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

Area of Science:

• Materials Science
• Optoelectronics
• Machine Learning Applications

Background:

• Printed metal halide perovskites offer cost-effective, high-throughput manufacturing for optoelectronics like solar cells.
• Spatial heterogeneity in perovskite films hinders the performance of large-area devices despite current patterning capabilities.

Purpose of the Study:

• To accelerate the development of printed perovskite solar cells using a spatial-uniformity-driven Bayesian optimization (BO) approach.
• To enhance the performance of large-area perovskite solar cells by mitigating spatial heterogeneity.

Main Methods:

• Leveraged a Bayesian optimization (BO) surrogate model to explore a 6D design space of ink chemistry and printing physics.
• Utilized iterative experimentation (≈100 trials) guided by an objective function measuring spatial photoluminescence (PL) variance.
• Performed rheological comparisons of ink formulations to address Saffman-Taylor artifacts and improve film uniformity.

Main Results:

• Optimizing for uniformity significantly improved photovoltaic performance.
• Achieved ≈20% power conversion efficiency (PCE) in small-area (0.134 cm²) devices.
• Demonstrated >16% PCE in large-area (1 cm²) devices, showcasing scalability.

Conclusions:

• Uniformity-driven BO is an efficient method for optimizing printed perovskite solar cells.
• This approach effectively identifies key printing physics and mitigates spatial heterogeneity for scalable devices.
• The machine learning strategy accelerates the development and improves the performance of perovskite-based optoelectronics.