High-throughput Computation of Defects in Semiconductors

The properties of realistic materials are controlled by their defects. In CODIS we will perform a large- scale computational screening of defects in semiconductors, thereby potentially saving years of trial- and-error effort to find the optimal candidate for a specific application. The project will be the first to inter-relate growth conditions, defect types and concentrations, and functional transport properties. The study will combine several approaches developed by the two project partners and streamline them into an automated flow linked to a material-defects database. New methods from the artificial intelligence, especially machine learning classification and regression algorithms, will be applied to unveil hidden trends and relationships in order to accelerate the screening process. A novel understanding of semi-conductors according to their prevalent defect types will thus emerge. Specific focus will be put on the lattice thermal conductivity to elucidate the characteristic relationships between defects and the materials functional properties. In addition to its fundamental interest, this project will impact many industrial technologies where semiconductor defects are key to their performance.

Computer chips generate heat that must be dissipated as quickly as possible so that the chip is not destroyed. This requires special materials with particularly good heat conduction properties. Neither the electrons nor the lattice vibrations can propagate completely unhindered through the material. They can be scattered by irregularities in the material, so-called defects. At the same time external impurities must be introduced into these semi-conducting materials for their ability to function as logic gates. In the FWF project "Computation Of Defects In Semiconductors" (CODIS) researchers at the TU Wien, together with colleagues in France, China and the US set out to predict the influence of said impurities. Despite boron and phosphorous doped silicon forming the basis of almost all transistors the role impurities versus electrons in heat conduction was discussed controversially in the literature. Within CODIS found that at low carrier concentrations and temperatures phonon scattering by electrons is dominant, whereas at higher doping concentrations the scattering by point defects dominate. By scanning a large number of semiconductors relations between structure and effect on the thermal conductivity of specific defects was predicted. So-called DX impurities were shown to severely limit the thermal conductivity. These are also known as killer defects which limit the electronic conductivity and should be avoided. The structure of these defects was unknown for decades (D stands for defect and X for unknown). A combination of machine learning and an evolutionary strategy was developed, so that the atomic detail of these defects can now be revealed in a matter of weeks.