A Data-Driven Approach for Metasurface Design Using Conditional Genertaive Adversarial Network (cGAN)

Deep learning has become the very fascinating and popular tool to solve complex and non-linear databased problems in artificial intelligence. The inverse design of the metasurface is a very challenging task due to the non-linear relationship between spectrum and unit cell design. Conventional design of metasurface for target spectrum mainly is done using a trial and test basis which is very time-consuming and requires a lot of effort. In this paper, a Conditional Generative Adversarial Network (cGAN) has been used for the inverse design of a metasurface unit cell. Training of cGAN framework is done to generate metasurface designs for targeted sparameters with the rapid synthesis of structures to meet desired electromagnetic criteria. Mapping of sparameters and random noise to metasurface designs are done by the Generator while the discriminator ensures that the fabricated designs closely match the real structures by evaluating their authenticity. Results of the work reveal that the cGAN successfully models the relationship between spectrum data and metasurface configurations, achieving high design accuracy and diversity of structure. The proposed approach effectively increases the design process and provides an ascendable solution for engineering applications in photonics, antennas, and wavefront manipulation.