Optimizing etching process recipe based on Kernel Ridge Regression

Exploring optimal recipes to reduce dimensional variations is critical in etching processes. Variations in critical dimensions that were acceptable previously can become problematic because of smaller node sizes and more complex structures. Dry etch can be a major source of variations and will be the focus of this research. Advanced Process Control (APC) has been widely studied in semiconductor manufacturing. Even though different APC methods have been developed to adjust recipes, it is challenging to explore an optimal recipe to achieve multiple critical dimensions. In this paper, a learning method based on Kernel Ridge Regression (KRR) is proposed to generate optimal recipes for multi-input multi-output (MIMO) systems. A KRR parameter optimization method is developed. To improve the recipe optimization process, a feedback fine tuning method is proposed. Experimental data in a dry etch process were collected and processed for model construction and recipe optimization. The results demonstrate the effectiveness of the proposed method in exploring optimal recipes for MIMO systems.

Automated summary

Research on optimizing recipes to reduce dimensional variations in etching processes is critical. A learning method based on Kernel Ridge Regression (KRR) is proposed to generate optimal recipes for multi-input multi-output (MIMO) systems. Experimental data from a dry etch process were used to demonstrate the effectiveness of the proposed method in exploring optimal recipes for MIMO systems.