Researchers at Cornell University have unveiled a microchip that processes information directly in the microwave frequency range, bypassing the usual step of converting radio signals into digital format. The team says it is the first processor of its kind to combine high-speed data and radio signal processing in a single device.
The chip relies on what the researchers call microwave tokens. The approach mirrors tokenization used in large language models: instead of transmitting long sequences of digital data, the system encodes information as compact microwave pulses that preserve the relationships between data fragments.
In tests, the chip transmitted a satellite image of a tropical cyclone, reducing the data volume by roughly eight times while keeping the details needed for analysis intact. The chip consumes less than 200 milliwatts.
The researchers see potential uses in small satellites, drones, robots, and other autonomous systems where bandwidth and power consumption are critical. They also suggest that devices sharing the same architecture could exchange microwave tokens in a form only compatible hardware could interpret, adding a layer of hardware-level data protection.
Cornell has filed a patent application and is pursuing commercialization through a university program.
