Reviewed by Lexie CornerMay 6 2025
Researchers at the University of Chicago have introduced a new framework for identifying molecules that optimize three key characteristics of an ideal battery electrolyte: ionic conductivity, oxidative stability, and Coulombic efficiency.
Ritesh Kumar, an Eric and Wendy Schimdt AI in Science Postdoctoral Fellow, is the first author of a new paper that outlines how artificial intelligence and machine learning can help discover new, powerful electrolytes. Image Credit: The University of Chicago
Identifying new, high-performance electrolytes remains a major challenge in developing next-generation batteries for electric vehicles, consumer electronics, and grid-scale energy storage.
The most stable electrolytes are not always the most conductive, and the most efficient batteries are not always the most stable, highlighting the complex trade-offs involved.
The electrodes have to satisfy very different properties at the same time. They always conflict with each other.
Ritesh Kumar, Eric and Wendy Schimdt AI, Science Postdoctoral Fellow, University of Chicago Pritzker School of Molecular Engineering (UChicago PME)
By analyzing data from 250 research papers covering the history of lithium-ion battery research, the team used AI to calculate an “eScore” for different molecules. This score balances three key properties—ionic conductivity, oxidative stability, and Coulombic efficiency—to highlight top-performing candidates.
The champion molecule in one property is not the champion molecule in another.
Chibueze Amanchukwu, Principal Investigator and Neubauer Family Assistant Professor, Molecular Engineering, UChicago PME
The researchers validated their AI-driven method by identifying a molecule with performance comparable to today’s leading electrolytes. This achievement represents a significant step forward in a field that has traditionally relied on trial and error.
Electrolyte optimization is a slow and challenging process where researchers frequently resort to trial-and-error to balance competing properties in multi-component mixtures. These types of data-driven research frameworks are critical to help accelerate the development of new battery materials and to leverage advancements in AI-enabled science and laboratory automation.
Jeffrey Lopez, Assistant Professor, Chemical and Biological Engineering, Northwestern University
The Music of Batteries
Artificial intelligence is helping scientists streamline the search for better battery materials by identifying the most promising candidates for lab testing, saving time, energy, and resources. At the University of Chicago’s Pritzker School of Molecular Engineering (PME), researchers are already applying AI to speed up advances in cancer therapies, water purification, quantum materials, and more.
When it comes to battery research, the challenge is immense. The number of possible electrolyte molecules is estimated to be as high as 1060, which is far too many to explore through traditional methods. AI offers a way to narrow this vast field to the most likely candidates.
“It would have been impossible for us to go through hundreds of millions of compounds to say, ‘Oh, I think we should study this one,’” Amanchukwu noted.
Amanchukwu compares the use of AI in battery research to how streaming services recommend music. Think of each person’s music taste as their own “eScore.” The current AI can scan a playlist and predict which songs someone might like. The next step is building AI that can generate an entirely new playlist based on those preferences.
Ultimately, Amanchukwu’s lab is aiming for something even more advanced: AI that can not just select or predict, but actually design new molecules from scratch that meet all the required performance criteria, like composing new music instead of just recommending it.
To support this work, Amanchukwu received a Google Research Scholar Award last year to help move closer to that goal of truly generative electrolyte AI.
A Quirk of Graphic Design
The research team began manually assembling the AI's training data back in 2020.
“The current dataset has thousands of potential electrolytes, which we extracted from literature that spanned over 50 years of research,” said Kumar.
One challenge lies not in chemistry but in how scientific papers are designed. Much of the numerical data needed to calculate eScores is embedded within images—charts, diagrams, and illustrations saved as .jpeg or .png files rather than included in the main text.
Because most large language models used to train AI only process written text, the UChicago PME team still relies on manual data extraction and expects this to continue.
“Even the models today really struggle with extracting data from images,” said Amanchukwu.
Despite the size of the dataset, it represents only the first step in the larger effort.
“I don't want to find a molecule that was already in my training data. I want to look for molecules in very different chemical spaces. So we tested how well these models predict when they see a molecule that they've never seen before,” said Amanchukwu.
The team found that the AI performed well when predicting molecules similar to those it had already encountered. However, it was less effective when presented with unfamiliar compounds—a limitation they now aim to overcome in the next phase of their work.
Journal Reference:
RiteshKumar, R., et al. (2025) Electrolytomics: A Unified Big Data Approach for Electrolyte Design and Discovery. Chemistry of Materials. https://doi.org/10.1021/acs.chemmater.4c03196.