1. Introduction

Since the discovery of superconductivity in 1911, it attracts great research interest and finds abundant and irreplaceable applications [1], especially in weak photo signal detection [2]. Compared with other photodetection technologies, superconducting photodetection offers an extremely wide band coverage over the entire electromagnetic spectrum, ranging from the very low frequency microwave to very high frequency X-ray [3,4]. Remarkably, in the short wavelength range, ultimate sensitivity to single photon level can be achieved. As an example, one most striking superconducting detector in the past decade is superconducting nanowire single photon detectors (SNSPDs) [5,6] owing to their outstanding performance, which benefit from the superconducting to normal conductance phase transition. The recently developed SNSPD exhibit high system detection efficiency (SDE) close to unity at near infrared [7], enabling numerous application such as optical quantum computation [8] and quantum key distribution [9,10]. However, in the longer wavelength infrared region, the SDE is quite low, which is for example lower than 3% at 5$\mathrm{\mu m}$ in [11]. The SDE can be decomposed as [12] SDE=${\eta _{tansmi}}{\eta _{couple}}{\eta _{absorb}}{\eta _{QE}}$, where ${\eta _{tansmi}}$ is the transmission, $\textrm{}{\eta _{couple}}$ is the geometric coupling efficiency, ${\eta _{absorb}}$ is the absorption efficiency and ${\eta _{QE}}\textrm{}$ is the internal quantum efficiency. Since the internal quantum efficiency dropped significantly when the wavelength went beyond about ∼3$\mathrm{\ \mu m}$ [11,12], extensive effort therefore had been made to improve internal quantum efficiency in the longer wavelength range [13,14]. In these recent works [13,14], the internal quantum efficiency had reached almost unity at 5$\mathrm{\ \mu m\ }$ for Mo_0.8Si_0.2 [13] and at 10$\mathrm{\ \mu m}$ for WSi [14]. On the other hand, however, the SDE was still very low [12–14] due to very weak optical absorption, for instance, only 5% at 5$\mathrm{\ \mu m}$ [13]. Therefore, improving the optical absorption is highly demanded in order to improve SDE at longer wavelength. In the near infrared, many methods have been implemented to improve the absorption efficiency such as nanoantenna [15], optical cavities [16] and perfect metamatrials absorber [17]. In the mid-infrared, however, very few works [18,19] have attempted to improve the coupling efficiency, in particular, using the metamaterial absorber. What’s more, these works [18,19] are limited to the enhancement of single band photodetection despite of the urgent demand of dual- or multi-color photodetection.

In the infrared spectral range, dual color photodetection turns out to be very useful for many practical applications since it can identify targets in a complex environment by processing signals from two different wavebands [20,21]. For instance, this technology can accurately detect absolute temperature of targets with unknown emissivity [22] and hence enable widely application in thermal image of accurate temperature distributions. Alternatively, it can measure atmospheric species such as NO₂ by using a dual color infrared detector without the bulky, complex optical component, thereby minimizing size, cost and complexity [23]. As a result, dual color photodetection constitutes an important branch in the infrared optoelectronics. So far, however, very few works on superconducting detectors have addressed dual color photodetection [24].

In this paper, a dual color superconducting detector with nearly perfect absorption in the mid wavelength infrared (MWIR) and long wavelength infrared (LWIR) is realized by utilizing the surface plasmon resonance of metamaterials and the Fabry-Perot (F-P) like resonant cavity mode. We demonstrate that the peak responsivity reaches ∼1.2 × 10⁶V/W at 36.6 THz and ∼3.2 × 10⁶V/W at 104 THz at near ${T_C}$ (8 K), which are both enhanced appreciably compared to that of non-resonance frequency. Our work provides an innovative way to improve the light coupling/harvesting efficiency of multi-color superconducting detector in MWIR/LWIR, which may find important applications in very sensitive thermal image and gas sensing etc. In particular, the dual bands at 36.6 THz and 104 THz in our work correspond respectively to the C-N bond in amine and the common function group of OH bond.

2. Detector design and fabrication

Our designed dual color MWIR/LWIR niobium nitride (NbN) superconducting detector consists of three layers as shown in Fig. 1(a), in which the bottom layer is a 100 nm thick continuous niobium (Nb) reflector and the top one is a 20 nm thick NbN patterned metamaterial, with a 600 nm thick silicon (Si) dielectric spacer in between them. The substrate is Si with 285 nm thick of silicon oxide layer. The thickness of the Nb reflector is greater than the penetration depth of the target light, implying that light transmission is practically zero. Therefore the device’s absorption (A) is nearly equal to A = 1-R, where R is the reflection. All of these materials were deposited on the substrate by magnetron sputtering. The device pattern is then created by electron beam lithography (EBL) and reactive ion beam etching (RIE). The scanning electron microscope (SEM) images of the actual device in Fig. 1(b) and (c) show a periodic cross metamaterial structure with length L = 1.2$\mathrm{\ \mu }$m, width W = 0.6 $\mathrm{\mu }$m, period P = 1.6 $\mathrm{\mu }$m and a total of 64 periods. This structure constitutes a polarization-independent perfect metamaterials absorber in the infrared [25–27]. The electrical connection between all neighboring cross structures is realized using the Hilbert fractal curve with a linewidth of 0.1 $\mathrm{\mu }$m, in order to reduce polarization sensitivity [28]: These fractal curves would eliminate the orientation of the nanowire on the global view of the device, making absorption and photodetection remain polarization-independent. The first-order Hilbert fractal curve can connect 2 × 2 cross structure. Specifically, the geometry of the nanowire schematically shown in Fig. 1(a) is a second-order Hilbert fractal curve with 4 × 4 cross structures connected in series. Our actual device is a sixth-order Hilbert fractal curve with 64 × 64 cross structure.

 

Fig. 1. (a) Illustration of a structure of a dual color NbN device. (b)(c) SEM microscopic image of dual color NbN device. White color scale bar 20$\mathrm{\mu }$m. Red color scale bar 1$\mathrm{\mu }$m. (d) The resistance-temperature curve of device with superconducting phase transisition at ${\textrm{T}_c}$. = 8.8 K

Download Full Size | PDF

3. Dual color perfect absorption

Our dual color device is targeted at transparency atmospheric window in LWIR (8∼13 $\mathrm{\mu }$m) and atmospheric molecule infrared spectroscopic fingerprints in MWIR (2.5∼3 $\mathrm{\mu }$m), such as H₂O, N₂O and CO₂ [29]. The transparency window of the atmosphere at 8∼13$\mathrm{\mu }$m has a large spectral overlap with the blackbody radiative spectrum at typical ambient temperatures near 300 K [30], resulting promising application for thermal image at this range. Taking the advantage of the high tunability of metamaterial, the perfect absorber structure here was designed to enhance the light coupling/harvesting efficiency simultaneously in LWIR and MWIR. Finite-difference time-domain (FDTD) method was implemented for the simulation in which we set the permittivity of Si $\mathrm{\varepsilon }$=11.9, and the permeability constant $\mathrm{\mu }$=1. The permittivity of Nb was adopted from reference [31]. The permittivity of NbN was extracted from the fitting curve of reflection spectrum of 200 nm thick continuous NbN film with Kramer-Kronig (K-K) transform in RefFit software [32]. Experimentally, the reflection spectrum was measured by Fourier transform infrared spectrometer (FTIR, Bruker HYPERION 1000). As shown in Fig 2(a), both the calculated and measured absorption spectra show two resonance peaks in LWIR and MWIR. The absorption reaches near unity at two resonant frequencies. The slight difference in resonant frequency between simulation and experiment may originate from the additional Hilbert fractal curve and/or structural imperfection of fabricated device. The resonant wavelength can be approximately estimated according to F-P like mode with the equation $\mathrm{\lambda } = 4\textrm{nd}/({2\textrm{k} - 1} )$, where n, d and k are the refraction index of Si, thickness of Si and the order number, respectively. In the LWIR range, the resonant frequency ${\textrm{f}_1}\sim 36\,\textrm{THz}$ corresponds to the lowest order of FP-like mode (k = 1, $\textrm{d} = \mathrm{\lambda }/4\textrm{n}$) being hybridized with local surface plasmon of metamaterial [33]. Figures 2(b) and 2(c) display the x-y and y-z plane E_z/E₀ electric field distributions in LWIR band (${\textrm{f}_1}\sim 36\,\textrm{THz}$). From Fig. 2(c), it can be seen that the electric field of LWIR light is primarily confined on the top NbN surface, with maximum E_z field magnitude reaching over 150 times that of incident E₀ field. In the MWIR range, the resonant frequency ${\textrm{f}_2}\sim 107\,\textrm{THz}({\textrm{}{\mathrm{\lambda }_2} \approx 2.8 \mu m\,\textrm{in wavelength}} )$ corresponds to k = 2 ($\textrm{d} = 3\mathrm{\lambda }/4\textrm{n}$). Figures 2(d) and 2(e) show that the MWIR resonant electric field is distributed between the top and bottom metals and one node exists in the E_y-profile as depicted in the right cartoon inset of Fig. 2(a).

 

Fig. 2. (a) The simulation (black dotted-line) and experiment (black circle dot-line) absorption spectra at room temperature. Insert The different orders FP-like cavity mode in our dual-color NbN device (b)(c) The simulation x-y and y-z plane LWIR resonant E_z/E₀ electric field distributions. The electric field is primarily confined on the NbN surface. (d)(e) The simulation x-y and y-z plane MWIR resonance E_z/E₀ electric field distributions.

Download Full Size | PDF

It is noteworthy mentioning that the reported traditional gold antenna design [19] has the apparent polarization-dependence and can enhance the photoresponse only for one polarization. Also the single-antenna-layer design in conventional work cannot suppress the transmission of the incident light and therefore limits the absorption efficiency. While in our work, the tri-layer design with the bottom mirror layer suppresses the transmission down to zero and meanwhile the reflection is suppressed to almost zero at resonant wavelengths. As a result, coupling efficiency approaches the ultimate performance so that all energy of incident photons can be absorbed to contribute to the (bolometric) photoresponse of our device.

4. Detector characterization

The electrical and photoelectrical properties were tested in cryostat (PHYSIKE: Qcryo-Scryo-S-300) with the base temperature <3 K. A four-point probe method measurement with ${\textrm{I}_{Bias}} = 1\textrm{}\mu A$ was performed to determine the superconducting phase transition temperature (T_C) of the device, which is 8.8 K as indicated in Fig. 1(d). Below this ${\textrm{T}_C}$, the current-voltage (I-V) curve and photo-response were acquired simultaneously at two representative temperatures, 6 K and 8 K. The schematic of measurement setup was show in Fig. 3(a). The photo-response was generated under the excitation of a 331 Hz-chopped quantum cascade laser (QCL) with tunable frequency (23 THz∼58 THz) and detected by a lock-in amplifier.

 

Fig. 3. (a) Schematic drawing of the experimental setup for photoresponse and IV measurement. (b) The I-V curve shows hysteresis at 6 K. The black curve and red curve represent the current ramp-up and ramp-down, respectively. (c) Responsivity plotted against the bias current at 6 K.(d) The I-V curve at 8 K. (e) Responsivity plotted against the bias current at 8 K (f) The temperature and current dependence of responsivity for current ramp-up. The scatter black line and scatter red line represented the critical-current and retrapping current, which were fitted by using GL theory (scatter black line) and self-heating hotspot model (scatter red line), respectively. (g) The temperature and current dependence of responsivity for current ramp-down.

Download Full Size | PDF

Depending on the operating temperature being either lower (6 K) or close (8 K) to ${\textrm{T}_C}$, the I-V curve and photo-response show remarkably different behavior. At low temperature 6 K (Figs. 3(b) and 3(c)), clear hysteresis can be seen with sweeping the bias current, while no apparent hysteresis can be discerned at near critical temperature 8 K (Figs. 3(d) and 3(e)). As a result, a bistable region is formed at 6 K in the bias current range 9-14 $\mu A$, as indicated by the grey region in Figs. 3(b) and 3(c). Accordingly, the photo responsivity exhibits different peaks depending on current sweeping direction in the gray region.

When ramping-up the current from 0 → 20 $\mathrm{\mu }$A, two separate peaks occur at different bias currents, as indicated by black line in Fig. 3(c). The first peak occurs at a bias current (${\textrm{I}_C}$) that just precedes the onset of a finite resistance in the I-V cure. Delacour et al also observed the appearance of two photo response peaks with current bias sweeping from zero to over critical current [34]. In our case, the first peak is consistent with the exponential dependence of the photo-response upon current bias, a feature usually be attributed to phase slip mode [34–36]. Therefore, the photo detection mechanism for the first peak can be ascribed to this phase-slip model, that is, (multiple-)photons destroy large quantities of Cooper pairs and then reduce the free-energy barrier of phase slips, thereby resulting in proliferation in the phase-slip events and eventually leading to the formation of sufficiently large resistive hotspots in the weak-link region of our fractal structure. Above the critical current (${\textrm{I}_C}$), significant portion of the device switches from superconducting state to resistive state, and at the second peak where resistance increases sharply with the current, a small temperature change will cause a large change in resistance. This second peak is therefore similar to the working point for superconducting bolometer detectors, such as transition-edge sensors (TES) [37]. In this resistive state, if the current is increased further, all the Cooper-pair will be annihilated and the device will turn into the normal state and hence will become less sensitive in photo response.

In contrast, during the current ramp-down from the normal state to superconducting state, the global Joule heating and the subsequent temperature rise of the sample require a significant reduction of applied current before the sample goes back to the superconducting state, leading to the observed hysteresis (Fig. 3(b)). This hysteretic behavior has been previously observed [38] and the physical reason behind is the quantum phase-slip process governs the local resistance of the NbN wire at a given current I and local temperature T, but classical heat flow determines the local temperature distribution generated by this heating by phase-slip processes. During the current ramp-down, the responsivity first increases until the retrapping current I_r (∼9 $\mu A$) and then decreases dramatically in accordance with the resistance transition. As a result, only one peak appears in the current back scan of 0 $\mathrm{\mu }$A ← 20 $\mathrm{\mu }$A. For the near-phase transition temperature (8 K, Figs. 3(d) and 3(e)), large portion of the device lies in the resistive state, and the hysteresis vanishes. Accordingly, only one same responsivity peak occurs for both current ramp-up and ram-down sweep, showing also no hysteresis. We note that the hysteretic current-voltage curve is undesirable for the photodetection since it will lead to unstable photo response. This hysteresis can be suppressed by introducing shunt resistance or optimizing the working temperature.

Figures 3(f) and 3(g) display the two-dimensional mapping of responsivity, as a function of operating temperature and bias current. During the current ramp-up from the superconducting state, the increasing current induced an increasing magnetic field. When the induced magnetic field exceeds the critical magnetic field, the superconducting state will be destroyed. Within Ginzburg-Landau (GL) theory one can calculate the expected temperature-dependence for the depairing-critical current [39,40]${I_C} = {I_C}(0 ){({1 - {t^2}} )^{3/2}}{({1 + {t^2}} )^{1/2}}$, where $\textrm{t} = \textrm{T}/{\textrm{T}_C}$, ${\textrm{T}_C}$ is superconducting phase transition temperature, ${I_C}(0 )$ is critical current at zero temperature. As shown in Fig. 3(f) the black line is a fitting curve from which we estimate ${\textrm{T}_C}$=8.7 K, ${I_C}(0 )$=27 $\mathrm{\mu }$A. This justifies the photo-response peak in the current forward scan ( 0 $\mathrm{\mu }$A ← 20 $\mathrm{\mu }$A.) and suggests the underlying mechanism to be GL depairing with multiple-photon-induced phase-slip events. For the current backward scan from the normal state to the superconducting state, the photo-response increases until reaching the retrapping current which can be described by self-heating hotspot mode [39] ${I_r} = {I_r}(0 )\sqrt {({1 - t} )} $, with the fitting parameters ${I_r}(0 )$ = 17 $\mathrm{\mu A}$, ${\textrm{T}_C}$=8.7 K. This implies that the photo-response in the backward scan (0ß20 $\mathrm{\mu A}$) is dominated by the Joule heat generation. It is interesting to notify that, as temperature increases, ${\textrm{I}_C}$ and ${\textrm{I}_r}$ cross at the temperature T_h (∼6.9 K). When the operating temperature is lower than T_h, there is an obvious hysteresis in the region of ${\textrm{I}_r} < I < {\textrm{I}_C}$, with distinct response mechanisms depending on current sweep direction (Fig. 3(c)). When the operating temperature exceeds T_h, the hysteresis disappears and the photo responsivity becomes independent of current sweep direction (Fig. 3(e)). In this case, since ${\textrm{I}_C}$<${\textrm{I}_r}$, GL depairing takes place just below the retrapping current. As a result, the depairing induced local hotspot and the global self-heating (like TES) mechanisms contribute cooperatively to the photo response when the biasing current is in the range of ${\textrm{I}_C} < I < {\textrm{I}_r}$ and hence make the photo responsivity the highest.

Figure 4 shows the spectral response of our device measured by an FTIR step scan system at 8 K. Two responsivity peaks are observed in MWIR (104 THz) and LWIR (36.6 THz) respectively, which agree well with the measured and calculated absorption spectrum (Fig. 2). Compared to the nonresonant response (∼67 THz), the peak responsivity is enhanced about ∼8 and ∼22 times, respectively, in the LWIR and MWIR. Comparing the photo response at two resonances, the responsivity in MWIR (104 THz) is about 2.8 times higher than that in LWIR (36.6 THz), with the photon energy ratio of 104 THz$/$36.6 THz∼2.7 between MWIR and LWIR photons. Noting that absorption is near unity at both MWIR and LWIR, the intrinsic detection efficiency for each photon breaking Cooper pair is about ∼2.8${\times} $2.7 = 7.56 times higher for MWIR photons than for LWIR photons, as well expected for the hotspot spreading photodetection model.

 

Fig. 4. Photo responsivity spectra of dual-color NbN device measured by FTIR step scan system at 8 K.

Download Full Size | PDF

Figure 5 displays the response time of the device under the irradiation of LWIR resonant light (36.6 THz, Fig. 5(a)) and non-resonant light (57 THz, Fig. 5(b)), respectively. It can be seen that the response time to resonant light is much faster (∼0.3 ms) together with a much larger amplitude of photo-induced voltage than that of non-resonant light (∼20 ms). The fast resonant response may arise from the large local electromagnetic field enhancement in top NbN photoactive layer as shown in Figs. 2(b) and 2(c) while non-resonant electromagnetic field is more dispersed in the entire cavity.

 

Fig. 5. (a) The photo-induce voltage time trace under the irradiation of LWIR resonance light (36.6 THz). (b) The photo-induce voltage time trace under the irradiation of non-resonance light (57 THz)

Download Full Size | PDF

Since thermal Johnson noise dominates typically in a superconducting bolometer-based detector working at temperatures near ${\textrm{T}_C}$, the noise equivalent power (NEP) can be calculated according to $\textrm{NEP} = \frac{{\sqrt {4{k_B}TR\varDelta f} }}{{{R_V}}}$, where, ${k_B}$,T, R, $\varDelta f$, ${R_V}$ is Boltzmann constant, temperature, resistance, measurement bandwidth and photo responsivity, respectively. We evaluate the minimum NEP of our device is 6.4$\textrm{} \times {10^{ - 15}}W/\sqrt {Hz} $ at 8 K with a typical bias-current of 3$\mathrm{\ \mu }$A. The normalized detectivity ${D^\mathrm{\ast }} = \frac{{\sqrt {A\varDelta f} }}{{NEP}}$ of our detector is thereby calculated to be 1.1${\times} $10¹² $\textrm{cm} \cdot \sqrt {Hz} /W$ using the real detector area (A∼100 $\mathrm{\mu }$m x 100 $\mathrm{\mu }$m). This ${D^\mathrm{\ast }}$ value is readily 3∼4 orders of magnitude higher than conventional MCT detectors and can be further improved by, for example, optimizing the film quality of top NbN layer.

5. Conclusions

In conclusion, we have demonstrated a dual color LWIR/MWIR superconducting detector with near perfect absorption at 36.6 THz and 104 THz, which were attributed to hybridized resonance between local surface plasmon and Fabry-Perot cavity like mode. The responsivity reaches highest at the working temperature of 8 K which is slightly below ${\textrm{T}_C}$∼8.8 K, under the cooperative contribution of GL depairing induced local hotspot and the global Joule heating photodetection mechanisms. The photo responsivity at the dual color resonant frequencies (36.6 THz and 104 THz) is roughly 8 and 22 times that of non-resonant frequency (67 THz), respectively. Our work provides a new strategy to improve the sensitivity of superconducting photodetectors in infrared and terahertz spectral range in multispectral applications.