A 1984-Pixels, 1.26nW/Pixel Retinal Prosthesis Chip with Time-Domain In-Pixel Image Processing

Abstract

The retinal prosthesis (RP) chip restores basic visual function in patients with retinal degenerative disease and improves patient's oriental skills and ability to localize objects. Since the RP chip has a low resolution due to a limited number of pixels, image processing such as edge-extraction is necessary to improve visual acuity [1]. The edge-extraction is also essential to generate power efficient stimulation patterns in the RP chip where heat generation due to power consumption should be minimized [2]. Since each pixel generates a photocurrent according to the intensity of the input light, the edge-extraction can be performed by directly processing photocurrents of neighboring pixels. However, current-domain analog processing requires multiple current branches for mirroring and consumes large static power [3-4]. Counter-assisted all digital processing with light-to-frequency converter demonstrates low power consumption for the edge processing, but it requires area-consuming multi-bit registers and large computational logic circuits [2]. In this paper, we propose a time-domain edge extraction technique for power and area-efficient in-pixel image processing for RP chips.