EMCCDs contain an extra structure in the serial register that multiplies the pixel charge through an avalanche multiplication mechanism. The image area of the EMCCD, where the photons are detected, is of conventional design.
Sub-electron read noise is easily achieved. Scientific observations where the intrinsic noise of the detector is the limiting factor will benefit from the use of EMCCDs. In higher signal regimes , where the observation is limited by statistical photon noise then there will be no benefit. Indeed, the EMCCD will actually perform worse than a conventional CCD in the photon noise limited regime, due to so called “multiplication noise”. This noise appears as higher than expected variances in flat field images, it is less important at low signal levels. It is possible, however, to eliminate multiplication noise by using a photon-counting analysis of the images. Here a threshold is set above the read noise of the detector and any pixels exceeding this threshold are counted as 1 photon. This will only work in very low signal regimes where there is a negligible chance of two photons landing in a single pixel.
The video below demonstrates the very-low noise levels of an EMCCD. Here, a pinhole has been placed just in front of the detector. The pinhole is illuminated by a distant LED whose flux is so low that much less than one photon, on average, is en-route from LED to CCD. As the LED light passes through the pinhole, a diffraction pattern is generated, consisting of a central Airy disc surrounded by a sequence of faint rings. This is what is expected based on the wave nature of light. This faint ring structure can be seen gradually appearing in the right hand panel, which shows the accumulated sum of the individual, thresholded EMCCD (i.e. photon counted) frames. These individual frames are shown in the left-hand panel. Here the ring structure is not really visible, instead what we see is a sprinkling of single photon events.
The above video sequence is remarkable in that it simultaneously demonstrates both the wave nature and the particle nature of light.
EMCCDs should be the choice for any “photon-starved” application. This starvation could be due to very low source-flux levels or it could be due to very high frame rates. There are two obvious applications: Adaptive Optics wavefront sensors, where frame rates of ~1kHz are needed and High Dispersion Spectroscopy where the light from an already faint source will be spread by a grating across many pixels. An example of a spectrum taken with an EMCCD is shown below. The target (SDSS J1433+1011) is so faint that individual photon events are visible in its spectrum. The image was taken during a full moon so the background is contaminated by a scattering of moonlight photons.
Another example of an EMCCD observation is shown below. This sequence of images was taken of the Crab Nebula pulsar with an e2v CCD60 on the 4.2m William Herschel Telescope. These are raw frames, no phase-binning has been used. The 30Hz flash from this rotating Neutron Star is clearly visible in every 7th frame. The third frame also gives a hint of the weaker secondary pulse.
One widely used EMCCD is the E2V CCD220. It is an 8-output device intended for wavefront sensing in adaptive optics systems. It can achieve sub-electron read noise at a frame rate of 1kHz and will be used extensively on the European Extremely Large Telescope project.
Further information:
SPIE, 2008, Modelling the suitability of EMCCDs in spectroscopic applications
MNRAS, 2011, On the use of electron-multiplying CCDs for astronomical spectroscopy
Thesis 2010, Astronomical Spectroscopy with Electron Multiplying CCDs