Machine learning workflows for optimization of plasma-based devices

In this study, we employed machine learning (ML) techniques to optimize operating parameters of plasma-based devices, including low-temperature plasma reactors used for microelectronics manufacturing as well as advanced laser-based accelerators. Our strategy utilized multi-fidelity training of unified surrogate models using simulation and experimental data. We leveraged available high-energy experimental data obtained from the BELLA iP2 beamline at LBNL to train our Gaussian Process (GP) models. By implementing Bayesian optimization and Gaussian processes using the PyTorch and BoTorch libraries, we efficiently explored parameter spaces. The results demonstrated that our ML-driven approach significantly enhances the predictive capabilities and optimization efficiency of plasma laser systems, providing valuable insights and improvements in plasma dynamics simulations.