Introduction

Nonradiative surface recombination (SR) at the boundaries of a semiconductor device can be a major factor limiting the efficiency in light-emitting and laser diodes (LEDs and LDs), photovoltaic cells, and photodetectors. The detrimental effects of SR are usually tackled by adding a surface-passivating layer or, in case of a heterostructure, an interface layer. Carrier-confining interfaces have enabled a substantial increase in efficiency in arsenide- and phosphide-based photonic devices. In bare-surface GaAs the surface recombination velocity (S) can reach 107 cm/s1, but a suitably engineered interface can reduce the S value down to 18 cm/s, as reported for AlGaAs LD heterostructures2. While the focus on the interface quality in arsenide- and phosphide-based photonic devices has led to tangible progress, the nonradiative recombination at the interface layer in nitride devices have been often overlooked, mainly due to the secondary role of SR in the high-efficiency InGaN LEDs. The SR velocity is generally lower in nitrides3,4,5,6,7 than in other III-V materials, and the diffusion length in the active InGaN region is relatively short8 due to both carrier-localizing composition fluctuations, and short carrier lifetime at high operating carrier density9. However, a recent theoretical study6 shows a noticeable impact of SR towards the efficiency of a nitride-based LED structure, and the negative impact of SR is predicted to be even higher for AlGaN-based UV LEDs or nanostructured photonic devices.

In this work, we observe and investigate an unconventional trend in AlGaN/GaN interfaces: the adverse influence of the AlGaN barrier on the interface recombination in the underlying GaN buffer. The effects of decreasing AlGaN barrier quality on carrier diffusivity and recombination were explored by time-resolved photoluminescence spectroscopy and light-induced transient grating (LITG) techniques. The density-dependent values of carrier lifetime were obtained in different depths of the GaN buffer using different excitation wavelengths. The terms of the effective carrier lifetime (interface recombination velocity, Shockley-Read-Hall and radiative recombination rates) are evaluated.

Methods

Samples

Twelve c-plane AlGaN/GaN/sapphire heterostructures were grown by the metalorganic chemical vapour deposition (MOCVD) technique. The 130–300 nm thick AlxGa1-xN barriers were grown at 1090 °C and at constant tri-methyl-aluminium (TMAl) flow rate, while the Al content was controlled by changing the tri-methyl-gallium (TMGa) flow rate. Three sample sets with different barrier Al content were obtained: x = (0.13, 0.2, or 0.34, estimated from the XRD spectra), with ≤10% relative variation within a sample set. Variance of the ammonia flow rates (500–5000 sl/min) during the barrier growth resulted in different barrier structural quality between the samples. All barriers were grown on identical GaN/Sapphire templates with 4 µm thick GaN buffers. XRD reciprocal space maps show all barriers to be strained (see XRD data in the Supplementary Figs I and II). The XRD-estimated dislocation density (total of edge and screw DD) shows little variation in both barriers and buffers between samples and is equal to ~ 4 × 109 cm−2.

Measurement methods

Light-induced transient gratings (LITG) and time-resolved photoluminescence (TRPL) techniques were employed to investigate carrier dynamics in AlGaN barriers and GaN buffers at room temperature. The diffusion coefficient in GaN buffer was measured by LITG, whereas the carrier lifetime in GaN buffer and the AlGaN barrier was obtained from the TRPL and LITG transients. Different excitation configurations of the LITG and TRPL experiments are demonstrated in Fig. 1, where the excitation beams (λ = 266 nm or 355 nm, indicated) represent pulses (25 ps, 10 Hz) of an Ekspla Nd:YAG laser.

Figure 1
figure 1

Simplified scheme of the excitation configurations: (a) LITG in AlGaN barrier, 266 nm excitation; (b) LITG in GaN buffer, 355 nm excitation; (c) 266 nm excitation for TRPL in the barrier and the shallow interface (initial ~50 nm in GaN); (d) 355 nm excitation for TRPL in the deep interface (initial ~100 nm in GaN).

Full size image

In the LITG experiment, two coherent pump pulses intersected at the angle ϴ on the sample creating a transient pattern of photoexcited carriers spatially modulated with spacing Λ = λ/sinϴ10. The carrier density modulation causes a spatial modulation of the refractive index with the amplitude proportional to the photoexcited carrier density (N), thus recording a transient diffraction grating. The grating decay was monitored by a diffracted and delayed probe pulse in the transparency region of the sample at 1064 nm. The decay of the diffraction grating was caused by two effects: recombination of the excess carriers and their diffusion, which tends to homogenize the carrier spatial distribution. Evaluation of grating decay time (τG) at various grating periods Λ enabled the determination of excess carrier ambipolar diffusion coefficient (D) and recombination time (τ): 1/τG = 1/τ + 4π2D/Λ2 10. Carrier dynamics were selectively observed in the AlGaN barrier and the GaN buffer by employing excitation pulses of 266 nm or 355 nm wavelength, respectively (see Fig. 1a,b). The selective excitation of the GaN buffer is possible as the AlGaN barriers (x = 0.13–0.34; Eg = 3.7–4.2 eV) are transparent to the 355 nm photons (hν = 3.49 eV). Excess carrier density N (5 × 1018 cm−3–2 × 1020 cm−3) was controlled by varying the energy density of the excitation pulse (0.03–1 mJ/cm2).

The TRPL experiment was performed using a Hamamatsu streak camera with an Acton monochromator. Above-bandgap 266 nm excitation of both AlGaN and GaN and the available spectral resolution of the TRPL technique allowed to extract carrier lifetime in both barrier and buffer simultaneously (see Fig. 1c), while employing 355 nm excitation provided greater absorption depth in GaN (Fig. 1d). As a result, carrier recombination was investigated in different depths of the GaN buffer: in the initial ~50 nm and in the initial ~100 nm (from here on, ‘shallow’ and ‘deep’ interface, respectively). Carrier recombination was observed at excess carrier densities ranging from 1 × 1017 to 3 × 1019 cm−3.

The lifetime extraction from LITG- and TRPL-measured transients was carried out from the trailing transient end. The TRPL transients for buffer or barrier were obtained by integrating the full spectrum of band-to-band transitions in GaN or AlGaN, respectively. See Supplementary Material for typical LITG and TRPL transients (Supplementary Fig. III,VI) and a more detailed description of the lifetime extraction procedure.

Results and Discussion

Carrier lifetimes in AlGaN barriers (τAlGaN) with various Al content (x) are demonstrated in Fig. 2 as a dependence on the V/III molar ratio, which was calculated from the gas flow rates: NH3/(TMGa + TMAl). A non-monotonous trend of initial increase and eventual decrease is observed for all sample sets and may be attributed to the different origin of the growth-related defects, discussed in the following paragraphs.

Figure 2
figure 2

Carrier lifetime τAlGaN in AlxGa1-xN barriers with different Al content x as a function of V/III ratio; τAlGaN was assessed from LITG transients recorded with pulses of λ = 266 nm.

Full size image

The initially low τAlGaN at low V/III ratio (NH3 flow rate) is attributed to nitrogen deficiency-related defects: N vacancy-bound defects (such as VN-decorated dislocations), and impurities (such as carbon occupying the N-sites). The growth of carrier lifetime with V/III ratio is due to diminishing of nitrogen deficiency and the resulting decrease in defect concentration. Increase in the NH3 flow rate has been shown to cause a decrease in the VN concentration in GaN11,12, AlGaN13,14 and AlN15; a similar decrease is expected for the carbon concentration in GaN16.

At the highest V/III ratios the carrier lifetime decreases due to the excess-nitrogen induced point defects and degraded structural quality. Overflow of N can saturate the lattice sites and hinder the surface mobility of Ga17 and Al adatoms13,18, thus generating Ga19 and Al vacancies (VGa, VAl), or deteriorating surface morphology18,20,21. VGa and VAl-related deep gap defects can be attributed to VGa-O complexes22,23, dislocations decorated with VGa-O24,25,26 or VAl-O complexes27,28.

Further discussion of the carrier lifetime τAlGaN concerns the samples with different Al content grown at respectively optimized V/III ratios (see larger symbols in Fig. 2). Peak carrier lifetime decreases from 0.9 to 0.4 ns in the AlGaN barriers with increasing Al content (from 0.13 to 0.34; see Fig. 2); this result is consistent with the previously reported trend in AlGaN epilayers with wide Al composition range29. Evidently, the optimization of growth conditions does not compensate for the decrease in structural quality with growing Al content. This decrease is typically attributed to low Al adatom mobility30 resulting in extended and point defects, large and fine scale lateral phase separation31,32 or spontaneous phase modulation31. It can be speculated that Al content-induced decrease in τAlGaN is governed by native or impurity related point defects. Chichibu et al. observed33 the increasing concentration of VIII (or VIII complexes) with increasing Al content in AlGaN, and argued34 that VGa are the main carrier lifetime killers in GaN. Meanwhile, studies on impurities in AlGaN demonstrate the increasing oxygen13,16,31,35 and carbon16 concentrations with increasing Al content.

While a definitive interpretation of mechanisms governing τAlGaN at different growth conditions and barrier compositions is unavailable, it can be stated that the nonradiative carrier recombination in the studied samples is driven by defects of numerous origins.

The data on carrier recombination at different AlGaN/GaN interfaces is presented in Fig. 3 as carrier lifetime in GaN buffer (τGaN) dependences on carrier density; the lifetime curves for shallow and deep interface excitations are depicted as dashed and continuous lines, respectively. The rise-and-fall behaviour of carrier lifetime can be attributed to the competition between the nonradiative and radiative terms: the initial saturation of the nonradiative channel and the eventual emergence of the radiative term with increasing N. The nonradiative recombination may include contributions from Shockley-Read-Hall (SRH) and interface recombination channels. The SRH channel saturation and lifetime increase was previously observed in synthetic diamonds36, where defects acting as centres of nonradiative recombination were saturated at high photoexcitation levels.

Figure 3
figure 3

Carrier lifetime τGaN in GaN buffers with AlxGa1-xN barriers of various composition as a function of photoexcited carrier density; τGaN was assessed at the shallow and deep interface of the AlGaN/GaN heterostructures using excitation pulses of 266 nm and 355 nm wavelength, respectively.

Full size image

The dominance of either SRH or interface recombination channel can be determined by varying the excitation wavelength37,38. In this study, this is achieved by measuring carrier lifetime in shallow and deep interface.

In GaN buffer with Al0.13Ga0.87N barrier the lifetime τGaN increases by switching from deep to shallow interface (see blue lines in Fig. 3). This is an indication that the defect concentration is decreasing with increasing effective layer thickness (decreasing depth), and that the interface has a minor impact on carrier recombination. A similar carrier lifetime increase with increasing layer thickness occurs in bulk GaN due to a decrease in dislocation density39,40,41,42,43.

The opposite case is observed in the heterostructures with higher Al content (see gray and yellow lines for x = 0.2 and 0.34, respectively). Here, switching to the shallow interface results in a decrease in the carrier lifetime τGaN. Moreover, the ratio between the deep and the shallow interface lifetimes τGaN increases with the barrier Al content: from 1.2 (x = 0.2) to 1.4 (x = 0.34) at the lowest carrier density. These trends indicate the increasing role of the interface recombination channel.

The most important and non-intuitive feature of the carrier recombination is the considerable change in the τGaN(N) curves (seen in Fig. 3) with the increasing Al content in the AlGaN barrier. As the Al content increases from 13% to 34%, the lifetime in the GaN buffer decreases by a factor of ~2 (calculated as the average shallow interface τGaN). Additionally, the onset of the increase in τGaN(N) shifts to higher carrier densities. At first glance, the Al content in the AlGaN barrier has a direct effect on the carrier recombination in GaN buffer.

To determine if the variation in lifetime τGaN between the samples is related to the barrier composition only, deep interface τGaN was assessed additionally in samples with different defect origins (each sample corresponding to one point in Fig. 2). The relation between τGaN and τAlGaN is demonstrated in Fig. 4a, where τGaN scales linearly with τAlGaN for all samples. This correlation also holds for samples of same barrier composition and different structural quality (symbols of matching color in Fig. 4a). Apparently, the carrier lifetime decrease in GaN cannot be attributed to a particular barrier defects and is controlled by the overall AlGaN barrier quality.

Figure 4
figure 4

(a) Carrier lifetime τGaN at fixed carrier density N = 5.4 × 1019 cm−3; (b) diffusion coefficient DGaN at various N in GaN buffers as a function of carrier lifetime τAlGaN in AlGaN barriers; τGaN, DGaN were assessed from the LITG transients recorded with pulses of 355 nm wavelength, while τAlGaN – with pulses of 266 nm wavelength.

Full size image

The effects of barrier quality on carrier scattering in GaN were assessed by the diffusion coefficient DGaN measurements; the values of DGaN as a function of τAlGaN at various carrier densities N in GaN are demonstrated in Fig. 4b. No dependence of carrier diffusivity on barrier lifetime τAlGaN was observed throughout the explored N range (from 5.4 × 1018 to 1.8 × 1020 cm−3). The lack of such dependence in the studied GaN buffers with different AlGaN barriers indicate the absence of barrier effect on carrier scattering in GaN, contrary to the influence on carrier lifetime τGaN. It could be speculated that the carrier diffusivity in GaN is adversely affected by the AlGaN barrier, but any measurable change is hidden behind the dominating scattering mechanism in room temperature GaN (carrier - optical phonon).

The average DGaN value depends on carrier density and increases from 1.7 to 2.8 cm2/s with increasing N. Similar diffusion coefficients at comparable carrier densities are reported for thick high-quality HVPE-grown GaN44, where increase in carrier diffusivity with N was attributed to carrier degeneracy44. Degeneracy could also play a role in masking any adverse effects of AlGaN barrier on carrier scattering in GaN buffer.

The extracted DGaN values are required for a more detailed analysis of diffusion-limited interface recombination pathway at the AlGaN/GaN interface. The interface recombination velocity (Si) was assessed by fitting LITG transients with the following model:

$$\frac{\partial N(x,z,t)}{\partial t}=\nabla [{D}_{{\rm{GaN}}}\nabla N(x,z,t)]-\frac{N(x,z,t)}{{\tau }_{{\rm{GaN}}}^{{\rm{SRH}}}}-{B}_{{\rm{GaN}}}{N}^{2}(x,z,t)+G(x,z,t)$$
(1)
$${\frac{\partial N(x,z,t)}{\partial z}|}_{z=0}=\frac{{S}_{{\rm{i}}}N(x,0,t)}{{D}_{{\rm{GaN}}}}$$
$${\frac{\partial N(x,z,t)}{\partial z}|}_{z=d}=0$$

where \({\tau }_{{\rm{GaN}}}^{{\rm{SRH}}}\) – SRH lifetime in GaN, BGaN – radiative recombination coefficient in GaN, G(x, z, t) – spatially modulated carrier generation rate in GaN, and d – the GaN buffer thickness. The radiative recombination coefficient BGaN is considered as identical between the buffers. The BGaN value was calculated in the sample with the highest quality barrier and equals to 1.5 × 10−11 cm3/s, which is close to the previously reported values in thick ELO- and HVPE-grown GaN layers44,45.

Three LITG transients recorded at N = 5.4 × 1019 cm−3 in GaN buffers with various AlGaN barrier quality are depicted with fits (according to Eq. 1) in the inset of Fig. 5; the corresponding Si and \({\tau }_{{\rm{GaN}}}^{{\rm{SRH}}}\) values are illustrated as larger coloured circles in Fig. 5a,b. These values (and the fitting results for all samples and carrier densities) are shown as a function of the carrier lifetime in the AlGaN barrier τAlGaN. A strong Si dependence on barrier quality is observed: Si increases from 2 × 103 to 2 × 105 cm/s as τAlGaN decreases from 0.9 to 0.2 ns at N = 5.4 × 1019 cm−3 (circles in Fig. 5a). Compared to bare-surface GaN layers, where surface recombination velocity S typically reaches 1–5 × 104 cm/s3,4,5,46, buffers with high quality barriers (τAlGaN > 0.7 ns) display improved boundary properties. Such improvement is observed in a wide range of III/V and II/VI material interfaces, e.g., S ≈ 1 × 107 cm/s in bare-surface GaAs1 is greatly reduced with p+ GaAs layer (Si = 1.5 × 104 cm/s)37, Ga2O3 layer (Si ≈ 4.5 × 103 cm/s)1, GaAs/AlAs superlattice (Si = 40 cm/s)47, AlGaAs layer (Si = 18 cm/s)2, or GaInP layer (Si = 1.5 cm/s)48. However, the GaN buffers with moderate and low AlGaN barrier quality (τAlGaN < 0.7 ns) suffer a substantial decrease in interface quality (Fig. 5a). This trend has not been reported for nitride interfaces.

Figure 5
figure 5

Interface recombination velocity Si (a) and SRH carrier lifetime \({\tau }_{{\rm{GaN}}}^{{\rm{SRH}}}\) (b) at various carrier densities N in GaN buffers as a function of carrier lifetime τAlGaN in AlGaN barriers; the Si and \({\tau }_{{\rm{GaN}}}^{{\rm{SRH}}}\) values signified as larger colored circles are extracted from the LITG transients demonstrated in the inset; inset transients (points) were recorded with pulses of 355 nm at N = 5.4 × 1019 cm−3 in GaN buffers; inset lines denote fitting with the (1) model.

Full size image

Interface recombination velocity depends not only on the interface (barrier) quality but also on the carrier density. In the buffer with low quality barrier (τAlGaN ≈ 0.2 ns), Si gradually decreases from 3.5 × 105 to 7.5 × 104 cm/s with N increasing from 1.8 × 1019 to 1.8 × 1020 cm−3. However, increase in barrier quality weakens this dependence (see Fig. 5a). According to our calculations, the discussed carrier density range is sufficient for screening of the electric polarization field in the interface. Therefore, density-driven Si decrease was attributed to the saturation of interface recombination channel, mirroring the τGaN(N) dependences discussed in the section of Fig. 3.

Compared to the Si dependence on τAlGaN, the impact of barrier quality on the \({\tau }_{{\rm{GaN}}}^{{\rm{SRH}}}\) term is diminished: \({\tau }_{{\rm{GaN}}}^{{\rm{SRH}}}\) features a ~20% decrease as τAlGaN decreases from 0.9 to 0.2 ns at N = 5.4 × 1019 cm−3 (circles in Fig. 5b). Meanwhile, growing carrier density N causes a ~30% increase in the lifetime \({\tau }_{{\rm{GaN}}}^{{\rm{SRH}}}\) in the buffer with low quality barrier (τAlGaN ≈ 0.2 ns), but has no perceptible effect on \({\tau }_{{\rm{GaN}}}^{{\rm{SRH}}}\)(N) dependence in buffers with higher quality barriers.

The depth-wise expansion of the barrier influence is implied by the shifting balance between the volume- and interface-bound recombination channels (characterized with \({\tau }_{{\rm{GaN}}}^{{\rm{SRH}}}\) and Si, respectively). Moderate and high quality samples (τAlGaN > 0.4 ns) show little to no change in \({\tau }_{{\rm{GaN}}}^{{\rm{SRH}}}\), while demonstrating significant variation in Si (see respective dependences on τAlGaN in Fig. 5a,b), pointing to an interface-bound defective area. The expansion of the barrier influence for the low quality samples (τAlGaN < 0.4 ns) is observed as a decrease in \({\tau }_{{\rm{GaN}}}^{{\rm{SRH}}}\) with decreasing τAlGaN. A mechanism of quality deterioration for the GaN buffer volume may be attributed to the diffusion of interface-degrading impurities such as carbon and oxygen or point defects such as VN and VGa from the barrier (or the interface) to the buffer volume. For instance, multidirectional diffusion of VN and VGa in GaN is expected at MOCVD-characteristic temperatures49,50. Such speculation is further supported by the saturation of the Si increase with decreasing τAlGaN paired with the simultaneous and accelerating decrease in \({\tau }_{{\rm{GaN}}}^{{\rm{SRH}}}\) (see Fig. 5): this juxtaposition implies a possible defect redistribution from the interface to the volume.

To finalize the characterization of the interface recombination, we provide an insight into the relation between the recombination velocity Si and the internal quantum efficiency (IQE). For a model SQW heterostructure with one electrically active interface, the IQE can be roughly estimated using a simple rate equation3:

$${\rm{IQE}}=\frac{BN}{BN+1/{\tau }^{{\rm{SRH}}}+{S}_{{\rm{i}}}/d}$$
(2)

where d is the thickness of the active region. Following this model, the IQE in a d = 5 nm SQW (AlGaN/GaN/AlGaN) drops by two orders of magnitude with the decrease in interface quality (Si increase from 2 × 103 to 2 × 105 cm/s) at N = 1.8 × 1019 cm−3. In comparison, the adverse impact of surface recombination at the bare-surface areas of an InGaN based LED has been previously theoretically evaluated at 5–7% of the device wall-plug efficiency, and predicted to be even stronger in UV LEDs6. We show that in AlGaN/GaN heterostructures the drop in the IQE may be significantly higher due to carrier recombination at the heterointerfaces.

In conclusion, we study carrier dynamics in AlGaN/GaN heterointerfaces with numerous defect origins in AlGaN barriers, grown on identical GaN buffers. We demonstrate that the barrier alters and governs the carrier dynamics in the underlying buffer: carrier lifetime in the GaN buffer decreases with the AlGaN barrier quality. The majority of photo-generated carriers in the affected GaN buffers recombine close to the heterointerface, and the dominating decay mechanism is attributed to interface recombination. The interface recombination velocity increases from low-103 to mid-105 cm/s with decreasing barrier quality. The carrier recombination in the GaN buffer is not governed by a particular type of defect in the AlGaN barrier, and is controlled by the overall barrier quality. Contrary to the influence on carrier recombination rate, the AlGaN barrier has no observable effect on the carrier scattering and diffusivity in the GaN buffer. Finally, we show that the interface recombination, as a major carrier loss mechanism, may substantially limit IQE in nitride-based UV LEDs.