MODELING THE PROPERTIES OF PECVD SILICON DIOXIDE FILMS USING OPTIMIZED BACKPROPAGATION NEURAL NETWORKS

Silicon dioxide films deposited by plasma-enhanced chemical vapor depo sition (PECVD) are useful as interlayer dielectrics for metal-insulato r structures such as MOS integrated circuits and multichip modules. Th e PECVD of SiO2 in a SiH4/N2O gas mixture yields films with excellent physical properties. However, due to the complex nature of particle dy namics within the plasma, it is difficult to determine the exact natur e of the relationship between film properties and controllable deposit ion conditions. Other modeling techniques, such as first principles or statistical response surface methods, are limited in either efficienc y or accuracy. In this study, PECVD modeling using neural networks has been introduced. The deposition of SiO2 was characterized via a 2(5-1 ) fractional factorial experiment, and data from this experiment was u sed to train feed-forward neural networks using the error back-propaga tion algorithm. The optimal neural network structure and learning para meters were determined by means of a second fractional factorial exper iment. The optimized networks minimized both learning and prediction e rror. From these neural process models, the effect of deposition condi tions on film properties has been studied, and sensitivity analysis ha s been performed to determine the impact of individual parameter fluct uations. The deposition experiments were carried out in a Plasma Therm 700 series PECVD system. The models obtained will ultimately be used for several other manufacturing applications, including recipe synthes is and process control.