Scholarlink

<p>Voids generated during soldering significantly influence the reliability of power semiconductor packages by accelerating fatigue crack initiation and propagation under thermal cycling. Conventional x-ray inspection identifies such voids but provides limited predictive insight into degradation behavior. This study presents a machine learning-based framework that integrates nondestructive x-ray void characterization with thermomechanical loading data to predict solder joint degradation. Using morphological and spatial void descriptors—including cluster density, edge proximity, and spatial entropy—as input features, classification models were developed and experimentally validated. The random forest classifier achieved a prediction accuracy of 94.8%, outperforming the support vector machine model (91.1%), and revealed that void clustering and edge-adjacent defects are the most influential predictors of solder fatigue. Beyond predictive accuracy, the framework offers interpretability through feature importance analysis, providing physical insight into failure mechanisms. This morphology-aware, data-driven approach demonstrates a practical pathway for early failure prediction, process optimization, and reliability assurance in electronic packaging.</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-Based Prediction of Solder Joint Degradation from Void Morphology Characterized by X-ray Inspection under Thermal Cycling

  • 期刊论文

  • 发表日期：

    2025年12月11日

查看全文
  • Suleiman Ibrahim Mohammad,
  • Asokan Vasudevan,
  • K Venkadeshwaran,
  • Dhirendra Nath Thatoi,
  • A Karthikeyan,
  • Ripendeep Singh,
  • Yashwant Singh Bisht

摘要

Voids generated during soldering significantly influence the reliability of power semiconductor packages by accelerating fatigue crack initiation and propagation under thermal cycling. Conventional x-ray inspection identifies such voids but provides limited predictive insight into degradation behavior. This study presents a machine learning-based framework that integrates nondestructive x-ray void characterization with thermomechanical loading data to predict solder joint degradation. Using morphological and spatial void descriptors—including cluster density, edge proximity, and spatial entropy—as input features, classification models were developed and experimentally validated. The random forest classifier achieved a prediction accuracy of 94.8%, outperforming the support vector machine model (91.1%), and revealed that void clustering and edge-adjacent defects are the most influential predictors of solder fatigue. Beyond predictive accuracy, the framework offers interpretability through feature importance analysis, providing physical insight into failure mechanisms. This morphology-aware, data-driven approach demonstrates a practical pathway for early failure prediction, process optimization, and reliability assurance in electronic packaging.