KIT’s high-performance computer optimizes materials for medical technology

The Karlsruhe Institute of Technology (KIT) has unveiled a cutting-edge €1.2 million computer dedicated to materials research. This powerful tool can predict material properties even before they are produced, significantly enhancing the potential for optimization, including in the realm of COVID rapid tests.

Dr.-Ing. Michael Selzer from the Institute for Applied Materials (IAM) leads the development of the high-performance software and, in parallel, is developing the research data management system Kadi4Mat (Karlsruhe Data Infrastructure for Materials Science) at the Institute for Nanotechnology (INT). In an interview with COMPAMED-tradefair.com, he describes the possibilities of the computer.

How does this high-performance computer work?

Dr.-Ing. Michael Selzer: Several computers are connected in a network and linked by a high-performance network. The computing power is equivalent to that of 10,000 laptops. For this purpose, algorithms must be developed that decompose a problem into corresponding subproblems and distribute them to different nodes. The intermediate results are exchanged among each other via the network.

The software has been developed by us for 25 years. It comprises about one million program lines, 60 employees are dedicated to the development of the physical models and the software. Particularly complex are tasks with a 3D-question. These can only be solved quickly using HPC systems (High Performance Computing).

Particularly demanding are flow-dependent variables, for example the calculation of the properties of membrane structures in medical technology, which are found in pregnancy tests or in COVID tests. We use simulations to investigate how different variants of the membrane structure influence the test process.

What values have to be entered to obtain a material forecast?

Selzer: The models we use are all physical. That means there are equations that describe the behavior based on natural laws. In addition, material coefficients are used, which are obtained from different methods. These can be simulative methods, atomistic methods or fundamental methods that provide parameters. Experimentally determined parameters are also used, which include uncertainties.

How could the COVID-19 rapid test be improved in the test trial by the computer?

Selzer: We have investigated the liquid spreading in the test strips: What geometric properties, what wetting properties does the membrane with the corresponding structure have? For the quality of the test, it is important how fast a liquid rises to a certain height in the membrane. This speed influences the reaction possibilities of the test liquid.

We know a similar behavior from pregnancy tests, drug tests and sugar tests for diabetes. All these applications have different requirements because they function differently. A kind of portfolio of materials is needed, which we are now building. In a structure database, you can then look up which membrane is best suited for which application, just like in a catalog.

What role does AI play in this construction?

Selzer: When building such databases, one uses the possibility of applying machine learning. This allows us to move through the parameter spaces in a more targeted way. With changes to the geometry or the chemical wetting agents, we can turn many adjusting screws. Since this takes a lot of time, we try to work out variations in a more targeted way using machine learning. To do this, the data must be available in a structured form, because this is the only way to automate the determination of parameter variations.

If we can predict parameters quickly and in a targeted manner, this leads to an acceleration of measurement development and to a more effective use of costs and resources and means greater sustainability.