Spatial-Uniformity-Driven Bayesian Optimization for Rapid Development of Printed Perovskite Solar Cells

Printed metal halide perovskites can enable rapid, roll-to-roll manufacturing of a broad class of optoelectronics-flexible solar cells and imagers among them-while promising cost and speed advantages over incumbent silicon. However, though current methods offer high throughput and patterning capabilities, perovskite films' spatial heterogeneity remains a challenge for large-area devices. Here, a spatial-uniformity-driven Bayesian optimization (BO) approach is leveraged to accelerate the development of printed perovskite solar cells and improve large-area device performance. Using a BO surrogate model, a 6D design space of ink chemistry and printing physics is explored via extensive iterative experimentation (≈100) informed by an objective function capturing spatial photoluminescence (PL) variance. It is discovered that optimizing for uniformity drives rapid advances in photovoltaic performance, yielding ≈20% power conversion efficiency (PCE) for small area (0.134 cm²) devices and > 16% for large area (1 cm²) devices. This machine-learning approach simultaneously enables rheological comparison of ink formulations that accelerate the leveling of Saffman-Taylor artifacts and improve film uniformity. This showcases uniformity-driven BO as an efficient approach for uncovering the key printing physics and mitigating spatial heterogeneity to enable device scaling beyond small cell areas.