Oxygen-incorporated single-photon sources observed at the surface of silicon carbide crystals

The formation of high-brightness single-photon sources (SPSs) that emit single photons at room temperature was recently confirmed in oxygen-annealed SiC semiconductors (surface SPSs.) However, the defect structure of surface SPSs remains unclear, which makes device fabrication and property control difficult. To verify the incorporation of oxygen in surface SPSs, we fabricated SPSs using stable 18O isotopes as oxidants. By comparing this to the case of natural oxygen annealing, we found that the SP emission spectra for the 18O sample tended to have shorter peak wavelengths, slightly narrower peak widths, and higher intensities. Thus, it appeared that, in the case of the 18O sample, the phonon sideband was located closer to the zero-phonon line and that oxygen was incorporated into the defects attributed to the surface SPS.


Introduction
Recently, quantum-based technologies such as quantum computing, quantum cryptographic communication, and quantum sensing have led to innovative advances. Single-photon sources (SPSs) that are indispensable to this field have been developed in various materials. Silicon carbide (SiC) is one of the most promising SPS host materials because high-quality and large-diameter wafers are in mass production and various device processes are matured. In addition, several kinds of stable and high-brightness SPSs have been identified at room temperature [1][2][3][4], which will enable us to construct quantum devices for practical use. Furthermore, MEMS technologies for SiC enable us to significantly enhance the SP emission rate by introducing an optical resonator around the SPS [5,6].
It has been reported that high-brightness SPSs (hereafter 'surface SPSs') are formed in the vicinity of the SiC-SiO 2 interface upon oxygen annealing of SiC [7]. In addition to room-temperature photo-excited SP emission, electrical-current-excited SP emission at room temperature was also demonstrated from surface SPSs embedded in a pn junction diode [4,8], which is the first achievement after the diamond NV center [9]. However, there are some issues to be resolved, such as the broad emission spectrum and dispersive emission wavelengths of 600-800 nm from the surface SPSs. With the defect structure of the surface SPSs still unclear, it is difficult for us to address these issues. Furthermore, even the presence of electron spin in surface SPSs has not yet been demonstrated. Thus, it is important to elucidate the defect structure of surface SPSs. On the other hand, given that the polarization of the SP emission agrees with the crystallographic symmetry of the SiC substrate [7], surface SPSs are located near the SiC-SiO 2 interface but completely inside the SiC crystal. Therefore, since the defect structure is likely complex (i.e., a point defect pair), the broad spectrum is presumably due to the wide phonon sideband (PSB) emission, as in the case of diamond NV centers. In addition, Lohrmann et al suggested that the single defects attributed to surface SPSs were carbon-or oxygen-related defects because they exhibited Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence.
Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. similar behavior, even though they were formed in different polytypes having different bandgaps [7]. We speculate that the inclusion of oxygen is more likely for a surface SPS because carbon-related defects are usually not single-photon sources (e.g., C vacancies [10], D1 centers [11].) The use of an oxygen marker such as a stable oxygen isotope ( 18 O) may help to settle this issue. On the other hand, Lohrmann et al also suggested that the variation in emission wavelength is due to the variation in the distance between surface SPSs and stacking faults, because the presence of stacking faults produces a linear change in the defect levels of surface SPSs, as shown by ab initio studies [4]. Since, as proposed by Matsushita and Oshiyama [12], the SiC-SiO 2 interface may include many stacking faults, it is thought that stacking faults can account for this wavelength variation. A further discussion on the wavelength variation in surface SPSs can be found elsewhere [13].
It is well known that the optical transitions for color centers such as SPSs obey the Frank-Condon principle [14,15]. Figure 1 shows the coordinate diagram for a color center; E 0 and E 1 denote the energy diagrams for the ground state and excited state, respectively, and n¢ and n are the phonon quantum numbers for these states. In general, when an electron is excited from E 0 to E 1 , nuclear coordinate of the E 1 state shifts owing to the Coulomb force of the excited electron (so-called 'Frank-Condon shift'). Since such a nuclear displacement occurs much more slowly than electron transitions, photon absorption (emission) entails phonon absorption (generation) before the recovery of nuclear displacement. Thus, if we employ stable 18 O isotopes, which are heavier than natural oxygen ( 16 O), the Frank-Condon shift should decrease, as should the phonon energies, resulting in a shift of PSB toward the zero-phonon line (ZPL) or in a disappearance of PSB. Moreover, the transition energy of the ZPL for 18 O might be higher than that for 16 O because the degree of change in structural relaxation after electronic excitation is presumably smaller owing to the heavier mass of 18 O and the E 1 of the 18 O is higher than the E 1 of the 16 O. 13 Thus, by replacing 16 O with 18 O, we may be able to determine whether the defects attributed to the surface SPS contain oxygen.
In this study, we fabricated three types of samples: Ar-, 16 O 2 -and 18 O 2 -annealed. By comparing the radiation properties of these samples, we attempted to determine whether the generation of the surface SPS requires oxidation and the single defects attributed to the surface SPS contain oxygen.

Materials and methods
Epitaxial n-type 4H-SiC wafers with a 4°off-oriented (0001) Si-face and a net donor concentration N d -N a of 1.6×10 16 cm -3 were used in this study. The samples were oxidized in an infrared furnace at 800°C in a dry natural or stable-isotope oxygen ( 16 O 2 / 18 O 2 ) atmosphere at 5 Pa for 30 min. Another specimen was annealed in an Ar atmosphere at 800°C and 100 kPa for 30 min. Photoluminescence (PL) from the samples was observed using a confocal laser scanning fluorescence microscope (CFM) (WITec) with an NA0.9 100×air objective lens (Nikon), single-photon counting modules (Laser Components), and a spectrometer with a CCD detector (Princeton Instruments). The measurement durations of the photon counter and spectrometer were 0.01 s and 0.1 s, respectively. A DPSS laser with a 532-nm wavelength and 1-mW net output was used on the sample as an excitation source. During PL intensity mapping and photon correlation measurements, a 600-nm long-pass filter was placed in front of the photon counter. Low-temperature PL measurements were performed by cooling the sample mounted in a vacuum chamber with a liquid N 2 flow. Photon correlation measurements were performed with a standard Hanbury-Brown-Twiss interferometer.

Results and discussion
First, we carried out PL intensity mapping for the Ar-annealed sample. There were numerous radiation points on the in-plane PL map of the sample surface (not shown here). Next, we randomly selected 68 radiation points and performed photon correlation measurements on each. We found that none of them exhibited antibunching characteristics. Thus, we concluded that a surface SPS cannot be generated only by heating a SiC substrate. It is known that Ar annealing of SiC substrates often produces carbon byproducts, such as graphitic C [16] and C clusters [17]. Therefore, carbon-related defects can be ruled out as a candidate for surface SPSs. Figures 2(a) and (b) show cross-sectional and in-plane PL maps of the 18 O sample, respectively, and 2(c) shows an in-plane PL map of the 16 O sample. As shown in the figure, several dozen radiation points were observed on each sample surface, i.e., the SiC-oxide interfaces. Figure 3 shows representative PL spectra from the radiation points marked in figures 2(b) and (c). The peak around 590 nm appearing in all the spectra is the second-order Raman shift from the 4H-SiC substrate [18]. It was found that there were two kinds of spectra for both samples: one with a sharp peak and the other with a broad peak, similar to a previous study [19]. Figure 4 shows the low-temperature PL spectra from a typical radiation point at specimen temperatures of 80 K and RT. The sample used here was fabricated by oxidization in a dry natural-oxygen ambient at 800°C. The twin peaks in the 580-590 nm range and the intense peaks at 560 nm, observed in all spectra, were due to the above mentioned second-order Raman shift and the LO phonon line, respectively. The low-temperature PL spectrum at the radiation point exhibits a very sharp peak at 580 nm, while a broad peak is observed around 610 nm at RT. The spectra in the inset of figure 4, derived by subtracting the background spectrum, also indicate that the broad peak at RT is red-shifted from the sharp peak at 80 K. Thus, the broad peak observed at RT is entirely composed of PSB, which is observed as the ZPL solely at 80 K. This interpretation corroborates a previous study [20] but contradicts another study [21]. This will be discussed further in a future report.
To determine whether the radiation points corresponded to an SPS, photon correlation measurements were carried out for the 16  where g f 2 is the degree of purity of the SP emission. The condition   g 0.5 1 f 2 corresponds to SPS, a is the extent of transition to the metastable state, and t i (i=1 and 2 for anti-bunching and bunching, respectively) is the radiation lifetime. The solid curve in figure 5 represents a fit to the experimental data obtained by using equation (1). From this fit, g , f 2 a, and t 1 were found to be 0.685, 1.40, and 5.67 ns, respectively. Thus, this radiation point is clearly an SPS. In addition, the a value suggests that a three-level transition is predominant. It should be noted that the radiation rate t / 1 , 1 which corresponds to the radiation intensity, is roughly twice that of a diamond NV center [9]. However, the radiation intensity is generally lower than those of the 4H-SiC surface SPSs in previous reports [4,7,20], which could be attributable to the low-pressure oxidation occurring in the  . Low-temperature PL spectra for a dry-oxidized sample. The solid and broken lines denote the PL spectra at 80 K and RT, respectively. The inset shows the PL spectra at the radiation points minus the background spectra. present case. We also measured the t ( ) g 2 for the 16 O sample, as well as for other radiation points in the 18 O sample. All the results obtained were similar to those shown in figure 5.
We obtained such PL spectra at room temperature from about 60 radiation points for each 18 O sample and 16 O sample and made histograms of peak wavelengths, peak widths, and radiation intensities to compare these samples ( figure 6). The sharp and broad peaks for each sample are distinguished by color. As discussed above, the sharp peak and the broad peak can respectively be regarded as the ZPL alone and the superposition of ZPL and PSB. The radiation intensities were derived from the peak height normalized by the peak height for the secondorder Raman shift (see inset in figure 6(c)). Figure 6(a) reveals that the 18 O sample tended to have a larger number of sharp peaks, widely distributed between 600 and 680 nm, compared to the 16 O sample, which had few sharp peaks, mostly distributed around 640 nm. However, the sharp peaks of the 18 O sample were concentrated around 615 nm, which is lower than the 640 nm for the 16 O sample. The photon energy difference between 615 nm (2.016 eV) and 640 nm (1.934 eV) roughly corresponds to that of the square root of their masses. Therefore, as mentioned above, the E 1 level of the 18 O sample was presumably elevated owing to its heavier mass. In the case of the broad beaks, the peak wavelengths did not differ significantly between the 18 O and 16 O samples. Figure 6(b) reveals that the broad peaks were slightly narrower in the case of the 18    also seen from figure 6(c) that the 18 O sample has stronger radiation intensities, especially in the case of the broad peak. This means that, in the case of the 18 O sample, the PSB disappears or is closer to the ZPL.
To compare the radiation intensities, the polarization of the SP emitter should be considered. Since a surface SPS is probably a complex structure similar to a diamond NV center, it should have linear polarization. However, since the defect structures of the 18 O and 16 O samples are presumably the same, we ignored the influence of polarization on the statistics of the radiation intensity data.
Our experimental results strongly support the presence of oxygen in the single defects attributed to surface SPSs. Next, we consider the candidates for the surface SPS. As mentioned in the Introduction section, the structure is thought to be a complex point defect. Thus, as shown in figure 7, several candidates can be inferred for the surface SPS structure. In the upper two structures, an O atom is inserted into the Si (C) site and the neighboring C (Si) is missing, resulting in an O-vacancy pair. Another possibility is that an O atom becomes interstitial, i.e., bonded to Si (C), as shown in the lower two structures in figure 7. In this case, the bond next to the O interstitial may be broken, and the emission wavelength will depend on which bond is broken. By conducting an ab initio study of these defects, the defect structure of the surface SPS can be determined from the calculated defect levels. We are currently also performing first-principles calculations and will be presenting the results elsewhere [13].

Conclusions
We investigated the radiation properties of the radiation points formed by Ar annealing, 16 O oxidation, and 18 O oxidation, to determine whether formation of surface SPSs requires oxidation and whether the defects attributed to surface SPSs contain oxygen. For the Ar-annealed sample, there were no SPSs among the 68 radiation points. This indicates that thermal treatment is insufficient and that oxidation is needed to form surface SPSs. Thus, carbon-related defects could be ruled out as candidates. The radiation spectra showed that the 18 O-oxidized sample tended to have a greater number of sharp peaks, smaller sharp peak wavelengths, slightly narrower broad peaks, and stronger intensities as compared to the 16 O-oxidized sample. These results revealed that in the case of the 18 O sample, the zero-phonon lines on the shorter wavelength side were easier to generate and that the location of the phonon-side band was closer to the zero-phonon line because of the larger mass of 18 O, evidencing the incorporation of oxygen in the defects attributed to surface SPSs. Finally, we proposed some candidate defect structures for surface SPSs to determine whether there is an electron spin and to establish a control method for SP emission from surface SPSs.