Data Mining Helps to Find Organic Semiconductors

Researchers at the Technische Universität München (TUM) are using a new process to search for organic compounds that could, among other things, replace silicon in electronic components.

Algorithms can be used to calculate the probable conductivity. Both the carbon skeleton of the molecules and the functional groups can influence the conductivity.Photo: C. Kunkel / TUM

Algorithms can be used to calculate the probable conductivity. Both the carbon skeleton of the molecules and the functional groups can influence the conductivity.

Photo: C. Kunkel / TUM

Classic solar cells are made of silicon. The material has proven itself, but is not considered the optimal solution. For one thing, silicon is relatively rigid and brittle. On the other hand, a lot of energy has to be spent to win it over the different procedures. Organic semiconductor materials are lighter and more flexible in comparison. However, they can only become a serious alternative for the production of solar cells if they can achieve sufficient efficiency and keep up with regard to their service life. A team at TUM has now found a way to accelerate the search for suitable organic semiconductors.

Researchers calculate conductivity on the computer

Suitable organic compounds could not only contribute to better results in solar cells. For example, they are also used for displays or for light-emitting diodes (OLED). The team headed by the chemist Karsten Reuter focuses on organic compounds whose central framework is based on carbon atoms. For scientists, however, research is like finding a needle in a haystack. Because the molecules and the materials formed from them have very different properties, depending on the exact structure and composition of the molecules. Accordingly, there are countless possible combinations and probably many promising candidates for the desired applications. “So far, it has been a big problem to track them down,” says Reuter. “It takes weeks to months to produce, test and optimize a new material in the lab. “With computational screening, we can accelerate this process enormously.”

So the researchers initially put aside test tubes and Bunsen burners. Instead, they work with a powerful computer and use algorithms to analyze data. These already exist and come from the Cambridge Structural Database. What is new, however, is the targeted search for relationships and patterns – called data mining. The term mining comes originally from the mining industry and translates as “mining” or “extraction”. So it’s about getting important information from the data. “The key to data mining is that you know what you are looking for,” explains Project Manager Harald Oberhofer. “In our case, it’s the electrical conductivity. High conductivity, for example, causes a lot of current to flow in the photovoltaic cell when sunlight excites the molecules. “

Artificial intelligence designs new connections

Scientists are looking for extra-programmed algorithms certain physical parameters related to conductivity. An example is the so-called coupling. The larger this value, the faster an electron moves from one molecule to the next. Another factor is the reorganization energy: when a molecule is supplied with electrical charge, it has to adapt its structure to this new charge, so it has to reorganize itself. In turn, it uses energy for this process. The less it takes, the better the conductivity, because less energy is lost.

In this way, the researchers have already analyzed structural data from about 64,000 organic monocrystals and grouped them in clusters. In doing so, they found that two factors influence conductivity, namely both the carbon-based framework of the molecules and the “functional groups,” that is, the compounds that hang laterally on the central framework. Accordingly, the clusters show which frameworks and which functional groups allow good charge transport and are thus particularly suitable for the development of electronic components. “Not only can we now predict the properties of a molecule, but we can also use artificial intelligence to design new compounds in which both scaffold and functional groups promise very good conductivity,” explains Reuter. In The next step will be practical implementation: the molecules redesigned on the computer are produced in the laboratory and their conductivity is tested in practice.

Further contributions to research topics:

  • Particle accelerator on a microchip
  • New supercomputer “SpiNNaker”
  • Waveguide technology solves space problem in the car cockpit