MIT researchers report an AI model that identifies and measures multiple atomic-scale defects in solid materials using noninvasive neutron-scattering data, a task that has long challenged conventional techniques. Trained on spectra from 2,000 semiconductor materials, the system can simultaneously classify up to six point defects and estimate concentrations down to 0.2%, offering a potential boon for manufacturers of semiconductors, batteries, and solar cells seeking tighter process control and higher yields. The model relies on attention-based pattern recognition to parse subtle differences in vibrational signatures between pristine and doped samples. While the current approach uses neutron measurements that are not yet practical for routine factory deployment, the team plans to adapt it to Raman spectroscopy, a more widely available tool. The work, published in Matter and supported by the Department of Energy and the National Science Foundation, underscores growing momentum for AI-driven, non-destructive quality assurance across advanced materials and electronics.
Related articles:
Raman spectroscopy
Nondestructive testing
Spallation Neutron Source (SNS)
