High-Speed 4D-STEM using Electrostatic Subframing

The microscopy technique known as four-dimensional scanning transmission electron microscopy “4D-STEM”[1] has grown dramatically in recent years. The trend is driven largely by (1) the increased availability of high-frame-rate cameras and high-throughput data processing and (2) the recognition of the tremendous amount of information that can be produced by this technique. Conventional STEM uses a small number of bright-field, annular dark-field, and/or segmented detectors, each gathering different information about a sample. 4D-STEM extrapolates this to the extreme, answering the question, “What if we could do STEM with a million parallel detectors?” The result is a measurement that unambiguously contains much more information than is possible with traditional approaches, assuming one's computing infrastructure can handle the load. This enables a huge range of advanced analysis and reconstruction techniques. 4D-STEM often uses highly specialized cameras with high frame rates and low pixel counts compared to standard TEM cameras. Cameras used for conventional TEM imaging typically have many millions of pixels, allowing large fields of view and exploiting the massively parallel nature of conventional imaging. With so many pixels, various hardware bottlenecks typically limit continuous full-frame acquisition to tens or hundreds of frames per second. For 4D-STEM, most standard TEM cameras have vastly more pixels than you need and vastly less speed than you want. Many 4D-STEM analysis techniques only need at most a few tens of thousands of pixels per diffraction pattern, while performing a reasonable-sized STEM scan with pixel dwell times of tens of milliseconds is at best far from ideal. In contrast, dedicated 4D-STEM cameras may collect thousands of patterns per second at pixel counts appropriate for later data processing. Here we present an alternative approach, using electrostatic subframing (ES) [2] to subdivide each camera frame into an array of subframes, effectively multiplying the frame rate while proportionately reducing the number of pixels per subframe. This enables high-speed 4D-STEM using a conventional TEM camera. The example in Figure 1 uses an 8×8 subframe array, effectively converting the 2k x 2k 100 fps camera (a Gatan OneView IS [3] using 2×2 hardware binning) into a 256×256 6400 fps camera. Using the same hardware with 8×8 binning, we have captured 4D-STEM scans at 19,200 patterns per second, and the system is capable of operating at still higher rates. The ES system can use subframe array sizes up to 16×16, effectively multiplying the camera frame rate by a factor of 256, which translates to tens of thousands of patterns per second for most modern cameras. Additionally, ES with appropriate camera synchronization naturally eliminates rolling-shutter artifacts that may complicate analysis of 4D-STEM data captured without ES. The data analysis requires us to separate the captured images into individual patterns, correcting the offsets and distortions to each subframe, then assemble the resulting set of images into a four-dimensional array (Figure 2). Once this is done, the data can be exported into standard formats for further analysis using any of the standard 4D-STEM software packages [3]. (Left). Example capture of 64 diffraction patterns in one image, from a 4D-STEM scan of a polycrystalline Au/Pd sample. The exposure time per frame was 10 ms, thus this represents a capture rate of 6400 diffraction patterns per second. At higher binning, we have demonstrated rates up to 19,200 patterns per second. (Below). Conceptual example (final pictured result is a schematic only) of the data analysis process converting an image stack into a 4D-STEM dataset with one diffraction pattern associated with each spatial pixel.