This might be a stupid question, but I have just been interested in low light photography and how these new cameras have clean ultra high ISO and such. Why don’t these low light cameras use bigger sensors too? From crop sensor to full frame you gain a full stop of light. Why not just make the sensor even bigger and collect even more photons?
With a large sensor
- the sensor is more expensive (the production cost of a sensor grows at least as fast as its physical area).
- the whole device is bigger, and requires bigger lenses, that add to the overall cost.
There are cameras with sensors larger than full-frame, not mentioning the instruments in any modern astronomical observatory, that are mostly a huge sensor(*) coupled to a huge lens.
(*) Not always for the visible light range, though
What happens when you double a camera in all dimensions?
- the basic image stays the same (once you adjust the focus plane)
- the aperture number stays the same
- the camera gathers 4 times more light but distributes it over 4 times the sensor area. Which means that for the same exposure, you need the same ISO value
- with 4 times the area per pixel (assuming equal MP count), photon noise gets reduced by a factor of 4. However, thermal noise stays the same per sensor area, and sensitivity is the same per area. That means that the kind of noise that is countered by "long exposure noise reduction" will stay similar.
- unsharpness doubles
- the camera and lens weigh 8 times more
Now the tricky thing is the "unsharpness doubles". The photon gains by increased sensor size cannot be achieved otherwise if you are already at maximum feasible aperture, like f:1.4. If you don't have scenes that allow taking the aperture to its maximum (like not being able to get both eyes and nose, or eyes and eye glasses sharp), that additional load of glass needed to support a large sensor is wasted with regard to low-light performance. You'd be better off with a smaller sensor allowing you to take the aperture to its sensible maximum. Also once you get into long exposure noise reduction ranges, the gains by a larger sensor are diminishing.
The issue is not the number of pixels in the frame. If one wants higher sensitivity of each pixel, then the area of the pixels need to be increased. For many fab strategies, which will result in a larger well, which will increase the quantum efficiency of the sensor.
So full frame vs DX or other formats does not make the difference that is realized with an array which has fewer pixels per given area.
Of course this can impact the cost of optics for a similar system performance.
Stated differently, the sensitivity curve (V/uJ/cm**2) has three nominal regions. The dark area, the linear area and the saturation area. The linear area is bounded by the Noise Equivalent Exposure (NEE) and the Saturation Equivalent Exposure (SEE). Larger wells, and more sensor area on the array (as opposed to mask) tends to expand the linear area of the curve. It also depresses the dark area. The dark area of the saturation curve is noise limited, which is why many cameras with high sensitivity will also perform noise reduction on low light exposures.
Sensitivity is not the only metric, and noise is a system limiting factor. And for a given sensor, the sensitivity can be modified with gain, but that will impact system noise.