Digital Imaging

High-resolution, digital imaging tools help you to understand ultrastructure of biological and inorganic specimens.

Overview Products Media Library Research Spotlight Publications Resources Back to top

There are a number of options and technologies available for digital imaging in transmission electron microscopy applications today. Traditionally, high energy electrons could not be directly exposed to a sensor due to excessive damage so the beam is first exposed to a scintillating film that converts the incoming electrons into light (photons). These photons are then transferred to the sensor, either through a series of optical lenses or a coupled fiber optic plate. Finally, the light is collected by a sensor where the image is created pixel-by-pixel based on the amount of light detected at each position in the sensor.

Coupling options

The lens and fiber coupling options both transfer light to the sensor but with different performance tradeoffs. The fiber coupling, while technically more difficult to manufacture and bond to the sensor, offers better performance with fewer distortions and loss of signal-to-noise ratio. Lens coupling is simpler to implement, but suffers from a greater loss of signal and lens aberrations, such as vignette that produces distortions and signal loss depending on the optical path. Fiber coupling avoids these problems as the fibers themselves do not have any position-dependent performance variation.

There are two basic types of sensors used for scintillator-coupled devices: charge-coupled devices (CCD) and complementary metal-oxide semiconductors (CMOS). Each type of sensor has inherent advantages and limitations given the differences in their operation.

CCDs operate by transferring charge from one pixel to another (historically compared to a ‘bucket brigade’ transferring buckets of water (charge) to the edge of the sensor where it can be read out (converted to a digital signal). There are two design options for transferring the charge between pixels. Interline CCDs have a protected (covered) area for each pixel where the accumulated charge can be transferred and then read out while the next frame is being accumulated. The advantage of this is that the sensor has 100% duty cycle; there is no need to blank the beam in the column. The other type of CCD (frame transfer) has 100% light sensitive area per pixel (100% fill factor) so captures all the information in space. This means that as the sensor is read out, the beam must be blanked (shuttered) by the TEM. 

CMOS sensors by contrast do not transfer charge between pixels; there are A/D converters at each pixel (and transistor circuitry to perform this) that convert the charge to a digital signal, and each pixel is then addressed and read out on a per row and column basis. One advantage is that there is no need to shutter the beam during frame read out, however this comes at the expense of reducing fill factor, as the transistors in each pixel reduce the photon-sensitive area.

Direct detection

There is a class of sensors that have been recently developed that can be exposed directly to the electron beam to collect signal, eliminating the need for a scintillator and optical transfer of information. These are referred to as direct detection, and are based on a CMOS sensor design, similar to the coupled CMOS sensors but design differences increase the radiation hardness of these sensors to varying degrees. 

Due to the elimination of photon conversion and the need to optically transfer signal, direct detection sensors typically produce a much higher signal-to-noise ratio output compared to scintillator-coupled sensors. This is best quantified with detective quantum efficiency (DQE), which is represented by the ratio of the output signal-to-noise ratio (SNR) compared to the input SNR. 

The DQE of direct detection sensors can be up to a factor of 5x greater compared to scintillator-coupled sensors when operating in a linear accumulation mode. Electron counting mode, which boosts the DQE even further by eliminating both read noise as well as the variability in deposited energy per incident electron, both of which significantly reduce SNR in the output signal.