Scientists have developed an artificial intelligence system that can create complete battery electrolyte recipes, a breakthrough that could speed up the development of better batteries for electric vehicles, electronics, and renewable energy storage.
The new tool, called ElectrolyteGPT, was developed by researchers at the University of Chicago Pritzker School of Molecular Engineering.
Unlike many AI systems that focus on finding individual chemicals, ElectrolyteGPT can design an entire electrolyte formulation from scratch.
Electrolytes are a critical part of batteries.
They allow charged particles to move between the battery’s electrodes, enabling the battery to store and release energy. However, electrolytes are not made from a single ingredient.
They are complex mixtures of salts, solvents, and additives that must work together while balancing many competing properties.
For example, an ideal electrolyte needs to conduct electricity efficiently, remain stable under high voltages, flow easily, and support long battery life.
Improving one property can often make another worse, making electrolyte design a difficult challenge.
The researchers trained ElectrolyteGPT to handle these trade-offs. Instead of selecting a few promising ingredients, the AI determines the full recipe, including ingredient types, concentrations, and mixing ratios.
After generating new formulations, the team synthesized several of them in the laboratory and tested their performance in lithium metal batteries.
The results were encouraging. Some of the AI-designed electrolytes performed as well as today’s best commercial and research-grade electrolyte formulations.
According to the researchers, this demonstrates that AI can successfully create battery recipes comparable to those developed by experienced scientists.
One reason this technology is needed is the enormous size of the chemical search space. Scientists estimate there may be around 10⁶⁰ possible molecules that could potentially be used in battery electrolytes. That number is greater than the estimated number of stars in the universe.
Even more challenging is the nearly endless number of ways those molecules can be combined into mixtures. Exploring all those possibilities through traditional laboratory research would be impossible.
To overcome this problem, the researchers trained their AI using a carefully selected database of electrolyte-related compounds. This prevented the system from generating molecules designed for other purposes, such as pharmaceuticals.
A major innovation in the project was the creation of a new chemical language called fLine. Traditional chemical notations describe individual molecules, but fLine can describe entire formulations.
It includes information such as solvent ratios, salt concentrations, temperature conditions, and other factors that influence battery performance.
This allows the AI to understand and generate complete electrolyte systems rather than isolated chemicals.
The researchers believe the technology could eventually be useful for designing many types of chemical mixtures beyond batteries. More importantly, it brings scientists closer to creating truly generative AI systems capable of inventing entirely new materials.
Although the current model still relies on limited data and requires experimental verification, the team has already shown that its suggestions can work in real batteries.
Their next goal is to make the system larger, smarter, and capable of discovering electrolyte formulations that outperform the current state of the art.
