1. INTRODUCTION

The ability to control the polarization state of light from a laser is fundamental, as well as being desirable for a variety of applications, including polarization-sensitive imaging, measurements of Faraday rotations, and circular dichroism spectroscopy. Typically, external elements (e.g., polarizers and waveplates) are used to control polarization. Less common is the direct switching of the polarization state of the laser itself. Classical methods of changing the polarization direction of lasers generally rely on mechanical means, such as movable optics or electromechanical piezoelectric components [1–3], which are bulky and expensive. One exception is in some vertical-external-cavity surface-emitting lasers (VECSELs), which exhibit optical bistability and can be switched between nearly degenerate orthogonally polarized modes using an injection current [4,5]. However, since the dependence upon the injection current can be hard to predict, injection locking or an adjustable external cavity is often needed to improve stability [6–8]. Meanwhile, several passive metamaterial/metasurface structures have displayed the ability to select a specific circular polarization using chiral plasmonic structures [9–11]. A compact semiconductor laser with a dynamic polarization control capability integrated on chip is desirable, especially in the terahertz (THz) range, where many basic optical components are not readily available.

There have been several efforts to engineer the polarization of quantum-cascade (QC) lasers. In the mid-IR, plasmonic polarizers have been integrated onto the QC-laser facet to generate linear or circularly polarized light [12]; in another case, a modified ridge waveguide acted as a TM-to-TE polarization mode converter [13]. In the THz region, circularly polarized emission from a THz QC laser has been obtained by patterning surface-emitting gratings comprising sets of orthogonally oriented slots [14]. However, these are essentially static devices, with the polarization determined at the time of fabrication. To our knowledge, only two examples of dynamic polarization tuning of a QC laser have been demonstrated: one instance in which a TE-to-TM waveguide mode converter was biased to tune linear polarization over 45° [15], and one instance in which two side-by-side and $\sim \pi /2$ phased-shifted QC lasers feed surface-emitting gratings composed of orthogonally oriented antennas, which demonstrated linear to near-circular polarization tuning controlled by the current injection difference between the two lasers [16]. Nevertheless, dynamic modulation of THz polarization still mostly relies on external modulator elements, such as rotating polarizers/waveplates [17], active THz metamaterials [11,18,19], and liquid-crystal-based waveplates [20,21], all of which are subject to insertion loss and modulation speed limits. Photoelastic techniques for rapid polarization modulation are not available at THz frequencies. In this paper, we report a THz QC laser with built-in electrically controlled polarization switching between linear polarization states, obtained while maintaining a high output power, a single-mode spectrum, and an excellent beam pattern, all of which are nearly unaffected as the polarization is switched.

The foundation of this work is our recently demonstrated metasurface-based QC-VECSEL, which exhibits simultaneous high-power/efficiency and excellent beam quality [22,23]. The enabling component of a QC-VECSEL is the active reflectarray metasurface, which is made up of an array of low-quality-factor resonant antennas. Each antenna sub-cavity is a metal–metal waveguide loaded with an electrically biased QC active material so that the incident THz radiation is coupled in, amplified, and re-radiated. The active metasurface provides a highly flexible platform for integrating new functionality into a QC laser. In this work, we design a metasurface based around interleaved sets of cross-polarized antennas, a concept that has been previously demonstrated in the microwave range using leaky-wave antennas [24]. Figure 1(a) shows an SEM image of a fabricated polarization-selective metasurface. It is most intuitive to consider each “zigzag” antenna as a set of patch antennas that couple to the incident electric field polarized along the patch width (13 μm in this case) and are resonant at a frequency that corresponds approximately to when the width is equal to half of the wavelength within the semiconductor. Sets of patches are rotated either at an angle of 45° or 135° from the $x$-axis; these patches are then connected by narrower segments needed to allow a continuous dc injection current path for each antenna [as seen in Fig. 1(a)]. Patches of one orientation type are all electrically connected to one wire-bonding area and thus can be biased separately from the other type. By switching the electrical bias between the two sets, we can select the polarization preference that the metasurface amplifies and reflects. By pairing such a metasurface with an output coupler that is insensitive to polarization [see Fig. 1(b)], we create a QC-VECSEL with electrically controlled polarization switching capability. Because the cavity mode profile does not depend upon the detailed antenna structure, high power and an excellent beam pattern can be consistently maintained as polarization is switched.

 

Fig. 1. (a) SEM image of the fabricated metasurface. The zigzag metasurface covers an area of $2\mathrm{mm}\times 2\mathrm{mm}$. Only a center circular region of 1.5 mm diameter is biased, shown by the red dashed circle. The portions outside the circle have an ${\mathrm{SiO}}_2$ layer beneath the top metallization to prevent the quantum well medium from being biased. The tapered terminations serve both as the wire-bonding region and help suppress the reflection of traveling waveguide modes to prevent self-lasing. Antennas preferring one polarization direction are electrically connected together through the tapers on the top left of the metasurface, while others preferring the orthogonal polarization direction are connected together on the bottom right side. The inset shows a zoomed-in SEM image. (b) A schematic of the plano–plano VECSEL cavity. (c) Top view of a portion of the metasurface illustrated with dimensions given in micrometers. One set of antennas—the ones interacting with radiation linearly polarized at 45°—is shown in dark blue, while the second set of antennas, which interacts with radiation linearly polarized at 135°, is shown in light blue. For brevity, the former set will be referred to as Set 1, while the latter will be referred to as Set 2. The region inside the green dashed rectangle is one unit cell.

Download Full Size | PDF

2. DESIGN

The design goal for the metasurface is straightforward: first, when the injection current is applied to a single set of antennas, we wish to obtain a net reflectance gain at a narrow range of frequencies for a single incident polarization (i.e., $|{\mathrm{\Gamma }}_{45°-45°}|>1$, while $|{\mathrm{\Gamma }}_{135°-135°}|<1$ when Set 1 is biased and Set 2 is unbiased). Second, to ensure polarization purity, we must suppress cross-polarized scattering at those same frequencies ($|{\mathrm{\Gamma }}_{45°-135°}|\approx |{\mathrm{\Gamma }}_{135°-45°}|\approx 0$). Figure 1(c) shows a top view of the metasurface design and dimensions. Due to their geometry, Set 1 patches (drawn in darker blue) respond to light polarized at 45°, while Set 2 patches respond to light polarized at 135°. These patches, appearing as the thicker portion of the zigzags (13 μm wide), have a width of approximately $\lambda /2n$, where $\lambda$ is the free-space wavelength of the designed frequency, and $n$ is the index of refraction in the GaAs/AlGaAs quantum-well medium. The period is 71 μm in the horizontal direction and approximately 46.1 μm in the vertical direction (with the exact value being a function of the patch length of 39 μm, the connector width of 6.4 μm, and the right angle they form). The periodicity of the metasurface is chosen to be sufficiently small to avoid Bragg scattering at 3.4 THz for normally incident waves.

While the concept of using rotated sets of patch antennas is straightforward, the detailed implementation has several subtleties involved with minimizing the effect of the connector segments needed to provide a continuous top metallization and with preventing cross coupling between the two sets of antennas. First, cross-polarization scattering is reduced by clipping off the right-angled “elbow” bends along the zigzags, shown as dashed red lines in the upper corner of Fig. 1(c). Second, we suppress cross coupling between adjacent, separately biased antennas by introducing a vertical offset $\mathrm{\Delta }\approx 30.1\mathrm{\mu m}$ [with the exact value being a sum of the variable $\delta =7\mathrm{\mu m}$ defined in Fig. 1(c) and a geometric constant rounded to 23.1 μm]. Zero offset leads to a structure with adjacent ridges being mirror images with respect to the $y$-axis [i.e., the 90° axis shown in Fig. 1(c)]. The value of the offset and amount of clipping have been optimized by a parametric study of the effect of each parameter on the amount of cross coupling using single-cell periodic simulations (see Supplement 1 for details on simulations with different offset values), with the goal of suppressing the unwanted cross-polarized scattering (i.e., ensure $|{\mathrm{\Gamma }}_{45°-135°}|\approx |{\mathrm{\Gamma }}_{135°-45°}|\approx 0$) to ensure the purity of the generated linear polarization.

Modeling was performed using full-wave 3D finite-element simulations (Ansys HFSS). Reflection mode simulations at normal incidence were performed for a unit cell using periodic boundary conditions to simulate infinitely periodic arrays. The Drude model was used to describe the free carrier scattering loss in GaAs/AlGaAs material and gold metallization (the free carrier density used for metal is $5.9\times {10}^{22}{\mathrm{cm}}^{-3}$, and for the active medium, it is $5\times {10}^{19}{\mathrm{cm}}^{-3}$; the Drude lifetime used for metal is 39 fs, and for the active medium, it is 0.5 ps). The QC-laser material gain coefficient $g$ was modeled using an anisotropic permittivity and was assumed to be frequency independent over the simulation range. For clarity, we define gain coefficients $g_1$ and $g_2$ to represent the mount of gain supplied to Set 1 and Set 2 patches, respectively. Figure 2(a) shows the simulated co-polarization and cross-polarization reflectances ${|{\mathrm{\Gamma }}_{ij}|}^2$ of the metasurface for two cases: the metasurface is passive ($g_1=g_2=0$) as well as the case where Set 1 patches only are supplied with a gain of $g_1=30{\mathrm{cm}}^{-1}$ (emulating “turning on” Set 1 patches using bias current) and Set 2 patches kept passive ($g_2=0$). When gain is supplied to Set 1 only, net gain is observed for the incident E-field polarized at 45° near the target frequency of 3.4 THz, while the orthogonal polarization of 135° is almost unchanged compared to the fully passive case. A closer investigation of the reflectance change against the gain in Set 1 at 3.4 THz confirms the effectiveness of selectively amplifying one specific polarization via the bias switch [see Fig. 2(b)]; indeed, examination of the E-field profile shows the field is mostly localized to the Set 1 antennas. Furthermore, the design shows high effectiveness in suppressing cross polarization near the target frequency (${|{\mathrm{\Gamma }}_{45°-135°}|}^2$ and ${|{\mathrm{\Gamma }}_{135°-45°}|}^2<0.01$ across a bandwidth of 51 GHz). The asymmetric reflectance lineshape is likely a characteristic of a Fano resonance owing to the interactions and coupling paths between the complex set of resonances present within the metasurface lattice. One consequence of this is the strong cross-polarization scattering seen at 3.48 THz, particularly when gain is applied (see Supplement 1 for the field profile at 3.48 THz).

 

Fig. 2. (a) Top: Co-polarization and cross-polarization reflectance of the metasurface when Set 1 and Set 2 are both passive. The 45°–45° reflectance ${|{\mathrm{\Gamma }}_{45°-45°}|}^2$ designates the reflectance of light linearly polarized at 45° [defined according to the coordinates given in Fig. 1(c)] into light linearly polarized at 45°, and so on. Bottom: Co-polarization and cross-polarization reflectance of the metasurface when a QC gain of $g_1=30{\mathrm{cm}}^{-1}$ is assumed for Set 1 patches and Set 2 is kept passive. (b) The peak reflectance for 45°–45°and 135°–135° reflectances plotted against the gain $g_1$ supplied to Set 1, with Set 2 kept passive throughout. Inset is the simulated E-field intensity pattern of a unit cell at 3.4 THz for an incident E-field polarized at 45°.

Download Full Size | PDF

The expected polarization state of this laser is calculated by applying the polarization repeatability to the cavity round-trip propagation matrix, i.e., $\gamma \overrightarrow{E}={\mathrm{\Gamma }}_M\overrightarrow{E}$, where the matrices for the output coupler, free space, and the cryostat window are ignored, since they have no effect on the polarization. The metasurface field reflection matrix ${\mathrm{\Gamma }}_M$ is defined as

 

Fig. 3. (a) Simulated polarization eigenstate ellipses for the output beam when operating at the peak reflectance frequency of 3.397 THz. The two selectable polarization states, shown in blue (Set 1 switched on) and red (Set 2 switched on), differ by a rotation of 89.2°, and the intensity axial ratio of the ellipse’s major axis to its minor axis is 55 dB for both. The polarization eigenstate ellipses for the output beam when operating at 3.384 and 3.41 THz are plotted in (b) and (c). For these calculations, the co-polarized reflectance is held constant at its value at 3.397 THz in order to simulate all three cases with the same lasing threshold, with only the cross-polarized reflectance varied.

Download Full Size | PDF

We terminate the patch arrays with lossy tapers to suppress any self-lasing of the antenna ridges in the propagating mode, similar to the previous metasurface designs [24,25]. Additionally, to ensure that self-lasing of the antenna sub-cavity does not occur before the VECSEL lasing, we performed 3D Floquet–Bloch eigenmode finite-element simulations, including both radiative and material losses, to ensure that there were no high-quality-factor modes close in frequency to the design frequency.

3. FABRICATION AND CAVITY SETUP

A resonant phonon depopulation active region design very similar to [25] around 3.4 THz is used in this work, which was grown by molecular beam epitaxy in the $\mathrm{GaAs}/{\mathrm{Al}}_{0.15}{\mathrm{Ga}}_{0.85}\mathrm{As}$ material system (wafer number VB0739). The fabrication of polarimetric metasurface followed the standard procedures for making metal–metal waveguides, such as those described in [26]. The 10-μm-thick QC active layer was bonded to a receiving GaAs wafer via Cu–Cu thermocompression bonding, followed by substrate removal. Then, a layer of 200-nm-thick ${\mathrm{SiO}}_2$ layer was deposited and patterned to provide insulation between the top metal contact and the active region in the area outside the red circle, as shown in Fig. 1(a), including the taper, wire-bonding area, and part of the array area. A Ti/Au/Ni metal layer was evaporated and lifted off to provide both top metallization and a self-aligned etch mask for the subsequent chlorine-based dry etching to define the ridges. Finally, the Ni was etched away, and Ti/Au was evaporated onto the wafer backside. Dies are then cleaved and indium soldered to a Cu heat spreader.

The demonstrated VECSEL is based on a plano–plano Fabry–Perot cavity defined by the polarimetric metasurface mounted inside a cryostat and an output coupler in parallel mounted externally. The cryostat has a 3-mm-thick high-resistivity silicon window that introduces an intracavity etalon filter effect. The cavity length is approximately 9 mm. The output coupler used here is an inductive metal mesh on a 100-μm-thick crystal quartz substrate, whose transmission is approximately 20% around 3.4 THz (see Supplement 1 for the measured transmission spectrum).

4. RESULTS ON POLARIZATION-SWITCHABLE VECSEL

Electrical, power, and spectral characteristics were evaluated for a fixed VECSEL cavity alignment, where Set 1 and Set 2 antennas were switched on one at a time. As shown in Fig. 4, pulsed power-current-voltage (P-I-V) curves were measured at 77 K for each set with 0.25% overall duty cycle (500-ns-long pulses repeated at 10 kHz, modulated by a 150 Hz pulse train with lock-in detection). The power is measured using a pyroelectric detector and calibrated by a Thomas–Keating THz absolute power meter with 100% collection efficiency (given the directive beam pattern achieved). The measured P-I-V curves are very similar for the two sets, both exhibiting a threshold current density of $420\mathrm{A}/{\mathrm{cm}}^2$, a slope efficiency of $190\mathrm{mW}/\mathrm{A}$, and a peak power of $\sim 93\mathrm{mW}$. For comparison, a conventional metal–metal waveguide fabricated from the same wafer during the same process run had a threshold current density of $330\mathrm{A}/{\mathrm{cm}}^2$ at 77 K (see Supplement 1). A single-mode lasing operation was observed for both sets at an identical frequency of 3.37 THz (within the FTIR spectrometer’s resolution of 7.5 GHz), which did not change with electrical bias.

 

Fig. 4. Pulsed P-I-V curves measured for Set 1 and Set 2 in the same cavity setup at 77 K. The inset is the spectra measured for the two sets.

Download Full Size | PDF

To analyze the polarization state of the output, we placed a wire-grid polarizer (Infraspecs model P03) in between the output coupler and the detector and measured the power as the polarizer was rotated. This measurement was conducted for Set 1 and Set 2 by switching on one set at a time. The results shown in Fig. 5(a) demonstrate a polarization switching between linear polarized states separated by 80°. The slight deviation from the ideal value of 90° is likely due to non-ideal cross-polarized scattering from the metasurface, the reasons for which are discussed below. Far-field beam patterns were characterized using a 2-axis spherical scanning pyroelectric detector for Set 1 and Set 2 at the same bias point near the maximum power output. As can be seen in Figs. 5(b)–5(e), the measured beams are almost identical as the bias is switched between 2 sets, both exhibiting a directive and narrow near-Gaussian pattern with a full width at half-maximum (FWHM) angular divergence of $\sim 3°\times 3°$. Hence, the electrically controlled polarization switching is conducted without significantly affecting the output beam, spectrum, or total power.

 

Fig. 5. (a) Measured total power through the polarizer versus the polarizer angle for the two sets. 80° linear polarization angle switching is shown by the arrow. The circles are the experimental data, and the solid lines in red and blue are the fitted curves [to Eq. (2)]. The schematic on the bottom right of (a) shows the 2-axis far-field beam pattern measurement scheme. The measured 2D beam patterns for Set 1 and Set 2 are shown in (b) and (c), with an angular resolution of 0.5°. The 1D cuts along the $x$- and $y$-directions through the beam center for Set 1 and Set 2 are plotted as colored circles in (d) and (e), with the Gaussian curve fitting results plotted as solid colored lines.

Download Full Size | PDF

The uniformity of polarization over the beam was further evaluated by measuring the power versus polarizer orientation at different beam spots. Figure 6 presents the mapping results at five different beam spots, including one at the center and another four that are 2° away from the center in four directions. It can be seen that the polarization is very uniform across the $x$-axis. It is slightly less uniform across the $y$-axis, with a maximum deviation of 22° for Set 1 and 13° for Set 2 rotated towards the $y$-axis. We further assessed the linear polarization purity by fitting the power against the polarizer angle data to the expression,

 

Fig. 6. Measured power through the polarizer against the polarizer angle at five spots on the output beam for Set 1 (a) and Set 2 (b). The dots are the experimental data, and the solid lines are the fitted curves [to Eq. (2)].

Download Full Size | PDF

Table 1. Axial Ratios of Field Intensity (in Units of dB) for the Total Beam and Different Beam Spots from Set 1 and Set 2 Patches

View Table

5. RESULTS OF METASURFACE SELF-LASING

The metasurface QC-VECSEL is designed to operate with only one antenna set switched on at a time. Nonetheless, a natural question arises: what if both antenna sets are switched on together? For example, might one be able to continuously vary the polarization of the output state, as in the case of passive cross-polarized antennas? Unfortunately, the answer is no: due to the cross coupling between antenna sets, the introduction of gain into both sets significantly changes the reflectance spectrum and results in lasing at different frequencies than the original design at 3.40 THz. However, this behavior is interesting in its own right.

In the experiment, when both sets of antennas were biased together, we observed self-lasing of the metasurface alone, without any output coupler. The measured P-I-V curve and spectra at different biases are shown in Fig. 7(a). Single-mode lasing at a frequency of 3.75 THz is observed and is unchanged with bias within the resolution of the FTIR. The power peaks at 144 mW with a slope efficiency estimated at $139\mathrm{mW}/\mathrm{A}$. The threshold current density is $400\mathrm{A}/{\mathrm{cm}}^2$, lower than the threshold for the polarimetric VECSEL, although the larger bias area results in a larger total threshold current. The power output is in the surface normal direction and exhibits a directive and narrow beam of $3°\times 4°$ FWHM divergence, as shown in Fig. 7(b), which is measured with the metasurface positioned as the origin of the measurement setup as Fig. 5(a) shows. Some excess power below the main lobe is observed; we speculate that this may result from scattering from the wire bonds. The polarization over the output beam is found to be largely linear over the beam, but with a very non-uniform distribution of polarization direction, as shown in Fig. 7(c). We speculate that this behavior is associated with the complex and spatially varying phase relation between the coupled modes in Set 1 and Set 2, along with inhomogeneities in fabrication, structure, and biasing which break the expected symmetry between 45°- and 135°-polarized radiation. This is not fully understood, and will require further detailed simulation and experimental study.

 

Fig. 7. (a) Measured P-I-V curve for the polarimetric metasurface without an external cavity. Nominally identical bias is provided to both Set 1 and Set 2. The inset shows the spectra at different injection current levels. (b) Measured far-field beam pattern for the metasurface self-lasing near the peak power, with FWHM divergence angles in the $x$- and $y$-directions labeled. The polarization is mapped at different beam spots within the white dashed box. (c) Measured power through the polarizer at different spots on the beam pattern indicated by the circled white cross marks.

Download Full Size | PDF

The self-lasing phenomenon is consistent with behavior predicted both by eigenmode type simulations and driven reflection mode simulations if both antenna sets are supplied with the same amount of gain (see Supplement 1 for details on simulation results). The simulated reflectance spectra show the emergence of a sharp Fano-type resonance at 3.80 THz for $g_1=g_2=20{\mathrm{cm}}^{-1}$ in both the co- and cross-polarization reflectances. The eigenmode simulation shows that the metasurface exhibits self-lasing in a surface-emitting mode at $\sim 3.8\mathrm{THz}$, with a low threshold gain of $\sim 17{\mathrm{cm}}^{-1}$ (see Supplement 1 for details on the simulated gain thresholds for several modes found between 3 and 4 THz). This gain threshold is lower than the value of $28{\mathrm{cm}}^{-1}$ estimated for the metasurface VECSEL lasing. This self-lasing is attributed to a standing wave mode resonance along the antenna patch length; the mode is “bright” in that it radiates strongly in the far field with even parity. If only one set is biased, the simulations predict that the gain threshold is much larger: approximately $30–35{\mathrm{cm}}^{-1}$.

This self-lasing phenomenon can be considered to be an unintentional realization of a laser array phase locked through mutual antenna coupling, similar to that reported in [27]. The self-lasing mode relies on strong coupling between sets of antennas, as is revealed by the strong cross-polarized scattering shown in the simulation results at 3.80 THz and the fact that the gain threshold is much higher when only one set of antennas is biased (Figs. S5 and S6 in Supplement 1). The fact that such a narrow far-field beam was obtained indicates that locking is obtained over the entire 1.5-mm diameter bias area (19 wavelengths), which encompasses $\sim 1080$ patch elements. The performance of this self-lasing phenomenon could be further improved if the metasurface was designed by intention to bring the self-lasing frequency down to better match the QC material optimal gain region of 3.3–3.5 THz.

6. CONCLUSION

We have demonstrated a THz metasurface QC laser with electrically switchable polarization. Dynamic switching of the linear polarization angle of the laser between two states separated by 80° is observed, while maintaining a consistent high output power, threshold, single-mode spectrum, and a high-quality beam pattern. This represents the flexibility of the metasurface QC-VECSEL approach to customize the polarization response. Such a polarization-switchable THz laser is relatively simple to assemble and offers compactness, robustness, and speed over other designs that use moving parts. Furthermore, because our approach is based upon the switching of the gain within a laser cavity, it avoids issues with insertion loss associated with external polarimetric modulators, which can be particularly large in the THz region. Indeed, there does not appear to be a significant performance penalty of this device compared to conventional QC-VECSELs [23]. In addition, the polarization purity is observed to be greater than 13 dB for the total output, and the polarization state is mostly uniform across different positions within the beam pattern. It is likely this can be further improved by optimization of the VECSEL cavity to have the lasing frequency better match the metasurface resonance and by exploration of advanced metasurface designs with improved broadband suppression of cross-polarized scattering. The biasing may also be switched with a high modulation frequency: in principle, the switching speed is limited only by the buildup time of the laser oscillation (∼nanosecond or less) and the RC time constant associated with the electrical bias. Such a high-performance QC laser with electrically switchable polarization may find use within a multitude of THz applications, such as polarimetry, spectroscopy, and cancer detection [28–30]. Furthermore, simply by adding a quarter-wave plate in the path of the output beam, the laser can be converted into one that switches between circular polarizations of opposite handedness. Alternately, a future chiral metasurface design could selectively amplify specific circular polarization states directly.

Finally, it is important to comment upon the prospects for continuous-wave (cw) operation for the polarimetric metasurface laser architecture. The devices reported here were tested at a low duty cycle. The large bias area (1.5-mm diameter) draws a great deal of current; as a result, the total thermal load is too large to allow cw operation at 77 K. However, we would emphasize that the metasurface VECSEL architecture is very advantageous for cw or high-duty cycle operation once the bias area is reduced to bring down the total thermal load. In particular, the narrow ridge geometry of each antenna is an advantageous geometry for heat removal. For example, in a separate work, 40 mW cw power at 6 K was observed from a QC-VECSEL with a 1-mm bias area diameter [23], and 5–7 mW of cw power at 83 K was reported in a QC-VECSEL with a 0.7-mm bias area diameter [31].