This paper discusses an experimental and numerical study of the nucleation and growth of particles during low-pressure (∼1.0 Torr) thermal decomposition of silane (SiH4). A Particle Beam Mass Spectrometer was used to measure particle size distributions in a parallel-plate showerhead-type semiconductor reactor. An aerosol dynamics moment-type formulation coupled with a chemically reacting fluid flow model was used to predict particle concentration, size, and transport in the reactor. Particle nucleation kinetics via a sequence of chemical clustering reactions among silicon hydride molecular clusters, growth by heterogeneous chemical reactions on particle surfaces and coagulation, and transport by convection, diffusion, and thermophoresis were included in the model. The effect of pressure, temperature, flow residence time, carrier gas, and silane concentration were examined under conditions typically used for low-pressure (∼1 Torr) thermal chemical vapor deposition of polysilicon. The numerical simulations predict that several pathways involving linear and polycyclic silicon hydride molecules result in formation of particle "nuclei," which subsequently grow by heterogeneous reactions on the particle surfaces. The model is in good agreement with observations for the pressure and temperature at which particle formation begins, particle sizes and growth rates, and relative particle concentrations at various process conditions. A simplified, computationally inexpensive, quasi-coupled modeling approach is suggested as an engineering tool for process equipment design and contamination control during low-pressure thermal silicon deposition.
Bibliographical noteFunding Information:
The authors thank Dr. Pauline Ho at Sandia National Laboratories (SNL) for helpful discussions on the role of cyclic molecules on particle formation. We also acknowledge the help of Dr. Harry K. Moffat at SNL for modifications to the SPIN code. This research was partially supported by Semiconductor Research Corporation under contract SRC/97-BJ-442, by the Sandia National Laboratories under contract DE-AC04-94AL85000, and by the National Science Foundation under Grant CTS-9909563. Modeling work was supported by a research grant from the Minnesota Supercomputer Institute. Sandeep Nijhawan was also supported during this research by a fellowship from the Graduate School of the University of Minnesota.