Automated Electron Microscopy X-Ray Spectroscopy for Compositional Characterization of Materials

AutoEMX is a fully automated framework for SEM-EDS workflows — from spectral acquisition and quantification, to principled filtering and compositional analysis — all in one click.

🎥 Watch AutoEMX in action on a desktop SEM-EDS system at https://youtu.be/Bym58gNxlj0

🧪 Try quantification now: upload a .msa or .emsa spectrum at https://autoemx-singlespectrum.streamlit.app

📖 This work is described in:
A. Giunto et al., Accurate SEM‑EDS Quantification, Automation, and Machine Learning Enable High‑Throughput Compositional Characterization of Powders, Nature Communications 17, 9735 (2026).
DOI: https://doi.org/10.1038/s41467-026-76633-x

✨ Key Features

• Fully automated SEM-EDS phase-level compositional analysis workflow, which includes:

  • Acquisition of EDS spectra, including particle localization if sample is powder. Compatible also with bulk samples, or manual navigation.
  • Automated peak identification and element assignment for unknown samples
  • Quantification of compositions using the peak-to-background method
  • Rule-based filtering of compositions to discard poorly quantified spectra from the analysis
  • Unsupervised machine learning–based analysis to identify the compositions of individual phases in the sample

• AutoEMX GUI — acquire spectra with the microscope, quantify the samples of a project, run their clustering analysis with every parameter and explore the clusters in interactive 3D / ternary / 2D plots linked to the spectra and SEM images, and fit/quantify single spectra. Works also with spectra acquired and exported with commercial software (.msa, .emsa, .msg)

• Automated experimental standard collection scripts

• Automated particle size distribution measurements scripts

• Extensible architecture — adaptable to other techniques such as

  • Wavelength Dispersive Spectroscopy (WDS)
  • Scanning Transmission Electron Microscopy (STEM) with EDS

• Extensible hardware support — includes driver for ThermoFisher Phenom Desktop SEM series, and can be extended to any electron microscope with a Python API

📊 Performance

• Benchmarked on 74 single-phase samples with compositions spanning 38 elements (from nitrogen to bismuth), it achieved <5–10% relative deviation from expected values
• Machine learning compositional analysis detects individual phase composition in multi-phase samples, including minor phases
• Intermixed phases can also be resolved
• Works with spectra from any microscope, although recalibration is recommended for maximum compositional accuracy

🧪 Supported Use Cases

• Scanning Electron Microscopy (SEM) with Energy-Dispersive Spectroscopy (EDS)
• Powders and rough samples, e.g. rough films, or pellets, with automated segmentation.
• Bulk, flat samples, probed by defining a grid of points.
• Manual navigation of any sample.

⚙️ Requirements

• Cross-platform: runs on Linux, macOS, and Windows
• Quick installation via pip

---

📑 Table of Contents

• 🧪 Try quantification on your spectrum
• 📘 Documentation
• 📦 Requirements
• 📂 Project Structure
• 📁 Scripts
• 🤝 Contributing
• 📄 License
• 📖 Citation
• 📂 Paper Data
• 📬 Contact

---

🧪 Try quantification on your spectrum

Upload a .msa or .emsa spectrum acquired at 15 kV and test quantification in the browser:

👉 https://autoemx-singlespectrum.streamlit.app

---

📘 Documentation

Installation instructions, usage examples, and workflow descriptions are available in the AutoEMX documentation:

👉 https://cedergrouphub.github.io/AutoEMX/

---

📦 Requirements

• Python 3.11 or newer
• All dependencies are installed automatically via pip or conda.
• Tested versions of dependencies are specified in pyproject.toml.

  The package may work with more recent versions, but these have not been tested.

---

Electron Microscope Support

• ✅ Developed and tested for Thermo Fisher Phenom Desktop SEMs.
• ✅ Compatible with any Phenom microscope equipped with PPI (Phenom Programming Interface).
• ⚠️ For other microscope models, the driver must be adapted to the appropriate API commands (easy to do with modern LLMs).

---

📂 Project Structure

The repository is organized as follows:

AutoEMX/
├── autoemx/                 # Main package source code
│   ├── config/                 # Configuration files, including default values to employ during measurements.
│   ├── core/                   # Core objects and source code
│   ├── data/                   # Libraries of X-ray data
│   ├── microscope_drivers/              # Electron Microscope driver (⚠️ adapt to your own instrument)
│   ├── runners/                # Runner functions calling on core objects
│   ├── scripts/                # Scripts to run acquisition, quantification, etc. (see full list below)
│   ├── web/                    # Streamlit app of the online single-spectrum demo
│   ├── gui/                    # Local GUI: acquisition, quantification, analysis, single spectrum (`python -m autoemx.gui`)
│   ├── calibrations/             # Detector models of supported microscopes (your instrument's calibrations are kept in ~/.autoemx)
│   ├── utils/                  # Utility functions and strings employed by the program
│
├── examples/                  # Example scripts for fitting, quantification and compositional analysis of example data
├── tests/                     # CI smoke suite (`test_ci_smoke.py`) + offline workflow/unit tests
│                               # Hardware-only scripts stay capitalized (`Test_EM_driver.py`)
├── paper_data/                # Raw paper data uploaded on Git LFS (Dowload instructions in Paper Data section below)
│
├── LICENSE.txt
├── README.md
└── pyproject.toml

---

📁 Scripts

This repository includes a collection of scripts that streamline the use of AutoEMX. Each script is tailored for a specific task in spectral acquisition, calibration, quantification, or analysis. Below are the main scripts available in autoemx/scripts/ and their purposes:

🔬 Acquisition, Quantification & Analysis

• Run_Acquisition.py — Acquire X-ray spectra from the microscope (supports automated and manual modes).
• Run_Quantification.py — Quantify acquired spectra (single or multiple samples) and perform machine-learning analysis.
• Run_Analysis.py — Launch customized machine-learning analysis on previously quantified data.
• Set_Active_Analysis.py — Choose which quantification run and analysis of a sample are active (reused by later runs, shown in the GUI).
• Fit_Quant_Single_AutoEMX_Spectrum.py — Fit and optionally quantify a single spectrum measured with AutoEMX. Prints fitting parameters and plots fitted spectrum for detailed inspection of model performance.
• Fit_Quant_Single_MSA_Spectrum.py — Fit and optionally quantify a single spectrum exported by proprietary software.
• quantify_external_spectra.py — Quantify spectra acquired outside AutoEMX (e.g., from other SEM-EDS systems).
• Annotate_Particle_Images.py — Draw the positions of the measured spectra on the particle images of a sample.
• Online demo — upload .msa / .emsa and test quantification at https://autoemx-singlespectrum.streamlit.app, without installing AutoEMX. Locally, use the Single spectrum tab of the AutoEMX GUI.

📊 Particle Size Distribution Measurements

• Collect_Particle_Statistics.py — Analyse sample, collecting particle size statistics and distribution.
• Process_Particle_Stats_Files.py — Filter out particles from acquired particle size data and recompute the statistics.

🛠️ Calibration of Your Instrument

• Run_SDD_Calibration.py — Calibrate the energy scale of the SDD detector.
• Run_Experimental_Standard_Collection.py — Acquire and fit experimental standards, to obtain reference P/B values.
• Refit_Experimental_Standards.py — Refit previously acquired experimental standards.
• Modify_EDS_Standards_File.py — Delete standards, or exclude them from the reference P/B values.
• Extract_Experimental_Standards_Details.py — Summarize the available standards (elements, peaks, P/B values).
• Calibrate_Peak_Sigma_Params.py, Calibrate_Peak_Shape_Params.py, Calibrate_Background_Params.py — Calibrate the detector model of a new microscope (peak widths, peak tails, background).

⚗️ Characterize Extent of Intermixing in Known Powder Mixtures

(see Chem. Mater. 2025, 37, 6807−6822 for example)
Use the same scripts as regular composition characterization, as described in the docs Tutorial.

👉 All scripts can be executed directly from the command line or imported into a Python environment, making them accessible from anywhere on your system.

---

🤝 Contributing

Contributions are welcome! Need a feature or guide that is not available? Please ask for it in the GitHub Discussions: requests help us prioritize.

Open to collaborations to extend this package to different tools or to different types of samples, for example thin films. Please contact me at agiunto@lbl.gov

---

📄 License

Free use for non-commercial use only. Contact IPO@lbl.gov for commercial purposes.

This project is licensed under a NON-COMMERCIAL USE ONLY license — see the LICENSE file for details.

---

📖 Citation

If you use AutoEMX in your research, please cite the following publication:

A. Giunto, Y. Fei, P. Nevatia, B. Rendy, N. J. Szymanski & G. Ceder, Accurate SEM‑EDS Quantification, Automation, and Machine Learning Enable High‑Throughput Compositional Characterization of Powders, Nature Communications 17, 9735 (2026). DOI: https://doi.org/10.1038/s41467-026-76633-x

BibTeX

@article{Giunto2026AutoEMX,
  author  = {Giunto, Andrea and Fei, Yuxing and Nevatia, Pragnay and Rendy, Bernardus and Szymanski, Nathan J. and Ceder, Gerbrand},
  title   = {Accurate SEM‑EDS Quantification, Automation, and Machine Learning Enable High‑Throughput Compositional Characterization of Powders},
  journal = {Nature Communications},
  volume  = {17},
  pages   = {9735},
  year    = {2026},
  doi     = {10.1038/s41467-026-76633-x},
  url     = {https://doi.org/10.1038/s41467-026-76633-x},
}

---

📂 Paper Data

The raw data used in the associated publication is stored in the paper_data/ directory.
These files are tracked with Git LFS (Large File Storage).

🔽 Download with Git LFS

The repository is automatically cloned without Git LFS; you will only see placeholder files instead of the actual datasets inside paper_data/.
To download the full data, on the terminal go to the repo directory and:

# 1. Install Git LFS (only needed once per machine)
git lfs install

# 2. Fetch the data files
git lfs fetch --all
git lfs checkout

Alternatively, download manually from the github repo Download button.

After downloading, run the run_analysis.py or run_quantification_analysis.py scripts within the folder.

---

📬 Contact

For questions or issues, please open an issue on GitHub.

---