Anomaly Detection in Float-Zone Crystal Growth of Silicon

Float-Zone (FZ) crystal growth can manufacture silicon crystal with high purity and low oxygen concentration. However, due to the limited capability of FZ machines, oxidation contamination cannot be avoided, and an oxide layer sometimes would form on unmelted polycrystalline silicon surface. Oxide layers, so-called watermark, would negatively affect the crystal quality. Consequently, it is desirable to have an automated watermark detection in order to take corrective action in the early stages of the process. This paper proposed a complete framework for watermark detection on FZ images, including pre-processing, feature selection and classification. The results show it can act as a promising quality assurance tool for the FZ process.