Design and Manipulation of Electromagnetic Metasurface Based on Bidirectional Deep Neural Network

In the current research field, metasurface technology has attracted much attention because of its excellent electromagnetic wave manipulation ability, and is widely used in the field of micro and nano photonics. In the lower frequency band, the megahertz band, metasurface technology also shows great potential for development, especially in the field of medical imaging, which is becoming increasingly important. However, the design process of metasurfaces is intricate and computationally costly. To tackle this challenge, this paper introduces a deep learning-based approach for the electromagnetic control and design of metasurfaces in the megahertz frequency band. By developing a Deep Neural Network (DNN), we are able to rapidly predict the electromagnetic response of metasurfaces, substantially reducing the time and memory consumption required for simulations. Moreover, the application of deep learning technology in metasurface design extends beyond forward prediction to inverse design, allowing for the prediction of metasurface structures that meet specific electromagnetic response requirements from the design objectives. This method overcomes the limitations of traditional design methods, offering new possibilities for the advancement of metasurface technology.