Scholarlink

<p>While femtosecond laser trepan drilling (LTD) is a promising fabrication method of through-silicon via (TSV), simultaneously achieving high circularity and low taper is challenging, due to the inevitable heat-affected zones and inherent Gaussian laser beam. In present work, we demonstrate the effectiveness of applying the multi-objective optimization strategy, based on the integration of Backpropagation (BP) neural network and non-dominated sorting genetic algorithm III (NSGA-III), in promoting the fabrication accuracy of TSV with 41&#xa0;μm diameter and 200&#xa0;μm depth by femtosecond LTD. Firstly, 125 sets of orthogonal experiments of femtosecond LTD of TSV are conducted to derive the relationships of four input parameters, as laser fluence, overlap ratio, focus deviation and assistant gas flow, with three output results of entrance, exit circularity and taper of TSV. Secondly, leveraging the BP neural network trained on the experimental data, the expanded sampling data are obtained to analyze the coupled effects of two significant input parameters on three output results. Thirdly, based on the functional relationships derived by BP neural network, multi-objective parametric optimization for three output results is performed with NSGA-III. The optimal result of TSV quality improves by 1.1048% for entrance circularity, 1.9247% for exit circularity and 20.0630% for taper, respectively, yielding corresponding optimal input parameters selected from Pareto solutions for the highest precision of TSV. Finally, a 40 × 40&#xa0;mm array of high-precision TSVs, with an entrance circularity of 0.9908, an exit circularity of 0.9859 and a taper of 0.8336°, is successively fabricated by femtosecond LTD based on the proposed optimization strategy.</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

Multi-objective optimization of femtosecond laser trepan drilling for high-precision through-silicon via based on integrated BP neural network and NSGA-III framework

  • 期刊论文

  • 发表日期：

    2026年6月24日

查看全文
  • Yuanjin Cong,
  • Wenqi Ma,
  • Oleg Pasichnyi,
  • Qianhe Wang,
  • Junjie Zhang

摘要

While femtosecond laser trepan drilling (LTD) is a promising fabrication method of through-silicon via (TSV), simultaneously achieving high circularity and low taper is challenging, due to the inevitable heat-affected zones and inherent Gaussian laser beam. In present work, we demonstrate the effectiveness of applying the multi-objective optimization strategy, based on the integration of Backpropagation (BP) neural network and non-dominated sorting genetic algorithm III (NSGA-III), in promoting the fabrication accuracy of TSV with 41 μm diameter and 200 μm depth by femtosecond LTD. Firstly, 125 sets of orthogonal experiments of femtosecond LTD of TSV are conducted to derive the relationships of four input parameters, as laser fluence, overlap ratio, focus deviation and assistant gas flow, with three output results of entrance, exit circularity and taper of TSV. Secondly, leveraging the BP neural network trained on the experimental data, the expanded sampling data are obtained to analyze the coupled effects of two significant input parameters on three output results. Thirdly, based on the functional relationships derived by BP neural network, multi-objective parametric optimization for three output results is performed with NSGA-III. The optimal result of TSV quality improves by 1.1048% for entrance circularity, 1.9247% for exit circularity and 20.0630% for taper, respectively, yielding corresponding optimal input parameters selected from Pareto solutions for the highest precision of TSV. Finally, a 40 × 40 mm array of high-precision TSVs, with an entrance circularity of 0.9908, an exit circularity of 0.9859 and a taper of 0.8336°, is successively fabricated by femtosecond LTD based on the proposed optimization strategy.