Light interference detection on-chip by integrated SNSPD counters

SWIFTS is an innovative concept for integrated spectrometers based on the readout of light interferences by nanodetectors and the reconstruction of the original spectrum by inverse Fourier transform.^1,2 Inspired by G. Lippmann's works on color photography in late 1800's,^3,4 the concept now opens the path to extremely compact and motionless spectro-detectors. Existing prototypes with remote lecture systems exhibit promising performances,⁵ offering a good idea of the potential offered by such integrated spectrometers. Nevertheless, it has not been possible until now to directly sample the interferogram using integrated arrays of sensitive nanodetectors: in current devices, diffusive gold nanowires were used to project the pattern for a remote lecture by micrometric Coupled-Charge Device (CCD) detectors, implying a sampling spatial period much larger than the standing wave period. As a consequence, undersampling techniques were required to reconstruct the spectrum by inverse Fourier transform, with the inherent aliasing drawback. We aim to demonstrate that these limitations can be overcome thanks to the use of SSPD nanowires, leading to an extremely compact SWIFTS microspectrometer operating at its full potential, as pictured in Fig. 1. The near-infrared range, around 1.55μm wavelengths, is convenient for this investigation in regard to SNSPD detection and SiN waveguide properties, as well as spectral range relevancy for SWIFTS.

The SNSPD detectors are known for their high speed and very low noise, and can be operated in single photon counting mode at near-infrared wavelengths when biased close to their critical current,^6,7 which makes them suitable for accurate intensity measurements. Their nanometric dimensions make them the ideal candidate among detectors for localized sampling of light patterns. Moreover, their relatively low intrinsic coupling to light, i.e. quantum efficiency (QE) about 1% for classical 5x5μm²-pixel bare meander detectors, becomes an asset for a measurement with minimum impact on the analyzed standing wave. The interferences are created by a counter-propagative interferometer, constituted by a loop-waveguide dividing the incident wave into 2 paths joining each other.

The next section describes the SWIFTS-SNSPD chip design, featuring an interferometric loop deposited over an array of 24 nanodetectors. The preliminary studies concerning waveguide development and NbN nanowire studies, already described in previous works,⁸ are briefly recalled in section three. The third section details the fully integrated device fabrication, while its characterization is discussed in section four. Section five reviews the advances done so far and concludes on the possibilities shown by the prototype.

In order to achieve interferogram sampling in agreement with Shannon's criterion, one needs to use photon detectors smaller than the standing wave period. This requires dimensioning adequately the standing wave with respect to the implementation of the narrow width SNSPD nanowires.

Epitaxial quality of the superconducting NbN nano-layer is mandatory, and could only be obtained by deposition onto a suitable single crystal substrate, R-plane sapphire heated above 500°C in our case⁹ as shown on Fig.2.

The accessible technology led to the choice of a Si₃N₄ ridge waveguide, with a refractive index of 1.96 measured by ellipsometry, deposited over a 1-D array of 24 epi-NbN nanowires grown on R-Sapphire substrate. Similar optoelectronic stacks, containing deposited SiN waveguides over NbN SNSPDs, had been previously studied and simulated,¹⁰ but not fabricated and tested as operational device. Our experimental device allows the sampling of a ∼400nm-period stationary wave (at 1.55μm wavelength in free space) by an array of 40nm-wide, 120nm-spaced SNSPD, as shown in Fig.3. Such a configuration permits a spectral resolution of ∼9, and a spectral bandwidth of ∼2μm.

The device has been designed taking into account the available thin films, processing technologies and characterization techniques, which led us to the final chip disposition shown in Fig. 4(a) and 4(b), with 6 mm x 8 mm size. Input light is injected through end-fire fiber coupling into the interferometer, on the chip's cleaved edge, with the help of a coupling taper. NbN nanowires are terminated by large 200 x 6 μm², low contact resistance, connecting surfaces under the Ti-Au coplanar tracks, which lead to the Ti-Au contact pads.

The interferometric loop dimensions were chosen after a careful study of SiN waveguides with different ridge widths and heights, having straight or curved geometries, in order to obtain single-mode waveguides at 1.55μm: multimode waveguides would induce different optical paths values for the present modes. When each mode interferes with another, several zero path difference (ZPD) positions will appear, resulting into a blurring of the interferogram. Present modes were calculated using the effective index method with the help of Wave-Matching Method (WMM) tools,¹¹ and losses induced by curvatures were estimated by the Aperiodic Fourier Modal Method (AFMM).¹² Test waveguides were fabricated to validate the simulation results and to measure the effective losses: 1μm-thick Si₃N₄ was deposited by plasma enhanced chemical-vapor deposition (PECVD) at 280°C on R-Sapphire substrate, followed by Deep-UV optical lithography (UV5 resist) and SF₆/O₂ reactive ion etched (RIE) to define the ridges. Samples were measured on an optical bench setup injecting light by end-fire fiber coupling, and outbound signal was collected using a commercial infrared camera. It was established by the cut-back method that a high level of propagation loss (∼ -15dB/cm) was present, mainly due to line edge roughness of optical lithography masks inducing irregularities along the ridge. To address this issue, we chose better quality photo-masks for our final SWIFTS device fabrication.

End-fire fiber coupling also induced important losses (∼ -15dB), strongly dependent on the fiber/waveguide mode mismatch and the quality of the cleaved facet. The latter has been improved with adequate cleaving technique: wafers are partially cut with a diamond blade on their reverse side, to form thin trenches about 380μm deep in the 550μm-thick, 4-inches R-plane Sapphire substrate. The chips can then be manually cleaved in an effective and reproducible way. Despite the strong losses, the waveguides are functional with monomode guiding properties as required for the application.⁸ 

The resulting configuration of the interferometric loop, considering all the losses induced by coupling and propagation, imposes a total light signal attenuation of the order of -32dB. The ridge waveguide, featuring a SiN core of about 1μm thick over very thin (∼4nm thick) NbN nanowires, leads to an evanescent coupling of light with the SNSPD counters. As detailed in Ref. 13, the mapping of the guided fundamental mode shows an exposition of the counters to approximately 0.25% of the light power integral in the ridge above. In this way, we can expect an exposition of the nanowires to roughly 2.10-6 of the input laser light power, which in our case corresponds to a few nW.

The NbN nanowires were designed to comply with two main requirements: Shannon sampling criterion (impacting on nanowire widths and spacing) and non-latching requirement (impacting on nanowire length). While Shannon criterion is addressed by the ebeam lithography capacity in defining nanowire patterns narrow enough compared to the interferogram period, the non-latching issue needs careful dimensioning of the nanowire length to avoid the creation of self-heating hotspots.¹⁴ For this matter, we chose to increase the nanowires’ inductance up to classical 5x5μm²-pixel meander SNSPD inductance levels, by increasing its number of squares to about 1000 considering the R-plane sapphire substrate used, reaching typical levels of approximately 90nH.¹³ In order to keep the nanowire sensitivity at its highest level only on its exposed parts to light, unexposed “inductive tail” parts are 5 times wider, as shown in Fig. 4(d). The resulting lower SNSPD switching speed and reduced maximum count rate expected from large inductances SNSPD does not constitute a drawback for the SWIFTS functionality, as SNSPD should operate at medium count rates, sequentially, to extract relative variations of light intensities along the waveguide.

The fabrication process of the SWIFTS-SNSPD is described in Fig. 3. A 4nm-thick epitaxial NbN is first DC-magnetron sputtered on the heated R-plane Sapphire 4-inch wafer,⁹ and patterned afterwards by electron-beam lithography with a VB6 UHR tool from VISTEC, using the negative tone resist NEB35 from Sumitomo. This resist, with a thickness of 45nm, allows good reproducibility of ∼40nm wide nanowires and 160nm pitch with 90μC/cm² dose to size (±10% process window). After O₂ RIE plasma descum of residual resist between nanowires, the SNSPD are defined by SF₆/O₂ RIE etching, and the resist is stripped with EKC-LE bath.¹⁵ We then proceed to the lift-off of evaporated Ti-Au (30-150nm thick) contacts.

To ensure a good chemical protection of the NbN nanowires for the following fabrication steps, a protective MgO thin film is RF-sputtered on the whole wafer before processing of the optical layer. It has been demonstrated that this nano-cladding preserves the NbN from being oxidized, which can occur during the wafer introduction into the PECVD machine (loading onto a 280°C hot plate under air for 30 seconds) for the following step of 1μm-thick SiN deposition.⁸ 

Afterwards, the interferometric loop is patterned by Deep-UV optical lithography and SF₆/O₂ RIE partial etching to reach a ridge height of about 350nm. Finally, remaining SiN and MgO layers over the contact pads are entirely etched respectively by RIE and diluted H₃PO₄ bath to open the vias.

Once the chips fabricated, the waveguide dimensions can be measured by ellipsometry and profilometry to verify the monomodicity by simulation.

Aside the waveguide's modal properties, another issue concerns the guiding of fundamental TE and TM polarizations: the different effective indices N_eff induce an intrinsic birefringence. Therefore, the induced optical paths difference slightly changes the interference periods with a δλ shift:

For our waveguide dimensions (typically 1.1±0.05μm core thickness, for 0.35±0.05μm ridge height and 1.6-2μm ridge widths), we typically have N_eff(TE0)∼1.88 and N_eff(TM0)∼1.86, and equation (1) gives δλ∼4.3nm for λ₀=1.55μm, which is about 1% of the standing wave periods λ(TE0) and λ(TM0). Considering that a fringe blurring occurs for a cumulated shift of ∼25% (a quarter period), corresponding to 25 periods from the central fringe, we can establish the maximal detector distance to the ZPD to ∼10μm before signal jamming. The distance of analysis being shorter (3.84μm between the 1^st and 24^th detectors), the shift should remain under 5% for the most remote detectors considering a ZPD located in the middle of the array (if we consider the interferometric loop to be strictly symmetrical). Moreover, it has been demonstrated that SNSPD are about 4 times more sensitive to TE modes, for which the electrical field is parallel to the nanowires, than to TM ones.¹⁶ The effect would be to attenuate the shift visibility by diminishing the TM signal contribution.

In this way, we conclude that the waveguide intrinsic birefringence should have minimal impact on the detected pattern. Nevertheless, this factor will have to be treated carefully in future devices covering a longer distance of analysis, for example by designing polarization-independent waveguides.¹⁷ 

Compared to our previous achievements of 80nm and more NbN nanowires widths,⁷ the achievement of much narrower dimensions with a section of about 40nm x 4nm may be detrimental to the nanowires’ superconducting properties. In order to check that issue, the NbN critical temperatures have been measured on different nanowires and typically display Tc values of 11.7K as shown in Fig. 5(a), which indicates a good quality of the patterned stripes, close to the blank film Tc value.

Moreover, the I-V characteristics measured on 40nm-wide nanowires and 80nm-wide meander SNSPD, shown in Fig. 5(b), exhibit very high critical current densities, above 9MA/cm² at 4.2K, as well as low critical current dispersion among the nanowires, confirming the excellent state of the NbN despite drastic dimension reductions of the nanowires, which assesses for the quality of the fabrication process and the effectiveness of the MgO protecting layer.

The end-fire fiber alignment to couple light into the interferometer, inside a Helium Dewar without direct visual control, is extremely challenging. For this purpose, a new insert has been designed as already detailed in a previous report,⁸ to fit with the existing electro-optical setup.¹⁸ 

The alignment protocols take advantage from the reflection of the laser beam by the surface facing the fiber: interferences are created by incident and reflected light as in a Fabry-Pérot cavity, which induces periodic variations in the spectrum of the resulting beam that are inversely proportional to the fiber-surface distance. The readout of these variations with a spectral analyzer gives a reference about the fiber's position in respect to the chip's edge.

We then measure the photon counting characteristics independently on each nanowire. The sequential operation of SNSPD avoids crosstalk risks despite their proximity. One nanowire is kept as a reference from one helium cycle to another, to trace possible experimental condition variations.

With the experimental setup mentioned above, we used a continuous wave (CW) broadband ASE laser source (Thorlabs-FL7002, 1.52-1.61μm). The first step is to adjust the laser beam position in front of the SiN layer, not necessarily the precise entrance of the interferometric loop. Above that SiN layer (in air or helium), the beam reflection on the cleaved surface disappears and spectrum variations are not observed. Otherwise, light is diffused across the planar waveguide, and can be detected by the NbN nanowires as shown in Fig. 6. A large signal-to-noise ratio (SNR) up to 10⁴ is observed when the nanowire DC current bias ratio to critical current is about 0.7, thanks to the very low dark counts rate of SNSPD.

However at this stage, we cannot quantify precisely the light power to which each SNSPD is locally exposed without an in-depth optical study of the SiN waveguide on the device. Such a characterization will be necessary to evaluate the detectors’ QE, the overall system detection efficiency, as well as determining the photon counting regime in a reliable manner.^6,7 The latter will be critical for the envisioned precise measure of light intensities. Although the very narrow widths of the nanowires along with low light powers and current biasing levels above 0.85Ic state in favor of a predominant single-photon regime, the photon counting modes will need further investigation with the use of higher power laser with precise attenuation levels.

The proof of the detection capability of such narrow nanowires under the SiN plane constitutes a validation of the fabrication process and the alignment setup, but is insufficient to reveal the presence of the interference pattern. For that purpose, one needs to place the laser beam at the SiN interferometer loop entrance, by sweeping the fiber along the SiN layer after the preliminary alignment on the chip edge described in the previous paragraph. A good ridge-fiber alignment is indicated by a sharp variation (several orders of magnitude) of the counting rate of the SNSPD biased close to its critical current, for very small movements of the piezoelectric positioners. Once this alignment is achieved, we used a second laser source (Thorlabs-S3FC, narrow band 1.548±0.0005μm) at same power but different wavelength spectrum, which induces a distinct interference pattern as shown in Fig.7: with the standing-wave period shift directly impacting on the SNSPD photon counting rate, it is possible to characterize with a single detector the presence of interferences, which proves useful when nearby detectors are defective or bridged.

Applying this method on the SWIFTS-SNSPD chip, we show in Fig.8(a) that in a random alignment situation, for which the light signal is not injected into the interferometer, the two laser sources tuned at same power induce the same level of detected signal. As anticipated, the light diffused across the SiN plane is detected in an identical way for the two sources, with level variations according to the beam position.

On the opposite, a correct ridge-fiber alignment exhibited a very different behavior as shown in Fig.8(b): the two lasers induced clearly distinct counting characteristics for current biasing levels ranging from 0.7 to 0.98, reflecting the formation of different interference patterns. This demonstrates that, by using only one nanowire, we have been able to discriminate very locally (probing ∼40nm) at 4.2K non-identical light levels issued from the two optical sources, tuned at same power but with different spectra, in accordance with the expected shift of the interferogram depicted in Fig.7. The counts modulation, in addition to the detection level sharp variation with the beam positioning, constitutes a strong indication of light guiding and formation of interferences within the loop waveguide.

Due to time-consuming ridge-fiber alignment, and large number of chips to be tested, we have not yet been able to address the full achievable optical characterization of the interferogram and of this new device effective spectroscopic performance. So far, with more than 40 SWIFTS-SNSPD chips achieved on a single 4-inch wafer and a yield of more than 30% operating nanowires at 4.2K (others being bridged through e-beam resist or etched defects¹³), the selection, test and calibration of nanowires have been done as preliminary tasks.

Eventually, the demonstration of a more accurate and useful chromatic signal detection and reconstruction issued from a phantom optical source would require the selection and one-by-one calibration of several non-defective nanowires among the 24 on the same chip. An extensive optical characterization of the interferometric loop will also be necessary in order to evaluate the light intensity. Such campaigns will open the path to spectrum reconstruction as well as in-depth fringe pattern analysis, potentially revealing new photon-photon intrication phenomena within the waveguide. Additional effects such as light diffraction by nanowires or photon-waveguide coupling modes will also have to be investigated in order to consolidate the understanding of the device and possibly provide correction factors.

In this paper, we described the new SWIFTS-SNSPD device design, fabrication, and characterization. This micro-spectrometer combines advantages of guided optics (a SiN waveguide interferometer) integrated on-top of an array of 24 superconducting SNSPD photon counters which are suitable to sample a colored standing wave at a 160nm step.

The collective fabrication process of devices with a good yield on 4-inch wafer preserves the compatibility between optical and electronic parts of the device by applying a MgO protecting nanolayer over the NbN nanowires. State-of-the-art epitaxy and e-beam technologies have allowed the realization of ∼40nm wide nanodetectors, ∼120nm spaced, exhibiting excellent superconducting properties (Tc∼12K, Jc∼9MA/cm² at 4.2K). The SiN ridge waveguide dimensions ensure monomode guiding of light around 1.55μm wavelengths in the interferometer. The light patterns formed within are detected by the SNSPD, revealing a clear difference in relation to the laser source used, in agreement with the expected interferogram modulation induced by distinct spectra.

The displayed ability of revealing interferences at such small spatial dimensions, with unequalled sampling precision obtained from the NbN-SNSPDs integrated under a SiN-interferometer, constitute a notable breakthrough in the field of superconducting optoelectronics as well as in integrated spectrometry. Immediate developments involve process stabilization and further characterization of the SWIFTS-SNSPD device, to take benefit from the promising results achieved.

Future investigation of SWIFTS micro-spectrometers with competitive spectral resolution would involve a larger array of SNSPDs to increase the standing-wave distance of analysis. With the nanowire dimensioning and sampling period obtained with the SWIFTS-SNSPD, which already allows a very high spectral bandwidth, an array of 128 detectors would cover a distance of analysis up to 20.48μm that would imply a resolution of about 50. The on-chip development of multiplexing and readout electronics based on NbN SQUID or RSFQ circuits¹⁹ would constitute a major step for large scale integration of SNSPD into commercial micro-spectrometer devices.

From a broader point of view, we have been able to demonstrate that SNSPDs are well adapted for a SWIFTS application, with interesting evolution perspectives offered by light sensitivity ranging from visible to mid-infrared wavelengths. The SWIFTS-SNSPD technology is expected to enable new instrumental concepts in astrophysics²⁰ or telecommunications,²¹ thanks to its unique interferogram in-situ oversampling capability with integrated nanodetectors. This specific asset can raise large interest in many other research communities ranging from medical (optical tomography) to environmental (DIAL-LIDAR) fields, where accuracy, sensitivity and speed for a relatively low cost are key factors for measuring ultimately small patterns of light.

The authors thank E. Le Coarer and P. Kern from IPAG for insightful discussions on SWIFTS prototypes and on instrumental perspectives. One of the authors (P. Cavalier) acknowledges a Grant co-financed by CEA-INAC and by the Centre National d’Etudes Spatiales (CNES).