DiSCO—A Human Knowledge Embedded Automated [Di]scovery, [S]ynthesis, [C]haracterization, and [O]ptimization Platform for Identifying Synthesizable Inorganic Materials

Advent of generative design algorithms have successfully helped create databases with millions of inorganic material candidates. However, identifying synthesizable candidates and successfully synthesizing them remain to be outstanding challenges. In our current body of work, we aim to address the challenge of synthesizability by a two-fold approach of<br/>1. predicting synthesizability of novel materials by embedding human knowledge in the form of “Filters” in material discovery pipeline and,<br/>2. validating the ground truth of novel materials by implementing a medium-fidelity high-throughput automated material synthesis and characterization.<br/><br/>Our two-fold solution aims to tackle the problem of novel material synthesizability in two ways. First, we embed chemical rules, practical constraints (raw material availability, ease of synthesis, etc.), and human intuition drawn from tacit knowledge as post-generation “Filters” into our materials discovery pipeline. Each filter embodies a single chemical rule, constraint, or human intuition which needs to be satisfied for the successful synthesis of a novel material. We use these individual filters in different configurations to develop unique screening pipelines. The “Filter” pipeline in combination to material generation algorithm forms the discovery part of our DiSCO platform. The intention of use of such “Filter” pipelines is to screen generated material libraries and identify materials which have a higher probability of being synthesizable. Such post-generation screening pipeline also enables the downselection from million or more potential candidates in generated material libraries to a few hundred compounds which are actually chosen for synthesis, and hence helps with better resource allocation. The downselected materials are sent for synthesis in the next step.<br/><br/>In the second part of our workflow, we synthesize the downselected novel materials, and validate their ground truth. The physical synthesis of the downselected compounds presents several challenges as the formation of the target compound is a function of the expertise of the experimentalist, precursor temperature, precursor molarity, synthesis temperature, substrate temperature, environmental humidity, etc. To address these challenges, we developed the later part of DiSCO featuring islands of automation performing automated synthesis, characterization and optimization. The system combines (a) high-throughput pipetting robots with temperature control for synthesis, thus reducing human introduced variances, (b) automated characterization tools for faster cycles of learning and validating the ground truth, and (c) a graph database system for tracking the meta-data and results. The intention of use of this high-throughput automated platform is to enable very fast synthesis and characterization of the downselected material candidates at different processing conditions, and identify which of the attempted material candidates was successfully created and under what conditions. The purpose of the graph database management system is to seamlessly store the data and meta-data to enable future learning and prediction.<br/><br/>Post high-throughput synthesis and characterization, we used tradition high-fidelity characterization methods to validate the actual composition and structure of the formed compounds to establish further trust in our medium fidelity high-throughput system. As a use case of this material exploration platform, we explore halide “perovskite-inspired” materials system involving 60 phase spaces, to identify materials of bandgaps within the 1 to 2 eV range, by following the process of discovery, synthesis, characterization, and optimization, and creating a database consisting of information on precursor solubility, material processing condition, characterization, and validation data. We identify 2 novel material compositions by following this process.