Scientific data are approaching acquisition volumes larger than ever before. Electron microscopy generates exponentially more information as detector technologies advance. In particular, direct electron detectors (DED) are approaching 200 TB/hour.1 To date, electron microscopy holds the world record for resolution at 39 pm.2 This is attained through solving the NP-hard problem of phase retrieval with ptychographic algorithms. Obtaining this resolution requires relatively radiation-robust materials of near-atomic thinness, limiting applicability of the technique.3 To take advantage of DED for low-dose imaging, the quality and usability of every data point must increase. We present a new method to process DED data with machine-learning and high-performance computing infrastructures. Increasing accuracy is attained while compressing information by orders of magnitude.
1I.J. Johnson et al., "Next-Generation Electron Microscopy Detector Aimed at Enabling New Scanning Diffraction Techniques and Online Data Reconstruction," Microsc. Microanal., vol. 24, no. S1, pp. 166-167, 2018.
2Highest resolution microscope. Guinness World Records, 2018.
3Y. Jiang et al., "Electron ptychography of 2D materials to deep sub-angstrom resolution," Nature, vol. 559, no. 7714, pp. 343-349, 2018.