Multifunctional Movable-Type Coding Metasurface Enabling Reconfigurable diffractive neural network

With the rapid development of artificial electromagnetic (EM) media, metamaterials have achieved remarkable progress over the past two decades, demonstrating powerful capabilities in manipulating the phase, amplitude, wavelength, and polarization of EM waves. In this context, microwave diffractive neural networks (DNNs) have recently emerged as a promising paradigm for intelligent EM computing, showing significant potential in tasks such as logical operations, target detection, and parameter sensing. However, current metasurface-based EM computing schemes predominantly rely on photoelectric or thermal modulation mechanisms, which typically require frequency-specific materials and are often accompanied by high fabrication costs, large power consumption, and complex control systems, thereby limiting their practical deployment in DNNs. Mechanical reconfigurable metasurfaces, which achieve dynamic EM control through simple operations such as meta-atom rotation or flipping, offer a more flexible and low-complexity alternative. Nevertheless, the stringent rotation-angle requirements of conventional mechanical designs further complicate reconfiguration. Consequently, there is a strong need for mechanically reconfigurable metasurfaces that support flexible assembly and reusability, so as to enhance the efficiency, scalability, and practicality of microwave diffractive neural networks.

 

In a recent paper published in Light: Science & Applications, a team of scientists led by Professor Tiejun Cui and Jianwei You from the Department of Electronics and the State Key Laboratory of Millimeter Waves at Southeast University, China, report a reconfigurable transmissive metasurface inspired by ancient Chinese movable-type printing technology. The proposed design adopts a modular architecture composed of detachable meta-atoms, in which eight categories of meta-atoms with distinct EM responses serve as functional “molds.” These meta-atoms can be mechanically reassembled into arbitrary configurations, enabling efficient post-fabrication reconfiguration while significantly reducing fabrication complexity and cost.

 

Owing to its modularity, reconfigurability, and reusability, the proposed “movable-type coding metasurface” supports multiple EM functionalities within a single hardware platform. By cascading multiple metasurface layers, the researchers construct a movable-type reconfigurable diffractive neural network (MT-RDNN), which can be rapidly adapted to new tasks by replacing only a small subset of meta-atoms, rather than redesigning or retraining the entire network. This unit-level reconfiguration strategy enables efficient task transfer while maintaining high performance. In addition, a single-layer movable-type coding metasurface provides versatile wavefront-shaping capabilities and supports a range of EM applications, including holographic imaging and contactless human vital-sign sensing. For example, by selectively replacing specific meta-atoms, EM waves can be dynamically focused at different spatial locations to enhance the signal-to-noise ratio for multi-person vital-sign monitoring.

 

The researchers summarize the working principle of their approach as follows:

“We design a movable-type coding metasurface composed of reusable meta-atoms with discrete EM responses. By selectively assembling these meta-atoms, the metasurface can be configured to perform diverse tasks, including diffractive neural network-based classification, electromagnetic holography, and contactless human vital-sign sensing. Furthermore, a movable-type reconfigurable diffractive neural network (MT-RDNN) is constructed with multiple metasurface layers cascaded, which can be efficiently adapted to new learning tasks by replacing only a small subset of meta-atoms.”

 

“Thanks to the unit-level modularity of the meta-atoms, task adaptation can be realized with substantially reduced computational and physical reconfiguration overhead,” the scientists added. “For instance, the MT-RDNN can be transferred from handwritten digit recognition to letter classification by modifying only the final metasurface layer, while maintaining high classification accuracy.” they further explained.

 

The proposed modular and reusable metasurface paradigm provides a new pathway toward low-cost, scalable, and task-adaptive intelligent EM systems. “By further integrating mechanical reconfiguration with electrical or optical tuning mechanisms, the proposed strategy could enable multifunctional and reprogrammable metasurface platforms for future applications in wireless communications, human-machine interaction, healthcare monitoring, and intelligent sensing, without the need for complex hardware redesign,” the scientists forecast.

---

---