New Brain-Like Chip Design Lets Cameras Sort Footage on the Spot

Two researchers, Charles Rizzo and James Plank, have published a new paper that upgrades how brain-inspired computer chips handle messy real-world data. The star of the show is DBSCAN, a decades-old tool that sorts data into groups and ignores random junk. Think of sorting a pile of photos into stacks of similar ones while tossing the blurry accidents. Until now, doing that on "neuromorphic" chips — chips modeled on how brains work, which sip electricity instead of gulping it — didn't account for timing.

That's what the new work fixes. "Spatiotemporal" is jargon for "where and when." Instead of treating sensor readings as a static pile, this approach handles them as a stream of events with timestamps. That matters because modern "event cameras" don't shoot 30 frames per second — they only report when something actually changes, like a person walking past a locked door. The paper offers two versions: a "flat" design built for raw speed, and a "systolic" design that trades some speed for a smaller, cheaper chip. There are even chunked versions for when hardware space is tight.

So why should you care? Because doing this sorting inside the camera, rather than shipping every second of video to the cloud, changes the economics. Less data uploaded means lower bills, faster alerts, and better battery life for doorbells, drones, fitness trackers, and cars. It also means more of your footage never leaves your home in the first place — a real privacy win, since the boring footage is discarded before it travels anywhere.

The catch is timing. This is a research paper with open-source chip designs, not a product you can buy. Neuromorphic chips are still rare and mostly live in labs, so don't expect your Ring doorbell to get a brain transplant this year. And the speed-versus-size tradeoff is real: you can have the fast version or the small version, not both. Still, it's a clear step toward cameras that think locally instead of phoning home.