Two-step machine learning enables optimized nanoparticle synthesis

Abstract In materials science, the discovery of recipes that yield nanomaterials with defined optical properties is costly and time-consuming. this study, we present a two-step framework for machine learning-driven high-throughput microfluidic platform to rapidly produce silver nanoparticles desired absorbance spectrum. Combining Gaussian process-based Bayesian optimization (BO) deep neural network (DNN), algorithmic able converge towards target spectrum after sampling 120 conditions. Once dataset large enough train DNN sufficient accuracy in region spectrum, used predict colour palette accessible reaction synthesis. While remaining interpretable by humans, proposed efficiently optimizes nanomaterial synthesis can extract fundamental knowledge relationship between chemical composition properties, such as role each reactant on shape amplitude