Scholarlink

<p>The increasing demand for high-performance and high-frequency electronic devices necessitates advanced packaging technologies that minimize wire loop height and improve signal integrity. In this study, we propose an integrated process optimization framework for wedge bonding by combining machine learning (ML) models with Taguchi-based experimental design. A dataset was efficiently constructed using a Taguchi orthogonal array, enabling the comprehensive evaluation of multiple ML models. An ensemble framework was developed by integrating a random forest (RF) classification model for predicting process feasibility and an artificial neural network (ANN) regression model for loop shape prediction, achieving a high prediction accuracy (R<sup>2</sup> = 96.24%). Using the optimized parameters derived from the model, an ultra-low loop height of 85.07&#xa0;μm was successfully fabricated. Application of the optimized process to a C-band monolithic microwave integrated circuit (MMIC) package improved signal transmission performance (S21) by 13.4%. These findings demonstrate that the proposed ML–assisted optimization approach effectively overcomes the limitations of traditional experimental design and provides a scalable methodology for next-generation high-frequency packaging technologies.</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-driven process optimization for ultra-low loop wedge bonding in monolithic microwave integrated circuit packaging

  • 期刊论文

  • 发表日期：

    2026年7月16日

查看全文
  • Byung-Geon Park,
  • Jae-Hak Lee,
  • Ah-Young Park

摘要

The increasing demand for high-performance and high-frequency electronic devices necessitates advanced packaging technologies that minimize wire loop height and improve signal integrity. In this study, we propose an integrated process optimization framework for wedge bonding by combining machine learning (ML) models with Taguchi-based experimental design. A dataset was efficiently constructed using a Taguchi orthogonal array, enabling the comprehensive evaluation of multiple ML models. An ensemble framework was developed by integrating a random forest (RF) classification model for predicting process feasibility and an artificial neural network (ANN) regression model for loop shape prediction, achieving a high prediction accuracy (R2 = 96.24%). Using the optimized parameters derived from the model, an ultra-low loop height of 85.07 μm was successfully fabricated. Application of the optimized process to a C-band monolithic microwave integrated circuit (MMIC) package improved signal transmission performance (S21) by 13.4%. These findings demonstrate that the proposed ML–assisted optimization approach effectively overcomes the limitations of traditional experimental design and provides a scalable methodology for next-generation high-frequency packaging technologies.