Introduction

Silicon quantum technologies based on integrated photonics1 offer a promising path to scaling by leveraging advanced semiconductor manufacturing and integration capabilities2,3. Current approaches to fault-tolerant photonic quantum computation use weak material nonlinearities and measurements to probabilistically generate photon pairs and implement two-qubit gates1,4. The lack of deterministic quantum light sources5, photon-photon gates6, and quantum memories7 in silicon photonics poses a major challenge to scalability and requires very large resource overheads8. Coherently controlled quantum emitters in a reconfigurable photonic circuit can enable hardware-efficient universal quantum computation9,10 and time-multiplexed quantum networking11. Silicon photonics provides a mature platform for low-loss reconfigurable integrated photonics12. However, an atomic source of indistinguishable photons in silicon has been missing13. We address this challenge by demonstrating telecom-band indistinguishable photon generation from an artificial atom in silicon photonics.

Artificial atoms in solids enable single-photon level optical nonlinearities for realizing deterministic single-photon sources, as well as two-photon gates and long-range spin-spin entanglement7,14,15,16. While defect-based photoluminescence (PL) in silicon has been studied for decades17,18, bright telecom-band single-photon emission from a broad diversity of artificial atoms in silicon was only recently shown19,20,21,22,23. In order for silicon artificial atoms to function as quantum-coherent light sources, their emission has to satisfy spatiotemporal indistinguishability24. In this work, we integrate a silicon color center into a photonic waveguide, show pulsed single-photon generation, and demonstrate that successive photons emitted are indistinguishable.

Results

An artificial atom in a silicon waveguide

Our device consists of a G center created in a silicon photonic waveguide (Fig. 1). The G center is a complex defect in silicon that consists of two substitutional carbon (CS) atoms and an interstitial silicon (Si). It emits in the telecommunication O-band with a zero-phonon line (ZPL) at 1278 nm17,18. We create G centers inside the 220 nm device layer of a silicon-on-insulator (SOI) wafer by 12C ion implantation at 36 keV and a fluence of 1012 cm−2, followed by rapid thermal annealing at 1000°C for 20 seconds. Following device fabrication, these parameters result in the creation of approximately one G center in a 100 μm-long waveguide, allowing investigation of a single spatially isolated defect. The creation yield of G centers can be dramatically increased via proton irradiation25 or by alternative fabrication steps as discussed in Supplementary Information (SI) Note 2.

Fig. 1: An atomic quantum light source in silicon photonics.
figure 1

a Device and measurement schematic. (i) A G center is created inside a silicon photonic waveguide via ion implantation. (ii) Silicon waveguide cross section. G center transition dipole \(\overrightarrow{d}\) is along 〈110〉. Laser excitation (orange) of the G center results in single-photon emission into the waveguide (red) which is collected using a lensed fiber. b Scanning electron micrograph of the photonic waveguide near the broadband Bragg reflector. c PL image of the waveguide shows bright emission from an isolated G center.

Full size image

Upon above-bandgap excitation, the G center emits photons via the radiative recombination of electron-hole pairs at localized defect levels (Fig. 2a). The dipole emission is guided by a single-mode silicon waveguide coupled to a single-mode lensed fiber with 50% efficiency using an adiabatic mode converter (Fig. 1a, Fig. S3). The waveguide is terminated with a Bragg reflector for single-sided measurements and coupling efficiency calibration. The collected photons are detected using a spectrometer (Fig. 2b) or superconducting nanowire single-photon detectors (SNSPD) with a quantum efficiency of 60%. The sample is housed in a cryostat and measured at 3.4 K. Materials, fabrication, photonic design, setup, and first principles calculation details are provided in the Methods and SI.

Fig. 2: Photodynamics of a G center in a waveguide.
figure 2

a Above-bandgap excitation (Γexc) creates excess carriers in the conduction and valence bands (CB, VB) which recombine at the localized defect levels (ΓG) and produce single-photon emission. b PL spectrum of the ZPL of the G center. c Excitation power dependence of the ZPL PL shows saturated emission. d PL lifetime measurement with pulsed excitation.

Full size image

Optical properties of a G center in a waveguide

To locate a single G center, we spatially scan a free-space excitation beam at 635 nm and detect photons emitted into the waveguide through the lensed fiber. Figure 1c shows the resulting PL image of the waveguide where we observe an isolated emitter with a measured photon rate of 18 kHz using a bandpass filter (1280 ± 6 nm) centered at the G center ZPL (emission spectrum shown in Fig. 2b). In the following experiments, we probe the linear and nonlinear optical responses of this G center using time- and spectrally- resolved single-photon detection. We study the saturation response of the G center by measuring the power dependence of the ZPL emission rate RZPL on the excitation power P. The power dependence is modeled by RZPL = RsatP/(Psat + P) + αP, where the two terms correspond to a two-level atomic response and a weak linear background. The fit yields a saturated count rate of Rsat = 35kcps and a saturation power Psat = 2.4 μW (Fig. 2c). Next, we use a pulsed laser at 705 nm to measure the PL lifetime of the emitter to be \({\Gamma }_{G}^{-1}=4.6(1)\,{{{{\rm{ns}}}}}\) (Fig. 2d). For each excitation pulse, we detect a ZPL photon with a probability of 0.4(1) × 10−3 (SI Note 5). We calibrate the losses in our setup and use the ZPL branching ratio of 0.18 to estimate the probability the G center emits into the waveguide β = Γ1D/ΓG = 0.014, where Γ1D is the radiative emission rate into the waveguide. We estimate a radiative lifetime upper bound of 260 ns. Our first principles calculations in SI Note 4 predict a radiative lifetime of 225(75) ns.

Optical coherence

We probe the optical coherence of the G center by measuring the ZPL emission spectrum using a tunable Fabry-Perot (FP) cavity with a linewidth of κFP/2π = 3.4 GHz (Fig. 3a). The resulting spectrum, which is a convolution of the ZPL emission and the FP transmission, shows a total linewidth of 6.2 GHz. After deconvolving the cavity response, we find the G center emission linewidth to be Γ/2π = 2.8 GHz (Fig. 3b). Next, we characterize the photon statistics of the G center emission by measuring the normalized intensity correlations g(2)(τ) under pulsed 705 nm excitation at a repetition period Δτ = 25 ns (Fig. 2c). We observe antibunched intensity correlations at zero delay g(2)(0) = 0.15(2) < 0.5 which confirm single-photon emission. The g(2)(0) value is limited by contributions from exponential tails originating from the ratio of the repetition period and the excited state lifetime (\(\Delta \tau /{\Gamma }_{G}^{-1}\)), imperfect extinction in pulsed laser downsampling, and dark counts (SI Note 6). We benchmark the long-term stability of the G center emission by analyzing the intensity correlations up to seconds of delays under CW excitation. The results in Fig. 3d, e show a flat response, indicating stable single-photon emission without any excess intensity fluctuations for τ > 50 ns. We observe bunching at shorter timescales which has been attributed to the presence of a metastable state19,22.

Fig. 3: Quantum coherence of G center emission.
figure 3

a Emission from the G center is collected through a single mode (SM) fiber and analyzed with a tunable Fabry Perot (FP) cavity. BP: Bandpass filter, BS: beamsplitter. b Measured emission linewidth (red fit): 6.2(1) GHz, FP linewidth (black, using reference laser): 3.4(1) GHz. Calculated G center linewidth after deconvolving the FP response: 2.8(1) GHz. c Normalized intensity correlations at the detectors g(2)(τ) show single-photon emission g(2)(0) = 0.15(2) < 0.5. d, e Long term intensity correlations under continuous wave excitation show stable single-photon emission. c–e Error bars represent Poissonian noise, and the error of reported values is the error of the fits.

Full size image

Time-resolved two-photon quantum interference

Photon indistinguishability requires a high degree of spatio-temporal overlap between single-photon wavepackets24. We use a time-delayed Hong-Ou-Mandel (HOM) interference experiment to test the indistinguishability of successive single-photon pulses from the G center26,27. We interfere successive single photons (red and blue pulses in Fig. 4a) using a fiber-based time-delayed Mach-Zehnder interferometer (MZI) where one path has an additional delay Δτ = 25 ns, matched to the laser repetition period. We adjust the relative polarization between the two MZI paths to control the mode overlap and photonic indistinguishability at the second beam splitter.

Fig. 4: Quantum interference of single photons from a G center in silicon photonics.
figure 4

a Successive photons (Δτ = 25 ns) interfere via a time-delayed Mach-Zehnder interferometer. Indistinguishability of the interfering single photons is adjusted by setting their relative polarizations using a polarization controller (PC). b Two-photon interference measurement. g(2)(τ) for orthogonal (blue) and parallel (red) polarizations. (c) When the two photons are indistinguishable, quantum interference results in the antibunching dip at short time delays. We observe nearly identical correlations outside of the quantum interference window. b, c Error bars represent Poissonian noise, and errors in g(2)(τ) values are extracted from the error of the fits.

Full size image

The results of HOM interference between parallel and orthogonally polarized single-photon pairs from a G center are shown in Fig. 4b, c. When the polarizations of the two interfering photons are parallel (red data, indistinguishable case), we see the characteristic HOM dip28 resulting from two-photon quantum interference at short time delays with \({g}_{\parallel }^{(2)}(0)=0.26(4) \, < \, 0.5\). When we tune the photons to be orthogonally polarized so that they are intentionally distinguishable (blue data), the HOM dip disappears and we obtain \({g}_{\perp }^{(2)}(0)=0.69(5)\). A comparison of the normalized coincidence probability at zero time delay yields an HOM interference visibility of \(\chi (0)=1-{g}_{\parallel }^{(2)}(0)/{g}_{\perp }^{(2)}(0)=0.62(4)\). We identify a 0.37(18) ns window over which χ(τ) > 0.5. The integrated visibility over the entire wavepacket is 0.004, which can be improved to close to 1 with future advances discussed below.

The temporal dynamics of the experimentally measured \({g}_{\parallel }^{(2)}(\tau )\) can be described using the formula \({G}^{(2)}(\tau )=\frac{{e}^{-\tau / {T}_{1}}}{4{T}_{1}}(1-\chi {e}^{-({\Gamma }^{{{{{\rm{HOM}}}}}}\tau )})\) which is derived in SI Note 6. In the model, we use the value of T1 = 4.6(1) ns from the lifetime measurements in Fig. 3. ΓHOM describes an effective linewidth of the emitter which represents the average detuning between successive single photon pulses separated by 25ns. We find \({({\Gamma }^{{{{{\rm{HOM}}}}}})}^{-1}=0.4 \pm 0.1\,{{{{\rm{ns}}}}}\), corresponding to an effective short-timescale linewidth of ΓHOM/2π = 0.4(1) GHz. We note that this effective linewidth is about an order of magnitude smaller than the measured linewidth of the emission spectrum in Fig. 3, and filtering effects from the FP cavity have a negligible impact (2% reduction) on ΓHOM. We attribute the smaller effective linewidth to the time dynamics of spectral diffusion in the system. Fig. 3b was acquired over a ten-minute period, whereas we interfere two subsequently emitted photons separated by 25 ns in our HOM experiment. These results show that the emission spectrum is stable over short timescales, a feature favorable for on-chip multi-photon experiments. The zero-delay visibility is limited by the timing jitter of our detector pair (250 ps), imperfect polarization overlap, finite lifetime-to-repetition period ratio, and dark counts (SI Note 6).

Discussion

Our results show that color centers in silicon can generate indistinguishable photons at the telecom-band in silicon photonics. This experiment is enabled by the large transition dipole moment (2.8 ± 0.5 Debye calculated from first principles in SI Note 4), the optical coherence of the G center, and efficient collection of single photons from the G center using silicon photonics. In the following, we discuss open questions and approaches to use this platform to develop high-purity waveguide-integrated single photon sources in silicon.

Photonic indistinguishability is a prerequisite for achieving strong photon-photon interactions15 and building large photonic quantum states29,30. The primary limitation of our results is the small temporal window over which indistinguishable photons may be post-selected, caused by optical decoherence of the G center. To relax the post-selection requirement, the decoherence time scale must be sufficiently long when compared to the excited state lifetime. This can be achieved by reducing the excited state lifetime and improving the optical coherence of the emitter. The coherent interaction rate between the emitter and photons can be enhanced in high Q/V silicon photonic resonators31,32,33,34 with mode volumes V ~ 0.1λ3 and quality factors Q ~ 106. Resonant excitation, dynamic stabilization35, and embedding emitters in p-i-n junctions can suppress charge noise36, reduce spectral diffusion, and improve optical coherence. Additionally, the discovery of new centrosymmetric artificial atoms21 in silicon will make these systems more robust against spectral diffusion37. With such improvements, we estimate that G centers in high Q/V cavities can reach cavity-QED parameters of {g, κ, γ}/2π ≈ {1.3, 0.25, 0.2} GHz and enter the strong-coupling regime with a cooperativity C = 4g2/κγ ~ 100. For indistinguishable photon generation in the Purcell regime, the cavity external decay rate can be intentionally increased to achieve high efficiency and high indistinguishability5.

Emerging silicon artificial atoms with electron and nuclear spins23 can also be introduced to this device platform to implement quantum processor and repeater building blocks, as was shown in other material platforms7,16,38,39. The realization of such high-fidelity spin-photon gates with silicon artificial atoms will open up the possibility of scaling spin2 and photonic1 quantum processors and repeaters using advanced CMOS manufacturing and integration capabilities2,3,13.

Methods

Device fabrication and design

The sample was prepared from a 1 cm  × 1 cm chip diced from a 200 mm, high-resistivity (Float-zone, ≥3000 Ωcm), 220 nm SOI wafer prepared using the SmartCut method. The fabrication process is summarized in Fig. S2. The backside of the chip was partially diced to enable cleaving of the chip at desired locations to expose the waveguide facet for fiber coupling. Carbon implantation and high temperature annealing were carried out before any lithography steps. An etch mask for photonic structures was defined through electron beam lithography on HSQ resist. The mask was developed using a NaOH/NaCl developer chemistry. The 220 nm device layer was etched in a Cl2/HBr chemistry. Finally, approximately 3 μm of silicon oxide cladding was deposited on the surface of the chip through PECVD. The resulting waveguides have the cross section shown in Fig. S3. We used Finite Element Method (FEM) and Finite-Difference Time-Domain (FDTD) solvers to design a Bragg reflector for the fundamental TE mode based on a photonic crystal with lattice constant of 370 nm and ellipsoid holes with the principle axes of {170, 200} nm. This design yields near-unity reflection for TE polarization over a 150 nm band centered at 1330 nm. The center of the band was biased towards wavelengths longer than the G center ZPL to include the phonon sideband. The waveguide width of 300 nm was optimized to achieve the maximum electric field intensity at the center of the waveguide, and therefore the coupling strength to the emitter. The waveguide width was tapered down to 130 nm over 50 μm for maximum coupling to a 2.5 μm lensed fiber (Fig. S3). The waveguide is fabricated along the 〈100〉 crystal axis.