Random forest based heating power prediction for single crystal furnace

The thermal output of the heating element undergoes temporal variations throughout the fabrication of monocrystalline silicon via the Czochralski process. The isodiameter stage is the most critical link in the whole crystal pulling process. This marks the shift from the neck growth phase to the phase where the crystal maintains a consistent diameter, significantly influencing the final quality of the crystal rods. The heater power has a direct influence on the heat field distribution inside the furnace. Given the difficulty of measuring this internal temperature distribution directly, achieving control objectives requires indirect methods, such as optimizing the temperature regulation path. Throughout the isodiameter phase, adjustments to the heater’s power output should be made dynamically using real-time production data to ensure stable hot zone temperatures. This study introduces a model for forecasting heater power, utilizing the random forest algorithm to enable real-time predictions. Test outcomes indicate that the model exhibits strong prediction accuracy and robustness, offering a novel technological solution for controlling crystal growth.