A High-Efficiency and Versatile Reconfigurable Intelligent Surface Design Paradigm with Novel Topological Representation

• This work presents a groundbreaking and versatile design paradigm for the automatic design of Reconfigurable Intelligent Surfaces (RIS), distinguished by its efficiency and adaptability across various functions and structures. At the core of this paradigm is the innovative Non-Uniform Rational B-Splines Topological Representation Method, which transforms complex 100-dimensional continuous patterns into simplified 5-dimensional NURBS control points. This transformative approach reduces the pattern solution space by a remarkable 20-fold, ensuring smooth excitation current flow and maximum metallic pattern resonance capability. • Further amplifying its impact, this paradigm introduces a separate design architecture that leverages terse neural networks alongside multiport network theory. This architecture elegantly resolves multistate mapping challenges inherent in RIS design, enabling rapid, within-one-second predictions of multistate electromagnetic responses and allowing for seamless reuse of datasets and models across different RIS configurations. • The effectiveness of this paradigm is validated through three typical designs: two high-performance RIS units and an ultra-wideband multilayer RIS. These examples clearly demonstrate the paradigm’s ability to meet design targets while significantly reducing reliance on traditional, experience-based methods. This paradigm not only addresses the complex problem of multistate electromagnetic responses but also establishes itself as a leading solution for efficient, scalable, and practical RIS design in real-world applications. With digital coding technology, reconfigurable intelligent surfaces (RISs) become powerful real-time systems for manipulating electromagnetic (EM) waves. However, most automatic RIS designs involve extensive numerical simulations of the unit, including the passive pattern and active devices, requiring high data acquisition and training costs. In addition, for passive patterns, the widely employed random pixelated method presents design efficiency and effectiveness challenges due to the massive pixel combinations and blocked excitation current flow in discrete patterns. To overcome these two critical problems, we propose a versatile RIS design paradigm with efficient topology representation and a separate design architecture. First, a non-uniform rational B-spline (NURBS) is introduced to represent continuous patterns and solve excitation current flow issues. This representation makes it possible to finely tune continuous patterns with several control points, greatly reducing the pattern solution space by 20-fold and facilitating RIS optimization. Then, employing multiport network theory to separate the passive pattern and active device from the unit, the separate design architecture significantly reduces the dataset acquisition cost by 62.5%. Through multistep multiport calculation, the multistate EM responses of the RIS under different structural combinations can be quickly obtained with only one prediction of pattern response, thereby achieving dataset and model reuse for different RIS designs. With a hybrid continuous-discrete optimization algorithm, three examples—including two typical high-performance RISs and an ultra-wideband multilayer RIS—are provided to validate the superiority of our paradigm. Our work offers an efficient solution for RIS automatic design, and the resulting structure is expected to boost RIS applications in the fields of wireless communication and sensing.