Single photon generation from AlGaN exciton localization centers exhibiting narrow spectral linewidths

III-nitride semiconductors are one of the most important electronic materials for modern society. Among the various advantages of III-nitrides, researchers have been attracted by the materials’ high potential as a practical platform for solid-state single photon sources operating at and above room temperature.^1–4 Recent demonstrations, such as single-photon emission from a site-controlled single GaN quantum dot (QD) at temperatures up to 350 K^5,6 and bright room-temperature single-photon sources in the telecom range consisting of localized defects in GaN,⁷ have further stimulated research and development of III-nitride nanostructure-based quantum light sources.^8–11 

However, there are still several technological challenges to be overcome before practical devices can be realized. Although epitaxially grown QDs are promising candidates due to properties such as the possibility of site-control during growth, current injection, coupling with optical cavities, or emission wavelength tunability,³ the development of high-quality single-photon sources based on III-nitride QDs has been hampered by their relatively broad emission linewidths due to a considerably large degree of spectral diffusion, which is exacerbated by the presence of a large internal electric field in polar III-nitrides. In this regard, we have demonstrated single-photon emission with a high degree of purity [g⁽²⁾ (0) = 0.02] from single GaN interface-fluctuation QDs exhibiting sharp emission lines with narrow linewidths.¹² The reported linewidths (down to 87 µeV) presented the state-of-the-art for III-nitride QDs and are attributed to the lower defect density in the surrounding AlGaN matrix.^12,13 On the other hand, although progress has been made in understanding the physics of spectral diffusion in III-nitride QDs,^14–17 it is obvious that further improvements are necessary to realize practical nonclassical light sources. Here, we present the identification, analysis, and discussion of an alternative nanostructure that is spontaneously formed in AlGaN and has the potential for realizing ordered arrays of single-photon sources with extremely narrow spectral linewidths.

Samples for investigation were grown using metalorganic chemical vapor deposition on sapphire (0001) substrates. Trimethylgallium, trimethylaluminum, and NH₃ were used for the group-III precursors and the nitrogen source. First, a thin (∼40 nm) GaN nucleation layer was deposited at 480 °C after a 4-min thermal cleaning of the substrates at 930 °C. Next, a 1.5 μm-thick GaN buffer layer was grown at 1071 °C. The layer was grown under atmospheric pressure from the initial stage until this buffer layer had formed, and then the pressure was rapidly reduced down to 2.67 × 10⁴ Pa before an Al_xGa_1−xN layer was finally grown on the buffer layer at 1100 °C. Several samples with different AlGaN thicknesses and Al compositions x were grown as described in the following. Photoluminescence (PL) measurements were performed by using a continuous wave diode-pumped solid-state (DPSS) laser (wavelength: 266 nm) as an excitation source. Surface emission images were monitored through an objective lens. A Hanbury Brown and Twiss setup consisting of a beam splitter, two photomultiplier tubes, and a time-correlated single photon counting system was used for photon autocorrelation measurements. Details on the experimental setup can be found in our previous reports.^9,12,13

A typical low-resolution micro-PL spectrum of 200-nm thick Al_0.23Ga_0.77N (sample A) is shown in Fig. 1(a). The dominant peak at around 3.92 eV (with accompanying LO phonon replicas on its low energy side) corresponds to the emission from the AlGaN bulk. In between the AlGaN peak and the GaN band-edge emission (∼3.49 eV), we occasionally see several sharp peaks at around 3.6 eV. As described in detail below, such sharp peaks appear only at spatially localized spots.

Figures 1(b) and 1(c) show low temperature (∼7.5 K) surface emission images taken from two samples: sample A and a similar sample with a 100-nm thick Al_0.27Ga_0.73N layer (sample B), respectively. The excitation area was rather wide (∼1500 μm²), and the excitation conditions were adjusted for each sample so that the localized emissions can be clearly seen. An optical bandpass filter (window: 330–349 nm) was used to differentiate the localized emissions from the bright background (emissions from AlGaN bulk and GaN) for both the samples. Compared to the homogeneous background emission (filtered fraction of LO phonon replicas of AlGaN bulk emission) in sample B, many bright spots are clearly observed in sample A. It is also seen that the bright spots are well aligned along lines which are parallel to the ⟨1$1̄$00⟩ crystal axes. Several other samples with different AlGaN thicknesses (150–375 nm) and Al contents (0.17–0.24) were also investigated, revealing that only the thicker samples (thicker than ∼180 nm) exhibit the bright spot arrays. It is known that misfit dislocations (MDs) are generated along the ⟨1$1̄$00⟩ directions to relax the misfit strain in AlGaN/GaN.¹⁸ Although the reported threshold thickness for MD generation is ∼5 times thinner than the above-mentioned thickness,^19,20 the results strongly suggest that the bright spot arrays are closely related to the existence of MDs.

To further investigate the localized emission peaks, we carried out micro-PL mapping on various regions with bright spots from sample A. Figures 2(a) and 2(b) depict PL integrated intensity maps taken along an array of bright spots (area: x 2.5 × y 25 µm²). The PL spectra recorded at each position (0.5 µm step in both x and y axes) were integrated within the energy ranges of 3.5500–3.6500, 3.5733–3.5743, and 3.5748–3.5758 eV to produce the maps shown in Figs. 2(a) and 2(b) (left), and Fig. 2(b) (right), respectively. Almost all of the inspected localization centers are situated along a straight line [corresponding to the dashed vertical line in Fig. 2(a), x = 1.0 µm], which is parallel to ⟨1$1̄$00⟩, as described earlier. The PL spectra taken along this line are shown in Fig. 2(c). Spectrally sharp and spatially localized emission peaks are again clearly seen, and it is also noticeable that the spectra vary considerably from position to position. As an example, the spectra measured at the three representative positions [P1–P3 in Fig. 2(a)] are separately displayed in Fig. 2(d). The nominal spatial broadening of each localized emission in the maps (∼2 µm) corresponds to the spatial resolution limit of the PL mapping (determined by the size of a confocal-pinhole spatial filter that was inserted into the detection optical path). This implies that the actual lateral dimension of each center is far smaller than the width of the detection window (∼2 µm). From Fig. 2(c), the number of the sharp emission peaks is ∼100 in total, and we note that there also exist a small number (less than 10) of additional peaks outside the depicted emission energy region. Considering that all the localization centers investigated here are almost linearly aligned, the average spacing between the adjacent centers is estimated to be ∼0.2 µm, though there exist denser and sparser sections. However, this number could be an underestimate, as it is possible that more than one emission peak originates from a single localization center, as discussed later.

Typical linewidths of the emission peaks observed from the localization centers studied in this work are ∼150 µeV to several hundreds of μeV, which is much narrower than typical self-assembled GaN QDs.¹⁵ The spectrum of the localization center exhibiting the narrowest linewidth is presented in Fig. 3. When fitted with a Gaussian function, the linewidth of the peak is estimated to be as small as 40 ± 0.9 µeV, which is even narrower than the emission from state-of-the-art III-nitride interface-fluctuation QDs.¹² This value is close to the resolution limit of the setup (∼35 µeV¹²), and it is therefore not easy to accurately analyze the peak width assuming a Voigt line shape (which, being composed of both a Lorentzian and a Gaussian component, would provide information on the homogenous linewidth, γ_h, of the emitter and also the extent of inhomogeneous spectral diffusion effects, γ_inh). Nevertheless, it is possible to get some information under the assumption of two special cases: (1) spectral diffusion in the emission spectrum is negligible and (2) spectral diffusion completely dominates the emission spectrum. If we assume that the Gaussian component of the line shape is determined entirely by the instrument response (i.e., spectral diffusion in the emission can be neglected, γ_inh ≈ 0), we can extract an upper limit on the Lorentzian linewidth γ_h of 6 ± 1 µeV. On the other hand, if we assume that the Gaussian-fitted linewidth of 40 µeV is comprised entirely from a convolution of the instrumental response and a purely Gaussian emission peak (i.e., spectral diffusion dominates the emission linewidth), an upper limit of the inhomogeneous broadening γ_inh of 19 ± 2 µeV is obtained. Regardless of the exact analysis used, the strikingly small inhomogeneous broadening means that spectral diffusion is heavily suppressed in this localization center. The 6 µeV upper limit on the homogeneous linewidth for this emitter is still much broader than would be expected solely due to lifetime broadening (the lifetimes of these emitters are typically measured to be ∼1–3 ns). This result therefore also places a lower bound of ∼200 ps on the pure dephasing time, T₂, which we note is comparable with the dephasing times measured in some studies on III-As QDs or compositional fluctuations in InGaN quantum wells.^21,22 It should be stressed that this result represents the most direct evaluation of the homogeneous linewidth and dephasing times of III-nitride QDs ever reported, thanks to the heavily suppressed spectral diffusion (at most 19 ± 2 µeV) of this emitter.

Among the various sharp emission peaks, a pair of peaks were further investigated and their excitation power dependences are plotted in Fig. 4. As can be seen in the figure, the peak X at ∼3.596 eV has linear dependence on the power, whereas the other peak XX at ∼3.599 eV indicates quadratic dependence. It is also confirmed that the locations (center positions) of the localized emissions corresponding to those two emission peaks are identical (according to a PL mapping analyzed similarly to that in Fig. 2). These properties strongly suggest that both excitons and biexcitons can be localized in a single localization center. The estimated biexciton binding energy for this case is −3.0 meV, which is comparable to the reported values for various GaN QDs emitting in the same energy range.^2,12,23 Although the biexciton binding energy in GaN QDs has a non-trivial dependence on the QD geometry,²³ the similarity between the value presented here and the values found in the literature could support the assertion that the confinement potentials of these emitters are similar to those of self-assembled GaN QDs (in terms of aspect ratio and physical size). However, we note that this relatively small biexciton binding energy would likely prohibit the measurement of single-photon emission at higher temperatures, where increased thermal broadening would result in peak overlap and prevent the spectral isolation of an isolated transition. In the future, it may be possible to tune the magnitude of the biexciton binding energy to larger values in order to facilitate high-temperature single-photon emission by somehow controlling the sizes (and/or aspect ratios) of the emitters. Increasing the Al composition in the AlGaN would also be beneficial in this regard by increasing the confinement potential.

Finally, second-order photon-autocorrelation measurements were carried out on an excitonic emission from a localization center. Figure 5(a) shows the measurement data, which clearly exhibits antibunching at time delay zero. The estimated g⁽²⁾(0) value is 0.47 ± 0.10. Although this raw value is at the limit for verifying single-photon emission, we note that the non-zero value can be largely attributed to the degree of background emission which is present under the relatively high excitation power used [see Fig. 5(b)]. Note that a relatively low-resolution grating was used in order to obtain a reasonably high photon counting rate. The signal to total intensity ratio (ρ) of the measured spectral range (indicated by the red dashed rectangular box in the figure) is estimated to be ∼0.77. As the lower limit for a measured g⁽²⁾ (0) value for such a case with uncorrelated spectral background can be expressed as 1−ρ²,²⁴ it is fair to state that the background-corrected value for the g⁽²⁾ (0) in this study should be considerably smaller than 0.47, which confirms the single-photon nature of the emission from the localization center itself. According to the result, it is also confirmed that single excitons can be confined in a localization center.

Based on these findings, we now discuss possible structural origins of the localization centers. As described earlier, it is likely that the centers are related to MDs generated when the accumulated tensile strain in AlGaN is relaxed. First, we examine the possibility that the MDs themselves are origins of the localization centers. The MDs in AlGaN/GaN are known to be dissociated into two partial dislocations, 1/2a + 1/2c.²⁵ The atomic structure of a 1/2a dislocation is also proposed to be very similar to a-plane stacking faults,^26,27 which have been reported as being related to radiative emission.²⁸ Moreover, emission from MDs in some other materials, such as ZnSe/GaAs, has been previously reported.²⁹ It stands to reason, therefore, that MDs in AlGaN may also facilitate radiative recombination processes. However, a MD is a one-dimensional defect (two-dimensional confinement), and thus, there must be an additional factor to provide the three-dimensional carrier confinement that would allow single-photon emission. At this stage, we suggest the possible presence of impurities, Al compositional fluctuations, or the existence of (nearly) intersecting dislocations as candidates for the additional dimension of confinement.^29–32 We note that Medvedev and Vyvenko have proposed that intersection nodes of basal screw dislocations in GaN can act as optically active QDs.³³ Similarly, intersecting MDs might also create QDs at the nodes. However, it should be also noted that the observed luminous centers seem to be concentrated along a linear structure, whereas the intersection nodes should, in principle, be distributed randomly over the sample.

Next, we discuss the possibility that the localization centers may nucleate during growth along surface steps that are induced by the MDs. Indeed, it is well known that self-assembled QDs can be ordered along the misfit-dislocation-induced steps or surface structures in various material systems, such as SiGe/Si³⁴ or InAs/GaAs.³⁵ In this study, however, we have not intentionally grown any QD layers or similar heterostructures during the growth of the AlGaN. We performed atomic force microscopy studies on these steps in relaxed AlGaN layers (see the supplementary material for details) to investigate their heights and density. In general, the steps are found to have a height of ∼0.5 nm, exactly corresponding to the c-axis component of the reported Burgers vector, a + c, of a MD.²⁵ However, we also found that a subset of the steps is larger and exhibits a height of 2c (∼1.0 nm). Such a 2c-high step will reveal a semipolar (11$2̄$2) facet, which could lead to preferential nucleation of Ga adatoms rather than Al counterparts on the step edge during overgrowth (whereas a c-high step formed from a standard MD does not generate any well-defined facet on the growth surface as the uppermost atoms at the step edge may deviate from their natural sites due to reconstruction). The preferential Ga nucleation will result in the formation of Ga-rich localization centers aligned along the 2c-high MD steps (which occur at a lower density than the usual c-high MD steps). Indeed, in the literature, there are several reports that show a local increase in the Ga concentration of AlGaN quantum wells at macrosteps.^36,37 Such a scenario is consistent with our experimental results that the observed localization centers form along linear structures that are well aligned with MDs, which exhibit a density that is apparently lower than the total density of MDs. In addition, it is plausible that a 2c-high MD could relax more strain and, thus, have a thicker threshold thickness than a standard MD, further explaining the observed property that localized emitters form in relatively thicker AlGaN samples.

As for the suppressed inhomogeneous broadening, three possible factors may play important roles; a reduced point defect density in the AlGaN matrix, a moderate electric field due to a rather small lattice mismatch between the surrounding AlGaN and the localization centers themselves, and a preferable spatial distribution of charged defects near a localization center. These factors are roughly the same, as has been discussed with respect to interface-fluctuation QDs.¹² Note that the typical emission lifetimes of these localization centers is 1–3 ns (not shown), which is very similar to the lifetimes of other GaN QDs emitting in the same wavelength range. It is therefore likely that the dipole moments of excitons in the localization centers studied here are not largely different from those in other GaN QDs. It is possible that the quasi-one-dimensional nature of the localization center arrays may influence the distribution of charged defects which can critically determine the degree of spectral diffusion.¹⁶ 

In summary, we have discovered that exciton localization centers are spontaneously formed in AlGaN layers grown on GaN. Those emitters are ordered along misfit dislocations running along ⟨1$1̄$00⟩ directions. Both excitons and biexcitons can be confined within a single center, as revealed by micro-photoluminescence measurements. The emission peaks from individual localization centers are very sharp, and a strikingly narrow spectral linewidth, as narrow as 40 µeV, is observed from a localization center. It is revealed that an almost spectral-diffusion free state with a record-low inhomogeneous broadening (γ_inh ≤ 19 ± 2 µeV) is realized in polar III-nitride quantum nanostructures. Thanks to the low γ_inh, a lower bound on the pure dephasing time, T₂, of ∼200 ps for this emitter is also evaluated. Furthermore, single-photon generation from a localization center is demonstrated. Possible structural origins of the localization centers are discussed, and some plausible factors that lead to the heavily suppressed inhomogeneous broadening are suggested. These findings demonstrate that there is a certain way to dramatically suppress the spectral diffusion in excitonic emissions from III-nitride low dimensional nanostructures and will pave the way to further explore their electronic properties, leading to the development of sophisticated quantum emitting devices based on III-nitrides.

See the supplementary material for discussions and the additional data on (i) reported linewidths for various III-nitride quantum nanostructures, (ii) atomic force microscopy observations of misfit-dislocation induced steps, and (iii) differences in structural and optical characteristics between GaN interface-fluctuation QDs and the localization centers studied here.

This work was supported by the Project for Developing Innovation Systems of the Ministry of Education, Culture, Sports, Science and Technology (MEXT) and the New Energy and Industrial Technology Development Organization (NEDO), Japan. The authors thank M. Nishioka for technical support.