Removing constraints of 4D-STEM with a framework for event-driven acquisition and processing.

Area of Science:

• Materials Science
• Physics
• Electron Microscopy

Background:

• Scanning transmission electron microscopy (STEM) generates massive datasets (tens to hundreds of GB per scan).
• Efficient data representation is crucial for scalable advancements and widespread adoption of STEM technologies.
• Event-driven detectors, like Timepix3, show promise for low-dose and real-time processing in electron microscopy.

Purpose of the Study:

• To compare sparse and dense data representations in 4D-STEM regarding size and computational cost.
• To demonstrate the benefits of an event-driven data format for 4D-STEM acquisition and processing.
• To introduce a framework for optimizing the entire 4D-STEM pipeline using event data.

Main Methods:

• Comparative analysis of sparse vs. dense data representations in various 4D-STEM scenarios (high-resolution imaging, nano-beam electron diffraction).
• Development of a data acquisition and processing framework optimized for event-driven data.
• Demonstration of live, event-driven 4D-STEM processing, including analytical ptychography.

Main Results:

• Event-driven 4D-STEM significantly reduces memory, bandwidth, and computational demands compared to dense representations.
• Optimizing the pipeline for event format avoids intermediate dense data, fully leveraging detector advantages.
• Successful live processing of event-driven 4D-STEM data was achieved, showcasing the framework's capability.

Conclusions:

• Event-driven data processing is key to fully realizing the potential of advanced electron detectors in STEM.
• A dedicated framework for event-based acquisition and processing enables efficient, real-time analysis in 4D-STEM.
• This approach facilitates scalable adoption of cutting-edge electron microscopy techniques.