The other main source of image noise is the sensor itself, and in most cases this is unavoidable. As we've seen in a previous tutorial, the individual photocells on a digital camera sensor are incredibly small, especially with high-resolution compact camera sensors. In the diagram below I've drawn a series of camera sensors at approximately their actual sizes (if your monitor is set to 96dpi).
Most compact camera sensors are of the 1/2.3 inch type, the smallest size on this chart, but commonly have over 10 million individual photocells crammed into an area of less than 30 square millimetres. These light sensing cells are so small that in low light conditions they may only be collecting a few thousand photons (individual light 'particles') during an exposure, so the level of the electrical signal produced by the cell can be affected by random statistical fluctuations in photon density. This is the main reason that physically larger sensors are much better than smaller ones. The individual photocells are larger and collect proportionately more photons during the exposure, reducing the effects of these statistical variations and producing an inherently higher signal to noise ratio.
The level of noise produced by the sensor and other components in the camera is usually fairly constant and hopefully at a fairly low level. In normal operation, when taking photographs in good light, the level of the signal vastly outweighs the level of noise, in other words the signal to noise ratio is very high, and consequently noise isn't a problem. The problems start when light levels fall. When shooting in low light the level of the signal drops nearer to the constant noise level, producing a lower signal to noise ratio. At extremely low light levels the signal may be entirely drowned out by the noise.
This problem is made worse when shooting at higher ISO settings. When we set a higher ISO sensitivity, what we are really doing is increasing the amount by which the signals from the sensor photocells are amplified. Unfortunately the amplification process can't tell the difference between image signal and noise, and so the noise gets amplified as well. If the signal to noise ratio was already very low then this just produces more noise without improving the image. This is why high ISO images are always more noisy than ones taken at lower settings.
Another type of sensor noise can also be a problem when using longer exposures, usually over a couple of seconds. Sometimes, due to irregularities in the manufacturing process, errors in signal transmission and other factors, the photocells that make up the sensor may not all respond to light to an equal degree, causing single pixels to appear very bright or very dark. The charge build-up over a longer exposure makes this problem more noticeable. Since the pattern of these "dead" or "hot" pixels is usually consistent from one frame to the next this type of noise can usually be remedied by applying a masking filter during image processing. Most modern cameras do this automatically, but it can be a problem on many older models. The shot below was taken on a Fujifilm FinePix 6900 Zoom in 2002, and shows a large number of "hot" pixels. The exposure time was 2.8 seconds at 200 ISO.