Smart design of crystal growth furnaces and processes

• Applicants: Dr. Natascha Dropka; Dr. David Linke
• Subject Area: Thermodynamics and Kinetics as well as Properties of Phases and Microstructure of Materials ; Automation, Mechatronics, Control Systems, Intelligent Technical Systems, Robotics ; Metallurgical, Thermal and Thermomechanical Treatment of Materials
• Term: since 2021
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 467401796

In recent years, there has been remarkable progress in the research of artificial intelligence (AI) and in the research of electro mobility based on power semiconductor devices and single crystals as their key components. Still, the process and equipment for the growth of single crystals by the preferred Czochralski (Cz) method are traditionally derived from time consuming and cost intensive computational fluid dynamics (CFD) simulations, combined with over simplified model experiments on a small scale. Based on the added value obtained through the combination of AI and crystal growth research fields, we aim to assess the potential of artificial neural networks (ANN) in derivation of the unknown relationships between the furnace design and process parameters from the one side and crystal size and targeted temperature distributions in the grown material from the other side, with a final goal to predict the optimal dimensions of the hot zone parts, optimal choice of the construction materials, heating powers, ambient gas pressures, crucible and crystal rotational rates etc. for any bulk grown material with any size, all in the real time. The novelty of this proposed approach consists in the hybrid combining CFD and ANN where CFD will be used to simulate transport phenomena taking place in the furnace during the crystal growth and generate a database for ANN training. In this way, the time consuming CFD simulations will be restricted to the singular step of data generation. Emphasis will be put on the investigation of the potential of dynamic deep neural networks, especially on the long short-term memory (LSTM) and its later descendants for modelling dynamic dependences characterizing Cz growth. Rare studies on the application of ANN in crystal growth in the literature used static feedforward networks of the kind multilayer perceptron to model dynamic dependences pertaining to crystal growth process.Through the bilateral cooperation between two research groups specialized in AI and crystal growth, their competences and complementarities, a basis for better understanding of the crystal growth processes and relationships between numerous parameters is feasible and project success highly plausible. Moreover, this project will offer opportunities for early career researcher to pursue an outstanding research program in the field of AI and material science/crystal growth that crosses the boundaries of disciplines and institutions.

• DFG Programme: Research Grants