6.5 25.3μW at 60fps 240×160-pixel vision sensor for motion capturing with in-pixel non-volatile analog memory using crystalline oxide semiconductor FET

A vision sensor used to capture motion must operate with very low power when used in sensor networks where power is very limited. Frame-based [1] and event-driven [2,3] vision sensors have been reported. The former captures motion from the difference between captured data of a previous frame and that of a current frame; thus, it is difficult to capture motion of a slowly moving object. An event-driven sensor captures motion of a slowly moving object; however, its pixel configuration is complicated, and it is difficult to perform both motion capturing and image capturing. In this paper, we report a vision sensor for motion capturing having in-pixel non-volatile analog memory utilizing a c-axis aligned crystalline In-Ga-Zn oxide (CAAC-IGZO), a crystalline oxide semiconductor based FET that demonstrates very low off-state current [4] and retains captured data of a given reference frame. Although an electronic global shutter image sensor with improved floating diffusion (FD) charge retention characteristics is reported in [5], our vision sensor realizes normal global shutter and motion capturing depending on the presence or absence of differences with respect to a given reference frame in each pixel. The sensor has 3 modes: an imaging mode to output captured data in pixels, a motion-capturing mode to process differential data using an analog processor, and a standby mode to reduce power after motion capturing for each frame. Power consumption is reduced by operating only circuit blocks needed for each mode. The sensor (240×160 pixels), fabricated by a 0.5μm CAAC-IGZO FET/0.18μm p-channel Si FET (no n-channel Si FET) hybrid process shows power consumption of 25.3μW and 1.88μW at 60fps in the motion-capturing and standby modes, respectively, which equal 1/140^th and 1/2000^th of the power consumption of the imaging mode.