Athena Seminar Series: A Wireless In-Physics Computing Architecture Using Frequency Mixers for Deep Learning at the Edge

Modern edge devices, such as cameras, drones, and Internet-of-Things nodes, rely on deep learning to enable a wide range of intelligent applications. However, this deep learning inference is usually disaggregated: the model is stored on the cloud, while the inputs/outputs are obtained/required on the edge. To this end, we present a novel disaggregated computing architecture for wireless edge networks with two key innovations: disaggregated model access via over-the-air wireless broadcasting for simultaneous inference on multiple edge devices, and in-physics computation of general matrix-vector multiplications directly at radio frequency driven by a single frequency mixer. Using an experimental software-defined radio platform, it achieves 95.7% image classification accuracy with ultra-low energy consumption of more than two orders of magnitude improvement compared to traditional digital computing.