Machine Learning Applied to Electron Beam Lithography to Accelerate Process Optimization of a Contact Hole Layer

Area of Science:

• Materials Science
• Computer Science
• Electrical Engineering

Background:

• High-resolution patterning is critical for advanced chip manufacturing.
• Lithography process optimization is complex and time-consuming due to nonlinear system dynamics.
• Current methods require extensive experimental trials, creating bottlenecks in development.

Purpose of the Study:

• To develop an efficient process optimization solution for contact layers in metal oxide nanoparticle photoresists.
• To leverage machine learning to predict and classify optimal lithographic process conditions.
• To accelerate the optimization and control of lithographic processes for semiconductor manufacturing.

Main Methods:

• Integration of electron beam lithography (EBL) experiments with machine learning algorithms.
• Development of a long short-term memory (LSTM) network for contact hole imaging prediction.
• Implementation of a support vector machine (SVM) model for process condition classification.

Main Results:

• The LSTM network accurately predicted horizontal and vertical contact widths, showing high consistency with EBL experimental data.
• The combined SVM and LSTM approach effectively screened process conditions meeting critical dimension uniformity tolerances.
• The classification model demonstrated good accuracy in identifying optimal process parameters.

Conclusions:

• The proposed machine learning-based solution significantly enhances the efficiency of lithographic process optimization.
• This approach provides a reliable reference for optimizing contact layers in semiconductor fabrication.
• Combining EBL with LSTM and SVM offers a powerful strategy for overcoming bottlenecks in chip manufacturing.