New AI Modeling Approach Accelerates the Development of Advanced Materials

A research team at Lawrence Berkeley National Laboratory (Berkeley Lab) has developed an advanced AI modeling approach that accurately predicts how solid materials react over time. This novel model accounts for the movement of atoms within materials, a factor known as kinetics, which significantly impacts reaction outcomes. Kristin Persson, a senior scientist at Berkeley Lab and professor at UC Berkeley, explained that the model enables material scientists to create new materials faster, with higher purity, and at lower costs, accelerating their commercialization.

The study, published in Nature Materials, highlights the importance of discovering better inorganic solid materials for various technologies such as batteries, sensors, and medical devices. Traditional methods of synthesizing these materials often involve trial-and-error experimentation, taking weeks to years to find the right recipe.

However, the new AI model can simulate the entire reaction pathway, including intermediate compounds, final products, and impurities, in just minutes. It incorporates machine learning to predict the speed at which atoms move through materials at reaction sites, which are highly disordered regions. By considering both thermodynamics and kinetics, the model provides a more comprehensive understanding of solid-state reactions.

The researchers tested the model on barium-titanium oxides, a material with important applications in electronics. The model successfully predicted the reaction outcomes, confirming its accuracy. In the future, the team plans to train the model on more solid-state materials, eventually creating a foundation model applicable to various technologies and industries.