Research on Classification Algorithm of Silicon Single-Crystal Growth Temperature Gradient Trend Based on Multi-Level...

Area of Science:

• Materials Science and Engineering
• Semiconductor Manufacturing
• Artificial Intelligence in Industrial Processes

Background:

• Silicon single-crystal preparation requires precise control of furnace conditions for high-quality output.
• Abnormal thermal field conditions can lead to crystal defects or complete growth failure.
• Timely detection and adjustment of anomalies are critical for successful silicon crystal production.

Purpose of the Study:

• To develop an algorithm for early identification of abnormal conditions in silicon single-crystal growth.
• To accurately classify temperature gradient trends using multi-level feature fusion.
• To enhance the stability and quality of silicon single crystals in industrial production.

Main Methods:

• A silicon single-crystal growth temperature gradient trend classification algorithm based on multi-level feature fusion.
• Categorization of temperature gradient trends using expert knowledge and qualitative analysis.
• Fusion of original, shallow statistical, and deep learning features with mutual information-based weighting.
• Classification of trend changes using a Deep Belief Network (DBN) model.

Main Results:

• The proposed algorithm effectively predicts changing trends in thermal field temperature gradients.
• Demonstrated ability to capture process dynamics and classify trend changes accurately.
• Experimental validation confirms the algorithm's efficacy in identifying temperature gradient shifts.

Conclusions:

• The developed algorithm significantly improves the accuracy of fault trend prediction in silicon crystal preparation.
• Early detection of thermal field anomalies minimizes product quality issues and production interruptions.
• This approach supports the stable growth of high-quality silicon single crystals for industrial demands.