High-Performance Computing in Plasma Reactor Scale Simulations: A Review and Prospects

Non-equilibrium plasmas phenomena are notoriously difficult to model accurately and even more difficult to simulate computationally. This results from variety of physical and chemical processes that constitute the plasma phenomena. For example, electron kinetics, ion dynamics, fluid flow, gas and surface chemistry, conjugate heat transfer, and electromagnetics comprise some of the basic processes that define a typical plasma reactor operation. These processes have very different mathematical character that cannot be represented within a single discretization approach. For example, while the mass and energy transport processes can be represented as transient convection-diffusion-reaction equations, electrostatic coupling is purely elliptic character, while electromagnetic waves are hyperbolic: all of which require different computational solution approaches. The large number of governing equations and domain sizes have meant that computational problem sizes with tens of millions of unknowns is now typical. High-Performance Computing (HPC) is therefore a necessity.

This talk will discuss recent progress in HPC for plasma reactor simulations. While both task parallelism (domain decomposition) and data parallelism (thread) frameworks can be used in HPC of plasmas, a pure data parallel approach has inherent limitations owing to the disparity in the solution approaches for the different equations resulting from their mathematical character described above. A hybrid parallel approach that exploit task and data parallelism has shown promise subject to HPC hardware being used. We will discuss examples of HPC for reactor simulations that exploit the above frameworks. Upcoming hardware architectures and their implications for HPC for plasma simulations will also be discussed.