Global Inverse Design across Multiple Photonic Structure Classes Using Generative Deep Learning

• DOI: 10.1002/adom.202100548
• Publication Date: 2021-07-17T07:45:43Z
• Authors: Christopher Yeung; Ryan Tsai; Benjamin Pham; Brian King; Yusaku Kawagoe; David Ho; Julia Liang; Mark W. Knight; Aaswath P. Raman
• Keywords: Generative model; Structural Coloration

Understanding how nano- or micro-scale structures and material properties can be optimally configured to attain specific functionalities remains a fundamental challenge. Photonic metasurfaces, for instance, spectrally tuned through choice structural geometry achieve unique optical responses. However, existing numerical design methods require prior identification of material-structure combinations, device classes, as the starting point optimization. As such, unified solution that simultaneously optimizes across materials geometries has yet realized. To overcome these challenges, we present global deep learning-based inverse framework, where conditional convolutional generative adversarial network is trained on colored images encoded with range parameters, including refractive index, plasma frequency, geometric design. We demonstrate that, in response target absorption spectra, identify an effective metasurface terms its class, properties, overall shape. Furthermore, model arrive at multiple variants distinct nearly identical spectra. Our proposed framework thus important step towards photonics strategies combinations categories, parameters which algorithmically deliver sought functionality.