Quantized Inverse Design for Photonic Integrated Circuits.

Area of Science:

• Photonics and Computational Electromagnetics
• Integrated Circuit Design
• Materials Science and Engineering

Background:

• Photonic integrated circuit (PIC) inverse design faces significant computational memory challenges.
• Advanced fabrication techniques like two-photon polymerization (2PP) require more flexible optimization algorithms for multimaterial 3D structures.
• Existing methods struggle with complex constraints from novel fabrication processes.

Purpose of the Study:

• To develop a memory-efficient reverse-mode automatic differentiation framework for finite-difference time-domain (FDTD) simulations.
• To enable the design of complex, multimaterial 3D photonic structures using novel fabrication methods.
• To address computational limitations in the inverse design of PICs.

Main Methods:

• Implemented a memory-efficient reverse-mode automatic differentiation framework for FDTD simulations.
• Utilized straight-through gradient estimation to handle nondifferentiable shape parametrizations.
• Developed optimization algorithms capable of managing complex constraints from advanced fabrication processes.

Main Results:

• Successfully designed increasingly complex photonic structures to solve coupling problems in PICs.
• Demonstrated the framework's effectiveness in handling memory-intensive simulations.
• Validated the approach for creating 3D multimaterial structures with unique properties.

Conclusions:

• The proposed memory-efficient framework significantly advances PIC inverse design capabilities.
• The method facilitates the creation of complex photonic devices using novel fabrication techniques.
• This approach holds substantial potential for future PIC design and practical applications.