Scholarlink

<p>While machine learning holds remarkable potential for designing high-quality (Q) photonic crystal (PC) cavities, its effectiveness heavily relies on the availability of thousands of data samples. This requirement necessitates substantial simulation resources and considerable time. To tackle the challenge of data scarcity in high-Q microcavity designs, we propose an innovative intelligent model for efficient data augmentation that entails merely a few hundred original samples. Notably, our novel structural reshaping strategy, involving the groundbreaking Euler-bend air-hole structure, significantly enhances the fabrication robustness, addressing the consistency difficulty associated with large-scale manufacturing of high-Q PC microcavity arrays. Silicon PC nanobeam cavities are experimentally demonstrated, featuring record-breaking loaded Q factors, large tolerance for the Euler-bend holes and extremely compact sizes of 6 μm<sup>2</sup>. Importantly, to emphasize the on-chip high-resolution signal processing, the cavity-based microwave photonic filters (MPFs) offer unprecedented capabilities, including ultra-narrow bandwidths, an unlimited frequency tuning range and ultra-high rejection ratios using a micrometer-scale cavity. This breakthrough truly transcends the traditional limitations between the filter size, frequency resolution and tuning range. These exceptional characteristics position our MPFs with a cavity-based record-breaking <i>Q</i><sub>MPF</sub>/<i>S</i> ratio (<i>S</i>: device size).</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

Machine-learning-powered efficient design of photonic crystal cavities

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月7日

查看全文
  • Li Liu,
  • Yangcan Long

摘要

While machine learning holds remarkable potential for designing high-quality (Q) photonic crystal (PC) cavities, its effectiveness heavily relies on the availability of thousands of data samples. This requirement necessitates substantial simulation resources and considerable time. To tackle the challenge of data scarcity in high-Q microcavity designs, we propose an innovative intelligent model for efficient data augmentation that entails merely a few hundred original samples. Notably, our novel structural reshaping strategy, involving the groundbreaking Euler-bend air-hole structure, significantly enhances the fabrication robustness, addressing the consistency difficulty associated with large-scale manufacturing of high-Q PC microcavity arrays. Silicon PC nanobeam cavities are experimentally demonstrated, featuring record-breaking loaded Q factors, large tolerance for the Euler-bend holes and extremely compact sizes of 6 μm2. Importantly, to emphasize the on-chip high-resolution signal processing, the cavity-based microwave photonic filters (MPFs) offer unprecedented capabilities, including ultra-narrow bandwidths, an unlimited frequency tuning range and ultra-high rejection ratios using a micrometer-scale cavity. This breakthrough truly transcends the traditional limitations between the filter size, frequency resolution and tuning range. These exceptional characteristics position our MPFs with a cavity-based record-breaking QMPF/S ratio (S: device size).