US-made world’s first ‘microwave brain’ chip learns language to slash satellite data by 8x

Cornell researchers built a silicon chip that computes directly on microwave and wireless signals, using under 200 mW of power.

The chip learned microwave token embeddings, encoding data into pulse-like units and reconstructing a satellite storm image while sending only about one-eighth of the original data.

Published in Nature Communications, the work points to far more efficient data compression and processing for satellites and edge devices.

The team has filed a patent and is working through Cornell’s commercialization pathway to bring the technology closer to market.