Study On SPC Of Furnace Temperature Series For SiC Homoepitaxy

As one of the key technology of Si C devices manufacturing, Si C homoepitaxial process is directly related to the performance and quality reliability of Si C devices. So, to ensure the thickness uniformity as well as possible is very important. In an actual Si C homoepitaxial process, the furnace temperature, as one of the key factors whice affect the thickness uniformity of the epitaxial layer, varies with the epitaxial process. The change of the furnace temperature directly affects the quality of Si C epitaxial layer.Under the influence of the temperature feedback control system, there is certain correlation between the furnace temperatures, instead of independence. Then the furnace temperature cannot be satisfied with the condition of IIND, and the traditional Shewhart control chart cannot work with it directly. In this paper, the control chart for autocorrelated process parameters is studied, and then SPC of furnace temperature series for Si C homoepitaxy is implemented. The main contents are summarized as follows:1. Traditional Shewhart control chart is applied to realize statistical process control of real furnace temperature series. The result shows that there are a lot of points act abnormally. This result cannot consist with the fact. It shows that in this case, traditional Shewhart control chart cannot work well. By using autocorrelation test, the characteristic of the autocorrelated temperature series is analyzed.2. Residual control chart to analyze the autocorrelated temperature series is applied. Firstly, ARMA model is established for autocorrelated series, and the fitted values series and the residual series are calculated. Then the autocorrelation of the original furnace series is eliminated. The residual series is a white noise series, and is satisfied with the condition of IIND, the prerequisite of traditional Shewhart control chart. Traditional Shewhart control chart is applied to analyze the residual series, to monitor the Si C homoepitaxial process.3. According to the ARMA model of Si C homoepitaxial process, the average run length of residual control chart is studied. The result is compared with the ARL of traditional Shewhart control chart to study the performance of residual control chart. Residual control chart can effectively detect the process mean shift occurs, but its detection capability is weaker than conventional Shewhart control chart.4. Because of the complex and tedious operation which is used to establish the ARMA model by Minitab, a new completed and accessible method is put forward using MATLAB. The method contains how to determine the model type, how to determine the model order, and how to determine the model coefficients. By this method, the ARMA model for Si C homoepitaxial process is obtained, as so as the residual series. Then Shewhart control chart is applied. Finally, both the model results and the control results are compared to the previous results obtained from Minitab to prove the feasibility of this method.5. To some extent, different amounts of data may result in different ARMA modeling results,and this may further affect the accuracy of the ultimate control results. So the minimum amount of data which is used to accurately establish the ARMA model is studied. The result shows that, for some models, just dozens of data are needed to accurately establish the ARMA model.