ABSTRACT
Spintronic nano-synapses and nano-neurons perform neural network operations with high accuracy thanks to their rich, reproducible and controllable magnetization dynamics. These dynamical nanodevices could transform artificial intelligence hardware, provided they implement state-of-the-art deep neural networks. However, there is today no scalable way to connect them in multilayers. Here we show that the flagship nano-components of spintronics, magnetic tunnel junctions, can be connected into multilayer neural networks where they implement both synapses and neurons thanks to their magnetization dynamics, and communicate by processing, transmitting and receiving radiofrequency signals. We build a hardware spintronic neural network composed of nine magnetic tunnel junctions connected in two layers, and show that it natively classifies nonlinearly separable radiofrequency inputs with an accuracy of 97.7%. Using physical simulations, we demonstrate that a large network of nanoscale junctions can achieve state-of-the-art identification of drones from their radiofrequency transmissions, without digitization and consuming only a few milliwatts, which constitutes a gain of several orders of magnitude in power consumption compared to currently used techniques. This study lays the foundation for deep, dynamical, spintronic neural networks.
ABSTRACT
Energy-efficiency is crucial for modern radio-frequency (RF) receivers dedicated to Internet of Things applications. Energy-efficiency enhancements could be achieved by lowering the power consumption of integrated circuits, using antenna diversity or even with an association of both strategies. This paper compares two wideband RF front-end architectures, based on conventional low-noise amplifiers (LNA) and low-noise transconductance amplifiers (LNTA) with N-path filters, operating with three transmission schemes: single antenna, antenna selection and singular value decomposition beamforming. Our results show that the energy-efficiency behavior varies depending on the required communication link conditions, distance between nodes and metrics from the front-end receivers. For short-range scenarios, LNA presents the best performance in terms of energy-efficiency mainly due to its very low power consumption. With the increasing of the communication distance, the very low noise figure provided by N-path LNTA-based architectures outperforms the power consumption issue, yielding higher energy-efficiency for all transmission schemes. In addition, the selected front-end architecture depends on the number of active antennas at the receiver. Hence, we can observe that low noise figure is more important with a few active antennas at the receiver, while low power consumption becomes more important when the number of active RF chains at the receiver increases.
