基于机器学习电子封装互连材料研究现状

Current Status of Electronic Packaging Materials Using Machine Learning

随着“后摩尔时代”到来,先进封装技术使芯片性能不断提升,传统的材料开发制备以及封装互连研究难以适应时代需要。根据工业4.0的愿景,采用机器学习的算法在封装材料开发预测、封装工艺优化、焊点可靠性等方面进行高效预测工作,是未来发展的趋势之一。材料领域应用机器学习的过程,包括收集数据、数据预处理、调参优化三方面。封装领域中大多数预测问题可以归类为分类问题,传统监督学习应用最为广泛,常用算法有支持向量机、人工神经网络等。深度学习在封装问题的预测中大放异彩,如卷积神经网络模型等深度学习算法已经逐渐应该用于预测中,并且获得极佳的预测效果。尽管机器学习辅助电子封装研究尚处于萌芽状态,但仍然能够看到机器学习在电子封装领域具有巨大研究及实用价值。

With the advent of“post-Moore era”,advanced packaging technology improves the performance of chips,and traditional materials development and preparation and research of packaging interconnect are difficult to meet the needs of the era.According to the vision of industry 4.0,it is one of the future development trends to use machine learning algorithm to efficiently predict packaging material development,packaging process optimization,solder joint reliability.The process of applying machine learning in material field includes data collection,data preprocessing and parameter tuning optimization.Deep learning shines in the prediction of packaging problems.Deep learning algorithms such as convolutional neural network model have gradually been applied to prediction and achieved excellent prediction results.Although the research of machine learning-assisted electronic packaging is still in its infancy,machine learning still has great research and practical value.