The first sCMOS (scientific Complementary Metal Oxide Semiconductor) cameras were introduced some 10 years ago and are now the dominant sensor technology used in microscopy imaging cameras. In recent years, a new generation of sCMOS camera models based on back-illuminated sCMOS sensors has appeared. This new iteration of sCMOS camera technology promises even higher sensitivity than prior designs, and therefore may displace previous generations of front-illuminated sCMOS cameras and further push EMCCD (Electron Multiplying Charge Coupled Device) cameras to more niche applications.
The reason for the success of sCMOS cameras for microscopy, and indeed the wider majority of imaging applications, is largely due to the highly parallel architecture of the CMOS sensor. CMOS sensors are active pixel sensors, meaning that each pixel element has its own on-pixel amplification circuit that then feeds a column amplifier before being read out. This arrangement allows for a very fast readout from the sensor, ideal for high speed imaging and larger sensor sizes with wide fields of view.
CCD devices do not have on-pixel processes, meaning that the readout process of CCDs is serial in nature — restricting speed and consequently sensor size. Interline CCD and frame transfer CCD variants did provide some increase in speed, but the CCD design was inherently limited in speed compared to CMOS. However, early CMOS sensors were not developed for scientific applications and were associated with low quality imaging products such as camcorders. CCD cameras provided good image quality and they remained the preferred choice for microscopy applications for many years.
The breakthrough for CMOS in microscopy came from the development of “scientific” CMOS (sCMOS) by Andor Technology, PCO and BAE Fairchild Imaging. The first of these cameras was the Andor Neo 5.5 sCMOS camera. It offered a 5.5-megapixel sensor with 6.5 μm pixels and 60% quantum efficiency. It could operate in a rolling shutter mode whereby exposure proceeded row-by-row, for the highest frame rates and lowest noise. A global shutter mode provided a “snapshot” mode that avoided image smearing, at the expense of frame rate and noise level. Additionally, a dual amplifier configuration meant that high dynamic range images up to 16-bit could be achieved. In this dual-amplifier arrangement, outputs from the high gain channel and a low gain channel could be stitched together in a single full-range image. This resulted in a camera that was faster, had better resolution, lower noise, a wider field of view and a higher dynamic range than the common interline CCD cameras at that time.
Until recently all sCMOS cameras had what is called a “front-illuminated” sensor architecture (Figure 1). In a front-illuminated sensor each pixel has a microlens to help direct the light past a wiring network and through to the photosensitive region of the underlying silicon where photons are converted to electrons. Several revisions to these sensors over time saw sensitivity improve due to quantum efficiency (QE)-the efficiency at which photons are converted to electrons increased from ~60% up to 82%.
A process called “back-thinning” or back-illumination had been used for CCD cameras to provide even higher QEs, but it is only in recent years that this process has been successfully implemented in sCMOS sensors for microscopy. With back-illumination, the sensor has essentially been turned upside down, then processed to remove much of the bulk silicon substrate. Incident light can then directly reach the photosensitive area without obstruction from the pixel circuitry.
EMCCD cameras effectively have <1 e- using EM gain.
As a result of back-illumination, sCMOS cameras such as the Andor Sona models can offer QE of up to 95%-on par to that of back-illuminated CCD and EMCCD cameras. This means that within the visible spectrum almost all the incident light will be converted into electrons. Despite having the same QE as EMCCD cameras they do not offer the same sensitivity.
Signal = QE at wavelength x number of photons
Dark Current = noise generated by the sensor itself without signal
Read Noise = noise generated during the sensor readout process
ENF = For EMCCD an additional noise factor of √2 is due to the EM amplification, however this process can reduce read noise to below 1e-.
EMCCD cameras can exploit electron multiplication (EM gain) to amplify the signal many times before readout via a process called impact ionisation. This process introduces an additional noise factor, but it allows EMCCD cameras to mitigate read noise and operate at light levels well below that of back-illuminated sCMOS cameras. This is particularly relevant to single molecule studies with very low photon regimes and why EMCCD cameras are still recommended.
Effective cooling of the sensor is another consideration for all cameras including back-illuminated sCMOS models. Cooling can reduce the dark current — the noise generated by the sensor itself without any signal — and reduce hot pixels, which are pixels with higher dark current. As exposure times increase, cooling becomes more important. Vacuum sealed sensors help provide deeper cooling while liquid cooling, if available, may be used to obtain the deepest cooling and, thus, the lowest dark current. A further benefit of liquid cooling is the elimination of cooling-fan-induced vibration. This can be important for fast, high magnification imaging applications, e.g. localisation-based super resolution.
Another aspect of cameras that is related to sensitivity and resolution is pixel size. In simple terms, a larger pixel has a larger area to collect more photons; therefore it can allow for an improved signal to noise ratio (SNR). However, if the pixel size is too large, then we will not be able to resolve the full detail of the image. Conversely, going too small on pixel size to achieve high resolution will impact the SNR. Ideally, we need to have a pixel size that will sufficiently oversample the image so that we meet Nyquist criteria for the optical system. For a microscope this can be expressed as:
Some example combinations of typical objectives matched with camera pixel size are shown in Table 2. A 10-11 μm pixel size will be suitable for 100× while a 6.5 μm pixel will be suited for the widely used 40× and 60×. It is possible to use additional post objective magnification to reduce the effective pixel size and improve oversampling. Pixels can also be binned-i.e. sum the values of a number of pixels into larger “superpixels”. Using 2×2 binning july provide a SNR improvement by the square root of 2 at the expense of spatial resolution.
Sensors are composed of millions of small silicon pixel elements, so they are inherently fragile. Moisture and other airborne contaminants will quickly degrade the sensor, meaning that protecting the sensor is essential. One way of doing this is “backfilling” the sensor enclosure. In this process the atmospheric air is replaced with a positive pressure of an inert dry gas such as nitrogen or argon and sealed using O-ring seals. This is the most common approach but inevitability the seals will fail over time, gas will be lost from the chamber (may be seen as condensation on the camera window) and a camera seal service will be required.
The best strategy for long-term protection of the sensor is, therefore, a permanent vacuum seal. The only back-illuminated sCMOS cameras available for microscopy with a permanent vacuum seal are the Andor Sona camera series.
Improved sensitivity will continue to be a key driver for microscopy cameras since this helps allow lower exposure times, lower illumination intensity and lower fluorophore concentrations. All of which help to provide research scientists with more accurate information of the underlying cell biology under study. The increase in QE provided by the latest generation of back-illuminated sCMOS cameras is, therefore, a welcome addition to the strengths of sCMOS-speed, field of view and dynamic range which have positioned sCMOS as the detector technology of choice for most microscopy applications.