ERI: Multi-Scale Machine Learning for Mixed-Cation Perovskite Solar Cells: Integrating Materials Data, Device Simulation, and Nanoscale Characterization

Data-Driven Design of Efficient and Stable Perovskite Solar Cells Using Published Data, Device Simulations, and Nanoscale Characterization Abstract: Perovskite photovoltaics has advanced rapidly as a high-efficiency technology, yet device stability remains a roadblock to large-scale deployment. It is challenging to identify optimal combinations of perovskite composition, processing conditions, and device architecture that simultaneously promote efficient carrier extraction, suppress recombination losses, while preserving stable operation under varied environmental conditions, including illumination, electrical bias, moisture, and temperature. Performance loss in these devices originates from coupled degradation mechanisms, including composition-dependent phase instability, defect-assisted recombination, interfacial energy losses, ion migration, and nanoscale electrical heterogeneity. These mechanisms link materials design to device performance across multiple length scales, making traditional material and device optimization slow and difficult. This award supports a data-driven and machine-learning-enabled approach to establish quantitative relationships among material composition, fabrication processes, device architecture and performance. The mechanistic origins of instability in mixed-cation perovskite solar cells will be investigated, and strategies to mitigate degradation will be proposed. The project aims to combine photovoltaic device databases, physics-based device modeling, and nanoscale electrical characterization to discover material and device design rules. By reducing trial-and-error in materials selection and device design, the work supports photovoltaic technologies manufactured using high-throughput processing and cost-effective materials. Penn State Behrend, located in Erie in northwest Pennsylvania, will serve as a regional site for advanced energy-materials research and workforce development. The project will train undergraduate researchers and a postdoctoral scholar in device fabrication, nanoscale characterization, scientific computing, and machine-learning methods for materials analysis, while disseminating results through peer-reviewed publications, conference presentations, open research workflows, and outreach that connects a regional undergraduate-serving campus to the broader photovoltaics and materials communities. The research will integrate literature-reported device data, physics-based drift-diffusion modeling, nanoscale electrical measurements, and interpretable machine learning in a closed experiment–theory–design loop. A curated photovoltaic device database assembled from published perovskite studies will be used to train predictive models that relate perovskite absorber composition, transport-layer selection, interfacial modification, and fabrication conditions to efficiency and stability metrics. Drift-diffusion simulations will quantify how band offsets, trap-state density, carrier mobility, interfacial recombination, and ionic redistribution shape current–voltage response and operational loss pathways. These predictions will be corroborated through thin-film and device characterizations using conductive atomic force microscopy and Kelvin probe force microscopy, which will map local conductivity, surface potential, and defect-associated heterogeneity across grains, grain boundaries, and interfaces. Interpretable machine-learning tools, including feature-attribution methods and latent-space generative models, will then be used to extract design rules, rank controllable variables, and propose additive engineering strategies and material combinations for further experimental validation. The proposed framework is to link global database trends, device-scale transport models, and in-house nanoscale measurements to establish a mechanistic basis for the rational design of perovskite solar cells, rather than relying on traditional trial-and-error optimization. Expected outcomes include validated multiscale descriptors of loss mechanisms, testable design heuristics for materials and device architectures, and a transferable framework for accelerating perovskite photovoltaics deployment. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.