UVB-emitting InAlGaN multiple quantum well synthesized using plasma-assisted molecular beam epitaxy

Compact, low-cost, and efficient solid-state light sources emitting in the deep ultraviolet (UVB: 280 nm – 320 nm) are needed for a range of applications including water purification, and industrial waste photocatalytic decomposition.^1,2 In comparison to InGaN light emitting diodes (LEDs) operating in the visible, high Al-content (y > 0.4) Al_yGa_(1-y)N-based LEDs are less mature and have low efficiencies and short lifetimes due to the relatively large dislocation density.^1–4 In addition, the difficulty in achieving high p-type conductivity in AlGaN using metalorganic chemical vapor deposition (MOCVD) growth leads to poor current spreading and serious thermal management challenges.^4,5 InAlGaN has emerged as a promising alternative because it is known to be less sensitive to dislocations than AlGaN is.^6–11 On the other hand, molecular beam epitaxy (MBE) has been used recently to demonstrate p-type Al_xGa_(1-x)N with a hole concentration up to $2×1019cm−3$ for x = 0.27.¹² However, there are no reports on the MBE growth of high Al-content (y > 0.4), low In content (x<0.05) In_xAl_yGa_(1-x-y)N multi-quantum-well (MQW) structures. In addition, the relationships between materials properties, especially defect density and type, and optical characteristics of high Al-content (y > 0.4) In_xAl_yGa_(1-x-y)N MQW structures are poorly understood, so optimal high Al content (y > 0.4) quaternary MQW structures have yet to be identified.

Here we report on a comprehensive investigation of the structural and optical properties of a high Al-content (y > 0.4) In_xAl_yGa_(1-x-y)N MQW grown using MBE. The insights gleaned from this investigation, along with numerical simulations of the MQW layer, allow for a better understanding of the material properties needed to achieve the target of an efficient InAlGaN-based deep UV emitter.¹ 

A Veeco Gen II rf plasma-assisted molecular beam epitaxy (PAMBE) system was used to synthesize InAlGaN thin films and a MQW on 5 μm thick hydride vapor phase epitaxy GaN templates. Each sample growth was initiated with a 50 nm thick GaN buffer layer grown at a substrate temperature of 680°C using Ga-rich conditions. InAlGaN and AlGaN reference samples were grown, in addition to a MQW sample, all with thicknesses of 64 nm at a substrate temperature of 600°C. The N₂ plasma condition was 350 W and 1.0 sccm for all samples. The compositions of the two reference samples were determined with Multi-voltage multi-layer electron probe microanalysis by using a Cameca SX51 electron probe with 7 and 12 keV beam energy, and the N Ka was analyzed with a 100 Å pseudo-crystal and a curved background fit utilizing Probe for EPMA software. The well and barrier compositions of the MQW were designed to match the two reference samples: the MQW is comprised of five 3 nm thick In_0.05Al_0.47Ga_0.48N wells separated by six 8 nm thick Al_0.5Ga_0.5N barrier layers. Well and barrier compositional changes were achieved by opening and closing the In shutter, while the N-rich growth condition of the MQW ensured all the In adatoms were incorporated in the well regions. The substrate temperature and metal fluxes were held constant in order to achieve abrupt interfaces without any growth interruption. Consequently, well and barrier compositions can be maintained throughout the MQW; however, due to differing strain and strain relaxation in the multilayered structure, absorption and emission wavelengths are expected to shift compared to the thin films of similar compositions. The absorption characteristics of the samples were measured using a Horiba-HJ UVISEL spectroscopic ellipsometer (SE) with photon energy 1.5 eV to 6.5 eV. Room temperature photoluminescence (RTPL) spectra were recorded using an Acton SpectraPro 300i 30cm spectrometer with a liquid nitrogen-cooled back-illuminated charge coupled device detector from Roper Scientific. The samples were excited by a continuous wave 244 nm laser from Cambridge Laser Labs. Although an actual emitter would use a high excitation intensity, we used a low excitation irradiance of ∼ 60 mW/cm² for a 1 mm diameter spot in order to explore the underlying physics.

In order to ascertain the role of strain relaxation in the MQW sample, Fig. 1 (a–c) compares its (015) X-ray diffraction (XRD) reciprocal space map (RSM) with those from the In_0.05Al_0.47Ga_0.48N and Al_0.5Ga_0.5N reference samples.²² Fig. 1 (a) shows the 0^th order peak of the MQW, which represents the average composition of the structure. The relaxation of the MQW is similar to that of the Al_0.5Ga_0.5N reference sample, with slightly higher Al and In content in the QW barrier and well layers, respectively, than in the corresponding thin films. The Al_0.5Ga_0.5N reference sample with a lattice mismatch of 1.2% is partially relaxed, as can be inferred from the peak center shift to R ≈ 0.1 ∼ 0.3 with the most relaxed part reaching R ≈ 0.8. In contrast, the diffraction peak of the In_0.05Al_0.47Ga_0.48N reference sample is symmetric and fully strained to the GaN template with a lattice mismatch of 0.6%.

A further indication of how the MQW structure modulated strain and altered the defect and dislocation density was gleaned by using AFM to characterize its surface morphology and that of the two reference samples (Fig. 1). Measurements were taken after etching the samples in 50% HCl for ten minutes to reveal the surface depressions. All three samples retain relatively smooth surfaces with RMS ranging from 1.08 nm to 1.69 nm, comparable to a bare GaN template (RMS = 1.10 nm). The mottled surface of the In_0.05Al_0.47Ga_0.48N reference sample is distinctively absent in the other two samples, a consequence of the different surface chemistry associated with etching In-containing surfaces. Larger surface depressions with diameters of 10 - 50 nm and an average depth of 2 nm are observed for both the In_0.05Al_0.47Ga_0.48N and Al_0.5Ga_0.5N reference samples with densities of $∼108cm−2$ and $∼109cm−2$⁠, respectively. These depressions can be attributed to open-core threading dislocations (TDs).^13,14 The higher density observed for the Al_0.5Ga_0.5N sample is due to relaxation, which is supported by the XRD data shown in Fig. 1 (b) and (c). The TD density of the MQW is estimated to be on par with that of the Al_0.5Ga_0.5N reference: the (002) and (015) plane ω scan full widths at half maximum (FWHMs) of these two samples are comparable. These dislocations may act as midgap traps reducing the internal quantum efficiency.¹⁵ On the other hand, as shown in Figs. 1 (d) and (e), the MQW sample has a surface depression density similar to that of the In_0.05Al_0.47Ga_0.48N. The density reduction is likely due to strain modulation resulting from compositional variations intrinsic to the MQW structure which preferentially nucleates closed-core TDs or facilitates TD bending at the barrier/well interfaces.¹⁶ It is also possible that the addition of In flux during MQW growth increases the III/V ratio, effectively lowering the formation energy of Ga-filled core TDs or nitrogen vacancies, in comparison to that of open-core TDs.^17–20 

Spectroscopic ellipsometry was used to estimate the dielectric functions of the reference samples and the MQW using the methodology described in previous reports.^21–23 The refractive index $n$ and the absorption coefficient $α$ were derived from the real and imaginary parts of the dielectric function $εr(ℏω)$ and $εi(ℏω)$⁠. The bandgap energy E_A is obtained by linearly extrapolating the function $α2(ℏω)$ to zero, as shown in Fig. 2 (a).²⁴ As such, the band edge energy $EA$ of the well and barrier of the MQW are separately measured to be 4.21 eV and 4.70 eV. In comparison, the $EA$ of wells is slightly lower than the reference sample In_0.05Al_0.47Ga_0.48N (⁠$EA$ = 4.24 eV), and the $EA$ of the barriers is slightly higher than that of the Al_0.5Ga_0.5N (⁠$EA$ = 4.57 eV) reference.

Room temperature photoluminescence spectra in Fig. 2 (b) in indicate UV emission peak energies $EPL$ of 3.92, 3.75, and 4.24 eV, for the MQW, In_0.05Al_0.47Ga_0.48N sample, and Al_0.5Ga_0.5N sample, respectively (Table I). The emission peaks from the GaN templates are the near bandedge peak (3.42 eV), while the yellow luminescence spans 1.5 - 3.2 eV. Using the $EA$ determined by SE, the Stokes shifts of the In_0.05Al_0.47Ga_0.48N and Al_0.5Ga_0.5N samples are 0.49 eV and 0.33 eV, respectively. Clearly, the introduction of In induces a larger Stokes shift, which has routinely been attributed to carrier localization at In centers.^6–11 On the other hand, the calculated emission energy of the MQW is 4.390 eV, with a Stokes shift of 0.467 eV similar to the In_0.05Al_0.47Ga_0.48N sample. This suggests In-related carrier localization also plays a role in the In-containing MQW. In addition, the RTPL intensity of the MQW is ∼10× higher than the In_0.05Al_0.47Ga_0.48N sample and ∼20× higher than the Al_0.5Ga_0.5N sample. Such emission enhancement is due to the strong carrier confinement in the MQW that effectively reduces the radiative recombination lifetime. In addition, the In-related carrier localization in the well region further reduces non-radiative recombination at dislocation sites by confining the carriers at the In centers. Intense yellow luminescence in the visible range (1.5 ∼ 3.2 eV) is also observed for all three samples, the origin of which will be discussed below.

The band structure, carrier distribution and recombination characteristics of the MQW was studied using a Silvaco Atlas simulator. Experimentally-determined material parameters were used when possible, with the exception of the electron affinities $χInN=5.34eV$⁠, $χGaN=4.31eV$⁠, and $χAlN=2.42eV$⁠,²⁵ the carrier mobilities $μn=200cm2V−1s−1$ and $μp=10cm2V−1s−1$⁠,^26,27 the non-radiative Shockley-Read-Hall (SRH) lifetime $τSRH=1ns$⁠,^28,29 and the radiative combination constant $B=1.1×10−8cm3s−1$⁠.³⁰ The strain of each layer assumes a relaxation of $R≈0.8$⁠, as indicated by the XRD results shown in Fig. 1 (c). To account for the reduction of the piezoelectric polarization field in the structure due to relaxation, the polarization surface charge is calculated and reduced to a value consistent with R.³¹ The band edge energy and dielectric function of the well and barrier regions are chosen with the values based on SE characterization and modeling (Table I). The sensitivity of simulation results to material parameters is discussed below.

Fig. 3 (a) compares the simulated energy band diagrams of the MQW structures under $60mW/cm2$ excitation, with the solid curve representing the structure with $R=0.8$ relaxation and the dotted curve representing a fully strained R = 0 structure. Band bending with a steep triangular slope in the well and barrier layers is observed, caused by the strong polarization field in III-N heterostructures. In the well regions, the electron density (Fig. 3 (b)) is primarily a result of polarization-induced carriers and ranges from $1017∼1018cm−3$⁠, while the density of optically generated holes spans $1014∼1015cm−3$⁠. Due to the quantum confined Stark effect, the electrons and holes are concentrated at opposite QW interfaces, thus reducing carrier overlap and recombination efficiency. Fig. 4 (a) shows the calculated QW radiative recombination rate $rrad′$⁠, integrated across the QWs, as a function of well width w under different polarization fields. The result shows the optimum w decreases as the polarization field increases. A width of $w∼4nm$ is required to achieve the highest $rrad′$⁠, assuming a fully strained MQW. Deviation from the optimum w will result in a lower $rrad′$ due to either reduction in the confinement of carriers (⁠$w<4nm$⁠) or carrier overlap (⁠$w>4nm$⁠). A non-polar MQW with zero polarization field is not affected by this carrier overlap issue. On the other hand, a higher optimal $rrad′$ is found for a polarized structure due to an increase in the effective well depth (Fig. 3 (a)) which produces a higher carrier density for a given carrier injection level. The effect of well depth d on $rrad′$ is shown in Fig. 4 (b). Initially, $rrad′$ increases with increasing d until $d∼0.4eV$ due to better carrier confinement and increasing carrier density in the wells. A further increase of d leads to a reduction in $rrad′$ and an increase of non-radiative SRH recombination.

The solid curve in Fig. 3 (c) is the simulated local radiative recombination rate $rrad(z)$ at locations z within the structure for an excitation power density of $60mW/cm2$⁠. Note that the QW closest to the surface dominates the optical emission, with $rrad(z)$ approximately two orders of magnitude higher than that of the fifth QW. Since $rrad(z)$ is directly proportional to the local electron density, this is the result of a much higher electron concentration in the first QW caused by the downward band bending, as shown in Fig. 3 (a) and (b). For similar reasons, the $rrad(z)$ at the MQW - GaN buffer interface is comparable to that of the first QW due to the presence of a polarization-induced electron accumulation on the order of $1019cm−3$⁠. This explains partially the bright yellow luminescence in the MQW and reference samples (Fig. 2 (b)), both of which are induced by recombination through dislocation-related midgap states at the buffer interface. As such, the yellow luminescence can be suppressed by substituting the GaN buffer layer with AlGaN of comparable or higher Al content, thereby reducing the interfacial electron accumulation. The limited penetration depth of the excitation laser suggests that reabsorption of MQW emission may also contribute to the yellow luminescence, something our model of the MQW behavior does not consider. Further, if the injected carrier density is increased to a level on par with the polarization-induced carrier density, the radiative recombination in the MQW will dominate, as shown by the dotted curve in Fig. 3 (c).

On the other hand, the excessively high dislocation density is another factor that limits efficient radiative recombination. The blue curve in Fig. 4 (c), which simulates integrated QW radiative (non-radiative) recombination rate $rrad′(rSRH′)$ as a function of $τSRH$ of the epitaxial material, shows that $rrad′(rSRH′)$ increases (decreases) rapidly until $τSRH$ reaches $10ns$ and becomes constant for $τSRH>100ns$⁠. Given the $τSRH$ of $10ns$ and the electron mobility $μ$ of $200cm2V−1s−1$⁠, the diffusion length of electrons is 2.3 $μm$⁠, which translates to a non-radiative recombination center density of $∼107cm−2$⁠. As shown in the AFM measurement, both the epitaxial single layers and the MQW possess a dislocation density on the order of $108∼109cm−2$ which falls into the dislocation-limited regime, with a quantum efficiency $η=rrad′/(rrad′+rSRH′)$ of $0.1∼0.2$ (Fig. 4 (c)). More than a tenfold $rrad′$ enhancement and $η∼0.8$ can be achieved if the dislocation density can be reduced to $107cm−2$⁠. This would require further optimization of the epitaxial material, such as growth on a buffer layer with a low dislocation density and small lattice mismatch.

In conclusion, a high-Al content (y > 0.4) MQW with quaternary In_xAl_yGa_(1-x-y)N active layers was synthesized by PAMBE. UVB RTPL was enhanced by an order of magnitude compared to an In_xAl_yGa_(1-x-y)N thin film of comparable composition and thickness. Simulations suggest that the UVB emission efficiency, limited by dislocation-related non-radiative recombination in the MQW and at the buffer interface, may be improved by reducing the dislocation density, optimizing the QW width and depth, and replacing the GaN buffer layer with unstrained AlGaN of comparable Al content.