Online Monitoring Mechanisms for Manufacturing Processes with Randomness and Fuzziness Information
In semiconductor industry, although the major function of an IC lies in its electronic properties, the marking content plays a key role in its identification because it is one of the major defect modes of returned material (RMA). A bad marking condition may spread to more than one spot and more than one type of defect mode may occur on a single unit. In this thesis, by applying a demerit control chart for a weighting system to the monitoring of demerit statistics, the process abnormalities may become detectable and controllable. However, in many cases, the recognition of the defects is achieved by the visual inspection of practitioners, which is highly reliant on the experience of human inspectors. Hence, human subjectivity is unavoidable due large part to limitations of human sensitivity, visual perception, inconsistent detection, and visual fatigues. In this situation, not only does randomness provide one aspect of uncertainty for defects data but also the occurrence of fuzziness contributes another uncertainty that should be taken into account when dealing with collected defect data. Therefore, in this paper a fuzzy-weighting defect assignment that represents a degree of seriousness of defects is allotted in accordance. A demerit-fuzzy rating mechanism and monitoring chart is proposed by first incorporating a fuzzy linguistic weight in response to the severe degree of defects. Then the resolution identity property in construction of fuzzy control limits is employed. Moreover, a new fuzzy-number ranking method in differentiation of the underlying process condition developing is developed. Finally, the proposed fuzzy-demerit chart methodology is illustrated by an application of TFT-LCD manufacturing processes for monitoring their Mura-inspection defects conditions. It is well-known that conventional process control can only categorize the process as “in control” or “out of control”. This kind of two binary classifications is sometimes too restrictive to provide the proper decision for decision-makers, especially, when each sample demerit, upper and lower control limits are all fuzzy numbers in the demerit-fuzzy monitoring chart. To determine if the process needed to be adjusted or not, a methodology for ranking/comparing these fuzzy numbers is required. In this paper, the proposed new fuzzy ranking method provides ability for the monitoring chart classifying process conditions linguistically like “rather in control” or “rather out of control”. This kind of intermediate classifications hopefully can supplement the shortcomings of binary classifications when randomness and fuzziness coexist in our modern business environment. Index Terms- Online Manufacturing Process; Randomness and Fuzziness; Fuzzy Set Theory; Discriminatory Classification; Resolution Identity; Monitoring Mechanism.