基于多示例学习径向基函数神经网络的刻蚀设备异常侦测

Multi-Instance Learning Based Radial Basis Function Neural Network for Etching Fault Detection

针对晶圆制造系统刻蚀设备工作过程中的"时间-晶圆-设备参数"三维结构数据导致常规分析方法难以应用的难题,提出了一种基于多示例学习径向基函数(RBF)神经网络的异常侦测方法.该方法应对常规点间距离无法解决示例间不匹配程度衡量问题的不足,改进了距离计算方式,可用既含正常数据又含异常数据的数据包来训练和侦测.运用实验设计理论优化了神经网络参数,使用交叉验证方法将采集到的刻蚀设备运行数据按晶圆加工批次划分为不同示例,作为RBF网络的训练和测试数据,来侦测刻蚀设备异常,提高侦测的稳定性.实验结果表明,该方法可识别多批次多变量情况下的设备异常,适合于刻蚀设备运行过程中多种产品并存情况下的异常侦测.

In a semiconductor manufacturing system, wafers are etched piece by piece, and the etching process data is scattered around different parameters, wafers, and lots. A novel fault detection method based on the radial basis function （RBF） network with a multi-instance learning mechanism was proposed. As the processing data of each wafer was indicated as an instance, the batch data was grouped as training bags. Because Euclidean distance could only suit for pairs of points, the Hausdorff distance was introduced to match different instances. A multi-instance learning RBF network was constructed to detect etching faults using both positive and negetive data bags, and the parameters of network were optimized by design of experiment. The historical data were used to train the proposed network. The k-fold cross validation results show that the proposed method can detect faults in a multi-variable situation with a simple algorithm and fast solution.