>>3216600traditionally, sensors are built with the interconnect (layers of metal conductors that act as internal wiring for the sensor) sandwiched between the microlenses and the light sensitive photodiode (the part of the sensor that detects light).
Especially with small pixel sizes, that interconnect can block photons before they reach the photodiode, resulting in a less efficient sensor. With REALLY small pixel pitches (camera sensors, etc), the odds of a photon making it to the photodiode is only around 60% or even less. That's pretty bad in low light where your photon-per-second count is really low.
You can improve that by flipping the photodiode and interconnect 180 degrees (so that the photodiode sits in front of the interconnect and directly behind the lenses) and illuminating the sensor from the "backside". Photons then have a direct path from the microlenses to the light sensitive layer, improving efficiency from around 60% to upwards of 90%. You then get drastically improved low light sensitivity compared to the exact same sensor with a traditional layout.
This all is slightly less true for massive full frame sensors, where the proportionate obstructing size of the interconnect layer is much smaller compared to the size of the pixels. The OTHER benefit of a BSI sensor is that you increase the apparent f/stop of the microlenses on your sensor by shortening the distance that light has to travel before it reaches the photodiode. Your microlenses on a traditional sensor might have an apparent f/stop of, say, f/2.4. That means that your sensor might actually be slower than the lens mounted to it. Manufacturers cover that up with secret iso boosts built into the firmware, which means lessened image quality. A BSI sensor might improve the apparent f/stop of the same microlens to f/1.4, meaning less ISO boosts and better quality.
