📊RHEED Analysis

How can AtomCloud help me get more from my reflection high-energy electron diffraction (RHEED) data?

After uploading your RHEED files, AtomCloud will immediately get to work extracting key information from your data using our curated pipelines (see consolidated example literature). You can see the status of your files on the Data Management page, in the Your Data section. For RHEED data, the analysis pipeline is broken down into a few steps outlined below:

For data visualization see: Clustering and Pattern Analysis

Data Cleaning and Standardization

  1. Automated rotating vs. stationary stage detection.

If the data is taken from a rotating stage, AtomCloud will automatically extract the rotational frequency using fast-fourier transform (FFT) analysis, and average over the rotation period to create a semi-stationary RHEED video that is a moving average of over one rotational period.

  1. Automated video cropping: reduce the field of view of the raw RHEED data to the illuminated pattern region only.

This standardizes the data across different detector configurations and reducing the size of the RHEED data.

  1. Time cropping: remove sections of the input data without active signal. This can occur when the detector is recording but the shutter is closed.

  2. Data can be exported at this stage for independent analysis. Use the 'Export Processed Selected' button on the Data Management page to export the transformed video file for the selected data.

Step 2 - Dimensionality Reduction and Clustering

Principal component analysis (PCA) is a dimensionality reduction technique which aims to find efficient representations of data while maintaining the variance in the original dataset. The new representation is calculated by diagonalizing the covariance matrix.

In this step, we apply PCA to the RHEED images, where each frame of a RHEED dataset is treated as an observation and each pixel in the processed images is a feature. By retaining enough components to preserve 95% of the variance in the processed RHEED data, we can often reduce the size of the RHEED dataset by ~10,000x. This allows us to use more complex data analysis techniques on the smaller dataset while retaining the core information.

Step 3 - Physical Feature Extraction


Clustering algorithms are applied to identify and group statistically similar frames within the RHEED recording. This clustering is done without knowledge of the frame sequence (time order) providing robust statistical grouping.

Clusters are used to identify changes in the growth phase as they signal significant changes in the diffraction pattern's evolution.

Machine learning analysis of perovskite oxides grown by molecular beam epitaxy (Sydney, Et Al.)

Skill-Agnostic analysis of reflection high-energy electron diffraction patterns for Si(111) surface superstructures using machine learning (Asako Yoshinari, Et Al.)

Streak to spot Ratio

This value quantifies whether you are closer to an island-like growth mode (high streak/spot ratio) or a layer by layer growth mode (low streak/spot ratio)

Application of machine learning to reflection high-energy electron diffraction images for automated structural phase mapping (Liang, Et Al.)

Last updated