|M.Sc Student||Goldenberg Ira|
|Subject||Capacity Planning in Semi-Conductor Industry:|
A Robust Optimization Approach
|Department||Department of Industrial Engineering and Management||Supervisor||Mr. Michael Masin|
|Full Thesis text|
Semiconductor manufacturing is a highly competitive industry with rapid innovation rate and long and hard to predict lead time. In turn, long lead time and high innovative rate cause very volatile and uncertain demand. The unmet demand is usually lost both through the specific order that produced in another facility, but also through future sales, meaning loosing the customer.
The common policy in the semiconductor industry is to build a big building, prepare all needed facility and then to ramp it up (purchase equipments in sequence). The reasons for purchasing new equipments are replacement of obsolete equipment and introduction of new products / technologies that require innovative dedicated tools. A typical process requires between 400 to 600 different operations with a variety of tool-types needed. One tool can cost between one to eight million dollars.
The capacity planning in Semiconductor industry is about to determine the timing and amount of equipment to be purchased from each equipment group. It happens under constraint of budget, highly uncertain demand forecast, existing tools from each group, and available space. Obviously, the goal is to find tool-set that is robust for demand uncertainty and other uncertain parameters, while taking into account high capital investment costs. In other words: to balance the tradeoff between the risk of low utilization (excess of capacity) and unmet demand (the shortage of capacity).
In this research, we concentrate on a single location (one facility) problem, with complex tool group / operation relationship, meaning that a specific operation is performed on more than one tool group and each group can perform more than one operation. The model is a multi-product model. We formulate a Mixed Integer Linear Programming (MILP) model for both the nominal problem and its robust counterpart problems. The nominal model aggregates and extends all known models for purchasing planning of equipments. To the best of our knowledge, our robust counterpart is the first model for purchasing planning with uncertain demand dealt with by robust optimization. We also develop a Dynamic Robust Counterpart model, by taking a rolling/folding horizon approach. At each time-period, a decision regarding tool’s acquisition is implemented only when it cannot be postponed. Then, the model is updated with the realized demand and rerun for the remaining periods.
Computational experiments, including real world case study, provide insights on the problem complexity and the importance of robust solutions.