Multivariate Statistical Modelling to Correlate PECVD Layer Properties with Plasma Chemistry during Silicon Nitride Deposition

Optical emission spectroscopy (OES) was performed during the deposition of SiNx layers in an industrial-scale plasma-enhanced chemical vapor deposition (PECVD) tool. Emission spectra peaks were correlated to deposition properties: deposition rate, refractive index, and bond densities. A statistical model was built to predict layer properties for silicon nitride deposition using different process parameters. This model was used to build a multivariate linear correlation between process parameters and film properties. The accuracy of the results showed a deviation of less than 10.3% for layer thickness, 1.7% for refractive index and 35.8% for bond densities. Based on these results, we propose OES as a powerful method that can be used not only to increase the process control but also for statistical modelling. Moreover, the inversion of the multivariate correlation allows calculation of the multiple sets of process parameters for depositing silicon nitride layers with given properties.