Remotely Programming the Weights of a Spintronic Neural Network by a Radiofrequency Broadcast Signal

Jul 16, 2026

Upload by Himbo

Remotely programming non-volatile synaptic weights is a key challenge for in-memory computing scalability. The work experimentally demonstrates remote programming of synaptic weights in series-connected chains of 11 vortex-based magnetic tunnel junctions using broadcast radiofrequency signals via a shared strip line. Distinct resonance frequencies enable frequency-selective vortex-core polarity reversal without per-synapse access lines or selector devices. Reconfiguring chain states reshapes spectral transfer functions for frequency-multiplexed RF inputs. A two-chain, 22-synapse network performs handwritten-digit classification and drone RF-signature identification with 94.91 ± 0.26% and 97.33 ± 0.62% accuracy after remote reconfiguration.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages20

Reading Time50 min

Overview

Remote broadcast programming architecture

Frequency-multiplexed RF inference

Experimental results and task switching

Remotely programming the weights ofa spintronic neural network by a radiofrequency broadcast signal M. Menshawy 1, D. Sanz-Hernández 1, L. Mazza2, V. Puliafito2, G. Finocchio3, A. Jenkins4, R. Ferreira4, L. Benetti4, J. Grollier 1*, and F.A. Mizrahi 1 * 1-Laboratoire Albert Fert, CNRS, Thales, Université Paris-Saclay, Palaiseau, France 2-Department of Electrical and Information Engineering, Politecnico di Bari, 70126, Bari, Italy 3-Department of Mathematical and Computer Sciences, Physical Sciences and Earth Sciences, University of Messina, Messina, Italy 4-International Iberian Nanotechnology Laboratory, Braga, Portugal * [Corresponding authors: julie.grollier@cnrs-thales.fr](Corresponding authors: julie.grollier@cnrs-thales.fr), frank.mizrahi@cnrs-thales.fr Selectively programming a large number of non-volatile synaptic weights without compromising scalability is a key challenge for in-memory computing. Here, we experimentally demonstrate remote programming of synaptic weights in series-connected chains of 11 vortex-based magnetic tunnel junctions using broadcast radiofrequency signals applied through a shared strip line. Because each junction is engineered with a distinct resonance frequency, programming relies on frequency-selective reversal of the vortex-core polarity and therefore does not require individual access lines or selector devices. Reconfiguring the binary states of these chains strongly reshapes the continuous spectral transfer functions that determine the weighted sums performed on frequencymultiplexed RF inputs. This allows compact networks built from such chains to process high-dimensional RF-encoded inputs. We experimentally perform handwritten-digit classification and drone RF-signature identification using a two-chain network comprising only 22 synapses in total, reaching 94.91 ± 0.26% and 97.33 ± 0.62% accuracy, respectively, after remote reconfiguration for each task. Broadcast RF programming thus provides a compact and scalable route to rapidly reconfigurable spintronic neuromorphic hardware. In-memory computing constitutes a promising route toward more energy-efficient artificialintelligence hardware as it reduces data transfer between memory and processing units and allows physical devices to directly perform vector-matrix operations 1–4. A central challenge, however, is to program the individual non-volatile memory components storing the weights in these dense networks without sacrificing compactness, energy efficiency, or scalability. Existing approaches face important trade-offs. Passive crossbar arrays require complex biasing schemes across rows and columns, which increase circuit overhead and programming energy 5,6 . As illustrated in Figure 1a, adding a transistor-based selector restores individual control but increases the cell area by factors up to two orders of magnitude in current demonstrations, while also adding local access routing, and complicating dense 3D integration7–10,5 . In addition, programming and inference typically still rely on distinct peripheral circuits for write driving, sensing and computation. More generally, architectures that rely on individual synapse access preserve programmability, but their wiring overhead makes scaling to dense large networks difficult. This motivates approaches in which many channels can instead be routed through shared physical interconnects. Among neuromorphic hardware platforms, architectures that exploit frequency multiplexing are especially attractive because many input channels can be routed and processed on shared physical interconnects. This opportunity has been explored for inference in photonic systems based on wavelength-selective microring resonators, microring weight banks and phase-change photonic memories4,11–16, as well as in spintronic systems including radio-frequency (RF) networks based on magnetic tunnel junctions 17–23. Yet in all these platforms, frequency multiplexing applies solely to network inputs for

FAQs

Why is remote programming of non-volatile synaptic weights important for in-memory computing?

It enables scalable in-memory computing without compromising compactness, energy efficiency, or scalability. The goal is to configure weights stored in dense networks efficiently.

How does the proposed method program synaptic weights without individual access lines or selector devices?

Each magnetic tunnel junction in a series chain is engineered with a distinct resonance frequency. A broadcast RF signal selectively reverses or reads the vortex-core polarity of targeted junctions based on frequency selectivity.

What tasks and accuracies are demonstrated with the remotely reconfigurable two-chain network?

The network is used for handwritten-digit classification and drone RF-signature identification. After remote reconfiguration, it reaches 94.91 ± 0.26% accuracy for digits and 97.33 ± 0.62% for drone signatures.

Jul 16, 2026

Upload by Himbo