The Polymer Society of Korea

Abstract

Unraveling nanocrystal formation is essential for advancing optoelectronic materials, yet reaction pathways remain obscured by transient intermediates and complex chemistry. We present a machine learning framework that integrates transformer-based spectral learning with topological manifold analysis (UMAP) to map reaction pathways directly from raw UV–vis data. Applied to InAs nanocrystals, this approach reveals four distinct growth routes, identifies hidden intermediates, and clarifies how additives dictate pathway selection—establishing direct links between reaction dynamics, nucleation density, and final material properties.
To translate these insights into synthesis, we employ Bayesian optimization to design experimental strategies that directly target high-performing material characteristics. This closed-loop, data-driven workflow moves beyond empirical trial-and-error, offering a scalable paradigm where ML not only uncovers hidden chemistry but also accelerates the rational discovery of advanced nanomaterials.