Abstract
Practical quantum communication between remote quantum memories rely on single photons at telecom wavelengths. Although spinphoton entanglement has been demonstrated in atomic and solidstate qubit systems, the produced single photons at short wavelengths and with polarization encoding are not suitable for longdistance communication, because they suffer from high propagation loss and depolarization in optical fibres. Establishing entanglement between remote quantum nodes would further require the photons generated from separate nodes to be indistinguishable. Here, we report the observation of correlations between a quantumdot spin and a telecom single photon across a 2km fibre channel based on timebin encoding and backgroundfree frequency downconversion. The downconverted photon at telecom wavelengths exhibits twophoton interference with another photon from an independent source, achieving a mean wavepacket overlap of greater than 0.89 despite their original wavelength mismatch (900 and 911 nm). The quantumnetworking operations that we demonstrate will enable practical communication between solidstate spin qubits across long distances.
Introduction
Quantum physics can empower current network infrastructures with fundamentally new functionalities^{1} such as quantum key distribution, distributed quantum computation, quantum clock synchronization and verylongbaseline interferometry, motivating the research for a quantum internet^{2}. Essential to the quantum internet is establishing entanglement between remote quantum nodes, which is practically realized by extending spinphoton entanglement^{3,4,5,6,7,8} using indistinguishable photons^{9} in Bell state measurements^{10,11,12,13}. Once entanglement is established, information can be transferred between the nodes using quantum teleportation. However, in a quantum internet that comprises multifarious nodes and spans long distances, such a protocol may fail because of photon attenuation, whichpath information leakage, decoherence and distinguishability. First, photon attenuation over optical fibres worsens at the native wavelengths of quantum nodes with demonstrated quantum computing capability^{14}. The attenuation exponentially adds up to more than 30 dB at a node spacing of 10 km, as is commonly assumed in quantum repeater research^{1}. Second, photons that are entangled with spins may leak whichpath information associated with the energy difference between nondegenerate spin states. The eventual registration of photons then causes the spin states to mix incoherently. Third, photon polarization in standard optical fibres changes uncontrollably owing to birefringence, which is unavoidable in practice because of stress and temperature variations and renders the common polarization encoding of photonic qubits^{3,4,5,6,7} prone to decoherence^{15,16}. Converting orthogonally polarized photons with equal efficiencies would also require nontrivial engineering of frequencyconversion devices that are generally polarization dependent. Fourth, photon distinguishability due to the heterogeneity among quantum nodes hinders twophoton interference that is crucial to the remote establishment of entanglement; this problem has motivated research on both intrinsic strain^{17} and electric field^{18} tuning as well as extrinsic quantum frequency upconversion^{19,20}. However, none of these techniques can be generally applied to heterogeneous quantum nodes while simultaneously achieving long communication distances.
In the following, we address these challenges by employing four operations that are broadly identifiable as quantum networking (Fig. 1): wavelength conversion, quantum erasure, timebin encoding and mediated twophoton interference. A quantum dot (QD) generates an entangled spinphoton pair^{7,8}, whose propagating photonic component is then timebin encoded. Both basis states of the timebin qubit can be supported by a single quantum frequency downconverter, which allows for wavelength conversion to telecom wavelengths and concurrently for quantum erasure^{7} of the whichpath information in photon energy using ultrafast/broadband pulses^{7}. These timebinencoded photons at telecom wavelengths propagate in optical fibres with minimal loss and decoherence, coalescing with other photons of heterogeneous origins to exhibit twophoton interference as a result of the mediation provided by the downconverters, which render them indistinguishable in terms of wavelength and wavepacket. Upon a successful Bell state measurement, the QD spin states can eventually be swapped to a remote spin or photon. Of these four central quantumnetworking operations, we have previously demonstrated the former two^{7,21}; in this work, we demonstrate timebin encoding and mediated twophoton interference.
Results
Overview
Starting from a QD as the end point of a quantum network, we demonstrate spinphoton correlations that relate a spin qubit to a photonic timebin qubit, singlephoton downconversion that facilitates a photon to transfer quantum information, and twophoton interference between a QD photon and another photon that originates from an independent source, all through telecom wavelengths.
Spintimebin correlations
Our demonstration of longdistance spinphoton correlations was achieved based on charged InAs QDs in a magnetic field, which served as a source of entangled spinphoton pairs. Their propagating photonic components would be timebin encoded and downconverted to preserve their correlations with the QD spins across long distances. In the Voigt geometry, in which the magnetic field is perpendicular to the growth direction/optical axis, the charged QDs have the level structure shown in Fig. 2a. The level structure can be inferred from their magnetophotoluminescence spectrum, shown in Fig. 2b. The ground states contain an electron, and the excited states are the trion states, in which the added electronhole pair forms a threebody composite with the original electron. Each trion state is optically connected to the two underlying electron spin states, thereby forming a socalled Λ system, which constitutes the basis for alloptical spin control^{22}. Moreover, the QD photon that is spontaneously emitted from the excited state has been shown to be entangled with the QD spin^{7,8}; the combined spinphoton state is written in equation (1).
The entangled spinphoton pair represented in equation (1) was generated from a QD in the following sequence: the QD spin was first initialized into the state through optical pumping on the transition and was then rotated to the state with a 4pslong, 1nm reddetuned, σ^{+}polarized optical pulse. A third optical pulse, which was 30 ps long and on resonance with the transition, excited and triggered the QD to spontaneously emit the entangled photon, which could be subsequently downconverted to extend its propagation distance. This sequence concluded with spin readout, which was also performed via optical pumping and whose projection basis could be changed to by applying an additional preceding spinrotation pulse.
Our quantum downconverter, illustrated in Fig. 4a, frequency downconverted photons in the signal wavelength range of λ_{s}=900–911 nm to a target wavelength λ_{t} in the telecom Lband (1,565−1,625 nm). The core operation, a differencefrequency generation (DFG) process in a periodically poled lithium niobate (PPLN) waveguide^{21}, was pumped by a strong pulse that was independently tunable in terms of wavelength and wavepacket. The pump pulse allowed translation of λ_{t} by more than 30 nm. Unlike a continuouswave (c.w.) pump, this pulsed pump could be used to shape the target wavepacket either generically or specifically to a squared hyperbolic secant (sech^{2}) shape with nominal settings of 30 or 100 ps for our subsequent demonstrations. The pump pulse gained sufficient power for downconversion from the holmiumdoped fibre amplifiers that we adopted (Methods). We present further studies of the downconverter later in the text.
The entanglement that existed between the QD spin and the photon polarization was encoded into the spintimebin format shown in equation (2) using the two interferometers depicted in Fig. 3a. An unbalanced polarization Michelson interferometer introduced a 1.2ns difference in the optical path length (OPL) between the QD H and Vphotons. Correspondingly, the pump pulses for downconversion were also passed through an unbalanced fibre interferometer with the same OPL difference, prepared in both time bins to downconvert the QD photons. The downconverted QD photons in the s and l bin are shown in Fig. 3b together with the downconverted laser leakage, which acted as background noise. The actual QD signal was identified by comparing the cases with the spinrotation pulse switched on and off; the absence of this pulse would cause the spin to be trapped in the state and prohibit further spontaneous emission. By taking the difference between the two traces shown in Fig. 3b, we estimated the signaltonoise ratio to be 1–3 under various experimental conditions. The downconverted laser leakage originated from the finite extinction of the spininitialization pulses and entanglementgeneration pulses in their nominal ‘off’ states (Methods).
We measured the spinphoton correlations in different bases to confirm their persistence after transmission through a 2km fibre link. Before downconversion, a strong anticorrelation between the spin state and the photon H state was observed (Fig. 3d), in agreement with equation (1); the conditional probability after correction for memory effects was 0.02±0.00 (Methods). The V polarization was not measured concurrently because of the strong reflection from the resonant, entanglementgeneration pulse. After being encoded into the timebin format and downconverted, the QD photons in orthogonal states (s and l bins) could be measured concurrently because the reflection from the entanglementgeneration pulse was largely avoided via time gating: the downconversion pulse was 30 ps long and delayed with respect to the reflection by 400 ps (Fig. 3b). In agreement with equation (1), (anti)correlation between the spin state and the photon (l) s state was observed (Fig. 3e, f). The conditional probabilities and were 1.25±0.13 and 0.14±0.03, respectively, without correction; the latter deviated from because the downconverted laser leakage resulted in false photondetection events that were uncorrelated with the QD spin state. After correction for memory effects^{23} and laser leakage (Methods), the probabilities were 1.16±0.14 and 0.01±0.03, respectively. For completeness, we also measured the correlations between the spin state and the photon H, s, and l states by applying an additional spinrotation pulse before spin readout. Although these measurements further suffered from unwanted trion excitation by the spinrotation pulse, which lowered the fidelity of the subsequent spin readout^{8}, they were not essential to the entanglementswapping protocol. The overall correlation results, summarized in Fig. 3c, confirmed the persistence of spintimebin correlations after transmission through the 2km fibre link.
Backgroundfree downconversion
The communication of the QD with the next node was further facilitated using the downconverter, which could improve photon statistics and indistinguishability and mediate twophoton interference. We quantify these effects in the lowphotonnumber limit: A quantum state of light, with its photon wavepacket given as x(t), can be approximated as , where is its mean photon number and g measures its twophoton probability relative to that of a Poissonian source. The photon wavepacket creation operator, defined as with , satisfies the commutation relation . The mean wavepacket overlap, , measures the indistinguishability of photons among one another, and vanishes between photons unless they are matched in terms of both wavelength and wavepacket. When the quantum states of light are not pure, the mean wavepacket overlap can readily be shown to become , where is the density operator of the quantum state x. The overlap V approaches 1 only when both and are in pure states, in which case this generalization agrees with the original definition of V (see Supplementary Note 1 for further discussion).
Practically, to verify that single photons could be downconverted with high efficiency and without background noise, a QD was excited via pshell excitation at a wavelength of 6 nm below its main emission line at 910.85 nm (Fig. 5a). The QD photons exhibited an exponentially decaying pulse shape with a time constant of 681 ps, which were then downconverted to a wavelength of 1,610 nm and a sech^{2}shaped pulse with a time constant of 120 ps, as shown in Fig. 5b and verified in a crosscorrelation measurement (Fig. 5c; Methods). The downconverter generated 34.9k photon counts per second (c.p.s.) with 88% conversion efficiency; the overall system efficiency, 5.3%, was determined primarily by the finite transmission and the pulseduration mismatch (Methods). While converting a singlephotonlevel signal, the downconverter did not generate observable background noise; this noise level could be precisely quantified with the noisecount probability per pump pulse at peak downconversion. We lowered this probability from the level of 10^{−8} in our previous study^{21} to less than 10^{−11}. This low level of probability represents the demonstration of backgroundfree quantum frequency downconversion (QFDC; Methods).
By applying backgroundfree downconversion, the quantum downconverter could improve the photon statistics and indistinguishability of a singlephoton source, which ultimately determine the fidelity to transfer quantum information. The QD photons, both before and after downconversion, were directed to Hanbury Brown–Twiss (HBT) setups for photoncorrelation measurements (Fig. 5e). Extracted from the resulting histograms (Fig. 5e), the secondorder autocorrelation function at zero time delay, g^{(2)}(0), was improved from 0.11±0.01 to 0.03±0.02 before and after downconversion, respectively. To test the photon indistinguishability, consecutive photons^{24} were generated from the QD and subjected to twophoton interference before and after downconversion, as shown in Fig. 4b. In the resulting timecorrelated photon coincidence measurements shown in Fig. 5f, the diminished centre peaks (at a delay of 0 ns) correspond to the coincidence events that occurred when the leading (single) photon followed the long arm of the interferometer and the trailing (single) photon followed the short arm. In these events, the two photons coalesced at the second beamsplitter and were thus less likely to generate a coincidence. Further analysis based on the relative peak areas showed that the mean wavepacket overlap, V, was improved from 0.73±0.04 to 0.89±0.07 before and after downconversion, respectively (Methods). We attribute this improvement in photon statistics and indistinguishability to a timefiltering effect in the mixing of the pump pulse with the QD photons: the short pump pulses downconverted comparatively less of the unwanted background emission than they did of the QD emission, which jittered less in time; the broader bandwidth of the pump pulses also improved V by dominating the pure dephasing rate of the QD photons (Methods).
Mediated twophoton interference
Moving on to a heterogeneous quantum network, we generalize previous treatments^{18,25} of twophoton interference and mediate it using quantum downconverters. For two nodes a, b that generate quantum states of light with general g and wavepackets, we derived the normalized photon coincidence number (Methods):
where R and T are the reflection and transmission intensity coefficients of the beamsplitter. In the simplest case of R=T=1/2, and V=1, n_{c,d} goes to zero. That is, two indistinguishable single photons coalesce at a 50/50 beam splitter without generating a coincidence. More generally, both the numerator and denominator of n_{c,d} can be interpreted as a probability sum of twophoton and onephoton events weighted by the probability associated with the paths that the photons take. As long as V is nonzero, twophoton interference results in a reduction of n_{c,d}. Also commonly quoted in the literature is the twophoton interference visibility, which follows from equation (3): when R=T. The visibility does not exceed V but may further degrade because of finite g; this effect, however, could be offset by decreased .
Quantum downconverters can render photons largely indistinguishable regardless of the original separation between the photon sources in terms of wavelength, wavepacket, and distance. For illustration, an attenuated c.w. diode laser was used as a second photon source, in addition to the 911nm QD. The laser emitted at 900 nm and had a nominal 1MHz bandwidth, which was at least two orders of magnitude smaller than the QD bandwidth. Single photons from the QD and diode laser were downconverted via two quantum downconverters that were separated by a 2km fibre, as shown in Fig. 4c, to nearly the same wavelength, as shown in the highresolution spectra presented in Fig. 6a. From the spectra (Fig. 6a), bandwidths of 7.10 and 4.46 GHz could be extracted for the downconverted QD and laser photons, respectively, based on a Gaussian fit. As naively estimated from the transformlimited timebandwidth product, the bandwidth of the pump pulse (2.62 GHz) dominated those of the QD (0.23 GHz) and the diode laser and thus predominantly determined the convolved spectra of the downconverted photons. Practically, the overall bandwidth was broadened because of the technical noise of the Ti:sapphire laser and may have been further broadened by the longterm spectral diffusion of the QD. The mean wavepacket overlap V, generalized to account for the mixed states that were ensemble averaged over the spectral broadening, could theoretically have reached 0.92 based on the measured spectra, in contrast with the original zero overlap in the absence of downconversion.
Consequently, twophoton interference could be mediated between heterogeneous quantum nodes of general photon statistics. The second downconverter tuned the downconverted laser photons to several frequency detuning points; at each of these points, the normalized photon coincidence number n_{c,d} was extracted from the corresponding histogram measured using the HBT setup. The mean number of downconverted laser photons was set to to offset the degradation of visibility because of the finite . As shown in Fig. 6b, when the second downconverter eliminated the frequency detuning in the mean wavepacket overlap V, a reduction in n_{c,d} was observed as the result of twophoton interference. We arrived at quantitative bounds on the reduction in n_{c,d} by evaluating equation (3) with the experimentally determined , V and spectra (Methods). The measured n_{c,d} overall approached the lower bound closely. At the nominal zerofrequencydetuning point, the overlap V, 0.89±0.11, approached the upper bound 0.92; the corresponding twophoton interference visibility, 0.75±0.10, was primarily limited by the finite . Based on the high overlap, we conclude that the downconverter effected twophoton interference by further mitigating the degradation because of QD dephasing and longterm spectral diffusion, which has previously limited the overlap V to be ∼0.33 or less even between two wavelengthtunable QDs^{17,18}. We refer the reader to Supplementary Note 2 for further discussion on the possible causes of the nonideal wavepacket overlap.
Discussion
We have demonstrated correlation between the QD spin and the photon arrival time and mediated twophoton interference at kilometre scales. The next steps towards practical communication between quantumdot spin qubits readily follow this work. The demonstration of spintimebin entanglement would further include measurements in the offdiagonal basis: and . We have previously demonstrated complete control of a quantumdot spin over the Bloch sphere^{7,22}. For timebin qubits, we designed and realized actively stabilized interferometers (Supplementary Fig. 1; Supplementary Note 3) that maintain the phase relationship between the s and lbin photons as they traverse different optical paths during encoding, downconversion and readout. Moreover, our conversion of polarization encoding to timebin encoding circumvented the nontrivial engineering of frequencyconversion and/or fibreoptic devices that otherwise would be required. (See Supplementary Note 3 for what the nontrivial engineering entails.) The 50% signal loss at the +45° polarizer can be avoided by rotating instead of projecting the polarizations of the s and lbin photons using a fast Pockels cell. On the other hand, although our demonstration of mediated twophoton interference involved only one QD, a second QD would not be affected by the multiphoton events and bandwidth mismatch of an attenuated laser, thereby circumventing the condition that we set to preserve interference visibility. Mediating the twophoton interference between two quantum nodes is therefore within the demonstrated experimental capability.
Our backgroundfree downconversion results hold broad implications for the applicability of QFDC for quantum communication, ultimately allowing heterogeneous quantum nodes to exchange quantum information across long distances. The range of downconvertible wavelengths, beyond the 11nm range that we demonstrated, can cover the visible and nearinfrared through suitable phasematching engineering of the frequency converter^{26}. Complementarily, the available wavelength range to acquire sufficient pump power has been expanded as a result of our successful adoption of holmiumdoped fibre amplifiers, which, in turn, may indicate that other rareearthdoped fibre amplifiers that are currently under development^{27} may also be suitable for use in quantum optics experiments. The system efficiency can be improved to 27.5% based on our current device by using a longer pumppulse duration, and even further by optimizing the coupling into PPLN waveguides. Background noise can be essentially eliminated by judiciously choosing the pump wavelength to be sufficiently longer than the target wavelength, and cascaded downconversion^{28} may be adopted to increase wavelength separation if necessary (Supplementary Note 4). Photon coherence, which is implied in our study of photon indistinguishability, renders QFDC useful also in other nonentanglementswappingbased schemes^{29} for quantum repeaters. Temporal mismatch, which prohibited the efficient downconversion of the laser photons in this work, can be overcome by combining our downconverter with pulse shapers to effect quantum optical waveform conversion^{30}.
The quantumnetworking operations may further include entanglementassisted twophoton interference^{31}. In the midpointsource scheme^{31}, an entangled photonpair source facilitates the speedup for entanglement distribution, and twophoton interference occurs adjacent to each QD (Fig. 1). The source transmits many photon pairs within one roundtrip time window, thereby eliminating the communication delay and the dead time for QD operations. In this scheme, successful Bell state measurements would rely further on twophoton interference being mediated between photons of heterogeneous and even Poissonian origins, an effect that we have demonstrated. Timebinentangled photon pairs can be generated from a QD via twophoton resonant excitation^{32} in addition to converting polarization encoding to timebin encoding, or by pumping a QD^{33} or a PPLN waveguide^{34,35} with coherently superposed consecutive laser pulses between which the time delay is longer than the coherence time of the generated photons. We refer the reader to Supplementary Fig. 2 and Supplementary Note 5 for a concrete example.
In summary, we have demonstrated correlation between the QD spin and the photon arrival time, and mediated twophoton interference at kilometre scales. These quantumnetworking technologies, together with wavelength conversion^{21} and quantum erasure^{7}, will enable practical quantum communication between solidstate spin qubits across long distances.
Methods
Quantumdot spectroscopy and spin control
The QD sample and the lowtemperature, magneto, confocal microscope used in this study were similar to those used previously to investigate spinphoton entanglement^{7}; the microscope was based on a 0.68 numerical aperture aspheric lens inside a superconducting magnetic cryostat (Oxford Spectromag). The QD emission was collected into a singlemode fibre that was routed to a spectrometer for spectroscopy, singlephoton counting modules (SPCMs) for photon counting or PPLN waveguides for downconversion.
The optical spincontrol pulses were generated using three lasers as shown in Fig. 2e: A narrowband c.w. laser (New Focus Velocity), on resonance with the transition (910.416 nm at 6 T), was used for spin initialization and readout through optical pumping. One Ti:sapphire modelocked laser (Coherent Mira), centred at 911.42 nm, generated 4ps pulses to rotate the QD spin. Another Ti:sapphire modelocked laser (SpectraPhysics Tsunami), which was also used for downconversion, generated 30ps pulses to excite the QD and trigger entanglement generation. The master clock for the experiment was derived from the Mira laser, to which the Tsunami laser was synchronized through a SpectraPhysics LoktoClock system. The clock was further frequency multiplied to 10 GHz to serve as the reference frequency of a pulsepattern generator (Anritsu PPG) using a phaselocked frequency synthesizer (Valon Technologies) and an electronic quadrupler (Marki Microwave). The pulsepattern generator drove fibrebased electrooptic modulators (EOSpace) to shape and pick optical pulses from each laser; in particular, the c.w. laser for spin initialization was externally modulated to produce 4nslong pulses.
Reflected laser light was separated from the QD single photons through a combination of spatial, polarization, time and wavelength filtering. The last filtering was implemented using ultranarrow bandpass filters (Alluxa), which rejected the reflection by at least 4 orders of magnitude at a 1nm separation and by more than 6 orders of magnitude at large detuning. Resonant laser reflection, which was vertically polarized, was rejected by nearly 60 dB using a crossed polarizer^{36}.
The settings of the fibre electrooptic modulator and modelocked laser were finetuned to optimize the extinction of the spininitialization pulses and entanglementgeneration pulses in their nominal ‘off’ states. The former pulses had an extinction ratio of ∼33 dB. The latter pulses followed the temporal decay of a sech^{2}shaped pulse only within three times of the pulse duration from the pulse peak, and remained at an extinction level of ∼35 dB even 400 ps after the pulse peak. In the study of spinphoton correlations, the resulting downconverted laser leakage limited the signaltonoise ratio, which varied from 3 for the spin measurement to 1 for the spin measurement.
Downconversion
In this subsection, we present the generation and characteristics of the pump pulse. These characteristics allow evaluation of the efficiency of the waveguide DFG process. We close this subsection by summarizing our strategies to eliminate the background noise produced in the downconversion.
The strong pump pulse required in the waveguide DFG process, which was in the wavelength range of 2.04−2.1 μm, was generated using a master oscillator power amplifier configuration. We generated the seed pulses using a DFG process in a bulk MgOdoped PPLN crystal by combining the c.w. light from a 2W tunable amplified telecom laser with the optical pulses from a Ti:sapphire modelocked laser (SpectraPhysics Tsunami), whose wavelength was set according to the transition for entanglement generation or at 895 nm in the mediated twophoton interference experiment. After wave mixing, the residual 910/895nm and telecom light was filtered out using a combination of dichroic and absorptive filters. The resulting seed pulses were coupled into an optical fibre to be further amplified by holmiumdoped fibre amplifiers. The amplifier provided a gain of more than 10 dB and a maximum average power of 70 mW in the range 2.04−2.1 μm, thus ensuring that sufficient pump power could be generated to reach peak downconversion under various pulsing conditions. Crucially for singlephoton experiments, the amplifier had a large spectral separation between its opticalpumping and gain wavelengths (2.1 μm), thereby allowing for the complete removal of the residual opticalpumping light from the resulting 2.1μm light using a longpass filter on a Ge substrate.
The master oscillator power amplifier configuration also provided the downconverter with tunability in terms of pulse duration and wavelength. The pumppulse duration could be controlled by the Gires–Tournois interferometer of the modelocked Ti:sapphire laser and was selectable among three nominal settings: 3, 30 and 100 ps. We chose the nominal 30ps setting for entanglement generation, or the 100ps setting for mediated twophoton interference when the largest photon yield was desired. The pump wavelength was determined in the bulk DFG process and was therefore tunable through tuning of the telecom laser.
The (2.1μm) pump pulse characteristics were verified in a crosscorrelation measurement, where the 910nm light was divided to generate 2.1μm light via DFG inside the bulk PPLN crystal, as well as to inject directly into the waveguide. The 910nm and 2.1μm pulses then mixed in the waveguide; the DFG output was monitored while the arrival time of one of the inputs was scanned relative to the other. The crosscorrelation results are shown in Fig. 5c for a pulse duration setting of 100 ps. With this setting, the original sech^{2}pulseshape was largely preserved; a deconvolution indicated a pulse duration of 120 ps.
We used fibrepigtailed reverse protonexchange PPLN waveguides to achieve high efficiency for singlephoton downconversion experiments^{21} (Supplementary Note 6). In the waveguide DFG process, the nonlinear conversion efficiency η depends on the pump power P_{p}^{37} as follows:
where P_{max} is the pump power required for complete conversion. For corresponding illustration, we observed a DFGsignal peak when increasing the pump power as shown in Fig. 5d, thereby determining the sufficiency of pump power for downconversion. Practically, although the entire curve as shown in Fig. 5d (and also as shown in Supplementary Fig. 3) was not taken in every experimental run, a power sweep near the peak was always performed to ensure the most efficient downconversion, which occurred at a peak pump power (in the input fibre) of ∼1 W. We also mention that the data in Fig. 5d do not represent the best coupling of the pump power, therefore P_{max} corresponds to a peak pump power higher than 1 W.
We calculated the limit on conversion efficiency because of the duration mismatch between the QD pulse and the pump pulse. For this calculation, the pump pulse shape was substituted into equation (4) to arrive at a timeresolved conversion efficiency, with the assumption that the (instantaneous) efficiency approached unity at the peak of the pump pulse. An overlap integral could then be calculated based on the measured QD pulse shape (in Fig. 5b), and the inferred 120ps time constant in the crosscorrelation measurement of the sech^{2}shaped pump pulse (in Fig. 5c). From this calculation we found the conversion efficiency, limited by pulse duration mismatch, to be 22%, which was notably higher than the ratio between the corresponding time constants, 18%. The difference mainly resulted from the saturation behaviour of equation (4), leading to a conversion window wider than 120 ps.
The overall system downconversion efficiency, which was determined to be 5.3% from the measured 911nm and 1,610nmphoton count rates, was accounted for by other experimentally determined values: the finite time overlap between the QD pulse and the pump pulse (22%) and the transmission efficiency of the waveguide and filters (27.5%), which could be further subdivided into the input coupling and transmission (71%) at 910 nm, the output coupling and transmission (71%) at 1,610 nm, and the filtering setup transmission (55%). We thus estimated that the internal conversion efficiency approached 88%.
To eliminate the background noise produced in the downconversion, we adopted the following strategies: choosing the pump wavelength to be sufficiently longer than the target wavelength^{21}, restricting the detection bandwidth, and promptly removing the strong pump light from the downstream optics of the waveguide. The implemented filtering setup, which rejected the pump photons by 18 orders of magnitude in total, included a fibre Bragg grating, a longpass filter on a Si substrate and a fibrebased 2μm filter. (See Supplementary Fig. 4 and Supplementary Note 4 for the identification of the origins of the noise.)
Singlephoton counting and data analysis
Two models of SPCMs were used to detect 910nm single photons. The Micro Photon Devices model (PDM series) had a timing jitter of ∼48 ps and a quantum efficiency of ∼2% at 910 nm and was used to measure the 910nm photon pulse shape shown in Fig. 5b. Two additional PerkinElmer (SPCMAQRH14) SPCMs were used for all other 910nm photoncounting tasks during the experiment. They both had a quantum efficiency of ∼30% at 910 nm, but they had different timing jitters of ∼430 and ∼780 ps.
The telecom photons were detected using superconducting nanowire singlephoton detectors (SNSPDs)^{38}. The two SNSPDs, which were maintained at 2 K, had system detection efficiencies of 6 and 20%, ungated dark count rates of 100 and 200 Hz and fullwidthhalfmaximum timing jitters of 130 and 190 ps. Timecorrelated singlephoton counting was performed using a time interval analyser (Hydraharp, PicoQuant GmbH) in the timetagged timeresolved mode. Because the events registered by both the SPCMs and the SNSPDs were accurately time gated during postprocessing, the dark counts of the detectors were essentially negligible in this work. The finite number of coincidences, however, resulted in an uncertainty in determining the conditional probabilities as well as g and V due to Poissonian statistics.
The spinphoton correlation data were obtained from the histograms (for example, see Fig. 3d–f) of the coincidence counts between the downconverted single photons and the single photons that were used for spin readout. The coincidence counts were obtained through postprocessing of the timetagged timeresolved data stream. In principle, by comparing the coincidence counts registered within the same experimental cycle with those registered in subsequent, uncorrelated cycles, the conditional probability of detecting a spin state given the detection of a photon state could be calculated^{7}. For the particular QD that we studied, memory effects resulted in additional positive correlations (blinking^{23}) among the QD photon emissions that needed to be corrected for. We fit the binned coincidence counts C[n] in the delay range of (−528, 1,531.2) ns using a twosided exponential function plus a background:
where T_{rep} is the period of one experimental cycle (≈52.8 ns); α and are fitting parameters that characterize the amplitude and time scales, respectively, of the memory effect; A accounts for the (uncorrelated) coincidences of actual QD photons; and B accounts for the false coincidences caused by the downconverted laser leakage. The extrapolation of the fitting to C[t→0] then provided a spinindependent factor that could be used to properly normalize the conditional probabilities, which otherwise would have exceeded 1 in the case of . The ratio of the fitted value of A to that of B was also found to be consistent with the monitored signaltonoise ratio preceding each extended run of photon counting. We also verified that the memory effect was not caused by incomplete spin initialization: extending the duration of the spininitialization pulse from 4 to 13 ns and beyond did not mitigate this effect.
Twophoton interference
In this subsection, we first summarize the experimental conditions that were common to both of the twophoton interference experiments: the one between consecutive photons and the other between heterogeneous nodes. For the former, we then explain how the reported mean wavepacket overlap V was extracted from the histograms in Fig. 5f. For the latter, besides the experimental settings, we explain how the normalized photon coincidence number in equation (6) was derived, and how its reduction was limited by a wavepacket mismatch.
All the twophoton interference experiments were performed using polarizationmaintaining fibre beamsplitters to ensure good polarization and mode matching among interfering photons. To generate indistinguishable single photons from the QD, the excitation power was chosen to excite the QD only at 80–90% of its saturation count rate under pshell excitation such that the QD photon coherence could be largely preserved.
For twophoton interference between consecutive photons, the relative peak areas of the five peaks within each central cluster in Fig. 5f were determined by the probability aggregated over the various possible paths that the two photons could take. Each possible path was defined by whether the leading or trailing photon travelled along the short or long arm of the interferometer and then triggered the START or STOP detector. The total number of such possibilities, along with the light properties g and V, then determined the relative peak areas, as shown in the following equations:
where N is the number of repetitions, η^{(2)} is the combined efficiency of twophoton generation and detection, and R_{I}, R_{II}, T_{I} and T_{II} are the intensity coefficients of reflection and transmission of the first and second beamsplitters. These equations generalize the previous approach^{24}, accounting for the possibility that the two beamsplitters in the Mach–Zehnder interferometer may have different intensity coefficients. The parameter V(Δt) in the expression for A_{3} represents the mean overlap between the wave packets of the two consecutive photons at a time difference Δt. Using these equations, the overlap V between consecutive photons was extracted using the experimentally determined values of g, T_{I} and T_{II}.
To apply equation (5) to the data presented in the top panel of Fig. 5f, it was necessary to further extract the areas of individual peaks, which overlapped with one another because of the QD emission tail and the limited time response of the SPCMs used in this experiment. We fit the histogram as a series of peak functions separated by a time delay of 2.6 ns, as illustrated in Supplementary Fig. 5; each peak function, resulting from the START–STOP measurements, was a convolution of a twosided decaying exponential^{39} (the QD emission) with a Gaussian distribution (of the SPCM time response).
The areas of the five central peak functions could then be used to extract V based on equation (5), together with our measured g^{(2)}(0) in Fig. 5e. In the test of the indistinguishability of consecutive photons before downconversion (corresponding to the top panel of Fig. 5f), the fibre interferometer at 910 nm exhibited T_{I}=0.77 and T_{II}=0.43. With the measured g being 0.11±0.01, we could estimate V at zero time difference to be 0.73±0.04, which was affected by pure dephasing on top of the lifetimelimited bandwidth of the QD photons, 0.23 GHz (see Supplementary Note 2 for further discussion on the possible causes of the nonideal wavepacket overlap). For the test of the indistinguishability of consecutive photons after downconversion (corresponding to the bottom panel of Fig. 5f), we performed a similar estimation and arrived at V=0.89±0.07, using the experimentally determined values of T_{I}=0.50, T_{II}=0.46 and g=0.03±0.02. In addition, we note that the asymmetry among the five peaks in each cluster in Fig. 5f was due to the unbalanced splitting ratio of the fibre beamsplitters.
For twophoton interference between heterogeneous nodes, we used the HBT setup to quantify their properties such as , g and V. The schematic of an HBT setup is shown in Fig. 4c, in which the input ports are labelled as , and the output ports as , . The HBT setup measures the normalized photon coincidence number
where and represent the bra and ket of a generic state, respectively. We assume two quantum states of light entering the two ports of the beamsplitter separately, that is,
By applying the beamsplitter relationships and , the normalized photon coincidence number as shown in equation (6) is then calculated to obtain equation (3). Equation (3) not only incorporates all the previous scenarios^{18,25} in which the interference between two QDs, a QD and a laser, and two lasers have been studied but also treats the cases of finite g and superPoissonian sources, provided that the average photon number is much less than one.
During the QDlaser interference experiment, the sample was temperature stabilized at 2 K to stabilize the QD emission wavelength. To avoid any wavelength mismatch, the wavelengths of all lasers were verified using a wavemeter (Burleigh WA1100) to within a resolution of 1 pm, and are tabulated in Supplementary Table 1. The spectra of the downconverted photons were further measured using a fibre Fabry–Pérot interferometer with a bandwidth of ∼1 GHz. The second telecom laser was then detuned by ±3 and ±10 GHz, and the normalized photon coincidence number n_{c,d} at each of the corresponding frequencies was extracted from the corresponding histograms, examples of which are provided in Supplementary Fig. 6.
We analysed the reduction in n_{c,d} in Fig. 6b by evaluating equation (3) using the experimentally determined , V and spectra to arrive at quantitative bounds. Given only the amplitude but not the phase of the wavepackets from the measured spectra in Fig. 6a, we obtained the following upper bound on the overlap V versus the frequency detuning:
where I_{a,b}(f) is the normalized intensity spectrum of , such that . This upper bound is physically motivated by the wavepacket overlap between two (pure) states of constant phase that have individual wavepacket amplitudes of and . When the upper bound is approached, dephasing is negligible and the overlap V is limited only by the amplitude mismatch between the spectra. This upper bound on the overlap V also sets a corresponding lower bound on n_{c,d} according to equation (3). These bounds, which are also plotted in Fig. 6b, were adjusted only by an overall frequency shift (0.53 GHz) to offset the finite frequency resolution (1 GHz) of the measurement. The measured n_{c,d} overall approached the lower bound closely.
Additional information
How to cite this article: Yu, L. et al. Twophoton interference at telecom wavelengths for timebinencoded single photons from quantumdot spin qubits. Nat. Commun. 6:8955 doi: 10.1038/ncomms9955 (2015).
References
 1
Meter, R. V. Quantum Networking WileyISTE (2014) .
 2
Kimble, H. J. The quantum internet. Nature 453, 1023–1030 (2008) .
 3
Blinov, B., Moehring, D., Duan, L. & Monroe, C. Observation of entanglement between a single trapped atom and a single photon. Nature 428, 153–157 (2004) .
 4
Volz, J. et al. Observation of entanglement of a single photon with a trapped atom. Phys. Rev. Lett. 96, 030404 (2006) .
 5
Wilk, T., Webster, S. C., Kuhn, A. & Rempe, G. Singleatom singlephoton quantum interface. Science 317, 488–490 (2007) .
 6
Togan, E. et al. Quantum entanglement between an optical photon and a solidstate spin qubit. Nature 466, 730–734 (2010) .
 7
De Greve, K. et al. Quantumdot spinphoton entanglement via frequency downconversion to telecom wavelength. Nature 491, 421–425 (2012) .
 8
Gao, W. B., Fallahi, P., Togan, E., MiguelSanchez, J. & Imamoglu, A. Observation of entanglement between a quantum dot spin and a single photon. Nature 491, 426–430 (2012) .
 9
Hong, C., Ou, Z. & Mandel, L. Measurement of subpicosecond time intervals between two photons by interference. Phys. Rev. Lett. 59, 2044–2046 (1987) .
 10
Moehring, D. L. et al. Entanglement of singleatom quantum bits at a distance. Nature 449, 68–71 (2007) .
 11
Hofmann, J. et al. Heralded entanglement between widely separated atoms. Science 337, 72–75 (2012) .
 12
Ritter, S. et al. An elementary quantum network of single atoms in optical cavities. Nature 484, 195–200 (2012) .
 13
Bernien, H. et al. Heralded entanglement between solidstate qubits separated by three metres. Nature 497, 86–90 (2013) .
 14
Ladd, T. D. et al. Quantum computers. Nature 464, 45–53 (2010) .
 15
Brodsky, M., George, E. C., Antonelli, C. & Shtaif, M. Loss of polarization entanglement in a fiberoptic system with polarization mode dispersion in one optical path. Opt. Lett. 36, 43–45 (2011) .
 16
Antonelli, C., Shtaif, M. & Brodsky, M. Sudden death of entanglement induced by polarization mode dispersion. Phys. Rev. Lett. 106, 080404 (2011) .
 17
Flagg, E. B. et al. Interference of single photons from two separate semiconductor quantum dots. Phys. Rev. Lett. 104, 137401 (2010) .
 18
Patel, R. B. et al. Twophoton interference of the emission from electrically tunable remote quantum dots. Nat. Photonics 4, 632–635 (2010) .
 19
Takesue, H. Erasing distinguishability using quantum frequency upconversion. Phys. Rev. Lett. 101, 173901 (2008) .
 20
Ates, S., Agha, I., Gulinatti, A. & Rech, I. Twophoton interference using backgroundfree quantum frequency conversion of single photons emitted by an InAs quantum dot. Phys. Rev. Lett. 109, 147405 (2012) .
 21
Pelc, J. S. et al. Downconversion quantum interface for a single quantum dot spin and 1550nm singlephoton channel. Opt. Express 20, 27510–27519 (2012) .
 22
Press, D., Ladd, T. D., Zhang, B. & Yamamoto, Y. Complete quantum control of a single quantum dot spin using ultrafast optical pulses. Nature 456, 218–221 (2008) .
 23
Santori, C. et al. Submicrosecond correlations in photoluminescence from InAs quantum dots. Phys. Rev. B 69, 205324 (2004) .
 24
Santori, C., Fattal, D., Vucković, J., Solomon, G. S. & Yamamoto, Y. Indistinguishable photons from a singlephoton device. Nature 419, 594–597 (2002) .
 25
Bennett, A. J., Patel, R. B., Nicoll, C. A., Ritchie, D. A. & Shields, A. J. Interference of dissimilar photon sources. Nat. Phys. 5, 715–717 (2009) .
 26
Langrock, C., Kumar, S., McGeehan, J., Willner, A. & Fejer, M. Alloptical signal processing using χ^{(2)} nonlinearities in guidedwave devices. J. Lightwave Technol. 24, 2579–2592 (2006) .
 27
Jackson, S. D. Towards highpower midinfrared emission from a fibre laser. Nat. Photonics 6, 423–431 (2012) .
 28
Pelc, J. S. et al. Cascaded frequency upconversion for highspeed singlephoton detection at 1550 nm. Opt. Lett. 37, 476–478 (2012) .
 29
Munro, W. J., Stephens, A. M., Devitt, S. J., Harrison, K. A. & Nemoto, K. Quantum communication without the necessity of quantum memories. Nat. Photonics 6, 777–781 (2012) .
 30
Kielpinski, D., Corney, J. F. & Wiseman, H. M. Quantum optical waveform conversion. Phys. Rev. Lett. 106, 130501 (2011) .
 31
Jones, C., De Greve, K. & Yamamoto, Y. A highspeed optical link to entangle quantum dots. Preprint at http://arxiv.org/abs/1310.4609v1 (2013) .
 32
Müller, M., Bounouar, S., Jöns, K. D., Glässl, M. & Michler, P. Ondemand generation of indistinguishable polarizationentangled photon pairs. Nat. Photonics 8, 224–228 (2014) .
 33
Jayakumar, H. et al. Timebin entangled photons from a quantum dot. Nat. Commun. 5, 4251 (2014) .
 34
Tanzilli, S. et al. PPLN waveguide for quantum communication. Eur. Phys. J. D Atomic Mol. Opt. Phys. 18, 155–160 (2002) .
 35
Zhang, Q. et al. Generation of 10GHz clock sequential timebin entanglement. Opt. Express 16, 3293–3298 (2008) .
 36
Kuhlmann, A. V. et al. A darkfield microscope for backgroundfree detection of resonance fluorescence from single semiconductor quantum dots operating in a setandforget mode. Rev. Sci. Instrum. 84, 073905 (2013) .
 37
Pelc, J. S. et al. Longwavelengthpumped upconversion singlephoton detector at 1550 nm: performance and noise analysis. Opt. Express 19, 21445–21456 (2011) .
 38
Tanner, M. G. et al. Enhanced telecom wavelength singlephoton detection with NbTiN superconducting nanowires on oxidized silicon. Appl. Phys. Lett. 96, 221109 (2010) .
 39
He, Y.M. et al. Ondemand semiconductor singlephoton source with nearunity indistinguishability. Nat. Nanotechnol. 8, 213–217 (2013) .
Acknowledgements
We thank Darin Sleiter, Qiang Zhang, Naoto Namekata, Shuichiro Inoue, Mingwu Lu, Alireza Marandi, Nobuyuki Takei, Hayato Goto, Akira Ozawa and Hiroki Takesue for valuable discussions, comments and technical assistance. We gratefully acknowledge Valery Zwiller and Sander Dorenbos at TU Delft, the Netherlands, for providing the superconducting nanowire samples used. We thank Paul Hansen for performing a close reading of the manuscript. This work was supported by the JST through its ImPACT Program, NICT, NSF CCR08 29694, NIST 60NANB9D9170, Special Coordination Funds for Promoting Science and Technology, and the State of Bavaria. C.L. and M.M.F. acknowledge support through the AFOSR. C.M.N. acknowledges a SU2P Entrepreneurial Fellowship and R.H.H. acknowledges a Royal Society University Research Fellowship.
Author information
Affiliations
Contributions
S.M., C.S., M.K. and S.H. grew and fabricated the samples. L.Y. and Y.Y. designed the experiment. L.Y., C.M.N. and T.H. performed the optical experiments. J.S.P. designed and fabricated the PPLN waveguides. L.Y., C.M.N., C.L. and J.S.P. developed the 2.2μm setup and the 1,610nm filtering design. C.M.N., M.G.T. and R.H.H. packaged, characterized and implemented the SNSPD detectors. Y.Y., M.M.F., R.H.H. and S.H. guided the work. L.Y. and C.M.N. wrote the manuscript with input from all authors.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing financial interests.
Supplementary information
Supplementary Information
Supplementary Figures 16, Supplementary Table 1, Supplementary Notes 16 and Supplementary References (PDF 2579 kb)
Rights and permissions
This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/
About this article
Cite this article
Yu, L., Natarajan, C., Horikiri, T. et al. Twophoton interference at telecom wavelengths for timebinencoded single photons from quantumdot spin qubits. Nat Commun 6, 8955 (2015). https://doi.org/10.1038/ncomms9955
Received:
Accepted:
Published:
Further reading

Twophoton interference in the telecom Cband after frequency conversion of photons from remote quantum emitters
Nature Nanotechnology (2019)

Highfidelity entanglement between a trapped ion and a telecom photon via quantum frequency conversion
Nature Communications (2018)

Midinfrared coincidence measurements on twin photons at room temperature
Nature Communications (2017)

Polarisationpreserving photon frequency conversion from a trappedioncompatible wavelength to the telecom Cband
Applied Physics B (2017)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.