Scholarlink

<p>Biological systems process multimodal analogue sensory inputs with remarkable efficiency, seamlessly integrating sensing, encoding and learning. Replicating this capability in electronics has typically required complementary metal–oxide–semiconductor-based digital architectures with analogue-to-digital conversion, external power and off-chip processing, resulting in high energy consumption and complexity. Here we present a fully analogue, self-powered neuromorphic system that unites multimodal sensing, spike encoding and unsupervised learning entirely in hardware using drift and diffusive memristors. Built solely from sensors and memristors, the platform operates without digital circuitry or external power, enabling autonomous, real-time analogue processing. Diffusive memristors capture temporal correlations from diverse inputs, while drift memristors provide non-volatile synaptic weights. Supporting both homo-synaptic and hetero-synaptic plasticity, the system achieves unsupervised learning across multiple sensory channels. By integrating sensing, computation, memory and learning on a single printed circuit board, this work establishes a compact, energy-efficient hardware paradigm for neuromorphic processing and multimodal sensor fusion.</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

Self-powered analogue neuromorphic system for multimodal sensing, encoding and learning with diffusive and drift memristors

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月30日

查看全文
  • Seung Ju Kim,
  • Ruoyu Zhao,
  • Piyush Sud,
  • Yichun Xu,
  • Jian Zhao,
  • Han-Ting Liao,
  • Serena Zhang,
  • Rivu Midya,
  • Qinru Qiu,
  • J. Joshua Yang

摘要

Biological systems process multimodal analogue sensory inputs with remarkable efficiency, seamlessly integrating sensing, encoding and learning. Replicating this capability in electronics has typically required complementary metal–oxide–semiconductor-based digital architectures with analogue-to-digital conversion, external power and off-chip processing, resulting in high energy consumption and complexity. Here we present a fully analogue, self-powered neuromorphic system that unites multimodal sensing, spike encoding and unsupervised learning entirely in hardware using drift and diffusive memristors. Built solely from sensors and memristors, the platform operates without digital circuitry or external power, enabling autonomous, real-time analogue processing. Diffusive memristors capture temporal correlations from diverse inputs, while drift memristors provide non-volatile synaptic weights. Supporting both homo-synaptic and hetero-synaptic plasticity, the system achieves unsupervised learning across multiple sensory channels. By integrating sensing, computation, memory and learning on a single printed circuit board, this work establishes a compact, energy-efficient hardware paradigm for neuromorphic processing and multimodal sensor fusion.