Inspection of Lead Frame Defects Using Deep CNN and Cycle-Consistent GAN-Based Defect Augmentation

Inspection of the defects on lead frames is an important task in automated optical inspection (AOI) for the automaton of semiconductor packing industry. A lead frame is a thin layer of metal inside a chip package connecting a die to the circuitry on circuit boards. This chapter introduces the application of the faster region-based convolutional neural network (R-CNN) to detect and classify the defects on lead frames using AlexNet as a backbone. At the early stage of lead frame production, the available number of defects is typically limited, which degrades the inspection performance of the inspection network. To address this problem, this chapter proposes a defect augmentation technique by applying the Cycle-consistent Generative Adversarial Network (CycleGAN) to automatically generate different types of defects. The CycleGAN is characterized with unpaired input–output training images, which makes it possible to translate normal patches on lead frame images to defect patches. The augmented defect patches are then blended into the lead frame images by using a linear blending method to obtain augmented lead frame images in training the faster R-CNN. Experimental results show that the defect augmentation technique is effective in improving the defect inspection performance.