Machine Learning Enhanced High-Throughput Fabrication and Optimization of Quasi-2D Ruddlesden–Popper Perovskite Solar Cells

Organic–inorganic perovskite solar cells (PSCs) are promising candidates for next-generation, inexpensive solar panels due to their commercially competitive cost and high power conversion efficiencies. However, PSCs suffer from poor stability. A new and vast subset of PSCs, quasi-two-dimensional Ruddlesden–Popper PSCs (quasi-2D RP PSCs), has improved photostability and superior resilience to environmental conditions compared to three-dimensional metal-halide PSCs. To accelerate the search for new quasi-2D RP PSCs, this work reports a combinatorial, machine learning (ML) enhanced high-throughput perovskite film fabrication and optimization study. This work designs a bespoke experimental strategy and produces perovskite films with a range of different compositions using only spin-coating free, reproducible robotic fabrication processes. The performance and characterization data of these solar cells are used to train a ML model that allow materials parameters to be optimized and direct the design of improved materials. The new, ML-optimized, drop-cast quasi-2D RP perovskite films yield solar cells with power conversion efficiencies of up to 16.9%.