Predicting Active Species in Radio–Frequency Nitrogen Plasma with Quantum and Classical Supervised Learning Models

Plasma sources are often used during the epitaxial synthesis of nitride thin film material samples to produce active nitrogen species that incorporate into the resulting material. The various active nitrogen species in the plasma have a significant impact on the quality and structure of the grown sample; thus, control over the relative populations of these species is critical to producing high quality material samples. Previously, the optimal control of these nitrogen species and their populations has been investigated by interpolating between data points measured across the processing space; however, this approach is limited in its ability to generalize beyond the domains of the recorded data set. For a radio–frequency plasma source operated in a molecular beam epitaxy chamber, we have acquired optical emission spectra and from them measured the relative concentrations of the active nitrogen species for nearly 2,000 different combinations of plasma source operating parameters spanning large portions of the parameter space not previously studied. Using this data, we have explored the accuracy of a variety of supervised learning techniques, including some which incorporate quantum computation, to model the relative concentrations of nitrogen species within radio–frequency nitrogen plasma based on the chosen operating parameters. We assess the ability of each trained learning algorithm to generalize outside the range of the collected data and explore the underlying physical relationships therein.