稀疏重构SAM芯片焊点检测方法研究

提出了基于稀疏表示的声扫描显微镜(Scanningacousticmicroscope,SAM)图像超分辨率重构方法，以解决其空间检测分辨率受超声波频率和穿透深度的限制，原始SAM图像分辨率较低，不利于封装缺陷辨识等问题。通过字典设计训练和稀疏系数α求解获得了重构的高分辨率SAM图像，利用Levenberg-Marquardt算法改进BP神经网络(LM-BP)，并用于倒装芯片焊点缺陷识别。与原始图像及双三次插值图像相比，稀疏重构图像的峰值信噪比明显增大，提高了SAM图像质量，减小了芯片焊点的错误识别数目，错误率降至2.76%。试验结果表明稀疏表示的SAM重构算法和LM-BP神经网络训练速度快、识别精度高，可用于高密度半导体封装缺陷的检测及可靠性评估。

Solder Joint Inspection Using SAM Image through Sparse Reconstruction

An algorithm based on sparse representation was proposed to reconstruct high-resolution (HR) SAM images. Scanning acoustic microscope (SAM) is always used for defect detection of electronic packages, but the spatial resolution is limited by the frequency and penetration depth of ultrasound. The original SAM image with low resolution is adverse to defect recognition. The HR SAM image is obtained through dictionary training and sparse coefficient α solving. The Levenberg-Marquardt modified BP neural network (LM-BP) was used to classify the solder joints. Compared with the original image and the bicubic interpolation image, the peak signal-to-noise ratio of the sparsely reconstructed image is significantly increased, the quality of the SAM image is improved, the number of misidentified solder joints is reduced, and the error rate is reduced to 2.76%. The experimental results demonstrated that the sparse representation algorithm and the LM-BP neural network are effective and accurate for defect inspection of high-density packages.