AI-driven discovery of high-efficiency perovskite-organic hybrid materials for next-generation photovoltaics

This study introduces an innovative artificial intelligence (AI) framework designed to accelerate the discovery of highefficiency perovskite-organic hybrid materials for next-generation photovoltaics. We use a generative Crystal Diffusion Variational Autoencoder (CDVAE) and a predictive Graph Neural Network (GNN) to help us come up with new crystal structures from scratch and test them quickly in a large chemical space that can't be reached by traditional experimental methods. The CDVAE could make a lot of different perovskite-organic hybrids, up to a million. The GNN, on the other hand, was able to make accurate predictions about important properties like thermodynamic stability (formation energy) and power conversion efficiency (PCE). It did this with very high coefficients of determination (R² > 0.9) on test data that was not used for training. Through multi-objective optimisation and high-throughput virtual screening, we found a number of good candidates. These candidates were compositions that contained fluorinated aromatic organic cations and had DFT-validated PCEs greater than 23% and significantly negative formation energies. This meant that they could be made and would stay stable. This all-encompassing AI-driven strategy cuts the time it takes to find new materials from years to just weeks. This means that you don't have to use costly and time-consuming empirical methods as much. The results indicate that integrating generative and predictive machine learning models may address the stability-efficiency trade-off in perovskite solar cells. This could lead to the independent discovery of new energy materials.

Ҳали таржима қилинмаган