Convolutions with Radio-Frequency Spin-Diodes
Erwann Plouet, Hanuman Singh, Pankaj Sethi, Frank A. Mizrahi, Dedalo Sanz-Hernandez, Julie Grollier
TL;DR
The paper addresses energy-efficient RF signal classification by moving from complex MTJ-based spin networks to simple metallic spin-diodes (NiFe/Pt) that perform weighted sums with frequency-multiplexed inputs. It demonstrates three 2×2 convolutional filters implemented via three chains of four differential spin-diode synapses, achieving 88% top-1 accuracy on the first 100 Fashion-MNIST images and closely matching a noisy software baseline (88.4%) while approaching noiseless software performance (90%). The approach leverages differential synapses to cancel background and uses impedance-matched, frequency-encoded inputs to enable scalable hardware convolutions, highlighting the potential of spintronic RF hardware for energy-efficient, scalable neural processing with prospects for on-chip training and finer synapse integration.
Abstract
The classification of radio-frequency (RF) signals is crucial for applications in robotics, traffic control, and medical devices. Spintronic devices, which respond to RF signals via ferromagnetic resonance, offer a promising solution. Recent studies have shown that a neural network of nanoscale magnetic tunnel junctions can classify RF signals without digitization. However, the complexity of these junctions poses challenges for rapid scaling. In this work, we demonstrate that simple spintronic devices, known as metallic spin-diodes, can effectively perform RF classification. These devices consist of NiFe/Pt bilayers and can implement weighted sums of RF inputs. We experimentally show that chains of four spin-diodes can execute 2x2 pixel filters, achieving high-quality convolutions on the Fashion-MNIST dataset. Integrating the hardware spin-diodes in a software network, we achieve a top-1 accuracy of 88 \% on the first 100 images, compared to 88.4 \% for full software with noise, and 90 \% without noise.