Abstract: Yield was the key performance indicator in semiconductor wafer fabrication,and its prediction was very important for improving the chip quality and controlling production cost.However,the chip defect was the key factor affecting the yield level.Therefore,a hybrid Fuzzy Support Vector Machine (FSVM) approach based on Density-Based Spatial Clustering of Applications with Noise (DBSCAN) to yield prediction in semiconductor wafer fabrication was proposed.DBSCAN was used to analyze congeries characteristic of wafer defects,which could obtain input parameters on distribution pattern and density of defect data in the yield prediction model.Aiming at the fuzzy relation between defects and yield,a yield prediction model was constructed by incorporating fuzzy rules into Support Vector Machine (SVM) to improve prediction accuracy.The corresponding improvement measures were analyzed according to the yield prediction result and congeries characteristic of wafer defects.The simulation experiments showed that the proposed yield prediction method had better universality,and the accuracy was significantly higher than the Poisson model and binomial model.

Key words: semiconductor wafer fabrication, yield prediction, density-based spatial clustering of applications with noise, fuzzy support vector machine

摘要： 成品率是半导体生产线上的关键性能指标,对其进行预测分析能够有效控制芯片的生产成本、提高芯片质量,而芯片缺陷问题是制约成品率水平的关键因素。因此,研究一种密度聚类与模糊支持向量机相融合的半导体生产线成品率预测方法。首先,采用密度聚类方法对晶圆缺陷聚集特性进行分析,获取缺陷分布模式参数和密度参数,作为成品率预测模型的输入参数;然后,针对缺陷与成品率之间存在的模糊关系,利用模糊规则并结合支持向量机方法构建半导体生产线成品率预测模型;最后利用成品率预测结果对晶圆缺陷聚集特性进行定性分析,确定缺陷问题的来源,并提出相应的改善措施。通过仿真实验表明,所提方法的预测精度优于常用的泊松模型和二项式模型,具有更好的可行性。

关键词: 半导体生产线, 成品率, 基于密度的聚类方法, 模糊支持向量机