Data set of fully kinetic simulations of capacitively coupled plasma discharges in argon for development of surrogate models

Capacitively coupled plasmas (CCPs) are a key technology in the semiconductor industry that has been widely studied to optimize applications such as etching and deposition [1]. Particle simulations are especially applicable to capture relevant kinetic effects [2], however the processing requirements can be extreme. The computational burden can be reduced through model assumptions or by limiting the set of particle interactions, however this may cause desired physical effects, such as plasma densities and ion energy distributions, to no longer be correctly represented. Machine learning and data-driven methods provide a possible solution by replacing simulations or operations during simulations with less computationally expensive surrogate models. Surrogate models are optimized using large and expensive simulated data sets to provide a functional dependence between input and output parameters that is significantly cheaper to evaluate. In this work, we present a data set suitable for the development of surrogate models of CCPs that was created using the particle-in-cell with Direct Simulation Monte Carlo (PIC-DSMC) method [3]. The inter-electrode distance was fixed at 4 cm and a single RF waveform was applied at 13.56 MHz. In total, 900 1D3v simulations were performed for pressures from 0.124 Pa to 50 Pa and voltages from 100 V to 1000 V, requiring approximately 3000 processing cores used continuously for 9 months. Preliminary results of surrogate models that capture the space-time dependence of the plasma state across the simulated conditions are presented.

[1] M. A. Lieberman and A. J. Lichtenberg, “Principles of Plasma Discharges and Materials Processing,” New York: Wiley, (2005).

[2] Z. Donko, Plasma Sources Sci. Technol. 20(2), 024001, (2011).

[3] A. Lietz, S. Rauf, J. Kenney, P. Tian, and M. Hopkins, Technical Report, Sandia National Laboratories, SAND2022-2699R, (2022).