Hardware realization of the multiply and accumulate operation on radio-frequency signals with magnetic tunnel junctions

Abstract

Artificial neural networks are a valuable tool for radio-frequency (RF) signal classification in many applications, but the digitization of analog signals and the use of general purpose hardware non-optimized for training make the process slow and energetically costly. Recent theoretical work has proposed to use nano-devices called magnetic tunnel junctions, which exhibit intrinsic RF dynamics, to implement in hardware the multiply and accumulate (MAC) operation—a key building block of neural networks—directly using analog RF signals. In this article, we experimentally demonstrate that a magnetic tunnel junction can perform a multiplication of RF powers, with tunable positive and negative synaptic weights. Using two magnetic tunnel junctions connected in series, we demonstrate the MAC operation and use it for classification of RF signals. These results open a path to embedded systems capable of analyzing RF signals with neural networks directly after the antenna, at low power cost and high speed.

Quantum materials for energy-efficient neuromorphic computing

“…For this, antiferromagnets as well as synthetic antiferromagnets are promising due to their terahertz speed and spiking ability. 129,130 While new materials could provide efficient generation of magnetization dynamics as the basis for neuromorphic computing using spin oscillators, 21,27,131,132 they do not address adjusting synaptic weights or enabling high connectivity between individual oscillators. A neuromorphic circuit requires the output of a neuron to serve as the input to other neurons with learning associated with adjustment of the "synaptic" weights of the inputs to the neurons.…”

“…133 They can sense 141 and harvest 142 microwave signals, and embody synaptic devices that perform the multiply-and-accumulate operations directly on input radio-frequency signals without digitization. 132 Interfaces with phase change materials control their synaptic weights directly. 33 In addition to microwave signals, spintronic devices are also sensitive to a wide range of signals such as optical beams and acoustic waves.…”

Convolutional neural networks with radio-frequency spintronic nano-devices

“…A way to overcome the sneak-path in spintronic devices arrays is to use spintronic devices that communicate through radio-frequencies such that the path of the information is not solely determined by electrical connectivity, but also constrained by selectivity in the frequency domain [30][31][32]. Frequency selectivity can be achieved by encoding the input information into radio-frequency signals of different frequencies, and using spintronic nano-resonators as artificial synapses that rectify frequency specific signals [33,34]. Here we describe how we can perform several multiply-and-accumulate operations in parallel without the sneakpath currents arising in crossbar arrays of spintronic memories thanks to the topology of our architecture and the frequency selectivity of spintronic resonators.…”

“…This is promising to directly classify RF signals sensed from the environment with an antenna, as well as to establish communication between layers of neurons through RF signals rather than through wiring only. Furthermore, the RF multiply-and-accumulate operation has been demonstrated experimentally in a small system [34], and a simulated fully-connected perceptron has achieved classification as well as a software perceptron on an image benchmark of handwritten digits [33].…”

Introducing "Neuromorphic Computing and Engineering"

“…As Editor-in-Chief I'm pleased to announce the publication of the first content [1][2][3][4] in Neuromorphic Computing and Engineering (NCE). This editorial aims at motivating the need for creating such a journal by describing our view on some of the current challenges for Information and Communication Technologies (ICT) and the role that NCE can play to address them.…”