Node-Loss Detection of Silicon Single Crystal with Dual-Channel Cross-Modality Feature Enhancement

During the Czochralski (Cz) method production of silicon single crystal, dislocations are prone to occur. In the equal-diameter stage, dislocations manifest as node loss, and existing technologies struggle to detect the phenomenon of node loss.For this purpose, this study proposes a node loss detection method based on cross-modality feature enhancement and fusion. First, the growth process parameters of the Cz silicon single crystal are transformed into a time-frequency diagram using the Continuous Wavelet Transform. Then, a dual-channel feature extraction net-work is constructed, using the Temporal Convolution Module to extract the temporal features of the original signal, and using the Attention-based Multi-scale Convolution Module to extract multi-scale spatial features from the time-frequency diagram.Finally, to better integrate features from different modalities, a Cross-Modality Feature Enhancement and Fusion Module is proposed. This model enhances and fuses temporal features and spatial features to obtain defect features rich in node loss information. Experimental results show that this method can effectively detect node loss during the equal-diameter growth process of silicon single crystal and has high accuracy and robustness.