Image Based Automatic Defect Inspection of Substrate, Die Attach and Wire Bond in IC Package Process
There are many inspection stations in the IC packaging process to ensure that the quality in the manufacturing process is consistent and stable. In the past, the large amount of manpower was required to use a microscope to check for quality defects. We propose an inspection method for IC packaging processes to collect the images of the three visual inspection stations behind the welding line site, analyzes the differences in the characteristics of good and bad products, and uses these feature values to distinguish them. We hope to reduce labor costs, production cycle time, and misuse of defective products through automated inspections.
The inspection method is divided into three stages. In the first stage, we focus on incoming substrate defects. The image divided into the foreground and background by Otsu threshold selection method. The connected-component labeling method calculates the area of the image and distinguish defective products. In the second stage, we focus on the defects of die bonding, using the area of the marker to find the corners of the substrate lead, and the horizontal projection pixel accumulation method cuts out the upper and lower metal lead images, then cut out the IC by its proportion. Finally, use Canny edge detection, Hough line conversion method to find the edge of the IC and calculate the angle of the line to determine whether the IC chip is skewed, drained, or offset. In the last stage, we focus on the defects of wire bond. Use the connected-component method to determine whether the wire breaks or leaks, and use the ratio to find the solder joint on the IC to determine whether there is a connection error.
In this study, the image identification is segment into the three parts of defect recognition procedure respectively for calculating the recognition accuracy. From the experimental results, the recognition accuracy of abnormal substrate can reach 100% in first part of procedure, and the recognition accuracy of abnormal die bonding can reach 97% in the second part of procedure, and the recognition accuracy of abnormal wire bonding can reach 100% in the third part of procedure.
Index Terms - AOI, Image Process, Die Bonding Inspection.