Optimization of Surface Kinetics Schemes via Machine Learning Methods

Accurate modeling of surface kinetics is crucial for plasma-enhanced processes and plays a significant role in various scientific and technological domains. This study presents an optimization approach to refine surface kinetic parameters, primarily focused on model-free methods, such as DE and CMA-ES. The approach is applied to the surface kinetics of recombination of atomic oxygen in oxygen-containing plasmas using the LoKI (Lisbon Kinetics) simulation tool [1]. The refined parameters enable the minimization of the discrepancy between simulated and experimental results of atomic oxygen surface recombination probability. Moreover, uncertainty propagation via Monte Carlo sampling is conducted to assess the robustness and confidence of the predictions done by the kinetic model, accounting for parameter variability of input data such as species concentrations and temperatures. The methodology presented exhibits a significant improvement in both predictive accuracy and reliability, thereby facilitating a more comprehensive modeling of plasma-surface interactions. Additionally, the proposed workflow is agnostic to the underlying kinetic model and can readily be extended to the optimization of other reaction schemes with different observables.

[1] https://nprime.tecnico.ulisboa.pt/loki/