Abstract

AI-driven crystal structure generation has led to an explosion of hypothetical inorganic compounds, many of which will never be successfully synthesized in the lab. In this talk, I will present a reaction network-based framework for evaluating the synthetic accessibility of inorganic materials. Rather than relying solely on thermodynamic or heuristic “synthesizability” scores, our approach explicitly models the network of possible reactions leading to a target phase and scores pathways based on their driving forces, selectivity, and practical processing constraints, resulting in a more realistic picture of what can actually be made in the lab. We find that even thermodynamically favorable materials may lack viable synthetic routes under conventional conditions, but expanding the reaction and process search space can reveal new, experimentally accessible pathways. Our framework is implemented into a cloud-based application designed to accelerate synthesis planning and bridge the gap between prediction and realization within the materials research community.