pith. sign in

arxiv: 2605.18033 · v1 · pith:YR2FU3T2new · submitted 2026-05-18 · ❄️ cond-mat.mtrl-sci · cs.LG· physics.app-ph

Real-time Multi-instrument Autonomous Discovery of Novel Phase-change Memory Materials

Chih-Yu Lee , Haotong Liang , Ryan Kim , Austin McDannald , Carlos A Rios Ocampo , A. Gilad Kusne , Ichiro Takeuchi This is my paper

Pith reviewed 2026-05-20 09:51 UTC · model grok-4.3

classification ❄️ cond-mat.mtrl-sci cs.LGphysics.app-ph
keywords autonomous discoveryphase-change memorymulti-instrument integrationclosed-loop optimizationMn-Sb-Te ternaryX-ray diffractionelectrical resistancematerials discovery
0
0 comments X

The pith

The MAD framework merges live XRD and resistance data to map crystal structures and optimize phase-change memory properties simultaneously in a closed loop.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents the Multi-instrument Autonomous Discovery (MAD) framework to solve the problem of integrating heterogeneous and unsynchronized data from multiple instruments during ongoing experiments rather than after full collection. It applies this to the previously unexplored Mn-Sb-Te ternary system for phase-change memory materials by using a multi-output model with a co-regionalization kernel that links X-ray diffraction structural information to electrical resistance measurements. The shared probabilistic posteriors and uncertainty estimates support two concurrent goals: maximizing knowledge of crystal structure distributions through non-negative matrix factorization while locating the composition with peak resistance. In practice this identified promising electrical PCM candidates and the underlying synthesis-process-structure-property relationship after only 25 closed-loop iterations. The work indicates that future large-scale autonomous facilities can run experiments in parallel with shared knowledge instead of handling each instrument independently.

Core claim

The MAD framework combines structural property mapping and functional property optimization in real time by employing a multi-output model whose co-regionalization kernel merges heterogeneous XRD and resistance data streams, yielding shared posteriors that enable simultaneous non-negative matrix factorization of crystal structures and direct optimization of maximum resistance values, thereby identifying the undetermined SPSPR and promising PCM compositions in the Mn-Sb-Te system within 25 closed-loop iterations.

What carries the argument

The multi-output model with co-regionalization kernel that links unsynchronized XRD and resistance measurements to produce shared posteriors and uncertainty estimates for joint decision making across differing objectives.

If this is right

  • Promising electrical PCM candidates can be located in unexplored ternary composition spaces with far fewer total measurements.
  • The SPSPR for a material system can be determined while actively optimizing a key functional figure of merit in the same experimental campaign.
  • Decision making can draw on shared knowledge across instruments even when the immediate goals of each task differ.
  • Large-scale autonomous facilities can schedule parallel rather than sequential experiments across characterization tools.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same merged-data approach could accelerate discovery in other material families where structural and transport measurements are complementary but collected at different times.
  • Extending the framework to three or more instruments would require only additional output dimensions in the same co-regionalization structure.
  • The reported seven-fold reduction in iterations suggests that closed-loop multi-instrument methods scale favorably as the number of parallel experimental stations increases.

Load-bearing premise

The co-regionalization kernel can reliably combine heterogeneous, unsynchronized XRD and resistance data into accurate shared posteriors and uncertainty estimates that guide effective simultaneous structural mapping and resistance optimization.

What would settle it

Running the same Mn-Sb-Te search with independent per-instrument loops instead of the merged multi-output model and observing that more than 25 iterations are required to reach comparable identification of promising PCMs and the SPSPR.

read the original abstract

Autonomous labs enable the integration of automated experiment execution, data analysis and decision making. The main challenge remains the integration of diverse data streams from multiple instruments, where the data is often heterogeneous and unsynchronized. The standard learning process of undetermined synthesis-process-structure-property relationships (SPSPR) usually relies on post-experiment analysis after data is fully collected, not during live experiments, and decision making is carried out independently across characterization equipment. Here, we demonstrate the Multi-instrument Autonomous Discovery (MAD) framework -- combining structural property mapping and functional property optimization simultaneously in a closed-loop manner. As an example, we applied MAD to phase change memory (PCM) materials, and, in particular on the Mn-Sb-Te ternary, a previously unexplored materials system for PCM. A multi-output model is employed to merge data from x-ray diffraction (XRD) and electrical resistance measurements simultaneously through a co-regionalization kernel that models the relationship between them. The output probabilistic posterior and uncertainty quantification facilitate decision making with shared knowledge, while the goals are different across tasks. We aimed to maximize the knowledge of crystal structure distribution using non-negative matrix factorization (NMF), while in parallel, we find the composition with the maximum resistance value, an important figure of merit for PCM. Leveraging MAD, we found promising electrical PCMs and identified the SPSPR within 25 closed-loop iterations, corresponding to a seven-fold speed-up. The framework opens a new path of study in large-scale autonomous facilities, where future experiments can be run in parallel together, not independently.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces the Multi-instrument Autonomous Discovery (MAD) framework for real-time integration of heterogeneous data streams from multiple instruments in closed-loop autonomous materials discovery. Applied to the unexplored Mn-Sb-Te ternary system for phase-change memory (PCM) materials, it employs a multi-output Gaussian process model with co-regionalization kernel to simultaneously perform structural mapping via non-negative matrix factorization on XRD data and optimize electrical resistance. The framework identifies promising PCM compositions and the synthesis-process-structure-property relationship (SPSPR) in 25 iterations, reported as a seven-fold speedup.

Significance. If the performance claims hold, the work offers a meaningful step toward practical multi-instrument autonomous labs by demonstrating simultaneous structural and functional optimization from unsynchronized data. The experimental application to a new PCM system and use of standard multi-output GP machinery with fresh measurements provide a concrete example that could inform scaling to larger facilities.

major comments (2)
  1. [Abstract / Results] Abstract and Results section: the headline claim that the SPSPR was identified within 25 closed-loop iterations 'corresponding to a seven-fold speed-up' is load-bearing for the central contribution, yet no explicit baseline (random search, sequential single-instrument loops, or simulated comparator) is defined or reported with iteration counts, error bars, or supplementary data, preventing independent verification of the numerical factor.
  2. [Section 3] Section 3 (Multi-output model description): the co-regionalization kernel is stated to merge heterogeneous, unsynchronized XRD and resistance streams into shared posteriors that support simultaneous decision-making, but the manuscript provides no quantitative validation (e.g., cross-validation metrics or ablation on synchronization handling) showing that the shared uncertainty estimates remain reliable when the two tasks have different objectives.
minor comments (2)
  1. [Section 3] Notation for the co-regionalization kernel and NMF components could be introduced more explicitly with a small equation block to aid readers unfamiliar with multi-output GPs.
  2. [Figures] Figure captions should specify the exact number of iterations shown and whether error bars represent model uncertainty or experimental replicates.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and insightful comments on our manuscript. We have addressed each major comment point by point below, providing clarifications and committing to revisions where the concerns are valid and can be resolved with additional analysis or data.

read point-by-point responses

  1. Referee: [Abstract / Results] Abstract and Results section: the headline claim that the SPSPR was identified within 25 closed-loop iterations 'corresponding to a seven-fold speed-up' is load-bearing for the central contribution, yet no explicit baseline (random search, sequential single-instrument loops, or simulated comparator) is defined or reported with iteration counts, error bars, or supplementary data, preventing independent verification of the numerical factor.

    Authors: We agree that an explicit baseline comparison is necessary to substantiate the seven-fold speedup claim and allow independent verification. The reported figure was based on an internal estimate comparing the 25 iterations of the joint MAD framework against the iteration counts observed in our prior single-instrument sequential experiments on similar systems, but this was not documented with sufficient detail. In the revised manuscript we will add a new supplementary section that includes simulated random-search and sequential single-instrument baselines, each run with multiple random seeds to provide iteration counts, mean performance curves, and error bars. This will make the speedup calculation fully transparent and reproducible. revision: yes

  2. Referee: [Section 3] Section 3 (Multi-output model description): the co-regionalization kernel is stated to merge heterogeneous, unsynchronized XRD and resistance streams into shared posteriors that support simultaneous decision-making, but the manuscript provides no quantitative validation (e.g., cross-validation metrics or ablation on synchronization handling) showing that the shared uncertainty estimates remain reliable when the two tasks have different objectives.

    Authors: We acknowledge that the manuscript would benefit from quantitative validation of the multi-output model under unsynchronized conditions. The co-regionalization kernel follows the standard formulation of multi-task Gaussian processes and was chosen precisely because it permits joint posterior inference even when the two outputs (NMF-derived structural features and resistance) have distinct objectives and sampling times. To address the referee’s request, we will include in the revision (i) k-fold cross-validation metrics on held-out XRD and resistance data, and (ii) an ablation comparing the joint model against independent single-output GPs, quantifying the effect of synchronization handling on uncertainty calibration and decision quality. These additions will demonstrate that the shared uncertainty estimates remain reliable for simultaneous structural mapping and property optimization. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation relies on fresh experiments and standard GP machinery

full rationale

The paper applies a multi-output Gaussian process with co-regionalization kernel to merge unsynchronized XRD and resistance measurements in a closed-loop autonomous discovery loop on the Mn-Sb-Te system. The reported identification of SPSPR within 25 iterations and the associated seven-fold speedup are presented as empirical outcomes of the experimental campaign, not as quantities derived by algebraic reduction from the fitted model parameters or from any self-citation chain. No equation equates a claimed prediction to a fitted input by construction, and the central results rest on newly acquired physical measurements rather than on re-labeling of the model's own outputs.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The framework relies on standard assumptions of Gaussian process regression and non-negative matrix factorization without introducing new physical entities or ad-hoc constants beyond typical model hyperparameters.

axioms (2)
  • domain assumption Gaussian process assumptions hold for the joint distribution of XRD structural maps and electrical resistance values across the composition space.
    Invoked when the multi-output model with co-regionalization kernel is used to produce shared posteriors.
  • domain assumption Non-negative matrix factorization yields physically meaningful crystal-structure distributions from the XRD data.
    Used to maximize knowledge of crystal structure distribution as one of the parallel goals.

pith-pipeline@v0.9.0 · 5842 in / 1434 out tokens · 52346 ms · 2026-05-20T09:51:16.544292+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

8 extracted references · 8 canonical work pages

  1. [1]

    dueling robots

    Introduction Autonomous experimentation (AE) systems make informed decisions at each iteration based on data collected from automated instruments and learn from acquired data and explore previously hidden physical insights. This paradigm is revolutionizing materials science by fundamentally reshaping the pace and efficiency of optimization and discovery p...

    work page 2000

  2. [2]

    Ament, S. et al. Autonomous materials synthesis via hierarchical active learning of nonequilibrium phase diagrams. Science Advances 7, eabg4930 (2021). 8. Gregoire, J. M., Zhou, L. & Haber, J. A. Combinatorial synthesis for AI-driven materials discovery. Nat. Synth 2, 493–504 (2023). 9. Takeuchi, I., Lauterbach, J. & Fasolka, M. J. Combinatorial materials...

    work page arXiv 2021

  3. [3]

    Rahmanian, F . et al. Enabling Modular Autonomous Feedback-Loops in Materials Science through Hierarchical Experimental Laboratory Automation and Orchestration. Advanced Materials Interfaces 9, 2101987 (2022). 17. Guevarra, D. et al. Orchestrating nimble experiments across interconnected labs. Digital Discovery 2, 1806–1812 (2023). 18. Pendleton, I. M. et...

    work page doi:10.48550/arxiv.2311.06228 2022

  4. [4]

    Slautin, B

    N. Slautin, B. et al. Co-orchestration of multiple instruments to uncover structure–property relationships in combinatorial libraries. Digital Discovery 3, 1602–1611 (2024). 26. Kusne, A. G. et al. On-the-fly closed-loop materials discovery via Bayesian active learning. Nat Commun 11, 5966 (2020). 27. Xiang, X.-D. et al. A Combinatorial Approach to Materia...

    work page 2024

  5. [5]

    Suram, S. K. et al. Automated Phase Mapping with AgileFD and its Application to Light Absorber Discovery in the V–Mn–Nb Oxide System. ACS Comb. Sci. 19, 37–46 (2017). 35. Chen, D. et al. Automating crystal-structure phase mapping by combining deep learning with constraint reasoning. Nat Mach Intell 3, 812–822 (2021). 36. MaUettone, P . M., Daly, A. C. & O...

    work page doi:10.1002/adma.202102935 2017

  6. [6]

    & C h o n g , T.-C

    Song, W.-D., Shi, L.-P. & C h o n g , T.-C. Magnetic Properties and Phase Change Features in Fe-Doped Ge–Sb–Te. Journal of Nanoscience and Nanotechnology 11, 2648–2651 (2011). 43. Ding, D. et al. Origin of ferromagnetism and the design principle in phase-change magnetic materials. Phys. Rev. B 84, 214416 (2011). 44. Liu, J. Toward flexible memory applicati...

    work page 2011

  7. [7]

    Forrester, C. R. et al. Structural and magnetic properties of molecular beam epitaxy (MnSb2Te4)x(Sb2Te3)1−x topological materials with exceedingly high Curie temperature. APL Materials 12, 071109 (2024). 51. Liu, Y . et al. Site Mixing for Engineering Magnetic Topological Insulators. Phys. Rev. X 11, 021033 (2021). 52. Saxena, A. & Awana, V . P . S. Growt...

    work page doi:10.48550/arxiv.2403.10910 2024

  8. [8]

    V ., Smola, A

    Le, Q. V ., Smola, A. J. & Canu, S. Heteroscedastic Gaussian process regression. in Proceedings of the 22nd international conference on Machine learning - ICML ’05 489–496 (ACM Press, Bonn, Germany, 2005). doi:10.1145/1102351.1102413. 60. Zhou, F . et al. Heterogeneous Multi-Task Gaussian Cox Processes. Preprint at https://doi.org/10.48550/arXiv.2308.1536...

    work page doi:10.1145/1102351.1102413 2005