1. Introduction
Superconducting nanowire single-photon detectors (SNSPDs) have outperformed conventional semiconductor single-photon detectors in terms of detector performance metrics and have been widely used, especially in quantum information [1]. Large-scale SNSPDs with efficient active areas and large numbers of pixels promise high-efficiency photon detection and spatial and temporal resolution capability in photon-scarce environments, such as in dark matter detection [2], single-photon lidar [3], and astronomical detection [4]. Recently, significant efforts have been made to study large-scale SNSPD arrays. Many readout approaches have been proposed to reduce the number of radio frequency (RF) cables and thus the heat load to the cryocooler, such as row-column readout architecture [5], [6], frequency multiplexing [7], time-of-flight multiplexing [8], and digital multiplexing with a single-flux-quantum circuit [9]. These works have enabled SNSPD arrays to work at kilopixel levels and promise the readout of a larger number of pixels. Meanwhile, the active area was increased from tens of microns to the millimeter scale, enabling efficient light coupling from single-mode fiber to multi-mode fiber or even free space.
Among multiple multiplexing methods, thermally coupled multiplexing (TRC), which uses thermal coupling between two active SNSPD layers with nanowire pixels serially arranged in rows and columns, has many potential advantages. Since the pixel channels are electrically isolated at low frequencies, there will be no current redistribution and loss of electrical signal, which occurs in the electrically coupled row-column architecture [6], [12]. By measuring the coincidence events between row and column channels, the location of photon absorption can be determined by the intersection of those channels, forming a spatial pixel. Furthermore, compared with conventional one-layer SNSPDs, the orthogonalizing bilayer detector can provide a higher detection efficiency and reduce polarization dependence [10]. Depositing two layers of superconducting nanowires in parallel can increase the thickness of the absorbing layer, balance the relationship between light absorption and saturated quantum efficiency, and make it easier to achieve saturated detection efficiency [11].
Along with the technological development of the kilopixel SNSPD array, the material selection of the large-scale array remains a fundamental block that limits the system's performance and complexity. Amorphous superconducting materials with lower critical temperatures, such as WSi and MoSi, are often preferred for large-active areas and high yields of multiple pixels because of material uniformity, which, however, requires sub-Kelvin operating temperatures and suffers from low detection speeds and timing resolutions. Meanwhile, SNSPDs made of polycrystalline superconducting films with high Tc and working temperature (e.g., NbN and NbTiN) are also popular for their high switching current and, thus, high timing resolution [13] and detection speed [14]. Moreover, they have shown near-unity efficiency [11], large-active single-pixel SNSPDs [15], and ultra-large area superconducting nanowire array [16]. Our recent work also applied NbN detectors to infrared photon detection at 10 μm with an optimization of the NbN film [17]. These previous achievements suggest a large-scale SNSPD array that may operate at 2.3 K provided by a compact cryocooler.
Here, we report an NbN SNSPD array composed of 32 × 32 nanowire elements arranged in two layers with a SiO2 layer as a spacer. The bottom and top layer pixels were thermally coupled, with each layer pixels electrically connected in rows and columns, respectively. The unique 3D structure constitutes a detector array of 32 × 32 spatial pixels, covering an area of 0.96 mm × 0.96 mm, where the spatial position of the incident photons are determined by the coincidence count of row (x) or column (y) arranged nanowire pixels. Our array characterization at 2.3 K provided by a compact Gifford-McMahon (GM) cryocooler showed a pixel yield of 94% and maximal intrinsic efficiencies of 77% and 96% at 1550 nm and 405 nm, respectively.
2. Row-column multiplexing principle
The thermally coupled row-column multiplexing architecture was employed in our detector array. Fig. 1 shows the schematic of the array composed of an NbN/SiO2/NbN three-layer stack structure. Both the top and bottom NbN layers have 32 × 32 nanowire pixels, which are row or column connected. The nanowire channels in the row and column are then referred to as R1-R32 and C1-C32, respectively. The SiO2 layer was sandwiched between these two NbN layers and functions as an electrical spacer layer while allowing thermal coupling between the two layers. When a photon impinges on a certain layer (either the top or bottom layer) and triggers the nanowire pixel, the thermal phonons generated will pass through the spacer layer and trigger the pixel in the other layer. Then, the incident photon position can be determined by the pulses generated in the row and column channels [5].
Fig. 1. Schematic diagram of the row-column array. The device is composed of two-layer nanowires and a SiO2 isolation layer. Both the top and bottom NbN layers have 32 × 32 nanowire pixels, which are serially connected in rows and columns, as shown by blue and yellow solid lines, respectively. The nanowire channels in the row or column are then referred to as R1-R32 and C1-C32, respectively. When the photon triggers a certain row on the top layer, the top SNSPD is triggered, and a pulse signal is generated in the external reading circuit. Meanwhile, the phonons transmit to the bottom layer through the SiO2 spacer and trigger the bottom SNSPD. Purple diamonds represent non-activated nanowire pixel units, whereas red represents pixel units that produce hotspots. |
3. Device preparation
Fig. 2 shows an overview of the key steps of the device fabrication process:(1) The first layer of the NbN film with a thickness of 8 nm was grown on a 2-inch silicon substrate with double-sided oxidation using magnetron sputtering. (2) Then, the nanowire pattern was defined using electron beam lithography with ZEP520A electron beam photoresist and transferred to the NbN film through reactive-ion etching (RIE) with CF4 gas at a pressure of 4.0 Pa and an RF power of 50 W. The electrode part was made by a contact-type ultraviolet exposure machine MA6, and AZ703 was used as a photoresist, which was spun to a thickness of approximately 0.96 μm. Acetone was used to remove the residual photoresist AZ703 when completing the preparation of the first layer of nanowire patterns. (3) A 20-nm-thick Al protection layer was deposited on the pad region of the electrode using a lift-off process, which served as a cut-off for etching and conduction for external connections. (4) A SiO2 isolation layer of about 350 nm was deposited through plasma-enhanced chemical vapor deposition. The surface of the grown silicon dioxide layer presented uneven topography along with the undulation of the underlying layer. (5) Chemical mechanical polishing (CMP) was used to planarize the silicon oxide surface to ensure the growth of the second film. The CMP time was 21 s. The surface roughness of the SiO2 layer was reduced from 3 to 0.38 nm, measured using an atomic force microscope before and after the CMP process, and the thickness of silicon dioxide was reduced from 350 to 190 nm. The AFM scanning range was 5 μm × 5 μm. Overlay marks made of 5-nm Ti and 50-nm Au were used for the overlay alignment of the top and bottom layers. (6) A second layer of 8-nm-thick NbN film was deposited by magnetron sputtering for the preparation of the second layer of SNSPD. (7) The second layer fabrication was realized using the same process as the pattern preparation of the first layer. (8) The insulating layer that covered the bottom electrode connection area was removed using RIE with SF6 gas at a pressure of 12 Pa and an RF power of 100 W.
Fig. 2. Device fabrication flow. The SNSPD meander and layer thickness are not shown to scale. |
Fig. 3(a) shows the optical microscope image of the device. The red and white parts correspond to the electrode areas of different layers, and the green part in the middle is the nanowire area. Fig. 3(b) is an enlargement of the nanowire region in Fig. 3(a), showing the entire 0.96-mm × 0.96-mm photosensitive area of the device. Fig. 3(c) shows the scanning electric field microscopy (SEM) image of the fabricated array with the pixel pitch and size. We set the spacing of each pixel to 10 μm in the layout design to avoid the occurrence of triple coupling events(events with detections on three channels within a coincidence window); of course, appropriate spacing also plays a role in expanding the size of the photosensitive area of the device. The pixels of a single row and column are connected in series, and the upper and lower layers are orthogonal. In this arrangement, each pixel can be processed independently, and different rotation angles and different line width periods can be designed. The size of each pixel unit of the device is 20 μm × 20 μm, including a 10-μm × 10-μm photosensitive area and other auxiliary etching areas, and the interval between pixel units is 10 μm. The linewidth and period of the nanowire are designed as 150 nm/300 nm, as shown in Fig. 3(d).
Fig. 3. a) Optical micrograph of the fabricated array showing the bottom layer leads (light blue), top layer leads (purple), and nanowire region (green). b) Optical micrograph of the nanowire array region, with a size of 0.96 mm × 0.96 mm. c) The scanning electric field microscopy (SEM) image of the fabricated array showing the pixel pitch and size. Each pixel unit is composed of a nanowire area (light red), auxiliary exposure area (light yellow), dividing line (red), and electrode connection line (blue). The overall size is 20 μm, and the pixel pitch is 10 μm. d) SEM image of the fabricated array showing that the nanowire pitch and width are 300 nm\150 nm. |
4. Device characterization
4.1. Photoresponse and yield
As shown in Fig. 4(a), the device was placed on the designed 64-channel device holder, led out to the printed circuit board through the bonding wire, and then introduced into vertical and horizontal staggered placement mini Sub-miniature Push-On electrical channels; the package was then mounted to the cold head of a two-stage Gifford-McMahon cryocooler at a working temperature of ∼2.3 K. To introduce incident photons, a single-mode fiber was used, located about 8 mm away from the chip to accommodate a larger spot size coupled with the detector array. The input photon flux emitted by either continuous-wave or pulsed laser sources was controlled using two serial variable attenuators and calibrated using an optical power meter. To optimize the single-photon detection efficiency, the input photons were carefully tuned by using a polarization controller.
Fig. 4. a) Device packaging diagram. b) Transition current distribution of each channel in the top and bottom layers. c) Count rate curve vs. bias current of every channel. Herein, 1–60 indicates the number of channels to be tested; 1–32 are channels on the bottom layer, and 33–60 are channels on the top layer. The maximum value Isw is the transition current obtained by scanning in the absence of light, and Ib is the corresponding bias current. d) Variation of the normalized count rate curve as a function of the bias current for a channel fitted with a maximal 77% intrinsic efficiency at 1550 nm, with a maximal 96% intrinsic efficiency at 405 nm..In this paper, the term “Isw” specifically refers to the superconducting transition current of the device. It indicates the point at which the device transitions from a superconducting to a non-superconducting state. On the other hand, “Ib” refers to the actual bias current that is applied to the ends of the device. |
Each voltage pulse generated from the nanowires in the SNSPD array was individually amplified using a low-noise amplifier (RF Bay Inc., LNA-650) with a 50 dB gain, operating at room temperature. The input optical power used in this study was approximately −70 dBm. Since our cryostat system accommodated 16 coaxial cables, we characterized the electrical and optical properties of each of the 64 SNSPD channels by changing the connections between the channels and the coaxial cables during each cool-down process.
It should be noted that the leads and connections of the printed circuit board introduce some capacitance and inductance. However, based on our previous experience with multi-channel superconducting single-photon detectors, their impact on signals below 1 GHz is minimal and can be disregarded.
Fig. 4(b) shows the switching currents of 64 channels in the top and bottom layers with different distributions. The 32 channels of the bottom layer show a mean value of 15.1 μA and a standard error of 4.6 μA, whereas the 28 channels of the top layer show a mean value of 13.4 μA and a standard error of 5.4 μA. Moreover, four of the 32 channels on the top layer do not work, indicating a pixel yield of 60/64 (∼94%).
Fig. 4(c) shows the count rate curve of each channel as a function of the bias current. The maximum count rate (MCR) of a single channel reaches 0.35 Mcps, and the MCR calculated by 64 channels can reach 20 Mcps. Compared with the previous single-pixel SNSPD with an active area with a 100-μm diameter [18], the present active area was increased by ten times, while the MCR was improved over two times.
We then performed efficiency tests as the bias current on 60 channels. Fig. 4(d) illustrates the intrinsic efficiency of a row, which represents the device's capability within that specific row to convert absorbed photons into pulse signals. The intrinsic quantum efficiency of each channel was obtained using Boltzmann curve fitting, which achieved a maximal 77% intrinsic quantum efficiency at 1550 nm at a bias current of 18.3 μA, as shown in Fig. 4(d), and achieved near-saturation intrinsic quantum efficiency of 96% at 405 nm. Further optimization of the quantum efficiency can be performed on the superconducting material sputtering and nanowire fabrication.
4.2. Timing jitter
The timing jitter represents the uncertainty in the time delay (fluctuation in time delay) between the instant the photon is absorbed by the device and the instant the response pulse is detected, which can be measured using time-correlated single-photon counting technology. We measured the timing jitter of 23 bottom channels and 22 top channels. The average time jitter of the top layer, 274 ± 107 ps, is larger than the average time jitter of 220 ± 79 ps of the bottom layer. Both layer detectors show larger timing jitters than those of conventional single-pixel SNSPD, which can be attributed to its large geometric jitter [18]. Each row or column consists of 32 pixels connected in series, the electrical length of each pixel is 330 μm, and the transmission rate of electromagnetic waves is about 15 μm/ps [19]. The detection timing jitter is limited in part by the spatial variation of the photon detection events along the length of the wire, which can be suppressed by double-ended readout to achieve high time-accurate light detection [20].
Fig. 5(a) and (b) shows timing jitter for channels with different switching currents given the same low bias current; the jitter data for the entire channel may be mainly contributed by a single pixel. We observe that the timing jitter decreases as the channel switching current increases, showing the contribution of intrinsic jitter to the overall jitter at low bias currents. The lower timing resolution of the top layer may have resulted mainly from the material property of the second layer NbN, which can also be observed in Fig. 4(b).
Fig. 5. (a) and (b) Jitter curves for channels with different switching currents under the same low bias. (c) and (d) Broadening or multi-peaking jitter curves of different channels at high bias. |
Note that some time jitter curves show curve broadening or multi-peak phenomenon at high bias. The jitter data for the entire channel may be generated by more pixels, as shown in Fig. 5(c) and (d). We attribute this to the limitation of our nanowire processing quality. The overall length of each row is close to a millimeter, and the 32 pixels on the same row or column have differences in film thickness and nanowire width, which results in different intrinsic detection efficiencies for photons of the same wavelength. Thus, two non-adjacent pixels produce two characteristic peaks because of the distance between them. If other pixels between two pixels with high intrinsic efficiency have low intrinsic efficiency, most of the counts are generated by the two non-adjacent pixels during the jitter data collection process. The superposition of two or more characteristic peaks causes the broadening of the pulse peak or the phenomenon of multiple pulse peaks.
The other part is that the light introduced by the single-mode fiber fails to show Gaussian distribution on the pixels of a single row and column. The series connection of pixels leads to the length of the nanowire line being too long, and the photoresponse may be more concentrated in certain areas [21]. The above two reasons together lead to the broadening or multi-peak phenomenon of the time jitter peak.
4.3. Thermal coupling and coincidence counting
Fig. 6(a) shows the correlation between the top and bottom layer pulses, which verifies the thermal coupling between these two layers. Note the layer pulses generated 3 ns after the upper layer pulses, which is the transition time of the generated phonons of the upper layer passing through the spacer layer of about 190 nm. The transition time can be decreased using a thinner spacer layer. It is worth mentioning that in order to calibrate the pulse interval and thermal coupling probability, we used the same process to prepare a double-layer single-pixel device and ensured that the length of the external coaxial line was the same during the measurement process.
Fig. 6. a) Pulse correlation between the bottom and top layers. The red and black curves represent the pulse generated by the bottom and top layers, respectively. The inset is a local magnifying part of the two pulses during −5 to 5 ns. b) Curve of thermal coupling probability as a function of bias current. c) Distribution of the coincidence count histogram obtained by TSCPC with the time difference between the two channels. The legend represents the position of the counting measurement; R4 represents the fourth channel of the bottom layer, and C5 represents the fifth channel of the top layer. d) Histogram of the full width at half maximum for different pixels coincidence count curve (30 pixels in total). |
Fig. 6(b) shows the curve of thermal coupling probability as a function of the bias current, where the thermal coupling probability can be calculated as $ \mathrm{P}_{\mathrm{T}->\mathrm{B}}=\frac{\mathrm{N}_{\mathrm{Btm}}^{\text {Toponn }}-\mathrm{N}_{\mathrm{Bpm}}^{\mathrm{Top} \text {.off }}}{\mathrm{N}_{\text {Tap }}^{\text {Btm off }}} $f. In the formula, $ \mathbf{N}^{\text {Top_off }} Btm $ is for the bottom and $ \mathrm{N}_{\mathrm{Top}}^{\text {Btm_off }} $ is for top SNSPD; both were measured when the other detector was turned off. Then, both detectors were turned on, and their counting rates $ \mathrm{N}_{\mathrm{Top}}^{\mathrm{Btm}} $ and $ \mathrm{N}_{\mathrm{Bim}}^{\mathrm{Top}} $ were determined [22]. The thermal coupling probability increased with the increasing bias current of the top or bottom bias current.
Fig. 6(c) shows the row and column channel coincidence count curves tested by TCSPC, which determines the spatial position of the incident photons. The full width at half maximum (FWHM) in the counting curve corresponds to the thermal coupling time of the top and bottom layers. As shown in Fig. 6(d), FWHM for the coincident count peaks of different pixels were between 2.5 and 5.0 ns, which matched the pulse interval of the top and bottom layers measured using an oscilloscope in Fig. 6(a). The FWHM of the coincident count peaks was jointly determined using the row or column channels in both layers. In addition, experiments have proven that the thermal coupling time is proportional to the thermal coupling probability [22]. Therefore, we can judge the thermal coupling probability of each pixel position using the width of FWHM.
5. Conclusion and outlook
In summary, we have proposed a process scheme for fabricating double-layer thermally coupled array devices and fabricated a thermally coupled 32-row × 32-column multiplexed kilopixel SNSPD array based on an NbN material. The devices can work at 2.3 K provided by a compact GM cryocooler. The TRC architecture does not require biasing resistors or wiring within the device’s active area, as needed for electrical row-column arrays, which saves more space and tightens the arrangement of pixels, thereby achieving a higher fill factor. The array has a photosensitive area of 0.96 × 0.96 mm2 and a fill factor of 11.1%. Furthermore, the bilayer structure design may benefit optical designs such as polarization-insensitive structures.
In follow-up work, we aim to use NbTiN films instead of NbN films to fabricate devices and optimize parameters such as the linewidth and film thickness, among others, to address the uniformity and saturation of pixels [23]. We will also reduce the isolation layer thickness for a higher probability of thermal coupling. Composite imaging of intensity, depth, position, and spectrum is expected when each pixel is independently designed and integrated with different filters.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
We thank the National Natural Science Foundation of China (61971408, 61827823, 12033007), Shanghai Municipal Science and Technology Major Project (2019SHZDZX01), Shanghai Rising-Star Program (20QA1410900), Youth Innovation Promotion Association of the Chinese Academy of Sciences (2020241) for their financial support, and Shanghai Sailing Program (Grants No. 21YF1455700).
