Physics-separating artificial neural networks for predicting initial stages of Al sputtering and thin film deposition in Ar plasma discharges

Abstract

Simulations of Al thin film sputter depositions rely on accurate plasma and surface interaction models. Establishing the latter commonly requires a higher level of abstraction and means to dismiss the fundamental atomic fidelity. Previous works on sputtering processes addressed this issue by establishing machine learning surrogate models, which include a basic surface state (i.e., stoichiometry) as static input. In this work, an evolving surface state and defect structure are introduced to jointly describe sputtering and growth with physics-separating artificial neural networks. The data describing the plasma-surface interactions stem from hybrid reactive molecular dynamics/time-stamped force bias Monte Carlo simulations of Al neutrals and Ar+ ions impinging onto Al(001) surfaces. It is demonstrated that the fundamental processes are comprehensively described by taking the surface state as well as defect structure into account. Hence, a machine learning plasma-surface interaction surrogate model is established that resolves the inherent kinetics with high physical fidelity. The resulting model is not restricted to input from modeling and simulation, but may similarly be applied to experimental input data.

Review: Machine learning for advancing low-temperature plasma modeling and simulation

“…140 Second, a high-throughput, randomized data generation scheme was pursued to efficiently populate the relevant parameter space with hybrid MD/tfMC simulations. 138 Third, two CVAE were trained to predict either the PSIs or the diffusion processes. They were combined to form a PSNN 133 as PSI ML surrogate model, which can readily be used to complement either plasma simulations or experimental diagnostics.…”

“…This is of major importance since most PSIs (e.g., surface chemical reactions, ion radiation) are subject to excessive degrees of freedom and cannot straightforwardly be evaluated (e.g., reaction path ways, 0D to 3D defect structures). Remedies combine ML algorithms for clustering (e.g., Bayesian Gaussian mixture model), dimensionality reduction (e.g., harmonic linear discriminant analysis), or surrogate modeling (e.g., PSNN) with high-throughput simulations (e.g., randomized trajectories, 138 amorphous surface site melt-quenching 111 ) or enhanced sampling methods (e.g., variationally enhanced sampling, metadynamics). 106,111,138,220 Notably, GNNs have also been demonstrated to significantly accelerate atomic simulations.…”

“…Remedies combine ML algorithms for clustering (e.g., Bayesian Gaussian mixture model), dimensionality reduction (e.g., harmonic linear discriminant analysis), or surrogate modeling (e.g., PSNN) with high-throughput simulations (e.g., randomized trajectories, 138 amorphous surface site melt-quenching 111 ) or enhanced sampling methods (e.g., variationally enhanced sampling, metadynamics). 106,111,138,220 Notably, GNNs have also been demonstrated to significantly accelerate atomic simulations. 99 An active learning scheme as applied by Diaw et al 177 when conducting MD simulations of warm dense matter to setup a corresponding surrogate model may be considered to secure spending available resources most effectively.…”

“…States were found by detecting relevant positional changes of adatoms during MD simulations, clustering them (unsupervised), and eventually sorting them (supervised). Gergs et al 138 used a PSNN to generalize on previously sampled but disjoined system state trajectories. The system alternately underwent PSIs and diffusion processes for given experimental discharge conditions, continuously evolving the system state (e.g., composition, point defect populations).…”

Review and future perspective of feature scale profile modeling for high-performance semiconductor devices

“…After this improvement, fast calculation comparable to process time is the next issue as the second factor. Currently, machine learning, 100 – 104 fusion physical model with machine learning, 105 , 106 and surrogate models 107 , 108 are adopted to solve this issue. Figure 48 shows an example of a fusion model (or hybrid model) in which incident gas fluxes derived from machine learning using real-time monitoring of plasma and EES were used as an input and feature scale profiles, and damage distributions were simulated by a physical model.…”