1. Introduction

Superconducting nanowire single-photon detectors (SNSPDs) [1] are limited in detection efficiency primarily by optical loss due to the reflection from and transmission through the device. We have previously demonstrated a robust process for SNSPD fabrication [2, 3]. Here we demonstrate the integration of an optical cavity and anti-reflection coating (ARC) with these SNSPDs to overcome the problem of optical loss. Unlike previous work that used cavities (but not an ARC) to enhance detection efficiency in single-photon detectors, [4, 5], we have directly measured the enhancement on multiple devices tested before and after the integration of both the cavity and the ARC. The resulting measurements on fully-integrated devices demonstrate detection efficiencies as high as 67% at 1064-nm optical wavelength and 57% at 1550-nm wavelength.

Single-photon detectors with low jitter, high detection efficiency, high photon-counting rate, and low dark-count rate will enable a number of applications including: (1) quantum cryptography [6]; (2) integrated-circuit testing [7]; and (3) ultra-long-range optical communications [8]. While past work on SNSPDs have demonstrated sub-50-ps jitter [9], near-zero dark-count rates [10], and reset times less than 10 ns [3, 11], they have done so using detection efficiencies of only 17% at 1550-nm wavelength [12]. These detection efficiencies are no better than those of other (admittedly slower and higher-jitter) detectors such as InGaAs geiger-mode APDs [13]. The detection efficiencies demonstrated in our work are a key factor in enabling the adoption of SNSPDs as the single-photon-detector of choice for future high-performance single-photon optical systems.

The first step in the operation of a SNSPD is photon absorption: Fig. 1(a) shows a schematic cross-section of the device designed to optimize photon absorption in the NbN nanowire. Photons are incident onto the ARC, propagate through the sapphire substrate, and reach the niobium nitride (NbN) layer where they are reflected, absorbed, or transmitted. The nanowire is wound in a tight meander pattern to maximize its area of overlap with the optical beam. The area defined by the outline of the NbN meander, including the gaps between the wires, we will refer to as the active area of the detector. The NbN/hydrogen silsesquioxane (HSQ)/mirror structure forms a cavity, which is engineered so that the light reflected from the mirror interferes destructively with the reflection from the NbN/sapphire interface. By minimizing photon loss due to reflection and transmission that would occur without the cavity and ARC, the device exhibits an enhanced probability of absorption, P _a, for light incident within the detector’s active area.

 

Fig. 1. (a) Schematic cross-section of photodetector (not to scale) integrated with an optical cavity and anti-reflection coating (ARC) to reduce loss of photons from reflection and transmission. The device was illuminated from the back of the chip through the ARC and sapphire substrate. Photons that were not absorbed in the NbN wire at first pass entered the optical cavity, thereby having many chances to get absorbed by the NbN. The cavity thickness was chosen so that destructive interference reduced the reflectance from the NbN surface. (b) Transmission electron micrograph of cross-section of fabricated device with optical cavity. The cavity shown here was fabricated for calibration purposes and was thicker than those used to increase the detection efficiency. (c) Optical micrograph showing top-down view of optical cavity on photodetector. The outline of the detector is visible underneath the Ti/Au mirror due to a 9-nm-high step where the mirror goes from the substrate to the detector.

Download Full Size | PDF

The second step in the operation of a SNSPD is electrical detection. For each photon that is absorbed, there is another probability, P _r, that a voltage pulse will result. This probability is maximized by biasing the wire close to its critical current [1]. The voltage pulse can then be amplified and detected externally. Finally, the device cools and returns to its superconductive state.

We define the detection efficiency of a device, η, to be the probability that a photon incident in the active area will result in a voltage pulse, hence η = P _a P _r. We note that this definition neglects the light that is not incident on the active area of the detector.

2. Optical cavity and anti-reflection-coating fabrication

The process flow for fabricating an optical cavity on top of the detector is shown in Fig. 2. The process began with a fabricated photodetector [2] composed of a tightly wound meander structure with a 100-nm-wide wire and 100-nm spaces. This detector had an active area of 3.3 μm × 3.0 μm. First, HSQ was spin-coated on to the front of the detector and patterned to form the dielectric material of the cavity. Then titanium and gold were added to create a mirror for the optical cavity. Finally, HSQ was spin-coated onto the back of the device to act as an ARC. In this section, we detail this fabrication process.

 

Fig. 2. Process flow for integration of optical cavity and anti-reflection coating (ARC) onto photodetector. (a) The process began with a fabricated photodetector [2]. Note the residual HSQ etch mask left on the NbN wires from the photodetector fabrication process. (b) HSQ was spin-coated to achieve an optical cavity thickness of 195 nm and then electron-beam patterned. (c) 1 nm of Ti and 120nm of Au were evaporated onto patterned photoresist followed by lift-off, leaving the mirror for the optical cavity. (d) Finally, 277 nm of HSQ was spin-coated on the back side of the wafer and hardened in an O₂ plasma to form the ARC.

Download Full Size | PDF

The dielectric material used in the cavity was HSQ formulated as FOx-14 from Dow Corning. We chose HSQ as the cavity dielectric material for four reasons: (1) HSQ is a spin-on-glass and thus easy to deposit; (2) at infrared wavelengths the optical absorptance in HSQ is negligible [14]; (3) incorporating HSQ allowed us to simplify the cavity fabrication as we did not have to remove the remaining layer of HSQ from the previous process; and (4) HSQ was already used in the detector fabrication process, therefore, we knew it would not damage the device.

Figure 2 (b) shows the device with the cavity dielectric. First, the thickness of the HSQ remaining from the detector fabrication process was measured using atomic-force microscopy (AFM) and determined to be 45nm. We then spin-coated HSQ onto the sample at 8.6 krpm resulting in a total dielectric thickness of 195 nm on top of the detector active area¹. The samples were then soft baked at 90°C for 5 min to drive excess solvents out of the HSQ. Next, a layer of a conductive polymer (aquaSAVE [15]) was spin-coated onto the sample at 3.5 krpm to minimize charging during the subsequent electron-beam exposure. We used electron-beam lithography with 10 kV acceleration voltage and 300 μC/cm² exposure dose to expose the resist in a 40 × 40 μm² area centered at the detector. The sample was developed in 0.26 N TMAH for 8 min to remove unexposed HSQ. AFM imaging of the dielectric surface before mirror deposition revealed a peak-to-peak surface roughness of only 1 nm over the active area of the detector despite the underlying topography.

Figure 2 (c) shows the device after the addition of the cavity mirror. A metal liftoff process was used to deposit this mirror. This liftoff process was similar to the one used for depositing gold contact pads in the detector fabrication [2] except that 1 nm of Ti, rather than 10 nm of Ti, was deposited to promote adhesion of the reflector to the dielectric. The Ti thickness was minimized to optimize cavity performance: at IR wavelengths, gold has a greater perpendicular reflectivity than titanium [16]. Calculations revealed that the thin layer of titanium reduced the absorptance of the cavity by less than 0.1% relative to a gold only mirror that was 120 nm thick. The resultant mirror size was 14 μm × 22 μm. A transmission electron micrograph of the cross section of an optical cavity on top of a device is shown in Fig. 1 (b). An optical micrograph of the completed cavity structure is shown in Fig. 1 (c).

We used a single-layer ARC on the bottom surface of the sample to reduce optical loss due to reflections at the vacuum-sapphire interface. HSQ was chosen as the ARC in part due to its low index of refraction of 1.4 at 1550 nm wavelength [14]. This index is close to the 1.32 needed for a λ/4 thickness ARC that ideally couples light from vacuum into sapphire, which has an index of 1.74 [17]. For light at 1550 nm wavelength, 277nm of HSQ is required for a λ/4 ARC.

Figure 2 (d) shows the devices after the addition of an ARC. The ARC was applied to the back of the device by mounting the sample top-side down onto a silicon handle wafer with two drops of photoresist. The handle-wafer-and-sample pair were then soft-baked at 90°C for 15 min. This mounting method prevented physical damage to the devices on the top surface of the chip during the addition of the ARC. 277 nm of FOx-14 was added to the pair by spin-coating at 2.8 krpm and then baking for 20 min at 90°C. Next, the HSQ ARC was oxygen plasma treated for 10 min to form a hardened surface. The plasma treatment conditions were 100 W power and 2.8 kPa pressure of 80%He/20%O₂. Lastly, the sample was released from the handle wafer with a 2-min acetone soak, leaving a robust ARC.

3. Testing setup

The setup used to measure the electrical and optical response of the photon-counting devices is shown in Fig. 3. This setup included three major subsystems: (1) the cryogenic probing station; (2) the room-temperature electronics; and (3) the pulsed optical source. In this section, we describe each of these subsystems.

 

Fig. 3. Schematic of the experimental setup with the optical components on the left connected by optical fiber and the electrical components on the right connected by coaxial cable. The fiber probe illuminated the sample through the back side of the sapphire substrate while the RF probe contacted the gold contact pads on top of the NbN.

Download Full Size | PDF

First the sample was placed in the cryogenic probing station. Our sample consisted of an 8-mm-square chip, containing 154 individual devices. The cold head of the probing station was cooled to as low as 1.8K. The sample was attached using silver paint to a gold-plated copper sample mount that allowed optical access to the devices through the back of the substrate and electrical access from the front. Electrical contact with the devices was established using a 65-GHz RF probe connected to a coaxial cable and mounted on a micromanipulator arm, so that it could be touched down to any individual device on the sample. The probe was cooled to <30K using copper braids connected to the 4.2K stage of the probing station. An optical fiber and lens assembly was mounted to a second cooled micromanipulator, whose position was controlled using an automated, closed-loop three-axis positioning stage with submicron resolution (MICOS GmbH). This stage allowed the optical spot produced by the lens to be aligned automatically with any device using that device’s photon count rate as feedback.

Our room-temperature readout-electronics subsystem was connected to the cold RF probe through coaxial cable and a vacuum feedthrough. First, a 0.5 m length of coaxial cable was used to provide a delay that temporally separated any spurious electrical reflections from the output pulses of our devices. This cable was connected to a bias tee. Current bias was supplied to the devices through the DC port of the bias tee using a battery-powered voltage source in series with a 100 kΩ resistor. The AC port of the bias tee was connected to two cascaded wideband, low-noise amplifiers (MITEQ JS2-00100400-10-10A, 27 dB gain, 0.1-4 GHz) through a 3 dB attenuator. Without the isolation provided by this attenuator, we observed that the critical currents of our devices were suppressed by up to 10%; this was likely the result of electrical noise associated with reflections at the amplifier input (VSWR 2:1).

The amplifier output was sent through a DC block and split using a resistive splitter to be sent into both a photon counter and a 6-GHz, real-time oscilloscope. Without the DC block on the outer conductor, we again observed that the critical currents of our devices were suppressed, this time by up to 20%, likely due to ground noise. For the detection-efficiency measurements, the photon counter was used to count the number of electrical output pulses resulting from a fixed number of highly attenuated (<0.25 incident photons per pulse) optical pulses. The signal-to-noise ratio of the amplified voltage pulses was sufficiently high that the discriminator threshold level at the input of the counter could be varied over a wide range without changing the observed count rate. In order to reject counts not directly associated with the optical excitation, the counter was operated in a gated mode in which only those pulses arriving within a fixed 5-ns-long window were counted. This window was centered on the arrival time of detection events generated by the optical pulses by triggering the gate synchronously with the laser’s output pulses, and then adjusting the gate delay. In addition, a dark-count-noise baseline was taken of counts recorded in the gate window with the light blocked mechanically. This baseline was subtracted from each measurement.

Within the optical subsystem, the devices were illuminated with light from a laser that generated pulses at a 10 MHz repetition rate. This light was attenuated to the single-photon level using a precision optical attenuator and calibrated using an InGaAs power meter at power levels 30 dB above its noise floor. Varying the polarization of the light using a fiber polarization controller resulted in a change in the count rate. The polarization was set to maximize the count rate for all detection-efficiency measurements. All of the optical components were fiber-coupled and the measured losses were stable to less than 0.05 dB. The loss between the fiber connector at the input to the probing station and the output of the lens assembly was separately measured (at room temperature) using a free-space InGaAs power meter at a power level of 100 μW. Finally, the size of the optical spot was measured. Since the position of the lens could change as the probe is cooled, this spot size had to be measured in situ, at low temperature. To do this, we translated the optical probe while measuring the device count rate and mapped out the resulting profile, which consists of the convolution of the Gaussian beam profile and the rectangular active area of the detector. By fitting the measured profiles, the Gaussian beam waist could be extracted, and from this, the fraction of the total power incident on the device’s active area was determined. Detection efficiencies were measured using several spot sizes ranging from 20 μm to 90 μm, yielding consistent results.

4. Results and discussion

Here we present the results of measurements of 132 of these devices from one of these chips where measurements were made three times on each device: once after the initial device fabrication, once after the optical cavity was added, and finally after the ARC was added. This procedure enabled us to determine the detection efficiency of devices as well as the enhancement due to the cavity and ARC. These measurements were made at 1550-nm wavelength and a temperature of 1.8K. Additionally, measurements were also made at 1064 nm and 4.2K. Lastly, the timing jitter of the devices was also determined.

 

Fig. 4. Histogram of detection efficiencies for 132 tested devices on a chip measured (a) after initial fabrication of the bare photodetectors (b) after addition of the cavity structure and mirror on the devices and (c) after an anti-reflection coating was added to the back side of the sapphire substrate. Each measurement was made at 1.8K and at I = 0.975I_c.

Download Full Size | PDF

The results of detection-efficiency measurements made at 1550-nm wavelength are shown in Fig. 4. These histograms show the distribution of detection efficiencies measured for each device at 97.5% of its critical current. As shown in Fig. 4 (c), the highest detection efficiency observed was 57%. After the integration of both an optical cavity and ARC, fewer than 10 of the 132 devices had a detection efficiency below 20%; the median and mean detection efficiencies were 47.7% and 44.3%, respectively. This yield is clearly sufficient to fabricate large numbers of high-efficiency devices and even small (∼4-element) arrays of these devices.

The enhancement due to the addition of a cavity and an ARC was determined with a subset of the total devices tested at 1550 nm. Measurements taken over three days with the same sample under vacuum but cycled (three times) up to room temperature showed virtually no changes in the critical currents (<2%) or detection efficiencies (<4%). However, larger changes were observed for some devices over longer time scales (months) and in between processing steps.

In nearly all cases, a strong correlation was observed between changes in critical current and changes in detection efficiency; to distinguish the impact of the added optical elements from any changes in the detection efficiency that were correlated to changes in the critical current, we quantified the effect of adding the resonator and AR coating using only the subset of devices whose critical currents were nearly unchanged between processing steps. For the addition of the cavity, we obtained an enhancement factor of 2.44±0.005, based on the 21 devices whose I _c changed less than 3%, and whose initial DE was greater than 13%². The addition of the ARC provided an additional enhancement factor of 1.050±0.009, based on the 105 devices whose I _c changed less than 3% in this step of the processing.

To determine the detection efficiency at 1064-nm-wavelength, the five devices with the highest measured detection efficiency at 1550 nm were tested after the addition of the optical cavity and ARC. The highest detection efficiency observed was 67%. This result demonstrates the utility of our devices at another technologically important wavelength.

The detection efficiencies reported were measured at an operating temperature of 1.8K. Increasing this temperature resulted in a slight degradation of device performance. Specifically, at the easily accessible temperature of 4.2K the highest observed detection efficiency was 30% at 1550-nm wavelength and 57% at 1064-nm wavelength.

Lastly, to verify that previously reported excellent jitter performance [9] for these devices was unaffected by the cavity, the timing jitter of the electrical pulses relative to the optical excitation was determined. This experiment was carried out using 1-ps-wide pulses from a passively mode-locked 1550-nm source by measuring the time interval between optical excitation and electrical pulse response, using the scope to trigger on the sharp leading pulse edge at a fixed voltage threshold, for a large number of pulses and producing a histogram. This distribution was measured to be 41 ps FWHM for a device with an optical cavity at 1.8K. This result suggests that, as expected, the timing jitter of these devices is not degraded by the addition of a very short, low-Q resonant optical structure.

We note that observed results may be improved upon by increasing the thickness of the NbN film while decreasing the width of the nanowires or by increasing the fill-factor of the devices. Detection efficiency can also be increased by further optimization of the optical elements of the detector. We plan to pursue future work along these lines.

5. Conclusions

This paper reports a 57% detection efficiency of superconducting nanowire single-photon detector at 1550-nm wavelength and a 67% detection efficiency at 1064-nm wavelength through the integration of an optical cavity and ARC. In addition to this peak detection efficiency, we have found a median detection efficiency at 1550 nm of 47.7% over greater than 130 devices.