How Predictive Are Reactor-Scale Simulations of Wbg Semiconductor Epitaxy? a Sensibility Analysis of SiC and Nitride Epitaxy Models

Process modeling can be the perfect complement to process development: in addition to offering a cost-effective way of exploring the parameter space of a given process or new reactor configurations,  models are able to provide information regarding details of surface and gas phase chemistry that is not readily available experimentally. This information can be extremely useful in helping us rationalize trends in yield, epitaxial quality and defect concentration, or why a particular set of experimental conditions leads to the highest quality materials.

In the case of physical vapor and chemical vapor epitaxy, the fundamental aspects of reactor-scale simulations are well established. However, it is our knowledge of homogeneous and heterogeneous processes what limits the predictive nature of our models. Every reactor rate and transport coefficient comes with an associated error that propagates into the model solution.  In addition to this, the complexity of the overall kinetics of the growth process complicates our ability to identify rate limiting processes exclusively based on experimental trends of growth rates as a function of experimental parameters. Consequently, knowing how sensitive our model is to the different assumptions will help us both better understand the margin of error of model predictions and identify the key steps in the growth process.

In this talk I will summarize the sensitivity analysis studies that we have carried out on the heterogeneous and homogeneous processes during SiC, GaN, and AlN epitaxy. Using datasets compiled from different models available in the literature, we will compare the predicted results in both 0.5D and simplified 1D and 2D models that will allow both local and non-local sensitivity studies. The advantage of 0.5D models is that they can be used as a first order approximation to simulate epitaxy on planetary systems where wafer to wafer variability is lower than in-wafer variability. We will use these models to identify both  the key steps and  the main sources of error as a function of the experimental conditions. Finally, we will study the impact that reduction on complexity has on model predictions.