This paper presents in-pixel and near-pixel processing circuit techniques for multi-mode processing CMOS vision sensor. The new design includes a checkerboard patch-based pixel circuit and column-shared frame difference processing circuits that detect a patch-level temporal change and generate lb event signal. In-pixel time memory circuit is proposed to process a temporal derivative of the event in both area and power efficient way. The 324× 252 vision sensor is implemented in 0.18 μ m CMOS process and consumes 4.79 μ W @12.5fps showing FoM of 42.2pJ/pix. frame in the patch mode operation for frame difference-based motion detection and temporal derivative generation.
ACKNOWLEDGMENT This work was supported in part by the National Research Foundation (NRF) of Korea under Grants 2021R1A2C4002496 and 2019R1A5A1027055, and COCOSYS in JUMP2.0, an SRC program sponsored by DARPA. The EDA tools were in part supported by IDEC.