Tobias Gergs

This study examines the development of TiN/SiO_x/Cu/SiO_x/TiN memristive devices for neuromorphic applications using wedge‐type deposition and Monte Carlo simulations. Identifying critical parameters for the desired device characteristics can be challenging with conventional trial‐and‐error methods, which often obscure the effects of varying layer compositions. By employing an off‐center thermal evaporation method, a thickness gradient of SiO_x and Cu on a 4 inch wafer is created, facilitating detailed resistance map analysis through semiautomatic measurements. This approach allows for investigating the influence of layer composition and thickness while keeping other process conditions constant. Combining experimental data with simulations provides a precise understanding of layer thickness distribution and its impact on device performance. Optimizing the SiO_x layers to be below 12 nm, coupled with a discontinuous Cu layer with a nominal thickness under 0.6 nm, exhibits analog switching properties with an R_on/R_off ratio of >100, suitable for neuromorphic applications, while R × A and power exponent γ analysis show signs of multiple conduction mechanisms. The findings highlight the importance of SiO_x and Cu thickness in determining switching behavior, offering insights for developing high‐performance analog switching components for bioinspired computing systems.

Low-temperature plasmas (LTPs) consist of a mixture of neutral and charged species in thermal non-equilibrium. The governing discharge physics and chemistry lead to a zoo of (excited) species interacting with bounding surfaces. Their modification (e.g., activation, etching, functionalization) is exploited in the frame of plasma processing (e.g., plasma-enhanced etching, deposition, catalysis). LTPs also render a dominant part of the many steps required in semiconductor device manufacturing. This may involve a diverse portfolio of plasma discharges, chemical precursors, surface materials, and process recipe steps to achieve desired deposit/etch patterns.^1,2 It may as well include indirect plasma discharges, for example, in the generation of extreme ultraviolet light for state of the art 13.5 nm wavelength lithography, required for device feature scales of the order of 10 nm and beyond (for instance, termed 3 nm nodes).³ Modeling and numerical simulation contribute an integral component in the design and optimization of related plasma processes. In the recent past decade, data-driven methods such as machine learning (ML), deep learning (DL), and artificial neural networks (ANNs), among other computational statistics methods, have (re-)gained enormous interest—which may be subsumed under the umbrella term artificial intelligence (AI). Although these fields have seen a renaissance and have continued to rapidly develop in the past decade, many fundamental concepts have been well established in computational statistics and related disciplines. The past and present interest in the context of plasma processing may be attributed to the potential for hidden pattern detection in correlated data and, particularly, the efficiency in related optimization tasks.^{4 – 12}

Understanding and modeling plasma–surface interactions frame a multi-scale as well as multi-physics problem. Scale-bridging machine learning surface surrogate models have been demonstrated to perceive the fundamental atomic fidelity for the physical vapor deposition of pure metals. However, the immense computational cost of the data-generating simulations render a practical application with predictions on relevant timescales impracticable. This issue is resolved in this work for the sputter deposition of AlN in Ar/N₂ discharges by developing a scheme that populates the parameter spaces effectively. Hybrid reactive molecular dynamics/time-stamped force-bias Monte Carlo simulations of randomized plasma-surface interactions/diffusion processes are used to setup a physics-separating artificial neural network. The application of this generic machine learning model to a specific experimental reference case study enables the systematic analysis of the particle flux emission as well as underlying system state (e.g. composition, density, point defect structure) evolution within process times of up to 45 min.

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 (PSIs) 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 PSI 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.

Liquid‐infused surfaces are based upon the infusion of a liquid phase into a porous solid material to induce slippery and repellent character. In this context, porous SiOx plasma polymer films represent a relevant candidate for a robust nanoporous carrier layer. Intermittent low‐pressure plasma etching of O₂/hexamethyldisiloxane‐derived coatings is investigated to enhance the intrinsic porosity inherent to residual hydrocarbons in the silica matrix. Simulations of the resulting Si–O ring network structure using reactive molecular dynamics indicate formation of interconnected voids with Si–OH functionalized pore walls allowing water penetration with almost Fickian diffusive behavior. The corresponding porosity of up to 18%, well agreeing with simulations, Fourier‐transform infrared spectroscopy, and ellipsometry measurements, was found to be suitable for the liquid infusion of polyethylene glycol molecules into about 80 nm thick SiOx films providing ongoing lubricating properties, thus revealing their suitability as liquid‐infused surfaces.

Simulations of thin film sputter deposition require the separation of the plasma and material transport in the gas phase from the growth/sputtering processes at the bounding surfaces (e.g., substrate and target). Interface models based on analytic expressions or look-up tables inherently restrict this complex interaction to a bare minimum. A machine learning model has recently been shown to overcome this remedy for Ar ions bombarding a Ti-Al composite target. However, the chosen network structure (i.e., a multilayer perceptron, MLP) provides approximately 4×106 degrees of freedom, which bears the risk of overfitting the relevant dynamics and complicating the model to an unreliable extent. This work proposes a conceptually more sophisticated but parameterwise simplified regression artificial neural network for an extended scenario, considering a variable instead of a single fixed Ti-Al stoichiometry. A convolutional β-variational autoencoder is trained to reduce the high-dimensional energy-angular distribution of sputtered particles to a low-dimensional latent representation with only two components. In addition to a primary decoder that is trained to reconstruct the input energy-angular distribution, a secondary decoder is employed to reconstruct the mean energy of incident Ar ions as well as the present Ti-Al composition. The mutual latent space is hence conditioned on these quantities. The trained primary decoder of the variational autoencoder network is subsequently transferred to a regression network, for which only the mapping to the particular low-dimensional space has to be learned. While obtaining a competitive performance, the number of degrees of freedom is drastically reduced to 15 111 (0.378% of the MLP) and 486 (0.012% of the MLP) parameters for the primary decoder and the remaining regression network, respectively. The underlying methodology is very general and can easily be extended to more complex physical descriptions (e.g., taking into account dynamical surface properties) with a minimal amount of data required.

Variable charge models (e.g., electronegativity equalization method (EEM), charge equilibration (QEq), electrostatic plus (ES+)) used in reactive molecular dynamics simulations often inherently impose a global charge transfer between atoms (approximating each system as an ideal metal). Consequently, most surface processes (e.g., adsorption, desorption, deposition, sputtering) are affected, potentially causing dubious dynamics. This issue has been addressed by certain split charge variants (i.e., split charge equilibration (SQE), redoxSQE) through a distance-dependent bond hardness, by the atomic charge ACKS2 and QTPIE models, which are based on the Kohn–Sham density functional theory, as well as by an electronegativity screening extension to the QEq model (approximating each system as an ideal insulator). In a brief review of the QEq and the QTPIE model, their applicability for studying surface interactions is assessed in this work. Following this evaluation, a revised generalization of the QEq and QTPIE models is proposed and formulated, called the charge-transfer equilibration model or in short the QTE model. This method is based on the equilibration of charge-transfer variables, which locally constrain the split charge transfer per unit time (i.e., due to overlapping orbitals) without any kind of bond hardness specification. Furthermore, a formalism relying solely on atomic charges is obtained by a respective transformation, employing an extended Lagrangian method. We moreover propose a mirror boundary condition and its implementation to accelerate surface investigations. The models proposed in this work facilitate reactive molecular dynamics simulations, which describe various materials and surface phenomena appropriately.

External electric circuits attached to radio-frequency plasma discharges are essential for the power transfer into the discharge and are, therefore, a key element for plasma operation. Many plasma simulations, however, simplify or even neglect the external network. This is because a solution of the circuit’s auxiliary differential equations following Kirchhoff’s laws is required, which can become a tedious task especially for large circuits. This work proposes a method which allows one to include electric circuits in any desired radio-frequency plasma simulation. Conceptually, arbitrarily complex external networks may be incorporated in the form of a simple netlist. The suggested approach is based on the harmonic balance concept, which splits the whole system into the nonlinear plasma and the linear circuit contribution. A mathematical formulation of the influence of the applied voltage on the current for each specific harmonic is required and proposed. It is demonstrated that this method is applicable for both simple global plasma models and more complex spatially resolved Particle-in-Cell simulations.