We meet Professor Dr Makus Valtiner, the head of the Institute of Applied Physics in his big corner office at TU Wien Freihaus. Here, we get to enjoy the view over Vienna with him, while also discussing the methods in interface physics, the integration of old instruments into modern data flows, data sharing, and the future outlook on AI-ready data. The world Markus Valtiner studies begins exactly where mediums meet - at the subtle line between solid, liquid and gas. In his institute, an “interface” is not a sleek touchscreen, but the ultra-thin area where invisible boundaries control properties like hardness, stability, reactivity, and catalysis.

Interfaces are everywhere 

“I'm on the fundamental side of characterising interfaces, which is why I work with a lot of different fields. There are experts for friction, there are experts for materials, but there is always someone who looks at the surfaces and interfaces and develops methods. From my personal perspective, we research a lot on how these materials survive harsh environments like electrolysis, if they degrade, and what these mechanisms of degradation are.”

Whilst this sounds abstract at first, he then breaks it down: wherever there is a surface, it will always meet another medium. When a solid meets a liquid, the interface suddenly becomes a stage for friction and wear, for adhesion and the tiny interaction forces that decide whether two surfaces slide smoothly or grind themselves apart. The abstract notion of “interfaces” quickly turns into a tour through very concrete industrial problems. Their collaboration with Voestalpine, an Austrian steel manufacturer, focuses on ultra‑high‑strength steels, a tempting lightweight alternative to aluminium because they can be made extremely thin without losing strength. But these “super strong” steels become brittle when exposed to hydrogen. Here, instead of measuring their mechanical properties, the research focuses on how hydrogen actually crosses from the outside into the material and how that process can be slowed or even stopped.

Data-intensive methods

“Depending on which technique we are talking about, it can be 5 to 10 gigabytes per experimental day. That is a lot – more if it involves video with high framerate. As soon as you have images, it goes up, and it's very hard to compress many of our images, because you want to go back and look at details, and if you compress them out, you can't see many things that you would want to see.”

Valtiner’s institute uses atomic force microscopy (AFM) at solid-liquid interfaces to follow, almost atom by atom, how molecules arrange and move on a surface with extremely high spatial resolution. At vacuum interfaces, colleagues can even resolve individual atomic orbitals, turning abstract quantum objects into topographical images of the surface landscape. Imaging techniques - especially when videos with high frame rates are involved - can easily produce five to ten gigabytes per experimental day. By contrast, spectroscopic methods like X-ray photoelectron spectroscopy (XPS) produce countless individual spectra that are tiny in size, often just simple two-column datasets, but they accumulate into large series of measurements that have to be organised, stored and interpreted as carefully as the images themselves.

Challenges in sharing proprietary formats

"Most of the papers that come out of our department say data will be made available upon reasonable request on the repository – but we are setting up to do this differently in the future. But this takes time, of course, because we have a lot of different instruments, and some have proprietary software to even handle the data, and then without this software, it's a completely useless data anyway."

Valtiner explains that data storage starts with an institutional RAID server, offering redundancy across eight drives where up to three can fail without total loss – though last year’s failure tested that limit, requiring professional recovery that worked but ate up time. Sharing data remains a work in progress, since transitioning to open access lags due to proprietary instrument software that locks data into unreadable formats. To automate data flows, Valtiner’s team converts proprietary formats into open, machine-readable structures - either by exporting through vendor software when possible or evading the formats with custom scripts from a dedicated specialist. The goal is a seamless pipeline: data captured automatically, fed into self-built open-source analysis tools, and versioned on GitLab via Jupyter notebooks.

Here old machines pose the biggest hurdle - relics from 1994 or Windows XP era, air-gapped for virus protection, spitting out top data but demanding USB-stick shuttles that kill automation. For this legacy gear, they’re testing sealed containers as secure bridges. The aim is to automatically add DOIs to every data output, failed experiments included, stored without pre-sorting on infinite, cheap storage.

Future infrastructure for automated data flow

“Output needs to be structured in such a way that it also contains the metadata and everything to do machine learning and training - that is very complicated in experimental sciences. You have to explain what you did and why. This is nothing that automatically works. So, you have to force the students and researchers to enter metadata manually.”

Metadata automation remains tricky, especially for surfaces where context like preparation steps defies machine capture. One solution, proposed by data science students, uses near-field devices for user identification: before any measurement starts, researchers must spend five minutes logging what they plan to do - otherwise, the machine stays locked. Turning lab notebook scribbles into structured, appended metadata alongside automatic details like device ID, lab temperature, and timestamps, this seamless integration primes data for the AI era Valtiner sees rushing in, transforming analysis into material invention and parameter exploration. Leading the Institute of Applied Physics at TU Wien, he’s determined to pioneer the shift with ironclad databases, automated literature digestion and enough compute power to bridge atomistic simulations (currently stuck at 10x10 nanometers) to macroscopic scales - perhaps even cell‑level predictions. To Markus Valtiner all hinges on usable data: TU Wiens labs churn out terabytes yearly from spectra and other Datasets, “gold if made FAIR”, but worthless without proper flows and metadata - so new data must get it right from day one.

Contact

Markus Valtiner
Institute of Applied Physics
TU Wien
markus.valtiner@tuwien.ac.at

Center for Research Data Management
TU Wien
research.data@tuwien.ac.at