An autonomous laboratory for the accelerated synthesis of novel materials.

Szymanski NJ, Rendy B, Fei Y, Kumar RE, He T, Milsted D, McDermott MJ, Gallant M, Cubuk ED, Merchant A, Kim H, Jain A, Bartel CJ, Persson K, Zeng Y, Ceder G. · 2023

Read An autonomous laboratory for the accelerated synthesis of novel materials. by Szymanski NJ, Rendy B, Fei Y, Kumar RE, He T, Milsted D, McDermott MJ, Gallant M, Cubuk ED, Merchant A, Kim H, Jain A, Bartel CJ, Persson K, Zeng Y, Ceder G. (2023) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc10700133.txt.

Introduction

Szymanski and co-authors built an autonomous lab in 2023 that makes new materials with no hands on the bench. The lab, called A-Lab, plans and runs solid-state synthesis with robots. It draws on physics data banks, past papers, and machine learning to pick targets and recipes. Targets were new oxides and phosphates flagged as stable by large data sets.

Over 17 days of nonstop work, the lab made 36 compounds from 57 targets. Recipes came from language models trained on past papers, then tuned by active learning. Szymanski reports that the failed runs gave clear hints to fix screening and design. The high hit rate backs AI-run labs as a way to speed materials work.

Catalog record

CREATORSzymanski NJ, Rendy B, Fei Y, Kumar RE, He T, Milsted D, McDermott MJ, Gallant M, Cubuk ED, Merchant A, Kim H, Jain A, Bartel CJ, Persson K, Zeng Y, Ceder G.
YEAR2023
COLLECTIONPAPERS
FORMATTXT
FILE SIZE59 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC10700133

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

An autonomous laboratory for the accelerated synthesis of inorganic materials By Nathan J. Szymanski, Bernardus Rendy, Yuxing Fei, Rishi E. Kumar, Tanjin He, David Milsted, Matthew J. McDermott, Max Gallant, Ekin Dogus Cubuk, Amil Merchant, Haegyeom Kim, Anubhav Jain Published in: Nature DOI: 10.1038/s41586-023-06734-w ABSTRACT To close the gap between the rates of computational screening and experimental realization of novel materials1,2, we introduce the A-Lab, an autonomous laboratory for the solid-state synthesis of inorganic powders. This platform uses computations, historical data from the literature, machine learning (ML) and active learning to plan and interpret the outcomes of experiments performed using robotics. Over 17 days of continuous operation, the A-Lab realized 36 compounds from a set of 57 targets including a variety of oxides and phosphates that were identified using large-scale ab initio phase-stability data from the Materials Project and Google DeepMind. Synthesis recipes were proposed by natural-language models trained on the literature and optimized using an active-learning approach grounded in thermodynamics. Analysis of the failed syntheses provides direct and actionable suggestions to improve current techniques for materials screening and synthesis design. The high success rate demonstrates the effectiveness of artificial-intelligence-driven platforms for autonomous materials synthesis and motivates further integration of computations, historical knowledge and robotics. MAIN 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. Pioneering efforts have demonstrated autonomy in several aspects of materials research, including robotic and Bayesian-driven optimization of carbon nanotube yield3,4, photovoltaic performance5 and photocatalysis activity6. In contrast to conventional ML algorithms used for optimization, human researchers benefit from a wealth of background knowledge that informs their decision-making, and it is increasingly recognized7–9 that autonomy will require a fusion of encoded domain knowledge, access to diverse data sources and active learning. Here we present the A-Lab, an autonomous laboratory that integrates robotics with the use of ab initio databases, ML-driven data interpretation, synthesis heuristics learned from text-mined literature data and active learning to optimize the synthesis of inorganic materials in powder form. Although autonomous workflows based on liquid handling have been demonstrated in organic chemistry10–13, the A-Lab addresses the unique challenges of handling and characterizing solid inorganic powders. These often require milling to ensure good reactivity between precursors, which can have a wide range of physical properties related to differences in their density, flow behaviour, particle size, hardness and compressibility. The use of solid powders is well suited for manufacturing and technological scaleup, and the approach of the A-Lab to synthesis produces multigram sample quantities that facilitate device-level testing. Given a set of air-stable target materials (that is, desired synthesis products whose yield we aim to maximize) screened using the Materials Project14, the A-Lab generates synthesis recipes using ML models trained on historical data from the literature and then performs these recipes with robotics. The synthesis products are characterized by X-ray diffraction (XRD), with two ML models working together to analyse their patterns. When synthesis recipes fail to produce a high target yield, active learning closes the loop by proposing improved follow-up recipes. Over 17 days of operation, the A-Lab successfully synthesized 36 of 57 target materials that span 33 elements and 40 structural prototypes (Supplementary Fig. 1 and Supplementary Table 1). Inspection of the 17 unobtained targets revealed synthetic as well as computational failure modes, several of which could be overcome through minor adjustments to the lab’s decision-making. With its high success rate in validating predicted materials, the A-Lab showcases the collective power of ab initio computations, ML algorithms, accumulated historical knowledge and automation in experimental research. AUTONOMOUS MATERIALS-SYNTHESIS PLATFORM The materials synthesis and characterization procedures used in the A-Lab are shown in Fig. 1. All target materials considered in this work are new to the lab, that is, not present in the training data for the algorithms it uses to propose synthesis recipes (Methods). The experiments reported in this study represent the first attempts by the A-Lab to synthesize any of these targets. Each target is predicted to be on or very near (<10 meV per atom) the convex hull formed by stable phases taken from the Materials Project14 and cross-referenced with an analogous database from Google DeepMind. Because the A-Lab handles samples in open air, we only considered targets that are predicted not to react with O2, CO2 and H2O (Methods). Like all databases comprised of atomistic calculations15–17, the Materials Project represents its compounds as ordered structures, even in cases where these materials may form in experiments with partial disorder. This approach is commonly used in theory-driven materials screening efforts as ab-initio computations cannot represent partial atoms18,19. As such, we consider a synthesis procedure successful when it forms either an ordered or partially disordered version of its target material20,21.Fig. 1Autonomous materials synthesis with the A-Lab.Air-stable targets are identified using DFT-calculated convex hulls consisting of ordered ground states from the Materials Project and Google DeepMind. Synthesis recipes for each target are proposed using ML models trained on synthesis data from the literature. These recipes are tested using a robotic laboratory that automates (1) powder dosing, (2) sample heating and (3) product characterization with XRD. All sample transfer between these stations is performed using robotic arms, forming a fully automated sequence from chemical input to characterization. Phase purity is assessed from XRD, which is analysed by ML models trained on structures from the Materials Project and the ICSD, and confirmed with automated Rietveld refinement. In cases in which high (>50%) target yield is not obtained, new synthesis recipes are proposed by an active-learning algorithm that identifies reaction pathways with maximal driving force to form the target. For each compound proposed to the A-Lab, up to five initial synthesis recipes are generated by a ML model that has learned to assess target ‘similarity’ through natural-language processing of a large database of syntheses extracted from the literature22, mimicking the approach of a human to base an initial synthesis attempt on analogy to known related materials. A synthesis temperature is then proposed by a second ML model trained on heating data from the literature23 (Methods). If these literature-inspired recipes fail to produce >50% yield for their desired targets, the A-Lab continues to experiment using Autonomous Reaction Route Optimization with Solid-State Synthesis (ARROWS3), an active-learning algorithm that integrates ab initio computed reaction energies with observed synthesis outcomes to predict solid-state reaction pathways24. Experiments are performed under the guidance of this algorithm until the target is obtained as the majority phase or all synthesis recipes available to the A-Lab are exhausted. The A-Lab carries out experiments using three integrated stations for sample preparation, heating and characterization, with robotic arms transferring samples and labware between them (Fig. 1 and Extended Data Figs. 1 and 2). The first station dispenses and mixes precursor powders before transferring them into alumina crucibles. A robotic arm from the second station loads these crucibles into one of four available box furnaces to be heated (Methods). After allowing the samples to cool, another robotic arm transfers them to the third station, where they are ground into a fine powder

Read the complete text