A Perovskite Material Screening and Performance Study Based on Asymmetric Convolutional Blocks

Area of Science:

• Materials Science
• Computational Chemistry
• Renewable Energy
• Artificial Intelligence in Materials Discovery

Background:

• Developing novel perovskite materials is crucial for advancing solar cell technology.
• Traditional methods for material screening are often time-consuming and inefficient.
• Predictive modeling offers a promising avenue for accelerating materials discovery.

Purpose of the Study:

• To introduce an innovative asymmetric convolution block (ACB) method for identifying high-efficiency perovskite materials.
• To develop a precise predictive model for key perovskite properties like band gap and stability.
• To enable rapid material screening for photovoltaic applications.

Main Methods:

• Preprocessing of extensive perovskite oxide material data.
• Development of a predictive model utilizing an asymmetric convolution block (ACB).
• Utilizing the model for classification and regression tasks to predict material properties, bypassing traditional feature importance filtering.

Main Results:

• The predictive model achieved high accuracy (96.8%) and recall (0.998) in classification tasks.
• Regression tasks showed excellent performance with a coefficient of determination (R²) of 0.993 and a mean squared error (MSE) of 0.004.
• DyCoO₃ and YVO₃ were identified as promising candidates for photovoltaic applications based on predicted optimal band gaps.

Conclusions:

• The developed ACB method provides an efficient and precise framework for identifying high-performance perovskite materials.
• This approach significantly accelerates the screening and development of advanced materials for solar cells.
• The study demonstrates the potential of AI-driven methods in revolutionizing materials discovery for renewable energy.