AI-Driven Accelerated Discovery of High-Performance Perovskite Quantum Dots Via Predictive LightGBM Modeling

The
precise synthesis of stable, highly emissive all-inorganic
perovskite quantum dots (QDs), particularly for red and blue emission,
remains a formidable challenge due to complex phase transitions and
extensive synthesis parameter spaces. Traditional empirical methods
necessitate exhaustive trial-and-error experimentation, hindering
rapid optimization. Here, we introduce an end-to-end artificial intelligence
(AI)-driven framework based on a LightGBM model, trained on a curated
data set of 406 experimental hot-injection synthesis protocols, to
predict optimal synthesis conditions for achieving target photoluminescence
(PL) characteristics. Through a strategic reduction from 30 million
model-generated synthesis candidates to a select few experimentally
validated conditions, we successfully synthesized blue-emitting CsPbX3 QDs (PL peak ∼ 465 nm, fwhm <19 nm) and red-emitting
QDs (PL peak ∼ 630 nm, fwhm <30 nm) with significantly enhanced
spectral purity and narrow emission line widths. Our model notably
identifies critical synthesis parameters, including reaction temperature,
precursor ratios, and ligand concentrations, effectively accelerating
the experimental discovery timeline from weeks to less than a day.
This study establishes a robust methodology for the rational, AI-guided
design of perovskite materials, providing a generalizable platform
to expedite future innovations in optoelectronic devices.