|About this Abstract
||2021 TMS Annual Meeting & Exhibition
||Characterization of Materials through High Resolution Imaging
||AI-enabled High-throughput Three-dimensional Ptychographic Imaging
||Yi Jiang, Tao Zhou, Mirko Holler, Jeffrey Klug, Zhonghou Cai, Christian Roehrig, Mathew Cherukara
|On-Site Speaker (Planned)
X-ray ptychography has become a standard technique for high-resolution imaging at nanoscale. In combination with tomography or laminography, the technique can extend to quantitative 3D characterization and can achieve sub-20 nm spatial resolution. The fourth-generation light source and novel scanning techniques further allow 3D ptychography to image large objects at millimeter or centimeter scale. However, the increasing data acquisition rates and data volumes also bring tremendous burdens on data storage and image reconstruction. Here we demonstrate that a supervised deep convolutional neural network named LaminoNN can retrieve an object’s structure directly from scanning diffraction patterns, replacing more computationally expensive ptychographic reconstruction. We apply LaminoNN to an experimental ptycho-laminography measurement, which includes ~400,000 diffraction patterns and thus provide adequate training data for the neural network to achieve high prediction accuracy. The projection images recovered by LaminoNN are then used in subsequent laminographic alignment and reconstruction to produce 3D sturcture at high-resolution.
||Machine Learning, Characterization, Nanotechnology