Research On Automatic Detection Method For Chip Surface Defects

Surface defects caused by semiconductor chips in the production process will affect the service life and reliability of the chip.Therefore,chip surface defect detection is a key approach of chip quality control.The surface defect detection method based on machine vision has the advantages of high efficiency,high accuracy and high real-time performance,and has been widely studied and applied in the field of chips.However,due to the wide variety of defects,the features are difficult to define and occur only during production,making the chip surface defect detection method based on machine vision difficult.Therefore,this dissertation combines the actual needs of enterprises to carry out research on chip surface defect detection,which not only has important practical application value,but also has important theoretical research significance.The main tasks of this dissertation are as follows:(1)Chip surface defect detection method based on mathematical morphology: from chip against single background,the defect image on the surface of the chip is processed by binarization to obtain the defect candidate region.However,the chip surface image has problems such as uneven pixel grayscale and large area of spotty defects,so the defect candidate area has noise or fracture defects.Mathematical morphology can connect cracks and remove noise,and can acquire the geometric parameters and features of the target area.The experimental results show that the method can effectively detect defects that are obvious compared with the background area of the chip surface.(2)Chip surface defect detection method based on unsupervised learning: the types of chip surface defects are variable and the features are difficult to define,which makes it difficult for traditional surface defect detection methods.Deep learning,which can learn depth features,has been widely used in the field of image detection,and provides a new idea for chip surface defect detection.However,defect types are difficult to predict and only occur in the production process,and it is difficult to collect and mark a large number of defect samples,which poses a challenge for supervised chip surface defect detection methods.Therefore,this dissertation proposes an unsupervised chip surface defect detection method that uses only defect-free samples for model training.Firstly,this approach is carried out by reconstructing the defect image with convolutional denoising autoencoder network to obtain the image without defect;then the defect detection is realized by the residual image of the reconstructed image and the defect image.Therefore,overlapping block strategy is proposed to enhance the contrast between defect and non-defect regions.This method has two prominent characteristics.First,it is no need for human intervention in the detection process,which is completely unsupervised.Second,the overlapping block strategy is utilized in this method,which is capable of improving the robustness and accuracy of the method.The defect detection method in this paper has been applied in cooperative enterprises,which further verifies the effectiveness and practicability of this method.