Thermal field temperature prediction based on RTH-LSTM for the growth process of semiconductor silicon single crystals

The accurate monitoring of thermal field temperature is a key factor in achieving precise control of crystal diameter and real-time quality diagnosis during the growth process of silicon single crystals. However, direct measurement of the thermal field temperature using physical means is challenging in actual silicon single crystal growth. To address this, an RTH-LSTM-based method for thermal field temperature prediction is proposed. By incorporating the Red-Tailed Hawk (RTH) optimization algorithm, the core hyperparameters, such as the regularization coefficient, of the Long Short-Term Memory (LSTM) network are optimized, enabling automatic selection of the optimal parameter configuration. This approach eliminates the subjectivity involved in manual parameter selection and enhances the robustness of thermal field temperature prediction. Finally, the prediction performance of the proposed model is validated using actual semiconductor-grade crystal growth data, demonstrating that the model significantly outperforms other models in terms of accuracy and generalization.