I think BSI more about increasing the full well capacity rather than just preventing light from seeping between pixels.
Have to admit I was surprised that BSI is just now arriving at this level of sensor - my naivety - assumed it would have been here long ago.
My presumption was that gapless microlenses were supposed to make BSI unnecessary, but perhaps there's some marginal improvement using both.
Maybe one of our resident EE's can answer this question.
Hmmm...I'd like to hear more explanation of this: I had thought that fwc was not strongly dependent on the surface area.
FWC is dependent upon surface area...the surface area of the photodiode. With an FSI design, some of the pixel area is dedicated to wiring and transistors, which thus necessitates that the photodiode (the actual light-sensitive part of the silicon) become smaller. By flipping the die and etching on the back, the photodiodes can become larger, since the wiring is all on the other side. That increases FWC, which means for the same size sensor, with even the same size "pixel pitch", your actually gathering more light.
There are other benefits with BSI. With FSI, the structure of a pixel ends up being relatively deep. There are layers of wiring and transistors built up around the photodiode. The photodiode sits at the bottom of what is basically a physical "well" (Technically, the "photo well" refers to the potential well, the electronic charge capturing capacity of a photodiode, not a physical thing...but there is a physical "well" as well). The depth of that physical well affects the amount of incident light that can actually be captured, or more specifically, the amount of incident light on the pixel that actually reaches the photodiode and frees and electron. Microlenses helped with that, by bending the light at the periphery of the pixel around the wall of wires around the photodiode. Microlenses aren't perfect, though, and really need to be aspheric, to fully direct all light onto the photodiode. As such, even with them in place, you lose some light to heat as they strike the wiring walls, or they reflect off the walls and hit exposed substrate and don't actually reach the photodiode, etc.
Lightpipe designs improved FSI designs, by filling the well with a highly refractive material. This helped bend the light and keep more of it focused on the photodiode. Lightpipe designs also lined the wiring walls with highly reflective material, which combined with the high refractive index material, focused a lot more light onto the photodide with FSI designs. That's more complicated, though, and still wasn't as efficient as simply flipping the wafer and etching the photodiodes on the other side.
With BSI, you can basically flatten the light-sensitive side of the sensor. There is no longer a physical "well"...all you have is a microlens, a color filter, and the photodiode. It's a very short, flat "stack", the photodiodes are larger, basically covering the entire surface area of the sensor. It gives you the highest efficiency thanks to the very high fill factor.
Given what I've seen so far of NX1 data, the BSI design gives it about a stop better sensitivity than any Canon sensor. I don't know if it's as good as an Exmor, but it doesn't seem to have any banding that I could see...the noise is very random.