Machine-learning-enabled solvent engineering for uniform quantum dot packing in efficient and stable quantum-dot light-emitting diodes

Abstract Colloidal quantum dots (QDs) offer size-tunable optoelectronic properties and solution processability, yet achieving uniformly packed emissive layers remains a bottleneck for high-performance quantum-dot light-emitting diodes (QLEDs). Here, we report a machine learning -guided solvent optimization strategy to produce homogeneous QD films. Five representative solvent parameters are evaluated, and multiple regression models are trained against film uniformity derived from atomic force microscopy. Among them, support vector regression provides the highest predictive accuracy for film homogeneity. Guided by these predictions, we formulate a mixed solvent that closely matches the target profile, yielding superior packing homogeneity, which is confirmed by grazing-incidence small-angle x-ray scattering. QLEDs fabricated with this formulation exhibit an external quantum efficiency of 20.6% and an operational lifetime of 468.5 h at 20 000 cd m −2 , surpassing all single-solvent controls. These findings establish packing homogeneity as a decisive factor for device performance and introduce a generalizable and scalable framework for data-driven design of solution-processed devices, holding strong potential for next-generation optoelectronic systems.