Synthesizer: Machine Learning-Guided Perovskite Nanocrystal Optimization

Precise control over nanocrystal synthesis is crucial for tailoring their optical properties, yet traditional optimization remains labor-intensive and empirical. Here, we develop a data-efficient approach that combines experimental insights with machine learning to optimize the synthesis of quantum-confined nanocrystals. Using \ce{CsPbBr3} as a model system, we achieve precisely tunable photoluminescence (430–520 nm) with exceptionally narrow emission linewidths (70 meV) enabled through lateral size tuning and strategic choice of antisolvents. Our results also reveal that quantum yield is independent of antisolvent selection but correlates with surface trap density, offering insights into defect passivation strategies. A Gaussian Process regression model, incorporating a geometric encoding of antisolvent molecules, predicts photoluminescence peak wavelengths with high accuracy (< $\pm$4 nm), significantly reducing the number of required syntheses for optimization. This framework is transferable to other material systems, as we show for \ce{CsPbI3}, enabling accelerated synthesis optimization and expanding the tunability of functional nanomaterials beyond halide perovskites for optoelectronic applications.