Parallelization and performance of three-dimensional plasma simulation

Plasma-assisted materials processing is being applied for several applications including manufacture of integrated circuits. A need of these applications is the ability to computer simulate the plasma systems and plasma processes. The motion of particles in particle type plasma simulations is inherently parallel. Further, the solution to Poisson's equation and the calculation of local electric and magnetic fields (Maxwell's equations) requires usage of only data values from the immediate neighbors. This leads to the possibility of developing particle-type simulations on massively parallel processors. The primary goal of this work was to explore the avenues available to reduce the time taken to run a simulation to its convergence. We compared the execution times for completing some number of iterations on three different platforms: the HP 735, the Sun Sparc 10, and the MasPar MP2. We observed that by properly designing the parallel algorithms speed-ups of over 13 times more than an HP 735 and over 40 times more than a Sparc 10 could be achieved.<
>