主讲人：张勇兵

时间：2017年12月26日（周二）下午4点

地点：理北204

主办：科技处

微纳光学成像重点实验室

江苏省现代检测技术与智能系统重点实验室

物理与电子电气工程学院

报告内容简介：It is a complex and expensive task to detect surface defects and damages in manufacturing processes such as silicon films, steel plates, etc. These damages usually are inspected manually to determine whether they exceed some predefined thresholds that can be acceptable. Deep learning approaches such as deep neural networks, have been proven to be more efficient than traditional methods and can be applied to many fields including image recognition and computer vision. In this research, we apply a convolutional neural network (CNN) to the defect detection problem for silicon films in semiconductor manufacturing. In order to speed up the neural network computation, we implemented the neural network training on a GPU environment. We show that, with three classes of data images, i.e., defect free, bubble, and scratch, and each class with 500 randomly generated samples, the defect classification can be achieved with an accuracy higher than 99%. In order to improve the classification speed, we consider to first allocate the defect area and then apply the CNN to perform the classification.

报告人简介：Yongbing Zhang received his Ph.D. degree in Computer Science from the University of Electro-Communications, Tokyo. He joined the Institute of Policy and Planning Sciences, University of Tsukuba as an Assistant Professor in 1996 and is now a Professor of Graduate School of Systems and Information Engineering, University of Tsukuba. His research interests include distributed and parallel computer systems, communication networks and performance evaluation.