Deep learning improves performance of topological bending waveguides.

Area of Science:

• Photonics
• Materials Science
• Artificial Intelligence

Background:

• Topological photonics offers unique light manipulation properties.
• Propagation loss in photonic devices, especially at sharp bends, remains a challenge.
• Current design methods for complex photonic structures can be inefficient.

Purpose of the Study:

• To introduce design informatics, leveraging deep learning, for optimizing topological photonic systems.
• To apply this method to a topological waveguide with a sharp bending structure to minimize propagation loss.
• To demonstrate the versatility and applicability of the proposed design approach.

Main Methods:

• Utilized deep learning algorithms for parameter design within a topological waveguide system.
• Engineered a sharp bend using two photonic crystals with C6v symmetric dielectrics in hexagonal triangle lattices.
• Optimized 6x6 unit cells near the bending region through deep learning-driven parameter design.

Main Results:

• Achieved a 60% improvement in output compared to the initial waveguide structure.
• Successfully reduced propagation loss in the topological waveguide with a sharp bend.
• Demonstrated the effectiveness of deep learning in optimizing complex photonic structures.

Conclusions:

• Design informatics with deep learning provides an efficient method for optimizing topological photonic structures.
• The proposed method significantly reduces propagation loss in sharp bending waveguide designs.
• This approach shows high versatility and broad applicability for various photonic device designs.