Metasurface is a two-dimensional planar metamaterial composed of artificial sub-wavelength meta-atoms, and it can manipulate the wavefronts of electromagnetic waves. The rapid development of metasurfaces has spawned a variety of functional devices, including metasurface holography, metalens, vector vortex beam generator, polarization converter, etc. As can be seen, metasurfaces hold considerable promise for planar optical components and devices owing to their unprecedented ability to arbitrarily manipulate the amplitude, phase, frequency, and polarization of light. However, previous research on metasurfaces is based on passive metasurfaces with their intrinsic physical properties and structure parameters fixed and untunable after fabrication, which limits their practical applications such as dynamic control of the wavefront of light. As opposed to passive metasurfaces, the structure, properties, and functions of active metasurfaces can be tuned flexibly through external stimuli such as electricity, magnet, light, heat, and stress. With the help of active metasurfaces, the transmitted, reflected, or diffracted light can be dynamically manipulated by simultaneously or individually controlling the phase, amplitude, and polarization of light.

Recently, many efforts have been devoted to studying various tuning methods-based tunable metasurfaces with multiple functions to dynamically manipulate and control electromagnetic waves as needed. The working principle of tunable metasurfaces mainly involves two mechanisms: one is based on active materials, while the other is based on nano-mechanical structural reconfiguration.

The widely used active materials mainly include liquid crystals (LCs) and phase-change materials. LCs are active materials commonly used in optics owing to their broad tuning range of the refractive index. Recently, the combinations of LCs and plasmonic metasurfaces (Fig. 4) or dielectric metasurfaces (Fig. 5) are demonstrated with different functionalities including beam steering, zoom lens, and dynamic color display. The LC-based tunable metasurfaces have large modulation contrast and can cover the modulation range of 0-2π. In addition, LC-based metasurfaces have advantages such as high efficiency and low bias voltage requirements.

Another commonly used active material refers to phase-change materials including but not limited to chalcogenide-glass and vanadium oxide. Thanks to their tunable optical properties induced by external stimuli, phase-change materials have emerged as a class of active materials integrated into the metasurface. In recent years, various chalcogenide-glass (Fig. 6 and Fig. 7) or vanadium oxide-based (Fig. 8 and Fig. 9) tunable metasurfaces are investigated in the field of tunable optical switch, perfector absorber, and beam steering. The phase-change materials-based metasurfaces have large modulation amplitude and wide phase modulation range owing to a large refractive index change during phase transition.

Apart from active materials, tunable metasurfaces can be also achieved by nano-mechanical structural reconfiguration, such as flexible materials and microelectromechanical systems (MEMS). Recently, flexible materials (Fig. 10) and MEMS-based (Fig. 11 and Fig. 12) tunable metasurfaces are intensively studied, and their applications in zoom lenses, dynamic color display, and dynamic waveplates are explored. The flexible materials-based metasurfaces are relatively easy to be designed and fabricated, and they have several advantages such as a large phase modulation range (0-2π), large modulation depth, and high device efficiency. The MEMS-based metasurfaces have several advantages such as low loss, large modulation depth, the ability to cover the phase modulation range of 0-2π, high efficiency, and ease of integration with chips.

Although tunable metasurfaces have made significant progress, they still have low modulation speeds and are unable to control each element. Therefore, researchers need to make efforts in materials, fabrication technology, and design and optimization methods.

First, new optical materials with large refractive index, faster response speed, higher sensitivity to external stimuli such as electricity, heat, light, and mechanical forces, and better compatibility with nanofabrication technologies are needed, so as to achieve faster and more accurate dynamic control of metasurfaces. In recent years, new materials such as conductive oxides, transition metal nitrides, Ⅲ-V semiconductor compounds, chiral materials, black phosphorus, and conductive polymers have attracted researchers' attention due to their excellent optical properties.

Second, the nanofabrication technologies currently used for metasurfaces (such as electron beam lithography and focused ion-beam etching) are expensive and difficult to fabricate large-area metasurfaces with sub-wavelength-scale features due to long fabrication time and low yield. Therefore, a new nanofabrication technology that can balance yield, precision, fabrication time, and cost is urgently required. Nanoimprint lithography technology (especially ultraviolet roll-to-roll nanoimprint lithography) and projection lithography technology that can create large-scale nano-patterns in a short time are important research directions for subsequent metasurface fabrication.

Third, the design and optimization of tunable metasurfaces require a system design framework from a single meta-atomic to macroscopic optical systems, which connects all key stages of material characterization, fabrication, and system operation in a coordinated manner. Topological optimization is a typical example of local optical optimization. Global optimization algorithms such as genetic algorithms (GA), ant colony optimization algorithms (ACO), particle swarm optimization (PSO), covariance matrix adaptation evolution strategy (CMA-ES), and multi-objective optimizers (MOO) have played an important role in the design and optimization of tunable metasurfaces. In addition, the rapid development of machine learning in recent years has provided a promising tool for optimizing optical metasurfaces. Using extracted metasurface structures and their corresponding optical responses to train neural networks can help to study the relationship between geometric structures and optical responses. This approach enables accurate and fast forward prediction of the optical response for a given metasurface, as well as application in inverse design where the desired optical response is used to deduce the device layout and characteristic parameters of the metasurface.