Engineering near infrared single photon emitters in ultrapure silicon carbide

Quantum emitters hosted in crystalline lattices are highly attractive candidates for quantum information processing, secure networks and nanosensing. For many of these applications it is necessary to have control over single emitters with long spin coherence times. Such single quantum systems have been realized using quantum dots, colour centres in diamond, dopants in nanostructures and molecules . More recently, ensemble emitters with spin dephasing times on the order of microseconds and room-temperature optically detectable magnetic resonance have been identified in silicon carbide (SiC), a compound being highly compatible to up-to-date semiconductor device technology. So far however, the engineering of such spin centres in SiC on single-emitter level has remained elusive. Here, we demonstrate the control of spin centre density in ultrapure SiC over 8 orders of magnitude, from below $10^{9}$ to above $10^{16} \,$cm$^{-3}$ using neutron irradiation. For a low irradiation dose, a fully photostable, room-temperature, near infrared (NIR) single photon emitter can clearly be isolated, demonstrating no bleaching even after $10^{14}$ excitation cycles. Based on their spectroscopic fingerprints, these centres are identified as silicon vacancies, which can potentially be used as qubits, spin sensors and maser amplifiers.

Quantum emitters hosted in crystalline lattices are highly attractive candidates for quantum information processing 1 , secure networks 2,3 and nanosensing 4,5 . For many of these applications it is necessary to have control over single emitters with long spin coherence times. Such single quantum systems have been realized using quantum dots 6 , colour centres in diamond 7 , dopants in nanostructures 8 and molecules 9 . More recently, ensemble emitters with spin dephasing times on the order of microseconds 10,11 and room-temperature optically detectable magnetic resonance 12 have been identified in silicon carbide (SiC), a compound being highly compatible to up-to-date semiconductor device technology. So far however, the engineering of such spin centres in SiC on single-emitter level has remained elusive 13 . Here, we demonstrate the control of spin centre density in ultrapure SiC over 8 orders of magnitude, from below 10 9 to above 10 16 cm −3 using neutron irradiation. For a low irradiation dose, a fully photostable, room-temperature, near infrared (NIR) single photon emitter can clearly be isolated, demonstrating no bleaching even after 10 14 excitation cycles. Based on their spectroscopic fingerprints, these centres are identified as silicon vacancies, which can potentially be used as qubits 14 , spin sensors 15 and maser amplifiers 12 .
Silicon vacancy (V Si )-related defects in SiC can significantly exceed the performance of on-chip photonic networks and long-distance quantum communication systems, compared to many other solid-state single photon emitters. In particular, the zero-phonon lines (ZPLs) of V Si -related defects in 4H, 6H and 3C polytypes of SiC present spectrally narrow features at NIR wavelengths λ ZPL = 850 − 1200 nm. Rayleigh scattering losses in photonic structures are inversely proportional to the fourth power of the wavelength, giving almost one order of magnitude lower losses for these defects compared to the nitrogen-vacancy defect in diamond (λ ZPL = 630 nm) 16 or the carbon antisite-vacancy pair in SiC (λ ZPL = 660 nm) 17 . Similarly, scattering losses at interfaces and signal attenuation in optical fibers decrease with wavelength as well. Furthermore, V Si -related defects in SiC can be integrated with existing optoelectronic devices 18 and, in contrast to GaAs-based quantum dots 19 , operate even at room temperature.
The 4H-SiC unit cell with single V Si defect is shown in Fig. 1(a). The dangling bonds of four C atoms with the absent Si atom result in formation of energy levels within the forbidden gap (3.23 eV) of 4H-SiC 20,21 . In case of negatively-charged V Si , five electrons form a spin quadruplet (S = 3/2) in the ground state 12,22 . To excite these defects we use sub-band gap excitation of SiC at a laser wavelength of 785 nm (hν = 1.58 eV), which is close to their optimal excitation wavelength 23 . At room temperature, the V Si defects emit in the NIR spectral range from 800 to 1100 nm. At cryogenic temperatures, two distinct ZPLs at λ ZPL = 862 nm (V1) and λ ZPL = 917 nm (V2), associated with two different crystallographic sites in 4H-SiC, are observed in the photoluminescence (PL) spectrum 24 , which can be used as spectroscopic fingerprints of V Si . To control the V Si density in a high-quality 110-µmthick 4H-SiC epitaxial layer 23 , we used neutron irradiation (0.18 MeV < E n < 2.5 MeV) in a fission reactor. The irradiation dose was varied over more than 8 orders of magnitude, from 10 9 to 5 × 10 17 n/cm 2 (Fig. 1). Some part of the generated V Si defects are negatively charged due to the presence of residual N donors (5.0 × 10 14 cm −3 ). Additionally, the neutron transmutation doping 30 Si(n, γ) 31 Si → 31 P + β − may play a role.
A PL confocal raster scan (10 × 10 µm 2 ) on a sample irradiated with a low dose of n = 1 × 10 11 n/cm 2 is presented in Fig. 1(b). The PL is detected in the spectral range from 875 nm [owing to a longpass (LP) filter] to 1050 nm [limited by the sensitivity of Si avalanche photodiodes (APDs)]. Four, nearly diffraction-limited spots [full width at half maximum (FWHM) of ca. 500 nm] are clearly seen in this scan. With rising irradiation dose to n = 1 × 10 14 n/cm 2 the number of PL spots increases as well [ Fig. 1(c)]. For the highest irradiation dose of n = 5 × 10 17 n/cm 2 single PL spots cannot be resolved any more and the PL spatial distribution becomes highly homogeneous [ Fig. 1(d)]. Remarkably, in the negligibly weak irradiated sample (dose of 1 × 10 9 n/cm 2 ) only one PL spot is found in the 50×50 µm 2 raster scan [ Fig. 1(e)]. The single spots are also restricted along the optical axis as shown in Fig. 2(a). Below we unambiguously prove that these intensity spots are due to the emission from single V Si defects.
In order to find the density of single V Si photon emitters the following procedure is used. Up to the irradiation dose of n = 1 × 10 14 n/cm 2 we directly count the number of PL spots in the detection volume, given by the scanned area and the focus depth, the latter is about 1.2 µm ac- cording to Fig. 2(a). For this irradiation dose we also measure the integrated PL intensity collected from an area of about 100 µm 2 . This PL is used as a reference to calculate the emitter density in the strongly irradiated samples by comparing PL intensities. The results are presented in Fig. 1(f). For the lowest irradiation dose the defect density is N = 3 × 10 8 cm −3 , which can be taken as the upper limit of residual V Si concentration in our 4H-SiC sample. The concentration after the highest irradiation dose constitutes N = 7×10 15 cm −3 . The irradiation dose dependence follows quite well a polynomial scaling N ∝ n 0.8 , as shown by the solid line in Fig. 1(f).
To identify the type of generated defects, we measure the PL spectra for different irradiation doses. The spectrum from a single center is identical to the ensemble emission [ Fig. 2(b)]. Here we use a LP filter 850 nm to suppress the excitation light at 785 nm in the detection path. The LO phonon Raman line from 4H-SiC at 850 nm is independent of the irradiation dose and hence is masked by the stronger PL band for n = 10 15 n/cm 2 . We therefore use an additional LP filter 875 nm when investigating single centers. The PL spectra recorded at low temperature (T = 5 K) are presented in Fig. 2(c). Two characteristic lines at 861.4 nm and 916.3 nm are clearly visible for different irradiation doses. These lines coincide with the V1 and V2 ZPLs 24 , proving that the PL originates from the V Si defects in 4H-SiC.
As expected for single defect centers, the PL intensity I saturates with increasing excitation power density W [ Fig. 2(d)]. After subtracting the linear background contribution and APD dark counts, it follows Here, W 0 = 52 kW/cm 2 is the saturation power density exposed to the sample, corresponding to the laser power P 0 = 0.3 mW at the objective entrance aperture. The saturation PL intensity varies slightly from spot to spot and in Fig. 3(d) yields I max = 8.5×10 3 counts per second (cps). As a next step, we perform the Hanbury-Brown and Twiss (HBT) interferometry experiment, i.e., the time correlation measurement of photon detection by two APDs. This is a frequently used method to verify single photon emission 16,17 . The second-order correlation functions g (2) (τ ), recorded over several hours for different W , are shown in Fig. 3(b). The most important feature is the dip at zero time delay (τ = 0). For the lowest excitation, we obtain g (2) (0) = 0.23 ± 0.07 < 0.5, which denotes clearly the non-classical behavior of a single photon emitter. Additionally to the anti-bunching for |τ | < 15 ns there is also bunching for |τ | > 15 ns. In order to explain such a behaviour at least three levels should be involved [ Fig. 3(a)].
The second-order correlation function can be well described using as shown by the solid lines in Fig. 3(b). The power dependencies of parameters a, τ 2 and τ 1 are presented in Figs. 3(c)-(e), respectively. We use the same three-level model as for the colour centres in diamond 16,25,26 to fit these dependencies. This model describes reasonably well the bunching amplitude a(W ) [the solid line in Fig. 3(c)] and the anti-bunching decay time τ 1 (W ) [the solid line in Fig. 3(e)]. However, the relatively long bunching decay time τ 2 (W ) for W < 100 kW/cm 2 is not well reproduced within this model [the solid line in Fig. 3(d)]. The fit for τ 2 also provides significantly different transition rates k ij as that for a and τ 1 . A possible explanation is that a deshelving process of the metastable state |3 may occur under optical excitation 26 . In order to find the transition rates k ij of the threelevel model in Fig. 3(a), we take the limiting values for W → 0 and W = 800 kW/cm 2 W 0 as an approximation for W → ∞. We obtain τ 1 (0) = 1/(k 21 + k 23 ) = 5.3 ns, τ 2 (∞) = 1/(k 23 + k 31 ) = 14.5 ns, a(∞) = k 23 /k 31 = 6.4, and the corresponding lifetimes are summarised in table I. Remarkably, the value for τ 1 agrees well with the PL decay time of 6.1 ns observed in time-resolved experiments 23 and remarkably longer than that of the band-to-band transition in semiconductor nanostructures 27 . The excitation rate of V Si is proportional to the laser power density k 12 = σW , where the absorption cross section σ can be calculated from the saturation behaviour of Fig. 2(d) as σ = (hν/T W 0 )(k 23 k 31 + k 21 k 31 )/(k 23 + k 31 ). Here, we take for the transmission coefficient at the SiC surface T = 0.81 and the calculated value for σ is also presented in table I.
Photostability is an important characteristic of a single photon emitter. The PL time traces of a single V Si defect are shown in Fig. 4(a). For a sampling bin to δt = 100 ms the count rates remain constant over min-  utes. In order to examine the photostability on a shorter time scale, the sampling bin is reduced to δt = 1 ms. The number of detected photons per sampling bin is 10 in this case, and the time trace demonstrates statistical fluctuations without any indication of blinking [ Fig. 4(b)]. We have investigated a single V Si emitter over more than one week under continuous excitation and did not observe photobleaching. Assuming that the excitation occurs on average every 10 ns, this corresponds to 10 14 excitation cycles.
Finally we found that in the highly irradiated sample the PL intensity increases locally upon laser illumination. To demonstrate this effect, the laser of different intensities was focussed sequently on three different spots and remained there for 120 minutes, respectively. A confocal raster scan, performed at low laser power after such a procedure, demonstrates clearly a PL enhancement for each spot, as shown in Fig. 4(c). For the highest laser power [spot (iii) in Fig. 4(c)] this enhancement is ca. 25%. Re- markably, the generated pattern of Fig. 4(c) preserves at least one day.
Our interpretation is that the focused laser beam locally heats the sample, resulting in atomic displacements and thus in disappearance or/and transformation of some other types of intrinsic defects, generated upon neutron irradiation. These defects may serve as non-radiative recombination channels or as charge traps, switching off the V Si defects in close proximity. In order to corroborate this explanation, we perform complementary annealing experiments by increasing stepwise the temperature and monitoring the PL intensity after each step. The results are summarized in Fig. 4(d), showing the overall PL enhancement by a factor of 5.5. This corresponds to N = 3.9 × 10 16 cm −3 and demonstrates that V Si defects can be created at high density in a controlled manner, as required, for instance, for the implementation of a SiC maser 12 .
In our experiments, we precisely control the concentration of V Si defects in ultrapure 4H-SiC down to single defect level. This approach can be used to deterministically incorporate these atomic-scale defects in electronic 18 and photonic structures 28 as well as in nanocrystals. Together with their extremely narrow optical resonances (on the order of 10 pm at low temperature 14 ) and recently demonstrated optically-detected spin resonances at ambient conditions 15 , our results open exciting opportunities for various quantum applications with spin-photon interface.

Samples
The 4H-SiC sample has been purchased from CREE. A high purity (residual nitrogen doping below 5.0 × 10 14 cm −3 ) layer of 110 µm thickness was epitaxially grown on a 2-inch n-type 4H-SiC wafer. The layer is covered by a 5-µm-thick n-type 4H-SiC layer and a 2µm-thick p-type 4H-SiC layer. The wafer was diced in 4 mm × 2 mm pieces, which were then irradiated in a TRIGA Mark-II nuclear reactor, with neutron energies in the range of 0.18 MeV< E n < 2.5 MeV.
The sample with the highest neutron irradiation dose of 5 × 10 17 n/cm 2 was thermally annealed in several steps from 125 • C to 700 • C for a time of 90 minutes, respectively. The heating was either performed on a heat stage (125 • C-200 • C) or in an oven (300 • C-700 • C).

Experimental setup
Samples were investigated with a home-build confocal microscope and a cw 785 nm laser was used for excitation. A three dimensional piezo unit (nPoint) was used to move the SiC sample in lateral and axial directions. The excitation beam was focused onto the samples by a high aperture (NA = 1.49) oil immersion microscope objective (UAPON 100XOTIRF, Olympus). Collimated optical response of the sample was collected by the same objective and guided through a 30 or 75 µm pinhole (Thorlabs) followed by a 850 and 875 nm LP filter (Edmund Optics). Time correlated single photon counting (TC-SPC) was recorded by a HBT setup consisting of two APDs (Count-100C-FC, Laser-components GmbH) with a quantum efficiency of about 0.3 at the signal wavelength and less than 100 cps in the dark and a 16-channel photon correlator card (DPC-230, Becker&Hickl GmbH) with a time resolution of at least 165 ps.
The low temperature PL spectra were measured at 5 K in a cryostat (MicrostatHe, Oxford Instruments) built into a confocal Raman spectrometer (LabRAM HR, Horiba). The excitation wavelength was 633 nm.