Micro Solder Defect Inspection Using Infrared Sequence and Deep Learning Omni-scale CNN

The chip on PCBA is developing towards miniaturization and high density. It becomes more difficult to detect micro solder bump defects inside the electronic package and soldering defects cause 71% of electronic package problems. However, fault diagnosis in the industrial manufacturing scenario is more complex and diverse involving the detection and location of solder bump defects at the IC level. As an attractive option to package inspection, X-ray may seriously affect the service life of ICs. Our group proposed a nondestructive diagnosis method on the basis of active thermography. In the recorded thermal image, the solder ball was segmented for the classification analysis. Due to the large size of PCBA, the narrow gap laser is limited to industrial applications. Because of the inefficiency and low identification rate of current detection, here we propose a detection method using infrared thermal imaging and Omni-scale CNN deep learning algorithm. We take the real chip as the experimental target. The infrared thermal sequence is collected to analyze the temperature evolution of different defect types. We use the target region segmentation to extract features to distinguish the invisible solder defects in infrared images. With the hyperparameter optimization, the Omni-scale CNN classification model realizes efficient and highly accurate detection of different solder (400μm-500μm) defect types.