Machine learning for impurity charge-state transition levels in semiconductors from elemental properties using...

Area of Science:

• Materials Science
• Computational Physics
• Semiconductor Physics

Background:

• Accurate quantification of impurity charge-state transition energy levels is crucial for semiconductor optoelectronics.
• Current measurement and calculation methods are time-consuming, necessitating faster prediction techniques.

Purpose of the Study:

• To develop a rapid machine learning (ML) model for predicting impurity transition levels in semiconductors.
• To reduce the computational cost and time associated with predicting these critical energy levels.

Main Methods:

• Utilized a multi-fidelity dataset combining low-fidelity (LDA/GGA) and high-fidelity (HSE06) Density Functional Theory (DFT) calculations.
• Employed an ML approach with features based on elemental properties and impurity positions.
• Applied a modified band alignment scheme to correct low-fidelity DFT results.

Main Results:

• Achieved a root mean squared error of 0.36 eV (mean absolute error of 0.27 eV) against high-fidelity DFT values for II-VI and III-V semiconductors.
• Demonstrated improved accuracy with multi-fidelity data compared to models trained solely on high-fidelity data.
• Successfully predicted transition levels across all zinc blende III-V and II-VI systems.

Conclusions:

• The developed ML model offers a computationally efficient and accurate method for predicting impurity transition levels.
• This approach accelerates the discovery and engineering of semiconductor materials for advanced optoelectronic applications.
• The model provides a valuable tool for guiding future research in semiconductor materials design.