayak, Ganesh Kumar

AU - Srinivasan, Prashanth

AU - Todt, Juraj

AU - Daniel, Rostislav

AU - Nicolini, Paolo

AU - Holec, David

PY - 2025/1/6

Y1 - 2025/1/6

N2 - Ab initio calculations represent the technique of election to study material system, however, they presentsevere limitations in terms of the size of the system that can be simulated. Often, the results in the simulationof amorphous materials depend dramatically on the size of the system. Here, we overcome this limitation forthe specific case of mechanical properties of amorphous silicon nitride (a-Si3N4) by training a machine learning(ML) interatomic model. Our strategy is based on the generation of targeted training sets, which also includedeliberately stressed structures. Using this dataset, we trained a moment tensor potential (MTP) for a-Si3N4.We show that molecular dynamics simulations using the ML model on much larger systems yield elasticallyisotropic response and can reproduce experimental measurement. To do so, models containing at least ≈3, 500atoms are necessary. The Young’s modulus calculated from the MTP at room temperature is 220 GPa, which isvery well in agreement with the nanoindentation measurement. Our study demonstrates the broader impact ofmachine learning potentials for predicting structural and mechanical properties, even for complex amorphousstructures.

AB - Ab initio calculations represent the technique of election to study material system, however, they presentsevere limitations in terms of the size of the system that can be simulated. Often, the results in the simulationof amorphous materials depend dramatically on the size of the system. Here, we overcome this limitation forthe specific case of mechanical properties of amorphous silicon nitride (a-Si3N4) by training a machine learning(ML) interatomic model. Our strategy is based on the generation of targeted training sets, which also includedeliberately stressed structures. Using this dataset, we trained a moment tensor potential (MTP) for a-Si3N4.We show that molecular dynamics simulations using the ML model on much larger systems yield elasticallyisotropic response and can reproduce experimental measurement. To do so, models containing at least ≈3, 500atoms are necessary. The Young’s modulus calculated from the MTP at room temperature is 220 GPa, which isvery well in agreement with the nanoindentation measurement. Our study demonstrates the broader impact ofmachine learning potentials for predicting structural and mechanical properties, even for complex amorphousstructures.

UR - https://doi.org/10.1016/j.commatsci.2024.113629

U2 - 10.1016/j.commatsci.2024.113629

DO - 10.1016/j.commatsci.2024.113629

M3 - Article

VL - 249.2025

JO - Computational materials science

JF - Computational materials science

SN - 0927-0256

IS - 5 February

M1 - 113629

ER -