"Microwave Brain" Chip Developed

A research team at Cornell University has developed a low-power microchip called the "Microwave Brain," the first processor capable of simultaneously processing ultra-fast data signals and wireless communication signals using microwave physics principles. The findings were published in the latest issue of Nature Electronics.

The chip can perform real-time frequency-domain computations with less than 200 milliwatts of power, making it suitable for tasks such as radio signal decoding, radar target tracking, and high-speed digital data processing. The lead developer stated that due to its ability to instantaneously and programmatically modulate microwave signals across a broad frequency band, the chip can be flexibly applied to various computing scenarios. This design bypasses the extensive signal preprocessing and conversion steps required by traditional digital computers for similar tasks, significantly improving efficiency.

This breakthrough capability stems from the chip’s neural network architecture. Inspired by the brain, the neural network constructs complex interconnection patterns through tunable waveguides, enabling pattern recognition and data learning. Unlike conventional neural networks, which rely on clock synchronization and sequential digital instruction execution, this chip operates in an analog manner at microwave frequencies, leveraging nonlinear physical behavior to directly process signals. It can handle data streams up to tens of gigahertz, far surpassing the speed of most existing digital chips.

To achieve this, the team abandoned many traditional circuit design principles. Instead of replicating the structure of digital neural networks, they designed a system more akin to a controlled-frequency "chaotic system," which ultimately delivers high-performance computing capabilities.

The chip can perform both basic logic operations and complex tasks, such as recognizing specific bit sequences or calculating binary values in high-speed data streams. In classification tasks involving various wireless signal types, it achieves over 88% accuracy—comparable to traditional digital neural networks—while consuming only a fraction of the power and occupying minimal space.

The team emphasized that in traditional digital systems, increasing task complexity necessitates larger circuits, higher power consumption, and error-correction mechanisms to maintain precision. In contrast, the new probabilistic computing approach maintains high accuracy in both simple and complex tasks without additional system overhead.