Achieving sub-0.5-angstrom–resolution ptychography in an uncorrected electron microscope

This repo presents my personal collection of scripts and data used for the paper Achieving sub-0.5-angstrom–resolution ptychography in an uncorrected electron microscope published in Science (2024).

Our work demonstrates that with electron ptychography, one may achieve deep sub-angstron resolution on an uncorrected electron microscope.

While the zenodo record is meant for permanently archiving the raw data and relevant codes, I'll try to maintain this personal repo as much as possible.

Environment and dependencies

You'll (probably) need Matlab version >= 2019 to run fold_slice. I've only tested up to 2023b on Windows and 2021a on Linux
You'll need a CUDA-compatible GPU with ideally 8 or 16 GB of RAMs that is detectable by Matlab

Note: If you don't have a Matlab license, it is possible to run the reconstructions with other packages like py4DSTEM if you modify certain inputs like the CBED orientation, scan directions, etc.

Usage

If you download this folder through my github, the sub-folders under /data should be empty due to the file size limit on Github. You would need to download the actual experimental data from our zenodo record. Put the downloaded scan_x128_y128.raw into the corresponding folders and then you should be able to run reconstruction scripts from the main directory uncorrected_ptycho/.

For example:

Download and unzip Fig_01.zip from our zenodo record
Copy the Fig_01/Panel_c-d_Talos/scan_x128_y128.raw, and paste it to data/Fig_1d_10.3mrad_Talos/
Run reconstruction scripts Fig_1d_10p3mrad_Talos_runPtycho.m
The ptychographic reconstruciton output would be saved to data/Fig_1d_10.3mrad_Talos/

Additional notes

The first time executing fold_slice might takes longer because Matlab would need to compile the MEX functions for GPU
The script would first read the 4D-STEM data (.raw), do some preprocessing, and save it as .hdf5 for later reconstruction. Saving an upsampled .hdf5 (a few GB) would take a couple minutes.
Practically we only need to preprocess the .raw once and you may reuse the .hdf5 if you're reconstrucing the same data multiple times, but you'll need to modify the provided script accordingly because this script is meant for out-of-the-box experience.
Generally, a successful reconstruciton would need a couple stages of refinement to achive the best results, while these example scripts are meant for demo purpose so the result could be sub-optimal due to insufficient iterations / probe modes.
If you run into GPU memory issues, try reducing the grouping, resample_factor, or Nprobe.
There're a lot more useful information provided by Yi that's been compiled in fold_slice. I highly recommend cheking out the tutorials.

References

If you find the script useful for your research, please consider citing our original paper

@article{nguyen2024achieving,
  title={Achieving sub-0.5-angstrom--resolution ptychography in an uncorrected electron microscope},
  author={Nguyen, Kayla X and Jiang, Yi and Lee, Chia-Hao and Kharel, Priti and Zhang, Yue and van der Zande, Arend M and Huang, Pinshane Y},
  journal={Science},
  volume={383},
  number={6685},
  pages={865--870},
  year={2024},
  publisher={American Association for the Advancement of Science}
}

Created by Chia-Hao Lee on 2024.03.07

cl2696@cornell.edu

xiongh15/uncorrected_ptycho

Achieving sub-0.5-angstrom–resolution ptychography in an uncorrected electron microscope

Environment and dependencies

Usage

Additional notes

References