Is Electron Counting Feasible with Indirect Camera Systems for TEM?
Pushing the Limits of Conventional Fiber-Optically-Coupled Scintillator-Based CMOS Cameras
- Fig. 1: Sensors for direct and indirect detection camera systems are built in a layered stack with passivation, metal and active layers supported by a possibly back thinned substrate. For fiber optically coupled indirect detection cameras, the optical stack is glued onto the sensor, while direct detection systems are directly exposed to the vacuum.
- Fig. 2: Typical shape of single electron events (left). The accumulated event intensity from a 3x3 pixel region around the event center shows the expected Landau distribution (right).
- Fig. 3: Images formed by reconstructed single electron events for graphitized C and Au samples (ab) with Fourier spectrum of (b) in (c).
- Fig. 4: Diffraction image of polycrystalline Al. The switchable full-well capacity enables faithful recording of high intensities, as can be seen in the insets with equivalent dose potentials graphically marked. All images have been taken with identical microscope settings and exposure times.
- No Whitepaper has been uploaded yet
Direct detection camera systems received a lot of praise in the last decade and are even referred to fuel a “Resolution Revolution” particularly in low-dose experiments . This article compares the technological implications of direct and indirect detector designs and looks into the feasibility of implementing single electron (SE) counting acquisition schemes for a novel state of the art indirect CMOS camera system.
Apart from eliminating the scintillator as one source of noise, the superior performance of modern direct detection camera systems for TEM is traced back to their ability to clearly detect single electron events (SEE) at a fast rate, which enables a so-called counting-mode acquisition scheme. Therein, the illuminating electron dose rate is chosen such that with the given frame rate of the camera, SEEs appear spatially separated and the point of first interaction of the primary electron (PE) with the sensor can be determined. By incrementing a binary counter in a 2D array at each position the origin of a SEE has been found, the counting mode acquisition scheme normalizes the Landau noise affected contributions of each SEE for an improved detection quantum efficiency.
Since both properties– high speed and SE sensitivity can in principle also be achieved by an indirect camera, it is worth describing the implications of the different design choices.
Shot into a bulk material, a high-energy particle leaves an ionization trail along its path. For detection of such an event, the resulting charge needs to be separated before it is able to recombine. This is done in a depletion zone with a static electrical field, a photo diode. In case of direct detectors, the particles of interest are the high-energy PEs used for imaging the specimen. In case of indirect cameras, the energy of a PE undergoes a transformation into usually several hundred photons, which are then directed via an optical coupling onto a sensor. The optical system can be optimized by matching photon energy and peak sensitivity of the detector, as well as maximizing the transfer from scintillator to chip by using high numerical aperture fiber optical coupling and only a very thin layer of glue in between sensor and glass (fig.
To clearly detect an SEE, it is necessary for the signal of a PE to be well above the noise floor of the detector. Direct detection cameras reach this requirement, as a high-energy PE can generate hundreds of secondary electrons in the active layer. Indirect detectors can adjust the amount of photons generated by a PE by tuning the thickness of the scintillating material. However, while thick layers may emit a lot of photons, they only allow for a limited localization, which is detrimental for the resolution.
The resolution of a direct detection camera in counting mode depends on its ability to find the first point of interaction with the detector surface, which requires low dose rates adapted to the point spread function of the SEEs and achievable frame rate to suppress coincidence electrons. Pinpointing the location of the first interaction is not trivial, as a PE undergoes scattering in insensitive layers before reaching a photo diode (fig. 1). Furthermore, the largest energy deposition occurs at the position, where the PE is finally stopped, meaning the highest signal is not located at the point of first interaction with the detector. Current direct detection cameras therefore use back thinned sensors and allow high-energy PEs to completely traverse the bulk without stopping them .
As with direct detectors, a PE in the bulk of a scintillator also undergoes scattering. Monte Carlo simulations prove the interaction volume and therefore also the location of the source of the emitted photons is strictly confined to a certain characteristic projected area (not shown). Generally, to maximize direct resolution, it is desirable to match this area to the physical pixel size. For usual acceleration voltages and bulk materials, this results in rather large pixels in excess of 15x15µm2.
For pixel detectors, one source of noise is the kTC (or reset) noise, which is caused by the varying level of a pixel after it has been reset. Usually, this is accounted for by a correlated double sampling (CDS) acquisition scheme, where the reset signal is read out non destructively and subtracted after the exposure. This can be done either on- or off-chip, where the latter usually suffers from an inferior time correlation resulting in a higher noise level.
On-chip CDS requires each pixel to have its dedicated storage for the reset level. This can be implemented in a so-called floating diffusion (FD) node, which is charged to a certain voltage corresponding to the reset level and decoupled afterwards. In direct detectors, implementing such an FD node is impeded by the fact that the PEs are penetrating the sensor and may alter the signal.
Indirect detectors suffer from noise in the scintillator, since the resulting amount of photons depends on how a phosphor grain is hit by a PE. Generally, each pixel readout is hampered by a certain readout noise.
Performance of a Novel Indirect CMOS Camera System
Leveraging improvements in semiconductor manufacturing technology, design and implementation of a novel CMOS sensor became feasible. It features on-chip CDS for reduced kTC noise, a switchable full-well capacity for optimizing high sensitivity and extended dynamic range experiments while supporting high frame rates in rolling shutter (RS) movie acquisition. Albeit SE sensitivity has been demonstrated on previous systems , on-chip CDS now enables detection of SEE also at movie rates.
Figure 2 shows a detail of a single 4k frame taken in RS mode with an integration time of 40ms and low-dose conditions such that SEEs appear spatially separated on the detector. Each SEE is confined to an area of about 3x3 pixels. The intensity of the SEEs follows the expected Landau distribution. With a noise floor level of about 12 counts and a much higher most-probable SEE event intensity, a SEE can be easily separated from the background.
Figure 3(a) shows the result of counting localized 120kV SEEs for several hundred frames. The sample structure is clearly observable. The Fourier transform of the ultra-low dose exposure of polycrystalline Au using 200kV PE energy in (b) shows the 2.35 Å structure factor already at a dose of only 0.6 electrons/Å2. Also note the flat appearance of the Fourier spectrum with the amplitude of high and low spatial frequencies similarly represented. Figure 4 shows the effect of a switchable capacitor to adapt the dynamic range of the sensor to the requirements of the experiment. While a certain electron dose results in saturated pixels in (b), the same dose can be faithfully recorded by activating an additional capacitor in each pixel for a threefold increase in available dynamic range (c).
Summary and Outlook
The presented measurements indicate counting mode acquisition scheme is feasible with the studied indirect CMOS camera system due to its single electron sensitivity and fast acquisition rate. With its high dynamic range, it is also well suited for collection of diffraction data, as required for new methods like MicroED . Future research will focus on the algorithms needed to speed up the extraction of SEEs.
 W Kühlbrandt, Science 2014, p. 1443-1444
 M Stumpf et al, Microsc. Microanal. 16 Suppl 2 (2010), p. 856
 JA Rodriguez and T Gonen in “Methods in Enzymology Vol. 579” p. 369-392
Marco Oster1, Reza Ghadimi1, Andreas Wisnet1, Hans Tietz1
1Tietz Video and Image Processing Systems GmbH, Gauting, Germany
Tietz Video and Image Processing Systems GmbH
Tietz Video and Image Processing Systems GmbH