Datasets for "3D Atomic-Scale Metrology of Strain Relaxation and Roughness in Gate-All-Around Transistors via Electron Ptychography"

Through-focal series, 4D-STEM datasets and reconstructed phase stacks associated with the paper "3D Atomic-Scale Metrology of Strain Relaxation and Roughness in Gate-All-Around Transistors via Electron Ptychography." This repository includes the raw MEP datasets, MEP reconstructed phase stacks, and through-focal comparison images (iDPC and ADF). The reconstruction code is based on the fold_slice package (https://github.com/yijiang1/fold_slice), with tilt propagator extension (courtesy of Lopa Bhatt, Muller Group). A copy of the code is included with an example script and parameters to reproduce the reconstructions. Jupyter notebooks used for atom tracking, 3D interface shape extraction, and strain/roughness analysis are also provided. READ_ME.txt Associated manuscript: “3D Atomic-Scale Metrology of Strain Relaxation and Roughness in Gate-All-Around (GAA) Transistors via Electron Ptychography” Authors: Shake Karapetyan, et al. This repository contains the source code, reconstruction scripts, example datasets, and analysis notebooks used to generate the results presented in the manuscript. The repository includes: • 4D-STEM datasets (HDF5 format) • Through-focal ADF/iDPC image stacks • TIFF files of reconstructed MEP phase slices • MATLAB scripts for running multislice electron ptychography reconstructions (fold_slice with tilt-propagator extension; courtesy of Lopa Bhatt, Muller Group); original fold_slice package can be found at https://github.com/yijiang1/fold_slice • Jupyter notebooks for atom tracking, strain mapping, and surface morphology analysis • Parameter files documenting all acquisition and reconstruction settings ------------------------------------------------------------------------------------------------------------------------ 1. System Requirements A. Software dependencies a. MATLAB R2023b or later (tested also on R2023a, R2024a) b. Python ≥ 3.9 with the following packages: i. atomap ii. hyperspy iii. numpy iv. scipy v. scikit-image vi. matplotlib vii. tqdm viii. tifffile ix. open3d x. sklearn xi. prettytable c. Jupyter Notebook or JupyterLab d. No non-standard hardware required beyond a CUDA GPU for faster reconstruction. B. Systems tested a. Cornell Altas HPC cluster with NVIDIA A100 GPUs (MEP reconstructions) b. macOS Sonoma (M1/M2) (Python analysis notebooks) C. Hardware recommendations a. GPU recommended for reconstruction; CPU execution is possible but slow b. ≥ 16 GB RAM recommended for handling full 4D-STEM datasets ------------------------------------------------------------------------------------------------------------------------ 2. Installation Guide A. fold_slice MEP reconstruction (MATLAB) No installation is required. The provided folder fold_slice_tiltProp contains all code needed to run reconstructions. To install: 1. Copy the repository to any location. 2. Add the fold_slice_tiltProp folder and subdirectories to your MATLAB path. B. Python analysis environment Most users will find a conda environment easiest: conda create -n gaa_mep python=3.10 conda activate gaa_mep pip install atomap hyperspy numpy scipy scikit-image matplotlib tqdm tifffile open3d sklearn prettytable Typical installation time: 5–10 minutes on a standard laptop. ------------------------------------------------------------------------------------------------------------------------ 3. Demo Instructions A. Example dataset included Dataset: SupFig9a_MEP_Planar_Interface.hdf5 Reconstruction output: SupFig9a_MEP_Planar_Interface_Niter800_unwrapped.mat B. Steps to run the demo reconstruction Navigate in MATLAB to: fold_slice_tiltProp/ptycho/examples/Planar_Interface/ Run: prepare_data_Planar_Interface.m run_multislice_Planar_Interface.m Expected run time on an A100 GPU: ~24 hours (About 1 day per dataset, depending on slice thickness and number of iterations.) The reconstruction output is saved automatically in the results folder. Use view_multislice.m to generate the unwrapped phase to view. ------------------------------------------------------------------------------------------------------------------------ 4. Instructions for Use on User Data To run MEP on another dataset: 1. Place the dataset (HDF5 format) in the working directory. 2. Copy or adapt the example scripts in ptycho/examples/Planar_Interface/. 3. Update parameters (scan step, defocus, slice thickness, convergence angle, etc.) using values in the provided file: Parameters.txt 4. Run the same two MATLAB scripts. Python notebooks for atom finding, strain analysis, and geometry extraction can be adapted by replacing the input TIFF files. Typical notebook run time: 5–20 minutes, depending on dataset size. ------------------------------------------------------------------------------------------------------------------------ 5. Repository Contents DATASETS • planar Si/SiO2/HfO2 interface dataset: SupFig9a_MEP_Planar_Interface.hdf5 Dataset used in Fig. 6c,g; SFig. 7; SFig. 8a-c; SFig. 9a; SMovie 1; • second planar interface dataset: SupFig9b_MEP_Planar_Interface_2.hdf5 Dataset used in Fig. 4; SFig. 5; SFig. 9b; • GAA device 1: SupFig9c_MEP_GAA1.hdf5 Dataset used in Fig. 5; Fig. 6c; SFig. 6; SFig. 9c; SMovie 2, 4 and 6; • GAA device 2: SupFig9d_MEP_GAA2.hdf5 Dataset used in Fig. 1c-d; Fig. 2e; Fig. 6; SFig. 4a; SFig. 8d-h; SFig. 9d; SMovie 3 and 5. RECONSTRUCTED PHASE STACKS (TIFF) • MEP_phase_Planar_Interface_F.tiff • MEP_phase_Planar_Interface_2_F.tiff • MEP_phase_GAA1_F.tiff • MEP_phase_GAA2_F.tiff THROUGH-FOCAL DATA • tf-iDPC_GAA2_2nmStep.tif • tf-ADF_GAA2_2nmStep.tif PARAMETER FILE • Parameters.txt -- all experimental and reconstruction settings MATLAB RECONSTRUCTION SCRIPTS (located in fold_slice_tiltProp/ptycho/examples/) • prepare_data_Planar_Interface.m • run_multislice_Planar_Interface.m PYTHON ANALYSIS NOTEBOOKS • 01_Planar_Interface_atom_tracking.ipynb • 02_Planar_Interface_layer_tracking.ipynb • 03_Planar_Interface_strain_relaxation.ipynb • 04_Planar_Interface_3D_shape_roughness.ipynb