Uncertainty In The Semiconductor Companies In The Capacity Planning Process, The Sensitivity Analysis

One of the key points in the research of industry engineering filed is the capacity planning process of semiconductor manufacturing companies. On the one hand, enormous uncertainty exist in the capacity planning process due to complicated manufacturing process, rapid technology update, increasingly shortened products'lifecycle and changing enormous variability in market demand. On the other hand, those semiconductor companies face very high risk because of expensive equipment, workshop and long delivery time.Currently,"point forecast", that means the forecast value is a fixed number, is still widely used in the capacity planning process of many companies. It surely can not completely reflect the uncertainty, which a company will face in the future. So how to quantify these uncertainties and change the"point forecast"into the"range forecast"is the key effort of this research. Normally, the capacity planning of semiconductor manufacturing companies can be divided into three stages: long range planning (LRP), medium range planning (MRP) and short range planning (SRP). What this thesis focuses on is the uncertainty in LRP caused by factors of MRP (or SRP). We try to search effective methods to quantify the uncertainty of these factors and the effect of these uncertainties on the LRP.There are three parts in this thesis: uncertainty quantification, uncertainty propagation and global sensitivity analysis (GSA). For uncertainty quantification, we discussed data collection and preparation, outlier detection, data classification and uncertainty calculation. In the outlier detection part, we use an improved box-plot method, which is combined withβdistribution uniform expression, in order to better deal with the asymmetric sample data. For the uncertainty calculation, though the traditional parametric method is applied, the non-parametric method is also discussed for comparison. About the uncertainty propagation, Monte Carlo simulation method and error propagation equation are expatiated respectively. They are also the basis of GSA. GSA is used as the analysis method to quantify the contribution of each input variables on the uncertainty of output. Sobol method, which belongs to the variance-based GSA, is reviewed at beginning of this part. The Sobol method can only be used in the non-correlated variables. For its limitation in application, we advance a new method, which is based on the error propagation equation. The new method also belongs to the variance-based sensitivity analysis. Compared with Sobol method, its biggest advantage is that it is suitable for the correlated input variables. At last, an integrative example is used to present how to apply these methods in practice and what their advantages are for the actual work.In this paper, all the data are collected from a famous semiconductor assemble and test manufacturing factory. Therefore, what we want to offer in this research is an effective and convenient flow which can help the cooperated company make uncertainty and sensitivity analysis. Further more, based on those analysis results, we can help them improve their current forecast and decision-making methods, and reform their work flow in order to reduce repeated work.