The implant is the easy part. That sounds absurd to say about a chip surgically placed at the back of a human eye, but as a builder who spends most days wiring sensors to inference code, I think the retinal chip is the most finished component in Science Corporation’s PRIMA system. The company launched PRIMA in Europe after receiving CE marking for adults with severe central vision loss caused by geographic atrophy from dry age-related macular degeneration. It carries both Breakthrough Device and Humanitarian Use Device designations in the United States. The surgery is an hour-long outpatient procedure. By medical device standards, that is a shipped product.
The part that is still in motion is everything outside the eye. Science Corp has said it expects to keep improving PRIMA’s vision capabilities with a new chip and a new form factor for the glasses, which currently need a battery pack. That single detail tells you where the engineering debt sits.
Why the Glasses Are the Interesting Problem
Strip PRIMA down to its architecture and it looks familiar to anyone who has built a vision bot. There is a camera capturing the scene. There is processing that turns those frames into something the downstream hardware can act on. There is a transmission step. And there is an actuator, in this case photovoltaic pixels in the retina that stimulate surviving cells.
Swap the actuator for a motor driver and you have described half the projects on this site. Which is why the constraints feel so recognizable:
- Power budget. A battery pack means the compute and transmission draw more than a pair of glasses can carry comfortably. Every bot builder who has tried to run a detection model on a wearable knows this wall.
- Latency. Visual systems are unforgiving. When your robot’s camera-to-decision loop lags, it bumps into things. When a person’s does, the world feels wrong in a way no spec sheet captures.
- Form factor. The glasses are the human interface. If they are heavy or odd-looking, adoption suffers regardless of how well the implant performs.
- Asymmetric upgrade paths. This is the one that should keep engineers awake.
The Upgrade Problem Nobody Talks About
When I ship a bot with a bad model, I push an update. When Science Corp improves PRIMA with a new chip, the patients who already have the old one are not getting a download. They had surgery. The implanted layer is effectively frozen hardware with a service life measured in years, while the external layer can iterate on normal product timelines.
That split forces a design discipline most of us never have to practice. You have to decide, before the first implant, which functions live on the immutable side and which live on the replaceable side. Put too much intelligence in the eye and you have locked your users into version one. Put too little and the implant becomes a dumb display that depends entirely on glasses you have not finished designing.
Science Corp’s stated plan to improve both the chip and the glasses suggests they are iterating on both halves. The fact that the glasses are the thing getting called out for form factor work reads, to me, like confirmation that the external side is where the headroom is.
What Bot Builders Can Steal From This
You do not need a medical device to apply the lesson. Any system with a hardware component that is expensive or painful to replace has the same shape. An industrial sensor cast into concrete. A drone frame already in a customer’s hands. A microcontroller on a board you had fabricated in volume.
The pattern worth copying: keep the unreplaceable layer as simple and general as you can get away with, and push the logic that will change into the layer you can swap. If your implanted chip is a passive array that converts incoming light into stimulation, you have bought yourself years of improvement on the camera and processing side without touching the patient. If you had instead baked a specific image-processing pipeline into silicon inside the eye, you would be stuck with whatever you understood about vision in the year you taped out.
A Milestone Worth Watching Closely
PRIMA shipping in Europe matters because it moves brain-computer interface work from demo to product. The device is available to a defined patient group, through a defined procedure, under a defined regulatory clearance. That is a different category of achievement than a conference video.
For those of us building smaller systems, the useful takeaway is not the biology. It is the architecture: a fixed, dumb, durable layer in the place that is hard to reach, and all the fast-moving intelligence sitting where you can still get at it. Science Corp put the hard constraint in the eye and left themselves room everywhere else. That is not a compromise. That is the design.
🕒 Published: