Chipping value prediction for dicing saw based on sparrow search algorithm and neural networks

• Published in: The Journal of supercomputing Vol. 80; no. 6; pp. 7483 - 7506
• Main Authors: Shi, Jun, Zhang, Peiyi, Du, Sihan, Liang, Wanyong, Cao, Weifeng, Li, Qingbo, Hou, Hechao
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.04.2024; Springer Nature B.V
• Subjects: Algorithms; Chipping; Compilers; Computer Science; Integrated circuits; Interpreters; Machine learning; Neural networks; Principal components analysis; Processor Architectures; Programming Languages; Search algorithms; Wafer dicing; CLSSA; Dicing quality; BILSTM
• ISSN: 0920-8542; 1573-0484
• DOI: 10.1007/s11227-023-05728-9

Wafer dicing is a key step in the manufacturing process of semiconductor integrated circuits, and the quality of dicing directly affects the dimensional accuracy and economic benefits of the final product. However, the current inspection of dicing quality is usually carried out after the dicing is completed, which will lead to a decrease in dicing efficiency and an increase in costs. In order to solve these problems, we propose a deep learning-based method, which uses the improved sparrow search algorithm (SSA) to optimize the bidirectional long short-term memory (BILSTM) network to predict the dicing quality. Firstly, the importance of each feature is analysed using kernel principal component analysis for the data collected in the field, and the main feature components are selected as the feature input according to the contribution rate. Secondly, the SSA is improved using the idea of chaotic mapping and Levy flight, and the improved algorithm is used to optimize the hyperparameters of the BILSTM network. Finally, we evaluated the proposed method using the actual dicing dataset. Experimental results show that this method is superior to other prediction algorithms. The average prediction accuracy is 93.44%.