The Patent Landscape: 60+ Filings Across Five Jurisdictions

Over 60 patent records and technical disclosures filed between 2003 and 2026 document the accelerating application of artificial intelligence to lithium battery electrolyte formulation, spanning South Korea, Japan, China, the United States, and Europe. The dominant technical theme across this dataset is the displacement of traditional trial-and-error laboratory approaches by machine learning, neural networks, Bayesian optimization, and large language models (LLMs) — each targeting a distinct bottleneck in the formulation discovery pipeline.

The most strategically significant cluster within this dataset addresses AI-driven electrolyte formulation specifically, centred on predictive composition modelling, iterative Bayesian optimization loops, LLM-driven recipe generation platforms, and AI model-assisted viscosity and permeability analysis of electrolyte materials. A secondary but substantial cluster addresses AI for battery diagnostics, state-of-health estimation, and lifetime prediction — providing an adjacent technology context that informs formulation decisions at the cell level.

Among the most frequently appearing assignees are LG Chem, LG Energy Solution, and LG Management Development Institute; Samsung SDI; Asahi Kasei Corporation; Chungnam National University; Soochow University; Automat Solutions, Inc.; POSCO Holdings; and Korea Electronics Technology Institute. The geographic distribution of filings — with significant activity in South Korea, China, Japan, and the United States — reflects the global competitive intensity of next-generation battery development, as tracked by organisations including WIPO in its annual innovation trend reports.

AI-Driven Predictive Modelling for Electrolyte Composition

AI-driven composition-to-performance prediction eliminates the need for iterative experimental synthesis as the primary screening step: Asahi Kasei Corporation’s 2022 patent discloses a method that accepts electrolyte composition information — including the names and identities of each component — and generates two or more predicted values covering properties such as combustibility, maximum output, capacity, and internal resistance, all in a single inference pass. The same inventive concept was extended in a 2025 continuation filing, confirming the commercial importance Asahi Kasei places on multi-property simultaneous prediction.

Korea Electronics Technology Institute has extended this paradigm to the structural domain, disclosing an AI model-based computing device that inputs a tomographic image of calendared battery material into a two-stage AI pipeline: the first model analyses pore distribution and curvature, while the second derives viscosity and permeability of the electrolyte. This effectively bridges physical microstructure characterisation with electrochemical property prediction — linking calendaring-induced pore geometry directly to electrolyte transport properties.

LG Management Development Institute has filed a family of patents across Korea, the US, and Europe covering dual AI models for material feature extraction. The European patent discloses a system with a first AI model trained to output compositional feature data and a second AI model trained to output structural feature data, where the two models are mutually reinforced through cross-training. The composition model learns from structural outputs and vice versa — a co-training architecture designed to capture the complex, bidirectional relationship between molecular composition and material architecture. This is particularly valuable when electrolyte additives alter the solid-electrolyte interphase (SEI) structure in ways not predictable from composition alone.

Bayesian Optimization and Closed-Loop Electrolyte Formulation

Bayesian optimization (BO) has emerged as the dominant AI strategy for closed-loop electrolyte formulation, enabling efficient navigation of high-dimensional composition spaces with minimal experimental iterations. Chungnam National University has filed parallel patent families in both Korea and the United States covering a systematic BO-driven electrolyte optimization framework: the workflow begins with a first dataset linking formulations to measured experimental data, uses BO to propose a second formulation, updates the dataset with new measurements, and iterates until a termination condition is satisfied.

“Bayesian optimization uses the uncertainty estimates of the surrogate model to balance exploration of unknown composition regions and exploitation near known optima — achieving optimal or near-optimal formulations with far fewer experiments than exhaustive methods.”

Critically, the measured data in the Chungnam National University framework includes both first data from the manufactured electrolyte and second data from battery cells incorporating that electrolyte — meaning the loop captures both intrinsic electrolyte properties and in-cell performance simultaneously. The US counterpart filing expands the claims to explicitly cover the optimization system, the resulting electrolyte, and the battery itself, indicating a strategy to protect the entire value chain from algorithm to product.

The closed-loop BO concept is also demonstrated in zinc secondary battery electrolyte research, which provides a technically analogous template. A patent filed by Ewha Womans University constructs a surrogate model of the unknown function relating additive composition ratios to an objective function, updates the surrogate with new experimental results through a closed-loop Bayesian optimization network, and iterates to maximise performance. Though targeting zinc batteries, the architectural principle — surrogate model plus acquisition function plus automated experimental feedback — is directly transferable to lithium battery electrolyte systems, illustrating the broad cross-chemistry adoption of BO as the standard AI method for electrolyte optimization.

The efficiency advantage of BO over grid search or random search is well-established in the scientific literature, including work published by Nature on autonomous materials discovery: BO uses the uncertainty estimates of the surrogate model to balance exploration (testing unknown composition regions) and exploitation (refining near known optima), thereby achieving optimal or near-optimal formulations with far fewer experiments than exhaustive methods. When combined with high-throughput experimental platforms, this capability directly addresses the stated limitation of traditional electrolyte R&D — excessive time and cost — as explicitly recognised in the Chungnam National University disclosures.

Large Language Models and Robotics-Driven Autonomous Formulation Platforms

Large language models are being integrated as the human-machine interface layer for electrolyte formulation platforms, resolving the deployment and operability barriers of earlier narrow AI models. Soochow University has filed two closely related Chinese patents describing a comprehensive lithium battery electrolyte formulation prediction platform based on large language models, with a dual-LLM architecture comprising five modules: a data acquisition module for target electrolyte property parameters; a formulation prediction module using a first LLM to output candidate electrolyte recipes and theoretical viscosity values; a trajectory simulation module that pre-builds a battery model and simulates electrochemical reaction kinetics; a formulation properties calculation module computing system-level physicochemical properties; and a formulation output module using a second LLM to generate an optimised final recipe based on all prior outputs.

At the fully autonomous extreme, Automat Solutions, Inc. has patented a robotics-integrated AI system that closes the design-make-test loop without human intervention. Their 2021 patent and its 2025 continuation describe a system that predicts an objective function from a recipe using a machine learning model, generates multiple proposed battery material recipes, deploys a robotic preparation module to physically prepare and deposit each recipe into an electrochemical module, executes formulation characteristic tests via a robotic testing module, and feeds test results back to update the machine learning model in a continuous loop. This architecture eliminates the bottleneck of manual laboratory throughput and enables the machine learning model to learn from physically validated experimental data rather than literature data alone.

Machine learning is also being applied to inorganic battery material synthesis optimization beyond liquid electrolytes. A 2025 patent from Mitra Future Technologies utilises machine learning to define synthesis conditions and stoichiometry of lithium metal polyanion compounds, achieving cathode gravimetric capacities exceeding 170 mAh/g — illustrating that the same AI-for-materials paradigm extends from liquid electrolyte formulation to solid electrode compound design. This convergence of AI methods across both electrolyte and electrode domains is consistent with the broader trend in materials informatics tracked by institutions such as NIST and the OECD in their materials data frameworks.

Electrochemical Diagnostics and Ion Distribution Prediction

Supporting the electrolyte formulation pipeline, a separate class of AI methods addresses electrochemical characterisation and compound diagnostics that generate the training data needed by formulation models. LG Chem has patented an apparatus for diagnosing compounds that acquires XRD data of a lithium compound, calculates parameters related to compound state, inputs these into an AI model to extract main performance-correlated parameters, and uses these to diagnose compound performance. Identifying which structural parameters most strongly correlate with electrochemical performance is foundational to electrolyte design: a model that can interpret XRD data to extract performance-relevant structural fingerprints could dramatically reduce the analytical cost of screening candidate electrolyte salts or additives.

Ion transport behaviour inside battery systems has been targeted by AI prediction methods from Yonsei University. Their 2025 patent discloses a two-step system: first, boundary and initial conditions are fed into a numerical analysis model to generate ion distribution image data; second, an LSTM-based AI block processes the initial battery state and the ion distribution images to output a predicted ion distribution profile. For electrolyte researchers, accurate ion distribution prediction enables in silico testing of how a given formulation will affect lithium-ion transport homogeneity across the separator and electrode interfaces — a property directly linked to rate capability, cycling stability, and lithium plating risk.

Electrode characterisation using EIS data has been addressed by Yonsei University in a separate 2024 disclosure demonstrating that an artificial neural network can extract multiple electrode characteristics — including those normally requiring SEM — from EIS data alone. Since EIS is a rapid, non-destructive measurement, integrating this AI capability into the formulation screening pipeline enables high-throughput electrode characterisation of cells made with different electrolytes, generating richer training datasets at lower cost. This approach aligns with the high-throughput experimentation frameworks increasingly advocated by IEA in its battery technology roadmaps for accelerating energy storage deployment.

Key Players and the IP Landscape: Who Holds What

Analysis of the patent dataset reveals a clear stratification of players by scope and technical depth of AI-for-electrolyte innovation. LG Group — spanning LG Chem, LG Energy Solution, and LG Management Development Institute — is the most prolific assignee in the dataset, with filings spanning battery management AI, compound diagnostics, long-term characteristic estimation via artificial neural networks (with patents dating to 2009 on the system and method for estimating batteries’ long-term characteristics based on artificial neural networks), and dual-model material composition feature extraction. Their multi-jurisdictional coverage across Korea, Europe, and the US of the composition-structure dual AI model system signals broad IP protection intent.

Asahi Kasei Corporation has staked a distinctive position specifically on electrolyte composition-to-performance prediction, with two JP patents focused exclusively on this problem — making them the most focused assignee on direct AI electrolyte formulation prediction in the dataset. Chungnam National University is the leading academic innovator on Bayesian optimization for lithium battery electrolyte, with parallel KR and US filings covering the full BO loop, suggesting commercial licensing intent. The US counterpart filing’s explicit claim coverage of the optimization system, the resulting electrolyte, and the battery itself represents a comprehensive value-chain IP strategy.

Soochow University leads the LLM-for-electrolyte space with two active CN patents on the dual-LLM electrolyte prediction platform, representing the most architecturally sophisticated natural language processing application to electrolyte formulation in the dataset. Automat Solutions, Inc. holds the broadest claim scope for fully autonomous AI-robotics electrolyte and battery material optimization, with both 2021 and 2025 active US filings covering the closed-loop robotic testing and machine learning update system. Samsung SDI contributes AI methods for battery cell performance prediction and state-of-health estimation, including a 2026 pending patent applying machine learning models to design factor inputs to generate visual performance representations. Korea Electronics Technology Institute occupies a unique position with its AI model for deriving electrolyte viscosity and permeability from tomographic images, bridging physical characterisation and computational property prediction.

The geographic distribution of IP activity — concentrated in South Korea, China, Japan, and the United States — reflects the competitive dynamics of next-generation battery development that are closely monitored by the EPO in its annual patent index for clean energy technologies. For R&D teams and IP professionals seeking to navigate this landscape, PatSnap’s innovation intelligence platform provides structured access to the full patent corpus with AI-assisted analysis tools. Understanding the freedom-to-operate implications of these overlapping claim scopes — particularly the intersection of Chungnam National University’s BO method claims with Automat Solutions’ system claims — requires detailed patent analysis tools such as those available through PatSnap’s IP management solutions.