Application of artificial neural networks (ANN) and response surface model (RSM) in optical microlithographic process modeling

Optical microlithography represents one of the most sophisticated processes in the manufacturing of microelectronics devices. Accurate process models are highly desirable for process control, process optimization, yield improvement, and cost reduction. Design of experiments (DOE) and response surface model (RSM) are traditional tools for empirical modeling. This paper presents an alternative by using artificial neural networks (ANNs) to model the intricate relationship between the critical dimension (CD) and three key lithographic process variables, soft bake time, exposure stage speed, and develop time. A set of data obtained from a designed experiment is used to train a three-layer neural network. A comparison of the ANN model with the RSM model shows that ANN model provides higher accuracy and greater capability of generalization.