Functional design of metamaterial absorbers using a combined convolutional neural network-genetic algorithm approach

Area of Science:

• Metamaterials and Nanophotonics
• Computational Electromagnetics
• Materials Science

Background:

• Metamaterial absorbers are crucial for advanced photonic devices but face design challenges due to computational expense and structural limitations.
• Existing design methods often rely on predefined patterns, limiting exploration of optimal topological spaces.

Purpose of the Study:

• To develop an efficient functional design framework for metamaterial absorbers using a CNN-GA integration.
• To enable rapid exploration of topological space for high-performance absorber design.
• To reduce the computational cost associated with metamaterial absorber inverse design.

Main Methods:

• A shape grammar-based encoding scheme represents MXene resonator patterns as binary matrices.
• A CNN surrogate model is trained on finite-element simulation data to predict absorption spectra.
• A GA optimizes structural patterns and dielectric thickness guided by CNN predictions.

Main Results:

• The CNN accurately predicts absorption spectra (0.8-3.3 μm) with low mean absolute error (<0.02).
• Optimized absorbers achieve average absorption rates exceeding 94% with diverse patterns.
• The framework designs polarization-insensitive absorbers rapidly by modifying the fitness function.
• The surrogate-assisted GA reduces full-wave simulations by a factor of 30.

Conclusions:

• The integrated CNN-GA framework significantly accelerates the inverse design of high-performance metamaterial absorbers.
• The approach offers a reusable and efficient method for exploring complex metamaterial designs.
• This method enables the design of absorbers with tailored properties, such as polarization insensitivity, without symmetry constraints.