Scholarlink

<p>We present a reservoir computing system that utilizes the transient dynamics of photon-magnon coupling (PMC) for high-accuracy temporal data processing, specifically spoken-digit recognition, while minimizing hardware resources. By time-multiplexing a single physical node into ten virtual nodes, our approach achieves 89% classification accuracy, comparable to conventional systems using hundreds or thousands of nodes. This hardware-efficient design preserves the rich nonlinear dynamics essential for temporal processing while significantly reducing energy consumption. In addition, the inter-node connectivity derived from the transient response of PMC requires that the total sampling duration remains below the PMC decay time, and that the pulse interval be sufficiently small to enable overlapping oscillations for effective time-multiplexed reservoir operation. Our results demonstrate that this transient PMC approach can enable scalable, low-power neuromorphic computing for IoT devices, real-time edge computing, and other resource-constrained environments.</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

Time multiplexed single node reservoir computing enabled by transient photon magnon coupling

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年6月30日

查看全文
  • Haechan Jeon,
  • Bojong Kim,
  • Loïc Millet,
  • Sang-Koog Kim

摘要

We present a reservoir computing system that utilizes the transient dynamics of photon-magnon coupling (PMC) for high-accuracy temporal data processing, specifically spoken-digit recognition, while minimizing hardware resources. By time-multiplexing a single physical node into ten virtual nodes, our approach achieves 89% classification accuracy, comparable to conventional systems using hundreds or thousands of nodes. This hardware-efficient design preserves the rich nonlinear dynamics essential for temporal processing while significantly reducing energy consumption. In addition, the inter-node connectivity derived from the transient response of PMC requires that the total sampling duration remains below the PMC decay time, and that the pulse interval be sufficiently small to enable overlapping oscillations for effective time-multiplexed reservoir operation. Our results demonstrate that this transient PMC approach can enable scalable, low-power neuromorphic computing for IoT devices, real-time edge computing, and other resource-constrained environments.