Silicon Retina with Ganglion Cell Spiking Outputs as Neural Prosthesis
Around 2006, researchers developed a silicon retina implemented as an analogue VLSI chip that modelled four primary retinal ganglion cell types and generated 3,600 spiking outputs, designed as a neural prosthesis matched to the physical dimensions of the biological retina.
Background
By the mid-2000s, researchers working on vision restoration faced a problem that had nothing to do with surgery. The electronics did not speak the right language.
Early retinal prostheses mostly worked by converting a camera image into a grid of electrical pulses, then stimulating retinal cells in a pattern that matched the pixels. Crude but logical. The trouble was that the retina does not send pixel data to the brain. It sends something far more compressed and selective: the outputs of specialised cells called retinal ganglion cells, each of which responds to particular features such as edges, movement, or changes in contrast. Bypassing that processing meant the signal reaching the optic nerve looked nothing like what a healthy eye would produce. The brain then had to make sense of input it had not evolved to receive.
At the same time, the hardware itself posed a challenge. Implantable chips need to be small, power-efficient, and physically compatible with the tissue around them. Digital processors of the period consumed too much power and ran too hot for comfortable implantation. Something closer to the analogue, continuous behaviour of real neurons was needed, and building that in silicon was difficult.
What happened
Researchers at the Institute of Neuroinformatics, a joint institute of ETH Zurich and the University of Zurich, built a silicon retina implemented as an analogue VLSI chip. VLSI, or very-large-scale integration, is the process of packing many thousands of transistors onto a single piece of silicon; analogue VLSI means those transistors carry continuously varying signals rather than switching between just on and off. That made it possible to model the graded, ongoing electrical behaviour of biological neurons in hardware, rather than approximating it with rapid digital calculations.
The chip was arranged as a 60-by-60 grid of pixels, giving 3,600 output channels in total. Each channel was designed to mimic one of four primary retinal ganglion cell types, the cells that normally form the final output layer of the biological retina before signals travel along the optic nerve. Rather than simply reporting light intensity, each channel produced spikes, brief electrical pulses, in patterns timed to match what those ganglion cell types would generate in a real eye looking at the same scene. The chip also included both analogue and digital interfaces, which mattered for connecting it to downstream processing or stimulation hardware.
The physical dimensions of the chip were matched to those of the biological retina, a design choice that kept clinical implantation in mind from the start. Whether that would eventually translate into a working implant in a patient remained an open question in 2006; this was a demonstrated device, not yet a clinical one. But the Institute of Neuroinformatics had shown that analogue neuromorphic hardware, hardware modelled on the structure and behaviour of nervous tissue, could replicate retinal encoding at a scale that was at least plausible for prosthetic use.
Why it mattered
The device demonstrated that analogue neuromorphic hardware could replicate the spatiotemporal encoding of the biological retina at a scale relevant to clinical prosthetics. By producing ganglion-cell-type spiking outputs rather than simple pixel arrays, it offered a biologically plausible interface between a camera-like sensor and the optic nerve. This placed neuromorphic retinal prosthetics on a trajectory toward devices capable of restoring functional vision.
Organisations
Institute of Neuroinformatics, Eth Zurich University of Zurich
Sources
- A Retinal Prosthesis Using Spiking Silicon Retina.PubMed Central / National Institutes of Health.Secondary
- A 60×60 pixel neuromorphic retina chip with an analogue and digital interface.IEEE Xplore / International Symposium on Circuits and Systems 2006.Primary source
- Neuromorphic Visual Circuits, Institute of Neuroinformatics, ETH Zurich / University of Zurich.Institute of Neuroinformatics, ETH Zurich.institutional
Cite this page
AI Achievements. (2006). Silicon Retina with Ganglion Cell Spiking Outputs as Neural Prosthesis. Retrieved 2026-08-22, from https://achievements.ai/milestone/first-artificial-intelligent-silicon-retina
@misc{achievements_first_artificial_intelligent_silicon_retina,
title = {Silicon Retina with Ganglion Cell Spiking Outputs as Neural Prosthesis},
author = {{AI Achievements}},
year = {2006},
url = {https://achievements.ai/milestone/first-artificial-intelligent-silicon-retina}
}