Enhancing perovskite solar cell efficiency and stability: a multimodal prediction approach integrating...

Area of Science:

• Materials Science
• Renewable Energy
• Machine Learning

Background:

• Perovskite solar cell (PSC) performance is dictated by material composition, processing, and microstructure, impacting photovoltaic conversion efficiency (PCE).
• Conventional machine learning methods often fail to capture complex multi-parameter interactions.
• A multimodal approach is needed to accurately predict PSC performance and stability.

Purpose of the Study:

• To develop a multimodal machine learning model integrating diverse data sources for enhanced PSC performance prediction.
• To investigate the relationship between microstructural features, material composition, processing parameters, and PSC efficiency.
• To evaluate PSC stability using machine learning classification.

Main Methods:

• A multimodal model was developed, incorporating SEM-derived microstructural data, material composition, and processing parameters.
• A feature extraction network with a Convolutional Block Attention Module (CBAM) and adaptive feature fusion was employed.
• Gradient Boosting Regressor was identified as the superior algorithm for performance prediction; machine learning classifiers were used for stability assessment.

Main Results:

• The model achieved high accuracy for PCE prediction (R²=0.84) and bandgap estimation (R²=0.95).
• Machine learning models demonstrated robust performance in classifying PSC stability categories (AUC scores ranging from 0.76 to 0.81).
• A direct correlation between larger perovskite grain sizes and higher PCE was confirmed.

Conclusions:

• The multimodal model effectively predicts PSC performance and stability by integrating microstructural, compositional, and processing data.
• Larger perovskite grain size is a key factor for optimizing PCE.
• The framework is validated for scalability across different device sizes and mass production, advancing efficient and durable PSC development.