Abstract: Because of the complexity and non-linearity of microcircuit processes, the neural network method combined with the design of experiment(DOE) instead of the traditional statistical methods, is applied to the characterization of thermal oxide layer process equipments. Through only 15 experiments, the model of oxide layer thickness and uniformity is set up based on the generalized regression neural network(GRNN). The fitness and the predicting capability of the model are discussed using the function of the Signal Noise Ratio(SNR). Based on the result, the model can be used to characterize and control the process. The method could also be used for other microcircuit processes.

Key words: microcircuit process equipment, generalized regression neural network, signal noise ratio, design of experiment