Sequential experiments to characterize processing equipment-maximizing information content while restraining costs

As the semiconductor industry converts more of its fabs to relatively expensive 200 mm wafers, concerns for the costs of experimentation must increase. Many semiconductor processes involve 6 or more process inputs, and many of those inputs act together (factor interactions) or with nonlinear relationships to control outputs. Classical experimental design courses explain the need to build information sequentially to characterize a process fully. This paper contrasts alternative experimental scenarios regarding relative cost versus information content and describes an augmentation technique adopted for a number of process characterization investigations at SEMATECH and its member companies. The paper further illustrates the use of a commercial software product that fully supports these methods in a user-friendly environment and is available on a variety of computing platforms.< >