Neuromorphic Silicon Retinas: Analog and Digital Models

Abstract

Silicon retinas are artificial retinal devices designed to replicate the functionality of the biological retina using silicon-based technologies. These retinas aim to address visual impairment by capturing and processing visual information, serving as a key component in the development of visual prosthetics and artificial vision systems. While various artificial retina implementations exist, there is an ongoing need for a more realistic model that closely mimics the structure and function of the biological retina. The development of precise artificial retinas holds great significance owing to their potential to restore vision, improve visual prosthetics, and enhance computer vision systems.

This thesis, therefore, presents a biologically more accurate retina model implemented using analog circuits. The proposed retina incorporates spatio-temporal filtering in the outer plexiform layer (OPL), luminance adaptation, a contrast gain control mechanism, tonic and phasic cells, and spiking. The center-surround structure of the outer plexiform layer (OPL) enhances the retina's visual processing capabilities, acting as an edge and movement detector at the same time. The integration of contrast gain control and luminance adaptation ensures adaptability to varying light conditions, while the inclusion of tonic and phasic cells enriches visual processing capabilities. Moreover, the thesis also introduces a digital retina model on an FPGA, aiming to replicate biological features effectively. These models create biologically more realistic, and highly effective retina models for visual prosthetics, contributing to artificial vision system advancements and potential solutions for vision-related disorders.