Four-dimensional scanning transmission electron microscopy (4D-STEM) can yield a variety of atomically resolved measurements simultaneously, making it a widely used tool for both materials and life science applications [1]. In 4D-STEM, a sub-atomic focused electron beam is scanned across a thin sample while the full two-dimensional convergent beam electron diffraction pattern (CBED) is collected at every probe position. Each probe position contained within a 2D raster array, therefore, correlates to a 2D image in reciprocal space, hence the name 4D-STEM.
A key advantage of 4D-STEM is that it can be leveraged to extract high-spatial-resolution structural and property information from beam-sensitive materials by using low-dose illumination.
To understand why high-speed detectors are essential for low-dose 4D-STEM, one has to look at dose and sampling efficiency. In a 4D-STEM measurement, the typical dose, F (or fluence), can be expressed as:
F ~ I ∆t/∆R2
Where I is the probe current, ∆t is the pixel dwell time (or time spent at each probe position), and ∆R is the real space scanning step size. For low dose experiments, there are different options. Lower probe currents (I), faster pixel dwell times (i.e., lower ∆t, faster detectors) and longer scanning step size (∆R).
Assuming normal I (low enough to still provide enough counts at the detector) and ∆R (typically around 0.5-Å for atomic resolution), conducting ultra-low dose experiments at atomic resolution means that the detector speed (∆t) is critical.
Removing the bottleneck in speed for 4D-STEM
Ultrafast detectors in 4D-STEM are needed to handle the low electron budgets required by beam-sensitive samples that damage primarily through radiolytic pathways [2]. Therefore, an ultrafast detector’s performance under sparse conditions is critical for the retrieval of atomically-resolved structural information. Readout speeds of conventional frame-based detectors limit 4D-STEM to relatively longer dwell times, impeding its application to beam-sensitive materials and making the resulting datasets susceptible to drift.
The CheeTah T3 hybrid pixel electron detector overcomes this bottleneck by providing novel event-driven readout. Every pixel runs at 640-MHz and therefore guarantees dwell times as low as 1/640 MHz = 1.5625 ns. This is by far much lower compared to any frame-based detector existing on the market.
Recently, a Timepix3-based detector was used to collect 4D-STEM data on dichalcogenide and zeolite samples while avoiding unacceptable beam damage [3]. Single and multi-frame 4D-STEM acquisitions covering a 1024×1024-pixel scan area were performed at 200 kV and 60 kV using dwell times of 1 µs and 6 µs, respectively. Researchers also demonstrated successful 4D-STEM acquisition at 100-ns dwell time. Comparatively, a frame-based detector would need to operate at 10,000,000 frames per second to support such a dwell time.
How does event-based detection work with Timepix3?
Conventional frame-based detectors read out every pixel, even those with zero counts, to record a single two-dimensional array of intensity values in a single frame. Conversely, event-driven detection only performs readout on pixels that receive electrons. Under sparse conditions, this makes a huge difference in the data handling efficiency of the detector. For example, the size of a 4D-STEM dataset discussed in Verbeeck et al. is only 6.6 GB, compared to a 33-GB dataset that would be generated by a frame-based detector operating in 1-bit mode [3].
Event-driven readout takes advantage of sparse imaging conditions by ignoring pixels containing zero signal, resulting in higher effective frame rates. The Timepix3 ASIC chip, designed at CERN, is the first event-based hybrid pixel detector. Each pixel behaves independently to assign timestamps to individual electron hits. The Timepix3 sensor is composed of a 256×256 array of 55-um pixels. The CheeTah T3 Quad, which utilizes four Timepix3 chips, features a 2×2 sensor configuration (512×512 pixels) to support ultrafast 4D-STEM acquisition.
Electrons generated within the sensor layer are collected at individual pixels where an on-chip ASIC records each event as a time-varying signal that demonstrates a reverse sawtooth shape. Timepix3 has an internal digital timer that can measure time of arrival (ToA) with an accuracy of 1.56 ns. Simultaneously, the time-over-threshold (ToT), proportional to the energy of the incoming event, is also recorded. The threshold value influences the cluster size at a given accelerating voltage, or the number of neighboring pixels excited by a single event.
Each digital packet of data sent out by the detector contains the following data:
- Its “address”, which is its X and Y coordinate in the pixel matrix.
- The time of arrival (ToA) of the hit
- The time-over-threshold (ToT) of the hit
- 65-bits of information
The throughput of data is determined by the electron intensity on the detector. If no electrons arrive, no data is sent out. The maximum hit rate of the CheeTah T3 Quad is 120,000,000 counts per second or 457 counts per pixel per second. Comparatively, the Medipix3 frame-based sensor can handle over 1,000,000 hits per pixel per second. This means that the Timepix3 is limited to lower beam currents than that used in conventional STEM.
For extremely beam-sensitive materials, the Timepix3 opens up the possibility to always perform 4D-STEM to gain more detailed and localized structural and property information about the sample by guaranteeing the high-speed throughput necessary to minimize exposure (and damage) of the sample to the electron beam.
References
[1] C. Ophus, “Four-Dimensional Scanning Transmission Electron Microscopy (4D-STEM): From Scanning Nanodiffraction to Ptychography and Beyond,” Microscopy & Microanalysis, vol. 25, pp. 563-582, 2019.
[2] G. Li, H. Zhang and Y. Han, “4D-STEM Ptychography for Electron-Beam-Sensitive Materials,” ACS Central Science, vol. 8, pp. 1579-1588, 2022.
[3] D. Jannis, C. Hofer, C. Gao, X. Xie, A. Beche, T. Pennycook and J. Verbeeck, “Event driven 4D STEM acquisition with a Timepix3 detector: microsecond dwell time and faster scans for high precision and low dose applications,” Ultramicroscopy, vol. 233, p. 113423, 2022.
