Low light is not an ISO setting. It is a photon budget. Quantum efficiency says what fraction of the arrivals become charge. The well and the read noise say whether that charge is still a signal once you count it. Pixel size is the bucket. CFA is a tax. F-number is the window.
In one sentence. Night performance is QE times pixel area times aperture, minus the noise of the readout. Megapixels do not appear in that sentence.
What to look at, in order
- Pixel size. Larger wells collect more. Format and megapixels decide how large they can be.
- Mono versus Bayer. Filters throw light away. Night often wants mono, or RGB-IR with the cut removed.
- Technology. BSI usually helps QE. It is not magic if the pixel is still 1.0 µm.
- The lens. F/1.6 versus F/2.8 is a real stop. A faster lens can beat a fancier die.
SNR curves in a datasheet are the honest document. A slide that says "excellent low light" is not.
The porch light test
A doorbell at dusk has a face in shadow and a sky that is still bright. That is HDR plus low light, not one or the other. A warehouse at 22:00 with one sodium lamp is only low light. Do not buy the same part for both without saying so.
How to use the library
In the sensor library, start with format, then raise pixel size, then decide CFA. In the lens library, start with coverage, then lower F-number only as far as the depth of field allows. If the page has no QE plot, treat "low light" as a question, not a specification.