three dimensional(3D), massively parallel (MP), chemically reacting flow
code has been developed. This code, MPSalsa, allows simulation of complex
3D fluid flow, heat transfer, and mass transfer with complex bulk fluid
and surface chemical reaction kinetics. Current modeling development includes
the inclusion of turbulent fluid flow with reactions, enclosure and participating
media radiation, plasma chemistry and full multiphysics capabilities. Our
current numerical research focuses on efficient and robust iterative solution
methods. The techniques use an inexact fully-coupled Newton method with
preconditioned Krylov solvers . Currently we are exploring adaptive grid
techniques and multilevel preconditioners to increase solution accuracy
and the underlying convergence rate of the iterative solvers. An important
aspect of this work is the development of parallel dynamic load balancing
techniques to redistribute the load after adaptive mesh refinement.
MPSalsa has been used to simulate the deposition of a silicon carbide (SiC) mechanism with 19 chemical species undergoing over 40 chemical reactions. In addition, simulation results for Gallium Arsenide (GaAs) metal organic chemical vapor deposition (MOCVD) have been shown to agree with experimental data (see above figures). These CVD processes are of current interest to a number of US semiconductor companies and industrial CVD reactor suppliers (EMCORE, LAM, MOTOROLA). Using advanced MP algorithms, we have obtained over 65 billion operations per second performance in the solution phase of the SiC simulation. This is 46% of the peak performance of 1904 processors of Sandia's Intel Paragon and represents a significant increase in computational performance over state-of-the-art chemically reacting flow simulations. For this reason, we have recently been able to use MPSalsa in a numerical optimization of a prototype GaAS CVD reactor design. This optimization, using a gradient based technique (from DAKOTA) required the solution of about 100 full 3D reacting flow solutions. Each solution took 10 minutes on 512 processors of the Intel paragon.
The development of important enabling technologies, such as load balancing techniques, MP iterative solution methods, and chemical kinetics libraries was funded by DOE's Mathematical, Information, and Computational Sciences Division and by DOE Basic Energy Sciences.