Sequencing and Performance Analysis System Design for Semiconductor Manufacturing Cluster Tools: a PVD Example

• Main Author: 胡志豪
• Other Authors: Shi-Chung Chang
• Format: Others
• Language: zh-TW
• Published: 1999
• Online Access: http://ndltd.ncl.edu.tw/handle/62645735183929421569

碩士 === 國立臺灣大學 === 電機工程學研究所 === 87 === Cluster tools have the advantages of shorter cycle times, faster proc-ess development, and better yield for less contamination. Efficient use of cluster tools is naturally very significant to competitive fab operations.In this thesis, we develop a sequencing and performance analysis tool for cluster tools. In specific, we consider a PVD cluster tool, which has been widely used in today''s semiconductor fabs, as the conveyer to demon-strate the ideas and its validity in assisting manufacturing engineers to meet the fab operation goals.We present a modeling approach to model PVD cluster tools and the associated sequencing problems using Petri nets. Petri net modeling mechanism is adopted to capture problem features in resources sharing, concurrency, process sequences and flexible routing. Two performance met-rics of a PN model are analyzed in the performance analyzer: (1) cycle time per wafer and (2) tool bottleneck identification. Ramamoorthy and Ho’s Petri net cycle search and time calculation algorithm serves as the core tech-nique.An operation sequencing problem is formulated as a discrete, integer programming problem. We have demonstrated that cluster tool sequencing problem can be easily formulated from a PN model because PN model cap-tures many constraints in its structure. In our current implementation, tardi-ness is chosen as the objective function.In this thesis, we propose two heuristic algorithms to solve this sequencing problem of the PVD tool. One is called m-A* Search algorithm, and the other is called Fast Sequencing al-gorithm (FS). m-A* Search algorithm generates and searches a partial reachability graph to find an optimal or near optimal feasible schedule in terms of the firing sequence of the transitions of the production Petri net model. Fast Sequencing algorithm (FS) first sequenced the operation con-currently. It then checks if the Marking Number Constraint is satisfied. If the sequencing doesn’t meet the Marking Number Constraint, we will disable the transition whose successor marking with the largest estimate tardiness. Checking is recursive until the Marking Number Constraint is reached.We conduct some numerical cases to demonstrate the feasibility and optimality of our sequencing algorithms. In our test study, both m-A* sear-ch and FS algorithm achieves a better performance than using FIFO. And, FS algorithm outperforms m-A* search in computation efficiency.