SinoGreenTech Academic Portal
Official PDF TranslationSCIENCE CHINA Materials

Efficient Orbit-Torque Driven Spiking Neuromorphic Device Mimicking the Selective Attention Mechanism for Self-Adaptive Recognition

Authors: ZENG Junwei; QIU Shan; TANG Aihua; XU Teng; FANG Liang; GUO Yang; LIU Jiahao

DOI: 10.1007/s40843-025-3535-8Status: Verified Translated Edition
Sponsored AdvertisementAd Placement Area
reCAPTCHA Bot Shield Active

Preparing Secure Academic Download

Verifying human reader & generating high-resolution document...

Verifying Document Integrity15s remaining
← Back to Article
Protected by Google reCAPTCHA v3.PrivacyTerms
Sponsored ContentAdSense In-Feed Ad Slot

Key Findings in This Report

• • Write current density of 5 × 10^6 A/cm^2 is an order of magnitude lower than spin-orbit torque (SOT) devices (~5.5 × 10^7 A/cm^2), reducing energy consumption and enabling scalable on-chip integration for edge neuromorphic processors. • • The all-spin spiking neural network achieves 92% recognition accuracy on CIFAR-10 and 95% on MNIST under complex noise, surpassing state-of-the-art spin-based SNNs by 5 percentage points, demonstrating superior noise immunity for autonomous vision systems. • • Tunable peak action potential in ferromagnetic synapses and neurons mimics biological diversity in neuronal sensitivity, allowing adaptive filtering of sensory stimuli without external circuitry, which simplifies hardware complexity and enhances real-time self-adaptive recognition. • • Orbit torque (OT) arises from momentum-space orbital textures and operates universally in multi-orbital systems regardless of spin-orbit coupling strength, unlike SOT which requires heavy metals; this universality enables material flexibility and reduces reliance on scarce heavy metals, lowering manufacturing costs.