A modular neural network for R2R diagnosis of semiconductor fabrication equipment: a reactive ion etching application

With the objective of facilitating improved productivity and process control, this paper investigates the use of modular neural networks (MNNs) for malfunction diagnosis in reactive ion etching (RIE) using optical emission spectroscopy (OES) data. OES data acquisition is performed for 49 experimental trials in the etching of SiLK™ (a low dielectric constant polymer). The data collected is subsequently used for MNN modeling. MNNs consist of a number of local experts and a "gating" network, where the former map different regions of the input data space under the supervision of the gating network using a combination of supervised and unsupervised learning. 0.56% and 2.81% of errors were achieved from training and testing data set respectively: therefore, MNNs are found to be useful for diagnosis using OES data.