Dynamic Demand Conditions, Semiconductor Production Plan Optimization And Re-plan

Marketing environment changes dramatically in semiconductor industry. More complicated product mix and shorter life cycle enforce semiconductor manufacturing enterprise to act as a quick responder and to make more reasonable production planning. So various related factors should be considered for planning decision of numerous species of products, which contribute to the complexity of whole planning decision process.In this thesis, exploratory researches of production planning optimization and re-planning decision-making are carried out on the practical production planning of a famous semiconductor manufacturing factory. A two-level hierarchical model framework is proposed for production planning optimization, of which the higher level is a model for clustering of all products, the lower level is a model for re-planning decision-making of same category products. Demand forecasting is the crux of production planning decision-making. Then production planning can be re-planned by combining demand re-forecasting analysis, uncertainty quantification of re-forecasting and Value at Risk (VaR) analysis of production planning decision-making.The research flow is as follows:Firstly, Fuzzy Short Time-series (FSTS) clustering is proposed to well characterize the interactions of products because raw data are related with time sequence. Furthermore, this thesis presents an optimized FSTS clustering method to study the tendency of production planning, the feasibility and effectiveness of which has been proved through numerical experimental data.Secondly, demand re-forecasting analysis is proposed for different data structure typies, where time-series forecasting is employed for horizontal data, grey prediction for longitudinal data, and multiple regression prediction for two-dimensional data. Then weighted average method, median method and regression method are applied to further comprehensive analysis. The least of error sum of squares is the optimization result of re-forecasting. And then the optimization result has been tested by the absolute value of relative prediction error. Thirdly, uncertainty of demand re-forecasting is quantified. Parametric statistical and non-parametric statistical are two research methods of uncertainty quantification. Based on non-parametric statistical method, nonparametric density kernel estimate and cumulative probability distribution statistic have been related to non-parametric statistical interval estimating and distribution fitting. Through case study and comparative analysis, non-parametric statistical interval method has been selected.Finally, VaR is calculated by using variance-covariance method. The VaR model has been applied to estimate the risk of demand forecasting and inventory. Furthermore, re-planning decision-making is implemented by combining uncertainty quantification with VaR.It shows that the efficiency of production planning has been improved by using clustering and re-planning decision-making for production planning. And production planning has been optimized comprehensively.