Wafer Surface Semiconductor Defect Classification Using Convolution Neural Network Based Improved Faster R-CNN

Defect classification helps ensure the quality and reliability of semiconductor devices. Even a minor defect can lead to significant performance issues or failures in electronic components. Moreover, a recurring issue in this field is the identification of minute fault patterns, frequently measuring just a few $\mu \mathrm{m}$ in size, and pixels among enormous volumes of high-resolution imaging. In this paper, proposed the improved Faster Region-Based Convolutional Neural Network (R-CNN) by utilizing the Res2Net model can assign the backbone of the proposed model. The model can train and validating on WM811 K dataset. The dataset contains imbalance classes in the presented categories. Initially the data-augmentation technique is employed to create the synthetic data, and it solve the classes imbalance issue. Subsequently Build the Convolutional Neural Network (CNN) foe classify the categories of the defects by using the extracted features. The Res2Net is effectively extract the various features and the feature pyramid network is used to fuse the extracted information. The proposed method is evaluated and compared with various existing methods and metrices such as Accuracy of $97.86 \%$, precision of $96.50 \%$, Recall of $96.80 \%$ and $F 1$-score of $97.85 \%$ and the methods are namely Deep Convolutional Neural Network for Wafer Defect Identification (DCNN-WDI), ShuffleNet-v2-CNN and Transfer Learning Using Faster R-CNN.