Area Of ​​semiconductor Production Forecast Uncertainty Quantification And Risk Decision-making Research

In the industry manufacturing management, marketing demand, product requirement and equipment performance efficiency are the key variables determining factory's capacity. The uncertainty of these variables causes the variability of factory's capacity planning, especially for long range capacity planning in semiconductor manufacturing because of its shortening product life cycle, high equipment cost, etc.The uncertainty issue of long range capacity planning process in semiconductor manufacturing was studied in this thesis. The key uncertain variables in the capacity planning process should be defined firstly. Then the effect of these key uncertain variables to capacity should be quantified. Finally, a new decision-making method would be proposed to ensure that the operation manager of the firm could make a correct decision on whether the factory's capacity should be changed as well as its volume. Consequently the uncertainty of the long range capacity planning in semiconductor manufacturing could be controlled effectively.Then the high level uncertainty analysis of long range capacity planning—factory's space resource planning was discussed in detail. Based on the historical space forecast data and corresponding actual data provided by a global semiconductor assembly and test company, the uncertainty of space planning was defined. During this analysis process, linear regression, grey prediction, neural network back propagation algorithm and confidence interval were applied, respectively, to define the uncertainty. Compared with those methods, the confidence interval of historical space forecast error, calculated by mathematical statistics, was the reasonable method to define the space forecasting uncertainty.Based on the definition of the space forecasting uncertainty, Value at Risk (abbreviated as VaR) model was developed to identify the risk caused by the uncertainty of space forecasting. The most concern in the VaR method is to define the distribution of the population of space forecast error, while normal distribution was adopted to calculate the critical VaR value according to the acquired samples and expert experience. It was shown that the new decision-making method based on VaR was improved greatly...