Big Intelligence. Small Data.
Enabling AI to learn from limited data.
On-device. In real-time.
Our beliefs
-
The current paradigm of Cloud AI is a temporary necessity, not the final goal. True autonomy requires local intelligence to process data and learn in real-time.
In Autonomous Vehicles, the latency required to send data to and from the cloud could be fatal.
Unlocking advanced local processing will drive huge gains in safety, productivity and automation.
-
For decades, we have used digital hardware to mimic the brain's neural networks, costing enormous amounts of energy and time.
In mobile robotics, massive power draw from AI processing drains battery life in minutes.
We believe the next leap in AI won't come from heavier software, but from a fundamental hardware transformation that physically aligns with how the brain actually computes.
-
The internet has infinite data, but data in the physical world is scarce, expensive to obtain & annotate, and is full of rare edge cases that are crucial to detect.
A critical manufacturing defect might only happen once a month, or rare disease might only show in 1 in 10000 patients. Standard AI falls down in these scenarios.
We believe the future belongs to systems that can learn from extremely limited data.
-
A model trained in a lab is obsolete the moment it hits the real world. Conditions change, parts wear down and processes are dynamic.
In high-mix manufacturing, a robot arm trained to handle a specific part will fail catastrophically the moment the factory switches to a part with different shape, weight or reflectivity.
To survive the dynamics of the real world, systems must be able to learn & adapt to their environment continuously.
Technology
Brain-Inspired AI Processor For The Data-Scarce World
Born from 5 years of deep-tech research at Imperial College London, Rayd Technologies is developing a new class of AI accelerator. Our processors physically mimic the learning dynamics of the human brain, enabling machines to learn from limited data, directly on the device, in real-time.
Brain-inspired learning
We have re-designed computation at the fundamental level to process data more like the brain.
Our hardware extracts rich, non-linear transformations of data in a highly parallel and tuneable way. This allows our system to learn efficiently, requiring >30x fewer samples & >100x less parameters, breaking the reliance on huge datasets.
Photonics, but not as you know it
Traditional photonic computing promises massive gains in speed and energy efficiency. However, most photonic chips are simply designed to run the same data-hungry algorithms using light instead of electrons.
Our technology is a radically different take. We utilise non-linear photonics with brain-inspired learning dynamics embedded deep in the system physics - all in a device that is millions of times smaller than traditional photonic systems.
Fast, compact & efficient
We are condensing this massive computing power into a compact, low-power module designed to be integrated and embedded locally.
We are targeting orders of magnitude improvements in throughput & energy efficiency, bringing previously unattainable learning and inference capability to environments where latency, power consumption is critical and data is scarce or dynamic.
Our Team
Kilian Stenning
CEO & Co-Founder
Riccardo Sapienza
Co-CTO & Co-Founder
Kirsten Moselund
Co-CTO & Co-Founder
Jack Gartside
CSO & Co-Founder
Robert Swann
Chairman
Join us at the forefront of next-gen computing and help shape the future of this technology.
If you’re interested working with us, investing or partnerships, get in touch to learn more or explore opportunities.