Scholarlink

<p>In advanced semiconductor manufacturing, precise surface inspection of silicon wafers is crucial for achieving high yield and reliable device performance. Conventional surface inspection techniques predominantly detect local defects, constraining the evaluation of the global surface topography, an increasingly critical factor as the demand for large-diameter wafers and fine node scales increases. Interferometric metrology, particularly phase-shifting interferometry (PSI), provides nanometer-level precision in full-field surface inspection, but faces practical challenges in in-line environments owing to its sensitivity to environmental instabilities and the requirement for multiple phase-shifted interferometric images. To address these challenges, we propose a novel surface inspection framework that integrates a wavelength-tuning Fizeau interferometer with a novel deep learning-based phase reconstruction model, TridentNet. The framework reconstructs the wafer surface topography using only two arbitrarily phase-shifted interferometric images, significantly reducing acquisition requirements while preserving high reconstruction fidelity. The system comprises three interconnected modules: interferometric image acquisition, deep learning-based surface reconstruction, and quantitative evaluation based on industrial standards. Our method is trained on simulated datasets reflecting practical conditions, such as the phase-shift interval, additive noise, and nonlinear distortions. Extensive simulations and experimental validations confirm the robustness and accuracy of the proposed method compared with those of conventional techniques, achieving low errors (root-mean-square 0.0206<i>λ</i>) and a high repeatability (standard deviation 0.0185<i>λ</i>). The framework generates comprehensive analysis reports, including statistical metrics and conformity assessments, thereby facilitating process traceability and optimization. This methodology enables in-line, full-field metrology for highly reflective, large-diameter wafers, presenting a reliable solution for next-generation semiconductor manufacturing.</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

Surface topography reconstruction of highly reflective wafers for in-line inspection via deep learning and Fizeau interferometry

  • 期刊论文

  • 发表日期：

    2026年2月7日

查看全文
  • Jurim Jeon,
  • Yangjin Kim,
  • Naohiko Sugita

摘要

In advanced semiconductor manufacturing, precise surface inspection of silicon wafers is crucial for achieving high yield and reliable device performance. Conventional surface inspection techniques predominantly detect local defects, constraining the evaluation of the global surface topography, an increasingly critical factor as the demand for large-diameter wafers and fine node scales increases. Interferometric metrology, particularly phase-shifting interferometry (PSI), provides nanometer-level precision in full-field surface inspection, but faces practical challenges in in-line environments owing to its sensitivity to environmental instabilities and the requirement for multiple phase-shifted interferometric images. To address these challenges, we propose a novel surface inspection framework that integrates a wavelength-tuning Fizeau interferometer with a novel deep learning-based phase reconstruction model, TridentNet. The framework reconstructs the wafer surface topography using only two arbitrarily phase-shifted interferometric images, significantly reducing acquisition requirements while preserving high reconstruction fidelity. The system comprises three interconnected modules: interferometric image acquisition, deep learning-based surface reconstruction, and quantitative evaluation based on industrial standards. Our method is trained on simulated datasets reflecting practical conditions, such as the phase-shift interval, additive noise, and nonlinear distortions. Extensive simulations and experimental validations confirm the robustness and accuracy of the proposed method compared with those of conventional techniques, achieving low errors (root-mean-square 0.0206λ) and a high repeatability (standard deviation 0.0185λ). The framework generates comprehensive analysis reports, including statistical metrics and conformity assessments, thereby facilitating process traceability and optimization. This methodology enables in-line, full-field metrology for highly reflective, large-diameter wafers, presenting a reliable solution for next-generation semiconductor manufacturing.