1. Introduction

Terahertz (THz) spectroscopy has opened a wide range of scientific and practical applications [1–7]. In solid state physics, it enabled probing of inter-sub-band transitions in semiconductor quantum wells [1], energy gaps of superconductors [2], cyclotron resonances in high-mobility 2D electron gases [3] and crystal field splitting in rare earth element doped crystals [4,5]. In practical applications, THz spectroscopy has shown potential for atmospheric monitoring [6], quality-control [7], security, biomedical imaging and sensing [8]. Two approaches are commonly used to enable optical laser-driven THz spectroscopy: time-domain and frequency-domain approaches. While the more established THz time-domain spectroscopy (THz TDS) covers a broad spectrum of THz frequencies and provides amplitude and phase information with high signal to noise ratio (SNR), the spectral resolution is limited typically to the GHz range by the physical length of the delay line used [9]. In THz frequency-domain spectroscopy (FDS) the resolution is limited by the free running laser line widths [10]. A much higher spectral resolution, down to 10 Hz [5], can be achieved in THz FDS using stable 1550 nm wavelength telecommunication laser technology [5]. However, it remains challenging to simultaneously realize low-noise and wide bandwidth ultrafast mixers using photoconductors with a sufficiently narrow bandgap for absorbing 1550 nm photons, such as InGaAs due to its low dark resistivity [11–13].

Despite a larger bandgap, low-temperature (LT) grown GaAs can be photoexcited using 1550 nm light [14–17]. LT GaAs has been shown to weakly absorb light via two-step photon absorption and mid-gap defect states [18]. LT GaAs is also one of the best ultrafast photoconductors for THz detection [19] and its use in THz detectors with photoexcitation at 1550 nm have been proposed in several configurations [20–24]. However, the non-linear two-step photon absorption process is much weaker compared to direct inter-band absorption. To improve absorption efficiency, plasmonic electrodes [20], resonant cavities formed by gold mirrors [21] and LT GaAs metasurfaces [22,25] have been integrated into THz detectors to enable pulsed THz TDS systems. Despite these promising results, there has yet to be a demonstration of a CW THz FDS system based on the two-step photon absorption at 1550 nm in a LT GaAs metasurface. The challenge for CW systems stems from their operating conditions: in THz FDS systems, the 1550 nm pump peak intensity is orders of magnitude lower than the peak pulse intensity used in THz TDS systems, and therefore it is difficult to exploit the nonlinearity of the two-step absorption process.

In this article, we demonstrate detection of CW THz waves using the two-step photon absorption process in LT GaAs Mie metasurfaces. Despite the weak nonlinearity, we show for the first time CW THz detection up to 1 THz with 45 dB peak dynamic range [26] using an LT-GaAs photoconductive metasurface with a telecommunications wavelength (1550 nm) optical pump. We find that the LT-GaAs detectors have a photocurrent noise floor two orders of magnitude lower than commercial InGaAs detectors, but a lower peak dynamic range and bandwidth where commercial systems achieve 90 dB and 2.7 THz respectively. We also find that in contrast to the above-the-bandgap photoexcitation of LT GaAs metasurfaces, the 1550 nm photoexcitation requires a refinement of the metasurface design to counterbalance the effect of metallic electrodes, which could lead to an 50 nm shift of the metasurface peak absorption wavelength. This electrode effect is sufficiently large to detune the metasurface modes out of the telecommunications laser wavelength range, affecting CW THz detector efficiency. With this insight and refined design, LT GaAs metasurfaces can offer low-noise ultrafast photoconductive mixing exploiting the 1550 nm telecommunications laser technology and enable low-noise THz detection for THz FDS with increased spectral resolution in a compact and cost-effective solution, when compared to conventional THz TDS systems.

2. Metasurface design

The metasurface design is shown schematically in Fig. 1(a). It was adapted from a design previously reported to realize THz detection when pumped with 100 fs IR pulses [22]. The design uses optical resonances in the metasurface to enhance the two-step photoexcitation of LT GaAs at a wavelength of 1550 nm. The metasurface consists of rectangular LT-GaAs blocks which are elongated in x-direction and connected by narrow channels in the y-direction to form a grid structure (see Fig. 1(a)). This design enables charge carriers excited in the LT-GaAs to flow through the metasurface to electrical contacts (Fig. 1(b)), which form the antenna of the THz detector (shown in Fig. 1(c)). The LT-GaAs metasurface structure and the 60 $\mu m$ diameter gold bowtie antenna with a 3 $\mu m$ gap between the antenna electrodes are fabricated using electron beam lithography and lift-off. The device is transferred to a transparent sapphire substrate. Fabrication details can be found in the Supplement 1.

 

Fig. 1. a) Schematic of the metasurface unit cell element. L and W are the length and width of the metasurface block, $W_b$ is the width of the bar, T is the thickness, and $P_x$ and $P_y$ are the periods. b) A false color scanning electron microscope image of the metasurface with THz antenna electrodes. c) Artistic representation of the LT-GaAs THz detector illustrating incident optical excitation and THz waves, a unit cell supporting the electric dipole mode and charge carrier transport, and a schematic diagram of photoexcitation process.

Download Full Size | PDF

The metasurface resonances are the two lowest order Mie modes of opposite symmetries (odd and even with respect to the metasurface plane), the magnetic and electric dipole modes, which are ’tuned’ to at the wavelength of photoexcitation (1550 nm). An illustration of the electric dipole mode field profile at 1550 nm is shown in Fig. 2(a). In principle, a metasurface design with these two degenerate modes can be used to satisfy the degenerate critical coupling condition leading to 100% absorption [27]. However, for GaAs at 1550 nm, the degenerate critical coupling condition can be satisfied only for a narrow band of wavelengths (<10 nm) due to the low intrinsic absorption of GaAs at this wavelength, limiting the bandwidth of THz photoconductive detectors. To enable operation over a broader band of optical excitation, our metasurface design was optimised for absorption over the entire telecommunications C-band 1530-1565 nm. We therefore detuned the design from the critical coupling condition to broaden the bandwidth to 35 nm. We note that there is a trade-off between the level of absorption and the bandwidth. The effects of the metasurface parameters on its optical properties were discussed in [22].

 

Fig. 2. a) Simulated electric field distribution Ey showing the electric dipole mode in a periodic metasurface (unit cell). b) Measured transmittance spectra of LT-GaAs metasurfaces. c) Photocurrent spectra under a constant bias (200mV) for three tested metasurfaces. d) Numerically simulated transmittance and absorbance for a metasurface unit cell with infinitely periodic boundary conditions and a plane wave excitation.

Download Full Size | PDF

3. Photoconductivity in metasurfaces

The wavelength of the metasurface resonances can be experimentally determined from optical transmission properties. Transmittance spectra of three metasurface test samples (D1A, D2B and D3B; all 30 $\mu m$ X 30 $\mu m$ in size) were measured using a focused broadband white light source, linear polarizer, and a spectrometer. The metasurface dimensions are detailed in Fig. 1(b): thickness T = 320 nm, block width W=580 nm, block height L =420 nm, period in x-direction Px = 630 nm, period in y-direction Py = 660 nm (in D1A) and 690 nm (in D2B and D3B), bar width Wb = 210 nm (in D1A) and 240 nm (D2B and D3B). The annealing conditions for LT GaAs were slightly different: 60 seconds for D1A and 40 seconds for D2B and D3B.

Transmission spectra for the three samples are shown in Fig. 2(b). D1A shows a clear broad dip in transmission centered at 1500 nm whereas D2B and D3B show several dips centered around 1500 nm for D2B and 1550 nm for D3B. The dips (reduced transmission) indicate increased absorption at the wavelengths of the Mie modes supported by the metasurface (see numerical simulations in Fig. 2(d)). Where one dip is seen, the two modes overlap at approximately the same wavelength (degenerate), whereas multiple dips indicate that the modes do not overlap spectrally. The loss of mode overlap could have been caused by small changes in design B in comparison to design A: a larger bar width and the larger period Py. The variation in spectra between devices D2B and D3B, which were fabricated with the same nominal design dimensions (B), suggest that the optical properties are sensitive to the fabrication conditions and that the mode alignment to the excitation wavelength as well as the mode degeneracy can be easily lost. Despite the variation in transmission properties, all devices show reduced transmission in the 1550 nm band.

Since the observed reduced transmission indicated enhanced absorption only indirectly, we evaluated absorption and its dependence on the excitation wavelength using photocurrent spectroscopy. We directly measured the photocurrent for three THz detector devices containing similar metasurfaces D1A, D2B and D3B with electrical contacts deposited over the metasurfaces, as shown in Fig. 1(c). The metasurfaces were connected to a Source-Measure Unit (Keysight 2902B) and a 200 mV DC bias was applied to the contacts. Linearly polarized light from a tuneable laser (Keysight N7776C) was coupled from polarization maintaining fiber into a collimated free-space beam and then focused on the metasurfaces, in the 3 $\mu m$ wide gap between the metallic contacts. The measured photocurrent was recorded as the laser wavelength was swept from 1490 nm to 1600 nm. The photocurrent spectra are shown in Fig. 2(c). The measured responsivity peaks around 1480 nm (the short wavelength limit of the laser range) for all metasurfaces, and the responsivity of D2B and D3B is reduced relative to D1A. The latter is likely due to the loss of mode degeneracy, as suggested by the transmittance spectra where separate modes are seen at different wavelengths. It’s likely that the shifting and broadening of the absorption and thus the responsivity we observe are due primarily to the influence of the metallic contacts on the metasurface and variation of the unit cell dimensions due to fabrication tolerance ruling out any meaningful conclusions of the influence of annealing time on photoresponse.

4. Effect of metallic contacts on metasurface absorption

While the metasurfaces were designed to absorb maximally at 1550 nm, in Fig. 2(c) we observe for D1A and D2B that the peak responsivity appears to be blue-shifted for both device geometries in comparison to the peak absorption seen in numerical simulations (Fig. 2(d)). Moreover, the transmittance dips also exhibit a blue-shift and are slightly decreased in depth. To identify possible reasons for this discrepancy, we first compare the experimental conditions with the numerical simulation configuration. The metasurface design was developed numerically using a unit cell with periodic boundary conditions and a plane wave excitation at normal incidence. In contrast, in the experiment, the optical excitation is tightly focused to a spot diameter of 3 $\mu m$ in the photocurrent measurements and weakly focused to 30 $\mu m$ in the transmission measurements. Furthermore, the excited semiconductor metasurface region in the photocurrent measurements is adjacent to metallic contacts, with only approximately 5 x 5 metasurface unit cells between them. We therefore carried out numerical simulations using CST Studio transient solver of the full device to evaluate the impact of the focused excitation and metallic contacts on the optical response of the metasurface.

We first modelled the optical excitation as a Gaussian beam with a 3 $\mu m$ spot size with a 20 x 20 $\mu m$ region of metasurface and compared the metasurface absorption spectrum to the infinite periodic case. The two absorption peaks for the periodic case (orange line in Fig. 3(a)) merge and form a broader absorption band with a reduced peak value. Overall, the band is slightly blue-shifted. Secondly, we added metallic contacts to the model. The effect of the contacts splits the absorption band (green line in Fig. 3(a)) with the electric dipole mode peak appearing at 1530 nm and the magnetic dipole mode peak shifting to 1450 nm. An additional peak appears at 1630 nm. The latter is caused by an in-plane cavity mode formed between the metallic electrodes. Further discussion of each of the resonant modes can be found in the Supplement 1. The metallic contacts and focused beam excitation clearly affect the absorption spectrum, and therefore, the photocurrent response spectrum. The blue-shift in the simulations is consistent with the experimental results.

 

Fig. 3. a) Simulated absorbance for the finite size metasurface illuminated by a focused Gaussian beam, and the metasurface with THz antenna electrodes illuminated by a focused Gaussian beam. Simulated absorbance for an infinite metasurface (periodic boundary conditions) illuminated by a plane wave is shown by the orange line for comparison. Dashed vertical lines show approximate wavelengths of the electric (ED) and magnetic (MD) dipole modes. b-c) Normalised electric field distribution $E_y$ at 1540 nm ( 20 x 20 $\mu m$ area) for the metasurface without antenna electrodes excited with a 3 $\mu m$ Gaussian beam (b); for the metasurface with antenna electrodes at 1550 nm (c); for the metasurface with antenna electrodes at 1530 nm (d). The field distributions in (c) and (d) are normalized to the maximum field in (b). e) Normalised $E_z$ component of the electric field at 1630 nm showing the standing wave (SW) between the antenna electrodes. f) Simulated absorbance spectra for two electrode gaps: reducing the gap by 200 nm blue-shifts the SW mode by 70 nm from 1630 nm to 1560 nm.

Download Full Size | PDF

In addition, we evaluated the effects of the Gaussian beam and the metallic contacts on the optical field distribution within the metasurface. Figure 3(b) shows the field intensity distribution in the metasurface for the Gaussian beam at the peak absorption wavelength (1540 nm) predicted by the periodic simulations. The intensity distribution shows the characteristic electric dipole mode profile observed in the simulations with periodic boundary conditions, confirming that this mode is still observed despite the change in excitation source and boundary conditions.

Adding the metallic contacts, however, changes the field distribution and the peak field amplitude more noticeably (Fig. 3(c)). The reduced field amplitude in particular suggests reduced absorption at 1550 nm. Interestingly, the field distribution that closely resembles the distribution in the original metasurface design (periodic boundary conditions, Fig. 2(a)) can be now found at 1530 nm (Fig. 3(d)). We therefore deduce that the metallic contacts do not fundamentally change the modes supported by the structure: the electric dipole mode can be still recognized in the field profiles. However, the effect of the metallic contacts is significant for the optimal wavelength of excitation, and it can be approximately treated as a shift in the mode wavelength. Furthermore, there is an additional absorption peak observed at 1630 nm. The electric field distribution at this wavelength shows a clear standing wave localized between the antenna electrodes in Fig.3e. To confirm the standing wave nature of this mode, we reduced the gap spacing between the antenna electrodes and observed a blue shift of the resonance by 70 nm from 1630 to 1560 nm shown in Fig.3f.

The results of numerical simulations demonstrate the importance of including the full design of the active THz detector region when designing metasurfaces for integration with THz antenna electrodes. First, the focused beam excitation leads to peak broadening and reduced peak absorption compared to periodic simulations. Secondly, in the actual device, the metallic electrodes of the THz antenna further perturb the excitation field distribution in the metasurface: they split the magnetic and electric dipole modes and introduce an additional cavity mode due to standing waves formed between the electrodes. Accounting for these effects would ensure maximum absorption is obtained in the correct wavelength range of the pump laser [14].

5. CW THz detection

We tested the LT GaAs photoconductive metasurface devices as CW THz detectors using the optical heterodyne scheme with two continuous wave (CW) free running lasers. Since the photocurrent response spectrum (Fig. 2(c)) was stronger at the short wavelength side of the telecommunication band, we set laser 1 (Keysight N7776C) to a fixed wavelength emitting at 1520 nm and a tuneable external cavity laser (Santec TSL 710) with the operation window between 1510 and 1630 nm. To maximise the heterodyne beat note, a polarization controller was placed after each laser to set the polarization of both lasers to the same orientation. The two laser signals were combined using a 50:50 fiber coupler, as illustrated in the schematic diagram in Fig. 4, optical isolators (Luna Inc NISO 1550 nm) were placed before the couplers to prevent feedback to the tuneable laser.

 

Fig. 4. Schematic of the CW THz spectrometer for testing LT-GaAs THz metasurface detectors.

Download Full Size | PDF

A self-heterodyne configuration [28] was used to detect the THz signal: this configuration enables the recovery of the amplitude and phase of the transmitted THz signal using lock-in photocurrent detection. The optical frequency was shifted using a pair of electro-optic phase modulators and a differential sawtooth drive waveform at 1 kHz to generate a Serrodyne frequency shift [29,30]. The two phase modulators therefore translate the optical frequency modulation to the second harmonic of the sawtooth drive frequency (2 kHz), with a peak-to-peak drive voltage of 2$V_\pi$. After the frequency translation, an Erbium Doped Fiber Amplifier (EDFA) was used for both the transmitter and receiver optical paths to increase the optical power. As a THz source, we used a packaged Uni-travelling carrier photodiode (UTC-PD) integrated with a log periodic antenna (details of the UTC-PD were described in [31]). The EDFA output was directly connected to the UTC-PD. The THz beam from the UTC-PD was collimated and focused on the metasurface detector using two 50.8 mm diameter off-axis parabolic mirrors. The total path length of the THz beam in free space was 30 cm. A 4 mm hyper-hemispherical silicon lens was attached to the backside of the detector to improve coupling of the free-space THz beam to the THz detector bowtie antenna.

For the metasurface-based THz receiver, the second EDFA output was coupled from fiber to free space using a collimated fiber port, and then focused into the 3 $\mu m$ gap between the antenna electrodes using a 30x objective lens. To align the laser beam on the gap, a NIR camera was used to image the position of the laser spot relative to the antenna electrodes. The experimental arrangement is shown in Fig. 4. Firstly, we measured the dependence of the THz photocurrent on optical pump power at the peak output power of the UTC-PD source (120 GHz) [31]. The drive current of EDFA2 was swept to increase the optical power incident on the detector from 1 mW to 35 mW (measured using a free space optical power meter), while the amplitude of the THz photocurrent was recorded from the lock-in amplifier. We observed a super-linear but subquadratic dependence on optical pump power consistent with the two-step photon absorption process [14]. A peak THz photocurrent of 0.4 pA was observed for the highest pump power of 30 mW shown in Fig.5a. We limited the optical pump power at this level as one of the devices failed at a pump power of 40 mW; attributed to thermal degradation of the device.

We then characterized the THz spectral response of the detector by sweeping the heterodyne beat signal between laser 1 and laser 2. The peak dynamic range of 45 dB was observed around 0.12 THz corresponding to the peak output power of the UTC-PD [31]. Measurable THz power was observable up to around 1 THz where it dropped below the noise floor, which was measured by blocking the THz beam path with a metal sheet and performing an identical laser wavelength sweep.

We observed a dip in the THz detector response at 0.36 THz: this replicates a drop in THz power generation of the UTC-PD. This drop is due to the log-periodic antenna in UTC-PD which changes the THz emission polarization from horizontal to vertical within this band. The detector is fitted with a bow-tie antenna, which is sensitive only to the vertical polarization. Therefore, we observe a dip in the detector response at the frequencies where the source polarization is orthogonal to that of the detector antenna. An identical frequency response was observed by characterization of the UTC-PD using a commercial InGaAs THz detector, also integrated with a bow tie antenna [31]. The response of the InGaAs detector is plotted alongside the LT-GaAs device response in Fig. 5(b) for reference. We note that there are also periodic sharp dips in the spectrum. These features are due to the tuneable laser entering a multi-mode regime. We verified this by observing the laser emission at these points on an optical spectrum analyser. These spectral features develop more clearly in the high-spectral resolution scan that will be discussed later.

 

Fig. 5. a) Photocurrent vs. optical excitation power at 1550 nm. b) Normalized detected THz power showing a bandwidth up to 1 THz for the metasurface LT GaAs metasurface detector and a commercial InGaAs detector. c) RMS detector noise floor for the LT GaAs metasurface detectors and the commercial InGaAs detector (100 ms lock-in time constant). d) Fine resolution (500 MHz steps) spectrum of the 0.557 THz water vapor absorption line compared to the Hitran database model (the model spectrum is offset from the measured data by 5 dB for clarity).

Download Full Size | PDF

It is important to emphasize that the frequency response in Fig. 5(b) drops quickly at higher frequencies due to the THz UTC-PD source, rather than the metasurface detector. We characterized similar THz metasurface detectors using a THz-TDS system with 100 fs pulsed excitation and observed that the detector response peaks at 1 THz[20], the resonance frequency of the bow-tie antenna. We therefore expect that a THz source with higher power above 1 THz would enable a wider spectral window for the THz-FDS system.

Next, we compared noise in the metasurface detector to noise in a commercial InGaAs photoconductive receiver using the same experimental arrangement, as depicted in Fig. 4. The same operating conditions for both detectors were used in terms of transimpedance amplifier gain and offset frequency at 2 kHz the resulting RMS noise photocurrent is shown in Fig. 5(c). We observe approximately 2 orders of magnitude lower photocurrent noise for the LT-GaAs detector when compared with a commercial InGaAs photoconductive receiver due to the increased resistivity of the LT-GaAs material and the metasurface geometry. The main noise source in photoconductive antennas with a symmetrically pumped geometry is Johnson-Nyquist noise [15]. InGaAs, in general, has a low dark resistivity, which increases the noise level in the InGaAs photoconductive receiver. While there have been several approaches to mitigate this, for example using Beryllium doped InGaAs quantum wells with InAlAs trapping layers [32], or Iron doped [11,12] and Rhodium doped InGaAs [13], the dark resistivity of InGaAs THz detectors is still orders of magnitude lower than what can be achieved in LT GaAs THz receivers. For the LT GaAs detectors tested in this work, the dark resistance of was 15 $G\Omega$, compared to 1 $M\Omega$ in the commercial receiver [16].

Finally, to test the THz detector for spectroscopic analysis applications, we performed a high spectral resolution frequency sweep across the spectral line of atmospheric water vapor at 0.557 THz. An 8 dB dip in the detected power is clearly observed in Fig. 5(d) with a linewidth of around 4 GHz. This level of absorption agrees well with the numerically calculated absorption due to water vapour at similar temperature (27$^\circ C$) and humidity (RH = 62.5 %) conditions using the Hitran database parameters [33,34].

6. Conclusions

We have successfully demonstrated for the first time CW THz detection using photoconductive LT-GaAs metasurfaces with photoexcitation at 1550 nm. LT-GaAs was used for its superior optoelectronic properties: high dark resistivity, high carrier mobility and short photocarrier lifetime. We demonstrate that the detector showed two orders of magnitude lower noise than a commercial InGaAs photoconductive detector. A metasurface was designed to enhance the two-step photon absorption process, and we successfully demonstrate enhanced absorption of 1500-1550 nm light in the metasurface. Numerical modelling revealed the sensitivity of the metasurface spectral response to the presence of THz antenna electrodes, which can lead to a shift in the optimal wavelength for photoexcitation and a broadening of the absorption linewidth. Our numerical analysis demonstrates that THz antenna electrodes cannot be neglected in the design process. The experimental results and numerical analysis provide a route forward for a more advanced design to further optimize photoconductive LT GaAs metasurfaces for THz detectors operating with compact telecommunications band lasers.