scitomo/pfnc-gst-haadf-stem-eds-tomography-b2-d3

PFNC GST HAADF-STEM/EDS Tomography — Lamella b2_d3

Dataset summary

This dataset contains a limited-angle, multi-frame HAADF-STEM and EDS electron-tomography acquisition of a cross-sectional Ge-Sb-Te phase-change-memory lamella.

The data were acquired on a Thermo Fisher Scientific Titan Themis at the Platform for Nanocharacterisation (PFNC), CEA Grenoble, operated at 200 kV and equipped with a four-detector Super-X EDS system.

The original lamella identifier is:

b2_d3

The mapping of this lamella to the virgin or SET device described in the associated publication remains to be confirmed.

The dataset preserves:

• 16 original Velox EMD acquisition containers;
• 50 HAADF frames at each retained tilt;
• a 50-frame EDS spectrum stream at each retained tilt;
• integrated spectra and embedded elemental net-intensity maps;
• extracted HAADF and elemental tilt-series stacks;
• an aligned HAADF tilt series;
• per-projection pixel shifts referenced to the 0° acquisition;
• an explicit tilt-angle list.

Dataset name

Recommended public title:

PFNC GST HAADF-STEM/EDS Tomography — Lamella b2_d3

Recommended Hugging Face repository slug:

pfnc-gst-haadf-stem-eds-tomography-b2-d3

This name identifies the facility context, material system, acquisition modalities, tomography task, and original lamella identifier without making an unverified claim about the device state.

Provenance and credits

Role	Person
Original acquisition operator	Le-Duc Minh Tran — le-ducminh.tran@cea.fr
Dataset curator	Daniele Picone — daniele.picone@cea.fr
Scientific contact	Daniel Del Pozo Bueno — daniel.delpozobueno@cea.fr
PFNC contact	Zineb Saghi — zineb.saghi@cea.fr
Dataset organization	CEA-Leti, Grenoble
Acquisition facility	PFNC, CEA Grenoble

Instrument and acquisition

Field	Value
Facility	Platform for Nanocharacterisation (PFNC), CEA Grenoble
Microscope	Thermo Fisher Scientific Titan Themis
Exact commercial generation/submodel	Not documented
Operating voltage	200 kV
Corrector	Probe corrector
EDS system	Super-X, four silicon-drift detectors
Embedded detector identifier	SuperXG1
Acquisition software	Velox
Embedded Velox version	3.17.0.967-b5399cab
EMD format	Velox EMD version 11
Acquisition date	2025-07-07
Beam current	115 pA
Camera length	110 mm
Pixel size	10.67 Å/pixel
Frame size	400 × 300 pixels
Dwell time	40 µs
Frames per retained tilt	50
Time per frame	5.57 s
Cumulative acquisition time per tilt	278.5 s
Fluence per frame	2.52 × 10² e⁻ Å⁻²
Fluence per tilt	1.26 × 10⁴ e⁻ Å⁻²
Approximate total fluence	2.0 × 10⁵ e⁻ Å⁻²

Sample

Field	Value
Lamella identifier	b2_d3
Filename identifier	D3
Material system	Ge-Sb-Te phase-change-memory device
Specimen form	Cross-sectional FIB lamella
Operational state	Unknown: virgin or SET
Approximate GST-region thickness	50 nm, as reported for the associated study
Holder	Fischione single-axis tomography holder
Tilt-axis relation	Parallel to the GST layer
Wafer, die, and device lineage	Not available

Tilt series

The intended acquisition contained 17 views from −40° to +40° at nominal 5° increments. One corrupted acquisition was discarded.

The 16 retained angles are:

-40, -35, -30, -25, -20, -15, -10, -5,
  0, +10, +15, +20, +25, +30, +35, +40 degrees

The discarded acquisition is number 0010, corresponding to nominal +5°.

Do not synthesize, duplicate, or interpolate the missing +5° projection in the archived data.

Public tilt-series arrays use:

(A, V, U) = (16, 400, 300)

where A is tilt index, V is image row, and U is image column. metadata/angles_deg.txt is the authoritative angle coordinate.

Raw EMD acquisitions

The current raw filenames use the normalized convention:

D3_EDS-HAADF_0001_-40.emd
...
D3_EDS-HAADF_0017_+40.emd

The filename-to-angle mapping is recorded in metadata/acquisition_manifest.csv.

Each retained EMD contains:

HAADF-STEM frame series

shape: (400, 300, 50)
dtype: uint16
frames: 50

EDS spectrum stream

spectral bins: 4096
frames: 50
stream dtype: uint16
detector identifier: SuperXG1

An integrated spectrum of shape (4096, 1) and dtype uint32 is also present.

Embedded elemental maps

Velox-rendered float32 net-intensity maps are present for:

N, O, Si, Ti, Ge, Sb, Te

The embedded map-processing settings include a 3 × 3 mean filter. These maps are processed analytical products, not raw event data or calibrated concentrations.

Derived files

File	Shape	Dtype	Interpretation
HAADF_stack.tif	(16, 400, 300)	uint16	One extracted HAADF projection per retained tilt
HAADF_aligned.tiff	(16, 400, 300)	float32	HAADF tilt series aligned to the 0° acquisition
Ge_stack.tif	(16, 400, 300)	float32	Ge net-intensity tilt series
Sb_stack.tif	(16, 400, 300)	float32	Sb net-intensity tilt series
Te_stack.tif	(16, 400, 300)	float32	Te net-intensity tilt series
Ti_stack.tif	(16, 400, 300)	float32	Ti net-intensity tilt series
Si_stack.tif	(16, 400, 300)	float32	Si net-intensity tilt series
ali_haadf.json	16 integer pairs	—	Pixel translations used to align each HAADF projection to the 0° acquisition

The element-specific TIFF values are net intensities, not atomic concentrations.

HAADF alignment file

ali_haadf.json contains one pair of integer pixel shifts for each projection, in stack order. Each pair specifies the translation used to align that acquisition to the 0° HAADF projection.

Conceptually:

[
  [9, 2],
  ...
  [0, 0],
  ...
]

The [0, 0] entry corresponds to the 0° reference projection.

The following low-level convention remains undocumented:

• whether each pair is ordered as [x, y], [column, row], or [row, column];
• whether values encode the applied shift or the measured displacement;
• interpolation and boundary handling used to produce HAADF_aligned.tiff.

The shift units are pixels.

Elemental lines and later preprocessing

The associated publication reports net-count extraction using:

Element	Emission line
Ge	Kα
Te	Lα
Sb	Lα
Ti	Kα

The publication workflow reports:

1. translation-only alignment with Fiji MultiStackReg;
2. transform estimation from Ge maps;
3. application of the same transformations to Te, Sb, and Ti;
4. later crop/resampling to 176 × 112;
5. binning factor 2;
6. independent maximum normalization per element before reconstruction.

The distributed (16, 400, 300) TIFF stacks correspond to an earlier stage than the final reconstruction inputs.

ali_haadf.json describes HAADF-to-0° alignment and should not automatically be equated with the separate Ge-derived elemental-map transformations.

Repository organization

.
├── README.md
├── LICENSE
├── CITATION.cff
│
├── metadata/
│   ├── dataset.yaml
│   ├── acquisition_manifest.csv
│   ├── angles_deg.txt
│   ├── processing.yaml
│   ├── checksums.sha256
│   └── audit/
│
├── raw/
│   └── emd/
│       ├── D3_EDS-HAADF_0001_-40.emd
│       ├── ...
│       └── D3_EDS-HAADF_0017_+40.emd
│
├── derived/
│   ├── haadf/
│   │   ├── HAADF_stack.tif
│   │   ├── HAADF_aligned.tiff
│   │   └── alignment/
│   │       └── ali_haadf.json
│   └── eds/
│       └── elemental_tilt_series/
│           ├── Ge_stack.tif
│           ├── Sb_stack.tif
│           ├── Te_stack.tif
│           ├── Ti_stack.tif
│           └── Si_stack.tif
│
├── previews/
└── scripts/
    └── audit_themis_dataset.py

Loading the derived data

from pathlib import Path

import json
import numpy as np
import tifffile

root = Path(".")

haadf = tifffile.imread(root / "derived/haadf/HAADF_aligned.tiff")
angles = np.loadtxt(root / "metadata/angles_deg.txt", dtype=np.float64)

with (root / "derived/haadf/alignment/ali_haadf.json").open() as handle:
    haadf_pixel_shifts = np.asarray(json.load(handle), dtype=np.int64)

assert haadf.shape == (16, 400, 300)
assert angles.shape == (16,)
assert haadf_pixel_shifts.shape == (16, 2)

Intended uses

Potential uses include:

• limited-angle HAADF-STEM tomography;
• STEM-EDS chemical tomography;
• multichannel reconstruction;
• raw spectrum-stream reprocessing;
• frame-aware denoising and registration;
• irregular-angle reconstruction;
• joint HAADF/EDS methods;
• semiconductor-device characterization.

The dataset does not contain an independently validated ground-truth reconstruction.

Known limitations

• The meaning of the internal lamella code b2_d3 is undocumented.
• It remains unknown whether b2_d3 is the virgin or SET device.
• The exact Titan Themis generation/submodel is undocumented.
• Wafer, die, and device lineage are unavailable.
• The exact 50-frame HAADF combination method is undocumented.
• The component order and sign convention in ali_haadf.json remain undocumented.
• N and O maps are retained inside the EMD files but not exported as public TIFF stacks.
• Net-intensity maps are not quantitative concentration maps.
• No independently validated 3D ground truth is included.

Associated publication

This dataset is the experimental dataset associated with:

Daniel del Pozo Bueno, Serge Brosset, Theo Monniez, Gabriele Navarro, Philippe Ciuciu, and Zineb Saghi.
Unsupervised Deep Learning for Limited-Angle STEM-EDX Tomography — Application to 3D Chemical Analysis of Phase-Change Memory Devices.
arXiv:2606.10547, 2026. DOI: 10.48550/arXiv.2606.10547.

Associated repository

This dataset is associated to the repository:

https://github.com/CEA-MetroCarac/DL_etomo

License

This dataset is released under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International license:

CC BY-NC-ND 4.0

You may share the unmodified dataset with attribution for non-commercial purposes. Distribution of adapted or modified versions is not permitted under this license. Consult the license text for the authoritative legal terms.

Citation

Use the dataset citation in CITATION.cff and cite the associated publication when using these data in scientific work.

Contacts

• Dataset curator: Daniele Picone — daniele.picone@cea.fr
• Scientific contact: Daniel Del Pozo Bueno — daniel.delpozobueno@cea.fr
• PFNC contact: Zineb Saghi — zineb.saghi@cea.fr
• Original acquisition: Zineb Saghi - zineb.saghi@cea.fr
• Organization: CEA-Leti, Grenoble