Researchers at Duke University have developed a groundbreaking approach to energy-efficient artificial intelligence on edge devices using radio waves, offering a powerful alternative to traditional computing methods. The system, called WIreless Smart Edge networks (WISE), allows small, battery-powered devices like drones, sensors, and cameras to run complex AI models without storing them locally or relying on cloud processing. Led by Tingjun Chen, the Nortel Networks Assistant Professor of Electrical and Computer Engineering at Duke, and in collaboration with Dirk Englund’s team at MIT’s Research Laboratory of Electronics, the technology leverages in-physics analog computing. Instead of converting data into digital bits and processing them through power-hungry chips, WISE uses radio frequency (RF) signals to embed AI model weights directly into wireless transmissions. In this setup, a base station—such as a 5G tower, WiFi router, or future 6G infrastructure—broadcasts an RF signal encoding the AI model’s weights. Nearby edge devices receive this signal and mix it with their own input data using existing analog hardware like passive frequency mixers. This physical mixing process performs key mathematical operations, such as multiplication, directly in the radio domain, effectively completing part of the AI computation without digital processing. This method drastically reduces energy consumption and eliminates the need for large memory storage or high-powered processors. In lab tests, the WISE system achieved nearly 96% accuracy in image classification while using more than ten times less energy than conventional digital processors. Zhihui Gao, a Ph.D. student in Chen’s lab and lead author of the study published in Science Advances, emphasized that the technology builds on widely available components. Most wireless devices already include the necessary RF hardware, meaning no exotic or energy-intensive additions are required. The system can work with existing 5G, WiFi, and future 6G networks with minimal modifications. The implications are significant. A single base station could support a fleet of drones during a search and rescue mission, enabling real-time image recognition without draining batteries. Traffic sensors could coordinate signals intelligently, improving urban flow. Medical wearables could analyze vital signs on the spot, enhancing privacy and responsiveness. While still in early development, WISE faces challenges such as limited range and the need for efficient ways to broadcast multiple AI models simultaneously. However, researchers believe the approach offers a scalable path toward intelligent, distributed networks where computation and communication are seamlessly integrated. “This is the next evolution of wireless technology,” Chen said. “We’re not just sending data—we’re distributing intelligence. This could enable a new era of energy-efficient, real-time AI at scale, transforming how devices interact with the world around them.”