AI Model Predicts Solid-State Reactions to Speed Materials Synthesis

Breakthrough combines thermodynamics with kinetics modeling

Researchers at Lawrence Berkeley National Laboratory have developed an AI modeling framework that predicts the complete sequence of events in solid-state chemical reactions, including intermediate compounds, final products, and impurities. The approach addresses a major bottleneck in materials science: figuring out how to actually synthesize promising new materials once they've been identified.

The model runs simulations in minutes that reveal optimal synthesis recipes for advanced materials—work that traditionally requires weeks or years of laboratory trial and error. According to Kristin Persson, a senior scientist at Berkeley Lab and professor at UC Berkeley who co-authored the study, the framework "closes the gap between material discovery and new technologies that benefit society."

The research was published in Nature Materials, as first reported by Berkeley Lab.

Why it matters

Computational tools have become adept at identifying materials with desirable properties for batteries, sensors, and other technologies. But synthesis remains a stubborn practical barrier. Mixing and heating powders to create inorganic solids frequently produces unexpected compounds instead of target materials. This new model provides actionable guidance on temperature profiles, material ratios, and process conditions—information that can accelerate commercialization timelines and reduce development costs across multiple industries.

Accounting for how atoms actually move

Previous computational approaches relied solely on thermodynamics, predicting which reactions are energetically favorable and which products are most stable. That framework proved insufficient because it ignored kinetics—the rate at which atoms travel through solid materials to reach reaction sites.

"Atoms in solids move more slowly than in liquids, making it harder for them to reach reaction sites," Persson explained. Even when two materials are thermodynamically inclined to react, kinetic constraints can prevent the reaction if atoms cannot migrate to the necessary locations. Solid-state reactions often require temperatures as high as 800°C to increase atomic mobility.

The Berkeley Lab team's innovation was to incorporate a machine learning model trained to predict atomic diffusion rates. The model assumes that reaction interfaces between solids are highly disordered—"like the chaotic scene when a big concert is over, and crowds of people are exiting the arena," as Persson described it. Atoms must navigate these disordered regions for reactions to proceed.

Validation with barium-titanium oxides

The researchers tested their framework on barium-titanium oxides, a family of materials used in electronics. They deliberately selected these compounds because their reactions are strongly kinetics-driven. When barium oxide and titanium dioxide powders are heated together, multiple possible products have nearly identical thermodynamic stability, making kinetics the determining factor in outcomes.

The model simulated reaction pathways for various material ratios across different temperatures. When compared against decades of published experimental data, the predictions showed strong agreement across the full reaction sequence.

Path to broader application

The current machine learning model was trained specifically for barium-titanium oxide systems. The research team's next step is demonstrating the approach across other material classes. Their ultimate goal is a foundation model trained on large kinetics datasets that can handle virtually any solid-state material system, making the tool relevant across numerous technology sectors.

The work was supported by the Department of Energy's Office of Science. Details were first reported by Lawrence Berkeley National Laboratory.