- Email: [email protected]
GaN structure vs. single InGaN layer for solar cell applications: A comparative study
Available online at www.sciencedirect.com
ScienceDirect Acta Materialia 61 (2013) 6587–6596 www.elsevier.com/locate/actamat
Multilayered InGaN/GaN structure vs. single InGaN layer for solar cell applications: A comparative study Y. El Gmili a,b,c, G. Orsal a,d, K. Pantzas e,f, T. Moudakir b, S. Sundaram b, G. Patriarche e, J. Hester b, A. Ahaitouf a,b,c, J.P. Salvestrini a,d,⇑, A. Ougazzaden b,f a
Universite´ de Lorraine, LMOPS, EA4423, 2 Rue E. Belin, 57070 Metz, France b CNRS, UMI 2958, 2 Rue Marconi, 57070 Metz, France c USMBA, FST Fes, LSSC, Fes, Morocco d Supe´lec, LMOPS, EA4423, 2 Rue E. Belin, 57070 Metz, France e CNRS LPN, UPR, Route de Nozay, F-91460 Marcoussis, France f GIT, UMI 2958, 2 Rue Marconi, 57070 Metz, France
Received 11 May 2013; received in revised form 19 July 2013; accepted 20 July 2013 Available online 17 August 2013
Abstract We report a comparison of the morphological, structural and optical properties of both InGaN single-layer and multilayered structures, the latter consisting of periodic thin GaN interlayers inserted during InGaN growth. It is shown that such a structure suppresses the In concentration fluctuations and corresponding different states of strain relaxation with depth, both detrimental to solar cell applications. Measurements performed by X-ray diffraction, cathodoluminescence and photoluminescence demonstrate that this multilayer growth is a promising approach to increase both the InGaN layer total thickness and In content in InGaN epilayers. As an example, single-phase 120 nm thick InGaN with 14.3% In content is obtained and found to possess high structural quality. Ó 2013 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved. Keywords: Semiconductor materials; InGaN; Multilayer design; Deposition and fabrication; Thin films
1. Introduction InxGa1xN material, with its direct bandgap covering continuously the full visible range of the solar spectrum by varying only the In content, enables the production of high-efficiency solar cells based solely on nitride material [1]. To obtain this high efficiency, InGaN layers with a large In content are required. For instance, a solar cell with only one InGaN junction containing 68% indium, can achieve 30% efficiency [2]. In addition, despite a high absorption coefficient, InGaN layer thicknesses >100 nm are required for the absorption of more than 90% of the incident above-bandgap light [3]. To date, the maximum ⇑ Corresponding author.
E-mail address: [email protected] (J.P. Salvestrini).
In incorporation in InGaN-based solar cells with thicknesses >100 nm grown by metalorganic vapor-phase epitaxy (MOVPE) ranges from 12% to 25% [4–9]. At such high In contents, due to the low crystalline quality of the InGaN layer, bandgap energy fluctuations and carrier recombination at localized states arise. The conversion efficiency of such devices is thus limited, leading to low values of the open-circuit voltage and short-circuit current density [8,9]. Thus, it is challenging to investigate new InGaN layer growth techniques that allow high In incorporation in thick layers while maintaining high crystalline quality. The main issue arises from the broad immiscibility region between the two binary compounds InN and GaN, which originates from their large relative lattice mismatch, and leads to phase separations and/or In fluctuations with depth [10]. This has been observed in both thick InGaN layers [11– 14] and in InGaN multi-quantum-well (MQW) structures
1359-6454/$36.00 Ó 2013 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.actamat.2013.07.041
6588
Y. El Gmili et al. / Acta Materialia 61 (2013) 6587–6596
[10,12,15]. Furthermore, literature data report that for layer thickness and In incorporation >100 nm and >10%, respectively, distinct sublayers exist in the different InGaN layers [16–19,11]. Close to the InGaN/GaN interface, the film is two-dimensional (2-D), homogeneous and fully strained on the GaN template substrate. As the growth proceeds, we have shown that spatial fluctuations of the composition are observed and an accumulation of In arises especially around threading dislocation terminations leading to 3-D In-rich domains embedded in the 2-D InGaN matrix [19,14]. Such local In-rich regions contribute to the transition from 2-D to 3-D growth observed in thick epilayers. Recently, we proposed growing InGaN epilayers by periodically inserting ultrathin GaN interlayers during the growth process to absorb the excess In and relieve the InGaN compressive strain [20]. This approach was shown to allow the growth of thick single-phase InGaN layers of high crystalline quality. Thus, the aim of this paper is to compare in detail the properties of both InGaN/GaN multilayered (M-sample) and InGaN single-layer (C-sample) structures, grown with the same total InGaN layer thickness and nominal In composition. In particular, optical emission properties are investigated using room-temperature depth-resolved cathodoluminescence (CL) and temperature-dependent photoluminescence (PL). It is found that the multilayered approach effectively suppresses the distinct sublayers and enhances the InGaN material quality as compared to the conventional growth of a thick single-layer structure. Using this approach, we succeed in obtaining single-phase 120 nm thick InGaN containing 14.3% In that exhibited high structural quality. 2. Experiment Two set of InGaN samples were grown at 800 °C on GaN templates in a T-shape MOVPE reactor [21]. Nitrogen (N2) was used as the carrier gas and trimethylgallium (TMGa), trimethylindium (TMIn) and ammonia (NH3)
were employed as precursor sources for Ga, In and elemental N, respectively. The reactor pressure was 100 Torr and the V/III ratio equal to 8000. The ratio of TMIn to the sum of TMIn and TMGa in the vapor phase (TMIn/III) was kept constant at 12.5% for the first set of samples, and then increased to 45% for the second set of samples. For each set of samples, a reference InGaN single-layer (C-sample) and a multilayered structure (M-sample) were grown for comparison. A schematic representation of both structures under study is shown in Fig. 1. The M-samples were grown by periodically stopping the In precursor flow into the reactor. The two targeted structures consist of 5 1.5 nm (first set of samples) and 16 1.5 nm (second set of samples) thick GaN interlayers, inserted periodically between 6 21 nm (first set of samples) and 17 7 nm (second set of samples) thick InGaN layers, respectively, resulting in a total structure thickness of 133.5 nm (first set of samples) and 143 nm (second set of samples). The InGaN and GaN periodicity and layer thicknesses are based on simulations and previous experimental results on the relaxation of InGaN layers [19]. The layer thicknesses are deduced from in situ reflectometry and confirmed by (00.2) x 2h X-ray diffraction measurements by applying a fitting model to the Penddelosung fringes. Symmetric (00.4) x 2h scans in combination with X-ray reciprocal space maps (RSMs) of the asymmetric (10.5) planes are used to determine the lattice parameters (a, c), the degree of relaxation and the In content in the different samples. The input equations and parameters necessary to obtain these data are taken from Ref. [22]. The isocomposition lines are calculated according to the model proposed by Pereira et al. [23]. The surface morphology is observed by scanning electron microscopy (SEM) and atomic force microscopy (AFM). Microcompositional studies through energy dispersive X-ray spectroscopy (EDX) are carried out in an aberration-corrected JEOL 220FS microscope, operating at 200 kV with a probe current of 150 pA, and a probe size of 0.12 nm at the full width at half maximum (FWHM). Optical emission properties of InGaN structures are investigated by both CL
Fig. 1. Schematic representation of (a) C-sample, (b) M-sample of the first set of samples and (c) M-sample of the second set of sample structures. The InGaN total thickness and the nominal In composition are equal to 126 nm and 120 nm, and 10.2% and 14.3% for each set of samples, respectively.
Y. El Gmili et al. / Acta Materialia 61 (2013) 6587–6596
and PL measurements. Room-temperature CL investigations are performed in a digital scanning electron microscope. The CL emission is detected via a parabolic mirror collector and analyzed by a spectrometer with a focal length of 320 mm using a 1200 grooves mm1 grating and a spectral resolution of 0.06 nm. The signal is then recorded by a liquid N2-cooled Horiba Jobin Yvon Instruments Symphony 1024 256 CCD detector. This signal is treated mathematically to extract images. For this, each recorded spectrum is fitted using Gauss functions and the resulting fitting parameters (i.e. wavelength peak position, peak area, peak FWHM) are plotted vs. the position on the sample surface to produce 2-D maps. Temperaturedependent PL spectra are measured from 77 to 300 K using a 244 nm frequency-doubled Coherent Innova FreD argon laser as excitation source. For accurate determination of the composition from optical measurements (CL and PL) we have studied the band-edge emission energy dependence of the In content (varying in from 0% to 20% as deduced from XRD analysis) in different samples [24]. For strained and relaxed layers, the corresponding curves were fitted according to the equation:
InN GaN EInGaN NBE ¼ x E NBE þ ð1 xÞ ENBE b x ð1 xÞ
6589
ð1Þ
where b is the bowing parameter and ENBE the value of the optical emission energy as deduced from CL and PL data. This leads to bowing parameter values of b equal to 1.24 ± 0.40 and 2.7 ± 0.40 eV for strained and relaxed InGaN layers, respectively. 3. Results 3.1. First set of samples (10.5) RSMs and (00.2) x 2h scans as well as SEM surface images of both structures under study are presented in Fig. 2(c). One can clearly observe a broad InGaN diffraction spot for the C-sample. The diffraction maximum corresponds to a degree of relaxation of 13% with an average In content equal to 10.2% ± 0.1%. The extended profile observed along the isocompositional line indicates the presence of different strain relaxation states in the film. Furthermore, the scattering profile along Qz could indicate that the film is slightly compositionally inhomogeneous. For the M-sample, the InGaN diffraction spot is more intense and lies on the vertical line corresponding to pseu-
Fig. 2. (10.5) reciprocal space maps and (00.2) x 2h scans of the (a and b) control and (c and d) multilayered samples. The isocomposition lines connect the theoretical lines of fully strained (S) and fully relaxed (R) InGaN. The inset on (b and d) shows SEM images of the control and multilayered sample surfaces, respectively.
6590
Y. El Gmili et al. / Acta Materialia 61 (2013) 6587–6596
domorphic InGaN on GaN template substrate. The In content is calculated to be 7.3% ± 0.1% and the (00.2) x 2h scan reveals several orders of diffraction satellites which fit well with a 21 nm thick InGaN layer thickness. The insets of Fig. 2(c) show SEM measurements for both C-sample and M-sample surface. The C-sample exhibits rough 3-D surface growth mode with a large number of inclusions, while the M-sample shows 2-D surface growth mode with a small number of inclusions. Fig. 3(a) shows typical large area room-temperature depth-resolved CL spectra measured in the C-sample for electron beam energy varying from 3 to 11 keV. A double-peak structure, corresponding to the InGaN band-edge emission, is clearly shown irrespective of the beam energy. For clarity, the lower and higher peak wavelength positions are referred as InGaN#1 and InGaN#2 and correspond to values of k 410 nm and k 450 nm (as deduced from the data in Fig. 3(a) at a beam energy of 5 keV for which the CL intensity is maximum), respectively. For an electron beam energy ranging from 3 to 11 keV, the region of maximum energy loss moves progressively from the near surface to the near interface region, as shown in the upper scale of Fig. 3(b). This point of maximum excitation produces a maximum of luminescence intensity. Therefore, the related InGaN#2 and InGaN#1 CL emission peaks are attributed to the InGaN near-surface sublayer and the InGaN/GaN interface region, respectively. This is evidenced by the plot of the InGaN#1/ InGaN#2 intensity ratios vs. the electron beam energy (Fig. 3(b)), which increases with increasing electron beam energy up to 9 keV and then, due to the self-absorption effect [33], slightly decreases. Fig. 3(b) shows also the electron beam energy (and the corresponding depth of maximum electron energy loss in the material) dependences of both InGaN#1 and InGaN#2 band-edge emission energies, as deduced from the data of Fig. 3(a). Contrary to the InGaN#1 peak energy dependence, for which only a red shift is observed,
(a)
blue and red shifts of the InGaN#2 peak energy position with increasing electron beam energy are seen. Beyond a energy beam of 10 keV, both CL peak positions remain almost unchanged and the GaN template peak is observed (Fig. 3(a)). Fig. 4(a) reports the variation of the PL intensity in the C-sample as a function of the temperature in the 77–300 K range. The different spectra also exhibit a double-structure peak. In agreement with the CL data, the low and high intense wavelength PL peak positions are attributed to the underlying (InGaN#1) and top surface (InGaN#2) regions, and correspond to values of k 409 nm and k 439 nm (deduced from data of Fig. 4(a) at RT), respectively. The temperature dependence of both InGaN#1 and InGaN#2 band-edge emission energies, as reported in Fig. 4(b), shows a red, blue and red shift sequence. This typical “S-shaped” behavior has already been reported for InGaN MQWs [32] and InGaN single layers for In contents as low as 3% [30]. It was explained by spatially separated conduction band minima due to the In compositional fluctuation [26,27]. In fact, at low temperatures the photogenerated carrier lifetime is long enough so that the carriers can diffuse to the conduction band local minima. This leads to a blue shift of the PL peak position while such behavior is compensated by the red shift of the band gap as the temperature increases to room temperature. Based on the band tail model proposed by Eliseev et al. [28], the temperaturedependent emission energy above 70 K can be fitted according to: EðT Þ ¼ Eð0Þ
aT 2 r2 ; b þ T kBT
ð2Þ
which corresponds to the Varshni equation where localization effects are taken into account. E(0) corresponds to the energy gap value at 0 K, a and b are known as Varshni fitting parameters, kB is the Boltzmann constant and r corresponds to the degree of localization effect which is
(b)
Fig. 3. (a) Room-temperature CL spectra of the control sample measured in the range 3–11 deV; and (b) normalized InGaN#1/InGaN#2 intensity ratio as well as InGaN#1 and InGaN#2 band-edge emission energies variations vs. the electron beam energy and the depth of maximum energy loss (calculated using Monte Carlo simulation).
Y. El Gmili et al. / Acta Materialia 61 (2013) 6587–6596
(a)
6591
(b)
Fig. 4. (a) Temperature-dependent PL spectra of the control sample. The inset shows spectra obtained at a temperature close to RT. (b) InGaN#1 and InGaN#2 sublayer PL band-edge emission energies vs. temperature. The solid lines correspond to fitted curves based on the band tail model for each sublayer.
dependent on the carrier concentration, alloy composition and built-in electric field in the semiconductor material. In Fig. 4(b), the solid lines correspond to theoretical curves for which the parameters a,b and r are reported. The Varshni fitting parameter values are located between the corresponding values of the two binary compounds, GaN and InN [29], and are consistent with experimental data reported for thick InGaN layers [30] and MQW structures [31,32]. The values of r obtained for InGaN#1 and InGaN#2 are equal to 25 and 35 meV, respectively. Fig. 5(a) presents the temperature-dependent PL spectra of the M-sample. One can clearly observe two peaks for temperatures lower than 170 K. Contrary to the case of the C-sample, these two peaks are not related to distinct sublayers with different In concentrations. The high-wavelength PL peak (k 406 nm, RT data) can be attributed to the luminescence of the InGaN layers. The low-wavelength PL peak (k 378 nm, data at 150 K) could be attributed to the emission from the interlayers. This is confirmed by the
(a)
disappearance of this peak with increasing temperature. Indeed, since carriers photogenerated in the interlayers would be thermally injected into the InGaN layer at high temperature, the corresponding emission is expected to collapse. Moreover, since these interlayers are supposed to absorb the In excess, they should exhibit a luminescence peak shifted from the position of the GaN one. This indeed proves to be the case. Fig. 5(b) shows the temperature dependence of the InGaN near-band-edge emission energy. The solid curve corresponds to the fit based on Eq. (2). The value of the a parameter (1 meV) is close to that found in the case of the high crystalline quality InGaN#1 sublayer of the C-sample. The values of the b and r parameters are equal to 490 K and 17 meV, respectively, slightly different from those found in the case of the high crystalline quality InGaN#1 sublayer of the C-sample. As shown in Fig. 6(a), a large area CL study of the M-sample reveals at room temperature a single CL peak (k 400 nm, deduced from the data at an electron beam energy of
(b)
Fig. 5. (a) Temperature-dependent PL spectra of the multilayered sample. The inset shows peaks obtained at temperatures close to RT. (b) PL band-edge emission energy vs. temperature for the InGaN peak of the multilayered sample. The solid line is the fitted curve based on the band tail model.
6592
Y. El Gmili et al. / Acta Materialia 61 (2013) 6587–6596
(a)
(b)
Fig. 6. (a) Room-temperature CL spectra and (b) InGaN band-edge emission energy variation vs. the electron beam energy and the depth of maximum energy loss in the M-sample.
7 keV for which the CL intensity is maximum) which exhibits blue and red shifts when the electron beam energy varies from 3 to 11 keV (Fig. 6(b)). In a previous paper [14], we have demonstrated that the surface defect features observed on the sample surface (see inset of Fig. 2(d)) are 3-D In-rich InGaN domains embedded in a 2-D InGaN matrix. Hence, the blue shift of the luminescence peak can be explained by local change in the strain relaxation and In content with thickness. Both contributions can be responsible for the slight broadening of the diffraction spot corresponding to a pseudomorphic InGaN layer in Fig. 2. The red shift of the luminescence peak, found to be equal to 89 meV, can be attributed, as previously mentioned, to strong self-absorption effects occurring at the multiple InGaN/GaN interfaces [33]. The previous large-area CL measurements for both samples were recorded from a sample top surface size of 5 5 lm2. Obviously, the resulting CL spectra correspond to the average luminescence contribution of the entire analyzed surface. To locally study the spatial variation of the luminescence, we have recorded, in both structures, SEM images and corresponding CL hyperspectral mappings at different electron beam energies. Fig. 7 shows an example of mappings recorded at an electron beam energy of 7 keV. As expected, the C-sample exhibits two different phases: one, corresponding to the InGaN#1 sublayer (Fig. 7) and characterized by a luminescence peak wavelength centered at k 410 nm, and another one (Fig. 7e) corresponding to the InGaN#2 near-surface sublayer, and characterized by a luminescence peak wavelength centered at k 447 nm. On the contrary, the CL mapping in the M-sample (Fig. 7d) reveals only one phase characterized by a luminescence peak wavelength centered at k 402 nm and with rather low inhomogeneity, except for some variations correlated to nanoscopic surface defect features observed in the corresponding SEM image (Fig. 7b).
3.2. Second set of samples All the previous results clearly show an improvement of the InGaN layer quality when using the multilayered approach technique. This drastically reduces the inhomogeneities of In incorporation and related changes in strain relaxation. Only local variations of the luminescence due to a few inclusions are observed. Note that such surface defect features were also reported for InGaN layer thicknesses as thin as in the case of MQWs [34]. The multilayered approach seems thus to be helpful for increasing the total InGaN layer thickness and more especially the In incorporation in the InGaN material. This is assessed with the second set of samples, for which the thickness of each InGaN layer has been reduced and the TMIn/III ratio increased from 12.5 to 45. The asymmetric (10.5) RSM shown in Fig. 8 indicates a pseudomorphic growth of InGaN on GaN with an average In composition equal to 14.3% ± 0.1% in the case of the multilayered sample (Fig. 8(a)). In the case of the control sample (for which the In incorporation is equal to 15.4% ± 0.1% with a degree of relaxation of 18%), Fig. 8 shows two diffraction spots typically associated with InGaN epilayers that have undergone a transition from 2-D to 3-D growth (Fig. 8(b)). The 3-D surface morphology presented in Fig. 9(a) and the presence of the two main components in the corresponding CL mapping (Fig. 9(b) and (c)) confirm the previous XRD results. As can be seen, the spatial variations of the luminescence wavelength position for the control sample shows two different phases: one (Fig. 9(b)) characterized by a luminescence peak wavelength centered at k 435 nm, and another one (Fig. 9(c)) characterized by a luminescence peak wavelength centered at k 490 nm. In comparison, the multilayered sample (Fig. 9(d)) shows a 2D- growth morphology, and the CL mapping (Fig. 9(e) and (f))
Y. El Gmili et al. / Acta Materialia 61 (2013) 6587–6596
6593
Fig. 7. (a and b) SEM and (c–e) CL hyperspectral mapping images in the C-sample and M-sample, respectively.
Fig. 8. (10.5) Reciprocal space maps of the (a) multilayered and (b) control samples. The isocomposition lines connect the theoretical lines of fully strained (S) and fully relaxed (R) InGaN.
reveals only one phase characterized by a luminescence peak wavelength centered at k 433 nm with rather low inhomogeneity apart from some variations (Fig. 9(f)) correlated to a few local inclusions observed in the corresponding SEM images (Fig. 9(d)).
4. Discussion 4.1. First set of samples XRD characterizations (Fig. 2) clearly show an improvement of the InGaN layer structural quality when
6594
Y. El Gmili et al. / Acta Materialia 61 (2013) 6587–6596
Fig. 9. (a and d) SEM and (b, c, e, and f) CL hyperspectral mapping images of the control and multilayered samples.
using the multilayered approach growth technique. The Csample layer is partially (13%) relaxed, whereas the M-sample layer is close to being fully strained at the substrate. In the case of the M-sample, the In excess that accumulates at the InGaN layer surface is periodically absorbed in the GaN sublayers (as we will see in the next paragraph) during the growth process, leading to an In content value, as deduced from the XRD data, in the M-sample (7.3 ± 0.1%) lower than that in the C-sample (10.2 ± 0.1%) for the same growth parameters. AFM measurements (not shown here) reveal a decrease in the RSM roughness, taken for a 5 5 lm2 surface image, for the C-sample (RMS = 14 nm) compared to the M-sample (RMS = 7 nm). Together with SEM characterizations, these AFM results show that the surface morphology is clearly improved in the case of the M-sample. CL study of the C-sample (Fig. 3) shows that the InGaN layer is composed of two distinct sublayers with different strain-relaxation states and In contents. For the near-surface sublayer (InGaN#2), the blue shift of the electron beam energy dependence of the InGaN#2 band-edge emission energy (see Fig. 3(b)) can be related to the strain var-
iation and/or decrease in In content over the depth. Such a shift is not observed for InGaN#1, confirming the high quality of the initial layer which is fully strained on the GaN substrate and homogeneous in composition. Since the spatial CL excitation profile broadens (see Fig. 3(a)) with electron beam energy, the red shift of the CL peak position for InGaN#1 and InGaN#2 can be explained by self-absorption effects [25]. The energy splitting of both InGaN#1 and InGaN#2 peaks, according to data of Fig. 3(b), is equal to 235 meV. This value is larger than the energy splitting of both relaxed and strained InGaN peak positions obtained by Pereira et al. [25] for an In composition of 10% and equal to 150 meV. Thus, the splitting observed in Fig. 3(b) cannot be explained only by the strain relaxation. Hence, In incorporation >10% is expected in the near-surface region (InGaN#2). This is confirmed by the value of the In concentration derived from the InGaN#2 peak position at an electron beam energy of 5 keV, which is found to be equal to 12.6% ± 1%. In the same way, the In incorporation is found, using the InGaN#1 peak position, to be equal to 9.6% ± 1% in the InGaN near-interface sublayer, in a good agreement with
Y. El Gmili et al. / Acta Materialia 61 (2013) 6587–6596
6595
the corresponding value deduced from XRD measurements. PL study of the C-sample (Fig. 4(b)) confirms the conclusions of the CL study. According to the data of Fig. 4(b), the corresponding InGaN#1 and InGaN#2 sublayer In incorporations are found to be equal to 9.4% ± 1% and 11.2% ± 1%, respectively. All the In concentrations in the different samples as deduced from the various measurement techniques are summarized in Table 1 for clarity. Furthermore, the analysis of the temperature dependence of the InGaN#1 and InGaN#2 peak energies (Fig. 4(b)) leads to a larger value of the Varshni parameter r in the case of the InGaN#2 sublayer, indicating a stronger localization effect in this sublayer. This is expected since this sublayer has a higher In content than that of the InGaN#1 sublayer. Accordingly, the values of the Varshni parameter r obtained for the InGaN#1 sublayer of the C-sample is larger than that of the M-sample, as obtained from the data of the temperature dependence of the InGaN#1 peak energies (Fig. 5(b)) in the M-sample. PL and CL studies (Figs. 5 and 6) of the M-sample confirm the results of the XRD analysis and demonstrate the improvement of the InGaN layer quality when using the multilayered growth technique. Indeed, at room temperature, both PL and CL measurements reveal one luminescence peak corresponding to the existence of only one InGaN layer with 8.8% ± 1% (PL data) and 7.6% ± 1% (CL data) In concentration, in agreement with the value of 7.3% ± 0.1% deduced from XRD measurements. The low-temperature PL measurements reveal also (luminescence peak wavelength at 378 nm) that the GaN interlayers actually contain 2.8% ± 0.4% In, in agreement with the EDX measurements (not presented here), revealing an In content of around 1–2%. These In-poor InGaN interlayers may be responsible for the extended profile of the diffraction peak along Qz between the GaN and InGaN diffraction spots in Fig. 2. Finally, the hyperspectral mapping images (Fig. 7) indicate that both the InGaN#1 and InGaN#2 sublayers of the C-sample are spatially inhomogeneous with average In incorporations of 9.6% ± 1% and 12.2% ± 1% for InGaN#1 and InGaN#2 sublayers, respectively. However, the M-sample exhibits only one homogeneous phase with an In incorporation of 8% ± 1%.
multilayered growth technique. The C-sample layer is partially (18%) relaxed, whereas the M-sample layer is fully strained at the substrate. As observed in the first set of samples, the In composition, as deduced from the XRD data, is larger in the C-sample (15.4%) than in the M-sample (14.3%). This is due to a larger strain relaxation in the Csample which allows a higher In incorporation in the lattice. Nevertheless, the difference in the In composition between the C-sample and the M-sample, is larger in the first set of samples than in the second set of samples. This was expected since the growth parameters for the second set of samples have been slightly changed to decrease the absorption of the In from the InGaN layer surfaces to the GaN sublayers. The spatial variations of the luminescence wavelength position for the control sample observed in the CL hyperspectral mapping images (Fig. 9) show two different phases, as expected, but are surprisingly much more homogeneous when compared to the C-sample of the previous set of samples: one (Fig. 9(b)), with an average In incorporation of 14.4% ± 2% and another one (Fig. 9(c)) with an average In incorporation of 17.6% ± 2%. Data of PL measurements not shown here are in good agreement with this result and lead to In contents of 13.5% ± 1.5% and 16.4% ± 1.5% for the InGaN#1 and InGaN#2 sublayers, respectively. In comparison, the multilayered sample shows a 2-D growth morphology, and the CL mapping (Fig. 9(e) and (f)) reveals only one phase with rather low inhomogeneity (In incorporation of 14.1% ± 1.5%,14.8% ± 1.5% according to PL measurements not shown here), except for some variations (Fig. 9(f)) correlated to the few local inclusions observed in the corresponding SEM images (Fig. 9(d)). In a previous work [14], we have shown that two kinds of inclusions could be observed at the surface of InGaN layers: one for which there is no obvious shift in the CL peak position, and thus no In content variation, when going from the outside to the middle of such defect; and the other, which are shown to be local 3-D top surface In-rich InGaN domains embedded in an homogeneous InGaN matrix. The latter corresponds to the defect features observed in Fig. 9(d) and (f), while the former corresponds to the defect features observed in Fig. 7(b) and (d).
4.2. Second set of samples
5. Conclusion
XRD characterizations (Fig. 8) show a further improvement of the InGaN layer structural quality when using the
In this paper, we have compared in detail the properties of both InGaN/GaN multilayered and InGaN single-layer
Table 1 Indium content (%) in the control sample (C) and multilayered sample (M), for the two sets of samples, as deduced from XRD, CL (data of hyperspectral mapping measurements) and PL measurement techniques. Errors are given in the text. Sample
Sublayer
First set of samples XRD
CL
PL
XRD
CL
PL*
9.6 12.2
9.4 11.2
15.4 –
14.4 17.6
13.5 16.4
8.8
14.3
14.1
14.8
C
InGaN#1 InGaN#2
10.2 –
M
InGaN#1
7.3
*
Second set of samples
8
Data corresponding to the result of measurements not shown in the paper.
6596
Y. El Gmili et al. / Acta Materialia 61 (2013) 6587–6596
structures, grown with the same total InGaN layer thickness and nominal In composition. In particular, optical emission properties were investigated using room-temperature depth-resolved CL and temperature-dependent PL. Our results demonstrate that the multilayered approach greatly enhances the structural, morphological and optical properties as compared to InGaN single layers. In particular, it suppresses the distinct sublayers with different strain relaxations and In composition fluctuations with depth, which are both detrimental for solar cell applications. The multilayer approach appears to be helpful for increasing both the thickness of InGaN layers and In incorporation. Our current best InGaN structure is 120 nm thick, single phase and with 14.3% ± 0.1% In content. To our knowledge, this structure corresponds to the best performance achieved using MOVPE growth. We expect this approach also to work for higher In compositions simply by adjusting the interlayer thickness and/or frequency. Acknowledgment The authors wish to thank Jeremy Streque for his contribution to the PL measurements. This study has been funded by both ANR Habisol 2009 NEWPVONGLASS (Grant No. ANR-08-HABISOL-020-1), and NOVAGAINS (Grant No. ANR-12-PRGE-0014-02) projects. References [1] Wu J, Walukiewicz W, Yu KM, Shan W, Ager JW, Haller EE, et al. J Appl Phys 2003;94:6477. [2] Bremner SP, Levy MY, Honsberg CB. Appl Phys Lett 2008;92:171110. [3] Matioli E, Neufeld C, Iza M, Cruz SC, Al-Heji AA, Chen X, et al. Appl Phys Lett 2011;98:021102. [4] Jani O, Ferguson I, Honsberg C, Kurtz S. Appl Phys Lett 2007;91:132117. [5] Neufeld CJ, Toledo NG, Cruz SC, Iza M, DenBaars SP. Appl Phys Lett 2008;93:143502. [6] Jampana BR, Melton AG, Jamil M, Faleev NN, Opila RL, Ferguson IT, et al. IEEE Electron Device Lett 2010:31.
[7] Islam MdR, Kaysir MdR, Islam MdJ, Hashimoto A, Yamamoto A. J Mater Sci Technol 2013;29:128. [8] Cai X-m, Zeng S-w, Zhang B-p. Appl Phys Lett 2009;95:173504. [9] Tsai C-L, Liu G-S, Fan G-C, Lee Y-S. Solid-State Electron 2010;54:541. [10] Stringfellow GB. J Cryst Growth 2010;312:735. [11] Pereira S. Thin Solid Films 2006;515:164. [12] Ponce PA, Srinivasan S, Bell A, Geng L, Liu R, Stevens M, et al. Phys Status Solidi B 2003;240:273. [13] El-Masry NA, Piner EL, Liu SX, Bedair SM. Appl Phys Lett 1998;72:40. [14] El Gmili Y, Orsal G, Pantzas K, Ahaitouf A, Moudakir T, Gautier S, et al. Opt Mater Express 2013;3(8):1111. [15] Bertram F, Srinivasan S, Liu R, Geng L, Ponce FA, Riemann T, et al. Mater Sci Eng B 2002;93:19. [16] Liliental-Weber Z, Yu KM, Hawkridge M, Bedair S, Berman AE, Emara A, et al. Phys Status Solidi C 2009;6:433. [17] Wang H, Jiang DS, John U, Zhu JJ, Zhao DG, Liu ZS, et al. Thin Solid Films 2010;518:5028. [18] Sang L, Takeguchi M, Lee W, Nakayama Y, Lozac’h M, Sekiguchi T, et al. APEX 2010;3:111004. [19] Pantzas K, Patriarche G, Orsal G, Gautier S, Moudakir T, Abid M, et al. Phys Status Solidi A 2012;209:25. [20] Pantzas K, El Gmili Y, Dickerson J, Gautier S, Largeau L, Mauguin O, et al. J Cryst Growth 2013;370:57. [21] Gautier S, Sartel C, Ould-Saad S, Martin J, Sirenko A, Ougazzaden A. J Cryst Growth 2007;298:428. [22] Moram MA, Vickers ME. Rep Prog Phys 2009;72:036502. [23] Pereira S, Correira M, Pereira E, O’Donnell K, Alves E, Sequeira A, et al. Appl Phys Lett 2002;80:3913. [24] Orsal G, et al. Opt Mater Express, submitted for publication. [25] Pereira S, Correia MR, Pereira E, O’Donnell KP, Trager-Cowan C, Sweeney F, et al. Phys Rev B 2001;64:205311. [26] Schenk HPD, Leroux M, de Mierry P. J Appl Phys 2000;88:1525. [27] Naranjo FB, Fernanchez-Garcia S, Calle F, Calleja E, Trampert A, Ploog KH. Mater Sci Eng B 2002;93:131. [28] Eliseev PG, Perlin P, Lee J, Osinski M. Appl Phys Lett 1997;71:569. [29] Wu J. J Appl Phys 2009;106:011101. [30] Feng ZC, Liu W, Chua SJ, Yu JW, Yang CC, Yang TR, et al. Thin Solid Films 2006;498:118122. [31] Li J, Li S, Kang J. Appl Phys Lett 2008;92:101929. [32] Wang T, Saeki H, Bai J, Shirahama T, Lachab M, Sakai S, et al. Appl Phys Lett 2000;76:1737. [33] Knobloch K, Perlin P, Krueger J, Weber ER, Kisielowski C. MRS Internet J Nitride Semicond Res 1998;3:4. [34] Bruckbauer J, Edwards PR, Wang T, Martin RW. Appl Phys Lett 2011;98:41908.
ScienceDirect Acta Materialia 61 (2013) 6587–6596 www.elsevier.com/locate/actamat
Multilayered InGaN/GaN structure vs. single InGaN layer for solar cell applications: A comparative study Y. El Gmili a,b,c, G. Orsal a,d, K. Pantzas e,f, T. Moudakir b, S. Sundaram b, G. Patriarche e, J. Hester b, A. Ahaitouf a,b,c, J.P. Salvestrini a,d,⇑, A. Ougazzaden b,f a
Universite´ de Lorraine, LMOPS, EA4423, 2 Rue E. Belin, 57070 Metz, France b CNRS, UMI 2958, 2 Rue Marconi, 57070 Metz, France c USMBA, FST Fes, LSSC, Fes, Morocco d Supe´lec, LMOPS, EA4423, 2 Rue E. Belin, 57070 Metz, France e CNRS LPN, UPR, Route de Nozay, F-91460 Marcoussis, France f GIT, UMI 2958, 2 Rue Marconi, 57070 Metz, France
Received 11 May 2013; received in revised form 19 July 2013; accepted 20 July 2013 Available online 17 August 2013
Abstract We report a comparison of the morphological, structural and optical properties of both InGaN single-layer and multilayered structures, the latter consisting of periodic thin GaN interlayers inserted during InGaN growth. It is shown that such a structure suppresses the In concentration fluctuations and corresponding different states of strain relaxation with depth, both detrimental to solar cell applications. Measurements performed by X-ray diffraction, cathodoluminescence and photoluminescence demonstrate that this multilayer growth is a promising approach to increase both the InGaN layer total thickness and In content in InGaN epilayers. As an example, single-phase 120 nm thick InGaN with 14.3% In content is obtained and found to possess high structural quality. Ó 2013 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved. Keywords: Semiconductor materials; InGaN; Multilayer design; Deposition and fabrication; Thin films
1. Introduction InxGa1xN material, with its direct bandgap covering continuously the full visible range of the solar spectrum by varying only the In content, enables the production of high-efficiency solar cells based solely on nitride material [1]. To obtain this high efficiency, InGaN layers with a large In content are required. For instance, a solar cell with only one InGaN junction containing 68% indium, can achieve 30% efficiency [2]. In addition, despite a high absorption coefficient, InGaN layer thicknesses >100 nm are required for the absorption of more than 90% of the incident above-bandgap light [3]. To date, the maximum ⇑ Corresponding author.
E-mail address: [email protected] (J.P. Salvestrini).
In incorporation in InGaN-based solar cells with thicknesses >100 nm grown by metalorganic vapor-phase epitaxy (MOVPE) ranges from 12% to 25% [4–9]. At such high In contents, due to the low crystalline quality of the InGaN layer, bandgap energy fluctuations and carrier recombination at localized states arise. The conversion efficiency of such devices is thus limited, leading to low values of the open-circuit voltage and short-circuit current density [8,9]. Thus, it is challenging to investigate new InGaN layer growth techniques that allow high In incorporation in thick layers while maintaining high crystalline quality. The main issue arises from the broad immiscibility region between the two binary compounds InN and GaN, which originates from their large relative lattice mismatch, and leads to phase separations and/or In fluctuations with depth [10]. This has been observed in both thick InGaN layers [11– 14] and in InGaN multi-quantum-well (MQW) structures
1359-6454/$36.00 Ó 2013 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.actamat.2013.07.041
6588
Y. El Gmili et al. / Acta Materialia 61 (2013) 6587–6596
[10,12,15]. Furthermore, literature data report that for layer thickness and In incorporation >100 nm and >10%, respectively, distinct sublayers exist in the different InGaN layers [16–19,11]. Close to the InGaN/GaN interface, the film is two-dimensional (2-D), homogeneous and fully strained on the GaN template substrate. As the growth proceeds, we have shown that spatial fluctuations of the composition are observed and an accumulation of In arises especially around threading dislocation terminations leading to 3-D In-rich domains embedded in the 2-D InGaN matrix [19,14]. Such local In-rich regions contribute to the transition from 2-D to 3-D growth observed in thick epilayers. Recently, we proposed growing InGaN epilayers by periodically inserting ultrathin GaN interlayers during the growth process to absorb the excess In and relieve the InGaN compressive strain [20]. This approach was shown to allow the growth of thick single-phase InGaN layers of high crystalline quality. Thus, the aim of this paper is to compare in detail the properties of both InGaN/GaN multilayered (M-sample) and InGaN single-layer (C-sample) structures, grown with the same total InGaN layer thickness and nominal In composition. In particular, optical emission properties are investigated using room-temperature depth-resolved cathodoluminescence (CL) and temperature-dependent photoluminescence (PL). It is found that the multilayered approach effectively suppresses the distinct sublayers and enhances the InGaN material quality as compared to the conventional growth of a thick single-layer structure. Using this approach, we succeed in obtaining single-phase 120 nm thick InGaN containing 14.3% In that exhibited high structural quality. 2. Experiment Two set of InGaN samples were grown at 800 °C on GaN templates in a T-shape MOVPE reactor [21]. Nitrogen (N2) was used as the carrier gas and trimethylgallium (TMGa), trimethylindium (TMIn) and ammonia (NH3)
were employed as precursor sources for Ga, In and elemental N, respectively. The reactor pressure was 100 Torr and the V/III ratio equal to 8000. The ratio of TMIn to the sum of TMIn and TMGa in the vapor phase (TMIn/III) was kept constant at 12.5% for the first set of samples, and then increased to 45% for the second set of samples. For each set of samples, a reference InGaN single-layer (C-sample) and a multilayered structure (M-sample) were grown for comparison. A schematic representation of both structures under study is shown in Fig. 1. The M-samples were grown by periodically stopping the In precursor flow into the reactor. The two targeted structures consist of 5 1.5 nm (first set of samples) and 16 1.5 nm (second set of samples) thick GaN interlayers, inserted periodically between 6 21 nm (first set of samples) and 17 7 nm (second set of samples) thick InGaN layers, respectively, resulting in a total structure thickness of 133.5 nm (first set of samples) and 143 nm (second set of samples). The InGaN and GaN periodicity and layer thicknesses are based on simulations and previous experimental results on the relaxation of InGaN layers [19]. The layer thicknesses are deduced from in situ reflectometry and confirmed by (00.2) x 2h X-ray diffraction measurements by applying a fitting model to the Penddelosung fringes. Symmetric (00.4) x 2h scans in combination with X-ray reciprocal space maps (RSMs) of the asymmetric (10.5) planes are used to determine the lattice parameters (a, c), the degree of relaxation and the In content in the different samples. The input equations and parameters necessary to obtain these data are taken from Ref. [22]. The isocomposition lines are calculated according to the model proposed by Pereira et al. [23]. The surface morphology is observed by scanning electron microscopy (SEM) and atomic force microscopy (AFM). Microcompositional studies through energy dispersive X-ray spectroscopy (EDX) are carried out in an aberration-corrected JEOL 220FS microscope, operating at 200 kV with a probe current of 150 pA, and a probe size of 0.12 nm at the full width at half maximum (FWHM). Optical emission properties of InGaN structures are investigated by both CL
Fig. 1. Schematic representation of (a) C-sample, (b) M-sample of the first set of samples and (c) M-sample of the second set of sample structures. The InGaN total thickness and the nominal In composition are equal to 126 nm and 120 nm, and 10.2% and 14.3% for each set of samples, respectively.
Y. El Gmili et al. / Acta Materialia 61 (2013) 6587–6596
and PL measurements. Room-temperature CL investigations are performed in a digital scanning electron microscope. The CL emission is detected via a parabolic mirror collector and analyzed by a spectrometer with a focal length of 320 mm using a 1200 grooves mm1 grating and a spectral resolution of 0.06 nm. The signal is then recorded by a liquid N2-cooled Horiba Jobin Yvon Instruments Symphony 1024 256 CCD detector. This signal is treated mathematically to extract images. For this, each recorded spectrum is fitted using Gauss functions and the resulting fitting parameters (i.e. wavelength peak position, peak area, peak FWHM) are plotted vs. the position on the sample surface to produce 2-D maps. Temperaturedependent PL spectra are measured from 77 to 300 K using a 244 nm frequency-doubled Coherent Innova FreD argon laser as excitation source. For accurate determination of the composition from optical measurements (CL and PL) we have studied the band-edge emission energy dependence of the In content (varying in from 0% to 20% as deduced from XRD analysis) in different samples [24]. For strained and relaxed layers, the corresponding curves were fitted according to the equation:
InN GaN EInGaN NBE ¼ x E NBE þ ð1 xÞ ENBE b x ð1 xÞ
6589
ð1Þ
where b is the bowing parameter and ENBE the value of the optical emission energy as deduced from CL and PL data. This leads to bowing parameter values of b equal to 1.24 ± 0.40 and 2.7 ± 0.40 eV for strained and relaxed InGaN layers, respectively. 3. Results 3.1. First set of samples (10.5) RSMs and (00.2) x 2h scans as well as SEM surface images of both structures under study are presented in Fig. 2(c). One can clearly observe a broad InGaN diffraction spot for the C-sample. The diffraction maximum corresponds to a degree of relaxation of 13% with an average In content equal to 10.2% ± 0.1%. The extended profile observed along the isocompositional line indicates the presence of different strain relaxation states in the film. Furthermore, the scattering profile along Qz could indicate that the film is slightly compositionally inhomogeneous. For the M-sample, the InGaN diffraction spot is more intense and lies on the vertical line corresponding to pseu-
Fig. 2. (10.5) reciprocal space maps and (00.2) x 2h scans of the (a and b) control and (c and d) multilayered samples. The isocomposition lines connect the theoretical lines of fully strained (S) and fully relaxed (R) InGaN. The inset on (b and d) shows SEM images of the control and multilayered sample surfaces, respectively.
6590
Y. El Gmili et al. / Acta Materialia 61 (2013) 6587–6596
domorphic InGaN on GaN template substrate. The In content is calculated to be 7.3% ± 0.1% and the (00.2) x 2h scan reveals several orders of diffraction satellites which fit well with a 21 nm thick InGaN layer thickness. The insets of Fig. 2(c) show SEM measurements for both C-sample and M-sample surface. The C-sample exhibits rough 3-D surface growth mode with a large number of inclusions, while the M-sample shows 2-D surface growth mode with a small number of inclusions. Fig. 3(a) shows typical large area room-temperature depth-resolved CL spectra measured in the C-sample for electron beam energy varying from 3 to 11 keV. A double-peak structure, corresponding to the InGaN band-edge emission, is clearly shown irrespective of the beam energy. For clarity, the lower and higher peak wavelength positions are referred as InGaN#1 and InGaN#2 and correspond to values of k 410 nm and k 450 nm (as deduced from the data in Fig. 3(a) at a beam energy of 5 keV for which the CL intensity is maximum), respectively. For an electron beam energy ranging from 3 to 11 keV, the region of maximum energy loss moves progressively from the near surface to the near interface region, as shown in the upper scale of Fig. 3(b). This point of maximum excitation produces a maximum of luminescence intensity. Therefore, the related InGaN#2 and InGaN#1 CL emission peaks are attributed to the InGaN near-surface sublayer and the InGaN/GaN interface region, respectively. This is evidenced by the plot of the InGaN#1/ InGaN#2 intensity ratios vs. the electron beam energy (Fig. 3(b)), which increases with increasing electron beam energy up to 9 keV and then, due to the self-absorption effect [33], slightly decreases. Fig. 3(b) shows also the electron beam energy (and the corresponding depth of maximum electron energy loss in the material) dependences of both InGaN#1 and InGaN#2 band-edge emission energies, as deduced from the data of Fig. 3(a). Contrary to the InGaN#1 peak energy dependence, for which only a red shift is observed,
(a)
blue and red shifts of the InGaN#2 peak energy position with increasing electron beam energy are seen. Beyond a energy beam of 10 keV, both CL peak positions remain almost unchanged and the GaN template peak is observed (Fig. 3(a)). Fig. 4(a) reports the variation of the PL intensity in the C-sample as a function of the temperature in the 77–300 K range. The different spectra also exhibit a double-structure peak. In agreement with the CL data, the low and high intense wavelength PL peak positions are attributed to the underlying (InGaN#1) and top surface (InGaN#2) regions, and correspond to values of k 409 nm and k 439 nm (deduced from data of Fig. 4(a) at RT), respectively. The temperature dependence of both InGaN#1 and InGaN#2 band-edge emission energies, as reported in Fig. 4(b), shows a red, blue and red shift sequence. This typical “S-shaped” behavior has already been reported for InGaN MQWs [32] and InGaN single layers for In contents as low as 3% [30]. It was explained by spatially separated conduction band minima due to the In compositional fluctuation [26,27]. In fact, at low temperatures the photogenerated carrier lifetime is long enough so that the carriers can diffuse to the conduction band local minima. This leads to a blue shift of the PL peak position while such behavior is compensated by the red shift of the band gap as the temperature increases to room temperature. Based on the band tail model proposed by Eliseev et al. [28], the temperaturedependent emission energy above 70 K can be fitted according to: EðT Þ ¼ Eð0Þ
aT 2 r2 ; b þ T kBT
ð2Þ
which corresponds to the Varshni equation where localization effects are taken into account. E(0) corresponds to the energy gap value at 0 K, a and b are known as Varshni fitting parameters, kB is the Boltzmann constant and r corresponds to the degree of localization effect which is
(b)
Fig. 3. (a) Room-temperature CL spectra of the control sample measured in the range 3–11 deV; and (b) normalized InGaN#1/InGaN#2 intensity ratio as well as InGaN#1 and InGaN#2 band-edge emission energies variations vs. the electron beam energy and the depth of maximum energy loss (calculated using Monte Carlo simulation).
Y. El Gmili et al. / Acta Materialia 61 (2013) 6587–6596
(a)
6591
(b)
Fig. 4. (a) Temperature-dependent PL spectra of the control sample. The inset shows spectra obtained at a temperature close to RT. (b) InGaN#1 and InGaN#2 sublayer PL band-edge emission energies vs. temperature. The solid lines correspond to fitted curves based on the band tail model for each sublayer.
dependent on the carrier concentration, alloy composition and built-in electric field in the semiconductor material. In Fig. 4(b), the solid lines correspond to theoretical curves for which the parameters a,b and r are reported. The Varshni fitting parameter values are located between the corresponding values of the two binary compounds, GaN and InN [29], and are consistent with experimental data reported for thick InGaN layers [30] and MQW structures [31,32]. The values of r obtained for InGaN#1 and InGaN#2 are equal to 25 and 35 meV, respectively. Fig. 5(a) presents the temperature-dependent PL spectra of the M-sample. One can clearly observe two peaks for temperatures lower than 170 K. Contrary to the case of the C-sample, these two peaks are not related to distinct sublayers with different In concentrations. The high-wavelength PL peak (k 406 nm, RT data) can be attributed to the luminescence of the InGaN layers. The low-wavelength PL peak (k 378 nm, data at 150 K) could be attributed to the emission from the interlayers. This is confirmed by the
(a)
disappearance of this peak with increasing temperature. Indeed, since carriers photogenerated in the interlayers would be thermally injected into the InGaN layer at high temperature, the corresponding emission is expected to collapse. Moreover, since these interlayers are supposed to absorb the In excess, they should exhibit a luminescence peak shifted from the position of the GaN one. This indeed proves to be the case. Fig. 5(b) shows the temperature dependence of the InGaN near-band-edge emission energy. The solid curve corresponds to the fit based on Eq. (2). The value of the a parameter (1 meV) is close to that found in the case of the high crystalline quality InGaN#1 sublayer of the C-sample. The values of the b and r parameters are equal to 490 K and 17 meV, respectively, slightly different from those found in the case of the high crystalline quality InGaN#1 sublayer of the C-sample. As shown in Fig. 6(a), a large area CL study of the M-sample reveals at room temperature a single CL peak (k 400 nm, deduced from the data at an electron beam energy of
(b)
Fig. 5. (a) Temperature-dependent PL spectra of the multilayered sample. The inset shows peaks obtained at temperatures close to RT. (b) PL band-edge emission energy vs. temperature for the InGaN peak of the multilayered sample. The solid line is the fitted curve based on the band tail model.
6592
Y. El Gmili et al. / Acta Materialia 61 (2013) 6587–6596
(a)
(b)
Fig. 6. (a) Room-temperature CL spectra and (b) InGaN band-edge emission energy variation vs. the electron beam energy and the depth of maximum energy loss in the M-sample.
7 keV for which the CL intensity is maximum) which exhibits blue and red shifts when the electron beam energy varies from 3 to 11 keV (Fig. 6(b)). In a previous paper [14], we have demonstrated that the surface defect features observed on the sample surface (see inset of Fig. 2(d)) are 3-D In-rich InGaN domains embedded in a 2-D InGaN matrix. Hence, the blue shift of the luminescence peak can be explained by local change in the strain relaxation and In content with thickness. Both contributions can be responsible for the slight broadening of the diffraction spot corresponding to a pseudomorphic InGaN layer in Fig. 2. The red shift of the luminescence peak, found to be equal to 89 meV, can be attributed, as previously mentioned, to strong self-absorption effects occurring at the multiple InGaN/GaN interfaces [33]. The previous large-area CL measurements for both samples were recorded from a sample top surface size of 5 5 lm2. Obviously, the resulting CL spectra correspond to the average luminescence contribution of the entire analyzed surface. To locally study the spatial variation of the luminescence, we have recorded, in both structures, SEM images and corresponding CL hyperspectral mappings at different electron beam energies. Fig. 7 shows an example of mappings recorded at an electron beam energy of 7 keV. As expected, the C-sample exhibits two different phases: one, corresponding to the InGaN#1 sublayer (Fig. 7) and characterized by a luminescence peak wavelength centered at k 410 nm, and another one (Fig. 7e) corresponding to the InGaN#2 near-surface sublayer, and characterized by a luminescence peak wavelength centered at k 447 nm. On the contrary, the CL mapping in the M-sample (Fig. 7d) reveals only one phase characterized by a luminescence peak wavelength centered at k 402 nm and with rather low inhomogeneity, except for some variations correlated to nanoscopic surface defect features observed in the corresponding SEM image (Fig. 7b).
3.2. Second set of samples All the previous results clearly show an improvement of the InGaN layer quality when using the multilayered approach technique. This drastically reduces the inhomogeneities of In incorporation and related changes in strain relaxation. Only local variations of the luminescence due to a few inclusions are observed. Note that such surface defect features were also reported for InGaN layer thicknesses as thin as in the case of MQWs [34]. The multilayered approach seems thus to be helpful for increasing the total InGaN layer thickness and more especially the In incorporation in the InGaN material. This is assessed with the second set of samples, for which the thickness of each InGaN layer has been reduced and the TMIn/III ratio increased from 12.5 to 45. The asymmetric (10.5) RSM shown in Fig. 8 indicates a pseudomorphic growth of InGaN on GaN with an average In composition equal to 14.3% ± 0.1% in the case of the multilayered sample (Fig. 8(a)). In the case of the control sample (for which the In incorporation is equal to 15.4% ± 0.1% with a degree of relaxation of 18%), Fig. 8 shows two diffraction spots typically associated with InGaN epilayers that have undergone a transition from 2-D to 3-D growth (Fig. 8(b)). The 3-D surface morphology presented in Fig. 9(a) and the presence of the two main components in the corresponding CL mapping (Fig. 9(b) and (c)) confirm the previous XRD results. As can be seen, the spatial variations of the luminescence wavelength position for the control sample shows two different phases: one (Fig. 9(b)) characterized by a luminescence peak wavelength centered at k 435 nm, and another one (Fig. 9(c)) characterized by a luminescence peak wavelength centered at k 490 nm. In comparison, the multilayered sample (Fig. 9(d)) shows a 2D- growth morphology, and the CL mapping (Fig. 9(e) and (f))
Y. El Gmili et al. / Acta Materialia 61 (2013) 6587–6596
6593
Fig. 7. (a and b) SEM and (c–e) CL hyperspectral mapping images in the C-sample and M-sample, respectively.
Fig. 8. (10.5) Reciprocal space maps of the (a) multilayered and (b) control samples. The isocomposition lines connect the theoretical lines of fully strained (S) and fully relaxed (R) InGaN.
reveals only one phase characterized by a luminescence peak wavelength centered at k 433 nm with rather low inhomogeneity apart from some variations (Fig. 9(f)) correlated to a few local inclusions observed in the corresponding SEM images (Fig. 9(d)).
4. Discussion 4.1. First set of samples XRD characterizations (Fig. 2) clearly show an improvement of the InGaN layer structural quality when
6594
Y. El Gmili et al. / Acta Materialia 61 (2013) 6587–6596
Fig. 9. (a and d) SEM and (b, c, e, and f) CL hyperspectral mapping images of the control and multilayered samples.
using the multilayered approach growth technique. The Csample layer is partially (13%) relaxed, whereas the M-sample layer is close to being fully strained at the substrate. In the case of the M-sample, the In excess that accumulates at the InGaN layer surface is periodically absorbed in the GaN sublayers (as we will see in the next paragraph) during the growth process, leading to an In content value, as deduced from the XRD data, in the M-sample (7.3 ± 0.1%) lower than that in the C-sample (10.2 ± 0.1%) for the same growth parameters. AFM measurements (not shown here) reveal a decrease in the RSM roughness, taken for a 5 5 lm2 surface image, for the C-sample (RMS = 14 nm) compared to the M-sample (RMS = 7 nm). Together with SEM characterizations, these AFM results show that the surface morphology is clearly improved in the case of the M-sample. CL study of the C-sample (Fig. 3) shows that the InGaN layer is composed of two distinct sublayers with different strain-relaxation states and In contents. For the near-surface sublayer (InGaN#2), the blue shift of the electron beam energy dependence of the InGaN#2 band-edge emission energy (see Fig. 3(b)) can be related to the strain var-
iation and/or decrease in In content over the depth. Such a shift is not observed for InGaN#1, confirming the high quality of the initial layer which is fully strained on the GaN substrate and homogeneous in composition. Since the spatial CL excitation profile broadens (see Fig. 3(a)) with electron beam energy, the red shift of the CL peak position for InGaN#1 and InGaN#2 can be explained by self-absorption effects [25]. The energy splitting of both InGaN#1 and InGaN#2 peaks, according to data of Fig. 3(b), is equal to 235 meV. This value is larger than the energy splitting of both relaxed and strained InGaN peak positions obtained by Pereira et al. [25] for an In composition of 10% and equal to 150 meV. Thus, the splitting observed in Fig. 3(b) cannot be explained only by the strain relaxation. Hence, In incorporation >10% is expected in the near-surface region (InGaN#2). This is confirmed by the value of the In concentration derived from the InGaN#2 peak position at an electron beam energy of 5 keV, which is found to be equal to 12.6% ± 1%. In the same way, the In incorporation is found, using the InGaN#1 peak position, to be equal to 9.6% ± 1% in the InGaN near-interface sublayer, in a good agreement with
Y. El Gmili et al. / Acta Materialia 61 (2013) 6587–6596
6595
the corresponding value deduced from XRD measurements. PL study of the C-sample (Fig. 4(b)) confirms the conclusions of the CL study. According to the data of Fig. 4(b), the corresponding InGaN#1 and InGaN#2 sublayer In incorporations are found to be equal to 9.4% ± 1% and 11.2% ± 1%, respectively. All the In concentrations in the different samples as deduced from the various measurement techniques are summarized in Table 1 for clarity. Furthermore, the analysis of the temperature dependence of the InGaN#1 and InGaN#2 peak energies (Fig. 4(b)) leads to a larger value of the Varshni parameter r in the case of the InGaN#2 sublayer, indicating a stronger localization effect in this sublayer. This is expected since this sublayer has a higher In content than that of the InGaN#1 sublayer. Accordingly, the values of the Varshni parameter r obtained for the InGaN#1 sublayer of the C-sample is larger than that of the M-sample, as obtained from the data of the temperature dependence of the InGaN#1 peak energies (Fig. 5(b)) in the M-sample. PL and CL studies (Figs. 5 and 6) of the M-sample confirm the results of the XRD analysis and demonstrate the improvement of the InGaN layer quality when using the multilayered growth technique. Indeed, at room temperature, both PL and CL measurements reveal one luminescence peak corresponding to the existence of only one InGaN layer with 8.8% ± 1% (PL data) and 7.6% ± 1% (CL data) In concentration, in agreement with the value of 7.3% ± 0.1% deduced from XRD measurements. The low-temperature PL measurements reveal also (luminescence peak wavelength at 378 nm) that the GaN interlayers actually contain 2.8% ± 0.4% In, in agreement with the EDX measurements (not presented here), revealing an In content of around 1–2%. These In-poor InGaN interlayers may be responsible for the extended profile of the diffraction peak along Qz between the GaN and InGaN diffraction spots in Fig. 2. Finally, the hyperspectral mapping images (Fig. 7) indicate that both the InGaN#1 and InGaN#2 sublayers of the C-sample are spatially inhomogeneous with average In incorporations of 9.6% ± 1% and 12.2% ± 1% for InGaN#1 and InGaN#2 sublayers, respectively. However, the M-sample exhibits only one homogeneous phase with an In incorporation of 8% ± 1%.
multilayered growth technique. The C-sample layer is partially (18%) relaxed, whereas the M-sample layer is fully strained at the substrate. As observed in the first set of samples, the In composition, as deduced from the XRD data, is larger in the C-sample (15.4%) than in the M-sample (14.3%). This is due to a larger strain relaxation in the Csample which allows a higher In incorporation in the lattice. Nevertheless, the difference in the In composition between the C-sample and the M-sample, is larger in the first set of samples than in the second set of samples. This was expected since the growth parameters for the second set of samples have been slightly changed to decrease the absorption of the In from the InGaN layer surfaces to the GaN sublayers. The spatial variations of the luminescence wavelength position for the control sample observed in the CL hyperspectral mapping images (Fig. 9) show two different phases, as expected, but are surprisingly much more homogeneous when compared to the C-sample of the previous set of samples: one (Fig. 9(b)), with an average In incorporation of 14.4% ± 2% and another one (Fig. 9(c)) with an average In incorporation of 17.6% ± 2%. Data of PL measurements not shown here are in good agreement with this result and lead to In contents of 13.5% ± 1.5% and 16.4% ± 1.5% for the InGaN#1 and InGaN#2 sublayers, respectively. In comparison, the multilayered sample shows a 2-D growth morphology, and the CL mapping (Fig. 9(e) and (f)) reveals only one phase with rather low inhomogeneity (In incorporation of 14.1% ± 1.5%,14.8% ± 1.5% according to PL measurements not shown here), except for some variations (Fig. 9(f)) correlated to the few local inclusions observed in the corresponding SEM images (Fig. 9(d)). In a previous work [14], we have shown that two kinds of inclusions could be observed at the surface of InGaN layers: one for which there is no obvious shift in the CL peak position, and thus no In content variation, when going from the outside to the middle of such defect; and the other, which are shown to be local 3-D top surface In-rich InGaN domains embedded in an homogeneous InGaN matrix. The latter corresponds to the defect features observed in Fig. 9(d) and (f), while the former corresponds to the defect features observed in Fig. 7(b) and (d).
4.2. Second set of samples
5. Conclusion
XRD characterizations (Fig. 8) show a further improvement of the InGaN layer structural quality when using the
In this paper, we have compared in detail the properties of both InGaN/GaN multilayered and InGaN single-layer
Table 1 Indium content (%) in the control sample (C) and multilayered sample (M), for the two sets of samples, as deduced from XRD, CL (data of hyperspectral mapping measurements) and PL measurement techniques. Errors are given in the text. Sample
Sublayer
First set of samples XRD
CL
PL
XRD
CL
PL*
9.6 12.2
9.4 11.2
15.4 –
14.4 17.6
13.5 16.4
8.8
14.3
14.1
14.8
C
InGaN#1 InGaN#2
10.2 –
M
InGaN#1
7.3
*
Second set of samples
8
Data corresponding to the result of measurements not shown in the paper.
6596
Y. El Gmili et al. / Acta Materialia 61 (2013) 6587–6596
structures, grown with the same total InGaN layer thickness and nominal In composition. In particular, optical emission properties were investigated using room-temperature depth-resolved CL and temperature-dependent PL. Our results demonstrate that the multilayered approach greatly enhances the structural, morphological and optical properties as compared to InGaN single layers. In particular, it suppresses the distinct sublayers with different strain relaxations and In composition fluctuations with depth, which are both detrimental for solar cell applications. The multilayer approach appears to be helpful for increasing both the thickness of InGaN layers and In incorporation. Our current best InGaN structure is 120 nm thick, single phase and with 14.3% ± 0.1% In content. To our knowledge, this structure corresponds to the best performance achieved using MOVPE growth. We expect this approach also to work for higher In compositions simply by adjusting the interlayer thickness and/or frequency. Acknowledgment The authors wish to thank Jeremy Streque for his contribution to the PL measurements. This study has been funded by both ANR Habisol 2009 NEWPVONGLASS (Grant No. ANR-08-HABISOL-020-1), and NOVAGAINS (Grant No. ANR-12-PRGE-0014-02) projects. References [1] Wu J, Walukiewicz W, Yu KM, Shan W, Ager JW, Haller EE, et al. J Appl Phys 2003;94:6477. [2] Bremner SP, Levy MY, Honsberg CB. Appl Phys Lett 2008;92:171110. [3] Matioli E, Neufeld C, Iza M, Cruz SC, Al-Heji AA, Chen X, et al. Appl Phys Lett 2011;98:021102. [4] Jani O, Ferguson I, Honsberg C, Kurtz S. Appl Phys Lett 2007;91:132117. [5] Neufeld CJ, Toledo NG, Cruz SC, Iza M, DenBaars SP. Appl Phys Lett 2008;93:143502. [6] Jampana BR, Melton AG, Jamil M, Faleev NN, Opila RL, Ferguson IT, et al. IEEE Electron Device Lett 2010:31.
[7] Islam MdR, Kaysir MdR, Islam MdJ, Hashimoto A, Yamamoto A. J Mater Sci Technol 2013;29:128. [8] Cai X-m, Zeng S-w, Zhang B-p. Appl Phys Lett 2009;95:173504. [9] Tsai C-L, Liu G-S, Fan G-C, Lee Y-S. Solid-State Electron 2010;54:541. [10] Stringfellow GB. J Cryst Growth 2010;312:735. [11] Pereira S. Thin Solid Films 2006;515:164. [12] Ponce PA, Srinivasan S, Bell A, Geng L, Liu R, Stevens M, et al. Phys Status Solidi B 2003;240:273. [13] El-Masry NA, Piner EL, Liu SX, Bedair SM. Appl Phys Lett 1998;72:40. [14] El Gmili Y, Orsal G, Pantzas K, Ahaitouf A, Moudakir T, Gautier S, et al. Opt Mater Express 2013;3(8):1111. [15] Bertram F, Srinivasan S, Liu R, Geng L, Ponce FA, Riemann T, et al. Mater Sci Eng B 2002;93:19. [16] Liliental-Weber Z, Yu KM, Hawkridge M, Bedair S, Berman AE, Emara A, et al. Phys Status Solidi C 2009;6:433. [17] Wang H, Jiang DS, John U, Zhu JJ, Zhao DG, Liu ZS, et al. Thin Solid Films 2010;518:5028. [18] Sang L, Takeguchi M, Lee W, Nakayama Y, Lozac’h M, Sekiguchi T, et al. APEX 2010;3:111004. [19] Pantzas K, Patriarche G, Orsal G, Gautier S, Moudakir T, Abid M, et al. Phys Status Solidi A 2012;209:25. [20] Pantzas K, El Gmili Y, Dickerson J, Gautier S, Largeau L, Mauguin O, et al. J Cryst Growth 2013;370:57. [21] Gautier S, Sartel C, Ould-Saad S, Martin J, Sirenko A, Ougazzaden A. J Cryst Growth 2007;298:428. [22] Moram MA, Vickers ME. Rep Prog Phys 2009;72:036502. [23] Pereira S, Correira M, Pereira E, O’Donnell K, Alves E, Sequeira A, et al. Appl Phys Lett 2002;80:3913. [24] Orsal G, et al. Opt Mater Express, submitted for publication. [25] Pereira S, Correia MR, Pereira E, O’Donnell KP, Trager-Cowan C, Sweeney F, et al. Phys Rev B 2001;64:205311. [26] Schenk HPD, Leroux M, de Mierry P. J Appl Phys 2000;88:1525. [27] Naranjo FB, Fernanchez-Garcia S, Calle F, Calleja E, Trampert A, Ploog KH. Mater Sci Eng B 2002;93:131. [28] Eliseev PG, Perlin P, Lee J, Osinski M. Appl Phys Lett 1997;71:569. [29] Wu J. J Appl Phys 2009;106:011101. [30] Feng ZC, Liu W, Chua SJ, Yu JW, Yang CC, Yang TR, et al. Thin Solid Films 2006;498:118122. [31] Li J, Li S, Kang J. Appl Phys Lett 2008;92:101929. [32] Wang T, Saeki H, Bai J, Shirahama T, Lachab M, Sakai S, et al. Appl Phys Lett 2000;76:1737. [33] Knobloch K, Perlin P, Krueger J, Weber ER, Kisielowski C. MRS Internet J Nitride Semicond Res 1998;3:4. [34] Bruckbauer J, Edwards PR, Wang T, Martin RW. Appl Phys Lett 2011;98:41908.