Maximum contributed component regression for the inverse problem in optical scatterometry.

Area of Science:

• Materials Science
• Metrology
• Semiconductor Manufacturing

Background:

• Scatterometry is crucial for microelectronic manufacturing process monitoring.
• Inverse problems in scatterometry typically rely on model-based methods (e.g., library search, Levenberg-Marquardt, Artificial Neural Networks (ANNs)).
• Existing model-based and model-free methods require predefined geometric models or extensive labeled data, increasing costs and time.

Purpose of the Study:

• To develop a novel model-free scatterometry method for determining 3D profile structures.
• To address the limitations of existing methods regarding geometric model assumptions and data requirements.
• To reduce the cost and time associated with obtaining labeled data in microelectronic manufacturing.

Main Methods:

• Developed a novel model-free method named Maximum Contributed Component Regression (MCCR).
• Utilized Canonical Correlation Analysis (CCA) to estimate maximum contributed components from unlabeled and labeled data.
• Integrated maximum contributed components into conventional regression methods to solve the inverse problem.

Main Results:

• MCCR effectively estimates 3D profile structures without requiring a predefined geometric model.
• The method demonstrates high accuracy even with a small amount of expensive labeled data.
• Experimental results on synthetic and real-world semiconductor datasets validate the effectiveness of MCCR.

Conclusions:

• MCCR offers a cost-effective and efficient solution for the inverse problem in scatterometry.
• The proposed method significantly reduces the reliance on extensive labeled data, overcoming a major bottleneck in model-free approaches.
• MCCR shows strong potential for application in advanced microelectronic manufacturing process monitoring.