When we think of batteries, we usually picture lifeless boxes of metal and chemicals that power our phones, laptops, or electric cars. Yet, in many ways, batteries are more like living systems than we realize. They breathe in energy and release it when needed. They grow weaker with age. They can be pushed to exhaustion or nurtured into long and productive lives. Like humans, they sometimes need medicine to stay healthy.
For a battery, that “medicine” comes in the form of electrolyte additives. These are special molecules mixed into the liquid that ferries ions between a battery’s electrodes. Though they might sound like a minor ingredient, electrolyte additives play an outsized role in determining how long a battery lasts, how much energy it can store, and how efficiently it performs. They are the difference between a battery that quickly deteriorates and one that runs reliably for years.
But just as no doctor can prescribe the same medicine for every patient, no single additive works for every battery. Finding the right “treatment” requires careful testing, creativity, and now, thanks to modern advances, artificial intelligence.
The Prescription Challenge
The search for the perfect electrolyte additive is not simple. Hundreds of potential molecules exist, each with different shapes, chemical properties, and possible interactions inside a battery. Traditionally, scientists have had to test them one at a time in painstaking experiments. This is like a physician prescribing drugs blindly, without knowing whether the patient will improve—or suffer side effects.
Testing each possibility is slow and expensive. Every experiment requires building a battery cell, cycling it through charge and discharge, measuring resistance, energy output, and long-term stability. Doing this for hundreds of candidates could take years.
This bottleneck has limited how quickly new and better batteries can be developed, especially for advanced chemistries designed to power the next generation of electric vehicles, renewable energy storage, and portable electronics.
Enter Machine Learning
At Argonne National Laboratory, researchers are rewriting this story. They are using machine learning—a branch of artificial intelligence—to predict which additives will most likely improve battery performance. Instead of relying solely on trial and error, they are training computers to recognize patterns in chemical data and make informed predictions.
The process begins with a modest dataset. Rather than testing hundreds of additives, the scientists started with just 28, chosen for their diversity of chemical features. Each additive was tested in real batteries to measure performance: How much resistance did it introduce? How much energy did it allow the battery to store? Did it create a stable protective layer on the electrodes?
These experimental results became the “training data” for a machine learning model. The computer learned to connect the molecular structures of additives—their shapes, sizes, and chemical groups—with the observed effects inside the battery. Once trained, the model could look at entirely new molecules and predict their likely impact.
Batteries on the Operating Table
The need for better additives is particularly urgent for high-voltage batteries, such as those based on lithium nickel manganese oxide, or LNMO. These so-called “5-volt batteries” promise major advantages over conventional designs. They can store more energy, charge faster, and eliminate cobalt—a metal whose supply chain raises both environmental and ethical concerns.
But pushing a battery to such high voltages is like putting the system under extreme stress. The electrolyte, which is normally stable around 4 volts, begins to break down. The energized cathode also starts to corrode, leading to rapid capacity loss. It is as if the battery’s organs are overheating under pressure.
Here, the “medicine” of electrolyte additives becomes critical. The right additive will decompose early in the battery’s life, sacrificing itself to form a protective interface on the electrodes. This invisible coating acts like a shield, lowering resistance, preventing further damage, and allowing the battery to operate smoothly even at high voltage.
Data-Driven Discovery
By combining machine learning predictions with carefully chosen experiments, the Argonne team created a powerful cycle of discovery. The model identified 125 possible new additive combinations. Testing each one traditionally would have taken months and vast resources. Instead, the researchers focused only on the most promising predictions.
The results were striking. Several new additive combinations significantly outperformed the original ones, reducing resistance and extending energy capacity. These discoveries validated the model’s power: with just a small training set, it could forecast chemical behavior with remarkable accuracy.
Just as importantly, the study demonstrated that effective machine learning does not always require massive datasets. With carefully selected, high-quality data, researchers can build accurate models that accelerate discovery in ways previously thought impossible.
A New Way of Doing Science
This marriage of chemistry and computation marks a turning point in how materials science is done. Traditionally, progress has been driven by intuition, long experiments, and incremental improvements. Now, artificial intelligence is reshaping the landscape. By quickly sifting through vast chemical possibilities, machine learning points scientists toward the most fruitful directions, saving time, money, and effort.
It is not that the computer replaces the human scientist. Rather, it acts as a guide, highlighting promising paths that researchers can then explore in the laboratory. This synergy of prediction and experiment is what makes the approach so powerful. It is like having a physician who not only diagnoses but also consults a database of millions of past cases to recommend the most effective treatment.
Toward the Future of Energy
The implications of this work go far beyond one type of battery. As society transitions toward clean energy and electrified transportation, demand for better batteries is skyrocketing. We need systems that charge quickly, last for decades, and are made from abundant, sustainable materials. Meeting this demand will require innovation at every level, from electrode materials to electrolytes to the additives that fine-tune performance.
Machine learning offers a way to accelerate this innovation. It can be applied not only to electrolytes but also to cathode design, anode stability, and even manufacturing processes. The lessons learned in this study—using small, high-quality datasets to make powerful predictions—may soon reshape the entire field of energy storage.
Batteries with a Human Touch
In the end, the metaphor of batteries needing “medicine” is more than a clever analogy. It captures the essence of how we now approach energy technology: not as a lifeless engineering problem but as a living system requiring care, protection, and thoughtful treatment.
Electrolyte additives are the prescriptions that keep these systems healthy. Machine learning is the diagnostic tool that helps us find the right treatment faster than ever before. And together, they are paving the way for a future where batteries are not just more powerful but also more reliable, sustainable, and affordable.
The story of batteries is the story of modern energy. Just as medicine has extended human lifespans, the “medicine” of additives and machine learning may extend the lifespans of the technologies we rely on every day—bringing us closer to a world powered by clean, efficient, and enduring energy.
More information: Bingning Wang et al, Data-driven design of electrolyte additives supporting high-performance 5 V LiNi0.5Mn1.5O4 positive electrodes, Nature Communications (2025). DOI: 10.1038/s41467-025-57961-w
