(346n) High-Throughput XAFS for Machine-Learning Accelerated Discovery of ALD Ternary Oxides

Titanium dioxide, TiO₂, is an important catalytic material with a multitude of uses in the energy sector including electrolysis, carrier extraction, air purification, and water treatment. Most relevant to work in electrocatalysis is its application as a multifunctional coating, capable of electronic tuning and corrosion stabilization. Modification of TiO₂ through the intentional introduction of transition metals or main group elements allows for control over parameters such as band position, carrier transport, interfacial electron transfer, and surface active-sites.

In this study, titanium manganese oxide (TiMnO_x) thin films were grown using atomic layer deposition (ALD). The addition of Mn interlayers introduces a tunable intermediate band which can be modified through varying the concentration or oxidation state of the manganese.

The amorphous films were annealed in air or in an argon atmosphere at 400°C, 500°C, and 600°C resulting in various degrees of crystallinity, a parameter which is correlated with electronic transport and stability. Such a figure of merit, as summarized in the Figure, is quantified with high-throughput acquisition and processing of synchrotron x-ray absorption data. It will also be cross validated with x-ray diffraction and TEM characterization, but only for a limited subset in a vast parameter space of ALD processing conditions.

The biggest challenge is that even though a vast parameter space can be accessed via ALD synthesis, including compositions, porosity, and atomic structures which can match up with high throughput characterizations available to us, including synchrotron x-ray absorption and x-ray diffraction, we must pick a subset of samples instead of exploring a vast parameter space. So far, many properties such as intermediate-band electronic structures can only be measured at low throughput.

We propose to explore the use of machine learning (ML) to accelerate the discovery of this new multi-functional coatings, specifically, ternary metal oxides synthesized by ALD deposition with atomic-scale structural control. ML will help mine data and establish a correlation model from the local chemical environment suggested by XANES data and suggest precise processing condition to obtain desired structures. ML is particularly powerful when our understanding of structure-property relationship is not complete.