A 4.9Mpixel Programmable-Resolution Multi-Purpose CMOS Image Sensor for Computer Vision

Abstract

Today's mobile devices are capable of multiple tasks, including viewing and computer vision. However, these tasks are often implemented using multiple cameras due to divergent requirements, resulting in increases in form factor. Higher resolutions and frame rates are required to take pictures and video, while lower resolutions and frame rates are preferred especially for always-on computer vision to save power. To reduce the form factor, this paper presents a single-chip solution where both high-resolution viewing images and lower-resolution computer vision images are generated from one CMOS Image sensor (CIS). Full field-of-view programmable image resolutions are made possible with floating-diffusion (FD) binning through serial switches. In-frame dynamic voltage and frequency scaling (DVFS) is applied to minimize power at lower pixel resolutions and frame rates. 1/2, 1/4, 1/8, 1/16 and 1/192 binning modes are demonstrated using a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$2560(\mathsf{H})\times 1920(\mathsf{V})$</tex> pixel array that, at 1fps, consumes <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$790\mu\mathsf{W}$</tex> for full-color VGA and 120µW for motion-detection (MD) mode.