Berkeley Lab A-Lab: 17 Days, 36 of 57 Target Materials Synthesized Autonomously

Title

Berkeley Lab A-Lab: 17 Days, 36 of 57 Target Materials Synthesized Autonomously

Content

The A-Lab at Lawrence Berkeley National Laboratory is an autonomous solid-state synthesis laboratory that combines machine-learning synthesis planning with robotic powder mixing, heating, and X-ray diffraction characterization to discover and synthesize inorganic materials with minimal human intervention. The deployment was published in Nature in November 2023. The A-Lab paper describes the motivation: "Although promising new materials can be identified at scale using high-throughput computations, their experimental realization is often challenging and time-consuming. Accelerating the experimental segment of materials discovery requires not only automation but autonomy - the ability of an experimental agent to interpret data and make decisions based on it." The A-Lab was designed to close the loop between computational materials prediction and physical synthesis. At the AI layer, the A-Lab uses two machine-learning models working together to analyze diffraction patterns of synthesis products. When synthesis recipes fail to produce a high target yield, active learning closes the loop by proposing improved follow-up recipes. The system uses Density Functional Theory calculations and cross-references its predictions against an analogous database from Google DeepMind's GNoME project. All 57 target materials considered in the study were new to the lab, not present in the training data for the algorithms. At the hardware layer, the A-Lab uses robots to mix and heat powdered solid ingredients and characterize products via X-ray diffraction, all in open-air handling conditions. Each target is screened for predicted reactivity with O2, CO2, and H2O before synthesis. In production, the A-Lab successfully synthesized 36 of 57 target materials over 17 days of operation, spanning 33 elements and 40 structural prototypes. Inspection of the 17 unobtained targets revealed both synthetic and computational failure modes, several of which could be overcome through minor adjustments to the lab's decision-making. The 63% success rate (36/57) demonstrates the power of combining ab initio computations, ML algorithms, accumulated historical knowledge, and automation in experimental research. Looking ahead, the A-Lab team plans to expand the target library, integrate additional ML models for synthesis planning, improve failure-mode detection, and extend the autonomous platform to more complex material classes including battery cathodes and electrolytes.