The resolutions and frame rates of CMOS image sensors have increased greatly to meet demands for higher-definition video systems, but their design may soon be obsolete. That’s because photodetectors and signal processors lie in the same plane, on the substrate, and many pixels must time-share a signal processor. That makes it difficult to improve signal processing speed. NHK researchers developed a 3D parallel-processing architecture they call “pixel-parallel” processing, where each pixel has its own signal processor. Photodetectors and signal processors are built in different vertically stacked layers. The signal from each pixel is vertically transferred and processed in individual stacks. 3D stacking doesn’t degrade spatial resolution, so both high resolution and a high frame rate are achieved. 3D stacked image sensors have been reported previously, but they either didn’t have a signal processor in each stack or they used TSV/microbump technology, reducing resolution. NHK will discuss how photodiode and inverter layers were bonded with damascened gold electrodes to provide each pixel with analog-to-digital conversion and a pulse frequency output. A 64-pixel prototype sensor was built, which successfully captured video images and had a wide dynamic range of >80 dB, with the potential to be increased to >100 dB.
Illustration (a) shows a schematic diagram of the 3D integrated CMOS image sensor; (b) shows a conceptual diagram of the image sensor pixel; (c) is a cross-sectional scanning electron microscope image of a bonded CMOS image sensor pixel with no voids observed at the bonded interface and with the upper layer thinned to 6.5 µm; and (d) is a photograph of the bonded CMOS image sensor array, where 60-µm-square photodiodes (PD) are stacked on inverters.
(Paper #4.2, “Three-Dimensional Integrated CMOS Image Sensors with Pixel-Parallel A/D Converters Fabricated by Direct Bonding of SOI Layers,” M. Goto et al, NHK.)