Research On Quality Control For Manufacturing Process Of Silicon Nitride Films Deposited By Plasma Enhanced Chemical Vapor Deposition

Silicon nitride(SiN x) film prepared by Plasma Enhanced C hemical Vapor Deposition(PECVD) is confirmed to play a key role in promoting the conversion efficiency of crystalline silicon solar cells. With the rapid development of photovoltaic(PV) industry in the latest decade, customers and manufacturers have been upgrading the requirements of PV products with higher yield, qualified rate and merit factor. Therefore, research work on the quality control and quality diagnosis of PECVD deposited SiN x films is a significant issue in PV industry.On the basis of research work at home and abroad, this thesis proposes three main scientific issues with the quality control and quality diagnosis of SiN x films in the fabrication process by PECVD method:Fault diagnosis when process parameters have correlation, quality control at the initial state of manufacturing process, quality control based on double failure modes. The comprehensive application of Orthogonal Experimental Design, Bayes theory, multi- variant statistics and diagnosing technology, Path Diagram Theory and computer simulation helps establish four key technologies: a) parameter optimization method for SiN x films prepared by PEC VD technique; b) quality diagnosing method in multivariant statistical process based on path diagram. c) quality control method at the initial state of manufacturing process based on Bayes theory and d) quality control method in multivariant statistical process based on double failure modes. The paper establishes the numerical model on parameter optimization of SiN x films in PECVD process, Path Diagram Theory based multivariant fault diagnosing model, Bayes Estimator based quality control model at the initial state of manufacturing process, MEWMA(Multivariate Exponentially Weighted Moving-Average) control charts based multi- variant quality control model(constructed by the estimation of Population Covariance Matrix). The main contents of the thesis are spread in the following aspects:(1) At the beginning, PECVD deposition method and fabrication process of SiN x films are introduced. And the critical characteristics of quality control(i.e. quality evaluating indices) and the related influencing factors are specifically demonstrated. We have analyzed the problems of the application of the quality control and diagnosing methodology in the preparation process of SiN x films by PEC VD method. After reviewing historical and current research results, we conclude defects of current technology and present the four key technologies.(2) The first contribution of the thesis is to propose a set of numerical simulation approach for the rapid optimization of process parameters in the fabrication of SiN x films deposited by PECVD. By Single Factor Experiments, we deduce and derive the mathematical relations between principal influencing factors and quality evaluating indices, showing the general simulation procedures for the rapid parameter optimization. O n account of the ‘Comprehensive Scoring Method’, we obtain the optimal deposition parameters without a great deal of physical experiments. This approach dramatically reduces the workload of experiments. Moreover, the approach can not only get the optimal process parameters aimed to the best overall quality; but also conveniently obtain a new set of optimal process parameters without physical experiments as quality evaluating indices change.(3) It is a fact that the quality diagnosing method based on multi- variant statistical analysis cannot detect the real fault causes in the cases with fundamental reasons or casual relationships. The thesis proposes a new quality diagnosing method based upon Path Diagram Theory. At the first step, Path Diagram of SiN x films feature is figured out according to the experts opinions, professional knowledge and prior information. Secondly, multi- variant linear regression model and p independent statistic factorizations are obtained by MYT orthogonal analysis. This approach reduces the traditional p! factorization possibilities into a universal factorization way with p independent factors. And the Upper Control Limit of each decomposition factor is determined by F-distribution when the Process Mean Vector and Covariance Matrix are both unknown. With the Upper Control Limits, contribution value of each variable to the failure signal is confirmed. Accordingly, they are used to locate the roots which cause the out of control of the system. Finally, the quality diagnosis of multi- variant statistical process for PEC VD deposition has been accomplished.(4) Since the sample volume of current batch is too small to support sufficient information for Process Control Indices(PCIs) and Control Limits calculation at the initial state of fabrication process of SiN x films(similar to the multi-batch and low volume), we offer Bayes Estimation based method to calculate the μ and σ. Similar sample data is shifted from historical data and archived for subsequent uses by ANOVA and Bartlett Test method. Combined with the current batches, we can obtain estimators of the μ and σ by the Bayesian Theory and the EWMA coefficient. The introduction of EWMA factor modulates the weight of historical data in the whole database. O n the basis of the two estimators, we depict the SPC control charts and figure out the PCIs of PECVD deposition process for SiNx films preparation, solving the quality control problem of PEC VD deposition process at the initial state of fabrication process of SiN x films.(5) The PEC VD deposition process of SiN x films is featured with multiple- input-multiple-output process and double failure modes. As is known, the application of Hotelling 2 and MEWMA control charts requires a known covariance matrix. However, is usually unknown in real production process. It is necessary to replace with its estimator to figure the control charts. Obviously, different hypothesis and estimators result in control charts with different monitoring abilities. In order to debug the monitoring defects of current multi- variant control chart in the PEC VD fabrication process of SiN x films, we derive five different mathematical expressions for five estimators of and MEWMA statics. Under the condition of step offset failure mode and ramp offset failure mode for the mean vector, the thesis studies variations of monitoring abilities for the five MEWMA control charts. Therefore, the quality control method is established in the fabrication process for PEC VD deposited SiNx films.The research of the thesis is guided by practical problems and requirements in manufacturing enterprises. By using orthogonal experimental design, computer simulation and physical experiments, the thesis discusses the approach of quality control and quality diagnosis on the basis of statistical theory. The study of this thesis starts a new methodology of quality control and diagnosis for the fabrication process of SiN x films deposited by PEC VD method in PV industry. And it establishes a theoretical basis for further research in this field. The methodology of this research has been approved and confirmed by the practical cases in a manufacturing company. In conclusion, the research results of the thesis possess theoretical novelty, technical operability and economical utility.