Increasing the Productivity of the Czochralski Process Applying Machine Learning

Monocrystalline silicon for the production of solar cells has to be provided cost-effectively in large quantities. The Czochralski (Cz) technique is currently the only manufacturing method that can ensure the availability of high-quality material in sufficient quantities. Even if the crystal growth technology is highly developed, increasing productivity while reducing crystallization costs remains a major challenge. The outstanding advantage of Cz technology is the possibility of the visual control of the growing crystal. This means that disturbances during the crystal growth process that may result in a loss of monocrystallinity referred to as structure loss can be identified in an early stage and appropriate countermeasures can be initiated. Visual monitoring of the entire growth process has to be automated to such an extent that a failure is automatically detected and an alarm is triggered. A reliable visual assessment must still be given by an experienced operator for the further course of action. In this article, we report on an ongoing project in which we are testing different possibilities to ensure an effective automated monitoring of the Cz process. In a continuation of the project, we also want to apply methods of machine learning in order to find any meaningful correlations between process parameters and crystal growth disorders.