Switchable polarization rotation of visible light using a plasmonic metasurface

A metasurface comprising an array of silver nanorods supported by a thin ﬁlm of the phase change material vanadium dioxide is used to rotate the primary polarization axis of visible light at a pre-determined wavelength. The dimensions of the rods were selected such that, across the two phases of vanadium dioxide, the two lateral localized plasmon resonances (in the plane of the metasurface) occur at the same wavelength. Illumination with linearly polarized light at 45 ◦ to the principal axes of the rod metasurface enables excitation of both of these resonances. Modulating the phase of the underlying substrate, we show that it is possible to reversibly switch which axis of the metasurface is resonant at the operating wavelength. Analysis of the resulting Stokes parameters indicates that the orientation of the principal linear polarization axis of the reﬂected signal is rotated by 90 ◦ around these wavelengths. Dynamic metasurfaces such as these have the potential to form the basis of an ultra-compact, low-energy multiplexer or router for an optical signal. © 2016 Author(s). All article content, except where other-wise noted, is licensed under a Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/). [http://dx.doi.org/10.1063/1.4968840]


I. INTRODUCTION
Plasmonic metasurfaces that display dynamically modulated resonances have been demonstrated over a range of frequencies from the THz 1 to the visible. 2 This can be achieved by altering the material properties of the resonant metal, manipulating the dielectric environment, elastic/inelastic deformation of the device, or varying the method of illumination. Plasmonic modulation has already been demonstrated using a variety of these methods, such as photoexcitation of free carriers, 1 mechanical deformation, 3,4 incorporation within a matrix of a liquid crystal, 5 and phase change materials. 2,[6][7][8][9][10] The development of devices facilitated by these dynamic, reversibly tunable structures is one of the next steps in the translation of plasmonic structures into functional devices.
The localized surface plasmon (LSP) resonance of a subwavelength metallic structure is directly related to the composition and properties of its constituent materials, dielectric environment, and geometry. 11,12 For example, a red-shifting of the plasmon modes of a silver nanocube on an ITO-glass substrate was observed with increasing refractive index of the superstrate. 13 A rod on a substrate displays two LSP modes with their dipole moments parallel to the substrate, each aligned to one of the two axes of the rod, as well as a third whose dipole moment a Electronic mail: s.earl@student.unimelb.edu.au 2378-0967/2016/2(1)/016103/8 2, 016103-1 © Author(s) 2016 extends out of the plane of the substrate. 14 By varying the angle of incidence and polarization state of the incident light, a superposition of these different modes can be excited. 15 In the presence of a substrate, the resulting asymmetry in the dielectric environment of the rod leads to a strong separation in resonance frequency between the in-plane and out-of-plane LSP modes. The frequencies at which these modes occur are dependent on the value of the dielectric permittivity of the substrate/superstrate and the dimensions of the rod. 16 Thus, altering the optical properties of the substrate modifies the frequencies of the LSP resonances of the rod. This effect is the mechanism by which phase-change materials such as vanadium dioxide (VO 2 ) modulate plasmon resonances. VO 2 is considered an insulator at room temperature, but, if heated to temperatures higher than 68 • C, it becomes conductive, 17 its crystalline structure transitions from monoclinic to rutile, 18,19 and the real part of its refractive index decreases. This phase is commonly referred to as the metallic phase. The phase transition can be triggered not only thermally but also by application of mechanical stress, 20 applied current/voltage, 21,22 or optically. 23 Significantly, when triggered optically by femtosecond laser pulses, VO 2 has been shown to transition on a sub-picosecond time scale, indicating potential for ultrafast plasmonic modulation devices. [24][25][26][27][28] The insulator-to-metal transition (IMT) is percolative, 29,30 while the metal-to-insulator transition (MIT) process appears completely random. This process of individual crystal grains undergoing a transition leads to both phases occurring simultaneously in the same film, 31 enabling access to a range of intermediate effective material properties between the two extremes represented by the insulating and metallic phases. VO 2 has previously been used to modulate the extinction of a plasmonic waveguide 32 and nano-gap junctions, 33 as well as the spontaneous emission from a layer of quantum emitters via its manipulation of the local density of optical states. 34 Here we fabricate an array of silver nanorods on a thin film of VO 2, and harness the dielectric shift caused by the VO 2 phase-change to manipulate the resulting polarization axis of a reflected signal at a specified wavelength. Our results show that at the central operating wavelength of the metasurface, the principal polarization axis of the reflected light is rotated by 90 • when the phase of the underlying substrate is modulated. This is facilitated by the change in the real part (n) of the refractive index of VO 2 when transitioned from the insulating (ambient temperature) phase to its metallic (>68 • C) phase. Figure 1(a) presents a schematic showing this far-field behavior, highlighting how it is possible to rotate the principal polarization axis by modulating the phase of the substrate.

II. METASURFACE DESIGN
A nanorod on a substrate of the insulating phase of VO 2 will display LSP resonances that are redshifted relative to an identical rod on VO 2 in its metallic phase. This substrate-dependent dielectric shift therefore permits the dimensions of the short axis of a nanorod on a substrate of insulating VO 2 and those of the long axis on metallic VO 2 to be chosen such that the orthogonal LSP resonances are spectrally aligned.
This overlap of perpendicular LSP modes enables manipulation of the principal polarization axis of the reflected light via modulation of the substrate phase. It is well known that the orientation of the resonant axis of a rod supporting an LSP induces a polarization orientation in the reflected signal parallel to its electric dipole moment. Changing the VO 2 phase will therefore rotate the primary axis of the reflected signal by 90 • , creating a reversible, compact, and easily integrated method of changing the orientation of the polarization axis in a thin film device.

A. Design and simulation
To design the plasmonic metasurface, we found numerical solutions to Maxwell's equations via the finite element method (FEM) implemented in the RF module of COMSOL Multiphysics. 35 A square array of periodicity 250 nm of Ag nanorods on a silicon substrate supporting a thin film of VO 2 was simulated. The height of the modelled domain was 1.5 µm. Excitation was provided by a port boundary condition on the upper half-space (n = 1, air side), and periodic boundary conditions were used on the transverse boundaries. Since only normal incidence excitation was used, and since only the zeroth-order reflection was present in the wavelength range of interest, the port boundary was sufficient to terminate any reflections. The lower boundary (in the substrate) was simulated via a scattering boundary to absorb any transmitted signal, preventing any reflections from this lower surface. The optical properties of silicon and silver were taken from Palik, 36 while the optical constants of VO 2 were taken from Earl et al. 2 A parametric sweep was performed to ascertain a relationship between rod length, VO 2 phase, and plasmon resonance wavelength. As we have shown previously, 37 the thickness of the underlying VO 2 also affects the peak plasmon resonance wavelength, and so its thickness was set to a constant value of 110 nm to maximize the amplitude of the reflected signal around the wavelength range of interest. The outcome of these simulations, shown in Figure 2(a), found that a square-periodic array of 50 nm (140 nm) long nanorods on insulating (metallic) VO 2 supports an LSP centered on 750 nm.

III. EXPERIMENTAL RESULTS
Electron beam lithography (EBL) was used to fabricate silver nanorod arrays on thin films of VO 2 . Details of the deposition of the VO 2 thin films and subsequent EBL processes are given in the supplementary material. The surface structure of the VO 2 displayed a small level of porosity and an rms roughness of around 7 nm (measured using atomic force microscopy and incorporated into the ellipsometry fitting algorithm). Figure 1(b) shows an SEM of the metasurface, with color-coded arrows indicating the relevant polarization axes.
Normal incidence spectroscopy from the air side of the metasurface was performed as a function of both input and output angle of polarization. Figure S1 of the supplementary material presents a schematic and details of the experimental setup. Reflectance spectroscopy was used due to the supporting Si substrate being opaque at the wavelengths of interest. Figure 2(b) shows the normalised reflectance (R norm = R array /R substrate ) from the rod arrays for an input polarization of 45 • and an analyzer (the second linear polarizer immediately preceding the spectrometer) orientation parallel to each of the rod axes in turn. The normally incident, 45 • linearly polarized light excites both the dipolar LSP modes that lie in the plane of the substrate and facilitates simpler analysis due to its known polarization state. From the large shift in the peak reflected wavelength (as the analyzer, initially parallel to the rods' long axis, is rotated 90 • ) it is clear that the reflected signal is significantly linearly polarized. For the complete set of measurements refer to the supplementary material, Figure S2. For insulating VO 2 ("Cold VO 2 ," T ∼ 22 • C), the rod array resonances are measured at 690 nm (short axis) and 841 nm (long axis). For metallic VO 2 ("Hot VO 2 ," T > 80 • C) the rod arrays displayed resonances at 632 nm (short axis) and 679 nm (long axis). Hence, for 45 • linearly polarized (or unpolarized) incident light near wavelengths of 680 nm the primary polarization axis of the reflected light can be rotated by 90 • by switching the phase of the VO 2 substrate.

IV. DISCUSSION
Experimental measurements displayed a blue-shift of around 50 nm from the design wavelength, and the optimum rod dimensions for the LSP resonances also deviated from that predicted by COMSOL. While there are a number of explanations for this, from inaccuracies in evaporated film thicknesses, the assumed optical properties of silver (bulk properties were used), 36 the native oxide present on the silver, or inhomogeneities in rod dimensions, variations in the optical properties of the VO 2 film were thought to be the primary cause. VO 2 films are known to demonstrate a variation in optical properties dependent on deposition method and stoichiometry. [37][38][39] Ellipsometry was performed on the specific VO 2 films used here and numerical simulations using those results were performed; results of both are presented in the supplementary material ( Figures S3 and S4). A high level of agreement between the experimental and numerical results is apparent, suggesting that the optical properties of each individual film are the primary cause of the noted disparity. Despite this blue-shift, the overlap of the orthogonal LSP modes centred on 680 nm, however, enables the device to perform as desired.
While analysis of the normalised reflectance facilitates understanding of the spectral location of the plasmon resonances, a realistic device requires significant modulation of the absolute magnitude of the reflected signal to operate. The absolute signal from our device is compared to that from the bare substrate in the supplementary material, Figure S5. Analysis of the absolute signal permits identification of the operating range of the device. For these metasurfaces, the maximum functional range spans from 641 nm to 734 nm, with a central operating wavelength of 679 nm.

V. POLARIZATION ANALYSIS
Further insight into the polarization behavior of the metasurface can be obtained by characterizing the Stokes parameters of the reflected signal. The Stokes parameters (S 0 , S 1 , S 2 , and S 3 ) give detailed information about the polarization state of electromagnetic radiation. 40,41 Specifically, the S 1 value represents the degree of linear polarization parallel to the rod axes (the Stokes parameters are discussed briefly in the supplementary material). In relation to these measurements, an S 1 value of +1 indicates 100% linearly polarized wave with its polarization oriented parallel to the short axis of the rods while a value of −1 indicates 100% linearly polarized with orthogonal polarization. It is worth highlighting here that the presented values, S 1 , S 2 , and S 3 , are normalised to total intensity, S 0 .
The Stokes parameters were measured using normally incident light that was linearly polarized at 45 • to the primary nanorod axes in the manner of Collett et al. 41 The resulting intensity measurements are dependent on different polarized components within the incident beam and therefore permit calculation of the Stokes parameters. Figure 3 shows the S 1 value as a function of wavelength for the rod array with an incident linear polarization (a) parallel, (b) at 45 • , and (c) orthogonal to the short axis of the rods. The relative orientation of the arrows inset in the figure highlights the implication of the sign of the S 1 parameter. Below 500 nm and above 850 nm the experimental data are unreliable due to the limited operating wavelength ranges of the polarizing elements, beam splitters, and other optical elements. While The arrows inset on the plots show how the sign of the S1 parameter relates to the orientation of the polarization axis of the reflected signal, with a positive S1 value indicating "vertical" polarized light and a negative value indicating a "horizontal" polarization. input linearly polarized light at 0 • , 45 • , and 90 • relative to the short axis of the rods are presented in the supplementary material, Figure S6.
For incident light polarized at 45 • to the rod axes, in the vicinity of the short axis resonances (683 nm insulating and 630 nm metallic VO 2 ), S 1 increases to 0.1 before decreasing to −0.3 around the long axis resonances (850 nm insulating and 680 nm metallic VO 2 ). At the central operating wavelength of the array, a total change in linear polarization can then be found by calculating the difference in these two values at that wavelength, which for 679 nm is a change of over 13%. Significantly, the sign of the S 1 vector changes at this wavelength, indicating that the reflected signal goes from being partially vertically polarized (parallel to the short axis of the rods) to partially horizontally polarized (parallel to the long axis of the rods).
The largest difference between the two S 1 vectors occurs at 733 nm and is a much larger 29.6%. It is worth noting that at wavelengths longer than 750 nm there is a significant degradation in the response of the beam splitter (Thorlabs BS013), 42 and so the data used to calculate the Stokes parameters at these wavelengths must be regarded as questionable. They do, however, demonstrate that the array is indeed affecting the polarization of the reflected light, confirming that the LSP resonances are the driving mechanism behind the changes in reflected polarization observed.
A useful metric for assessing the performance of the polarization-modulating metasurface is the percentage of the reflected intensity in the desired polarization state. The incident light can be considered completely polarized at 45 • to the nanorod axes. Figure S6(e) of the supplementary material gives the S 2 parameter for the incident 45 • linearly polarized light. At wavelengths shorter than those of the LSP resonances of the nanorods, the S 2 parameter displays a value of +1, which indicates that 100% of the reflected light displays a +45 • linear polarization state. Since the metasurface displays no resonance in this wavelength range (refer to supplementary material Figure S2) it will reflect the incident polarization state essentially unperturbed. Hence, we can assume that the incident light across the entire visible spectrum exhibits an identical polarization.
A silver mirror (Thorlabs PF10-03-P01) was used in place of the VO 2 substrate to characterize the source, thus permitting calculation of the amount of incident light transformed into the desired polarization state. This figure of merit represents the absolute efficiency of the metasurface. At the central operating wavelength of 679 nm, the bare VO 2 film in its insulating (metallic) phase reflected 27.7% (11.8%) of the incident light. The normalised reflectance of the silver nanorod arrays on the VO 2 was 1.532 (insulating) and 1.913 (metallic). The metasurface therefore reflected 42.4% and 22.6% of the incident signal in the insulating and metallic phases, respectively.
The normalised Stokes parameters (S 1 ) at this wavelength were measured to be +0.08 (insulating) and −0.04 (metallic), which, translated into a percentage of the incident signal, indicates that 3.4% and 0.9% of the incident light were directed into the desired linear polarization states. The remainder of the incident light was either absorbed by the opaque substrate, the VO 2 thin film, and the nanorod array, or reflected into another polarization state.
The relatively low efficiency of this device can be improved in a number of ways. Here, we have focussed on a proof-of-concept device operating in the visible spectrum. The larger contrast between the optical phases of the enabling VO 2 in the near IR, however, suggests a superior device may be possible. An IR polarization rotating device is the subject of active research, although it represents a significantly different device to the one proposed here.
A challenge in the design of this device is the amplitude and width of the reflected signal from the two rod axes. Due to the lower absorption of silver at longer wavelengths, the two lateral resonances of the rods will always have different amplitudes. In addition, the higher absorption of the insulating phase of VO 2 will produce broader resonances compared to the metallic phase. There are several methods that could be used to ameliorate these issues. The incident polarization angle onto the array could be optimized to favor the weaker shorter-axis resonances, and the periodicity could be altered to increase the amplitude of the reflected signal for that axis. For simplicity here we have elected to restrict the experiment to the simpler parameters above to focus on the mechanism behind the polarization response.

VI. CONCLUSION
We have presented a proof-of-concept design of a metasurface that uses the orthogonal LSP resonances of a nanorod array to modulate the linear polarization state of the light it reflects. This was achieved by selecting the appropriate dimensions of the nanorods such that the lateral LSP modes occurred at the same pre-determined wavelength across the VO 2 phase change. This mechanism is repeatable, and easily reversed by simply lowering the temperature of the substrate.
We measure a 13% change in the total linear polarization for the signal reflected from our device at its central wavelength, as well as a sign change of the Stokes S 1 parameter, indicating a change from predominantly horizontally to vertically linearly polarized light. In conjunction with a linear polarizer to filter a specific linear polarization axis, we show this difference to be sufficient to clearly detect the change in signal. While the portion of incident light transformed to the desired polarization state is low, the signal switched between orthogonal states is sufficient to be detected, and hence, to form the basis of a useful device.
The operating wavelength of our device can be tuned by altering the dimensions of the metal nanorods in the array: increases in these dimensions could result in operating wavelengths in the infrared. This would increase the separation between the two lateral resonances of the rods, increasing the refractive index contrast across the VO 2 phases and therefore the effectiveness of the device. Devices such as this may therefore be of use in next generation optical circuits and other plasmonically enabled devices.

SUPPLEMENTARY MATERIAL
See supplementary material for details of the fabrication and characterization processes, normalised reflectance measurements for all combinations of input, and analyzing polarizer angles at 0 • , 45 • , and 90 • polarizer to array rods axes, the results of spectroscopic ellipsometry of VO 2 , and a comparison of numerical and experimental data. The absolute signal reflected from the sample is given. Also provided is a brief explanation of the Stokes parameters to give context for the discussion, and the full set of measured Stokes parameters for input linearly polarized light at 0 • , 45 • , and 90 • to the rod axes.