Toward the Commercialization of Perovskite Photovoltaic Technologies: From Human Intelligence to Artificial Intelligence

Abstract:  

The rapid growth of artificial intelligence, cloud computing, and data centers is driving an unprecedented increase in global electricity demand, placing significant strain on existing energy infrastructure and long-term sustainability. Among renewable energy technologies, photovoltaics offer a uniquely short deployment timeline, making solar energy an especially attractive solution for meeting this urgent and expanding demand. In this context, metal halide perovskite solar cells have emerged as leading candidates due to their exceptional power conversion efficiencies, compatibility with low-cost manufacturing, and potential for rapid scalability.

In this seminar, I will discuss how human intelligence (HI) and artificial intelligence (AI) can be integrated to understand and overcome critical challenges in the commercialization of emerging perovskite photovoltaic technologies. In the first part of the talk, I will show how HI-driven, first-principles understanding reveals the unique physical properties of metal halide perovskites and identifies key failure modes that limit device and module stability. I will also present a rational, molecule-based design strategy informed by this physical insight to mitigate these challenges. Notably, all molecules reported to enhance the stability of perovskite solar cells and modules already exist within PubChem, the world’s largest freely accessible chemical database. However, evaluating the applicability of the billions of molecules cataloged in PubChem is infeasible using HI alone, necessitating AI-assisted approaches.

In the second part of the talk, I will demonstrate how AI—including generative AI—can dramatically accelerate the search, discovery, and optimization of molecules critical to stabilizing perovskite solar cells and modules. Specifically, we employ first-principles density functional theory (DFT)–guided machine learning to refine force fields for systems involving lead halide perovskites and molecular species, ensuring accurate atomistic interactions. The resulting machine-learned force fields are then used to generate large-scale, high-fidelity datasets for molecular representation learning. These datasets enable geometric deep learning models to efficiently predict key material properties and rapidly evaluate the applicability of the entire PubChem molecular space. Finally, we leverage graph transformer–based diffusion models to generate molecules conditioned on targeted property ranges, thereby accelerating molecular discovery across diverse application requirements. Collectively, the complementary HI- and AI-driven strategies outline a pathway toward bankable, commercially viable perovskite photovoltaic technologies

Bio:

Yanfa Yan is an Ohio Research Scholar Endowed Chair and Distinguished University Professor in the Department of Physics and Astronomy at The University of Toledo, and a faculty member of the Wright Center for Photovoltaics Innovation and Commercialization. He previously served as a Principal Scientist at the National Renewable Energy Laboratory. His research focuses on the first-principles theory and physics of materials for energy applications, including thin-film photovoltaic technologies, solar fuel and catalytic devices, semiconductor defect and interface physics, and nanoscale structure–property relationships. His work integrates predictive theory with advanced characterization to address efficiency, stability, and reliability challenges in emerging energy technologies. Professor Yan is a Fellow of the American Physical Society and has been recognized for many years as a Clarivate Top 1% Highly Cited Researcher.