AI‐Enabled Inverse Design and Molecular Identification Using Phase Change Metamaterial Absorber

Abstract Tailoring metamaterial design for desired absorption in the mid‐infrared region is pivotal for molecular sensing and identification. Surface‐enhanced infrared absorption (SEIRA) spectroscopy with metamaterials promises accurate molecular fingerprint detection. However, challenges arise due to narrow resonant wavelengths and the extensive design space of the metamaterial, thus hindering its application in accurate molecular detection. This work introduces an integrated AI workflow for designing electrically tunable metamaterial absorbers using phase‐change materials for broadband molecular fingerprint retrieval and identification. A free‐form inverse design method based on a neural‐adjoint optimization technique is utilized to predict phase‐change metamaterial absorber (PCMA) designs for desired mid‐infrared absorption spectra. Using the trained inverse neural network model coupled with a genetic algorithm, nanostructure designs spanning the spectral range from 800 cm⁻¹ to 2000 cm⁻¹ are predicted. The phase change is achieved by integrating the designed PCMA with an energy‐efficient microheater, allowing resonant absorption peaks to be tuned via voltage pulses. The neural network‐predicted PCMA successfully detected molecular vibration fingerprints of four molecules and their mixtures. A machine learning model based on a support vector machine developed for molecular identification shows 100% identification accuracy. This AI‐driven approach for PCMA design and molecular identification via SEIRA marks a significant advancement in accurate molecular detection.