Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2023 Dec;624(7990):86-91.
doi: 10.1038/s41586-023-06734-w. Epub 2023 Nov 29.

An autonomous laboratory for the accelerated synthesis of novel materials

Nathan J Szymanski #  1   2 , Bernardus Rendy #  1   2 , Yuxing Fei #  1   2 , Rishi E Kumar #  3 , Tanjin He  1   2 , David Milsted  2 , Matthew J McDermott  1   2 , Max Gallant  1   2 , Ekin Dogus Cubuk  4 , Amil Merchant  4 , Haegyeom Kim  2 , Anubhav Jain  3 , Christopher J Bartel  2 , Kristin Persson  1   2 , Yan Zeng  5 , Gerbrand Ceder  6   7
Affiliations

An autonomous laboratory for the accelerated synthesis of novel materials

Nathan J Szymanski et al. Nature. 2023 Dec.

Erratum in

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 41 novel compounds from a set of 58 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 discovery and motivates further integration of computations, historical knowledge and robotics.

PubMed Disclaimer

Conflict of interest statement

The authors declare no competing interests.

Figures

Fig. 1
Fig. 1. Autonomous 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.
Fig. 2
Fig. 2. Outcomes from targeted syntheses of 57 inorganic materials.
Results summary from the synthesis efforts targeting 57 compounds, plotted against their decomposition energies (log scale). Arrows indicate values near zero. A total of 36 targets were successfully synthesized (blue bars), four were inconclusive from XRD (purple bars), and the remaining 17 were not obtained by the A-Lab (red bars). Targets optimized in the active-learning stage of the A-Lab are marked by diagonal lines; all other targets were only attempted using recipes proposed by ML algorithms trained on literature data. The scatter points above each bar represent the outcomes of attempted recipes for each target, ordered from top to bottom in the sequence in which they were performed. The inset pie charts show the fraction of successful targets (left) and recipes (right). Analyses performed after the fact suggest that the calculated decomposition energies for three targets, marked with stars, may be suspect owing to computational errors (see text).
Fig. 3
Fig. 3. Active learning with pairwise reaction analysis.
a, From a failed synthesis attempt, the A-Lab determines which pairwise reactions occurred. b, New precursors are recommended by substituting at least one precursor involved in the unfavourable pairwise reaction. In cases in which the new precursor set leads to identical intermediates as a previously tested set, it is not explored at any higher temperatures. c, The successful precursor set avoids all the unfavourable pairwise reactions. d, The free energy at each step in the reaction pathway, calculated using data from the Materials Project, which shows that the successful pathway maintains a large driving force at the target-forming step. e, Following this approach, the scatter plot shows the number of experiments required to exhaust all unique reaction paths for each target (red) or to identify an optimal path with high yield (blue), plotted with respect to the total size of each experimental search space.
Fig. 4
Fig. 4. Barriers to the synthesis of materials predicted to be stable.
The 17 target materials that could not be synthesized by the A-Lab, for which each is categorized by the feature that complicates its synthesis. One target (Ta4PbO11) is excluded from this list, as it is metastable and therefore was predictably unobtained in favour of its stable competitors. The challenges in synthesizing the remaining 16 stable targets fall under two categories: experimental barriers (blue, 13 targets) and computational barriers (green, three targets). We distinguish these barriers as four distinct failure modes: slow reaction kinetics, precursor volatility, product amorphization and limitations associated with DFT calculations performed at 0 K. A schematic explanation for each failure mode is provided.
Extended Data Fig. 1
Extended Data Fig. 1. A-Lab hardware setup.
Detailed overview of all physical components in the A-Lab, including the stations for precursor preparation, heating and product handling for XRD characterization.
Extended Data Fig. 2
Extended Data Fig. 2. Robotic installations for sample transfer in the A-Lab.
Grippers on the UR5e robotic arms that are used for sample preparation (a), loading/unloading of crucible racks to/from the box furnaces (b) and sample retrieval and characterization (c). d, Linear rail used to increase the working envelope of the robotic arm that loads/unloads crucible racks to/from the furnaces. e, Carousel used to organize and move samples in the sample preparation station.
Extended Data Fig. 3
Extended Data Fig. 3. Communication protocols connecting each module in the A-Lab.
A local area network (LAN) is built to connect all the pieces of the A-Lab with a control computer using an RS-485 interface (or DB25 for the box furnaces). Each module on the RS-485 interface has an Internet Protocol (IP) assigned to enable communication with the computer. For enhanced cybersecurity, only the control computer has access to the internet, whereas the LAN is isolated from it.
See this image and copyright information in PMC

References

    1. Jain, A., Shin, Y. & Persson, K. A. Computational predictions of energy materials using density functional theory. Nat. Rev. Mater.1, 15004 (2016). - DOI
    1. Sun, J. et al. Accurate first-principles structures and energies of diversely bonded systems from an efficient density functional. Nat. Chem.8, 831–836 (2016). - DOI - PubMed
    1. Nikolaev, P. et al. Autonomy in materials research: a case study in carbon nanotube growth. NPJ Comput. Mater.2, 16031 (2016). - DOI
    1. Chang, J. et al. Efficient closed-loop maximization of carbon nanotube growth rate using Bayesian optimization. Sci. Rep.10, 9040 (2020). - DOI - PMC - PubMed
    1. MacLeod, B. P. et al. Self-driving laboratory for accelerated discovery of thin-film materials. Sci. Adv.6, eaaz8867 (2020). - DOI - PMC - PubMed