A team at Cornell University has unveiled an experimental microchip that uses analog microwave signals, rather than traditional digital logic, to execute computing tasks. The device, called a microwave neural network (MNN), relies on a “frequency comb” of microwave spectral lines and interconnected electromagnetic nodes to recognize patterns, perform logic operations, and analyze data streams at tens of gigahertz speeds, tells Live Science.
In contrast to typical desktop or mobile CPUs that operate in the 2.5–4 GHz range, this chip functions in a much higher frequency band and processes input directly in the analog/physical domain, bypassing many steps of digitization, memory access, and clock-driven logic. It has demonstrated accuracy rates around 88% across wireless-signal classification challenges, while consuming less than 0.2 watts of power, orders of magnitude less than the 65+ watts typical of mainstream processors.
Its developers suggest this architecture is especially suitable for edge computing, wearable devices, AI inferencing in situ or communications/radar systems where both throughput and energy efficiency matter. Because operations occur in the microwave band, the chip also blurs the boundary between computing and signal-processing hardware.
However, the technology remains at prototype stage. Key engineering challenges include: scaling the device to broader task types, improving accuracy further, integrating with existing digital ecosystems, and ensuring manufacturability at scale. For engineers and systems designers, this development signals a potential paradigm shift, moving from purely digital architectures toward hybrid physical-signal computing models that may redefine performance, efficiency, and what “compute” means at the hardware level.