An Efficient Wafer Semiconductor Surface Defect Inspection Using Radial Basis Functional Neural Network

Semiconductor wafer discoveries widespread application in unmanned material delivery vehicles, repairing robots, spray androids, and recognition station sensors. the semiconductor cracker’s flaws, such as holes, husks, cuts, and stains that are created during the production process have a negative impact on the quality of products that are produced later. As a result, it was authoritative to inspect wafer defects. Conventional techniques for inspecting semiconductor wafer defects that rely on handcrafted features are limited in some application scenarios and strongly dependent on experience. This research proposes a unique approach for semiconductor wafer surface defect inspection using deep convolutional neural networks. Initially, in order to excerpt features and create feature maps, a innovative structure for feature pyramid networks with atrous convolution (FPNAC) is designed. Secondly, region proposals are generated by feeding the feature plots into the region proposal network (RPN). In order to correctly categorize and segment the flaws, the region recommendations are finally associated to matching size by way of the inputs of a Radial Basis Functional Neural Network (RBFNN), which consists of three branches. The suggested RBFNN produces good overall performance, as evidenced by the experimental findings, which show Mean Intersection over Union (MIoU) of 90.06% and Mean Pixel Accuracy (MPA) of 94.97%.