Autonomous Intelligent Agents for Accelerated Materials Discovery

Abstract

We present an end-to-end computational system for autonomous materials discovery. The system aims for cost-effective optimization in large, high-dimensional search spaces of materials by adopting a sequential, agent-based approach to deciding which experiments to carry out. In choosing next experiments, agents can make use of past knowledge, surrogate models, logic, thermodynamic or other physical constructs, heuristic rules, and different exploration-exploitation strategies. We show a series of examples for (i) how the discovery campaigns for finding materials satisfying a relative stability objective can be simulated to design new agents, and (ii) how those agents can be deployed in real discovery campaigns to control experiments run externally, such as the cloud-based density functional theory simulations in this work. In a sample set of 16 campaigns covering a range of binary and ternary chemistries including metal oxides, phosphides, sulfides and alloys, this autonomous platform found 383 new stable or nearly stable materials with no intervention by the researchers.

The materials tetrahedron has a “digital twin”

“…Acceleration of materials discovery and development presupposes the accompanying digital methods and workflows for collecting, curating, integrating, and analyzing data. Methods and workflows may include algorithms for achieving "inverse design" of processing parameters given a set of performance goals, 40 multiphysics simulations, 41 high-throughput methods that automate or semi-automate experimentation and data collection, [42][43][44] interpretable machine learning methods, 45,46 retrieval and management of large data sets facilitated by open-source toolkits, [47][48][49] methods for bridging length scales and imaging modalities via correlative characterization, 50 or mixed-initiative user interfaces that leverage automation to support better decision-making through human-computer interaction. 51 When applying the MITT paradigm to a materials data and informatics, one should consider how these methods and workflows combine to transform fragmented data into actionable information.…”

Toward autonomous design and synthesis of novel inorganic materials

“…Upon conducting a set of experiments and interpreting the resulting data, the next step in the closed-loop automation process is to learn from this information and make decisions regarding the subsequent experiments to be performed. These decisions are usually made with the goal of optimizing some quantity 82 ; for example, maximizing the yield of a product by modifying its synthetic procedure 83 or tuning the properties of a material with respect to its structure, composition, or processing conditions 12,48 . Alternatively, decisions can be made to formulate experimental tests that reveal information regarding a specific process 84 .…”

Rational Solid-State Synthesis Routes for Inorganic Materials

“…Today, this mapping can be achieved by efficiently populating the space with low-energy hypothetical compounds generated using data-driven methods. 17,[117][118][119] While the current implementation uses MP as the main data source, 29 extensions of the framework to work with other high-throughput DFT databases is also straightforward. [120][121][122][123] Finally, we should stress that while we have demonstrated that PIRO can provide guidance as an effective solid-synthesis planning tool for a wide array of materials, scientists would be the ultimate decision makers in reaction selection, considering many additional factors from safety or toxicity to decomposition temperatures, reactivities, volatilities, instabilities, moisture sensitivities, equipment constraints, and more critically, their field expertise and heuristics.…”