CD Laboratory for Advanced Computational Design of Crystal Growth

This CD Laboratory aims to develop novel modelling methods for the virtual description of crystal growth processes in the semiconductor industry. Physics-based and data-driven modelling approaches will be combined to achieve the most efficient and predictive approach possible.

Semiconductors with a wide band gap have a great deal of potential for future applications in power electronics, particularly in the field of e-mobility: the laboratory is focussing on silicon carbide, or SiC for short, which is particularly promising in this context. SiC-based components can be operated at higher voltages and temperatures than would be possible with silicon-based components. At the same time, switching frequencies can be increased and power losses can be reduced by as much as half.

However, in order to enable the production of high-quality SiC crystals as a series product by means of physical vapour deposition, modelling methods are required that are able to predict the crystal growth processes as precisely as possible. In order to develop such methods, a central problem must be solved: Physics-based models on the subject, i.e. those based on the latest knowledge of the physical processes of crystal growth, are already well advanced in principle, but must use approximations in order to be able to carry out the invoices. On the other hand, data-driven models could generate extremely helpful information for describing the growth of SiC crystals by utilising the results of completed experiments, but these in turn are limited by the current lack of "big data" in this subject area: In order to function optimally, models of this type require, as their name suggests, a large amount of data and require a corresponding number of experiments to be carried out - but these are extremely expensive, especially in the area of crystal growth.

The CD Laboratory for Computer-aided Design of Crystal Growth Processes is therefore working on solving this problem by combining both types of models: so-called hybrid models are to be developed that combine physics-based and data-driven models in order to achieve the greatest predictive power. The aim is to work comprehensively and efficiently towards a method to optimally describe and predict the growth processes of SiC crystals, which are particularly promising for the semiconductor industry - and to make said method transferable to other crystals in the future.