Quantum Efficiency, Read Noise, and why they are important

In essence CMOS sensors convert light falling on them in to numbers representing brightness levels, and their location on a grid of tiny little detectors called pixels.

Each step in a “unit” of brightness on a pixel is expressed as a positive number. The brighter, the higher the number, so for an 8bit pixel, 0 is black and 256 is white. For 12bit there are 4096 levels. These discreet “digital steps” in brightness are often referred to as “ADU” or “Analogue to Digital Units”.

In an ideal world, 1 photon would equal 1 ADU, which is 100% efficiency, but perfect efficiency isn’t possible, so we try to get as close to 100% as we can.

In a nutshell, Quantum efficiency, often referred to as “QE”, is a measure of sensitivity, expressed as a percentage, which describes how often a photon landing on a pixel gives off an electron which is then read out as a change in brightness expressed as a number – an ADU.

For a sensor with 70% Quantum efficiency, 7 out of 10 photons landing on a pixel will displace an electron, which has a chance of being read out as “signal”, amongst noise. The remaining 3 photons will not cause an electron to be displaced, and the information is essentially lost forever. The higher the QE, the more sensitive the camera. Modern CMOS sensors tend to have very high QE compared to CCD cameras versus price, however CMOS sensor manufacturers often do not release this information and express values in a relative sense.

Read noise can be thought of as a “threshold of uncertainty”, under which a pixel cannot accurately convert how many electrons it has collected into Analogue to Digital Units or ADU (Grey Values). Read noise is generated when a charge is read from the sensor and converted into ADU. Read noise is measured in electrons, which are ejected when a photon (the smallest measurable unit of light) lands on a pixel. The lower the read noise, the more accurately a pixel can record and differentiate faint signals from background noise. This improves the camera’s ability to detect weak signals. (Read noise decreases with gain, but before you think “well, I’ll just increase the gain to maximum” be aware that this would also decrease the number of available grey levels the pixel is able to record). If 1-2 photons land on a pixel with a read noise of -6e electrons, then you may not be able to spot such a weak signal amongst a much bigger amount of noise.

You will need more photons to bring your signal above this “threshold” of noise, and therefore a longer exposure time is required to isolate your signal (say the light from a faint star, or a faint area of nebulosity) from the background noise.

There is another way to get a good signal to noise ratio, provided your read noise is very low, and that’s to take a large number of short exposures, and take an average of them all, to determine more accurately what is noise and what isn’t. This technique works with CMOS cameras where the read-noise per frame is very low.

Of course, with higher sensitivity (QE) a greater proportion of photons landing on the pixel are being converted into electrons, therefore it can be said that the less the read noise, and the higher the sensitivity, the better a camera performs.