Meta-Learned and TCAD-Assisted Sampling in Semiconductor Laser Annealing.

Area of Science:

• Semiconductor Manufacturing
• Machine Learning
• Computational Modeling

Background:

• Machine learning (ML) application in semiconductor manufacturing is common, but efficient search space sampling remains underexplored for critical processes like lithography, annealing, deposition, and etching.
• Advanced semiconductor processes are costly, making efficient experimental design crucial for developing accurate predictive models with minimal trials.

Purpose of the Study:

• To propose and validate a novel technology computer-aided design (TCAD)-assisted meta-learned sampling approach for optimizing semiconductor manufacturing processes.
• To minimize experimental trials required for constructing accurate predictive models in complex manufacturing environments.

Main Methods:

• Development of a meta-learner that dynamically adjusts the hybridization strategy between TCAD simulations and ML for selecting optimal sampling points.
• Implementation and testing of the TCAD-assisted meta-learned sampling algorithm using laser annealing as a case study.

Main Results:

• The proposed TCAD-assisted meta-learned sampling method achieved significantly lower mean square error (MSE) within the initial 100 sampling steps compared to a pure ML approach.
• The TCAD-assisted sampling approach prevented MSE degradation observed in pure TCAD methods between 200-400 sampling steps, demonstrating enhanced model stability.

Conclusions:

• The TCAD-assisted meta-learned sampling approach offers a more efficient and accurate method for predictive modeling in semiconductor manufacturing.
• This methodology holds potential for broader applications in other manufacturing sectors and applied machine intelligence fields.