基于改进Unet的芯片凸点高度测量图像分割方法研究

打开文本图片集

中图分类号：TB9;TP391 文献标志码：A文章编号：1674-5124（2025）11-0066-08

Abstract： Image feature segmentation in chip bump height measurement image is an important step for chip bump height measurement based on image processing technology. Aiming at the problem that the traditional threshold segmentation method fails to segment the bump height measurement image with uneven illumination， an image segmentation algorithm for chip bump height measurement based on improved Unet is proposed. Animage segmentation dataset is produced based on the bump scanning experimental images， and the bump scaning images containing features such as light bars，light spotsand notches are firstly manually identified， box-selected and binarized as labeled maps.After that， based on the Unet structure，a Sobel convolutional layer is introduced to strengthen the high-pass filtering in the edge region and sharpen the edges of the light bar image，which improves the feature extraction capability and image segmentation performance of the Unet network.Finally，the efect of image segmentation using the improved Unet image segmentation algorithm is compared with the traditional threshold segmentation method and the traditional Unet method.The experimental results show that the improved Unet image segmentation algorithm can accurately and efficiently segment the chip bumps， with an average Dice similarity coefficient of 0.97，an mIOU of 0.97，a precision of 0.98，and a recall of O.98，and its segmentation performance is beter than that of other segmentation algorithms，and it can effectively solve the problem of image segmentation failure in the condition of uneven illumination.

Keywords： image segmentation; improved Unet; bump light bar image; Sobel operator

0 引言

在晶圆级封装和2.5D叠层封装等先进封装制程中，芯片与芯片之间或芯片与介质层之间利用凸点进行互连[川]。（剩余9419字）