Data-driven Wafer Yield Control Method

With the increasing demand for digital products and the popularization of intelligent technology,the importance of semiconductor products has increased day by day.Wafer yield is a key indicator to measure the quality of wafer manufacturing.How to effectively control its level is the core issue of wafer manufacturing enterprises.The Wafer Acceptance Test(WAT)is a test on the physical and electrical properties of the wafer after the wafer is manufactured to check the quality of the wafer product during the manufacturing process.Since each WAT parameter reflects the level of the manufacturing process at each stage,the WAT parameter is used as the control object to provide guidance for the wafer manufacturing process,which is conducive to reducing the investment in inspection equipment and continuously improving the wafer yield,but at the same time it will also trigger wafer production.The production line is stopped and the process is adjusted,which leads to excessively high process adjustment costs.Therefore,this paper studies the problem of wafer yield control,starting from WAT parameters,mining the mapping relationship between wafer yield,and accurately predicting wafer abnormal yield;comprehensively considering improving wafer yield and reducing adjustments Cost,seek reasonable control level of WAT parameters,and provide a reliable basis for wafer manufacturing process improvement.The research carried out includes:1.In view of the complex nonlinear relationship between WAT parameters and wafer yield,a wafer yield prediction method based on convolutional neural network-support vector regression is proposed.First,perform data processing on the wafer WAT parameters,convert the WAT parameters into the matrix input layer of the prediction model;secondly,use the convolutional layer and the pooling layer in the convolutional neural network to extract the features of the WAT parameters;finally,use the support vector Regression replaces the fully connected layer in the convolutional neural network,performs regression analysis on the WAT parameter characteristics,and outputs the wafer yield value.In the case verification,the effectiveness of the proposed method is verified by comparing the prediction error of this method with other methods.2.Aiming at the high dimensionality of WAT parameters and high adjustment cost,a wafer yield optimization method based on improved particle swarm algorithm is proposed.First,use the wafer yield prediction model as the evaluation basis to design a particle swarm algorithm process with adaptive algorithm parameters;then design a local search mechanism based on simulated annealing to enhance the limited optimization capabilities of the algorithm;finally,comprehensively consider improving the wafer yield and reducing Adjust the two optimization objectives of cost,construct a pareto solution set,and form a reasonable WAT parameter combination.In the case verification,the effectiveness of the proposed method is verified by comparing the objective function values obtained by the method in this paper and other methods.3.Based on the above research work,with a wafer manufacturing company in Shanghai as the engineering background,design and develop a wafer yield control prototype system.First,analyze the wafer yield control requirements in the wafer manufacturing process.Second,develop functional modules such as wafer data integration module,wafer yield prediction module,and wafer yield optimization module.Finally,the prototype system is verified by application examples.The yield control in the wafer manufacturing process provides a scientific basis.