Dual Conversion Gain
# Dual-Gain Pixel HDR Architecture
A single conventional pixel has to choose one conversion gain — how much voltage change it produces per unit of collected charge — and that single choice forces a tradeoff between capturing bright highlights without saturating and capturing dark shadow detail without the signal disappearing into noise, which is exactly the limitation dual-gain pixel architecture exists to escape. A high-gain setting amplifies small amounts of charge into a usable, distinguishable voltage signal, which is exactly what dim scene areas need to avoid disappearing into sensor noise — but that same high gain saturates and clips long before bright areas finish accumulating their full charge. A low-gain setting handles bright highlights without clipping, but the same small amount of charge from a dark area produces too small a voltage signal to distinguish reliably from noise. A single fixed gain setting can only ever be a compromise between these two competing needs.
Dual-gain pixels solve this by reading out each pixel's signal through two different gain paths simultaneously — one high-gain reading optimized for shadow detail, one low-gain reading optimized for highlight detail — and then combining the two into a single image that captures a dynamic range wider than either reading alone could achieve. Because both readings come from the same physical pixel and the same single exposure, there's no motion artifact or timing mismatch the way there would be combining two separate photographs taken at different exposure times. The high-gain reading supplies accurate shadow detail where the low-gain signal was too small to trust, and the low-gain reading supplies accurate highlight detail where the high-gain signal had already clipped — merging the two gives a combined signal with effectively wider dynamic range than a single-gain pixel could ever produce.
Implementing dual conversion gain requires each pixel to carry extra circuitry to support two gain settings and the readout logic to capture and combine both readings, which is exactly the kind of additional circuitry that stacked sensor architectures make practical by moving more processing onto a dedicated logic die rather than crowding it alongside the photodiode. A single-die sensor with limited area for both photodiode and readout circuitry would struggle to fit the extra gain-switching and dual-readout logic dual-gain pixels require without sacrificing fill factor; a stacked design's separated, more spacious logic die absorbs that extra complexity without competing with the photodiode for area the way it would on a conventional single-die sensor.
Deep trench isolation keeps light-generated charge confined to the correct pixel once captured; dual-gain pixel architecture is a separate improvement layered on top, extracting more usable dynamic range from whatever charge a pixel does correctly collect — together with stacked die separation, these three techniques represent a coordinated set of strategies all converging on the same underlying goal of extracting more accurate image information from an ever-shrinking pixel.