ICMCTF 2025 Session CM4-2-MoA: Simulations, Machine Learning and Data Science for Materials Design and Discovery II

The energy transition of our society requires an improved fundamental understanding of the chemical interaction of H with oxide materials, such as oxide membranes for H-purification, oxides for photocatalytic water splitting, and passivated oxides on steel. In particular, the effect of H impurities on the barrier properties of oxide layers grown by Atomic Layer Deposition (ALD) is of great scientific and technological interest, since hydrogen permeation barriers fabricated by ALD are broadly applied to address specific challenges for transport, handling, and storage of H, as well as for electronics, catalysis and gas sensing. However, resolving tiny changes in local chemical bonding states and structure of e.g. amorphous alumina oxides, as induced by H impurities originating from the ALD deposition process, still poses huge challenges for modern analytical tools up to date.

Here, the effect of hydrogen on the local chemical bonding states and structures of amorphous ALD alumina films is disclosed by predicting Auger parameter shifts, as measured by XPS/HAXPES, using a combination of atomistic and electrostatic modeling. First of all, it is demonstrated that a conventional melt quenching simulation procedure is not applicable for generating representative amorphous oxide structures with different H contents and densities, as observed in the experiment. Instead, a novel approach is proposed for simulating amorphous H-containing oxide structures by annealing reconstructed, highly defective crystalline hydroxide structures using a universal machine learning interatomic potential. As such, excellent agreement between the density, structure, and H-content was obtained between theory and experiment. Moreover, measured Auger parameter shifts for Al as a function of the H-content were accurately predicted by assuming all H atoms to be present in the form of hydroxyl ligands in the randomly interconnected 4-fold, 5-fold, and 6-fold nearest-coordination spheres of Al (by O). Combined atomistic and electrostatic modeling shows in detail how measured Auger shifts depend on the complex correlations between local coordination, bond lengths, bond angles, and ligand type(s) around the core-ionized atom, which equally applies to amorphous and crystalline compounds.

This work enables the computational design of new barrier coatings for hydrogen economy, providing a comprehensive computational characterization framework able to interpret even the tiniest Auger parameter chemical shifts obtained from experimental XPS/HAXPES techniques.

Crystalline zinc oxide thin films are important due to a combination of optical transparency, electrical conductivity and piezoelectric and pyroelectric properties. These functional properties largely depend on perfection of the crystalline structure. Reproducing the growth of thin films by molecular-dynamics (MD) simulations is very useful for the disentanglement of processes and phenomena which take place in parallel in the experiment and can yield a lot of atomic-scale information which is difficult to access experimentally. After introducing MD simulations in general, classical MD based on a reactive force field is used to study the atom-by-atom growth of ZnO_x films on a crystalline template. The effect of kinetic energy of fast atoms (E_fast) and fraction of fast atoms (f_fast) at varied elemental ratio (x = [O]/[Zn]) is analyzed in a wide range. Following the visual inspection, the crystallinity is quantified in terms of network ring statistics.

Simulations with fixed f_fast = 100% revealed that the highest crystal quality was obtained at x = 1.03 (not at the intuitive ratio x = 1.00) and E_fast = 3-12 eV with a maximum at E_fast = 10 eV. When only a low E_fast = 1 eV is available, even higher x = 1.10 leads to the relatively best results. Simulations with varied f_fast and fixed E_fast = 10 eV revealed that the crystallinity at f_fast ≥ 50% is saturated. The ratio x = 1.03 is optimum at all these f_fast values and it is followed by x = 1.05 which also leads to higher crystal quality than x = 1.00. However, the ratio x = 1.10 is for the present energy distribution functions too high, not only in terms of growth rate which decreases with increasing x, but also in terms of crystal quality.

Zinc gallium oxide is a metal oxide gas sensing layer with exceptional thermal and chemical stability, capable of detecting gases such as NO, NO₂, CO, CO₂, and H₂S. The work function of Zinc gallium oxide can be assessed through first-principles calculations based on Density Functional Theory, which allows for the prediction of the sensor's sensitivity. Although Density Functional Theory provides accurate computational results, its high computational cost and time requirements limit its applicability for large-scale surface screening. This study used a database based on a density functional theory-based zinc gallium oxide sensor model. We developed an automated workflow using Python programming to extract crystal structure features as input for the machine learning model. The processed and filtered input features were employed to predict the work function of the sensor model, achieving a mean absolute percentage error below 6% in the prediction results. This study presents a trained machine-learning model interface that allows users to input crystal structure files for the rapid and accurate evaluation of the work function of Zinc gallium oxide sensors.

Using first-principles calculations, we investigate electronic transport through carbon-, oxide- and transition metal dichalcogenide (TMD)-based nanojunctions for nanoelectronic applications. Effects of biasing, defecting, contacting and quantum interfering are addressed. We study noncollinear interlayer exchange coupling in magnetic trilayers for spintronic application. Effects of spacing are uncovered. We also study the efficiencies of hybrid-protected perovskite quantum dot films for LED backlighting and hydrogen evolution reaction in oxides for sustainable-energy applications. Effects of polymer-adsorbing and material-configuring are highlighted. Moreover, we explore the plasmonic properties of complex transition metal nitrides for photonic applications.