Neuromorphic compute
Intelligence that emerges from hardware itself
CyberSwarm is a deep-tech company building memristor-based neuromorphic processors for physical AI: energy-efficient computing for vehicles, robots and instruments that must sense and decide on the device.
Every processor ever built keeps memory in one place and computation in another, and spends most of its energy shuttling data between the two. For an AI workload that traffic is not overhead. It is nearly the whole job. CyberSwarm closes the gap: our chips compute inside the memory itself, so a value never has to travel to be used.
We took the design from biology. A brain does not keep the two apart. A synapse stores and computes at once, which is why an insect navigates an unfamiliar room on a fraction of a watt. Our arrays work the same way, and they learn the same way: by changing the physical component rather than rewriting a number held somewhere else.
The long-term aim is an artificial brain: a system that senses, learns and decides in hardware, continuously, wherever it happens to be. Each generation of the architecture is a step toward that, and each one already earns its place in a vehicle, a robot or an instrument today.
Physical AI
Intelligence has to work where the machine is
The last decade of artificial intelligence happened in data centres. That worked because a data centre has effectively unlimited power, active cooling, and no deadline a few hundred milliseconds cannot meet. A machine in the physical world has none of those things. It has a battery, a sealed enclosure, and a control loop that has to close before the thing it is steering hits something.
So intelligence stays in the building, and the machine keeps a thin connection back to it. That is workable for a thermostat. It is not workable for a car changing lane, a surgical instrument, a satellite over an ocean, or a robot on a factory floor where the network is one more thing that can fail.
The blocker was never the models. Those exist and they are good enough. It is that no processor built so far can run them inside the power and time budget a physical machine actually has, because every one of them keeps memory in one place and arithmetic in another and spends most of its energy moving numbers between the two.
CyberSwarm builds the processor that removes that trip. The weight is stored as a physical resistance in the same component that computes with it, so an entire layer of a neural network resolves as one pulse of current rather than as millions of trips to memory. Nothing is fetched, nothing is written back, and nothing has to be sent anywhere.
What that changes commercially is not speed. It is what becomes possible to ship. A product that works in a tunnel and at sea and inside a factory that will not put its line on the public internet. A feature that fits inside a battery budget instead of exceeding it. A camera that recognises a driver without the driver ever leaving the device. Privacy that is a property of the hardware rather than a clause in a policy.