【作者】 杨光；

【导师】 王宏伟；

【作者基本信息】 大连理工大学 ， 集成电路工程（专业学位）， 2016， 硕士

【摘要】 晶圆的清洗是半导体生产过程的重要工艺环节,晶圆对微污染物的存在非常敏感,晶圆的清洗质量直接关系到芯片的成品率,是关乎生产成本的关键因素之一。目前湿法清洗是被广泛采用的清洗技术,湿法清洗技术关键是对过滤器更换的控制,而控制的重要依据是清洗机内化学品的流量和浓度。论文研究了晶圆清洗机液态化学品过滤器的流量预测模型,并基于该流量预测模型对过滤器及生产线的更换进行优化,解决过早更换过滤器造成的过滤器未物尽其用的浪费和延迟更换过滤器导致流量EP (Excursion Prevention)报警造成的晶圆废品,有效的降低了生产成本。论文的主要研究工作如下：(1)对晶圆清洗机过滤器流量预测进行了研究。通过现场实际采样,选取构建模糊神经网络的训练样本集和检测样本集数据；对采集的数据进行标准化去除噪声处理。(2)进行模糊神经网络建模和学习算法的研究。建立模糊神经网络模型,并进行了实例验证；研究该模型的预测精度及误差原因,并提出改进措施。(3)对模糊神经网络模型的算法进行改进研究。提出了基于新目标函数的模糊聚类算法。该算法将输入量的变化引入到模型的输入空间,改善模型的结构。通过改进的聚类算法确定模糊神经网络的结构和网络参数、权值,确定模糊神经网络神经元个数和规则数,利用改进型模糊神经网络对晶圆清洗机过滤器流量进行预测研究。(4)搭建实验平台,进行实践平台的实际测试,包括模型结构的确定,参数的估计及实践平台的应用流程。在研究中,通过理论分析和实践验证,本文提出的基于模糊神经网络的预报方法可以实现自动预报,实现过滤器自动更换。更多还原

【Abstract】 Wafer cleaning is an important process in semiconductor manufacturing. As wafer are very sensitive to micro pollutants, wafer cleaning quality is directly related to the chip finish rate, and also, the key factor to the cost of production. Wet cleaning is the popular process at present. And the key technique to wet cleaning is the master of the filter replacing. The flow and concentration of the chemicals of the cleaning machine are the criterion of.the filer replacing. In this paper the flow prediction model of the liquid chemicals in the wafer cleaning machine was build., The replacement of filters and the optimization of the production line was executed based on the prediction model. The mode solved the waste that caused by premature replacement of the filters. The model also solved the waste of the wafer which caused by the delay of the filter. The main research work of this paper is as follows:(1) Study on the working principle of the filter of wafer cleaning machine. The training sample and the test sample were selected to construct the fuzzy neural network. The collected data were standardized by noise remove.(2) Study on the modeling and learning algorithms of the fuzzy neural networks. The fuzzy neural network model was established. And an example was given to verify the prediction accuracy and error of the model. The improvement measures were put forward based on the prediction.(3) An improved algorithm of fuzzy neural model was studied. A new fuzzy clustering algorithm based on the new objective function was proposed. In order to improves the structure of the model, the algorithm introduced the input variables into the input space of the model. The fuzzy neural network structure, network parameters, weight, neurons number and rules number were determined by the improved clustering algorithm. The predict study of the flow rate of the wafer cleaning machine filter was carried out by the modified fuzzy neural network.(4) The experimental platform was built in the paper. And the paper carried out the practical test of the practical platform, including the determination of the model structure, parameter estimation and application process of the platform.Based on the theoretical analysis and practical verification, the mode of fuzzy neural network based on fuzzy neural network was proposed, which can realize automatic prediction and automatic filter replacement.更多还原