Electrolytes used in batteries are far from simple compounds; they are carefully balanced mixtures of salts, solvents, and additives that constantly interact and influence one another. While AI has already proven useful in narrowing down suitable individual materials for these systems, researchers are now pushing the technology further.

A team from the University of Chicago Pritzker School of Molecular Engineering has developed an AI approach capable of generating entire electrolyte formulations rather than just selecting components. The work, published in JACS Au, represents an advancement of the Amanchukwu Lab’s ongoing AI platform for battery research, known as ElectrolyteGPT. 

According to first author Jaemin Kim, next-gen battery electrolytes must satisfy multiple and often conflicting performance requirements. The model’s ability to operate under varied conditions allows it to design new electrolyte candidates that simultaneously meet these demanding property targets.

AI design boosts battery electrolyte performance benchmarks 

Rather than simply selecting which materials should be included, the AI determines the full formulation details – covering concentrations, mixing ratios, and other key parameters of the electrolyte blend. In doing so, it works toward predefined performance targets spanning conductivity, stability, viscosity, and related properties.

When the researchers synthesized and tested the AI-generated recommendations, they identified several new electrolyte compositions that matched the performance of state-of-the-art lithium metal battery systems. According to corresponding author, Professor Chibueze Amanchukwu, this represents a meaningful step toward the broader objective of discovering electrolytes that can surpass today’s leading benchmarks.

The researchers reported that several AI-generated electrolyte compositions achieved performance comparable to state-of-the-art systems, reinforcing confidence in the model’s ability to replicate expert-level design outcomes. While the results are promising, they also underline that further refinement and exploration are still needed before the approach can consistently surpass existing benchmarks.

The number of possible molecules for battery electrolytes is estimated at around 10⁶⁰ – more than the number of stars in the observable universe. Testing each one for use in batteries, cancer treatments, or other advanced materials is far beyond a human lifetime. And that figure only covers individual molecules, not the near-infinite number of ways they can be combined into different formulations and mixtures.

Discovery expands possibilities for battery material design

According to Kim, it is not feasible to exhaustively explore the near-infinite space of possible electrolyte chemistries. However, generative AI can help navigate these uncharted regions of chemical space and propose entirely new molecules that may never have been synthesized before.

The system is capable of producing theoretical candidates at a pace far beyond human capability, selecting those it predicts – based on learned patterns from training data – could be suitable for specific applications. These AI-generated suggestions are then validated in the lab, where researchers test them using the same experimental procedures applied to materials designed by human scientists.

AI is widely used in drug discovery, which initially posed a challenge for Amanchukwu’s team, as most existing GPT models are trained on data that favors drug-like molecules, and not those suited for battery applications.

According to Amanchukwu, using standard literature-based datasets tends to produce irrelevant, drug-like outputs. To address this, the team built a curated dataset focused specifically on electrolyte-relevant compounds, effectively narrowing the model’s knowledge base. As a result, when prompted to generate new solvent molecules, the system produces candidates that resemble viable battery electrolytes rather than pharmaceutical compounds.

Recommended Articles