An innovative yield learning model considering multiple learning sources and learning source interactions

Abstract

Existing yield learning models do not separate the effects of different learning sources or consider the interactions among the sources. To address this problem, a multisource-with-interaction yield learning model was developed. In this paper, the properties of this multisource yield learning model are discussed from a theoretical and practical standpoint. In this study, the proposed methodology was applied to the manufacturing process of a dynamic random access memory product. The proposed model exhibited improved accuracy in estimating the future yield, evidencing its superiority over existing yield learning models. The proposed methodology can be generalized to model the learning processes of other performance measures in manufacturing or service systems.