http://www.ing.iac.es/~smt/CCD_Primer/CCD_Primer.htmSimon has kindly granted me permission to use this material -- thanks, Simon!
CCDs are not perfect detectors -- they do not record every single photon which strikes them. One reason is the change of index of refraction between air and the silicon itself.
One can reduce reflective losses by anti-reflection coatings. The idea is to cause destructive interference between a light ray which bounces off the front surface of the coating (labelled "A" in the diagram below) and a light ray which bounces off the back surface of the coating. If we can arrange it so that these reflected light rays interfere correctly, we can in essence prevent such reflection from taking place.
The first ray, "A", has its phase shifted by half a wavelength because it is bouncing off a material with higher refractive index than air. If we can cause the second ray, "B", to have an overall phase shift of zero, then the two rays will be perfectly out of phase and destroy each other.
So, in order to create destructive interference between rays "A" and "B", we want
Note that one can arrange for perfect destructive interference only at a particular wavelength; there will be some reflection at other wavelengths. For CCDs, the designer typically chooses a wavelength near the middle of the visible band, around 550 nm. This EEV chip, on the other hand, had its reflectivity optimized for 400 nm.
A good coating will have an index of refraction between that of air (n1 = 1.0) and silicon (n3 = 3.6). The best results occur when the coating's index is the square root of the silicon's: n2 ~ 1.9. One common choice is hafnium dioxide.
Our SBIG ST-8E camera has a Kodak KAF1602E chip in it, and our SBIG ST-9E camera has a Kodak KAF0216E chip. As you can see, the two have similar spectral response.
However, even CCDs start to distort the incoming signal when it becomes too large. When the number of electrons in a single pixel starts to approach the full-well capacity of the device, some of them are lost to inactive areas of the silicon or to neighboring pixels.
The gain on most CCDs is set so that the highest possible pixel value (65536 on RIT's camera) is near the full-well capacity; you should be wary of any values close to this limit. Below is a graph of the deviations from linear response for one of the 'z' CCD chips used in the Sloan Digital Sky Survey, as measured by Masaru Watanabe and Mamoru Doi.
- Stars "A" and "B" have true magnitudes 10.0 and 13.0, respectively. You take an image which is long enough that the response to star "A" is non-linear. What would the difference in the instrumental magnitudes be?
- less than 3.0 mag
- exactly 3.0 mag
- more than 3.0 mag
When the number of photons hitting a single pixel becomes much larger than the full-well depth, some of the pixels start to "bleed" into neighboring pixels in the same column:
The channel stops prevent electrons from migrating across columns, so they move up and down along a column, leaving a "bleed trail":
Here's an example of a CCD image with big bleed trails. There's not much one can do to avoid them when bright stars are embedded in faint nebulosity.
Here's a sample dark frame from the RIT CCD camera. The exposure time is only 0.12 seconds:
Another dark frame, 30 seconds in length, shows many more hot pixels.
How can one deal with this source of false signal?
Compare the raw 30-second image below with the same image after subtracting the "master dark frame". Click on the raw image to see the dark-subtracted version.
Copyright © Michael Richmond. This work is licensed under a Creative Commons License.