Photon-pair generation in a heterogeneous silicon photonic chip

Integrated photonics has emerged as a key technology in various applications such as data centers, telecommunication, photonic computing, and quantum computing. These applications require numerous integrated components including low-loss waveguide routing, laser sources, photodetectors, modulators, single photon and photon pair sources. However, due to distinct properties of different materials, it becomes necessary to incorporate various materials in heterogeneous platforms to incorporate different components onto a single platformDavanco et al. (2017); Schnauber et al. (2019); Moody et al. (2022).

Among those photonic components, correlated photon-pair sources are of great interest due to their applications in quantum computing, quantum teleportation, quantum information processing, quantum sensing, quantum cryptography and othersO’Brien (2007); Bouwmeester et al. (1997); Kimble (2008); Degen et al. (2017); Gisin et al. (2002); O’Brien et al. (2009). A photon pair source with a broad and switchable bandwidth is essential for applications such as quantum frequency multiplexing and quantum memoryLavoie et al. (2013); Sinclair et al. (2014). Over the past decade, there has been tremendous progress with chip-integrated photon pair sources by spontaneous four-wave mixing (SFWM) and spontaneous parametric down-conversion (SPDC), such as those in crystalline silicon (c-Si) Kues et al. (2017); Lu et al. (2016), silicon nitride ($\mathrm{Si_{3}N_{4}}$) Zhang et al. (2016), hydrogened amorphous Silicon (a-Si:H) Wang et al. (2014); Hemsley et al. (2016), gallium phosphide (GaP) Logan et al. (2018), gallium arsenide (GaAs) Chang et al. (2019), and lithium niobate (LiNbO3)Ma et al. (2020). Among these platforms, a-Si:H has achieved particular attention as a photonic platform due to its ultra-high Kerr nonlinearity ($n_{2}=7.43\pm 0.87x10^{-13}$ $cm^{2}/W$) comparing to other silicon-based material platformsWang and Foster (2012). With such a high Kerr coefficient, efficient SFWM can be achieved with only millimeters of waveguide propagation. Contending with crystalline Silicon (c-Si), a-Si:H can be easily deposited at a lower temperature and thus incorporated in massive chip manufacturing processesHemsley et al. (2016).

Despite the excellent Kerr nonlinearity of a-Si:H, the lowest reported a-Si:H linear propagation losses are about an order of magnitude higher than $\mathrm{Si_{3}N_{4}}$Oo et al. (2019). In quantum application, Even though the high nonlinearity of a-Si:H plays a key role in efficiently generating photon pairs, the high propagation loss of a-Si:H waveguide resuts in many of the photon pairs being lost on chip before off-chip detection. This dramatically reduces the coincidence- to- accidental rate that can be observed and severely limits its quantum applications.

In our previous work, to mitigate higher propagation losses in a-Si:H waveguides, $\mathrm{Si_{3}N_{4}}$ waveguides are used for on/off-chip coupling, and vertical coupling between a-Si:H and $\mathrm{Si_{3}N_{4}}$ occurs on-chipMacFarlane et al. (2019). With a careful design of vertical coupler, the coupling loss is optimized around 0.5 dB per transition. The replacement of the a-Si:H with a $\mathrm{Si_{3}N_{4}}$ waveguide for on-off chip coupling dramatically reduces fiber-to-fiber loss. This reduces the required pump power especially for photon pair generation and broadens the applications of such a platform to other quantum technologies.

In this paper, leveraging the low loss and high nonlinearity of the heterogeneous platform, we further explore the quantum performance of a-Si:H/$\mathrm{SiN_{x}}$ waveguides. We experimentally demonstrate the photon pair generation with the highest coincidence- to- accidental rate (CAR) of 1632.6 ($\pm$ 260.4) in non-resonant a-Si:H waveguides, showing significant improvement comparing to the previous state of the art value in non-resonant a-Si:H waveguideWang et al. (2014). The pure photon pairs are measured at multiple channels with a detuning bandwidth of 6 nm and CAR value consistently at 200. In addition, with the same chip, we perform heralded generation of single photons. The auto-correlation is measurement as low as 0.0126 $\pm$ 0.005. Our work combines the high nonlinearity of a-Si:H and the low propagation loss of $\mathrm{Si_{3}N_{4}}$. Our results demonstrate the benefits of a heterogeneous silicon platform for quantum applications by utilizing both low transmission loss and highly nonlinearity. With low-loss inter-layer coupler, such heterogeneous silicon platform also paves the way for integrating even more layers and materials to achieve high-quality components for various applications in quantum applications.

The multi-layer architecture is shown in Figure 1 (a-c). The device fabrication starts with a 100-mm silicon wafer with a 3-$\mu$m thick layer of thermal oxide on its surface. 240 nm of $\mathrm{SiN_{x}}$ is deposited using low-pressure chemical vapor deposition (LPCVD). The recipe parameters consisted of a deposition rate of 4-nm/min, a temperature of 775 $\mathrm{{}^{o}C}$, 1000-MPa tensile stress, and 250 mTorr chamber pressure. Platinum marks for alignment between layers were patterned using electron beam lift-off lithography. Weaveguides in the $\mathrm{SiN_{x}}$ layer were patterned and etched by e-beam lithography and reactive ion etching respectively. The Silicon dioxide ($\mathrm{SiO_{2}}$) cladding was then deposited with plasma-enhanced chemical vapor deposition (PECVD). Planarization of the $\mathrm{SiO_{2}}$ surface was performed using chemical mechanical polishing. 225-nm of a-Si:H was then deposited by PECVD. The a-Si:H layer was deposited using recipe parameters of a temperature of 300 $\mathrm{{}^{o}C}$, a RF power of 15-W, 1000-SCCM of 5% silane/helium gas flow, and 1800 mTorr chamber pressure. With the lower a-Si:H deposition temperature we are able to deposit it directly on top of the $\mathrm{SiN_{x}}$ waveguides without damaging them. Finally, a-Si:H waveguides are patterned and etched using e-beam lithography and reactive ion etching. The a-Si:H waveguides are clad with PECVD ($\mathrm{SiO_{2}}$) and individual chips are diced out of the wafer. Figure 1 (b) shows the geometry of the interlayer coupler. The optical modes at different coupler sections are simulated as figure 1 (b). Figure 1 (c) shows a dark-field microscope image of $\mathrm{SiN_{x}}$ and a-Si:H waveguides and the interlayer coupling region.

To characterize the Kerr coefficient of a-Si:H waveguides, we first perform classic degenerate FWM in our multi-layer devices. Light is coupled on/off chip via inverse taper couplers in the $\mathrm{SiN_{x}}$ layer as shown in Figure 1 (b). We couple on/off the chip using tapered lensed fibers with mode-field diameters of  2.5 $\mathrm{\mu m}$. The $\mathrm{SiN_{x}}$ layer offers improved fiber-to-chip coupling efficiency from better mode-matching with the lensed fiber compared to the a-Si:H waveguides. Once on-chip, the light propagates for 1.5 mm in the $\mathrm{SiN_{x}}$ layer and is then coupled to the a-Si:H layer via evanescent inverse taper couplers and propagates for varying a-Si:H lengths between 1 mm and 5.5 mm. The signal is then coupled back to the $\mathrm{SiN_{x}}$ layer, where it propagates for another 3.2 mm and is finally output via the $\mathrm{SiN_{x}}$ layer and coupled to a tapered lensed fiber. The on-chip interlayer coupling insertion loss is measured to be less than 1 dB per transition over a 100 nm bandwidth, while the propagation losses in the $\mathrm{SiN_{x}}$ and a-Si:H waveguides are 0.5 dB/cm and 4 dB/cm, respectively. The insertion losses as a function of wavelength are shown in Figure 2 (a). We measure the four-wave mixing conversion efficiency (CE), defined as the ratio of the output idler power to the input signal power, using multiple devices with varying propagation lengths in the a-Si:H layer. The measured conversion efficiency values as a function of a-Si:H propagation length are shown in Figure 2 (b). These results equate to a Kerr coefficient of $\sim 5$ x $10^{-13}$ $\mathrm{cm^{2}/W}$, which is comparable to the state-of-the-art value ($7.43$ x $10^{-13}$ $\mathrm{cm^{2}/W}$) Wang and Foster (2012).

The photon pairs are then generated through SFWM in the a-Si:H waveguides. In figure 3 (a), a 50MHz mode lock laser(Calmar Mendocino) is connected with a variable attenuator and four cascaded dense wavelength-division multiplexing (DWDM) filters at 1552.5 nm to accurately control pump power and suppress photons beyond pump wavelength, respectively. We couple to the TE waveguide mode for optimum SFWM efficiency and broad bandwidth by carefully tuning a fiber polarization controller (FPC) and couple to the $\mathrm{SiN_{x}}$ inverse taper coupler through a lensed fiber. After propagating in the $\mathrm{\mathrm{SiN_{x}}}$ waveguide, light is coupled through an interlayer coupler to a 2.5 mm long a-Si:H waveguide where the photon-pair is generated. The generated photon-pair is coupled back to a low-loss $\mathrm{SiN_{x}}$ waveguide and collected by a lensed fiber at the output. The total on-off chip loss is 9 dB. The remaining pump photons are immediately rejected by two cascaded DWDM filters with labelled 60 dB extinction. Two DWDM filters at 1545.3 nm collect Stokes photons and reject all other wavelengths. The anti-Stokes photon at 1559.7nm is collected by two other DWDMs. The bandwidth for the DWDM filters are aligned to be 200 GHz. The total losses by the output DWDM sets are 1.9 dB and 2.2 dB at Stokes and anti-Stokes channels, respectively. To ensure maximum detection efficiency, two FPCs are placed before entering superconducting nanowire single photon detectors (4 channels SNSPDs, ID281, ID Quantique). The SNSPDs have dark-count rate of 50-100 Hz and detection efficiency of 85%. The detected signals are lastly resolved by a time-tagging unit (Swabian Instruments) for photon counting and correlated detection.

The stimulated Raman scattering (SRS) noise is first characterized by measuring the power-dependent single photon counts. In the waveguide, the SRS generated photons are linearly proportional to the on-chip pump power $P_{in}$ while the SFWM generated photons are proportional to $P_{in}^{2}$. Thus, SRS noise can be abstracted at Stokes and anti-Stokes channels by fitting the curves as shown in figure 3 (b).

Figure 4 (d) depicts CAR measurement for multi-channel photon-pairs. To measure the CAR at different wavelength channels, instead of using DWDMs to select the signals and idlers, the photon pairs are filtered using a programmable waveshaper. The waveshaper introduces an additional 5 dB loss, which increases the required pump power. With on-chip pump power at -20.5 dBm, the CAR remains about 200 for the wavelength detuning from 4.2 nm to 10.2 nm. This showcases a wide bandwidth for photon-pair generation that can be dispersion engineered into these devices.

Table 1 shows the state of the art of photon pair generation through $\chi^{(3)}$ process. Comparing with the previous result in a pure a-Si:H platform, our device has exhibited 4 times higher CAR and 30 times higher PGR due to much lower propagation/coupling loss through the waveguide. With the CAR higher than 1000, our work features a high PGR of 1.94 MHz. This shows that our device can massively produce high quality photon pairs. Leveraging high the nonlinearity of a-Si:H waveguide, the photon pairs can be produced efficiently through our non-resonant waveguide without the nonlinearity enhancement by microring. This allows more freedom of wavelength selection and avoid the instabilities of resonance shifting due to the thermo-optic effect or nonlinear phase modulation from a strong pump. Our showcase this heterogeneous a-Si:H/$Si_{3}N_{4}$ platform as an excellent and stable platform for photon pair and single photon generation.

In this paper, utilizing the high nonlinearity of a-Si:H for efficient photon pair generation and low loss of $\mathrm{SiN_{x}}$ for the routing of photon pairs, we have experimentally demonstrated photon-pair generation with a record high CAR of 1632.6 ($\pm$ 260.4) for non-resonant a-Si:H waveguides, which shows 4-times improvement compared to the previous single-layer a-Si:H platform Wang et al. (2014). We also demonstrate multi-channel photon-pair sources with a CAR of around 200 across 10-nm bandwidth. Additionally, we have measured the heralded single-photon generation and demonstrated low correlation function of $g_{H}^{(2)}(0)$ at 0.0125 $\pm$ 0.005. Our results show that with the increased coupling efficiency and low propagation losses of the $\mathrm{SiN_{x}}$ waveguide, higher photon detection rate can be observed with much lower on-chip pump power. This enables us to achieve a high CAR and high photon pair generation rate. The multi-material deposition and low-loss interlayer couplers has enabled a way to combine different materials’ features onto the single platform and flexibly transfer photons and optical signals between different layers and materials with low transition loss. This solution offers the possibility for the future PICs to integrate components from other material platforms and further miniaturize the experimental setups used in this work, for example $\chi^{2}$ nonlinearity and electro-optic modulation from lithium niobate, laser source from III-V materials, and photodetectors from germanium Chen et al. (2019); Wang et al. (2018); Jin et al. (2021); Keyvaninia et al. (2013); Guo et al. (2019); Michel et al. (2010); Marris-Morini et al. (2018).

This research was supported in part by National Science Foundation (#1641094).