Abstract
Semiconductor manufacturing industry has the characteristics of high production costs and short life cycle of the products. It is therefore very important to reduce the amount of the defective products and find the way to enhance the production yield rate in order to bring more profits to the enterprises. To find the problems on the work station and activate the improvements immediately is one approach to improve the yield during the manufacturing process. But the defective products have already been produced at the same time during the long period of transforming the cause and effect information among work stations. Since it is time-consuming and lack of flexibility, this approach is obviously not effective in enhancing the product yield, especially when the product lift cycle becomes shorter and shorter. For this reason, if the task of yield enhancement can be moved from the manufacturing phase to the design phase, the yield can be more effectively improved since some potential manufacturing faults can be prevented through thorough design considerations. The products will also become more competitive in the constantly changing markets.In recent years, more and more yield enhancement efforts in semiconductor manufacturing have been put on the chemical mechanical polishing (CMP) process. It is believed that the product yield can be effectively improved if the dish effect happening in CMP process can be eliminated. The dish effect is due to the fact that the pad exerts different stresses upon the wafers because of the variability of the circuit density. Caves and defects may therefore be formed after the CMP process. To solve this problem, the common practice is to design dummy circuits into the dies such that the variability of the circuit density can be reduced, and the dish effect can hopefully be eliminated. Since this task is conducted in the IC design phase, the degree of yield enhancement it can achieve is expected to be much more effective than the in-process control techniques. In light of this, this research intends to propose a systematic approach to fill the dummy circuits such that the required variability of circuit density can be achieved with minimal amount of filled dummy circuits. To reduce the complexity of the circuit data, the K-means clustering algorithm was first applied to group the circuit data according to their characteristics. A bi-objectives nonlinear programming model was then proposed to find an optimal solution which can satisfy the requirement for circuit density variations within and between clusters of the circuit data while minimizing the amount of dummy circuits needed to be filled in each clustering area.