Taguchi method–ANN integration for predictive model of intrinsic stress in hydrogenated amorphous silicon film deposited by plasma enhanced chemical vapour deposition

An integration of Taguchi method and artificial neural network (ANN) technique for the prediction of intrinsic stresses induced during plasma enhanced chemical vapor deposition (PECVD) of hydrogenated amorphous silicon (a-Si:H) thin films is presented. Inputs to the ANN model are plasma power, hydrogen dilution ratio, chamber pressure and substrate temperature. Ninety-two data points were used for the network training, model validation and testing in a 2:1:1 relative proportion. An optimized model with a network architecture of 4-5-3-1, a Levenberg-Marquardt training algorithm and a learning rate of 0.1 was obtained from L9 (34) orthogonal array based on Taguchi approach. By using the optimized network, parametric studies were conducted to show how the intrinsic stresses are influenced by the deposition parameters. Analysis of variance (ANOVA) of the ANN variables indicates that the first hidden layer is the most significant parameter contributing about 39% to the changes in the network mean square error (MSE) while the second hidden layer contributes about 15%. Accuracies of the predictive model are within ±2.5% and ±13% error bound for compressive and tensile stress regimes, respectively. Also, results of the parametric study show a clear trend between the deposition parameters and the resulting intrinsic stresses, and are found to agree with published data. The results are discussed in the light of physics of PECVD process.