针对传统深度神经网络对混合型晶圆缺陷信息提取计算效率低的问题,提出了一种基于通道混洗和深度可分离卷积的轻量化深度神经网络,实现了混合型晶圆缺陷的高效识别.在晶圆图数据集Mixed-type WM38上的实验结果表明,所提出的模型对比于一些现有的深度学习模型,在耗费较少的训练和推理时间的同时取得了较高的模型精度,其平均正确率达97.32％,参数量仅有0.4786 M.
Mixed-type Wafer Map Recognition Based on Channel Shuffle and Depthwise Separable Convolution
A lightweight deep neural network based on Channel Shuffle and depthwise separable convolution is proposed to address the low computational efficiency of traditional deep neural networks in extracting mixed-type wafer defect information. Effi-cient recognition of mixed-type wafer defects is achieved through this approach. Experimental results on the Mixed-type WM38 wafer map dataset demonstrate that the proposed model achieves higher model accuracy while consuming less training and inference time compared to some existing deep learning models. The average accuracy of the proposed model is 97.32％, and its parameter count is only 0.4786 M.
万方数据
维普
