Deep-Learning-Enabled 3D STEM Tomography Captures Dislocations in Just 5 Seconds
A research team has developed a rapid 3D dislocation tomography method using scanning transmission electron microscopy (STEM) enhanced by deep learning–based noise filtering. Traditional electron tomography for thick specimens suffers from noise and long acquisition times, making real-time structural analysis difficult. By integrating machine learning to denoise and reconstruct tilt-series images, the new approach visualizes the three-dimensional arrangement of dislocations in a 300 nm thick steel sample using only five seconds of data acquisition. This rapid method dramatically reduces imaging time while maintaining high-quality 3D resolution, offering a powerful new platform for in situ and operando microstructural analysis of complex materials.

Journal: Scientific Reports

DOI: 10.1038/s41598-021-99914-5