Combining the Taguchi Method with an Artificial Neural Network to Construct a Prediction Model for Near-Field Photolithography Experiments

Abstract

For analysis using the Taguchi method, the L18 or L27 orthogonal array is usually adopted. However, this requires many experiments (18 or 27 runs, respectively), which consumes time and increases costs. In addition, while traditional analysis with the Taguchi model provides a better group of processing parameters, it cannot predict the unexperimented results. This article proposes a progressive Taguchi neural network model that combines the Taguchi method with an artificial neural network and constructs a prediction model for near-field photolithography experiments. This approach establishes a Taguchi neural network that requires fewer experimental runs, while achieving a high predictive precision. The analytical results of the progressive Taguchi neural network model show that, because there are few training examples in the stage 1 preliminary network, there is a significant fluctuation in the network prediction values. In the stage 2 refining network, the prediction effect in the region around the Taguchi factor level points is not bad, but the prediction in the region more remote from the learning and training examples has greater error. The stage 3 precise network can provide optimal prediction results for the full field.