Integrated operational decision-making in flexible manufacturing systems with considerations of quality and reliability
MetadataShow full item record
Inherent interactions between operational decisions and their impact on the product quality and equipment reliability necessitate those decisions being optimized concurrently, especially in highly flexible manufacturing systems, such as semiconductor manufacturing fabs, where those interactions are even stronger and where flexibility allows for such decisions to be implemented. This dissertation addresses multiple challenges associated with such integrated operational decisions in the manufacturing process with considerations of quality and reliability. The first part of this dissertation proposes an integrated decision-making policy for production and maintenance operations on a single machine under uncertain demand, with concurrent considerations and learning of yield dependencies on the equipment conditions and production rates. This policy is obtained through a two-stage stochastic programming model, which considers the variable demand, machine degradation, and maintenance times. This model incorporates outsourcing decisions and operational decisions regarding reworking, scraping of imperfect products to ensure the demand is adequately met. A closed-form reinforcement learning method is utilized to learn yield dependencies. Simulations confirm the necessity of yield learning and show the proposed method outperforms the traditional, fragmented approaches where the effects of production rates and machine conditions on the resulting yield rates are not considered. In the second part of this dissertation, a novel optimization framework that couples a recently introduced approach for robust control of overlay errors in photolithography processes with a strategic selection of overlay measurement markers to enable improved control of overlay errors using a reduced number of measurements. Application of this method to the data and models from an industrial-scale semiconductor lithography process shows that the newly proposed combination of the robust overlay control paradigm and optimized marker selection enables improved overlay control, even with a significantly reduced number of markers. Thus, the new methodology enables the reduction of measurement times and subsequent overall cycle time, without deteriorating the outgoing product quality. The measurement selection method suggested in the second part of this dissertation pursues optimality from a purely quality control aspect. Since any measurement down-selection procedure directly affects cycle-times of the resulting process and the understanding of yield rate behaviors, the final portion of this dissertation tackles the task of developing an optimization framework from a more system-level operational aspect. In this method, an optimal number of markers is decided that maximizes the profit considering revenue earned from perfect layer patterns, cost of misidentified bad layers, as well as production and measurement cost. At the same time, the distribution of those markers is optimized considering one’s ability to estimate actuator uncertainties and to understand the yield rate behavior. Application of this method to the data and models from an industrial-scale semiconductor lithography process, as well as sensitivity analyses, are presented to illustrate the proposed method and evaluate the effects of a variety of relevant parameters on the profit of the system.