Research On Qualification Management Decision In Semiconductor Manufacturing Based On Process Flexibility And Stochastic Programming

Qualification management is becoming one of the most significant scheduling decision problems in semiconductor manufacturing.However,it is least studied mainly due to the research difficulties and the complex and volatile enviromental factors,which leads the qualification management problem to be one of the theoretical research bottlenecks for schedulers and decisoners.In order to reduce the dismatch between the theoretical research and the practical requirements,this study focuses on addressing this problem.This paper mainly uses two types of theoretical methods: the process flexibility theory and the stochastic programming method,and meanwhile considers capacity loss and three types of uncertainty in model: capacity loss due to product-machine qualification,traditional random capacity loss,and stochastic product(recipe)demands,all of which are significant and practical in real systems.The processes of qualification test are often required by many operation steps in semiconductor manufacturing industry.On property,the process of qualification test is also a process of achieving flexibility without extra cost investment.Therefore,bringing in the theories and concepts in process flexibility field,such as the “long chain” concept,can help solving the qualification management problem and improving the system efficiency.However,no theories in past literature consider the capacity loss,which is very common in semiconductor manufacturing,resulting in inefficiency of most theories in practical applications.Therefore,one of the main contributions of this research is to address this problem by making a deep analysis,developing new theories on process flexibility field under capacity loss and extending these theories to a more practical situation.Furthurmore,this paper also provides insights on designing effective qualification structures based on capacity loss for qualification management problem.On the other hand,another contribution of these theories on process flexibility field is to to help developing an initial feasible solution for the coming stochastic programming algorithm,which speeds up the solution process.This paper makes another contribution for qualification management problem by presenting a two-stage stochastic mixed integer linear programming formulation for parallel machines.This formulation makes full use of features of semiconductor manufacturing processes,such as uncertainty and capacity loss.Moreover,this formulation uses the Lagrangian relaxation technique and the surrogate subgradient algorithm,which are proven to be powerful techniques to achieve near-optimal solutions,to improve the solution performance.Compared with the traditional stochastic programming alrotihms,such as the L-shaped method and the Lagrangian decomposition method,the developed algorithm in this paper has an advantage in solution efficiency because it is based on structural features of the model.Meanwhile,another advantage of this approach is that it allows the full use of distribution information of stochastic variables.Even if the distribution information for some random variables is unavailable,we can achieve an approximation of the problem's optimal objective by simplifying the proposed method.On the other hand,the designed algorithm has a huge application potential.It can be used in other models if the requirements of structural features are satisfied.