Fast Analysis Of Large-scale Wafer Inspection Data

Wafer manufacturing is a complex and expensive procedure with hundreds of process steps and hundreds of parameters.The defects can occur at any step because of various reasons.When the defect is located in the "critical area",it will cause functional fault and low manufacturing yield;even if defect does not cause a clear functional fault,there will be potential reliability problems.There are also inspection steps in wafer manufacturing.Experienced engineers can analyze the wafer inspection data and translate that information into manufacturing solutions,enhancing yield.Therefore,fast and effective analysis of wafer inspection data has great practical significance.With the development of the semiconductor industry,the scale of wafer inspection data is getting larger and larger.Because of the limitations of computer-aided systems in semiconductor companies and the lack of analysis solutions,it is impossible to effectively use the large-scale wafer inspection data.In this context,this thesis proposes a solution for fast analysis of large-scale wafer inspection data based on distributed computing clusters.The specific research work of this thesis is as follows:1.Based on the prior knowledge of wafer inspection data distribution,effective information extraction methods for inspection data at wafer,die and functional area level are proposed.On the premise of preserving the original data distribution,the amount of inspection data at wafer level has been reduced by nearly 99%.And the amount of inspection data at die and functional area level has been reduced by nearly 80%.In addition,the distributed processing platform is used to design and implement the parallelization process of these methods.2.The extraction method of the distribution feature of wafer inspection data is proposed and the wafer inspection data model is established.A self-adaptive selection method of DBSCAN clustering algorithm parameters is proposed to optimize the initial value problem of EM algorithm.The algorithm is combined with the Gaussian mixture model to fully automatically and accurately extract the distribution features of the inspection data functional area and model it.Finally,using the methods proposed in the thesis,a multi-level wafer inspection data model has been established.3.Designed and developed a fast visualization system for wafer inspection data.By analyzing the actual requirements,the overall structure of the system is designed.And set up a system development environment from both software and hardware aspects.Finally,the visualization module and user management module in the fast visualization system are realized.4.A method for generating ultra-large-scale wafer inspection data is proposed.Based on the multi-level wafer inspection data model constructed in this thesis,combined with the relevant information in the existing,limited wafer inspection data,the ultra-large-scale wafer inspection data is generated,which can be used for more subsequent work.