Engineers have long struggled to measure the microscopic "defects" that dictate how materials like steel and silicon perform. While some defects are essential for conductivity, unwanted impurities can cripple a product's efficiency. Historically, verifying these atomic-scale irregularities required destructive testing—essentially cutting open the material to see inside.

MIT researchers have now introduced a foundational AI model that changes the math on material science. By training on 2,000 different semiconductor materials, the model learns to interpret vibrational frequencies—the specific ways atoms wiggle within a solid—to identify structural flaws. The system uses a multi-head attention mechanism, the same architecture powering modern chatbots, to separate overlapping signals from up to six different types of defects simultaneously.

This non-invasive approach provides a "full picture" of a material's internal state without damaging the sample. While current testing requires specialized neutron-scattering facilities, the team is already working to adapt the model for Raman spectroscopy, a common industrial technique using light. This leap in "defect science" could soon accelerate the production of more efficient solar cells, batteries, and next-generation microelectronics by providing real-time quality control on the factory floor.

Engineers have long struggled to measure the microscopic "defects" that dictate how materials like steel and silicon perform. While some defects are essential for conductivity, unwanted impurities can cripple a product's efficiency. Historically, verifying these atomic-scale irregularities required destructive testing—essentially cutting open the material to see inside.

MIT researchers have now introduced a foundational AI model that changes the math on material science. By training on 2,000 different semiconductor materials, the model learns to interpret vibrational frequencies—the specific ways atoms wiggle within a solid—to identify structural flaws. The system uses a multi-head attention mechanism, the same architecture powering modern chatbots, to separate overlapping signals from up to six different types of defects simultaneously.

This non-invasive approach provides a "full picture" of a material's internal state without damaging the sample. While current testing requires specialized neutron-scattering facilities, the team is already working to adapt the model for Raman spectroscopy, a common industrial technique using light. This leap in "defect science" could soon accelerate the production of more efficient solar cells, batteries, and next-generation microelectronics by providing real-time quality control on the factory floor.