MIT researchers developed an AI model that can classify and quantify atomic-scale defects in materials using noninvasive neutron-scattering data. Trained on 2,000 semiconductor materials, the model uses a multihead attention mechanism to simultaneously detect up to six types of point defects at concentrations as low as 0.2%, something impossible with conventional techniques alone. The approach was validated on an electronics alloy and a superconductor, and the team plans to extend it to Raman spectroscopy for broader industrial adoption in semiconductor, solar cell, and battery manufacturing quality control.
Sort: