Abstract:Sn-based perovskites as low-toxic materials are actively studied for optoelectronic applications. However, their performance is limited by $p$-type self-doping, which can be suppressed by substitutional doping on the cation sites. In this study, we combine density functional theory (DFT) calculations with machine learning (ML) to develop a predictive model and identify the key descriptors affecting formation energy and charge transition levels of the substitutional dopants in CsSnI$_{3}$. Our DFT calculations create a dataset of formation energies and charge transition levels and show that Y, Sc, Al, Nb, Ba, and Sr are effective dopants that pin the fermi level higher in the band gap, suppressing the $p$-type self-doping. We explore ML algorithms and propose training the random forest regression model to predict the defect formation properties. This work shows the predictive capability of combining DFT with machine learning and provides insights into the important features that determine the defect formation energetics.
What problem does this paper attempt to address?
This paper aims to solve the problem of performance limitation in tin - based perovskite materials (such as CsSnI₃) in optoelectronic applications due to the self - doping effect. Specifically, the goal of the paper is to predict and identify substitute dopants that can effectively suppress p - type self - doping through a combination of density functional theory (DFT) calculations and machine learning (ML), thereby improving the optoelectronic performance of CsSnI₃.
### Background of the Paper
- **Research Object**: CsSnI₃, as a low - toxicity halide perovskite material, has potential optoelectronic application prospects, but its performance is affected by p - type self - doping, resulting in a low photoelectric conversion efficiency.
- **Existing Problems**: p - type self - doping causes the Fermi level to be fixed near the valence band, increasing the background hole concentration and affecting the optoelectronic performance of the material.
### Research Methods
1. **DFT Calculations**:
- Use HSE06+SOC (hybrid functional plus spin - orbit coupling) for DFT calculations to generate datasets of formation energies and charge transfer levels of substitute dopants.
- Calculate the formation energies and charge transfer levels of elements in different oxidation states (such as Mg, Ca, Sc, Y, Al, etc.) as Sn - site substitute dopants.
2. **Machine Learning**:
- Use the datasets generated by DFT to train machine learning models such as Random Forest Regression (RFR), Gaussian Process Regression (GPR), and Kernel Ridge Regression (KRR).
- By analyzing feature correlations and feature importance, determine the key descriptors that affect defect formation energies and charge transfer levels, such as oxidation state, heat of formation, density, and ionization energy.
### Main Findings
- **Effective Dopants**:
- Trivalent dopants (such as Al, Sc, Y) are more likely to be in the +1 charge state at the Sn site and have shallow or no charge transfer levels, which can fix the Fermi level at a higher position and effectively suppress p - type self - doping.
- Divalent dopants (such as Mg, Zn) are stable in the neutral charge state and have low formation energies.
- High - oxidation - state dopants (such as Nb) can also raise the Fermi level under I - poor conditions.
- **Machine Learning Models**:
- The Random Forest Regression model performs best in predicting defect formation energies and charge transfer levels, with training/test RMSE values of 0.22/0.26 eV, 0.16/0.23 eV, and 0.16/0.26 eV respectively.
- The model predicts that five dopants such as La, Ce, Pr, Sr, and Ba have low formation energies in the +1 or 0 charge state and are expected to be used to further optimize the optoelectronic performance of CsSnI₃.
### Conclusion
The paper successfully predicts substitute dopants that can effectively suppress p - type self - doping by combining DFT calculations and machine learning methods and provides the physical and chemical principles of key descriptors. These results provide an important theoretical basis and experimental guidance for optimizing the optoelectronic performance of CsSnI₃.