Antiferromagnetic Programmable Neuron: Structure, Training, and Pattern Recognition Applications
IEEE Journal on Exploratory Solid-State Computational Devices and Circuits · 2026 · 12 · 9-17
DOI: 10.1109/jxcdc.2025.3633490 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Journal on Exploratory Solid-State Computational Devices and Circuits, 2026. Verfasst von I. Sotnyk und 1 weiteren Autor:innen. Die Arbeit wurde bisher 5-mal zitiert (Crossref) und nennt 35 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Academy of Sciences of Ukraine, National Research Foundation of Ukraine, Ministry of Education and Science of Ukraine.
Abstract (Originalsprache)
Artificial neurons based on antiferromagnetic (AFM) spin Hall oscillators (SHOs) are promising elements for creating ultrafast, energy-efficient neuromorphic computing systems. These structures can generate picosecond spikes in response to dc and ac electric currents, thereby mimicking the reaction of biological neurons to an external stimulus. However, conventional AFM neurons have only one input, which significantly limits their applications. In this article, we propose an approach to the implementation of a programmable artificial neuron (P-neuron) based on conventional AFM neurons in the form of a simple, two-layer neural network. Each neuron in the first layer has an independent input, and all of their outputs are connected to a single main neuron in the second layer. This configuration allows the sensitivity of system to individual input signals to be changed independently and in real time by regulating the dc current applied to the first-layer neurons, which makes it possible to program the entire P-neuron structure. In addition, the P-neuron demonstrates the ability for controlled training. We demonstrate that a multi-input P-neuron can successfully classify small images ( $5\times 5$ pixels) of English alphabet symbols. Recognition is based on analyzing the time characteristics of the output neuron signal and comparing them with reference samples. We believe that the obtained results are important for the development and optimization of ultrafast neural network based on AFM nanostructures and AFM spintronic devices capable of generating and processing (sub)terahertz-frequency signals.
Förderung
- National Academy of Sciences of Ukraine
- National Research Foundation of Ukraine
- Ministry of Education and Science of Ukraine
Zitieren
APA
Sotnyk, I., & Prokopenko, O. (2026). Antiferromagnetic Programmable Neuron: Structure, Training, and Pattern Recognition Applications. IEEE Journal on Exploratory Solid-State Computational Devices and Circuits, 12, 9-17. https://doi.org/10.1109/jxcdc.2025.3633490
Vancouver
Sotnyk I, Prokopenko O. Antiferromagnetic Programmable Neuron: Structure, Training, and Pattern Recognition Applications. IEEE Journal on Exploratory Solid-State Computational Devices and Circuits. 2026;12:9-17. doi:10.1109/jxcdc.2025.3633490
BibTeX
@article{Sotnyk2026,
author = {Sotnyk, I. and Prokopenko, O.},
title = {Antiferromagnetic Programmable Neuron: Structure, Training, and Pattern Recognition Applications},
journal = {IEEE Journal on Exploratory Solid-State Computational Devices and Circuits},
year = {2026},
volume = {12},
pages = {9--17},
doi = {10.1109/jxcdc.2025.3633490}
}
EUVIMED