A trifunctional terahertz metasurface and a free-topology inverse design method thereof

Abstract

The application provides a trifunctional terahertz super surface and a free topology reverse design method thereof, which are sequentially arranged from bottom to top as an Au metal back plate layer, a PTFE dielectric layer and a top layer tunable mixed pattern layer; the top layer tunable mixed pattern layer is composed of a fixed function skeleton and a top layer silicon tunable region; the fixed function skeleton comprises a non-complete silicon arc-shaped open ring, a nested silicon square resonator and a plurality of VO2 stepped pixel blocks distributed at the edge positions of the top layer tunable unit; the nested silicon square resonator comprises a first silicon resonant unit located at the opening direction position of the non-complete silicon arc-shaped open ring and two second silicon resonant units located at the diagonal positions of the non-opening direction of the non-complete silicon arc-shaped open ring; and the top layer silicon tunable region comprises a second local control area located at the lower half of the non-complete silicon arc-shaped open ring, a third local control area located on the first silicon resonant unit and a first local control area located on the second silicon resonant unit.

Description

Technical Field

[0001] This invention relates to the field of terahertz metasurface devices, specifically a three-functional terahertz metasurface and its free topology reverse design method. In particular, it relates to a three-functional switchable terahertz metasurface based on a dual tunable material of silicon and vanadium dioxide, and a free topology reverse design method for this type of multifunctional terahertz metasurface, which can be applied to scenarios such as 6G wireless communication, integrated sensing, terahertz imaging, and electromagnetic control. Background Technology

[0002] With the large-scale deployment of cutting-edge technologies such as 6G mobile communication, integrated sensing, and terahertz imaging in the terahertz band, electromagnetically modulated metasurface devices are facing new performance requirements: devices need to possess comprehensive capabilities such as ultra-thin and compact structure, wide operating bandwidth, integration of multiple electromagnetic functions, and independent control of multiple operating modes. Reflective terahertz metasurfaces, with their advantages of no transmission loss, simple fabrication, and high degree of controllability, have become the preferred choice for core devices in absorption and polarization conversion. Specifically, broadband absorption can suppress stray reflections and parasitic electromagnetic interference in the system; linear-circular polarization conversion enables polarization multiplexing and circular polarization transmission and reception control; and linear-linear polarization conversion can eliminate link polarization mismatch loss and improve communication transmission stability. If a single device can integrate these three functions, it will significantly reduce the size of terahertz front-end hardware and lower system complexity, possessing extremely high engineering application value.

[0003] However, existing terahertz multifunctional metasurfaces generally suffer from the following problems: First, most existing multifunctional reconfigurable metasurfaces adopt a multilayered structure of alternating stacks of metal, dielectric, and phase change materials, typically consisting of 4 to 7 layers. This multilayer stacking not only increases the number of micro-nano processing steps such as thin film deposition, photolithography, and etching, thus increasing fabrication costs and process errors, but also creates complex interlayer electromagnetic coupling between the multilayer dielectric and metal layers. This can easily lead to mutual compression of the operating bandwidth of each function, making it impossible to achieve broadband stable operation of all three functions simultaneously, and thus failing to meet the broadband requirements of 6G terahertz systems. Secondly, most tunable metasurfaces rely on a single phase change material or a single photoelectric control method to achieve functional switching. The wave absorption and two types of polarization conversion functions share the same electromagnetic resonance channel. The electromagnetic responses of different working modes are deeply coupled, making it difficult to achieve relatively independent control of different working modes. Third, traditional metasurface design relies entirely on the electromagnetic experience of researchers and adopts a forward design approach that uses finite geometric parameters for traversal scanning. This approach can only fine-tune parameters for regular fixed topologies such as rectangles, rings, and crosses, resulting in two major drawbacks: First, the massive parameter scanning leads to huge consumption of full-wave simulation computing power, resulting in low design efficiency; second, it is constrained by simple regular geometric shapes and cannot generate resonant patterns with complex free topologies, making it difficult to explore the ultimate electromagnetic control performance of metasurfaces.

[0004] In summary, existing technologies lack a three-functional, dual-tunable terahertz metasurface with few layers, a compact structure, and weak coupling across multiple modes. Furthermore, they lack an intelligent reverse design method that adapts to free topologies, considers fabrication constraints, and enables high-precision, rapid inversion. Summary of the Invention The purpose of this invention is to provide a three-functional terahertz metasurface and its free topology reverse design method. By constructing a hierarchical control mechanism of "VO2 phase state controlling the main functional mode and silicon conductivity controlling the polariton mode", three working states of broadband absorption, linear-circular polarization conversion and linear-linear polarization conversion are realized on the same three-functional terahertz metasurface. It also provides an end-to-end free topology reverse design method that generates the top-level free topology pattern from the three-channel target response of the target absorption spectrum, linear-linear polarization conversion rate and linear-circular polarization conversion rate.

[0005] To achieve the above objectives, this technical solution provides a trifunctional terahertz metasurface, comprising: An Au metal backsheet layer, a PTFE dielectric layer, and a top tunable hybrid pattern layer are arranged sequentially from bottom to top along the z-direction. The top tunable hybrid pattern layer includes at least one top tunable unit, which is composed of a fixed functional skeleton and a top silicon tunable region. The fixed functional skeleton includes a non-complete silicon arc-shaped opening ring, a nested silicon square resonator, and multiple VO2 stepped pixel blocks distributed at the edge of the top tunable unit. The nested silicon square resonator includes a first silicon resonator located in the opening direction of the non-complete silicon arc-shaped opening ring, and two second silicon resonator units located diagonally opposite each other in the non-opening direction of the non-complete silicon arc-shaped opening ring. The top silicon tunable region includes a second local control region located in the lower half of the non-complete silicon arc-shaped opening ring, a third local control region located on the first silicon resonator unit, and a first local control region located on the second silicon resonator unit.

[0006] Compared with existing technologies, this technical solution has the following characteristics and beneficial effects: (1) This solution can realize three functions of broadband absorption, linear-circular polarization conversion and linear-linear polarization conversion on the same device platform by using only a three-layer structure of Au metal backplate layer, PTFE dielectric layer and top tunable hybrid pattern layer, which significantly improves the multi-functional integration and structural compactness.

[0007] (2) This invention proposes a hierarchical control mechanism of “VO2 phase state control main function mode and silicon conductivity control polaron mode”, which maps function switching and polaron mode formation to different physical degrees of freedom, thereby reducing the coupling between different functions and improving the controllability and bandwidth balance of multifunctional response.

[0008] (3) This invention transforms the free topology design problem from a high-dimensional pixel search to a constrained low-dimensional structure coding problem by using 45° axisymmetric constraints and 36-bit symmetric binary encoding. This significantly reduces the design space while ensuring manufacturability, and improves data utilization efficiency and model convergence stability.

[0009] (4) The free topology design model proposed in this invention introduces both FSCA dual attention mechanism and LSTM / Transformer dual branch coding structure, which can take into account both local spectrum detail modeling and global correlation expression, and realize high-precision inverse mapping of target multi-channel electromagnetic response to free topology pattern.

[0010] (5) This invention is applicable to scenarios such as 6G wireless communication, polarization multiplexing links, integrated sensing systems and terahertz imaging, and can provide methodological support for the rapid design and engineering implementation of high-degree-of-freedom, multifunctional terahertz metasurfaces. Attached Figure Description

[0011] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a top view of the overall structure and unit structure of the three-functional terahertz metasurface of this scheme.

[0012] Figure 2 This is an absorption spectrum of a three-functional terahertz metasurface in broadband absorption mode, according to one embodiment of this scheme.

[0013] Figure 3 This is an impedance matching analysis diagram of a three-functional terahertz metasurface in broadband absorption mode, according to one embodiment of this scheme.

[0014] Figure 4 This is a reflection coefficient curve of a three-functional terahertz metasurface in a linear-circular polarization conversion mode, according to an embodiment of this scheme, under co-polarization and cross-polarization conditions.

[0015] Figure 5 This is a phase difference curve of a three-functional terahertz metasurface in linear-circular polarization conversion mode, according to one embodiment of this scheme.

[0016] Figure 6 This is an ellipticity curve of a three-functional terahertz metasurface in linear-circular polarization conversion mode, according to one embodiment of this scheme.

[0017] Figure 7 This is a PCR curve of a three-functional terahertz metasurface in linear-to-linear polarization conversion mode, according to one embodiment of this scheme.

[0018] Figure 8 This is a diagram showing the phase and phase difference of the u and v reflection components of a three-functional terahertz metasurface in a linear-to-linear polarization conversion mode, according to one embodiment of this scheme.

[0019] Figure 9 This is a diagram of the co-polarization and cross-polarization reflection components of a three-functional terahertz metasurface in a linear-to-linear polarization conversion mode, according to one embodiment of this scheme.

[0020] Figure 10 This is a schematic diagram showing the decomposition relationship between the u / v intrinsic polarization coordinate system and the x / y coordinate system of a three-functional terahertz metasurface according to an embodiment of this scheme.

[0021] Figure 11 This is a surface electric and magnetic field distribution diagram of a three-functional terahertz metasurface at a representative frequency point, which is an embodiment of this scheme.

[0022] Figure 12 This is a schematic diagram of the top-level pattern binarization and symmetric encoding of a three-functional terahertz metasurface according to an embodiment of this scheme.

[0023] Figure 13 This is a network architecture diagram of a free topology reverse design model.

[0024] Figure 14 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0025] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of this specification. Rather, they are merely examples of apparatuses and methods consistent with some aspects of one or more embodiments of this specification as detailed in the appended claims.

[0026] It should be noted that the steps of the corresponding methods are not necessarily performed in the order shown and described in this specification in other embodiments. In some other embodiments, the methods may include more or fewer steps than described in this specification. Furthermore, a single step described in this specification may be broken down into multiple steps in other embodiments; and multiple steps described in this specification may be combined into a single step in other embodiments.

[0027] Example 1 like Figure 1 As shown, this solution provides a trifunctional terahertz metasurface, comprising the following: An Au metal backsheet layer, a PTFE dielectric layer, and a top tunable hybrid pattern layer are arranged sequentially from bottom to top along the z-direction. The top tunable hybrid pattern layer includes at least one top tunable unit, which is composed of a fixed functional skeleton and a top silicon tunable region. The fixed functional skeleton includes a non-complete silicon arc-shaped opening ring, a nested silicon square resonator, and multiple VO2 stepped pixel blocks distributed at the edge of the top tunable unit. The nested silicon square resonator includes a first silicon resonator located in the opening direction of the non-complete silicon arc-shaped opening ring, and two second silicon resonator units located diagonally opposite each other in the non-opening direction of the non-complete silicon arc-shaped opening ring. The top silicon tunable region includes a second local control region located in the lower half of the non-complete silicon arc-shaped opening ring, a third local control region located on the first silicon resonator unit, and a first local control region located on the second silicon resonator unit.

[0028] As mentioned earlier, the three-functional terahertz metasurface in this scheme adopts a three-layer reflective stacked structure. The Au metal backplane is used to suppress transmission and provide the basis for the reflective cavity. It is equivalent to an ideal conductor in the operating frequency band and can completely suppress transmission, achieving absorption and polarization conversion only through the reflection channel. The PTFE dielectric layer is used to provide electromagnetic coupling space and impedance matching conditions, that is, to provide the electromagnetic coupling space required for standing wave formation, power dissipation, and impedance matching. The top tunable hybrid pattern layer is responsible for realizing multimode resonance, polarization modulation, and function switching. In particular, the top tunable hybrid pattern layer is not a completely free random pixel plane, but is composed of a fixed functional skeleton and a top silicon tunable region: the fixed functional skeleton includes a non-complete silicon arc-shaped open ring and nested silicon square resonators, which can ensure that the device maintains a strong local field and obvious anisotropic response in the target frequency band; the free topology reverse design is mainly designed for the top silicon tunable region to adjust the filling / empty state of the silicon resonant structure; the VO2 pixel block is used for phase transition modulation and is not used as an independent degree of freedom for the 36-bit binary structure encoding.

[0029] In some embodiments, the first local control region, the second local control region, and the third local control region each contain independent units, and each independent unit is filled with empty silicon or not filled with silicon.

[0030] Regarding incomplete silicon arc-shaped opening rings: In some embodiments, the center of the incomplete silicon arc-shaped opening ring coincides with the center of the top-level tunable unit, and the outer periphery of the incomplete silicon arc-shaped opening ring is tangent to the VO2 stepped pixel block.

[0031] In some embodiments, the center of the non-complete silicon arc-shaped opening ring forms an opening on the top-level tunable unit, the opening direction has a deflection angle relative to the coordinate axis, and the opening is set towards the corner position of the top-level tunable unit.

[0032] In one specific embodiment, the deflection angle α is 240°.

[0033] In some embodiments, the body and outer contour of the non-complete silicon arc-shaped opening ring serve as a relatively fixed functional skeleton. The top silicon adjustable region includes a second local control region located in the lower half of the non-complete silicon arc-shaped opening ring. The second local control region contains 9 independent units, each of which is filled with empty silicon or not filled with silicon. The coding bit ci=1 indicates that the corresponding independent unit is filled with silicon, and ci=0 indicates that it is empty.

[0034] For ease of explanation, this scheme divides the top-level tunable unit into four corner regions along the center position. The corner of the top-level tunable unit relative to the opening direction of the non-complete silicon arc-shaped opening ring is defined as the first corner region. The three corners of the top-level tunable unit facing the non-opening direction of the non-complete silicon arc-shaped opening ring are defined as the second corner region, the third corner region, and the fourth corner region, respectively.

[0035] Regarding nested silicon square resonators: In some embodiments, the first silicon resonant unit of the nested silicon square resonator is located at the corner of the top-level tunable unit and is positioned toward the opening direction of the incomplete silicon arc-shaped open ring, and is symmetrically positioned with respect to the diagonal of the opening direction of the incomplete silicon arc-shaped open ring.

[0036] In some embodiments, the top silicon tunable region includes a third local control region located on the first silicon resonant unit, wherein the third local control region contains 21 independent units, each of which is filled with empty silicon or not filled with silicon, and the coding bit ci=1 indicates that the corresponding independent unit is filled with silicon, and ci=0 indicates that it is empty.

[0037] It should be noted that the first silicon resonator unit provides a square fixed frame, and different shapes of the first silicon resonator unit can be obtained by encoding the 36 independent units on it.

[0038] In a specific embodiment of this solution, such as Figure 1 As shown, the first silicon resonant unit includes an L-shaped silicon structure and a cubic silicon structure. The L-shaped silicon structure is formed orthogonally by two silicon edges of equal length, and the cubic silicon structure is formed orthogonally by four silicon edges of equal length. Furthermore, the cubic silicon structure is placed inside the L-shaped silicon structure, and the cubic and L-shaped silicon structures are spaced apart. The orthogonal angles of the L-shaped silicon structure and the diagonal of the cubic silicon structure are located on the diagonal of the opening direction of the incomplete silicon arc-shaped open ring.

[0039] In some embodiments, the two second silicon resonant units of the nested silicon square resonator are located at diagonal positions on the non-open side of the incomplete silicon arc-shaped opening ring, and the outer edge of each second silicon resonant unit is tangent to and attached to the surrounding VO2 stepped pixel block.

[0040] In some embodiments, two second silicon resonant units, located at diagonal positions on the non-open side of the non-incomplete silicon arc-shaped open ring, are mirrored along the diagonal of the opening direction of the non-incomplete silicon arc-shaped open ring.

[0041] In some embodiments, the second silicon resonant unit is spaced apart from the non-complete silicon arc-shaped open ring.

[0042] Similarly, in some embodiments, the top-level silicon adjustable region includes a first local control region located on each second silicon resonant unit. This first local control region contains six independent units, each filled with or unfilled silicon. The coding bit ci=1 indicates that the corresponding independent unit is filled with silicon, and ci=0 indicates it is empty. Furthermore, it should be noted that the first local control region located between two silicon resonant units is a mirror region; therefore, only one independent unit within the first local control region needs to be controlled.

[0043] It should be noted that the first silicon resonant unit provides a fixed framework, and different shapes of first silicon resonant units can be obtained by encoding the 36 independent units on it. In a specific embodiment of this solution, such as Figure 1 As shown, each second silicon resonator unit is designed as a stepped configuration. Each second silicon resonator unit is formed by a first silicon edge, a second silicon edge, and a third silicon edge connected end to end. The first silicon edge and the second silicon edge are right-angled sides that are perpendicular to each other. The third silicon edge intersects the first silicon edge and the second silicon edge perpendicularly. The third silicon edge has a multi-segment bent step structure, so that the second silicon resonator unit forms a stepped surface.

[0044] Furthermore, the stepped surface of the second silicon resonator is spaced apart from the non-complete silicon arc-shaped opening ring, and the first silicon edge and the second silicon edge of the second silicon resonator are attached to the VO2 stepped pixel block.

[0045] Regarding VO2 stepped pixel blocks: In this scheme, the VO2 stepped pixel blocks are discretely distributed along the edge of the top-level tunable unit and are symmetrically arranged with respect to the diagonal of the opening direction of the non-complete silicon arc-shaped opening ring.

[0046] In some embodiments, at least one VO2 stepped pixel block is independently constrained in each of the second, third, and fourth corner regions. That is, each of the second, third, and fourth corner regions contains at least one VO2 stepped pixel block, and the VO2 stepped pixel blocks in different corner regions are spaced apart.

[0047] In some embodiments, the VO2 stepped pixel blocks in each corner region are connected at the boundary of the corner region.

[0048] In some embodiments, three discretely distributed VO2 stepped pixel blocks are provided at the edge positions of the first and second sides that are connected on the top tunable unit, and one VO2 stepped pixel block is provided at the edge position of the third and fourth sides that are connected. The first and second sides are the sides where the non-open side is located relative to the non-complete silicon arc-shaped opening ring, and the third and fourth sides are the sides where the open side is located relative to the non-complete silicon arc-shaped opening ring.

[0049] Furthermore, the VO2 stepped pixel blocks on the first, second, third, and fourth sides are connected, but the corner position of the top tunable unit does not have VO2 stepped pixels, and the first corner area where the first silicon resonant unit is located also does not have VO2 stepped pixels.

[0050] As described above, the non-perfect silicon arc-shaped opening ring on the three-functional terahertz metasurface designed in this scheme has the center of the top-level tunable unit as the center, and its opening direction has a deflection angle relative to the coordinate axis, thus forming obvious anisotropic responses in the x and y linear polarization directions; the nested silicon square resonator participates in multimode resonance by adjusting the local inductance-capacitance distribution; the VO2 stepped pixel block is discretely distributed along the edge of the unit, and can change the overall loss characteristics and conductive bridging state under temperature control.

[0051] More specifically, the top-level tunable unit pattern design of this scheme has unique features: the incomplete silicon arc-shaped open ring mainly provides anisotropic resonant response, used to form the amplitude and phase difference between orthogonal polarization components; the nested silicon square resonator is used to introduce auxiliary LC resonance, which expands the working bandwidth after coupling with the arc-shaped open ring; the VO2 stepped pixel block is used to provide phase transition loss and conductive bridging channels, thereby realizing the switching between absorption mode and polarization control mode. Spacing is maintained between each structure, which on the one hand avoids direct short-circuiting between adjacent resonant units, and on the other hand allows for adjustment of coupling strength through gap capacitance, and provides tolerance for processing errors. Therefore, this top-level pattern is not a simple splicing, but rather achieves three functions—broadband absorption, linear-circular polarization conversion, and linear-linear polarization conversion—through the synergistic design of "silicon anisotropic resonant structure + auxiliary resonant structure + VO2 phase transition control unit".

[0052] In terms of the control mechanism, this scheme adopts a dual-tunable material system consisting of a silicon patterned layer (a non-complete silicon arc-shaped open ring + nested silicon square resonator) and a VO2 patterned layer (VO2 stepped pixel blocks). The VO2 patterned layer is a thermally induced phase-change tunable layer, whose phase state can switch between insulating and metallic states under temperature regulation, used to achieve the switching of the device's main functional mode. The silicon patterned layer is a photoexcitation tunable layer; under external illumination or optical pumping, its carrier concentration changes, thereby switching the equivalent conductivity between low, medium, and high conductivity states, used to achieve polarization conversion sub-mode switching. By synergistically controlling the equivalent conductivity of the silicon patterned layer and the phase state of the VO2 patterned layer, the same device can achieve three functional modes: broadband absorption, linear-circular polarization conversion, and linear-linear polarization conversion.

[0053] The switching of the VO2 phase state in the VO2 patterned layer determines whether the device is in absorption mode or polarization modulation mode. When the VO2 in the VO2 patterned layer is in an insulating state, the silicon conductivity of the incomplete silicon arc-shaped open ring and the nested silicon square resonator is adjusted to switch the three-functional terahertz metasurface to either a linear-circular polarization conversion mode or a linear-linear polarization conversion mode. When the VO2 in the VO2 patterned layer is in a metallic state, the three-functional terahertz metasurface is in a broadband absorption mode. It should be noted that this layered modulation mechanism maps the functional switching and polaron mode formation to different physical degrees of freedom, significantly reducing the coupling between different functions.

[0054] More specifically, when the VO2 of the VO2 stepped pixel block is in an insulating state, and the silicon conductivity of the incomplete silicon arc-shaped opening ring and the nested silicon square resonator is at a medium conductivity, the trifunctional terahertz metasurface is in a linear-to-circular polarization conversion mode; when the VO2 of the VO2 stepped pixel block is in an insulating state, and the silicon conductivity of the incomplete silicon arc-shaped opening ring and the nested silicon square resonator is at a high conductivity, the trifunctional terahertz metasurface is in a linear-to-linear polarization conversion mode; when the VO2 of the VO2 stepped pixel block is in a metallic state, and the silicon conductivity of the incomplete silicon arc-shaped opening ring and the nested silicon square resonator is at a low conductivity, the trifunctional terahertz metasurface is in a broadband absorption mode.

[0055] In some embodiments, a silicon conductivity of 1 S / m is defined as low conductivity for the incomplete silicon arc-shaped open ring and the nested silicon square resonator; a silicon conductivity of 1.3 × 10⁻⁶ is defined as low conductivity for the incomplete silicon arc-shaped open ring and the nested silicon square resonator. 5 At S / m, it has medium conductivity; the silicon conductivity of the non-perfect silicon arc-shaped open ring and the nested silicon square resonator is defined as 5 × 10⁻⁶. 6 At a conductivity of 20 S / m, it is considered high conductivity; a representative conductivity of 20 S / m is defined for a VO2 stepped pixel block, indicating it is in an insulating state; a representative conductivity of 2.0 × 10⁻⁶ S / m is defined for a VO2 stepped pixel block. 5 When S / m, it is in the metallic state.

[0056] The mechanisms and implementation of these three working modes are explained below: Broadband Absorption Mode: When the VO2 stepped pixel block is in a metallic state and the silicon of the incomplete silicon arc-shaped open ring and the nested silicon square resonator is in a low conductivity state, the VO2 region exhibits significant resistivity in the terahertz band. It can form a distributed current loop and resistive loss path together with the Au backplane. The silicon arc-shaped open ring and the nested square resonator form a multimode resonant cavity with the PTFE dielectric layer through geometric inductance and capacitance effects, thereby suppressing reflection and enhancing energy dissipation over a wide frequency band. Since the Au backplane almost completely shields transmission, the absorption rate can be expressed as A(f) = 1 - |S11| 2 At this point, the device's absorption rate remained above 90% in the 2.89–6.45 THz range, and approached or exceeded 99% at multiple frequency points.

[0057] Linear-to-circular polarization conversion mode: When the VO2 stepped pixel block is in an insulating state and the silicon conductivity is adjusted to a moderate level, VO2 essentially loses its conductive and high-loss properties. The top-layer response is mainly dominated by the anisotropic silicon arc-shaped open ring and square resonant structure. This structure can provide approximately equal amplitude reflection channels for two orthogonal linearly polarized components, with a phase difference of about a quarter wavelength, while maintaining high reflection efficiency. When |rxx|≈|ryx| and Δφ≈±90°, the reflected wave is formed by the superposition of two orthogonal linearly polarized vectors with equal amplitude and a phase difference of ±π / 2, thus achieving the conversion from linear polarization to left-handed or right-handed circular polarization. At this time, the operating bandwidth is 3.54–5.58 THz, and the absolute value of the ellipticity is close to 1 in this frequency band.

[0058] Linear-to-linear polarization conversion mode: When the VO2 stepped pixel block is in an insulating state and the silicon conductivity is further increased to a high conductivity level, the top silicon pattern approximately behaves as a highly conductive anisotropic grid, which can significantly enhance the cross-polarization scattering component. After rotating the intrinsic polarization coordinate system (u,v) relative to the (x,y) coordinate system by 45°, if |Ru|≈|Rv| and the phase difference Δφ between the two intrinsic modes ≈π within the working bandwidth, then the structure is macroscopically equivalent to a reflective half-wave plate, capable of rotating x-linear polarization to y-linear polarization. At this time, the linear-to-linear polarization conversion bandwidth is 2.16–5.32 THz, and the polarization conversion efficiency (PCR) is consistently greater than 0.9 in this frequency band.

[0059] In summary, when the VO2 patterned layer is in the metallic state, the VO2 patterned layer, Au backplane, and PTFE dielectric layer form a loss-type resonant cavity, and the device operates in a broadband absorption mode. When the VO2 patterned layer switches to the insulating state, the device exits the loss-dominated mode and enters the polarization modulation mode. At this time, by further adjusting the equivalent conductivity of the silicon patterned layer, the device can switch between the linear-circular polarization conversion mode and the linear-linear polarization conversion mode.

[0060] As mentioned earlier, this trifunctional terahertz metasurface achieves broadband absorption, linear-circular polarization conversion, and linear-linear polarization conversion in the ranges of 2.89–6.45 THz, 3.54–5.58 THz, and 2.16–5.32 THz, respectively.

[0061] Correspondingly, the state-function mapping relationship of this three-functional terahertz metasurface is shown in Table 1 below; Table 1. State-Function Mapping Relationship of Functional Terahertz Metasurfaces

[0062] Example 2 Specific designs such as Figure 1 The three-functional terahertz metasurface shown below has structural parameters as shown in Table 2. Table 2. Structural parameters of the three-functional terahertz metasurface

[0063] Simulation tests were conducted on the trifunctional terahertz metasurface, and the results are as follows: Figures 2-9 As shown, This embodiment uses CST Microwave Studio for frequency domain full-wave simulation. Periodic boundary conditions are applied in the x and y directions to create an equivalent infinite periodic array, and an open boundary is set in the z direction. A normal incident plane wave is excited at the upper port. The absorption rate, linear-circular polarization conversion index, and linear-linear polarization conversion rate are calculated using S-parameters under different material state combinations. The results are as follows: Figures 2-9 As shown.

[0064] Figure 2 The image shows the absorption spectrum of this trifunctional terahertz metasurface in broadband absorption mode, indicating that the absorption rate is higher than 90% in the VO2 metallic state and silicon low conductivity state within 2.89–6.45 THz. Figure 3 The impedance matching analysis diagram of the three-functional terahertz metasurface in broadband absorption mode shows that Re(Zr) is close to 1 and Im(Zr) is close to 0 in the main absorption band. Figure 4 The curves show the co-polarization and cross-polarization reflection coefficients in the linear-circular polarization conversion mode, indicating that the two components are approximately equal in amplitude within the range of 3.54–5.58 THz. Figure 5 This is a phase difference curve in linear-circular polarization conversion mode, indicating that the phase difference is close to ±90°; Figure 6 This is an ellipticity curve in the linear-circular polarization conversion mode, indicating that the reflected wave is close to circular polarization. Figure 7 The PCR curve in linear-to-linear polarization conversion mode shows that the PCR value is greater than 0.9 within the range of 2.16–5.32 THz. Figure 8This is a diagram of u and phase difference in linear-to-linear polarization conversion mode. Figure 8 This indicates that the intrinsic axis reflection components of u / v have equal amplitude and a phase difference close to π; Figure 9 This is a diagram of co-polarized and cross-polarized reflection components in the linear-to-linear polarization conversion mode. Figure 9 This indicates that the cross-polarization component is dominant and the co-polarization component is suppressed, which is consistent with the PCR results.

[0065] It should be noted that the three-functional terahertz supersurface, as described above... Figure 10 As shown, compared with common 4- to 7-layer multifunctional metasurfaces, this invention achieves broadband absorption, linear-to-circular polarization conversion, and linear-to-linear polarization conversion simultaneously using only a three-layer structure, exhibiting higher structural compactness and better overall functional integrity.

[0066] Furthermore, under the linear-to-linear polarization conversion mode where VO2 is in an insulating state, silicon is in a highly conductive state, and x-polarized waves are incident normally, a representative frequency point f=3.34 THz within the operating frequency band was selected. The electric and magnetic field distributions of the top layer of the unit cell were extracted in the CST, resulting in... Figure 11 The figure illustrates that the strong electric and magnetic fields are mainly concentrated in the silicon arc-shaped open ring and its two ends. The surface current propagates along the arc-shaped path and undergoes significant deflection in different regions of the unit cell. Under the influence of geometric anisotropy, it couples to the two intrinsic polarization channels u and v respectively. The phase accumulation of the two channels forms a difference close to π, which ultimately leads to the coherent cancellation of the copolarization component and the coherent enhancement of the cross-polarization component, thereby achieving broadband high-efficiency linear-to-linear polarization conversion.

[0067] Example 3 This solution additionally provides a free topology reverse design method for the three-functional terahertz metasurface shown in Example 1. This solution uses explicit symmetry constraints to confine the top silicon tunable region of the top tunable hybrid pattern layer within a 45° axisymmetric space and compresses it into a binary code of 36 independent design units to reduce the design space size and ensure structural manufacturability. Furthermore, a free topology reverse design model is used to extract key frequency bands, cross-channel coupling relationships, and local spectral dynamic features in the target multi-channel spectrum, achieving efficient mapping from the target response to the free topology pattern.

[0068] Specifically, such as Figure 13 As shown, this scheme provides a free topology reverse design method, including the following steps: A three-channel target response sequence is obtained, including the target absorption spectrum, linear-to-linear polarization conversion rate, and linear-to-circular polarization conversion rate. The three-channel target response sequence is obtained after normalization and sorting. The three-channel target response sequence is input into a pre-trained free-topology inverse design model, which outputs a free-topology designed three-functional terahertz metasurface. The free-topology inverse design model includes an FCSA module, an LSTM encoder, a Transformer encoder, a feature fusion module, a Transformer decoder, and a symmetric geometry decoder. The three-channel target is transformed into enhanced features by the FCSA module. The enhanced features are input into the LSTM encoder and the Transformer encoder to obtain two encoded features. The two encoded features are fused by the feature fusion module to obtain fused features. The fused features are input into the Transformer decoder to output a 36-bit binary structure code. The 36-bit binary structure code is input into the symmetric geometry decoder to output a pattern of the top-level tunable unit, including the top-level free-design region. The top silicon tunable region of the trifunctional terahertz metasurface includes a second local control region located in the lower half of the non-complete silicon arc-shaped open ring, a third local control region located on the first silicon resonant unit, and a first local control region located on the second silicon resonant unit. The two local control regions on the second silicon resonant unit constitute a mirror region. The first local control region, the second local control region, and the third local control region are confined within a 45° axisymmetric space.

[0069] Furthermore, the first local control region includes 6 independent units, and the first local control region corresponding to the mirror region does not introduce additional degrees of freedom; The second local control region is located in the lower left part of the non-intelligent silicon arc-shaped open ring. The second local control region includes 9 independent units, and uniquely recovers the units in the upper left part of the non-intelligent silicon arc-shaped open ring with a 45° axial symmetry relationship. That is, the second local control region occupies half of the non-intelligent silicon arc-shaped open ring and can fill the entire area of ​​the non-intelligent silicon arc-shaped open ring through a 45° axial symmetry relationship. The third local control region is located in the lower left part of the first silicon resonator unit, that is, the first silicon resonator unit is divided along the diagonal. The third local control region includes 21 independent units and uniquely restores the other half of the first silicon resonator unit with a 45° axial symmetry relationship.

[0070] like Figure 12 As shown, Figure 12 This is a schematic diagram of the top-level free design region of the three-functional terahertz metasurface, where R1 corresponds to the first local control region, R1' corresponds to the mirror region of the first local control region, R3 corresponds to the third local control region, and R2 corresponds to the second local control region.

[0071] For training the free topology reverse design model, a three-function terahertz metasurface with a known three-channel target is used as the training set. The top free design region of each three-function terahertz metasurface is converted into a 36-bit binary structure code and then input into the free topology reverse design model for training.

[0072] Furthermore, since pure random sampling simulation in 36-bit binary space is costly, this scheme adopts a hybrid strategy of Structured Scan Dataset (SSD) and Free Topology Augmentation Dataset (FAD). SSD forms 2880 sets of three-functional terahertz metasurfaces covering typical opening and filling patterns by performing bounded discrete scanning on low-dimensional coded blocks with clear physical meaning; FAD, while maintaining a fixed skeleton and 45° axisymmetric constraints, supplements the construction of three-functional terahertz metasurfaces with atypical patterns such as diagonals, checkerboard, and quasi-random filling, totaling 120 sets of samples. Finally, these are merged to obtain 3000 sets of hybrid data, which are then divided into training and test sets in an 8:2 ratio.

[0073] During the training phase, the spectral-structure inverse design problem is modeled as a multi-label binary classification task with 36 independent units in binary states. The binary cross-entropy loss function with L2 weight decay term is preferred, and the AdamW optimizer is used for training. In one embodiment, the batch size is 32, the maximum number of training epochs is 500, and early stopping is triggered when the test loss does not improve for 10 consecutive epochs. The model output is converted into a predictive code after Sigmoid activation and thresholding, and then restored to the complete top-level pattern by a symmetric geometric decoder. The spectrum of the generated pattern under three working states can be post-verified by full-wave electromagnetic simulation.

[0074] Regarding the design process of this free topology reverse design model: The normalized three-channel target response sequence is input into the FCSA module. The FCSA module then performs feature recalibration along the frequency and channel dimensions to obtain enhanced features. The frequency dimension attention is used to highlight the key resonant segments that are more sensitive to the discrimination of the target structure, while the channel dimension attention is used to assign different weights to the absorption response, linear-circular polarization conversion response, and linear-linear polarization conversion response. This strengthens the effective information that is highly correlated with the structure encoding and improves the feature expression ability of the multi-channel target response in the inverse mapping process, thereby enhancing the discriminability of the key resonant frequency band and key response channel.

[0075] In some embodiments, adaptive recalibration refers to the network automatically learning the importance weights of different frequency points and different response channels based on the characteristic distribution of the input target spectrum itself, and applying these weights to the original input features, thereby highlighting the key resonant segments and key functional channels that are more relevant to the structure encoding prediction. This process does not require manual pre-setting of key frequency bands, but is automatically optimized by backpropagation of the loss function during the training process.

[0076] Furthermore, the enhanced features are input into parallel LSTM and Transformer encoders to obtain their respective encoded features. The LSTM encoder is used to extract local continuity, resonance peaks, and band-edge abrupt changes as the spectrum unfolds along the frequency axis, while the Transformer encoder is used to capture wideband global dependencies and cross-channel coupling features. That is, the Transformer encoder uses a self-attention mechanism to establish long-range dependencies between different frequency points to model wideband global spectral changes and cross-channel coupling features, while the LSTM encoder is used to mine the contextual continuity and local dynamic changes of the spectral sequence as it unfolds along the frequency axis, providing a stronger representation of fine-grained features such as resonance peaks, band-edge abrupt changes, and local fluctuations. In this way, the two-way encoded feature output can take into account both global functional constraints and local spectral details.

[0077] Furthermore, the two enhanced features are input into the feature fusion module to obtain fused features, which are then input into the Transformer decoder to output a 36-bit binary structure code. The Transformer decoder models the potential correlation between different coding bits bit by bit, so that the prediction of each design unit not only depends on the input target response, but is also affected by the combined constraints of the coding combination, thereby improving the generation consistency, structural interpretability and inversion accuracy of complex free topologies.

[0078] Furthermore, the symmetric geometry decoder writes the states of the independent units corresponding to the lower left local grid region, the arc-shaped control region, and the lower right grid region in the 36-bit binary structure encoding, and then writes the states of the independent units in the mirror region using the mirror symmetry relationship about the u-axis, and restores the pattern of each top-level tunable unit based on the fixed function skeleton.

[0079] Regarding the 36-bit binary structure code output by this scheme, based on a fixed functional skeleton, the top-level free design region is represented as a 36-bit binary structure code satisfying a 45° axisymmetric constraint. A symmetry-sensing geometric decoder then restores the 36-bit binary structure code to a complete top-level free topological pattern. This method establishes a clear mapping relationship between "target multi-channel electromagnetic response—low-dimensional structure code—complete manufacturable pattern," which not only compresses the design space but also ensures that the output structure has clear physical meaning, fabrication feasibility, and verifiability in subsequent full-wave simulations.

[0080] Regarding the prediction quality of the free-topology inverse design model of this scheme, tests on the test set show that the bit-level prediction accuracy of the free-topology inverse design model reaches 0.9464, with an overall MSE of 0.00165. Compared with the reduced model, removing the FCSA module decreases the accuracy to 0.9318 while increasing the overall MSE to 0.00220; removing only the Transformer encoding branch results in an accuracy of 0.9246 and an overall MSE of 0.00253; removing only the LSTM encoding branch results in an accuracy of 0.9291 and an overall MSE of 0.00236. This demonstrates that the frequency-channel joint attention mechanism and the LSTM-Transformer dual-branch structure significantly contribute to the inverse generation of complex and multifunctional target spectra. In random sample validation, the overall MSE distribution is 2.07 × 10⁻⁶. -4 Up to 9.324×10 -3 Between these values, the average MSE is approximately 0.003878, indicating that the method has good stability and generalization ability.

[0081] Example 4 This embodiment also provides an electronic device, see reference. Figure 14 It includes a memory 402 and a processor 401, the memory 402 storing a computer program and the processor 401 being configured to run the computer program to perform the steps in any of the embodiments of the free topology reverse design method described above.

[0082] Specifically, the processor 401 may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0083] The memory 402 may include a large-capacity memory 402 for data or instructions. The memory 402 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 401.

[0084] The processor 401 reads and executes computer program instructions stored in the memory 402 to implement any of the methods for constructing a cross-condition health index prediction model for wind turbine gearbox components or a cross-condition health index prediction method for wind turbine gearbox components in the above embodiments.

[0085] Optionally, the electronic device may further include a transmission device 403 and an input / output device 404, wherein the transmission device 403 is connected to the processor 401 and the input / output device 404 is connected to the processor 401.

[0086] The transmission device 403 can be used to receive or send data via a network. Specific examples of the network described above may include wired or wireless networks provided by the communication provider of the electronic device. In one example, the transmission device includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 403 may be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0087] Input / output device 404 is used to input or output information. In this embodiment, the input information may be a three-channel target, etc., and the output information may be a three-functional terahertz metasurface, etc.

[0088] Optionally, in this embodiment, the processor 401 can be configured to perform the following steps via a computer program: A three-channel target response sequence is obtained, including the target absorption spectrum, linear-to-linear polarization conversion rate, and linear-to-circular polarization conversion rate. The three-channel target response sequence is obtained after normalization and sorting. The three-channel target response sequence is input into a pre-trained free-topology inverse design model, which outputs a free-topology designed three-functional terahertz metasurface. The free-topology inverse design model includes an FCSA module, an LSTM encoder, a Transformer encoder, a feature fusion module, a Transformer decoder, and a symmetric geometry decoder. The three-channel target is transformed into enhanced features by the FCSA module. The enhanced features are input into the LSTM encoder and the Transformer encoder to obtain two encoded features. The two encoded features are fused by the feature fusion module to obtain fused features. The fused features are input into the Transformer decoder to output a 36-bit binary structure code. The 36-bit binary structure code is input into the symmetric geometry decoder to output a pattern of the top-level tunable unit, including the top-level free-design region. The top-level free design region of the trifunctional terahertz metasurface includes the lower left local grid region where the second silicon resonator unit is located in the lower left, the mirror region where the silicon resonator unit is located in the upper right, the arc-shaped control region where the incomplete silicon arc-shaped opening ring is located, and the lower right grid region where the first silicon resonator unit is located. The lower left local grid region and the mirror region each contain 6 independent units, the arc-shaped control region contains 9 independent units, and the lower right grid region contains 21 independent units. The binary states of the independent units in the top-level free design region are sequentially written in the 36-bit binary structure code. When the state of each independent unit is 1, it indicates that the current independent unit is filled with silicon.

[0089] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated here.

[0090] Generally, various embodiments can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects of the invention can be implemented in hardware, while others can be implemented by firmware or software executed by a controller, microprocessor, or other computing device, but the invention is not limited thereto. Although various aspects of the invention may be shown and described as block diagrams, flowcharts, or using some other graphical representation, it should be understood that, by way of non-limiting example, these blocks, apparatuses, systems, techniques, or methods described herein can be implemented in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0091] Embodiments of the present invention can be implemented by computer software, which may be executable by a data processor of a mobile device, such as a processor entity, or by hardware, or by a combination of software and hardware. Computer software or programs (also referred to as program products), including software routines, applets, and / or macros, can be stored in any device-readable data storage medium, and they include program instructions for performing specific tasks. A computer program product may include one or more computer-executable components configured to perform embodiments when the program is run. One or more computer-executable components may be at least one piece of software code or a portion thereof. Additionally, it should be noted that any block in the logical flow of the figures may represent a program step, or interconnected logical circuitry, blocks and functions, or a combination of program steps and logical circuitry, blocks and functions. The software may be stored on physical media such as memory chips or blocks of storage implemented within a processor, magnetic media such as hard disks or floppy disks, and optical media such as, for example, DVDs and their data variants, CDs, etc. The physical medium is a non-transient medium.

[0092] Those skilled in the art should understand that the technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0093] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A trifunctional terahertz metasurface, characterized in that, include: An Au metal backsheet layer, a PTFE dielectric layer, and a top tunable hybrid pattern layer are arranged sequentially from bottom to top along the z-direction. The top tunable hybrid pattern layer includes at least one top tunable unit, which is composed of a fixed functional skeleton and a top silicon tunable region. The fixed functional skeleton includes a non-complete silicon arc-shaped opening ring, a nested silicon square resonator, and multiple VO2 pixel blocks distributed at the edge of the top tunable unit. The nested silicon square resonator includes a first silicon resonator located in the opening direction of the non-complete silicon arc-shaped opening ring, and two second silicon resonator units located diagonally opposite each other in the non-opening direction of the non-complete silicon arc-shaped opening ring. The top silicon tunable region includes a second local control region located in the lower half of the non-complete silicon arc-shaped opening ring, a third local control region located on the first silicon resonator unit, and a first local control region located on the second silicon resonator unit. The first local control region includes 6 independent units, the second local control region includes 9 independent units, and the unit that uniquely restores the upper right part of the incomplete silicon arc-shaped open ring with a 45° axial symmetry relationship, and the third local control region includes 21 independent units, and the unit that uniquely restores the other half of the first silicon resonant unit with a 45° axial symmetry relationship.

2. The trifunctional terahertz metasurface according to claim 1, characterized in that, The center of the incomplete silicon arc-shaped opening ring coincides with the center of the top-level tunable unit, and the outer periphery of the incomplete silicon arc-shaped opening ring is tangent to the VO2 pixel block.

3. The trifunctional terahertz metasurface according to claim 1, characterized in that, The first local control region, the second local control region, and the third local control region each contain independent units, and each independent unit may or may not be filled with silicon.

4. The trifunctional terahertz metasurface according to claim 1, characterized in that, The first, second, and third local control regions are confined within a 45° axisymmetric space.

5. The trifunctional terahertz metasurface according to claim 1, characterized in that, Two second silicon resonator units, located at diagonal positions on the non-open side of the incomplete silicon arc-shaped open ring, are mirror images of each other along the diagonal of the opening direction of the incomplete silicon arc-shaped open ring, with the second silicon resonator units spaced apart from the incomplete silicon arc-shaped open ring.

6. The trifunctional terahertz metasurface according to claim 1, characterized in that, The VO2 pixel blocks are discretely distributed along the edge of the top-level tunable unit and are symmetrically arranged with respect to the diagonal of the opening direction of the non-complete silicon arc-shaped opening ring.

7. The trifunctional terahertz metasurface according to claim 1, characterized in that, When the VO2 of the VO2 pixel block is in an insulating state, and the silicon conductivity of the incomplete silicon arc-shaped opening ring and the nested silicon square resonator is at a medium conductivity, the trifunctional terahertz metasurface is in a linear-to-circular polarization conversion mode. When the VO2 of the VO2 pixel block is in an insulating state, and the silicon conductivity of the incomplete silicon arc-shaped opening ring and the nested silicon square resonator is at a high conductivity, the trifunctional terahertz metasurface is in a linear-to-linear polarization conversion mode. When the VO2 of the VO2 pixel block is in a metallic state, and the silicon conductivity of the incomplete silicon arc-shaped opening ring and the nested silicon square resonator is at a low conductivity, the trifunctional terahertz metasurface is in a broadband absorption mode.

8. The trifunctional terahertz metasurface according to claim 1, characterized in that, The trifunctional terahertz metasurface achieved broadband absorption, linear-to-circular polarization conversion, and linear-to-linear polarization conversion in the ranges of 2.89–6.45 THz, 3.54–5.58 THz, and 2.16–5.32 THz, respectively.

9. A free topology reverse design method, characterized in that, The method for designing a trifunctional terahertz metasurface as described in any one of claims 1 to 8 includes the following steps: A three-channel target response sequence is obtained, including the target absorption spectrum, linear-to-linear polarization conversion rate, and linear-to-circular polarization conversion rate. The three-channel target response sequence is obtained after normalization and sorting. The three-channel target response sequence is input into a pre-trained free-topology inverse design model, which outputs a free-topology designed three-functional terahertz metasurface. The free-topology inverse design model includes an FCSA module, an LSTM encoder, a Transformer encoder, a feature fusion module, a Transformer decoder, and a symmetric geometry decoder. The three-channel target response sequence is converted into enhanced features by the FCSA module. The enhanced features are input into the LSTM encoder and the Transformer encoder to obtain two encoded features. The two encoded features are fused by the feature fusion module to obtain fused features. The fused features are input into the Transformer decoder to output a 36-bit binary structure code. The 36-bit binary structure code is input into the symmetric geometry decoder to output a pattern of the top-level tunable unit, including the top-level free-design region.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the free topology reverse design method of claim 9.

Images