## Introduction

The success or failure of quantum light sources in advanced commercial photonic quantum-network applications is strongly dependent on integrability of these sources with current infrastructure and technology. Regarding networks, operation at telecom wavelength is as essential as wavelength tuneability, required for interfacing and multiplexing with other sources of classical or quantum light over the same optical fibre. In terms of quantum light emitters, scalable manufacturing techniques for their production as well as compliance with low voltage driving electronics for safe and long-term reliable operation in remote non-laboratory environments are most desirable.

Shared entanglement between distant network users is an important resource for quantum-network applications going beyond conventional quantum key distribution (QKD)1 that makes use of weak coherent laser pulses2,3,4. Greater robustness to photon number splitting attacks5,6, an increase of communication distances7 and the prospect to link remote quantum processors8 are some of the immediate benefits. As currently employed sources are mostly based on spontaneous processes, intrinsically limiting efficiencies, further improvement is expected from sub-Poissonian photon-pair sources such as semiconductor quantum dots (QD), with the prospect for deterministic operation9. Such sources can also benefit highest level applications for scalable network architectures such as quantum relays10 and repeaters11 based on teleportation and entanglement swapping protocols, posing further constraints on photon indistinguishability.

Quantum light sources based on III–V compound semiconductor QDs embedded in a positive-intrinsic-negative (p-i-n) diode9 are considered a promising approach in terms of scalable manufacturability as they are sharing the same material platform as standard laser diodes. These devices generate single entangled pairs of photons on the so-called biexciton cascade after electrical injection of carriers, which is why they are also referred to as entangled light emitting diodes (ELED)12. Another approach for carrier injection is via optical excitation where one mainly distinguishes between non-resonant excitation above the bandgap of the surrounding semiconductor matrix and resonant excitation of the biexciton state13. The latter has recently become very popular, achieving good photon indistinguishabilities with high purity and most importantly, the potential for fully deterministic operation14. Recent progress in the field culminated in first demonstrations of entanglement swapping15,16 from this kind of source, an important step towards scalable quantum networks. However, these experiments require a sophisticated pulsed laser system combined with careful suppression of exciting laser light, making them at the moment not the first choice when it comes to field deployment of QD photon-pair sources in non-laboratory environments. In contrast, non-resonant excitation schemes are much simpler to implement and integrate, making them a welcome testbed for first steps into network integration of quantum light emitters. Of course, this comes at the expense of lower efficiency17 and photon purities, but with performances still good enough for experiments such as quantum teleportation with electrically18 and optically19 injected sources. For quantum communication applications, careful engineering of material compositions and wafer growth conditions have enabled fabrication of QDs emitting in the standard telecom wavelength bands20,21,22,23, recently enabling for the first time the transmission of photons from a single quantum emitter over a deployed standard network fibre24.

True network integration of quantum light sources requires co-existence with classical data traffic mostly in the telecom C-band over the same optical fibre. This favours operation of ELEDs in the telecom O-band as the large spectral separation combined with the generally reduced Raman scattering from the lower energy C-band to the higher energy O-band makes wavelength division multiplexing of quantum light at the single-photon level with classical communication signals feasible25,26,27. More sophisticated network architectures requiring higher bandwidth make use of the so-called coarse wavelength division multiplexing (CWDM) grid (ITU G.694.2), comprising a total of 18 channels from 1271 to 1611 nm with 20 nm spacing. For compatibility of ELEDs with this standard, spectral tuning ranges of 20 nm are desirable such that any emitter can be tuned to the closest CWDM channel or even changed to a different channel. Despite there has been good progress in the deterministic growth of QDs emitting at specific non-telecom wavelengths28,29, post-growth tuning mechanisms remain most important for enhancing the yield in manufacturability of devices.

The application of magnetic fields30 or strain31,32,33,34,35 has been demonstrated for tuning QD emission. While achieving only a few nanometres of tuneability, these approaches require large magnetic fields or very high voltages which is not practical for network deployment. In contrast, wavelength tuning via static electric fields36,37,38 is suitable for long-term safe operation, achieving good tuning ranges with only low applied voltages. In this work, we demonstrate a tuneable telecom ELED with low-voltage operation, large tuning range and strong single-photon character, which is compatible with networks containing classical communication traffic. This enables us to operate an ELED outside of a research laboratory and show sub-Poissonian entangled photon transmission with high fidelity multiplexed with classical data signals over the metropolitan fibre network in Cambridge.

## Results

### Tuneable ELED

A schematic of the tuneable device design is shown in Fig. 1a. Entangled photon pairs are emitted from the QDs located at the central circular region. These are optically excited by light emitted from the so-called ‘pumping diode’ surrounding it, when applying a forward bias to the top contact on the left. A reverse bias applied to the top contact from the inner diode is used to apply static fields for tuning the QD emission, making this the so-called ‘tuning diode’. The ring-circle structure enables homogeneous illumination of the tuning diode with pumping light from all sides. An optical microscope image of fabricated devices is shown in Fig. 1b. The central mesa for the design used in this work has a radius of 35 μm, typically containing around 120 QDs as can be seen in the camera image in Fig. 1c, showing luminescence of the QDs during operation of the device. Owing to the random nature of the type of self-assembled QDs used here, photons emitted at telecom wavelength generally exhibit a broad variation of parameters, requiring the selection of an emitter with suitable properties. For the device displayed, there is one QD meeting the requirements for brightness and entanglement fidelity. This is consistent with a typical yield of around 0.9% based on the examination of around 350 QDs.

The key features of the wafer structure are illustrated in Fig. 1d. The InAs/GaAs quantum dots used in this work were grown using the Stranski–Krastanov (S–K) growth mode in molecular beam epitaxy. Use of the so-called bi-modal growth mode21 results in QD emission in the telecom O-band. The QD layer is deposited on GaAs and capped with a thick layer of InGaAs (5 nm) which serves as optical pump and absorption layer, in contrast to structures operating at non-telecom wavelengths39,40. The optically active part of the wafer is finished with another layer of GaAs, forming a quantum well (QW). Additional barriers made from AlGaAs grown below and above these central layers inhibit escape of charge carriers when applying electric fields along the vertical direction. The structure further contains stacked distributed Bragg reflector (DBR) mirrors made from GaAs/AlGaAs at the bottom and top, for enhancing the emission in the telecom O-band with a weak vertical half-wavelength (λ/2) cavity. The top three repeats of the bottom DBR are n-doped and the whole of the top DBR is p-doped (both with 2 × 1018 cm−3), with doping extending a few nanometres into the cavity layer from both sides, forming a p-i-n diode.

The combined optical pump and absorption layer is used for increasing the on-chip optical excitation efficiency. In the pumping diode, charge carriers are injected into the InGaAs QW when applying a forward bias across the p-i-n junction. Carrier recombination in this 2D layer gives rise to broadband optical emission around 980 nm as can be seen in the blue spectral curve displayed in Fig. 1e, measured in a confocal microscope setup from the top, on a position with no QD on the pumping diode. The spectral signal looks rather weak as this light is non-resonant with the embedded vertical cavity which favours in-plane rather than out-of-plane emission. Pumping light is absorbed by resonant optical excitation of confined states in the QW of the tuning diode. Here, carriers can recombine in QDs resulting in emission of single or entangled photons in the telecom O-band around 1310 nm along the vertical direction as these are resonant with the embedded cavity. A corresponding spectrum is displayed as red curve in Fig. 1e. The geometric and extreme spectral separation of pumping light and QD emission of more than 300 nm enables an excellent suppression of background using standard optical filtering methods. This results in high-purity single-photon emission with values for zero time delay of the second-order correlation function (g(2)(0)) well below 10%, which is comparable with other QD photon sources which are non-resonantly optically excited with an external laser22,35,36 (a measurement of second-order autocorrelation functions is shown in the Supplementary Fig. 1). In comparison with devices where the on-chip optical excitation is only based on conventional wetting layer emission and absorption (thickness < 1 nm), we observe a significantly enhanced efficiency of the optical pumping process (see Supplementary Note 1). This is most likely caused by increased absorption efficiency of pumping light due to the much higher thickness (5 nm) of the pump and absorption layer.

The AlGaAs barriers surrounding the QD layer as shown in Fig. 1d inhibit escape of charge carriers, enabling the application of large electrical fields across the p-i-n junction. Growth of the QDs in the InGaAs QW further loosens the confinement of electron–hole pairs, resulting in a higher permanent dipole moment and polarizability of excitons, enhancing the tuneability of QD emission via the quantum confined Stark effect36,41,42. In the following, we selected a single QD with good entanglement fidelity and intensity of emitted biexciton (XX) and exciton (X) photons. While photon indistinguishability from telecom S–K QDs with densities as used here is not sufficient for scalable quantum communication schemes, with radiative lifetimes around 1 ns and coherence times around 50 ps, it is not relevant for the present study, focussing on field deployment of a sub-Poissonian photon-pair source and network distribution of entanglement. Figure 2a shows wavelength tuneability of the emission spectrum when driving the pumping diode with a forward bias and changing the reverse bias on the tuning diode from −3.8 to 0 V. The neutral X and XX states can both be shifted over the center of the telecom O-band (1310 nm) with a maximum range of more than 25 nm for the applied voltages. This significantly exceeds the tuneability performance of short-wavelength devices36,39 and meets the requirements for compatibility with the ITU CWDM grid. It has to be emphasised that the electrical driving conditions for excitation are comparable with directly electrically injected ELEDs12 indicating the optimised operation of the tuneable devices while measured photon intensities are comparable with what is typically achieved with non-resonant optical excitation by an external laser with this type of telecom QDs.

We also observe the characteristic tuning of the fine structure splitting (FSS) of neutral X and XX states as a function of bias voltage, as shown in Fig. 2b. The splitting parameter was extracted from measurements of the time-resolved quantum beat of X–XX correlations when filtering both photons with a linear polarizer (for details see Supplementary Methods and Supplementary Fig. 2). For entangled photon-pair emission it is important to achieve FSS values of 10 μeV or less such that indistinguishability of the resulting two spectral components of a photon is guaranteed with typical detection setups featuring 100 ps timing resolution or better (see “Methods” for details). The splitting of the selected QD stays significantly below that threshold over the entire tuning range, meaning that the device maintains the condition for entangled photon-pair emission. In the remainder of the paper, we operate the device with a voltage of −2.6 V applied to the tuning diode, setting the XX emission to 1310.00 nm (the exact center of the O-band) and X emission to 1321.45 nm and resulting in an FSS of 5.6 μeV.

### Field deployment

Simple electrical operation and compatibility with wavelength division multiplexing standards allow us for the first time to demonstrate the deployment and telecommunication network integration of the device. We operate a fully remotely controlled system that is installed in an office at a location in West Cambridge (CAM), which is connected via the Cambridge Network with two optical fibres of 15 km length each to the Toshiba Cambridge Research Laboratory (CRL) in the Science Park. Figure 3 shows the overall experimental setup and a map of Cambridge. The system consists of multiple sub-systems: an entangled photon transmission module containing the tuneable ELED, a spectral filter with polarisation reference for detector calibration, a polarisation stabilisation module and a quantum/classical data multiplexing module.

The ELED device is mounted in a closed-cycle cryostat, cooling it to 6 K. The emitted light from the quantum dot is coupled into single mode optical fibre and sent to a spectral filter module mounted in an instrument rack. Here, a diffraction grating is used to separate the entangled X and XX photons into two different output modes. Right after the spectral filter, the X photons pass through a polarisation discrimination setup that consists of a half waveplate (HWP), a quarter waveplate (QWP) and a linear polariser (LP), projecting their quantum state into different polarisation bases at the remote location before transmission to CRL for detection. The entangled XX partner photons are sent to CRL over a separate network fibre which is stabilized for changes of birefringence (see “Methods”) and additionally hosts a classical data connection for remote control of the deployed system. At CRL, their polarisation state is projected using an electronic polarisation controller (EPC) and a fibre polarizing beam splitter (PBS). Single-photon arrival times for X and XX photons are recorded using superconducting nanowire single-photon detectors (Single Quantum) and photon-pair correlations in different detection bases are measured for evaluation of the entanglement fidelity across the 15-km fibre link (see “Methods”).

A precise alignment of the detection basis to the eigenbasis of the QD emission is essential in these experiments. In this work, we make use of an innovative method for the precise measurement of said eigenbasis, that is based on analysing time-resolved photon-pair correlations in a set of randomly oriented detection bases (see “Methods”). For detector calibration, polarisation reference light is injected into X and XX output modes from the spectral filter module, more details are provided in the Supplementary Methods and Supplementary Figs. 2 and 3.

For demonstrating true network integration, the entangled quantum bits are multiplexed with classical communication traffic over the same optical fibre using CWDM with quantum light at 1310 nm and a bidirectional 1Gbit data connection at 1550 nm (see “Methods”). Previous field trials have illustrated the feasibility of multiplexing QKD qubits based on attenuated laser pulses with classical data traffic over installed network links43,44,45. However, the co-existence of entangled qubits from a sub-Poissonian photon source and classical data traffic over a real-world network has not been demonstrated yet. The classical communication channel is used to remotely control the ELED system, the polarisation stabilisation system and data acquisition during the experiments. Figure 4 shows a comparison of the single-photon purity of quantum light from the ELED before and after transmission over the lit field fibre, by measuring the second-order autocorrelation function. After transmission, the light is still strongly anti-bunched with a g(2)(0) value of 0.26 being significantly below the classical limit of 0.50. A more detailed analysis reveals that the increase from the laboratory-measured value of 0.08 is primarily caused by a reduction of the signal-to-background ratio due to photon loss in the network fibre and all system components, and not the presence of the classical data link (see Supplementary Note 2).

## Discussion

Figure 5a–c shows the results for co- and cross-polarised coincidences in the horizontal/vertical (HV), diagonal/anti-diagonal (DA) and right-/left-circular (RL) basis when projecting the quantum states of X photons at CAM and XX photons at CRL. The detection basis was changed every 10 min. In the HV basis, we observe a strong correlation of photon polarisations with a maximum contrast of 94.5%. The superposition bases DA and RL show high contrast quantum correlations as well, with the characteristic beat owing to the FSS of the QD as reported multiple times before21,46,47. The fidelity to the maximally entangled Bell $$\phi ^ +$$ state peaks at 94.4 ± 0.3% (see Fig. 5d) for a post-selection window size of 48 ps. This compares well with a maximum fidelity of 97.3% measured in a laboratory environment before deployment. The small drop in fidelity is mainly due to a decrease in signal-to-background when sending the photons over installed fibre. To demonstrate the good stability of the deployed ELED device, the experiment was continuously running over a period of 40 h. Figure 5e shows the evolution of the entanglement fidelity based on 120 min of data per displayed point. The bottom of the graph shows the corresponding quantum-bit error rate which is stable around 3.8%. These results are directly obtained from raw-data and no background subtraction of any kind has been done during processing. The classical communication link was running at all times. For a post-selection window of 200 ps, which is well above gate widths of commonly used communication-grade detector systems48, the maximum fidelity after transmission over the network fibre is 91.9 ± 0.2%, still more than 190 standard deviations above the classical limit.

To conclude, we have developed the first electrically operated tuneable ELED emitting in the main telecommunication wavelength band suitable for multiplexing with classical communication signals. Combination of on-chip optical excitation with dedicated layer structure and device design enable low-voltage operation (<5 V) and large wavelength tuneability (>25 nm). A high single-photon purity compatible with typical non-resonantly excited QD light sources was observed and entanglement with a fidelity of more than 94% was achieved after integration and deployment in a real-world fibre network. Continuous operation of the source showed stable entanglement over 15 km of installed fibre shared with classical data traffic used for remote control of all system components.

Compatibility of the device with standard low-voltage power supplies and compliance with laser-safety standards due to exclusively electrical operation pave the way for future integration of devices based on this approach in various end user applications. The large wavelength tuneability furthermore opens up the route for flexible wavelength division multiplexing of multiple tuneable ELED sources over the same optical network links and most importantly, multiplexing with classical data channels. This is of fundamental importance for the low-cost integration of quantum networks in classical network infrastructures.

While the reported results provide a proof-of-principle demonstration of practicality of the technology for network deployment, further work is required to make devices based on on-chip optical excitation that are suitable for high-level quantum-network applications. These are expected to require pulsed operation at GHz clock rates, high source efficiencies and good photon indistinguishability. Pulsed high-frequency operation is achievable via a reduction of the device capacitance by simply shrinking the dimensions of the pumping diode. Higher photon collection efficiencies could be achieved by implementing nano-photonic structures such as in-plane circular Bragg reflectors49,50,51, broadband optical antennas52 or microlenses53 while charge control of QDs54 via optimisation of the tuning diode might open up the route for higher photon generation efficiencies as required for deterministic operation. Furthermore, QDs grown by so-called droplet epitaxy are a highly promising alternative to S–K grown QDs as they show extremely good coherence properties even for non-resonant optical excitation19.

## Methods

### Calibration of detector polarisation basis

The measurement of the orientation of the linear QD eigenbasis is usually done by maximising the spectral shift of entangled photons when detecting after a rotatable LP. This generally requires placement of the polariser before light is coupled from the QD emitter into the first single-mode fibre55 which is not practical in compact setups. The following method that we employed is compatible with devices where free-space access to emitted photons on top of the emitter chip is not possible. We take X–XX correlation measurements (co- and cross-polarised) in three arbitrary but linearly independent detection bases. As explained in more detail in the Supplementary Methods, the time-dependent evolution of the correlation signal is then used to extract the exact orientation of the eigenbasis of QD emission in the chosen detector basis orientation. Using the standard procedure for Müller matrix evaluation56, this allows us to calculate the transformation between the two reference frames. In a second step, this information is used for calibrating the detection basis by injection of properly oriented polarisation reference states for a perfect match with the eigenbasis of QD emission.

### Measurement of entanglement

Two-photon correlations between X and XX photons are measured with a single correlator at CRL, providing intrinsic synchronisation. Entanglement measurements are carried out by recording photon-pair correlations $$c_{{\mathrm{PQ}}}$$ for co- and cross-polarized case in the three principal detection bases HV, DA and RL. The fidelity to the maximally entangled Bell $$\phi ^ +$$ state is then calculated as $${\cal{F}}\,=\,\left( {1\,+\,C_{{\mathrm{HV}}}\,+\,C_{{\mathrm{DA}}}\,-\,C_{{\mathrm{RL}}}} \right)/4,$$ where $$C_{{\mathrm{PQ}}}$$ are the so-called correlation coefficients with $$C_{{\mathrm{PQ}}}\,=\,(c_{{\mathrm{PP}}}\,-\,c_{{\mathrm{PQ}}})/(c_{{\mathrm{PP}}}\,+\,c_{{\mathrm{PQ}}})$$.

### FSS threshold for observation of entanglement

The entanglement fidelity follows a sinusoidal oscillation with the period given by $$h/\delta$$ (see Supplementary Methods) with $$\delta$$ being the FSS of the exciton level. Assuming that other dephasing processes are negligible at short time scales, for an ideal emitter the oscillation will cause the fidelity to drop to 50% (the classical limit) after a quarter of the rotation period. Therefore, as a general rule of thumb, for observing entanglement, the detector timing resolution should be equal to or better than this limit. Assuming a typical timing resolution of 100 ps, this results in a maximum affordable splitting parameter of $$\delta \,\approx\,10$$ µeV.

### Polarisation stabilisation system

The use of polarisation qubits over long installed fibre requires the stabilisation of changes in birefringence that naturally occur due to changing environmental conditions. Similar to our previous work24 we make use of a time-multiplexed polarisation stabilisation system operating at the XX photon wavelength of 1310 nm. Every 15 s, two polarisation references that are aligned along orthogonal axes on the Poincaré sphere are injected for 100 ms and measured over the field fibre. Feedback to a polarisation controller is then applied to compensate for detected polarisation rotations. In addition, the stabilisation is forced every 225 s and typically takes around 3 s to reach maximum alignment. The resulting overall duty-cycle for transmission of quantum bits is around 98%.

### Classical-quantum multiplexing

A bidirectional 1 Gbit classical communication link is established with standard 1550 nm transceiver modules. Up- and down-stream traffic is isolated by directional filtering, using circulators. Multiple CWDM modules in the quantum and classical channels and a diffraction grating spectral filter with a full width at half maximum of less than 1 nm installed in the quantum channel at CRL, are used to suppress the background from the classical light to a very low level. More details about this part of the experimental setup can be found in the Supplementary Note 3 and Supplementary Fig. 4.

### Loss information

The main optical path used for distributing the entangled photons goes through the polarisation stabilisation and classical communication multiplexing modules on both ends of the installed fibre. Including the additional filter at the receiver required to block residual light from the classical communication, and the EPC and PBS before the detectors, the total loss amounts to 3.3 dB for the transmission at 1310 nm. The fibre connection from the remote location at CAM to CRL has an overall length of 15 km and is partially privately owned and partially part of the University Network. The connection for transmitting entangled XX photons and establishing classical communication has a total loss of 8.5 dB at 1310 nm and the fibre used for sending the projected X photons has a loss of 12.5 dB. The latter is most likely caused by a sharp bent of the fibre cable in one of the multiple network nodes that the connection is routed through. In this fibre, we additionally experienced leakage of classical transceiver light at 1309 and 1313 nm probably from a neighbouring strand in the same optical cable. This required additional filtering of received X photons at 1321.45 nm at CRL using a tuneable narrowband filter, adding 2.5 dB to the overall loss budget.

### Photon rates

Photon pairs emitted from the ELED are coupled into single-mode optical fibre using a confocal microscope setup with an estimated extraction efficiency into the fibre around 3%. The raw rate of detected XX photons and X photons before transmission over the network fibre are 500 and 400 kHz, respectively, when driving the tuneable ELED with a DC current. At the output of the field fibre and the multiple system modules, we detect XX photon rates of 33 kHz. The detected X photon rate is around 13 kHz after sending the projected qubits to CRL over the high-loss fibre.