Siliconfile applied for a patent US20100252718 for a 4T-4S pixel structure which simplifies bad pixel recovery work in case a group of 4 shared pixel fails. It does so by arranging the pixels in such a way that the distance between the pixels of the same color grows, as compared with the regular linear or square sharing:
I'm not sure about advantages of bad pixel correction, but this structure creates at least two problems:
- There are 4 different kinds of transfer gates, instead of one or two in the prior art. It's harder to optimize the process with more kinds of Tx gate.
- There are two different pixels for each color, like two Gr, two Gb, etc. differing by the corner where Tx gate is located. It creates a nightmare with shading and color crosstalk correction across the sensor area.
TSMC patent application US20100252870 talks about dual-depth STI usege, so that one STI is optimized for periphery circuits, while the other one is for pixel area. The STI in pixel region is made shallower than that in the periphery region to reduce dark current. I think TSMC would have a serious problem with prior art here: