AI Speeds Up Solar-Cell Perovskite Discovery

An EPFL research project led by Haiyuan Wang and Alfredo Pasquarello, with collaborators in Shanghai and in Louvain-La-Neuve, has made significant strides in the search for optimal perovskite materials for photovoltaic applications. This innovative method combines advanced computational techniques with machine learning, offering the potential to revolutionize the solar industry by creating more efficient and cost-effective solar panels.

The researchers embarked on their journey by compiling a comprehensive dataset of band-gap values for 246 perovskite materials. This dataset was meticulously crafted using cutting-edge calculations based on hybrid functionals, a sophisticated computational approach that incorporates electron exchange. Unlike the traditional Density Functional Theory (DFT), hybrid functionals provide a more accurate representation of electronic properties, making them ideal for materials like perovskites where electron interaction plays a crucial role.

What sets this research apart is the utilization of "dielectric-dependent" hybrid functionals, which take into account the electronic polarization properties of the materials being studied. By integrating these properties into their calculations, the researchers were able to significantly enhance the accuracy of band-gap predictions. This precision is essential in the field of photovoltaics, where precise control over band-gap values is paramount for achieving maximum efficiency.

Building upon the foundation of the band-gap dataset, the team developed a machine-learning model trained on the 246 perovskites. This model was then applied to a vast database of approximately 15,000 candidate materials for solar cells, enabling the researchers to identify the most promising perovskites based on their predicted band gaps and stability. The outcome was the discovery of 14 entirely new perovskites, each exhibiting optimal band gaps and high energetic stability, positioning them as prime candidates for high-efficiency solar cells.

This groundbreaking work underscores the power of machine learning in expediting the discovery and validation of new photovoltaic materials. By streamlining the research process, this approach not only reduces costs but also accelerates the widespread adoption of solar energy. Such advancements are crucial in our collective efforts to combat climate change and reduce our reliance on fossil fuels, paving the way for a more sustainable future.