Optimisation of plasma enhanced chemical vapour deposition for silicon nitride photonics using optimal design of experiments

Abstract

There have been many advances in silicon nitride based integrated photonics, enabling a variety of interesting applications. The deposition of high quality and low loss silicon nitride (SiN) during the fabrication of waveguides is essential for creating useful devices. Low temperature alternatives to low pressure chemical vapour deposition (LPCVD) such as plasma enhanced chemical vapour deposition (PECVD) are required to enable back-end-of-line (BEOL) integration. The difficulty with optimising PECVD deposition of silicon nitride is that it includes many process parameters that affect the deposition and resulting properties of the film. Thus, optimisation of the PECVD recipe requires a strategic approach to the experimental design. In this work, a technique called optimal design of experiments (DoE) is used to obtain an overview of the PECVD factor’s influence on silicon nitride properties, and to predict the optimal PECVD recipe. With the help of the statistical software JMP the optimal DoE for PECVD deposition of SiN created, significant factors and correlations identified, and optimal PECVD factor combinations predicted. Ellipsometry measurements provide data regarding the responses of interest, namely thickness uniformity, refractive index, and extinction coefficient of the SiN film. It is found that the refractive index is correlated with the ammonia gas flow rate and the extinction coefficient. Most PECVD factors appear to be relevant, in particular the ammonia gas flow rate for the refractive index and the extinction coefficient, and the frequency mode for the thickness uniformity. In addition, two-factor interaction effects are present in the PECVD process and affect the relationship between process parameters and responses. Furthermore, in terms of stoichiometry, silicon nitride films with a refractive index close to two are found to be silicon rich. The testing of predicted recipes shows that the three responses, i.e. refractive index, extinction coefficient and thickness uniformity, cannot be optimised simultaneously to the desired outcome. However, when considering a stoichiometric Si/N ratio instead of the refractive index, predicted recipes result in improved and equally good responses compared to default PECVD and LPCVD recipes, respectively. This approach of optimal DoE shows promising potential and could be interesting to explore further, especially regarding the optimisation of other relevant fabrication steps affecting the propagation losses in waveguides.