Emission properties of Ga2O3 nano-flakes: effect of excitation density

In the quest of developing high performance electronic and optical devices and more cost effective fabrication processes of monoclinic β-Ga2O3, new growth techniques and fundamental electronic and optical properties of defects have to be explored. By heating of dissolved metallic Ga in HCl in a NH3 and N2 atmosphere, nano-flake films of monoclinic β-phase Ga2O3 were grown as confirmed by XRD. From optical measurements, we observe two strong emissions. A red band peaking at ~2.0 eV and a UV band at ~3.8 eV. The band at ~2.0 eV is attributed to donor-acceptor pair recombination where the donor and acceptor level is suggested to be related to VO and nitrogen, respectively. By studying the dependence of the intensity of the UV band at 3.8 eV versus excitation density, a model is suggested. In the model, it is assumed that local potential fluctuations forming minima (maxima), where the carriers would be localized with a summarized band offset for conduction and valence band of 1 eV. The origin of the fluctuations is tentatively suggested to be related to micro-inclusions of different phases in the film.

has almost completely vanished. This phenomena is illustrated in Fig. 3, where we compare the CL spectrum (red dashed line) taken in scanning mode from several nanoflakes shown in Fig. 3(a) with the spectrum for one nanoflake (blue solid line) measured from the point indicated at the SEM image in Fig. 3(b). In the visible band, two features at 1.8 and 2 eV can be resolved. The UV band has one maximum at ~3.7 eV though the line shape is slightly asymmetrical.
The increase of the electron beam current can have similar effect even if the measurements are done in a scanning mode. A definitive conclusion could not be done because the maximum current was limited by instrument to 3.73 nA. On the other hand, it seems that reducing the scan area is a more efficient way of excitation of the UV emission. The CL data are presented in Fig. 4(a) and (b) for three different beam currents for the case when CL spectra are collected from the smaller and larger scan areas of ~10 μ m × 10 μ m and ~50 μ m × 50 μ m, respectively. To minimize the contribution of possible inhomogeneity of the film, the sample position between measurements in (a) and (b) was not changed, only the magnification was different. We have examined several points and the effect was similar, i.e. at the higher excitation densities we observe an enhancement of the UV band compared to the red emission. Thus, it is clear that the excitation conditions influence the shape of the CL emission. Data in Fig. 4(a) are similar to the case of point excitation when the UV line dominates CL spectrum even at the lowest current of 0.25 nA. For the larger raster, see in Fig. 4(b), three emission bands, UV, blue and red, can be detected with maxima at ~3.8, ~3.0 and ~2.1 eV, respectively. However, now the red emission dominates the CL spectrum at all beam currents although the relative intensity of the UV band slightly increases with the beam current. The weakest band at ~3.0 eV (so-called blue emission) is not resolved under excitation by higher electron beam currents. Another observation is related to the spectral shift caused by excitation conditions. The maximum energy of the UV band (i.e. 3.8 eV in scanning mode) is shifted compared to the peak position of 3.7 eV observed under measurements in focused mode ( Fig. 3(b)). It is interesting to compare CL spectra with PL spectrum shown in Fig. 4(c), since excitation densities are much lower in PL than in CL, where one electron with the energy of 5 keV can create several hundred electron-hole pairs. Since the PL excitation was done using a laser wavelength of 266 nm (or 4.66 eV), i.e. inside the β-Ga 2 O 3 band gap, only intraband transitions are possible. We have observed a weak broad emission with a maximum at ~2.2 eV and with additional features at ~1.9 and ~2.5 eV, respectively. As mentioned, taking into account the lower excitation densities in PL compared to CL, the PL peak position is in line with the observed trend in CL under excitation using lower electron beam currents.
Temporal behavior of the red band was analyzed by time-resolved PL. We have found that PL recombination time strongly depends on the wavelength. Thus, for the high energy side the PL decay curves obey a bi-exponential law with fast and slow components, while for the low energy shoulder, the PL intensity decays rather slowly following a single exponential decay function. The examples of the PL decay curves are presented in Fig. 5 by black lines. Fitting using a bi-exponential decay law is also shown (red lines). For the peak position and for the higher energy side of the emission band, the fast recombination time is ~80 ps. The slow component becomes more dominant for the lower photon energies and for the feature at 1.98 eV the PL decay is characterized by only a slow recombination time exceeding 5 ns.  [14][15][16][17] to the recombination of carriers trapped by the oxygen and gallium vacancies forming deep donor and acceptor states, respectively. In contrary, the UV and red lines have been related to the presence of a subband formed by amorphous Ga 2 O 3 , which is in contradiction with results reported by other groups [14][15][16] . The UV luminescence (~3.45 eV), with a reference to the early paper of Harwig et al. 14 , is commonly assigned to a recombination of self-trapped excitons considering that this emission is absent under excitation below the band gap. The green emission (~2.3 eV) was usually observed in samples doped, for example, by Si 17 ; however, Villora et al. concluded that this band is associated with self-trapped or bound excitons 18 . The red band also detected in single crystal β-Ga 2 O 3 19 has been related to electron-hole recombination via the vicinity donors and acceptors 19 . However, it was also reported that the appearance of the red band at ~1.78 eV can be correlated with nitrogen doping 20 .
Intrinsic β-Ga 2 O 3 is unintentionally n-type doped, though the residual donor concentration is very low (~10 13 cm −3 ) and the origin of the n-type conductivity is still under debate. Binet et al. 16 suggested that oxygen vacancies (V O ) play a role of donors with ionization energy about 40 meV. This has been questioned by Onuma et al. 6 after electrical measurement studies correlated with CL. They suggested that V O acts as a deep level with an activation energy of ~1.0 eV. From theoretical calculations, it has been ruled out that V O is a shallow donor in intrinsic β-Ga 2 O 3 . It was suggested that Si can be used as a shallow donor 22 . Recently, it was shown using density function theory (DFT) calculations that C-H groups are deep acceptors in β-Ga 2 O 3 and compensate for the intentionally introduced donors 23 . Clearly, intrinsic point defects and impurities affect emission spectra in β-Ga 2 O 3 , however, it seems that a correct identification is complicated.
While our samples have not been specifically doped, common impurities such as Si, N, H and C could be presented. We have observed only two strong emissions: the red band peaking at ~2.0 eV and the UV band at ~3.8 eV. The last one is almost unknown except it was mentioned in ref. 13. Based on our investigations, the red emission band peaking at ~2.0 eV is likely due to donor-acceptor pair (DAP) recombination. The energy of the DAP emission depends on the distance between donor and acceptor meaning also that the recombination time for such emission is depending on photon energy. For the carrier pairs separated by large distances (and recombining emitting lower energy photons), the recombination time will be longer. We have observed those characteristics for the red band at ~2.0 eV. The energy of the DAP emission is written as: (1) Here r i is the donor-acceptor distance, the band-gap energy E g = 4.8 eV, E D and E A are donor and acceptor levels, respectively. If the energy of donor (i.e. V O ) is E D = ~1.0 eV 6 , then the acceptor levels energy can be estimated from Eq. (1) to E A = 1.8 eV, taking into account the CL peak position of ~2.0 eV for the red emission. The acceptor levels can likely be associated with nitrogen, since our samples were annealed in ammonia and nitrogen gases. Indeed, the energy dispersive X-ray spectroscopy (EDS) composition analysis of the Ga 2 O 3 nano-flake film has shown a presence of nitrogen with concentrations between 0.5 and 1.8 Wt %. Places on the film with lower nitrogen concentrations demonstrated the red emission peaking at 2.0-2.1 eV (Fig. 2(a)), while areas with high nitrogen concentration have shown red band centered at ~1.8-1.9 eV (Fig. 2(b)).
DFT calculations show that nitrogen forms deep acceptor levels at about the center of the band gap 24 , which is in qualitative agreement with our finds. Song et al. 25 indicated that there is a shift of the red emission to the lower energies with increasing nitrogen concentration. This is in line with our observations as shown in Fig. 2 for the CL measurements taken at the film surface, which was more exposed to nitrogen and to ammonia during annealing. Thus, we suggest that the red emission E R (r) at ~2.0 eV corresponds to the DAP recombination, where donors are formed by oxygen vacancies, while acceptor levels are formed by nitrogen impurities. See Fig. 6 showing schematic representation of the band diagram in Ga 2 O 3 , where levels D(V O ) and A 2 (N) have energies E D = ~1.0 eV and E A2 = 1.8 eV from corresponding band edges, respectively.
The assignment of the UV emission at ~3.8 eV is more complicated. With an increasing excitation densities, a strong enhancement of the UV emission compared to the red line is detected in our samples. Additionally, the shift of the peak position to lower energies (~0.1 eV) is observed for the highest excitation density, when the electron beam is focused to one point ( Fig. 3(b)). It is understandable that as more free electrons are created, the DAP recombination (i.e. red emission) decreases since the number of donor and/or acceptor levels is limited, while free carrier recombination should increase. It is known that the Mott transition can occur at high carrier densities leading to the band gap renormalization with corresponding red shift of the band-to-band transition 26,27 , which can explain the 0.1 eV shift to the lower energies for the UV line. However, the observed peak position of ~3.8 eV (corresponds to E UV in Fig. 6) is below the band gap energy in β-Ga 2 O 3 by ~1 eV. Such difference can hardly be explained by only the band gap narrowing, which is usually in the range of several meV, occurring at high excitation density or high doping [26][27][28][29] .
We assume now that the potential is not uniform and that there are local potential minima in the material. The CL data can be understood in the frame of processes schematically illustrated in Fig. 7. Under low excitation, created electrons (holes) are firstly relaxed to the band edges and trapped by the defect levels inside the band gap and then carriers recombine producing a photon with the energy hν 1 corresponding to the red emission as shown in Fig. 7(a). The recombination via the defect levels inside the band gap is the main recombination channel, which explains our observations that the red emission band is dominating in this case. When the excitation density is high, the defect traps are filled as depictured in Fig. 7(b). Consequently, the relative contribution of radiative transitions via the traps will be reduced compared to the band-to-band recombination. However, the peak energy of 3.8 eV for the UV line is lower than the band gap energy meaning that the transition origin is not directly related to the band-to-band recombination in β-Ga 2 O 3 . Likely, the potential profile is not uniform within the film and there are local potential fluctuations forming minima (maxima), where the carriers would be localized with a summarized band offset for conduction and valence band of 1 eV. It is not necessary that potential minima for electrons would correspond to the potential maxima for holes indicating that electron and hole recombination (with photon energy hν 2 ) in such indirect potential wells can be much less efficient compared to the emission via the defect levels. However, the UV luminescence starts to dominate CL spectra at higher excitation densities while the defect luminescence starts to saturate. This model describes rather reasonable the observed experimental data, though still the origin of the potential fluctuations should be determined. Tentatively, we suggest that such potential fluctuations can be caused by micro-inclusions of different phases as it was pointed by Zhou et al. 13 and/or structural defects in similarity with stacking faults building a quantum well-like formation with characteristic emission energy as observed, for example, in GaN 30-31 . High resolution transmission electron microscopy Optical measurements revealed two strong broad emission peaks. A red band peaking at ~2.0 eV and a UV band at ~3.8 eV. Due to the behavior of the recombination time with increasing photon energy and the peak position of the band at ~2.0 eV, we suggest it is related to donor-acceptor pair (DAP) recombination where V O is the donor. The acceptor level could be related with nitrogen, since the nano-flakes were grown in a nitrogen containing ambient.
A model of the origin of the UV band at ~3.8 eV was suggested. In the model, it is assumed that local potential fluctuations form minima (maxima), where the carriers would be localized with a summarized band offset for conduction and valence band of 1 eV. However, the origin of the potential fluctuations is still not verified; tentatively, we suggest that they can be caused by micro-inclusions of different phases as it was pointed by Zhou et al. 13 .

Methods
For growth of polycrystalline Ga 2 O 3 nanoflake films we used the following procedure: at first, metallic Ga was dissolved in HCl mixed with deionized water. Then, the gallium chloride solution was drop deposited on the substrates and placed in the oven for annealing at 600 °C in ammonia (NH 3 ) and nitrogen gas flow for 30 min in the halide vapor phase epitaxy (HVPE) reactor. As substrates, we used (0001) sapphire of optical quality, which is convenient due to its chemical stability and due to the absence of the luminescence signal simplifying optical characterization of Ga 2 O 3 .
X-ray diffraction (XRD) scans were performed in a Philips 1820 Bragg-Brentano diffractometer using Cu-Kα radiation. CL measurements were done using a standard Leo 1500 Gemini scanning electron microscope (SEM) combined with the MonoCL4 system. CL images were acquired using a Peltier-cooled GaAs photomultiplier tube. Spectral acquisition was done with a CCD detector using a 150 l/mm diffraction grating. The spectral resolution under the experimental conditions was ~10 nm. All measurements were done at room temperature and at acceleration voltage of 5 kV. Standard electron beam current was ~0.25 nA, in other cases it is indicated separately. Chemical composition was analyzed using energy dispersive X-ray spectroscopy (EDS) equipped with an Oxford LINK ISIS system with a Ge detector. TEM imaging was done with a high resolution FEI Tecnai G2 200 keV FEG microscope. Nano-flakes were removed from the substrate, suspended in ethanol and then placed on the 200 mesh copper grid. For detection of time-resolved photoluminescence we have used a Hamamatsu synchroscan streak camera with a temporal resolution of 20 ps. The available excitation wavelength of 266 nm (i.e. below the band gap of β-Ga 2 O 3 ) was a third harmonics of the Ti-sapphire femtosecond pulsed laser with a frequency of 75 MHz.