Optimizations of a photoresist coating process for photolithography in wafer manufacture via a radial basis neural network: A case study

This investigation applied a hybrid method, which combined a trained radial basis network (RBN) [S. Chen, C.F.N. Cowan, P.M. Grant. Orthogonal least squares learning algorithm for radial basis function networks. Neural Networks 2(2) (1991), 302–309] and a sequential quadratic programming (SQP) method [R. Fletcher, Practical Methods of Optimizations, vol. 1, Unconstrained Optimization, and vol. 2, Constrained Optimization, John Wiley and Sons Inc., New York, 1981], to determine an optimal parameter setting for photoresist (PR) coating processes of photolithography in wafer manufacture. Nine experimental runs based on an orthogonal array table were utilized to train the RBN and the SQP method was applied to search for an optimal setting. An orthogonal array table provided an economical and systematic arrangement of experiments to map the relationship between controlled parameters and desired outputs. In this study, a mean thickness and non-uniformity of the thickness of the PR were selected as monitored quality targets for the PR coating process. In addition, the PR temperature, the chamber humidity, the spinning rate, and the dispensation rate were four controlled parameters. The PR temperature and the chamber humidity were found to be the most significant factors in the mean thickness and non-uniformity of the thickness for the PR coating process from the analysis of variance (ANOVA).