Data-driven concurrent nanostructure optimization based on conditional generative adversarial networks.

Area of Science:

• Nanophotonics
• Computational Materials Science
• Machine Learning Applications

Background:

• Iterative numerical optimization is standard for designing optical nanostructures.
• A performance gap exists between simulated pristine structures and experimentally measured deformed profiles.
• Accurate simulation-to-fabrication mapping is crucial for reliable nanostructure design.

Purpose of the Study:

• To introduce conditional generative adversarial networks (CGAN) into the optimization loop for optical nanostructures.
• To learn and model the relationship between fabrication process parameters and resulting nanostructure shapes.
• To enable data-driven design of nanostructures that account for real-world fabrication imperfections.

Main Methods:

• Conditional Generative Adversarial Networks (CGAN) integrated into iterative optimization.
• Shifting optimization from structural parameters to process parameters (e.g., deposition rate, annealing time).
• Demonstration on designing metallic grating-based color filters (red, green, blue).

Main Results:

• CGAN learned complex nonlinear process-structure relationships.
• Generated simulation profiles closely matched experimental data across various fabrication conditions.
• CGAN-based optimization yielded higher figures of merit compared to standard optimization with pristine structures.

Conclusions:

• This data-driven approach accurately maps fabrication conditions to nanostructure designs.
• The method expedites nanostructure design by focusing on fabrication-accurate parameters.
• Optimal process parameters are automatically determined, improving experimental outcomes.