Characterizing Defect Dynamics in Silicon Carbide Using Symmetry-Adapted Collective Variables and Machine Learning...

Area of Science:

• Materials Science
• Quantum Computing
• Computational Physics

Background:

• Silicon carbide (SiC) divacancies are promising for spin-defect qubits due to long coherence times and optical addressability.
• Simulating SiC defect dynamics is computationally intensive due to high activation barriers, limiting first-principles molecular dynamics studies.

Purpose of the Study:

• To develop and implement machine learning interatomic potentials (MLIPs) for accelerated and accurate SiC defect dynamics simulations.
• To retain first-principles accuracy in simulations of defect formation, motion, and thermodynamic stability.

Main Methods:

• An active learning strategy was employed, combining symmetry-adapted collective variable discovery and enhanced sampling.
• Density functional theory (DFT) was used for calculating energies and forces to train an E(3)-equivariant MLIP (Allegro model).
• Simulations were performed on multidefect 216-atom SiC supercells.

Main Results:

• The trained MLIP achieved DFT-level accuracy for defect transition activation free energy barriers.
• Efficient and stable simulations of large supercells were enabled, allowing analysis of defect thermodynamics and kinetics.
• The temperature dependence of defect stability and kinetics was analyzed.

Conclusions:

• MLIPs significantly accelerate SiC defect dynamics simulations while maintaining high accuracy.
• The study proposes an optimal annealing temperature to maximize the stabilization of VV divacancies for qubit applications.
• This approach facilitates the study of complex defect phenomena in materials for quantum technologies.