> AFAIK you want the biggest pixels you can for the resolution you want in order to cut down on noise, no?
You want the biggest sensor, but nowadays, we're targeting smallest pixels at the same time.
The basic idea is that every pixel will have its own quantum and thermal noise, the noise on every pixel is independent. This means that to have the best measurement of the "real" light brightness at one place, you make the median of all the surrounding pixels (that's called binning). This obviously reduces resolution, hence we're making smaller pixels. (Nowadays we make better algorithms than median, but median is still pretty good)
Lately, one of the major reasons we've been losing light on the sensors, is the RG/RB matrices, we loose light at every transition from one color to another.
So, you combine those two pieces of information, about pixel binning and about color filtering, and that makes you modern sensors: 108MP (in Samsung Galaxy S22 Ultra case), where the RG/RB matrix only has 12M"spots", pixels being 9-binned (so you "create" one pixel, by measuring 9 at the same time).
So Galaxy S22 Ultra's 108MP is kinda "just" 12MP with much better light sensitivity.
In real life, it's a bit better than just 12MP, because even if you are in a R-only region, you can still find the changes of global brightness, and have sharpness further than 12MP. Hence Samsung touts "combining 9-binned image with non-binned image"