Autonomous discovery of solid electrolytes using large AI models ✛ ✦
Context
The development of all-solid-state batteries is held back by the challenge of simultaneously optimizing ionic conductivity, electrochemical stability, and mechanical compatibility of solid electrolytes. Traditional trial-and-error methods can't keep up with exploring the vast composition space, and scattered data hinder pattern discovery. Systems based on machine learning and numerical simulation offer a path to targeted discovery by mining knowledge from literature and automating the 'prediction-synthesis-analysis' loop.
Methods
The approach hinges on integrating several components. ML potentials (MLIPs) are trained on quantum mechanical calculations (DFT) to simulate ion transport at time and length scales unreachable by ab initio methods. Impedance spectroscopy, rooted in electromagnetic principles, and other experimental methods feed dynamically updated databases. The Boltzmann distribution of energy probabilities describes thermally activated ion hops, while quantum-wave effects are captured in density functional theory. Entropy from disorder in sublattices can boost conductivity, as shown for argyrodites, and hydride electrolytes reveal anion dynamics. Plasma spraying and other synthesis methods are also being optimized with AI. The closed loop includes candidate generation, multiscale modeling, uncertainty-aware selection, and robotic validation.
Results
The DigBat platform has compiled 3,736 experimental records with temperature-dependent conductivity. MLIP simulations revealed a non-monotonic dependence of conductivity in Li₆₋ₓPS₅₋ₓCl₁₊ₓ argyrodites on disorder degree, peaking at 37.5–50% site inversion. Screening 32 million candidates with ML and cloud computing led to identifying NaₓLi₃₋ₓYCl₆ halide electrolytes with ionic conductivity up to 10⁻³ S/cm at 100°C. Impedance measurements confirmed that the metal-organic framework NOTT-400, predicted through LLM-based literature analysis, has a conductivity of 3.1×10⁻⁴ S/cm at room temperature. The robotic platform can synthesize and test over 100 samples per day, speeding up iterations.
Implications
The shift to autonomous discovery changes the paradigm: solid electrolytes are no longer the result of chance discoveries but become products of systematic design. Integrating ML models and simulations reveals hidden correlations, such as the role of correlated hopping topology. This narrows the gap between prediction and practical implementation, while reducing reliance on intuition and expensive experiments.
Future development
Future systems will rely on data standardization following FAIR principles, including measurement context and negative results. Advances in interface-focused AI will enable modeling not only bulk transport but also interfacial degradation, critical for battery longevity. Discovering electrolytes beyond lithium systems (sodium, magnesium, proton) will broaden the raw material base, and high-entropy compositions may offer new property trade-offs.
Impact
The methodology will impact materials science broadly, accelerating the development of functional materials for energy, catalysis, and electronics. Automated feedback-driven labs will become a tool for validating physical models in complex multicomponent systems.
Next steps
Immediate next steps include establishing community-wide ontologies for describing the synthesis and properties of solid electrolytes, and developing generative models that can predict fundamentally new structures with desired transport properties and manufacturability.
Key open problems
This work directly tackles the problem of predicting material synthesizability—one of the key unsolved challenges in solid-state chemistry. The complexity of interfacial chemistry in solid-state batteries, including dendrite formation and space-charge layers, remains a challenge for current modeling methods and demands new approaches to collecting and analyzing experimental data.
The knowledge machine's view
This section is addressed to the author of the paper and their colleagues. Neighbouring works were found by meaning-based search across our archive, not by word overlap. We do not review: we show the field around the work.
What the work does
The paper presents a platform for autonomous discovery of solid electrolytes, combining ML potentials, DFT, impedance spectroscopy, LLM-based literature analysis, and robotic validation. Using argyrodites and halides as examples, it demonstrates the identification of a non-monotonic dependence of conductivity on disorder and the screening of 32 million candidates with experimental confirmation.
Where its strength is
A key strength is the integration of heterogeneous data and methods into a single loop, allowing microscopic characteristics (disorder, hopping topology) to be linked to macroscopic conductivity and predictions to be rapidly tested experimentally.
Where it could go next
- It may be interesting to see how the work on warm-starting quantum calculations uses pre-training of ML models on small data to accelerate DFT—this could help reduce computational costs when modeling ionic transport.based on: Warm Start for a Quantum Engine: How Machine Learning Accelerates Molecular Calculations arXiv:2607.09304
- A related approach trains a neural network on the errors of quantum chemistry to improve prediction accuracy; adapting such a feedback mechanism could enhance the reliability of MLIPs in describing rare events, such as ion jumps over high barriers.based on: How a Neural Network Learns from Quantum Chemistry's Mistakes arXiv:2604.24607
- There is a paper in the archive showing that for solid-state batteries, the secret lies not in the volume of data but in its quality; it might be worth looking at strategies for selecting informative experiments to make even better use of the robotic platform.based on: Solid-State Batteries: The Secret Isn't in Data Volume arXiv:2603.07425
Worth trying
- Test transfer learning from ML potentials pre-trained on general datasets to specific electrolyte systems to reduce the number of DFT calculations.
- Incorporate uncertainty quantification of predictions into the active learning loop so that the robot synthesizes the most informative compositions.
- Use LLMs to automatically extract not only compositions but also synthesis and measurement conditions from the literature to enrich the DigBat database.
Why this may matter
The development of such autonomous platforms could dramatically accelerate the search for materials for all-solid-state batteries, shortening the path from idea to device and opening access to compositions that are difficult to find with traditional methods.
Nearby works in our archive
- How Clever Sampling Teaches Metal to Predict Strengthsimilarity 0.66
- Warm Start for a Quantum Engine: How Machine Learning Accelerates Molecular Calculationssimilarity 0.65
- Solid-State Batteries: The Secret Isn't in Data Volumesimilarity 0.65
- A Conductor for Batteries: How Old Batteries Got a New Soundsimilarity 0.64
- A Quantum Compass for Your Batterysimilarity 0.63
- How a Neural Network Learns from Quantum Chemistry's Mistakessimilarity 0.63
The closest work in the archive sits at 0.66 by meaning — the usual level for our corpus.
Original abstract
as written by the authors arXiv:2606.24480
Solid electrolytes (SEs) are central to next-generation metal batteries, yet their discovery remains constrained by fragmented data, limited transferability of simulations, and slow experimental iteration. Unlike catalysis, where surface reactivity dominates, SEs require simultaneous optimization of bulk ion transport, defect chemistry, mechanical integrity, and interfacial stability. Here, we outline a framework for autonomous SE discovery enabled by large artificial intelligence (AI) models, including machine learning interatomic potentials (MLIPs) and large language models (LLMs). We discuss the evolution from static materials databases to dynamic, self-updating knowledge systems, the role of MLIPs in bridging density functional theory (DFT) and long-timescale ion migration, and the emergence of LLMs as engines for literature mining, hypothesis generation, and scientific reasoning. We further describe a closed-loop architecture integrating AI-driven candidate design, multiscale simulation, uncertainty-aware selection, and experimental validation. Such systems shift SE research from intuition-guided exploration to data-informed, self-improving cycles. We conclude by highlighting challenges in data standardization, interfacial complexity, and reproducibility, and we propose design principles for building autonomous laboratories for solid-state battery materials.
If all possible element combinations for solid electrolytes were tested manually, one sample per day, it would take billions of years to go through them all. Machine learning algorithms shrink that to months.
Key numbers
- number of experimental solid electrolytes in the DigBat database: 3736
- conductivity of NOTT-400 at room temperature: 3.1×10⁻⁴ S/cm
- number of candidates after initial screening out of 32 million: about 583
- throughput of the robotic platform: >100 samples/day
- synthesis and measurement temperature for optimized Li₇SiPS₈: below 500°C (down from >600°C)