Abstract
In semiconductor foundry, the variation of prosperity and the production strategy of various customers often affect the total quantity of customer order directly, thus create variability on fab input on production. It is critical to develop a systematic methodology to predict customer behaviors and help the decision maker to plan production strategy for the fabrication in the light of uncertainty. When the lots were finished in fab, there are many causes that would let the finished goods delay for customers. Thus the finished goods are stored in fab’s ware house. This increases the fab’s inventory and decrease the inventory turnover ratio. This kind of inventory is called the semiconductor vendor managed inventory in this thesis. The operation efficiency in supply chain is a key factor of enterprise. In supply chain management, it is very common to evaluate the internal efficiency of supply chain operation with key performance index. The inventory turnover ratio is one of the key performance indexes to understand the efficiency of supply chain operation. This research aims to develop a data mining framework and use neural network algorithms and distributed lags structure to analyze the relationships among customer finished goods, order, and other factors. We also conduct an empirical study for validation. The derived empirical rules from this approach can effectively help the decision maker in fab to make timely production decisions given different order situations while maintaining good fab utilization and controlling cycle time. Studying addition to literature review, this thesis consults the domain experts of one company to extract key indexes as the predictors of vendor managed inventory. Based on preliminary study, this study filters the unimportant indexes and focuses the critical ones to adopt the framework including utilization, customers hold stocks, SEMI B/B, etc. By applying this framework, the rules can be extracted to explain why and how customers queue the lots in the fabrication or find other causes that make the finished lots keep in fab. With knowing the information of customer behavior, the decision maker can have more advice by this decision support system. Based on the derived rules, the planning manager can effectively control the quantity of productions and optimize production allocation decisions in the light of involved uncertainty of customer behaviors and thus be able to reduce the waste of material and cost of inventory. These rules can also help the manager to track what goes wrong with these predict variables (i.e. key indices). The empirical study showed the practical viability of this approach and thus also would be able to help this company to better serve her customers.