Tunable broadband terahertz polarizer using graphene-metal hybrid metasurface

: We demonstrate an electrically tunable polarizer for terahertz (THz) frequency electromagnetic waves formed from a hybrid graphene-metal metasurface. Broadband ( > 3 THz) polarization-dependent modulation of THz transmission is demonstrated as a function of the graphene conductivity for various wire grid geometries, each tuned by gating using an overlaid ion gel. We show a strong enhancement of modulation (up to ∼ 17 times) compared to graphene wire grids in the frequency range of 0.2–2.5 THz upon introduction of the metallic elements. Theoretical calculations, considering both plasmonic coupling and Drude absorption, are in good agreement with our experimental ﬁndings.


Introduction
Components to enhance and extend the capability of THz spectroscopies have been intensively developed in the past few decades, including efficient antennas [1], switches [2], absorbers [3], and biosensors [4][5][6]. THz polarizers are essential for removing unwanted polarization components and / or of controlling the polarization state of THz waves and hence are important components in many THz systems and applications [7,8]. Several polarizers for THz radiation have previously been reported, including those formed from liquid crystals [9], semiconductors [10], metal wire grids [11], carbon nanotubes [12], and graphene [13]. Particular emphasis has been placed on graphene, owing to its electrically tunable optical (including THz) conductivity [14], which allows for tunable (Drude) absorption [15]. Graphene is also ideal for such applications since its plasmon resonance, which can be fine-tuned by changing its geometry [16][17][18], lies in the mid-infrared to THz frequency range [19].
Several different schemes for THz polarizers incorporating graphene have been presented previously [20]. Metasurfaces are comprised of two-dimensional arrays of elements capable of anisotropic manipulation of EM waves [21]. They offer a promising approach to achieving tunable THz polarizers, either in the form of pure graphene metasurfaces [13,22,23], or through the incorporation of sheets of graphene into metal metasurfaces [8,24]. These structures have shown their capability to control the polarization direction [8,23,24] and to manipulate polarization by converting between linear and circular polarization states, for example [13,22,23]. Compared with reports on the manipulation of the polarization state, THz amplitude modulation, and especially broadband modulation, is less common, despite the importance of being able to tune the amplitude of a desired polarization for the development of spatial light modulators [25] for THz communications [26] for example. More importantly, studying tunable THz polarizers is essential for developing THz switching and attenuator devices for THz communications [27].
THz polarizers formed from wire grids (either free-standing [28] or on a substrate [29]) have a low insertion loss (1 dB) and high polarization extinction ratio (PER) (up to ∼40 dB in the range 0.3-1 THz) [11]. Such metallic wire-grid polarizers typically show a low transmission for transverse electric (TE) polarized waves and a high transmission for transverse magnetic (TM) polarized waves [28]. The same is true for wire grids formed from graphene, both in the near-infrared [30,31], and in the microwave regions of the electromagnetic spectrum [32], while at THz frequencies, they typically show a low transmission for TM-polarized THz waves owing to THz wave-plasmon coupling [17,33]; this thus represents a limitation of pure graphene wire grid based polarizers. A further limitation is the low modulation (∼10%) of the TE transmission, which is caused by Drude absorption [17].
In this paper, we report on the design, fabrication and characterization of broadband tunable THz polarizers utilizing graphene-metal hybrid wire grids. Compared with previous hybrid graphene-metal metasurface polarizers that employ sheets of graphene covering the whole metal metasurface area [8,24], our design is that of a hybrid graphene-wire grid and metal metasurface which allows both the graphene and the metal elements of the design to contribute to the anisotropic properties of the polarizer, providing a significant PER enhancement. Our design also overcomes two drawbacks of graphene wire grid THz polarizers, namely their weak modulation and the THz wave-plasmon coupling, which restrict the TM transmission [17]. Our polarizer shows a broadband (up to several THz) polarization-dependent modulation, with an up to ∼17 times enhancement of modulation ability in the range 0.5-2.5 THz compared with a graphene-wire grid polarizer containing no metal.

Design and principle
Our graphene-metal hybrid wire grid structure, comprising a graphene wire grid overlaid on to a metal metasurface such that the graphene and metal elements are interleaved, is shown in Fig. 1(a). The metal metasurface is formed from a metal patch array. The introduction of metal makes the sheet conductivity of the graphene-metal hybrid wires larger than wires formed from graphene alone. The conductivity of the hybrid wires is given by where, σ graphene and σ metal are the sheet conductivity of graphene and metal, respectively, and L graphene , L metal , and Λ G/M are the dimension of graphene, metal patch and period of graphenemetal structure along the wires, respectively, as shown in Fig. 1(a). Since the sheet conductivity of the metal is much larger than that of graphene, we can write ∆σ s ≈ ∆σ graphene × Λ G/M /L graphene to a reasonable approximation. Thus, smaller graphene filling factors (GFF) along the wire, L graphene /Λ G/M , lead to a larger conductivity modulation, which in turn results in a larger THz transmission modulation. The strong TM-polarized plasmon mode present in graphene wires is suppressed by the introduction of the metal metasurface, which effectively acts to transform the graphene wires into separated graphene patches connected by metal. The EM wave-to-plasmon coupling efficiency of the graphene patches is determined by their dimensions, and for a certain width, the TM-polarized coupling is highly suppressed by decreasing L graphene [34]. As well as this suppression, the metal metasurface introduces a TE-polarized plasmon into graphene, whose interaction with THz wave is then tunable by modulating the conductivity of graphene; this TE-polarized plasmon also contributes to the THz transmission modulation [35]. The generated TE-polarized plasmon is induced by the THz wave coupling to the resonance of an inductive-capacitive (LC) circuit formed by the metal-graphene structure, in which the metal acts as a capacitive reservoir for charge accumulation and the graphene serves as an inductive channel. The transmission of the graphene-metal hybrid wire grid can be calculated by t = t d × t p , where t d and t p are the transmission after Drude absorption and TE-polarized plasmon absorption, respectively. t d can be calculated by t d (ω) = 2n 1 /(n 1 + n 2 + Z 0 σ s ) [36], where σ s is the sheet optical conductivity of the wires, and n 1 and n 2 are the refractive indices on the two sides of the wire grids, while t p can be calculated by , where Y 1 and Y 2 are the inverse of the characteristic impedance of materials on both sides of the wire grids, and Y(ω) is the complex admittance of the graphene-metal LC circuit. All theoretical results in this work were calculated based on the equations above.

Fabrication and measurement methods
Five devices of different geometries, as described in Table 1, were fabricated to investigate their effect on the polarizer performance. Devices A, B, and C were fabricated to show the effect of changing Λ G/M , while devices A, D, and E demonstrated the effects of changing GFF. Figure 1(b) shows a schematic diagram of the THz transmission experiment used to investigate the polarizer response. θ is the angle between the incoming THz wave electric field polarization and the wire orientation. θ = 0 • represents TE polarization, while θ = 90 • represents TM polarization. The upper two metal contacts connecting to either end of the wires were used to measure the DC conductance of the wire grid, which was then used to calculate the conductivity, based on the wire geometry. The lower metal contact was used to apply a gate voltage, enabling conductivity modulation. The polarizers were fabricated on a 0.35 mm thick ST-cut SSP quartz substrate. The metal patterns were fabricated first using direct-write DLP laser lithography (MLA150, Heidelberg Instruments) followed by electron-beam evaporation with Ti/Au (7nm/70nm) layers. CVD graphene samples on copper were then transferred onto the  [37,38]; Poly-(methyl methacrylate) (PMMA) was spin coated as a support, and the copper foil was then removed using 2% ammonium persulfate solution. The graphene initially covered the whole metal metasurface area, which was 1.5mm × 1.5mm. The graphene was then etched into the wire-grid structures using oxygen plasma at a power of 50 W for 2 minutes, with the graphene wires protected by S1813 photoresist, which was removed in acetone after etching. Devices were electrically tuned using an ion gel (a mixture of 2 parts of 1-ethyl-3-methylimidazolium bis(trifluoromethylsulfonyl)imide ([EMIM][TFSI]) and 1 part of poly(vinylidene fluoride-co-hexafluoropropylene) (PVDF-HFP) by weight) applied as the gate dielectric material; this gel has previously shown good transparency for THz waves [24] along with the ability to induce a large carrier density at low gate voltage [39]. The ion gel was prepared by dissolving PVDF-HFP in acetone with the aid of a magnetic stirrer for an hour, before adding the [EMIM][TFSI] into the solution and stirring for 24 hrs. The prepared ion gel was then drop-cast on to the gratings, and dried in ambient conditions for one hour [40]. The gate voltage-dependent DC conductivity of the graphene in each device was measured directly using a source-measurement unit, yielding the data shown in Fig. 1(c). The devices were imaged using SEM and characterized using Raman microscopy after THz transmission measurements. Examples of the SEM image of the wire grids, and the Raman spectrum of graphene in the polarizer, are shown in Figs. 1(d) and 1(e), respectively. The dark regions between the metal patches in the SEM image are graphene, as confirmed by the Raman spectra, in which the 2D peak was much larger than G peak, verifying the presence of single-layer graphene [41].
A THz time-domain spectroscopy (THz-TDS) system was used for transmission measurements. In this system both the THz emitter and detector were bow-tie pattern LT-GaAs photoconductive (PC) antennas, comprising a 2-µm-thick layer of LT-GaAs transferred on to a 5-mm-thick quartz substrate, fabricated using an epitaxial lift-off method [42]. The transmission characteristic of each polarizer device was obtained by dividing the Fourier transform of the transmitted THz pulse to a reference obtained from a quartz substrate with an identical layer of ion gel.  (Fig. 1(c)). The THz transmission was obtained for different gate voltages, as shown in Fig. 2(b), with examples of the time-domain THz pulses passing through the polarizer shown as insets. As anticipated, the largest transmission changes were observed for θ = 0 • ; the transmission gradually reduced with increasing θ up to θ = 90 • , at which the point transmission curves at different gate voltages essentially overlapped. The modulation depth (MD), shown in Fig. 2(c), was obtained by MD = (T D − T)/T D , where T D is the transmission when V g = V Dirac and T is the transmission of the polarizer, here obtained when V g = −2.1V. As expected, the largest transmission modulation was seen for the TE polarization, with the modulation gradually reducing as θ was increased to 90°. The inset of Fig. 2(c) presents the polarization-dependent MD at 0.5 THz. The MD of the TE transmission reached a value of ∼44% for device A for frequencies between 0.5-1.5 THz, while the MD of the TM transmission was found to be negligible. Figure 2(d) shows the suppression of the MD of the TM transmission when L graphene is reduced. For a graphene wire grid, device E, the MD of TM transmission was around 14% to 21% (red solid line), while when L graphene was reduced, the MD of TM transmission was also reduced (green and blue solid line). For the 25-µm-wide wires, the MD of TM transmission was found to be negligible when L graphene was smaller than 5 µm. Since, as expected, the MD of TM transmission was negligible, the following discussions will focus on the MD of the TE transmission. We note that the modulation speed of an ion-gel gating device such as ours is limited by the electrical characteristics of the ion gel. The operation speed of the ion gel we used in this work was found to be ∼ 300 Hz, reported by K. Chae et al [40]. This modulation speed is relatively slow compared to the conventional field-effect transistor based modulator [43], which can show several kHz operation, but the modulation speed of our device could be significantly increased (up to ∼ 10 kHz) by replacing the polymer in the ion gel from the PVDF-HFR used with PS-PEO-PS triblock copolymer [44].

Results and discussion
The working bandwidth of the polarizer was limited by the period of the graphene-metal grid, Λ G/M , since the metal patch array acts as a low-pass filter with a cut-off frequency determined by its period [45,46]. To investigate this, three devices (devices A, B, and C) with different Λ G/M were tested as shown in Fig. 3. The MD was obtained by tuning the gate voltage V g = V Dirac to V g = −2.1V in each case. The cut-off frequencies of the experimental data in Fig. 3 were found to be similar to the theoretical cut-off frequencies of the metal patch arrays as shown in Table 2, indicating that the period of the metal array was predominantly determined by the bandwidth of the polarizers. Here, the cut-off frequencies were calculated using their relationship with the period of the metal array (f cut−off =c/n eff ·Λ G/M , where c is the speed of light, and n eff is an effective refractive index). As such, to obtain broader bandwidth, a shorter Λ G/M would be required. Among the fabricated devices, the largest bandwidth was obtained by device C, with Λ G/M = 15µm. The bandwidth of this device extended across the whole bandwidth of our TDS system, ∼5 THz. The fractional bandwidths were also estimated in Table 2 to evaluate our devices. We obtained a fractional bandwidth of ∼ 1.6 with the device C, which may be categorized as ultra-wideband [47].  To investigate the modulation enhancement by introducing the metal metasurface into the graphene wire grid, device A and device D with different GFF to compare with device E (graphene wire only device). The MD of these devices as the gate voltage was changed from V Dirac to −2.1 V were measured and theoretically calculated as shown in Figs. 4(a) and 4(b), respectively. Compared with device E, in which no metal metasurface was introduced, device A showed an approximately four times greater MD at 0.5-1.5 THz, and device D showed an approximately ∼six times greater MD at 0.5-2.5 THz. This demonstrates that a smaller GFF induces a considerably greater modulation, as expected. The PER of device D was modulated from 3.7 dB to 10.3 dB obtained from PER = 20log(T TM /T TE ), where T TM and T TE were the TM transmission and TE transmission amplitudes, respectively. The maximum PER was obtained when V g = −2.1V, and minimum when V g = V Dirac , which was 0.9 V for device D. The PER value obtained here (∼10 dB) is below compared to that of metal wire grid (20 -40 dB), though still slightly better than tunable device using graphene/insulator stacks (9.5 dB) [48]. We note that our design thus shows a significantly enhanced PER when compared to devices that integrate sheet graphene with metal patch arrays, in which only ∼0.7 dB of PER has been obtained [49]. Comparing the experimental and theoretical transmission results (Fig. 4), we note several differences, notwithstanding their overall similarity. First, the measured MD is up to ∼3% larger than that calculated for low frequencies. This is likely to be a result of the contact resistance between graphene and metal, causing the dc conductivity used in the calculation to be smaller than the sheet conductivity. Second, the measured MD decays somewhat more slowly with frequency than the calculated results. This could be caused by impurities and vacancies in the graphene, which may affect the optical conductivity and plasmon excitations [50] by the introduction of additional carrier scattering. Additionally, we observe a slight oscillatory behavior in the experimental data, likely to be caused by interference of the THz pulse caused by the reflection the surfaces (substrate and ion gel). In addition, a more rapid MD decrease with respect to frequency was observed for frequencies above 3 THz in experiments owing to the limited system bandwidth, which rolled off beyond 3 THz.
The transmission modulation was found to be proportional to the graphene conductivity (Fig. 5), and we used the gradient of the curves of MD versus conductivity (∆MD/∆σ graphene ) to quantify the modulation ability of the polarizers. We note that the key factor here is the slope since we quantify the modulation ability using it. The maximum modulation ability enhancement came from device D, which had the smallest GFF, while the modulation ability enhancement compared to the device without any metal metasurface (device E) was found to be 17.8 × according to the experiment, and 18.0 × according to calculation. The MD and the working bandwidth of the polarizer could be improved by further reducing the GFF and the Λ G/M , which we investigated theoretically. We considered graphene dimensions from 10 nm to 10 µm, noting that the former could potentially be realized experimentally using e-beam lithography [51]. The GFF were set from 0.001 to 0.5 while the graphene conductivity was varied between 0.1 mS and 20 mS [36]. The calculated MDs at 0.5 THz are shown in Fig. 6. Based on these parameters, the maximum MD found to be possible was 99.997%, employing 10 nm graphene with a 10 µm period.
Aside from the MD, other performance parameters such as the bandwidth and transmission of the polarizer also change when the geometries are changed. To investigate this theoretically, several different geometries were investigated, realized to obtain a range MDs as shown in Fig. 6; the results are summarized in Table 3. It can be seen that either a shorter L graphene or larger GFF results in a broader bandwidth (reaching as high as >90 THz for geometry 1 in Table 3), while smaller GFF results in smaller TE transmission, indicating a higher PER (eg. up to ∼70 dB for geometry 5). Here, the modulation bandwidth here is determined by the structural factors such as GFF rather than gating method. The possible PER of our design is up to 30 dB higher than commercial metal wire grid THz polarizers (20-40 dB). By a suitable choice of geometry, the proposed polarizers can thus target either maximal bandwidth, maximal PER, or a compromise between these factors according to the needs of particular applications. We note that fabrication of the smaller devices could not be achieved using our current patterning method (direct-write DLP laser lithography) owing to its resolution limit of ∼ 1 µm. We anticipate that these devices could be fabricated using alternative methods such as high-resolution deep UV lithography in the future work, or by electron beam lithography, though we note the latter would be time-consuming for our device with patterned area of 1.5mm × 1.5mm.

Conclusions
A new tunable broadband THz polarizer with a geometry comprising a graphene-metal hybrid metasurface was designed, fabricated, and characterized. We experimentally verified its broadband, polarization-dependent, and electrically tunable THz transmission modulation. We demonstrated a significant enhancement of THz transmission modulation in this device, which gave a ∼17 times modulation ability enhancement in the frequency ranging from 0.5 THz to 2.5 THz compared to graphene wire arrays. Theory indicates that the performance of the polarizer could be further improved by reducing the dimensions, allowing the structures to satisfy applications where either ultra-broadband modulation or ultra-high PER are required. Our device could be applied as a polarization sensitive THz switch or attenuator in THz communication circuits, for example, and could also be extended to a wider range of operating frequency by controlling the geometrical parameters in the ways suggested, while noting that no current efficient THz source or detector combination is yet available which would extend across the entire potential working frequency range of this form of polarizer (from 0.1 to >90 THz).