1. Introduction

Spontaneous parametric downconversion (SPDC) has been used for decades to produce quantum entanglement in various photonic degrees of freedom, serving as a workhorse in emerging quantum optical technologies. Compared with most nonlinear processes, SPDC is relatively inefficient, requiring over one million pump photons to produce one pair of entangled photons in even the highest performing crystals. However, recent advances in nanophotonics, particularly in thin-film lithium niobate (TFLN), have enabled significantly more efficient frequency conversion and quantum state generation [1–3] through sub-$\mathrm{\mu}$m interaction areas, high nonlinearities, and low material losses [4,5]. By exploiting this platform, many recent demonstrations of SPDC in TFLN [6–9] have achieved efficiencies three orders of magnitude greater than that of bulk crystal-based sources [10] and one order of magnitude greater than that of large diffused waveguide-based sources [11]. To date, most TFLN-based photon pair sources are designed for SPDC at telecom wavelengths because of the low losses in optical fibers at 1550 nm [12,13] and back-compatibility for applications such as quantum communication [14], computing [15], and a globally connected quantum network [16].

Although telecom photons are preferred for quantum information applications, visible and near-infrared photons are generally better suited for imaging and spectroscopy. Experiments at these wavelengths can take advantage of multi-pixel detectors such as electron-multiplying charge-coupled devices (EM-CCD) and single-photon avalanche detector (SPAD) arrays, enabling the measurements needed for imaging [17–19] and characterization of high-dimensional entangled states [20,21]. Furthermore, the electronic transitions of molecules and atoms become accessible at near-IR wavelengths, allowing for fluorescence lifetime measurements [22–24], compatibility with quantum memories [25], and fundamental studies of few-photon nonlinearities [10,26,27]. More generally, near-IR and visible photons can be detected with high quantum efficiency and low dark noise using existing mature silicon technology at room temperature, compared with near-IR–IR detectors which require cryogenic cooling [28]. Despite these advantages, all demonstrations of nanophotonic pair production have resided in the telecom region, and the best near-IR and visible photon pair sources are still large-area waveguides [29–32] and bulk periodically poled crystals [10,26,29,33–36]. To date, lithium niobate is the only nanophotonic platform that exhibits high transparency at near-UV pump wavelengths and supports a $\chi ^{(2)}$ nonlinearity. Thus, despite the known challenges, TFLN is uniquely posed to address this wavelength range. Potential reasons for the lack of visible TFLN devices stem from the difficulty in fabricating visible nonlinear circuits on thin-film lithium niobate due to factors such as the ultra-short poling periods required for quasi-phase matching and losses from material absorption [37–39] and scattering [40]. In spite of these difficulties, high-performance visible devices in thin-film lithium niobate are becoming increasingly common for classical applications such as electro-optic modulation and second harmonic generation [41–45].

Here, we extend TFLN-based SPDC sources to shorter wavelengths to produce high-brightness photon pairs in the visible and near-IR. The device produces an on-chip efficiency of $(2.3\pm \,0.5)\,\times \,10^{11}$ pairs/s/mW, which corresponds to a per-photon conversion efficiency of more than 1 photon pair converted in every 10,000 pump photons [$(1.1\,\pm \,0.2)\,\times \,10^{-4}$ pairs/s/photon]. This efficiency is nearly two orders of magnitude better than visible-light diffused waveguide SPDC sources [31] and is on par with the highest-performing TFLN sources in the telecom regime [8]. The SPDC from this device exhibits a broad spectrum centered at 811 nm with a degenerate FWHM bandwidth of 117 nm and an average brightness of $(1.6 \pm 0.3) \times 10^{9}$ pairs/s/mW/nm, a number limited by at least an order of magnitude by the pump laser linewidth (0.8 nm). Consistent with this bandwidth, we measure an ultrashort coherence time of $\sim$40 fs for the entangled photons with an indistinguishability of $100 \pm 1{\% }$. These results are the shortest wavelength entangled photons generated in TFLN by nearly an octave to date. Our results therefore show that, although pumped at wavelengths near what would usually be considered its cutoff range, TFLN can equally be a platform for visible–near-IR entangled photon applications as it is at telecom wavelengths.

2. Device Design and Fabrication

The periodically poled lithium niobate waveguides [Fig. 1(a)] were simulated in Lumerical MODE to determine the quasi-phase-matching poling period. The guided modes at the design pump wavelength (406 nm) and SPDC center wavelength (812 nm) were simulated using the bulk Sellmeier coefficients of lithium niobate [46] and silicon dioxide [47] with the geometric parameters shown in Fig. 1(c) and a 60° sidewall, which is consistent with the fabrication process. To take advantage of lithium niobate’s largest nonlinear tensor element ($d_{33}$ = 28 pm/V) [48], only the fundamental quasi-TE modes of X-cut lithium niobate were considered. An etch depth of 420 nm, a top width of 1.5 $\mathrm{\mu}$m, and a total LN thin film thickness of 600 nm were targeted for ease of optical coupling, fabrication, elimination of slab-mode leakage, and near-2-$\mathrm{\mu}$m poling period while providing high performance. For these parameters, the effective refractive indices ($n_{\text {eff,pump}}$ = 2.29, $n_{\text {eff,SPDC}}$ = 2.09) result in a quasi-phase-matching poling period of $\Lambda =\lambda _{\text {pump}}/\Delta n_{\text {eff}}=2.03\,\mathrm{\mu} \text {m}$ at the target pump wavelength of 406 nm [Fig. 1(d)].

 

Fig. 1. (a) Schematic of the periodically poled lithium niobate nanophotonic waveguide. (b) Second harmonic microscopy image of the periodic poling. Note that the distance between the electrodes is 15 $\mathrm{\mu}$m while the waveguide top width is 1.5 $\mathrm{\mu}$m, allowing for multiple waveguides in the center of the poled region where the duty cycle is $\sim$50%. (c) Mode profiles and waveguide geometry of the fundamental quasi-TE modes at the designed pump and SPDC center wavelengths. (d) Refractive indices and corresponding poling periods for a range of SPDC wavelengths.

Download Full Size | PDF

The devices were fabricated from a 1 cm by 1 cm die of a 5% MgO-doped X-cut thin-film lithium niobate on insulator wafer (NanoLN), which consists of 600 nm of lithium niobate bonded to 2 $\mathrm{\mu}$m of silicon dioxide on a 0.4-mm silicon substrate. An MgO-doped film was chosen to lower the coercive field necessary for poling [49,50] and to reduce potential photorefractive effects from the violet laser diode [51]. MgO-doped LN has been demonstrated to lower propagation losses [38], but this doping may also result in irregular poled domains due to leakage current in the charged domain walls [50]. To quasi-phase match the SPDC sources, poling electrodes (7 mm long) were first fabricated by performing a metal lift-off through electron beam lithography with bilayer polymethyl methacrylate (PMMA) resist followed by electron beam evaporation of titanium (15 nm) and gold (55 nm). The electrodes were poled with a 490-V and 70-$\mathrm{\mu}$s square wave, and the poled domain formation was monitored with second harmonic microscopy [52] [Fig. 1(b)]. After poling, waveguides were defined through an aligned electron beam lithography step with hydrogen silesquioxane (HSQ) resist followed by argon inductive coupled plasma reactive ion etching to achieve an etch depth of 420 nm, verified through atomic force microscopy. The chip facets were manually polished to increase coupling efficiency, resulting in a final waveguide length of approximately 8 mm. A broadband oscillator was used to verify the phase-matching wavelength through second harmonic generation (Supplement 1 Section S2) and was found to match with the computationally predicted second harmonic wavelength within $\pm$3 nm. This small discrepancy was likely due to fabrication tolerances, particularly in the etch depth and film thickness.

3. Device Characterization

The spectrum, generation rate, and coherence properties of the entangled photon pairs produced from the fabricated device are characterized, as shown in Fig. 2. In these experiments, the room-temperature periodically poled waveguide is pumped with a free-running laser diode (Coherent OBIS LX 405 nm) to produce entangled pairs [Fig. 2(a)]. An antireflection-coated aspheric lens (NA = 0.58, Thorlabs C140TMD-A) couples the free-space pump beam to the fundamental TE mode of the waveguide. The photon pairs produced in the fundamental TE mode are collected off-chip and collimated using a similar aspheric lens (NA = 0.58, Thorlabs C140TMD-B).

 

Fig. 2. Schematic for coupling and detection of the photon pairs. I, isolator; ND, variable neutral density filter; HWP, half-wave plate; L, aspheric lens; LPF, long-pass filter; emICCD, electron-multiplying intensified charge-coupled device; BS, beamsplitter; D, single-photon avalanche detector; TT, time-tagger; M, mirror. (a) Optical setup to couple into and out of the TFLN waveguide. (b) Characterization scheme for measuring the photon pair spectra. (c) Characterization scheme for coincidence counting. (d) Optical setup for the Michelson interferometer.

Download Full Size | PDF

The spectra of the entangled photon pairs are measured to assess the phase-matching properties and tunability of the device (Fig. 3). Pairs collected from the waveguide are transmitted to a grating spectrometer and measured using an electron-multiplying intensified camera [Fig. 2(b)]. To tune the SPDC emission, the center wavelength of the pump wavelength is varied from 405 to 406.4 nm by changing the drive current of the laser diode. Variable neutral-density filters are used to keep the pump power for all the collected spectra at a consistent 10 $\mathrm{\mu}$W. Each spectrum was divided by the spectrometer wavelength response and then normalized to the maximum count rate to infer the relative intensity. Three distinct phase-matching regions are explored [Fig. 3(a)]: (1) at long pump wavelengths, the phase-matching condition is not satisfied and SPDC emission is not observed; (2) from 405.6 to 405.9 nm, the degenerate wavelengths are phase matched; and (3) at short pump wavelengths, the spectrum splits and nondegenerate emission extending to the cutoff wavelength of the filter is observed. The dip in intensity in Fig. 3(a) around 755 nm is likely due to on-chip loss, potentially from a mode-crossing [53], and is investigated in further detail with a transmission measurement (Supplement 1 Section S2). Due to the linewidth of the laser used in the experiment (0.8-nm FWHM), the spectra are considerably broadened compared with the spectra expected from a single-frequency pump laser (Supplement 1 Figure S4). Nevertheless, experiment and theory reach qualitative agreement by accounting for the pump linewidth [Fig. 3(b)], described further in Supplement 1 Section S6. Further spectral broadening can also arise from propagation loss at the pump or SPDC wavelengths [54] or index variations [55–57]. For all subsequent experiments, a laser center wavelength of 405.7 nm is used for degenerate phase matching. The resulting spectrum [Fig. 3(c)] is centered at 811.4 $\pm$ 0.7 nm with a FWHM bandwidth of 117 nm (53 THz) that accounts for 85% of the overall flux.

 

Fig. 3. (a) Measured and (b) theoretical SPDC spectra as a function of pump wavelength. Note that the arbitrary units for intensity could also be interpreted as counts per max counts due to the normalization of the spectra. (c) Lineouts of the measured and theoretical SPDC spectra at a pump wavelength of 405.7 nm, corresponding to the dashed horizontal lines in panels (a) and (b). Note that the frequency bandwidth per nm and the downconversion efficiency are wavelength dependent (Supplement 1 Eq. (S1)), resulting in higher intensities for shorter wavelengths. The theoretical spectra in panels (b) and (c) are calculated with the laser linewidth taken into account, described further in Supplement 1 Section S6.

Download Full Size | PDF

The pair generation efficiency of the device is measured through coincidence counting between two SPADs [Fig. 2(c)]. In this measurement, the pairs from the chip were split at a 50:50 broadband beamsplitter (Thorlabs BS014) and coupled to multimode optical fibers (Thorlabs M122L01) connected to the detectors. Coincidence detection events between the SPADs (Laser Components Count) were recorded with a time-tagger (Picoquant PicoHarp 300). Figure 4(a) shows a representative raw coincidence histogram recorded at 4.3 nW of on-chip pump power. The temporal correlation in this graph (3.4 ns FWHM) is given by the response time of the SPADs and not the entangled photon correlations (see Fig. 5 later in the text). The coincidence counts are corrected by background subtraction of the number of counts in a 9.5-ns window at the histogram peak from the number of counts in another 9.5-ns window in a background region far from the peak. Sweeping the laser power with a neutral density filter yields the curves in Figs. 4(b) and 4(c), which are linearly fit to determine the pair generation efficiency. Additional experimental details and the raw data from these measurements are given in Section S4 and Table S1. To account for the wavelength dependence of the SPAD quantum efficiency (Supplement 1 Section S3), all wavelengths in the spectrum are integrated over to calculate the average detection efficiency for single photons ($\eta _{1}=0.52$) as well as the average joint pair detection probability ($\eta _{12}=0.27$). Including a factor of 2 due to the probability of splitting pairs at the beamsplitter yields Eq. (1) for the measured efficiency of the source:

 

Fig. 4. (a) Raw coincidence histogram, including accidentals, at an input on-chip pump power of 4.3 nW. Measured (b) coincidence counts with accidentals subtracted and (c) singles counts while sweeping the input power. The vertical axes are measured in coincidence or singles counts per second (cps). The fitted slopes (lines) produce the pair generation efficiency of our device. Note that error bars are included in panels (b) and (c), but are smaller than the data markers. Additional experimental details and the raw data can be found in Supplement 1 Section S4 and Table S1.

Download Full Size | PDF

 

Fig. 5. (a) Measured and (b) simulated two-photon Michelson interferogram for the device. (c) Corresponding singles counts out of the interferometer, demonstrating single-photon interference within the coherence length of the source. (d) Coincidence counts and (e) singles counts far from the coherence length of the source. All measured coincidence counts [panels (a) and (d)] do not have accidentals subtracted. Note that error bars are included in panels (d) and (e), but are smaller than the data markers. Details of the simulated two-photon Michelson interferogram in panel (b) and the sinusoidal fit in panel (d) can be found in Supplement 1 Section S7.

Download Full Size | PDF

Finally, the two-photon interference is measured [Fig. 2(d)] to demonstrate the non-classical behavior of the produced photon pairs. Figure 5 shows the measured two-photon interferogram [Fig. 5(a)] obtained from the device without subtracting accidentals, as well as the one-photon interferogram [Fig. 5(c)] for comparison. Due to the aforementioned temporal resolution of the SPADs used here, the four unique paths through the interferometer are indistinguishable and combine to yield the interference pattern. The important features of the interferogram are as follows. (1) Near the zero time delay, photons are delayed within the coherence length of the source and exhibit both one- and two-photon interference. A visibility of $100 \pm 1$% is measured within the coherence length, with an uncertainty derived assuming Poissonian statistics. This near-perfect visibility indicates good mode overlap in the interferometer and indistinguishability of the photons within the pair. (2) Far from the coherence length of the source (delays greater than $\pm 20$ fs), interference between two photons taking different paths disappears, which explains why the single-photon interference [Fig. 5(c)] disappears in this region, shown in more detail in Fig. 5(e). Notably, interference between pairs of photons that travel together through the interferometer persists with a fringe period at half the pair wavelength [Fig. 5(d)]. This feature suggests quantum interference due to the energy-time entanglement of the pairs, and would not be observed if the light was generated from a coherent or a thermal source with a similar spectrum [58]. A fringe visibility of $43\pm 3{\% }$ is observed in this region far from the coherence length, which is close to the theoretical maximum of 50% for this experiment due to the temporal resolution of the detectors. The uncertainty in visibility is given from the standard error in the fit of Fig. 5(d), the details of which are outlined in Supplement 1 Section S7. The qualitative agreement between the measured and theoretical two-photon interferogram [Figs. 5(a) and 5(b)] suggests that SPDC and genuine energy-time entanglement are being produced.

4. Discussion

Compared with the state-of-the-art for visible photon pair sources, the device presented in this work exhibits substantially improved brightness and efficiency due to the small effective area of the nanophotonic waveguide. The device’s performance against reported literature devices spanning from the visible to IR is plotted in Fig. 6 using efficiency, brightness, and wavelength as the figures of merit. Our visible–near-IR device demonstrates improved efficiency and brightness even over other TFLN sources at the better-explored telecom band [6,7,9] and has comparable performance to the most efficient telecom TFLN source (Ref. [8]) to date. At short wavelengths, the phase-matching bandwidth decreases due to group velocity dispersion in lithium niobate, but the efficiency remains high because the downconversion spectral power density scales with the inverse fifth power of the SPDC wavelength ($\lambda _s^{-5}$) [29]. Thus, in addition to the benefits of higher-energy photon pairs, decreasing the SPDC wavelength to access higher efficiencies could enable single-photon nonlinearities when integrated with a resonator [59]. The high efficiency of this device has significant implications for practical uses of entangled photons, including allowing the use of low-power laser diodes for pair generation, reducing integration times, allowing high signal-to-noise coincidence measurements at low (nW) laser powers even with losses present, and reducing fluorescence and stray light.

 

Fig. 6. Comparison of relevant literature SPDC sources to this work with respect to (a) efficiency and (b) brightness against the SPDC wavelength. Horizontal error bars represent the reported bandwidth of the sources. Data for the efficiency, brightness, center wavelength, and bandwidth are taken from Refs. [10,26,29,33,35,36,60,61] for bulk crystal sources, Refs. [11,29–32,62–67] for large waveguide sources (including micromachined and diffused waveguides), and Refs. [6–9] for TFLN sources. Note that Refs. [11,64] are not included in panel (b) since the brightness and bandwidth were not reported. The shaded region represents the typical telecommunication wavelength window.

Download Full Size | PDF

Although the efficiency and brightness of the device compares well with literature, these values can be further improved by narrowing the bandwidth of the pump laser, as discussed further in Supplement 1 Section S6. A single-frequency pump is estimated to shrink the phase-matching bandwidth from 117 nm to 35 nm, increasing the brightness by a comparable factor. The SPDC process is also most efficient near degeneracy because the group velocity of the signal and idler are equal to first order, so the efficiency could increase by a factor of 5. Furthermore, compared with SPDC produced with a monochromatic pump, the radiation produced here with a multimode pump will exhibit reduced time-energy entanglement (and greater separability) due to the uncertainty in the energy of the pump photons. The coherence properties of the pump laser are also transferred to the SPDC state, which can reduce the N00N state interference visibility at delays longer than the coherence length of the pump laser. Note that this negative effect is not observed here since the path length differences in Fig. 5 cover more than the full coherence length of the SPDC source but less than the coherence length of the pump laser [68,69], indicating that the pump laser coherence does not impair the visibilities measured in this work. Conversely, one benefit of using a multifrequency pump is that the device sensitivity to the laser wavelength is reduced, yielding a higher stability average response with greater bandwidth. Another benefit of the multifrequency pump is in time-bin entanglement, which can be generated from revivals in the biphoton interference at integer multiples of the laser cavity length [70]. For these benefits, as well as for the greater accessibility of multimode lasers, a multimode pump laser was used in this work to represent an alternate and more applied use of photon pair sources. Furthermore, despite the aforementioned drawbacks of a multimode pump, the device was still able to achieve high efficiencies on par with the current state-of-the-art telecom TFLN pair sources.

5. Conclusion

Efficient photon pair generation has been demonstrated with an integrated thin-film lithium niobate waveguide at visible and NIR wavelengths (720–900 nm). An on-chip SPDC efficiency of $(2.3\pm 0.5) \times 10^{11}$ pairs/s/mW, which is on par with reported TFLN literature at the better-explored telecom wavelengths even with a multimode pump laser, has been produced near the usually associated cut-off wavelengths for the pump (406 nm) in TFLN. The photon pair spectra has an average brightness of $(1.6 \pm 0.3) \times 10^{9}$ pairs/s/mW/nm, centered at 811 nm with a 117-nm bandwidth. To date, these results are the shortest wavelength photon pairs generated in a thin-film platform by nearly an octave. The work opens up opportunities to exploit the quantum advantage of integrated entangled photon circuits beyond telecom to imaging and spectroscopy applications in the visible and NIR.