Multiscale models of the metalorganic vapor phase epitaxy process

Abstract

This thesis develops modeling techniques for chemical vapor deposition processes, specifically metalorganic vapor phase epitaxy (MOVPE). The difficulty in creating an overall modeling strategy for the MOVPE process is that important processes occur on a wide range of length and time scales. Gas phase heat and mass transfer affect the flux of species to the surface, while atomic processes affect the morphology of the growing film. In this thesis, new computational models are developed that work on specific length scales, models are linked together, and combined models are used to study the physics of actual deposition processes. A Kinetic Monte Carlo (KMC) model is developed in order to simulate surface morphology during epitaxial growth. Computational methods, such as binary trees, are used to improve the computational efficiency of the KMC algorithm. To extend the computationally accessible length and time scales, a new parallel algorithm is developed based on ideas from Parallel Discrete Event Simulations (PDES). Superlinear speedup is achieved using this algorithm. The methodology is used along with optimization routines to fit Temperature Programmed Desorption (TPD) spectra to experimental data of methyl desorption off Ga-rich GaAs and determine consistent desorption mechanisms. Physically based reactor scale models are linked to KMC models to gain an overall understanding of the MOVPE system. Initially reactor models that include surface unknowns are flux-split; the surface model is separated from the gas phase model and linked together through the flux to the surface. It is shown that flux-split models exactly match coupled models and in some cases offer better convergence. This linking methodology is extended with the use of a KMC model for the surface. A test case using GaAs growth is modeled, and both accurate growth rate and surface morphology estimates are achieved using the linked model. Neither model separately could predict both flux and surface morphology, but the linked models can be used to make a range of predictions from gas phase concentrations to surface morphology. A reactor used for Grazing-Incidence X-ray Scattering (GIXS) experiments is analyzed using the linked model. The linked model matches both gas-phase concentrations and surface morphology estimates in the reactor. The surface model is compared to experimental GIXS diffuse scattering. The model predicts that the complex reactions on the surface (As dimer and organic group adsorption and desorption from the surface) cause the surface morphology evolution to differ from that observed in molecular beam epitaxy.