Retinal implants and visual prostheses

Retinal implants and visual prostheses are devices that produce visual sensation in blind people by stimulating the surviving parts of the visual pathway with electric current. They target conditions in which the photoreceptors have died but the downstream retina and optic nerve remain largely intact — chiefly retinitis pigmentosa and the geographic atrophy form of age-related macular degeneration. Every device fielded so far produces vision built from phosphenes, discrete spots of light with no colour and little detail, and no system has restored acuity anywhere near the legal threshold for blindness.

§How it works

A camera, usually mounted on spectacles, captures a scene. A processor reduces the image to a low-resolution activation map and transmits it to an implanted array, which converts each pixel into a current pulse delivered to nearby retinal neurons. Because the retina is a layered circuit rather than a screen, where the electrodes sit changes what they excite.

Epiretinal arrays sit on the inner retinal surface and stimulate retinal ganglion cells, the output neurons whose axons form the optic nerve. This bypasses all intervening retinal processing, so the signal that reaches the brain is not the code the retina would normally send; axons of passage from distant regions are also stimulated, producing elongated or displaced phosphenes. Argus II was the leading epiretinal device.

Subretinal arrays sit under the retina in the space the dead photoreceptors vacated and stimulate bipolar cells, letting the remaining retinal circuitry perform some of its normal processing. Photovoltaic designs go further: each pixel is a photodiode that converts projected light into local current, so the implant needs no cables and the eye's own movements scan the image across the array, preserving the natural link between gaze and percept.

Cortical devices skip the eye entirely and stimulate primary visual cortex, which makes them the only option for people whose optic nerve is destroyed by glaucoma or trauma. They are also the most invasive, they require a craniotomy, and the retinotopic map they must exploit is folded into a sulcus, so much of it is difficult to reach with a surface array.

§Development history

Electrical stimulation of the occipital cortex was known to produce phosphenes from work in the mid-twentieth century, and Giles Brindley and William Lewin implanted an 80-electrode array over visual cortex in a blind volunteer in 1968, demonstrating that patterned percepts were possible and that the engineering was far from ready. Retinal approaches, which require far less surgery, dominated from the 1990s.

1. 1968
   First cortical visual prosthesis
   Brindley and Lewin implant an array of surface electrodes over the occipital cortex of a blind volunteer, who reports discrete phosphenes at predictable locations.
2. 2011
   Argus II CE-marked
   Second Sight's 60-electrode epiretinal implant is approved in Europe for retinitis pigmentosa, followed by a US humanitarian device exemption in 2013.
3. 2013
   Subretinal photodiode arrays
   Retina Implant AG's Alpha IMS, with roughly 1,500 light-sensitive pixels, receives European approval; the company later ceases operations.
4. 2019–2022
   Commercial collapse
   Second Sight halts Argus II production and nearly winds down; Retina Implant AG closes. Implanted patients lose software support and repair paths.
5. 2020–2021
   Cortical stimulation revisited
   A Utah array in the visual cortex of a blind volunteer supports letter identification, and a 1,024-channel array in monkey V1 evokes recognizable shapes.
6. 2024–2025
   Photovoltaic results and new ownership
   Pixium Vision enters receivership and its PRIMA subretinal system is acquired by Science Corporation; trial results reported in 2025 describe letter and word reading with electronic zoom in most implanted participants.

§Current state

As of 2026 the field is in an unusual position: its two commercially approved devices have been withdrawn, and its most promising results come from a system whose developer went insolvent.

The PRIMA subretinal photovoltaic chip is a wireless array of a few hundred pixels implanted under the macula, driven by near-infrared images projected from augmented-reality glasses. Feasibility work in geographic atrophy showed that patients could perceive prosthetic vision in the atrophic central field while retaining their natural peripheral vision, an unusual property of the approach.¹ Results from the pivotal European trial, reported in 2025, described most implanted participants regaining the ability to identify letters and read words when the projected image was electronically magnified. This is the strongest functional result any visual prosthesis has produced, and it should be read carefully: reading with a zoom function at a few pixels per letter is not restored sight.

Cortical work has been rebuilt on modern intracortical hardware. A 96-channel Utah array implanted in the visual cortex of a blind volunteer produced stable, reproducible phosphenes and supported identification of simple letter shapes.² A high-channel-count array in monkey V1 evoked percepts of shapes and letters assembled from many simultaneous phosphenes.³ Both establish feasibility; neither approaches a usable device.

§Why resolution is the binding constraint

The intuition that a prosthesis with n electrodes yields an n-pixel image is wrong in a way that governs the whole field. Current spreads, so adjacent electrodes excite overlapping populations and phosphenes merge. Epiretinal stimulation activates axons passing over the array from other retinal regions, so a phosphene may appear far from the stimulated site and be drawn out into a streak. Percepts fade under sustained stimulation, so images must be refreshed. And the phosphene map is idiosyncratic to each patient and must be characterized individually.

Simulation studies of prosthetic vision in sighted observers suggest that face recognition and fluent reading require on the order of hundreds to thousands of well-separated, independently controllable phosphenes. Fielded retinal devices have delivered tens. The gap is not merely manufacturing: packing electrodes more densely without proportionally increasing current confinement does not increase the number of distinguishable percepts, the same saturation that limits the Cochlear implant to roughly eight effective channels.

What "restored sight" means in this field

Reported successes are usually object localization, motion detection, and letter identification under favourable conditions. Users typically retain their cane and continue to rely on other senses. Press coverage that describes patients as "seeing again" overstates every result published to date.

§Risks and the abandoned-patient problem

Surgical risks include retinal detachment, hypotony, conjunctival erosion over the implant, and endophthalmitis; cortical devices add the risks of craniotomy and of a percutaneous connector where one is used. Long-term electrode-tissue stability is limited by the same foreign-body response that degrades cortical recordings in a Brain–computer interface.

The distinctive risk is commercial. When Second Sight stopped supporting Argus II, several hundred implanted patients were left with devices that could not be repaired or upgraded, and some lost function when hardware failed. The episode was widely reported and has become the reference case for the argument that manufacturers of implanted neurotechnology incur obligations that outlast their business models.⁴ Proposed remedies include escrowed device documentation, mandated open-sourcing on discontinuation, and explicit disclosure of company-failure risk in consent documents. None is required by regulators as of 2026, and the same exposure applies to every implant discussed under Neuroprosthetics and to the emerging commercial implants covered by Neurorights debates.

Cost compounds the exposure. Device, vitreoretinal surgery, and the months of rehabilitation needed to learn to interpret phosphenes place these systems among the most expensive interventions per patient in ophthalmology, for a functional gain that is real but small — the trade-off examined under Access and inequality. Parts of the blind community also reject the framing of blindness as a defect awaiting correction, an argument developed at length under Disability rights and enhancement and less prominent here than in deafness only because prosthetic vision has never worked well enough to force the question.

§Alternatives and outlook

Prostheses compete with biological approaches that did not exist when the field began. Somatic gene therapy delivered by AAV vectors is approved for one inherited retinal dystrophy caused by RPE65 mutations, which slows or partially reverses loss in a small patient population; it treats a cause rather than substituting for the sense. Optogenetics offers a middle path — making surviving retinal ganglion cells directly light-sensitive, then driving them with goggles that convert scenes into patterned light. Partial visual function was reported in a single patient with retinitis pigmentosa in 2021, and other opsin-based programmes have since been tested in early trials.⁵ Photoreceptor and pigment-epithelium replacement derived from Induced pluripotent stem cells has been attempted in small numbers of patients for macular degeneration, and retinal organoids grown by the methods described under Tissue engineering are a source of transplantable cells rather than a route to a working eye.

None of these will help someone whose retina is entirely gone or whose optic nerve is severed, which is why cortical prostheses continue despite their difficulty. The honest near-term expectation is a subretinal device that gives central-field pattern vision to people with geographic atrophy, a large and growing population, alongside continued failure to deliver anything resembling normal sight. The unresolved scientific question is whether the visual system can learn to interpret an artificial code at all, or whether prosthetic vision is permanently limited to what a naive read-out of phosphenes can convey — a question that Neural decoding research on the write side of the interface has barely begun to address.

§See also

• Cochlear implant
• Neuroprosthetics
• Optogenetics
• Brain–computer interface
• Sensory augmentation
• Utah array
• Somatic gene therapy
• Disability rights and enhancement

§References