Intelligent control of critical dimension in photolithography process

This paper focuses on the design of a control scheme for a photolithography process. The process requires a tight control to maintain a desired gate critical dimension (CD). The control approach, which is currently in use in most semiconductor facilities, is based on operator experience and does not provide satisfactory control on the CD variation. Implementation of an automatic feedback control system in the industry has been difficult because the CD cannot be measured in real-time but only after the process has been completed. In this paper, a neural network is used to predict CD based on the measurements of thickness, reflectivity, refractive index, and dose. The neural network is trained using historical data that are collected at a manufacturing facility. In addition, a neural network-based inverse model of the process is developed. The inverse model is cascaded with the process model to form a feed-forward controller. A feedback CD controller that provides a tighter control in the CD variation is obtained by including a fuzzy controller in the feedback loop.