Scholarlink

<p>Optoelectronic synapse devices hold enormous application prospects in neuromorphic machine vision but suffer from critical challenges including heteroepitaxial lattice mismatch and cumbersome fabrication processes. 4H-SiC offers distinct merits for high-performance neuromorphic devices due to its wide bandgap, high defect tolerance, and CMOS compatibility. Herein, we fabricated ultra-low-power photonic synapses based on MoS<sub>2</sub>/4H-SiC heterostructures, realizing faithful emulation of biological visual information processing. Under 405-nm illumination (5.05 mW · cm<sup>−2</sup>), the device exhibited amplified photoresponse with increasing light pulses: excitatory postsynaptic current change (ΔEPSC) rose progressively to saturation, mimicking biological reinforcement learning. It also achieved non-volatile long-term memory, with current remaining stable after light cessation. Pair-pulse facilitation (PPF) experiments showed a peak PPF index of 134.4% at 50 ms pulse interval, followed by exponential decay, consistent with biological short-term potentiation (STP). For handwritten digit recognition, the neural network (784 input, 300 hidden, 10 output neurons) achieved 86% accuracy in cycle 1, 93.7% in cycle 9 (2.3% below ideal), and sustained &gt; 94% accuracy in cyclic testing, verifying superior synaptic durability. This proof-of-concept work lays a solid foundation for integrated 2D/SiC photonic synapses and promotes the development of low-power intelligent information processing systems with future scalable fabrication compatibility.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A platform of MoS2/4H-SiC heterostructure photonic synapses for low-power neuromorphic machine vision

  • 期刊论文

  • 发表日期：

    2026年6月22日

查看全文
  • Gaotian Zhang,
  • Ruoyao Sun,
  • Yining Gong,
  • Sina Li,
  • Mengmeng Yang,
  • Xiao Liu,
  • Dongxiang Luo,
  • Xingfu Wang,
  • Feng Zhang,
  • Wei Gao

摘要

Optoelectronic synapse devices hold enormous application prospects in neuromorphic machine vision but suffer from critical challenges including heteroepitaxial lattice mismatch and cumbersome fabrication processes. 4H-SiC offers distinct merits for high-performance neuromorphic devices due to its wide bandgap, high defect tolerance, and CMOS compatibility. Herein, we fabricated ultra-low-power photonic synapses based on MoS2/4H-SiC heterostructures, realizing faithful emulation of biological visual information processing. Under 405-nm illumination (5.05 mW · cm−2), the device exhibited amplified photoresponse with increasing light pulses: excitatory postsynaptic current change (ΔEPSC) rose progressively to saturation, mimicking biological reinforcement learning. It also achieved non-volatile long-term memory, with current remaining stable after light cessation. Pair-pulse facilitation (PPF) experiments showed a peak PPF index of 134.4% at 50 ms pulse interval, followed by exponential decay, consistent with biological short-term potentiation (STP). For handwritten digit recognition, the neural network (784 input, 300 hidden, 10 output neurons) achieved 86% accuracy in cycle 1, 93.7% in cycle 9 (2.3% below ideal), and sustained > 94% accuracy in cyclic testing, verifying superior synaptic durability. This proof-of-concept work lays a solid foundation for integrated 2D/SiC photonic synapses and promotes the development of low-power intelligent information processing systems with future scalable fabrication compatibility.