TCN-Based Melting Stage Identification in Monocrystalline Silicon

To address the issue of seed-initiating failures caused by human misjudgment during the melting stage of monocrystalline silicon growth via the traditional Czochralski (CZ) method, this paper proposes an image sequence classification approach based on temporal convolutional network (TCN) to enhance melting stage identification accuracy. By extracting multi-frame image features from video sequences and modeling them with a TCN framework, this study achieves efficient classification between "unmelted" and "fully melted" states. Experimental results demonstrate high classification accuracy, outperforming traditional CNN methods.