晶圆表面缺陷在线检测研究

针对准确与实时检测晶圆表面缺陷的需求,提出了一种基于主成分分析(Principal Component Analysis, PCA)和贝叶斯概率模型(Bayesian Probability Model, BPM)的在线检测算法。首先,改进双边滤波方法以消除晶圆表面图像中的噪声和突出晶圆表面的模式特征。然后,提取晶圆表面缺陷的Hu不变矩、方向梯度直方图(Histogram of Oriented Gradients, HOG)和尺度不变特征变换特征(Scale Invariant Feature Transform, SIFT)。接着,采用PCA方法对特征进行降维。最后,在离线建模阶段构建各种缺陷模式的BPMs；在在线检测阶段采用胜者全取(Winner-take-all, WTA)法判断缺陷的模式和构建新缺陷模式的BPMs。提出算法在WM-811K晶圆数据库中得到了87.2%的检测准确率。单副图像的平均检测时间为40.5ms。实验结果表明,提出算法具有较高的检测准确性与实时性,可以实际应用到集成电路制造产线的晶圆表面缺陷在线检测中。

Research on Online Detection of Wafer Surface Defects

For accurate and real-time detection of wafer surface defects, an online detection algorithm based on principal component analysis (PCA) and Bayesian probability model (BPM) is proposed. Firstly, the bilateral filtering method is improved to filter the noise in the wafer surface image and to highlight the pattern characteristics of the wafer surface. Next, the Hu invariant moments, histogram of oriented gradients (HOG) and scale invariant feature transform (SIFT) features of wafer surface defects are extracted. Then, the PCA method is adopted to reduce the feature dimension. Finally, the BPMs of the various defect patterns are constructed in the off-line modeling phase. In the on-line detection phase, the defect patterns are judged by using the Winner-take-all (WTA) method, and the BPM of the new defect patterns are constructed. The detection accuracy of the proposed algorithm is 80.6% in the WM-811K wafer database. The average detection time of single image is 40.5ms. The experimental results show that the proposed algorithm has high detection accuracy and is provided with real-time performance. It can be really applied to the on-line detection of wafer surface defects in the manufacturing line of integrated circuits.