AI-Accelerated Synthesis of Targeted Nanoparticle Heterostructures Using Agent-Based Sequential Learning

We present a method to synthesize nanoparticles with specific structural features according to AI predictions. We trained a Gaussian process (GP)-based sequential learning agent using previously gathered scanning transmission electron microscopy (STEM) images and energy dispersive X-ray spectroscopy (EDS) maps to predict new nanoparticle compositions in the 4, 5, and 6-element chemical spaces with a high likelihood of exhibiting a single interface between two distinct phases. After synthesizing nanoparticle arrays according to the agent-predicted compositions using scanning probe block copolymer lithography, we characterized them using STEM and EDS, determining the composition and number of interfaces, and provided the GP-agent with the results. Here, we present the results from several iterations of this process in which 18 of 19 compositions successfully yielded a multi-element nanoparticle with the desired single-interface. We also discuss the prospects for further application of this methodology to predict new structural features, quantify catalytic figures of merit, and screen nanoparticle megalibraries for large-scale materials discovery.