A bright source of telecom single photons based on quantum frequency conversion

Semiconductor quantum dots (QDs) are a leading technology for bright, indistinguishable single-photon sources. QDs emitting in the 920 nm–980 nm window have been used as a source of single photons with count rates upward of 10 MHz, indistinguishability greater than 95%, and low multi-photon contributions on the order of 1%.^1–3 Yet, in order to be compatible with existing telecommunication technology, an ideal single-photon source should operate in the telecommunication C-band, around 1550 nm, where fiber loss is minimal. While there has been recent progress in producing QD sources that directly emit single photons in the C-band,^4–6 achieving high-quality single-photon emission at high rates remains an open challenge, with the best-performing single-photon sources to date being confined to near-infrared (NIR)⁷ wavelengths.

One route to bridge the gap to the C-band is quantum frequency conversion (QFC), converting single photons from an NIR QD to telecommunication wavelengths. This can be done in a nonlinear three-wave mixing process where a single-photon input is mixed with a strong seed beam producing a single-photon output at either the sum or difference frequency. QFC can, in principle, be noise free and therefore preserve the quantum statistics of near-infrared emitters. Initial experiments on quantum dot QFC focused on upconversion to leverage superior detectors at NIR wavelengths.⁸ This allowed the demonstration of noise free conversion and two-photon interference between independent dots at visible wavelengths.⁹ Frequency downconversion of QD sources has been demonstrated from 700 nm to the telecommunication O-band¹⁰ and from 900 nm to the C-band,^11,12 culminating in the remote two-photon interference between independent downconverted QD sources.¹³ Spin-photon entanglement between a 910 nm QD and a 1560 nm photon has been demonstrated using frequency conversion seeded by ultrafast pulses.^14,15 QFC of QDs has also been demonstrated in nano-photonic circuits using four-wave mixing in silicon nitride.¹⁶ However, a single photon source that is simultaneously bright, pure, and coherent has not been demonstrated at telecom wavelengths. Here, we demonstrate a frequency-converted InGaAs quantum dot source approaching 1 MHz count rates at 1550 nm, with $g(2)(0)$ around 4% and Hong-Ou-Mandel (HOM) visibilities of 60%.

For difference-frequency generation, energy conservation demands $(1/λin−1/λseed)−1=λout$⁠. Our 942 nm InGaAs QD source requires a seed wavelength of 2401 nm to generate output photons at 1550 nm. The seed beam is produced in a chromium doped zinc selenide (Cr:ZnSe) laser, while the difference frequency generation occurs in a periodically poled lithium niobate (ppLN) waveguide. Near-infrared photons are generated by a single self-assembled InGaAs/GaAs QD coupled to a high quality (⁠$Q≈4.4×104$⁠) oxide-apertured micropillar cavity, as shown in Fig. 1. The QDs are embedded in a p–i–n diode structure,^17,18 which enables charge control and tuning of the QD emission to the cavity mode via the quantum-confined Stark effect. The sample is kept at a temperature of $4 K$ in a closed-cycle helium flow cryostat. A dark-field confocal microscope is used to excite and collect the scattered photons from the QD while filtering in a cross-polarization scheme with an ≃$107$ extinction ratio to suppress the excitation laser.

Pulsed excitation of the QD is performed using a mode-locked titanium:sapphire laser with a repetition rate of 80.3 MHz and a pulse duration of 10 ps. The QD output is then detected using superconducting nanowire single-photon detectors (SNSPDs), with a nominal quantum efficiency of ≃$90 %$ at 950 nm. We exploit a neutral exciton QD transition (X⁰), resonantly coupled to the cavity with a Purcell factor of $≈4$ and an emission wavelength of 942.33 nm. The T₂ coherence time of the emission, measured using standard Fourier spectroscopy under π-pulse resonant excitation, shows $T2=0.348 (2) ns$⁠, corresponding to a linewidth of $915 (5) MHz$⁠. This value is $≈1.5$ times larger than the transform-limited linewidth (⁠$h/T1=607 MHz$⁠), with an independently measured lifetime of $T1=0.2622 (1) ns$⁠.

The seed laser for the QFC stage consists of a z-cavity resonator with a Cr:ZnSe crystal, a gain medium with an emission spectrum spanning 1900–3300 nm^19–21 (see inset in Fig. 1). This laser design allows for both continuous-wave²² and mode-locked operation generating pulses as short as 43 fs.²³ Here, we operate with a narrowband CW seed to drive the QFC. Our 2401 nm laser system is pumped by a thulium-doped fiber laser (IPG Photonics TLR-20-LP), which has a maximum CW output power of 20 W at 1900 nm. The pump light is focused into the Cr:ZnSe crystal using a 100 mm $CaF2$ lens. The cavity consists of a dichroic input coupler (50 mm radius of curvature (ROC), transmissive at 1900 nm, and reflective at 2400 nm), a gold mirror (50 mm ROC), two-plane silver mirrors, and the Cr:ZnSe crystal. The crystal is placed at the Brewster angle to minimize losses due to reflection. A diffraction grating (450 lines/mm) is inserted into the cavity to control the emission wavelength. The output coupler (Layertec) has a transmission of 60% at 2401 nm, allowing a good trade-off between the cavity enhancement and available output power.

Figure 1 shows our difference-frequency generation (DFG) setup. The DFG, one special case of QFC, takes place in a 48 mm ppLN crystal (NTT electronics). The chip contains multiple ridge waveguides with poling periods ranging from 26.00 μm to 26.25 μm. These poling periods are designed for type-0 DFG from 942 nm to 1550 nm. Quarter- and half-wave plates are used to align the polarization of the incoming single photons at 942 nm and the generated pump light at 2401 nm to the extraordinary axis of the crystal. Seed light from the laser is overlapped with single photons from the QD using a dichroic mirror (Omega Optical) and coupled into the waveguide using an NIR-coated aspheric lens with a focal length of 11 mm. A NIR-coated lens is used to match the beam size of the single photons to the seed beam and to compensate for the chromatic aberration of the aspheric lenses. The converted 1550 nm light is collimated with an 11 mm NIR-coated aspheric lens and sent toward a filtering stage.

The filtering stage consists of two shortpass filters at 2050 nm (>OD 4), which are used to remove seed light impinging on the collection fiber; a longpass filter at 1400 nm (>OD 5) to remove weakly phase-matched second-harmonic generation from the seed beam and unconverted quantum dot light; and finally, a 2.8 nm full-width-at-half-maximum (FWHM) bandpass filter (>OD 4) to isolate the converted single photons. The converted 1550 nm light is collected into a single-mode fiber with a coupling efficiency of 86% and sent to SNSPDs with a nominal quantum efficiency greater than 80%.

The DFG conversion efficiency is characterized by sending CW coherent light from a 942 nm laser (Toptica DL Pro) into the QFC setup. For CW-seeded QFC, the conversion efficiency is almost independent of the temporal mode of the input light.²⁴ This allows characterization with a CW beam despite the single photons' decaying exponential wavepacket. Under the approximation that the seed beam is unamplified and has considerably higher intensity than all other modes, QFC is expressed as a beam-splitter Hamiltonian between different frequency modes.²⁵ These two factors ensure that the conversion efficiency measured with a low power $(500μW)$ CW coherent field is equivalent to the single-photon conversion efficiency.

Figure 2 shows the internal conversion efficiency for 942 nm light, measured by comparing output 1550 nm light with the 942 nm coupled through the waveguide with the seed laser blocked. This factors outcoupling losses into the waveguide, which are measured to be 17%. The data is fitted with²⁷ 

where η_max is the maximum possible conversion efficiency, η_nor is the normalized conversion efficiency of the process, P is the input power, and L is the waveguide length. The fit gives a normalized conversion efficiency (to waveguide length in the limit of small pump powers²⁷) of $ηnor=44(1) % /(W cm2)$⁠. The maximal external conversion efficiency, the ratio of photons collected into single-mode fiber after the conversion stage to the number of NIR photons impinging on the waveguide, is $ηmax=38 %$⁠, leading to a maximum internal conversion efficiency of $56.7(4) %$ when taking losses into account. This η_max is higher than previously reported values for NIR QD frequency conversion to 1550 nm with similar waveguides.¹³ We would like to highlight that we achieve SNRs > 250 for all seed powers (inset in Fig. 2), meaning that the noise contribution of the DFG process toward the converted single photons is minimal.

We now compare the characteristics of the converted telecom photons with the QD NIR photons. We tune the QD into resonance with the first cavity mode and excite either resonantly or non-resonantly into the third cavity mode; see Fig. 3 for spectral properties of the cavity under 820 nm excitation. The inset in Fig. 3 shows the detected count rates for these two excitation scenarios as a function of power: Rabi oscillations are observed for resonant excitation, while a clear maximum is observed for non-resonant excitation. For non-resonant characterization, the QD photons are spectrally filtered with a grating filter with a 30 GHz FWHM to suppress the excitation laser. We detect a count rate of $1.85(5) MHz$ at an excitation power of $6.8 μW$⁠. The grating filter was removed when characterizing the converted photons as low-loss bandpass filters were used at 1550 nm. For resonant driving, we optimize the excitation power to the π-pulse and detect a count rate of $1.46(4) MHz$⁠. This value is slightly lower than for off-resonant excitation due to the presence of spectral fluctuations.²⁸ After QFC, the detected count rate at 1550 nm, for the off-resonant and on-resonant case, is $856(18) kHz$ and $456(14) kHz$⁠, respectively. Comparing the NIR and telecom counts under resonant excitation gives an end-to-end conversion efficiency of $≈35 %$⁠, after accounting for the difference in the detection efficiency of both NIR (⁠$≈90%$⁠) and telecom C-band (⁠$≈80%$⁠) detectors. This agrees well with the measured loss budget through the optical components including the conversion efficiency. The difference in efficiency for off-resonant excitation is accounted for by the loss of the grating filter.

Figure 4 shows the comparison between the performance of the QD signal before and after QFC. The lifetime measured under resonant excitation in Fig. 4(a) remains unchanged within experimental error after conversion. The oscillation in the time-resolved emission, indicative of the quantum beating of the X⁰ fine-structure splitting, shows a frequency of $4.807 (3) GHz$ extracted from a fit of $exp (−t/T1)(1+A sin (ΔfSSt))$ to the data.²⁹ The equivalent oscillation after the QFC process is unchanged (⁠$ΔfSS=4.803 (1) GHz$⁠), indicating that the CW-seeded frequency conversion preserves the temporal mode of the input photons.

Next, we measure the second-order intensity correlation $g(2)$ using a Hanbury-Brown and Twiss (HBT) interferometer. For a perfect single-photon source, $g(2)(0)=0$⁠, indicating the absence of multi-photon emissions. Under off-resonant driving, Fig. 4(b), we observe a slight increase from $g(2)(0)=0.045 (0)$ to $g(2)(0)=0.051 (1)$ before and after the QFC process, respectively. We observe similar values under resonant driving, Fig. 4(c), demonstrating near-ideal single-photon emission with $g(2)(0)=0.040 (0)$ and $g(2)(0)=0.043 (1)$ before and after the QFC process, respectively. The slight increase in the normalized coincidences in the uncorrelated side peaks in the HBT histogram is due to blinking of the emitters, a common effect resulting from QD coupling to the solid-state charge environment.³⁰ The imperfection in $g(2)(0)$ can be due to imperfect suppression of the cavity emission due to cavity feeding,^31–33 slight imperfection in the wave-plate retarders used in our confocal microscope, and the presence of multi-photon capture processes.^34,35 Nevertheless, with a modest increase in $g(2)(0)$ after the QFC process, we have demonstrated near background-free single-photon frequency conversion from the NIR to telecom C-band, with the photon-number purity predominately limited by the quantum dot.

To demonstrate that our QFC setup preserves photon coherence, we perform HOM interference between photons emitted from two consecutive excitation pulses. We use an unbalanced Mach–Zehnder interferometer with a delay of 12.5 ns to match photons temporally on a 50/50 beam splitter. We measure the coincidence counts for parallel and perpendicular polarized photons and evaluate the visibility as $VHOM=1−g∥(2)/g⊥(2)$⁠. For a pair of indistinguishable photons, $VHOM=1$⁠. For resonant excitation, we achieve an interference visibility of $VHOM=0.88 (1)$ before QFC. We expect a reduced $VHOM→T2/2T1≈0.66$ at longer delays, inferred from the coherence time, T₂ measured using a Michelson interferometer.^36,37 We calculate the single-photon indistinguishability M_s as $Ms=(VHOM+g(2)(0))/(1−g(2)(0))$⁠.³⁸ This gives an upper bound to the HOM visibility taking the finite $g(2)(0)$ into account. Before conversion, the $Ms$ value is equal to $0.95 (1)$⁠. After conversion, we find the raw visibility and corrected indistinguishability to be $0.60 (1)$ and $0.67 (2)$⁠, respectively. The results of lifetime, HBT, and HOM measurements are summarized in Table I.

The reduced interference visibility originates in spectral instability introduced by fluctuating power in multiple longitudinal modes of the seed laser. The linewidth of the seed laser is around 4 GHz with a free spectral range estimated to be 177 MHz, corresponding to approximately $22$ modes. Despite this, we show that our QFC setup indeed preserves the coherence of single photons as measured from HOM interference, which can be improved by increased control over the cavity dispersion and active stabilization of the cavity length.

Modest improvements to the current QFC system will allow us to improve the converted two-photon interference visibility to equal the unconverted visibility. The external conversion efficiency could be further improved with lower-loss filtering and improved mode matching between the single-mode fiber and the waveguide mode.

Quantum-dot sources emitting directly at telecom wavelengths have demonstrated $g(2)(0)≃4×10−4$ with count rates on the order of 200 kHz³⁹ but have not demonstrated the HOM interference visibilities of frequency-converted sources,^13,40 which typically have lower brightness and higher noise contributions. While a fair comparison between our source and other works is difficult to make due to the incomplete information in the reported data, we believe that the combined brightness, multiphoton suppression, and indistinguishability demonstrated by this source are marked improvements over other telecom-wavelength quantum-dot sources. This source will find applications in fiber-based quantum communication where a source of bright and highly pure single photons in the C-band is required. This can lead to demonstrations of various quantum communication protocols including measurement device-independent quantum key distribution, teleportation, and entanglement swapping between distant quantum nodes.

We acknowledge Pierre M. Petroff for his contribution to the sample design and fabrication. This work was supported by the EPSRC (Grants Nos. EP/L015110/1, EP/M013472/1, EP/P029892/1, EP/N002962/1, and EP/T001011/1), the ERC (Grant No. 725920), and the EU Horizon 2020 research and innovation program under Grant Agreement No. 820423. B.D.G. thanks the Royal Society for a Wolfson Merit Award and the Royal Academy of Engineering for a Chair in Emerging Technology. Y.M. acknowledges the support from the Chongqing Research Program of Basic Research and Frontier Technology (No. cstc2016jcyjA0301). The authors in K.I.S.T. acknowledge the support from the KIST institutional program, the program of quantum sensor core technology through IITP, and the IITP grant funded by the Korean Government (MSIT) (No. 20190004340011001).