Related Experiment Video
Updated: May 5, 2026
11:38
Influence of Hybrid Perovskite Fabrication Methods on Film Formation, Electronic Structure, and Solar Cell Performance
Published on: February 27, 2017
19.0K
Machine learning-guided optimization of lead-free perovskite solar cells: predicting PCE with high accuracy
Sanidul Islam1, Nikit Kundra2, Renu Thakur3
1Mahatma Gandhi Central University, Motihari, Bihar, 845401, India.
Environmental Science and Pollution Research International
|October 14, 2025
Summary
This study optimized lead-free perovskite solar cells (PSCs) by tuning absorber thickness and doping concentration, achieving high power conversion efficiency (PCE). Machine learning models, particularly XGBoost, accurately predicted PCE, accelerating device development.
Area of Science:
- Materials Science
- Renewable Energy
- Computational Physics
Background:
- Lead-free perovskite solar cells (PSCs) offer a sustainable alternative to traditional photovoltaic technologies.
- Optimizing absorber layer thickness and doping concentration is crucial for enhancing PSC performance.
- Machine learning (ML) presents a powerful tool for accelerating the design and optimization of complex material systems.
Purpose of the Study:
- To investigate the impact of absorber layer thickness and acceptor doping concentration on the photovoltaic performance of CsSn₀.₅Ge₀.₅I₃-based PSCs.
- To develop and validate ML models for predicting the power conversion efficiency (PCE) of PSCs.
- To identify key parameters influencing PSC performance using ML-driven analysis.
Main Methods:
- Device simulations were performed to analyze the effects of varying absorber thickness and doping levels.
- Three ML algorithms (Neural Network, Random Forest, XGBoost) were trained on a dataset of 1000 simulated PSCs.
- Model performance was evaluated using metrics like MSE, RMSE, MAE, and R², with SHAP analysis used for feature importance.
Main Results:
- An optimal absorber thickness of 0.8 µm was identified, enhancing light absorption and charge extraction.
- Maximum simulated photovoltaic parameters included a short-circuit current density (JSC) of 28.53 mA/cm², an open-circuit voltage (VOC) of 1.219 V, and a PCE of 31.29%.
- XGBoost demonstrated superior predictive accuracy (R² = 0.997) for PCE, with absorber thickness and doping concentration identified as dominant factors.
Conclusions:
- Optimizing absorber layer thickness and acceptor doping concentration is critical for maximizing PSC performance.
- ML models, especially XGBoost, significantly accelerate the optimization process for PSCs.
- This integrated approach of physical design and ML provides a robust framework for advancing lead-free PSC technology.
