GaN surface as the source of non-radiative defects in InGaN/GaN quantum wells

Blue light-emitting diodes based on III-nitride semiconductors are nowadays widely used for solid-state lighting. They exhibit impressive figures of merit like an internal quantum efficiency close to 100%. This value is intriguing when considering the high dislocation density running throughout the InGaN/GaN quantum well (QW) active region. This striking feature is currently ascribed to carrier localization occurring in the InGaN alloy, which hinders their diffusion toward dislocations. However, it was recently reported that another source of defects, disconnected from dislocations, dramatically decreases the radiative efficiency of InGaN/GaN QWs. Those defects, present at the surface, are usually trapped in an InGaN underlayer (UL), which is grown before the QW active region. To get insight into the trapping mechanism, we varied the UL thickness, In content, and materials system (InGaN or InAlN) and studied the photoluminescence decay time at 300 K of a single InGaN/GaN QW. Our data demonstrate that defects are incorporated proportionally to the indium content in the UL. In addition, we show that those defects are created during the high-temperature growth of GaN and that they segregate at the surface even at low-temperature. Eventually, we propose an intrinsic origin for these surface defects.

III-nitride white light-emitting diodes (LEDs), which consist of a blue LED covered by a yellow phosphor, recently achieved a record luminous efficacy of 300 lm/W. 1 This spectacular performance must be credited to InGaN/ GaN quantum wells (QWs), which form the active region of blue LEDs. When Nakamura et al. reported the first highbrightness InGaN based LED, 2 the scientific community was astonished since the device grown on sapphire was full of dislocations (density > 10 9 cm À2 ), which act as nonradiative recombination centers (NRCs). 3 The defectinsensitivity of InGaN/GaN QWs was then ascribed to the intrinsic properties of the InGaN alloy: carriers are strongly localized due to large potential fluctuations induced by random alloy disorder 4 hindering their diffusion towards NRCs. 5 Besides dislocations, point defects such as impurities or vacancies are another common source of NRCs in semiconductor optoelectronic devices. In this vein, Armstrong et al. pointed out the detrimental impact of deep level defects on the efficiency of blue LEDs. 6 In particular, they showed that the density of NRCs in InGaN/GaN QWs decreases if a thick InGaN layer is deposited before the growth of the active region. In fact, such an InGaN underlayer (UL) or InGaN/GaN superlattice is commonly used in high efficiency commercial blue LEDs grown by metal-organic chemical vapor deposition (MOCVD). Its role is to trap surface defects (SDs) and avoid their further incorporation into the InGaN/ GaN active region. [6][7][8] It was also noticed that this effect holds when InGaN QWs are grown on free-standing GaN substrates, which indicates that dislocations are not at play. 7 In this study, we show that the defects buried in the InGaN UL are intrinsic to the GaN surface. They form at high temperature (>900 C) and segregate at the surface even for low growth temperatures (<800 C). In turn, they are incorporated in indium containing layers (InGaN or InAlN) where they create NRCs.
Samples were grown by MOCVD in either a vertical Aixtron 3 Â 2 in. CCS system (reactor 1) or a horizontal Aixtron 200/4 RF-S system (reactor 2). Substrates are c-plane sapphire (dislocation density of about 10 9 cm À2 in the GaN buffer) or c-plane freestanding (FS) GaN (dislocation density about $10 6 cm À2 in the GaN buffer). The common structure for all the samples is a GaN buffer, an InGaN (InAlN) UL, a GaN spacer, an InGaN single (S)QW, and a GaN cap of 50 nm (see supplementary material Sec. 1). The buffer is grown using trimethylgallium (TMGa) and H 2 as carrier gas at a temperature of 1000 C. For the growth of the InGaN (InAlN) alloy, the GaN spacer, and the GaN cap, the carrier gas is switched to N 2 . The InGaN UL is grown using trimethylindium (TMIn) and triethylgallium (TEGa) at a temperature of 770 C. The InAlN UL is grown using TMIn and trimethylaluminum (TMAl) at a temperature of 750 C. The SQW growth temperature is about 755 C, the nominal indium composition is 12%, and the thickness is 2.7 nm. Time-resolved photoluminescence (TRPL) was performed using a streak camera operating in synchroscan mode. We used the third harmonic of a mode-locked Ti:sapphire laser emitting at 280 nm as an excitation source. The pulse width is 2 ps and the repetition rate is 80.5 MHz. To extend the time window for long PL decay times, we used a pulse picker. The QW effective lifetime is measured over a spectral window for which the contribution of the yellow band, the InGaN UL, and the GaN band edge emissions is negligible. Time-integrated PL (TIPL) was performed with the same excitation source as TRPL, but the signal was integrated over several pulses (with an excitation power density of 0.26 W/cm 2 ). As shown in previous reports, 6-8 the role of the InGaN UL in an LED structure is to bury SDs and ensure a high efficiency InGaN/GaN active region. We first study the capture mechanism of these SDs by the InGaN UL. For that purpose, we grew a series of identical 2.7 nm thick In 0.12 Ga 0.88 N/GaN SQWs (series A) except for the thickness (d) of the In 0.03 Ga 0.97 N UL, which was varied from 0 to 214 nm. PL measurements were performed at room temperature with a low excitation power density (0.26 W/cm 2 ). The PL intensity increases by more than two orders of magnitude between a bare SQW sample (without UL) and the SQW with an UL of 214 nm [ Fig. 1(a)]. The intensity first increases with the thickness and then saturates. Interestingly, the PL intensity of the InGaN UL also gets higher.
Additionally, we carried out TRPL experiments at 300 K to measure the QW effective lifetime and get insight into non-radiative (NR) processes. In agreement with the PL intensity behavior, the effective lifetime exhibits a huge improvement when the UL thickness increases [ Fig. 1 From 170 ps when no UL is inserted, it goes up to about 20 ns for the thickest one indicating very low NR recombination. Notice that the nonexponential behavior at long delays has already been observed in InGaN/GaN QWs grown on bulk GaN substrates and was ascribed to the saturation of slow dynamics NRCs. 9 The dependence of the InGaN/GaN SQW optical properties with the thickness of the InGaN UL suggests the following mechanism: SDs are incorporated if In atoms are present at the surface and, once buried in InGaN layers, they create NRCs. This is in agreement with former results, which showed, for instance, a strong improvement of the optical properties of multiple (M) QW structures compared to SQW ones. [10][11][12] Indeed, in this case, the first QWs in the MQW structure play the role of the InGaN UL by trapping SDs.
To account for the evolution of the QW decay time as a function of InGaN UL thickness, we developed a model where the NR lifetime is a function of the SD concentration in the InGaN/GaN QWs. 13 First, we suppose that SDs segregate at the surface, with a surface concentration given by where h 0 is the initial surface density after the GaN growth at high temperature, N is the number of monolayers (MLs) deposited (N ¼ d/e ML with e ML the InGaN ML thickness), and R is the segregation coefficient. 14 Previous observations by Akasaka et al. 6 indicate that the incorporation of SDs in GaN is negligible, i.e., R is close to 1. In contrast, SDs are incorporated in InGaN layers, i.e., R decreases. Assuming that SDs are buried upon interaction with In atoms, we can write R ¼ R GaN À xp with R GaN being the segregation coefficient of SDs in GaN, x the In composition of the InGaN layer, and p the interaction efficiency between In atoms and SDs. Then, we express the effective lifetime of the QW as 1/s eff ¼ 1/s 0 þ 1/s NR,SD , with s 0 being the QW lifetime when all SDs are trapped into the InGaN UL and s NR,SD the NR lifetime induced by the incorporation of SDs. The latter is given by 1/s NR,SD / [SD] QW , with [SD] QW being the concentration of SDs in the QW. Finally, the QW effective lifetime can be expressed as Next, a set of InGaN/GaN SQW samples (series B) was prepared keeping constant the InGaN UL thickness (55 nm) but varying the In content (up to 16% to ensure pseudomorphic layer). It is worth noticing that we used another reactor (horizontal reactor) and the substrate was sapphire. The InGaN UL was deposited under slightly different growth conditions (T g ¼ 750 C with H 2 and N 2 used as carrier gases). The QW effective lifetimes for this sample series are displayed in Fig. 2(b). Equation (1) still with p ¼ 0.7 reproduces well the QW effective lifetime. Note that s 0 is 3.3 ns, which is smaller than the value measured on FS GaN substrate, which might be due to a higher dislocation density. A piece of information is that SDs are present whatever the reactor type: vertical design with a stainless steel showerhead for reactor 1 or horizontal design with a quartz liner (no stainless steel) for reactor 2.
We then checked whether SD trapping is specific to InGaN layers. To this end, we prepared a series of InGaN/ GaN SQWs grown on FS-GaN substrate with In 0.15 Al 0.85 N UL (series C). Notice that the In 0.15 Al 0.85 N alloy is under slight tensile strain for this composition. 15 The PL decay time of the InGaN/GaN SQWs dramatically increases with the InAlN UL thickness [ Fig. 2(a)]. Due to the high indium (1) (with x ¼ 0) leads to R GaN ¼ 0.9991. This high R value confirms the strong surface segregation of SDs in GaN even at a growth temperature as low as 770 C. Consequently, 2 lm of GaN grown at 770 C would be required to trap all SDs. This highlights the need for indium in the UL, as already pointed out by Akasaka et al. 6 Although the phenomenological model we propose accounts well for the dependence of the QW effective lifetime on the InGaN (InAlN) UL thickness or composition, it does not provide any insights into the nature of those SDs. In the literature, capture coefficients for impurities and complexes range between 10 À6 and 10 À8 cm 3 s À1 , [16][17][18][19] corresponding to impurity densities comprised between 10 15 and 10 17 cm À3 . Thus, we carried out secondary ion mass spectrometry (SIMS) and searched for Li, C, O, Na, Mg, P, K, Ca, Fe, and Mo, which might be detrimental to the radiative efficiency of InGaN QWs. 16,20-24 SIMS measurements did not allow us establishing any straightforward relationship between those impurities and the QW effective lifetime. However, one could argue that the SIMS detection limit was too high (see supplementary material Sec. 5) or that we missed the relevant impurity.
Based on previous results, SDs should have the following properties: (i) once incorporated into InGaN layers, they create efficient NRCs, (ii) they strongly interact with indium atoms, and (iii) their concentration should range between 10 15 and 10 17 cm À3 . Another intriguing observation is that SDs are present whatever the MOCVD growth reactor type. It is also a common feature that InGaN/GaN SQWs grown without InGaN UL exhibit a poor radiative efficiency. [10][11][12] We therefore postulate that SDs could not be impurities but rather intrinsic point defects, which form during the growth of GaN at high temperature.
To investigate the dependence of the density of SDs on growth temperature, we prepared a series (series E) of InGaN/GaN SQW samples with InGaN UL but with a GaN spacer deposited at various temperatures T GaN (Fig. 3). Increasing the growth temperature T GaN from 770 C to 1020 C leads to a rapid collapse of the QW PL intensity (Fig. 3). The effective lifetimes measured at 300 K are displayed on a semilogarithmic scale as a function of 1000/ T GaN (inset, Fig. 3). Using the previous model and assuming that the SD density h 0 scales as exp ðÀ E a k B T GaN Þ, we extracted an activation energy of 3.6 eV. Interestingly, this value is close to the activation energy reported in the literature for the decomposition of GaN. 25,26 The dependence of the SD density on growth temperature suggests intrinsic point defects whose creation rate might be enhanced in the GaN near surface region. Indeed, the defect formation energy may be significantly different from the bulk formation energy. 27 In addition, vacancies tend to diffuse toward the surface, as observed in other materials. 28 We could thus infer that the GaN surface and near surface region provide a reservoir of point defects, which segregate at the surface until they react with indium atoms. Then, once buried, they form NRCs.
Let us consider the most likely points defects, which are gallium and nitrogen vacancies. Gallium vacancies (V Ga ) in GaN epilayers grown by MOCVD have been discussed back to 1996 and proposed to be responsible for the yellow-band luminescence. 29 In GaN homoepitaxial layers, two energy levels associated with V Ga complexes have been determined by deep level spectroscopy: E c À 0.56 eV and E c À 0.23 eV, 30 with E c the conduction band minimum. These values are far from the energy level E c À 1.62 eV that was identified to be related to the creation of NRCs by SDs in InGaN/GaN QWs. 6 Recently, theoretical calculations have pointed out that V Ga are unlikely in bulk GaN, but instead complexes with oxygen and hydrogen are. 19,31 This would mean that oxygen and hydrogen could be detected by SIMS if their density was high enough, which is not the case in our samples (see supplementary material Sec. 5). In addition, Dreyer et al. InGaN layers and found that they play a marginal role below 15% In content. 19 Thus, V Ga and related complexes are not expected to dramatically impact the efficiency of the present InGaN/GaN QWs, which have an indium content of 12%.
It is commonly admitted that the concentration of nitrogen vacancies (V N ) in bulk GaN is low due to their large formation energy 27 but not negligible. 32 Furthermore, the formation energy of vacancies at the semiconductor surface may be significantly smaller. 33 It was also proposed that during the decomposition of GaN, nitrogen atoms dissociate from the surface and form N 2 molecules, leading to a nitrogen-poor growth front, which could be a source of V N . 26,34,35 Therefore, [V N ] could be high at the GaN surface. Notice that the predominance of anion vacancies near the surface was already pointed out in common III-V semiconductors. 33 Another hint comes from the so-called two-flow MOCVD reactor invented by Nakamura et al. 36 This reactor design allows improving the III-N material properties thanks to a higher NH 3 pressure close to the surface using an additional N 2 flow.
The present study points out the strong affinity of SDs with indium atoms. Thus, the mechanism could be the following: V N are created near the surface during high temperature growth of GaN. Once formed, they segregate at the surface even for low-temperature growth GaN due to their high formation energy in bulk GaN. 37 In the presence of indium atoms, the segregation process is quenched: V N are incorporated in the bulk due to a decrease in their formation energy in InGaN, 38,39 inducing in turn NRCs. Potential NRC candidates could be a complex formed by V N and In as theoretically predicted, 40 and experimentally reported, 41 or a divacancy complex (V N , V III ) as recently observed in GaN. 42 In conclusion, we studied the origin of surface defects in GaN, which dramatically reduce the efficiency of InGaN/GaN QWs. We hypothesize that the rather low stability of GaN at high temperature induces nitrogen vacancies at the surface. Those V N strongly segregate, even during low-temperature growth of GaN. In contrast, in indium containing layers, V N are incorporated where they create NRCs. We thus propose that the role of the UL is to capture preexisting V N at the GaN surface. Additionally, InAlN nearly lattice-matched to GaN proved to be an extremely efficient UL material through its high indium content while avoiding strain-related issues. Hence, our results shed new light on the physics and fabrication of high performance III-nitride based optoelectronic devices and potentially high power electronics.
See supplementary material for complementary information.