基于高频超声的TSV三维封装微小缺陷检测研究

TSV-3D package technology, as the effective approach to achieve system package with high-performance, high-reliability and miniaturization, has been widely studied. The characteristics of small diameter, high density and large aspect ratio make the scale effect more apparent, which brings defects more easily. The traditional testing methods cannot meet the detection requirements. Nondestructive inspection of the defects becomes the new challenge. This project proposes a nondestructive testing method for micro defect inspection of TSV-3D package based on high frequency ultrasound. Through theoretical analysis, simulation modeling and experiment, the propagation law of high frequency ultrasound is investigated, and the mechanism of micro defects to high frequency ultrasound propagation is revealed. The adaptive sparse algorithm is introduced for super-resolution reconstruction of ultrasound echo signals and ultrasonic images. Feature extraction to weak echo signals is carried out by wavelet transform method. Kernel extreme leaning method is utilized for micro defect identification. Supplementing reliability testing and SEM results, the testing method is optimized to realize micro defect detection of TSV-3D package with high efficiency and precision. This project provides an effective reliability analysis technique for TSV-3D products and promotes theoretical and technical innovation in IC package and testing.

TSV三维封装技术是实现高性能、高可靠性、小型化系统级封装的有效途径，并向小孔径、高密度及大深宽比方向发展，封装缺陷更易产生也更难以检测，传统方法已无法满足检测要求，TSV三维封装微小缺陷的无损检测成为当前面临的新挑战。本项目提出基于高频超声的TSV三维封装微小缺陷无损检测方法。通过理论分析、仿真建模和实验，研究TSV三维封装的高频超声传播规律，揭示微小缺陷对高频超声传播的影响机制；引入自适应稀疏分解对高频超声回波信号和超声图像进行超分辨率重构，并综合小波变换和核极限学习等方法，对微弱回波信号进行特征提取且对TSV三维封装微小缺陷进行自动定量识别，辅之可靠性测试和SEM分析来验证、优化检测方法，最终实现TSV三维封装微小缺陷的高效、高精检测。项目研究成果为TSV三维封装产品的可靠性分析提供一种有效的技术手段，并将推动我国IC封装测试的基础理论和技术创新。

本项目围绕TSV三维封装缺陷高频超声检测方法开展研究，系统性地研究了高频超声传播理论，构建了TSV三维封装芯片内部的高频超声检测声场模型，分析了TSV三维封装芯片内部高频超声的传播规律和能量场分布，研究了微小缺陷对高频超声传播的影响，分析了高频超声场的变化规律，揭示了高频超声在微小缺陷界面的反射/透射现象；针对典型微小缺陷存在的TSV三维封装芯片构建了高频超声时域回波信号的稀疏模型，采用最小二乘法非线性拟合对信号进行了稀疏表征分析并进行内部微小缺陷的二维重构，有效地提高了检测的分辨率及弱化了边缘效应；基于高频超声检测平台对TSV三维封装芯片微凸点进行检测，利用经验模式分解算法对回波信号进行自适应降噪，并获取高频超声图像，提取微小缺陷多维特征，构建了微小缺陷特征数据库；建立了基于机器学习理论的微小缺陷自动识别模型，缺陷识别准确率高达97.2%，可有效克服人工检测中人眼疲劳、情绪因素等不足，并有助于推动IC检测的在线运行。