Description
The field of machine learning (ML) in plasma science has experienced an immense development in the past decade [1], advancing low-temperature plasma modeling and simulation, as well as experiments and diagnostics [2,3]. Part of this development relates to plasma processing, enabling fine control over tailored material properties at the nanoscale. In this study, plasma processing of SiO$_x$/Cu resistive switching devices [4] is investigated for reactive radio-frequency magnetron sputter deposition with Ar/O$_2$ through ML surrogate modeling, data-integrated physical simulations, and a data-driven classification of corresponding wafer-level measurements. Initially, Tridyn simulations [5] of the reactive plasma-surface interaction (PSI) during Ar and O$_2$ ion impingement on SiO$_x$ are conducted to establish an extensive data set. This is leveraged to devise a ML PSI surrogate model. The latter is subsequently utilized in an axially symmetric 2D particle-in-cell/Monte Carlo collision simulation with dynamical chemical surface coverage for the comprehensive prediction of discharge and surface conditions during reactive sputter deposition. Insights from this simulation study are correlated with a data-driven classification of measured electrical device characteristics. Specifically, a statistical analysis at the wafer-level is applied to over 50,000 devices to identify how processing conditions relate to device characteristics. The analysis reveals distinct device types attributed to the local process conditions during deposition, highlighting the importance of plasma process control in determining functional outcomes in nanoscale electronic devices.
[1] R. Anirudh et al., 2022 Review of Data-Driven Plasma Science, IEEE Transactions on Plasma Science 51, 1750 (2023)
[2] J. Trieschmann et al., Review: Machine learning for advancing low-temperature plasma modeling and simulation, Journal of Micro/Nanopatterning, Materials, and Metrology 22, 041504 (2023)
[3] M. He et al., Data-driven plasma science: A new perspective on modeling, diagnostics, and applications through machine learning, Plasma Processes and Polymers 21, 2400020 (2024)
[4] R. Lamprecht et al., Thickness-Related Analog Switching in SiO/Cu/SiO Memristive Devices for Neuromorphic Applications, Advanced Engineering Materials 27, 2401824 (2025)
[5] W. Möller and W. Eckstein, Tridyn - A TRIM simulation code including dynamic composition changes, Nuclear Instruments and Methods in Physics Research Section B: Beam Interactions with Materials and Atoms 2, 814 (1984)