New AI Modeling Approach Accelerates the Development of Advanced Materials

A research team at the Department of Energy’s Lawrence Berkeley National Laboratory (Berkeley Lab) has successfully demonstrated a powerful AI modeling approach that accurately and rapidly predicts how reactions between solid materials unfold over time. It is the first-ever predictive model that accounts for how atoms travel through materials during solid-state reactions. Importantly, its predictions provide practical insights into the best recipes for making advanced materials.

“Our new model enables material scientists and industry stakeholders to make promising new materials dramatically faster — and with higher purity and yield,” said Kristin Persson, one of the study’s authors. Persson is a senior scientist at Berkeley Lab and a professor in materials science and engineering at the University of California, Berkeley.

Persson added: “The model can help accelerate the advancement of solid materials to cost-effective manufacturing and commercialization. It closes the gap between material discovery and new technologies that benefit society.”

“The model can help accelerate the advancement of solid materials to cost-effective manufacturing and commercialization. It closes the gap between material discovery and new technologies that benefit society.”
– Kristin Persson

The research was published in Nature Materials.

Synthesis: a major bottleneck in material discovery

Technology innovation often relies on the availability of inorganic solid materials that perform valuable functions, such as storing energy, emitting light, and catalyzing chemical reactions. Discovering better materials can be a critical step toward commercialization of a broad range of next-generation technologies, such as batteries, sensors, and medical devices. 

Thanks to advances in AI-driven computational tools, scientists can identify materials with desirable properties for various technological applications. Yet, synthesizing these materials can be extremely difficult. Mixing and heating powders to high temperatures — a common approach to synthesize inorganic solids — often yields a mixture of unexpected compounds rather than the desired product. It can take weeks to even years of trial-and-error experimentation to figure out the right recipe to make these materials.

How models can help

Computational models can potentially address this synthesis challenge. Existing models predict the outcome of synthesis reactions based on thermodynamics. In other words, they account for which reactions are the most energetically favorable and yield the most stable products.

“Thermodynamics essentially refers to how much two solids ‘want’ to react with each other,” said Persson.

But thermodynamic-based approaches have struggled to make accurate predictions. That’s because they have not accounted for how atoms travel through materials as reactions proceed. This factor — known as kinetics — can play a key role in the outcome of solid-state reactions.

“Atoms in solids move more slowly than in liquids, making it harder for them to reach reaction sites,” said Persson. “Even if two materials want to react with each other, they may not ultimately do so if the right amount of atoms cannot move to the places where reactions occur.”

Because of kinetic factors, solid-state reactions often require heating the starting powders to temperatures as high as 800°C. This makes the atoms more mobile.

How the new model works

In addition to considering thermodynamics, the research team’s approach incorporates an innovative machine learning model trained to predict how quickly atoms travel through a material. The model’s design is based on the researchers’ hypothesis that the reaction interface between two solids is highly disordered.

“Imagine the chaotic scene when a big concert is over, and crowds of people are exiting the arena,” said Persson. “That’s how atoms are at reaction sites. For reactions to occur, atoms need to move through extremely disordered regions.”

Inputs into the model include the starting materials, their ratios, and the temperature ramp-up. In just minutes, the model simulates the full reaction pathway from start to finish, revealing both intermediate and final products as well as impurities.

Strong agreement with experimental data

The researchers tested the model on barium-titanium oxides. This is a family of well-studied solid materials with important technological applications in electronics.

“We deliberately chose these materials because the reactions are strongly influenced by kinetics,” said Persson. “When you heat mixtures of barium- and titanium-containing powders, the new solid compounds that form are extremely close in thermodynamic stability. They all want to form to the same degree, so the reaction outcomes are determined by how easily the atoms can travel to the reaction interface.”

The model simulated reaction pathways of various ratios of barium oxide and titanium dioxide over time and at different temperatures. The researchers then compared the results with decades of experimental synthesis data from the scientific literature and found strong agreement.

“Remarkably, the model was correct for the full sequence of events, including intermediate compounds, final products, and formation of impurities,” said Persson.

Taking the model to the next level

The researchers trained the machine learning model specifically to predict the kinetics of barium-titanium oxides. The next research step is to train and demonstrate the model on other classes of solid-state materials.

Eventually, the team plans to use large kinetics datasets to train a foundation model that users can apply to virtually any solid-state materials. This would make the model broadly relevant across numerous technologies and industries.

This research was supported by the Department of Energy’s Office of Science.

###

Lawrence Berkeley National Laboratory (Berkeley Lab) is committed to groundbreaking research focused on discovery science and solutions for abundant and reliable energy supplies. The lab’s expertise spans materials, chemistry, physics, biology, earth and environmental science, mathematics, and computing. Researchers from around the world rely on the lab’s world-class scientific facilities for their own pioneering research. Founded in 1931 on the belief that the biggest problems are best addressed by teams, Berkeley Lab and its scientists have been recognized with 17 Nobel Prizes. Berkeley Lab is a multiprogram national laboratory managed by the University of California for the U.S. Department of Energy’s Office of Science.

DOE’s Office of Science is the single largest supporter of basic research in the physical sciences in the United States, and is working to address some of the most pressing challenges of our time. For more information, please visit energy.gov/science.