Paper PS2-MoM8
Automated Reduction of Plasma Chemistry Sets

Monday, October 21, 2019, 10:40 am, Room B130

Simulating plasma reactors in multi-gas mixtures easily leads to chemistry sets comprising dozens of species and many hundreds of reactions. Including such complex chemistry sets in spatially resolved plasma models quickly becomes infeasible due to the high computational cost. Hence, it is desirable to keep a chemistry set simple while preserving the behavior of the plasma with regards to the density etc. of key species such as the radicals interacting with the surface. We are developing an algorithm within the Quantemol Database (QDB) [1] to automate this simplification for specified process parameters such as pressure, power, and gas mixture [2]. The algorithm will select a minimum set of species and reactions from the entirety of the database, which produce the same results with regards to user-specified species and the desired accuracy.

The challenge here is to find a reduction method, which can be automated reliably with minimum human input and is computationally cheap enough to run within the QDB framework. One method, which satisfies the need for quick calculation times is to run a 0D model with the full set of reactions and species, identify species with negligible densities, and remove these and associated reactions from the chemistry set. This requires only one run of the 0D model and a check of the species densities with the set threshold density. However, there are a few pitfalls concerning reliable automation. For example, a specie might have a low density in the steady state solution but act as a precursor for a more numerous specie, which would be missed by such an algorithm.

On the other side of the spectrum is the Morris method [3] based on Monte Carlo techniques. Here, the rate coefficients for the specific reactions are randomly changed for each run of the model. The effect of each reaction on the plasma system can be evaluated by the perturbations of, for example, densities of specified species caused by the variation in the rate coefficients. Reactions with low impact can be removed as well as species whose reactions showed no significant effect. This method is much more reliable without additional human input but requires a large number of simulation runs to gather enough data. Hence, it might be unfeasible to be used within the QDB infrastructure.

Given this, we require a method between these two extremes. Here, we will present our assessment of different methods, the current stage of development, and examples for chemistry reduction for specific process parameters.

[1] Tennyson J et al. Plasma Sources Sci. Technol. 26 (2017) 055014

[2] Ayilaran A, J Plasma Sci. Tech., 21 (2019) 064006

[2] Morris M D Technometrics 33 (1991) 161