Physical and numerical methods of speeding up particle codes and paralleling as applied to RF discharges

We demonstrate the means, both physical and numerical, for speeding up particle-in-cell (PIC) simulations of RF discharges. These include implicit movers, longer ion timesteps, lighter-mass ions, different weights for electrons and ions, and improved initial density profiles. By using these methods (singly or together) on Ar and O 2 RF discharges we were able to achieve speedups of six to 30 times with single-processor machines. In electrostatic 1d3v PIC simulations of RF discharges, the field solve is typically less than 1% of the work load. Even for 2d3v PIC simulations, the field solve can be a small percentage of the work load, especially when fast Fourier transform methods are used to solve the field. Thus, we can obtain significant gains by just paralleling particle processing (e.g., pushing/accumulating) without paralleling the field solve. We applied this simple scheme to conduct 1d3v and 2d3v PIC simulations of Ar RF discharges on two- and four-CPU symmetric multiprocessor machines and on a distributed network of workstations. For a fixed number of grid points, the speedup for this parallel particle processing became more linear with increasing number of particles. The combination of single-processor methods and paralleling makes run times for PIC codes more competitive with other types of codes.