Consuming under 200 milliwatts, it delivers high-speed, low-power performance for tasks from radar tracking to edge AI—potentially reshaping how smart devices process information.
Cornell University researchers have unveiled a first-of-its-kind microchip—dubbed the “microwave brain”—that can process both ultrafast data streams and wireless communication signals while consuming less than 200 milliwatts of power. The advancement introduces the first fully integrated microwave neural network on silicon.
Unlike conventional digital processors that rely on step-by-step instructions, the chip leverages analog, nonlinear microwave behavior to compute in real time across wide frequency ranges. This enables it to bypass many of the signal processing steps required by digital systems, opening new pathways for applications from radar tracking to wireless signal decoding.
The chip operates as a neural network, modeled loosely on the brain, using interconnected tunable waveguides that produce multiple modes. Instead of mimicking digital neural networks exactly, its design capitalizes on a programmable “mush” of frequency behaviors that can execute both simple logic and complex classification tasks. The processor has demonstrated up to 88% accuracy in identifying wireless signal types, matching digital neural networks but with a fraction of the size and power requirements.
Why it Matters
For industries dependent on high-speed signal analysis, the technology offers efficiency and flexibility. Its ability to adapt across a wide frequency band means one chip can be repurposed for diverse computational tasks without added circuitry or power. The processor’s extreme sensitivity also makes it suitable for hardware security—detecting anomalies across multiple microwave bands.
Researchers see strong prospects for edge computing. With further power optimization, the chip could be embedded in mobile devices like smartwatches and smartphones, enabling local AI model training without constant reliance on cloud servers. The work was co-led by doctoral students Bal Govind and Maxwell Anderson, under the supervision of professors Alyssa Apsel and Peter McMahon. Funded by DARPA and supported by the Cornell NanoScale Science and Technology Facility, the effort marks a step toward scalable analog-microwave computing platforms. The team has demonstrated that a radically different approach can yield compact, low-power, high-performance processors poised to reshape wireless and data-intensive computing.
