Statistics on Oxygen Vacancy Defects in Amorphous HfO₂: A Neural-Network Interatomic Potential Assisted...

Area of Science:

• Materials Science
• Computational Materials Science
• Solid State Physics

Background:

• Accurate prediction of defect properties in amorphous materials is crucial for developing functional devices but remains a significant challenge.
• Oxygen vacancies (Vo) are key defects influencing the properties of amorphous hafnium oxide (a-HfO2).

Purpose of the Study:

• To develop a highly accurate and computationally efficient method for predicting defect properties in amorphous materials.
• To establish reliable criteria for using supercell models in the statistical prediction of point defect properties.

Main Methods:

• Development of a graph neural network inter-atomic potential trained on extensive density functional theory (DFT) data for a-HfO2 and its oxygen vacancy defects.
• Utilizing the developed potential with supercell models for high-throughput calculations of neutral Vo defects across a wide range of supercell sizes.
• Analysis of the impact of supercell size on the statistical distribution and accuracy of Vo formation energies.

Main Results:

• Achieved high energy precision of approximately 1 meV atom-1 for defect calculations.
• Demonstrated that small supercells (<1000 atoms) introduce significant errors in the statistical distribution of Vo formation energies.
• Determined that converged calculations require supercells up to 1500 atoms, or alternatively, averaging results from multiple small supercells (e.g., 30 x 96-atom supercells).

Conclusions:

• The developed graph neural network potential enables DFT-level accurate and cost-effective prediction of defect properties in amorphous materials.
• Established quantitative criteria for selecting appropriate supercell sizes or averaging strategies to ensure accuracy in defect property predictions.
• Provided a clear statistical understanding of oxygen vacancy defects in amorphous hafnium oxide, paving the way for improved material design.