Structural Analysis in Low-V-defect Blue and Green GaInN/GaN Light Emitting Diodes Mingwei Zhu1,2, Theeradetch Detchprohm1,2, Yong Xia1,2, Wei Zhao1,2, Yufeng Li1,2, Jayantha Senawiratne1,2, Shi You1,2, Lianghong Liu3, Edward A. Preble3, Drew Hanser3, and Christian Wetzel1,2 1Future Chips Constellation, Rensselaer Polytechnic Institute, 110 Eighth Street, Troy, NY, 12180 2Department of Physics, Applied Physics, and Astronomy, Rensselaer Polytechnic Institute, 110 Eighth Street, Troy, NY, 12180 3Kyma Technologies, Inc., 8829 Midway West Road, Raleigh, NC, 27617

ABSTRACT

In this study, we characterized the structural defects in blue and green GaInN/GaN LEDs grown on c-plane bulk GaN and sapphire substrates. Low density large V-defects with diameters around 600 nm were found in the blue LEDs on bulk GaN. They were initiated by edge-type threading dislocations (TDs) around the homoepitaxial growth interface. On the other hand, a high density 7×109 cm-2 of smaller V-defects with sidewalls on }0111{ facets was observed in the active region of green LEDs on sapphire. Their diameter ranges from 150 to 200 nm. Misfit dislocations (MDs) generated in the quantum wells are found to initiate these V-defects. With optimizing the epitaxial growth conditions, the generation of MDs and their smaller V-defects was largely suppressed. As a result, the light output power improved by one order of magnitude. For green LEDs on bulk GaN, another unique type of defect was found in the active region: an inclined dislocation pair (IDP). In it a pair of dislocations propagate at a tilt angle of 18 to 23º from the [0001] growth direction towards >< 0011 . This defect seems to be a path of strain relief in the high indium composition quantum wells. INTRODUCTION

The group-III nitride material system has made rapid progress in recent years allowing the realization of light emitters in the UV and visible spectral range. High efficiency blue light emitting diodes (LEDs) have been demonstrated by using GaInN/GaN heterostructures. In order to achieve longer emission wavelength in the green and deep green spectral range, a higher InN fraction in the GaInN quantum wells (QWs) is essential. This, however, also results in a higher strain produced by the lattice mismatch between the GaInN well on the GaN template. Consequently, defects such as misfit dislocation (MD), stacking fault, V-defect, etc. could be created in the active region. For LEDs grown on sapphire, an additional high density of threading dislocations (TDs), generated at the sapphire/GaN interface, typically propagates into the active region. It is believed that by reducing the density of such defects and achieving higher lateral morphological homogeneity in the active region, the performance of green LEDs would be further improved.1 With recent advances of hydride vapor phase epitaxy (HVPE), bulk GaN with TD densities as low as 105 cm-2 provide an ideal template for homoepitaxial growth.2

Here we will use transmission electron microscopy (TEM) to analyze the structural properties of the blue and green LED epistructures grown on c-plane bulk GaN and sapphire substrates. We will discuss the relationship between defect density and LED optical performance.

Mater. Res. Soc. Symp. Proc. Vol. 1040 © 2008 Materials Research Society 1040-Q03-02

Figure 1 High resolution TEM picture of the MQWs in blue LED on bulk GaN. Rather homogenous QWs and barriers, as well as reasonably sharp QW edges could be seen.

We also will explore a new type of defect generated during the epitaxial growth on bulk GaN material. EXPERIMENT

The blue and green LED epi samples we study here were grown on c-plane HVPE bulk GaN and c-plane sapphire substrates by metalorganic vapor phase epitaxy (MOVPE). They are referred to as LED on GaN and LED on sapphire, respectively. The growth details can be found in references 3 and 4. Typical sample structures include 0.5 to 2 µm n-type GaN, five to ten periods of GaInN/GaN QWs, and 0.2 µm p-type GaN. The nominal thickness of QWs and quantum barriers (QBs) are 3 nm and 6 to 23 nm, respectively. The TEM sample of the blue LED was prepared from a fully fabricated die, while the samples of the green LED epi were prepared from unfabricated LED epi. The peak wavelength of blue and green LEDs are 420-440 and 535-565 nm, respectively in photoluminescence measurement. The InN-mole fractions were determined to be 9-13% for blue LEDs, and 16-25% for green LEDs by fitting X-ray diffraction data.

Cross-sectional TEM samples were prepared by mechanical polishing, followed by low angle ion milling in a Fischione Model 1010 Ion Mill. After perforation, the ion energy and milling angle were gradually decreased in order to minimize the surface damage and maximize the observable sample area. Bright field (BF) and high resolution TEM images were recorded on PHILIPS CM12 TEM and JEOL 2010 TEM operating at 200 kV. DISCUSSION Blue LEDs on bulk GaN

For blue LEDs grown on bulk GaN, no growth boundary is observed in TEM at the interface of the bulk GaN template and the epitaxial GaN layer. No defect is observed along a lateral length of the QWs for over 1.2 µm, indicating a low structural defect density and good epitaxial crystalline quality.5 Figure 1 shows a high resolution TEM image of the QWs taken along the ]0011[ zone axis. The dark layers are GaInN QWs. Rather homogenous QWs and barriers, as well as reasonably sharp QW edges on both sides can be observed.

However, some large V-defects initiated by edge-type TDs were observed in the sample, as shown in Figure 2(a). These V-defects have diameters from 600 to 650 nm at the top and depths range from 460 to 480 nm. The angle between the side-walls of the larger V-defect and the c-plane is about 55º when looking from the zone axis of ]0011[ . Their origin beneath the QW

region corresponds very well to the thickness of the overgrown n-layer. We therefore conclude that they should form at the beginning of the epitaxial growth. During the growth of the n-type GaN, there must have been no or only very little growth of GaN around the vicinity of the TDs. However, after the deposition of the GaInN QWs, the p-type material begins to fill in the V-shaped gap. Since the p-type GaN cannot fill in the gap entirely, pits with diameter from 300 to 500 nm remain at the surface and can be observed in scanning electron microscopy (SEM) images.3 Thus the density of such large V-defects was determined to be around 3×105 cm-2 by counting the pits in the SEM images. The density of TDs from the substrate should be on the same order of magnitude.

Besides TDs from substrate, more TDs originating in the QWs are observed as well. The total TD density is estimated to 108 cm-2 or less from cross-sectional TEM images. Despite the existence of large V-defects, the TD density has been reduced by more than one order of magnitude compared to blue LED epi on sapphire.5 This must be the reason for the 20-fold stronger photoluminescence intensity in this homoepitaxial material as described in reference 3.

Green LEDs on sapphires

Two groups of green LEDs on sapphire6 were studied here. Figure 2(b) and 2(c) show the bright field TEM images of LED in each group taken along the ]0211[ zone axis. The growth direction coincides with the [0001] c-axis as indicated. For LEDs in the first group, many V-defects initiated by MDs were found in the active region (figure 2(b)). Each V-defect was decorated by sets of thinner QWs on its sidewalls. The angle between sidewall and c-plane is about 59º, indicating V-defects should be on the }0111{ facets. As more V-defects were generated, some merge into each other and form a larger one. The density of MDs in the active region was determined to be 6-8×109 cm-2 by cross-sectional TEM. Most of the MDs were generated in the QWs. After the formation of V-defects, MDs still propagate within the V-defect and some of them extend to the free surface. MDs generated in the active region are determined

Figure 2 Bright field TEM images of (a) blue LED on GaN; large V-defect with diameter around 600 nm was shown; they are initiated by edge-type TDs from substrate; (b) green LED on sapphire with high density of small V-defects initiated by MDs; (c) green LED on sapphire without V-defect. White arrows indicate growth direction.

Figure 3 TEM image of green LED on bulk GaN taken along

]0211[ zone axis. IDs and IDPs were observed to be generated within the first three QWs. All of them are tilted certain angle off [0001], the growth direction.

to be edge-type by using the g�b=0 invisibility criterion.6 In addition, we observed that the material quality deteriorates as the number and size of V-defects increases, especially after the growth of the 7th QW. The widths of QWs and QBs on the c-plane also increases.6

Figure 2(c) shows the TEM image for another group of green LEDs on sapphire. The growth optimization was described in reference 3. No generation of V-defects was observed in the active region. The width of QWs and QBs remains uniform throughout the ten periods of the active region.

The peak emission wavelength for the samples with and without V-defects are 565 - 580 nm and 535 - 545 nm, respectively. For five-QW LEDs, the PL intensity of those without V-defects is about twice that of those with V-defects. When the number of QWs in the structure is increased to ten, the PL intensity doubles for V-defect free LEDs, but it drops by 75 % for LEDs containing V-defects. This can be explained as follows: V-defects disturb the periodicity of the QWs and degrade the active region's crystalline quality and homogeneity, thus leading to a drop of PL intensity. On the other hand, for LEDs without V-defects, the crystalline quality of the active region remains high from the first to the last QW. The high homogeneity of QWs throughout the structure leads to a highly uniform light emission. In this case, more QWs could be grown that ultimately enhance the radiative recombination and PL intensity. As a result, the PL intensity could improve by an order of magnitude for LED with ten QWs after suppressing those V-defects. Green LEDs on bulk GaN

For green LEDs on bulk GaN, we could not find any TDs propagating from the substrate. However, inclined dislocations (IDs) were ob-served in the active region, as shown in figure 3. They usually start within the first three QWs and propagate along directions with certain angle inclined off [0001]. Interestingly, most of the IDs come in pairs. For example, IDs 1a and 1b, 3a and 3b, 4a and 4b form three inclined dislocation pairs (IDPs). The twin IDs in one IDP start in close places and propagate to the different sides of [0001]. Some IDs extend to the free surface while some disappear in p-GaN. Despite of its high density of 6×109 cm-2, IDs and IDPs do not lead to the formation of V-defects in the active region.

In order to determine the propagating direction of IDs and IDPs in the QW region, we tilted the TEM sample and take images along different crystal orientations. Figures 4(a) and 4(b) show TEM images of two IDPs recorded along ]0211[ and ]0121[ zone axis, respectively. We measured the angle between the each ID and [0001] (direction perpendicular to QWs) in each image. Then we project each ID back and reconstruct their propagating directions in the three dimensional crystal, as illustrated in figure 4(c). The ID should lie in the plane which is

perpendicular to the projected plane and contains the projection of the ID. The interface of such two planes is the propagating direction of the ID.

The result is summarized in figure 4(d). Figure 4(d) is the top view of all the IDs and IDPs. For comparison, all the starting points of IDs were intentionally moved to the center of the hexagon and the twin IDs in one IDP share the same starting point. The dashed circle is the projection of a right circular cone with a cone angle of 40º. Therefore, any arrow ending on the dashed circle represents a direction 20º tilted off [0001]. We find that almost all of the IDs are inclined towards the >< 0110 direction. The inclination angles are between 18 and 23º. When seen from the top, the IDs of the same IDP, branch off at unique angels only: some show a 180º separation, while others show a 120º separation. Consequently, all the IDs and IDPs are highly geometrically orientated. Further TEM studies show that most of the IDs come out of contrast when the sample is tilted to g=[0002], (images not shown here), indicating they are edge-type.

IDs in AlGaN have been reported before.7,8 It was found that TDs inclined from [0001] to >< 0011 directions by about 15-25º could provide stress relief for compressively strained

AlGaN.7 Meanwhile, for high InN-fraction GaInN QWs, strain relaxation could be realized by edge-type MDs9-11, stacking faults10 or V-defects12. For the blue LED on bulk GaN, since the InN fraction in the QWs is lower, fewer or no relaxation is needed. That explains why only a few TDs or V-defects were generated. While for green LEDs on sapphire, some of the strain could be relaxed by the TDs propagating from the n-GaN underneath, or MDs and V-defects generated in the active region. For the case of green LEDs on GaN, due to the absence of TDs propagated from the n-GaN, the only possible way for relaxation is generating additional defects, like IDPs, in the active region. The observation, that no IDPs are generated after the fourth QW, could be explained as no other relaxation approach is needed for the following QWs once IDPs have been formed. Theoretical models predict that for QWs with nominal thickness of 3 nm, strain relaxation would only happen when the InN content is larger than 35%.13 However, this is not the case in our study.

Figure 4 TEM images of IDPs recorded along (a) ]0211[ zone axis; (b) ]0121[ zone axis. (c) Reconstructing the propagating direction of IDPs in the three dimensional crystal by projections of IDPs in different orientations. (d) The IDs and IDPs in top view. Almost all of them are inclined from [0001] to the >< 0011 direction by 18 to 23º.

In spite of the high density of IDPs in the active region, the PL intensity of this green LED on GaN is about 7 times higher than that of the LED on sapphire grown side by side under identical conditions. More studies are needed to understand the formation mechanism of IDPs and what role IDPs play in the optical performance. CONCLUSIONS

In conclusion, we observed large V-defects with diameters around 600 nm in the blue LEDs on bulk GaN at a density as low as 3 ×105 cm-2. They were initiated by edge-type TDs at the epitaxial boundary. Besides those from the substrate, TDs on the order of 108 cm-2 were generated in the active region. For green LEDs on sapphire, after growth optimization, the MDs and V-defects were suppressed, leading to a 10 times increase of the PL intensity. IDPs with high geometry orientation were found in the active region of green LED on GaN. They are inclined from ]0001[ to >< 0110 by about 20º. The defects seem to provide an alternative way to relax the strain in the active region. ACKNOWLEDGMENTS

This work was supported by a DOE/NETL Solid-State Lighting Contract of Directed Research under DE-FC26-06NT42860. REFERENCES 1. C. Wetzel, T. Salagaj, T. Detchprohm, P. Li, and J. S. Nelson; Appl. Phys. Lett. 85, 866 (2004). 2. O. H. Nam, M. D. Bremser, T. S. Zheleva, and R. F. Davis, Appl. Phys. Lett. 71, 2638 (1997). 3. T. Detchprohm, Y. Xia, Y. Xi, M. Zhu, W. Zhao, Y. Li, E. F. Schubert, L. Liu, D. Tsvetkov, D. Hanser, and C. Wetzel; J. Cryst. Growth, 298, 272 (2007). 4. T. Detchprohm, M. Zhu, W. Zhao, Y. Xia, Y. Li, J. Senawiratne, and C. Wetzel, Phys. Stat. Sol., submitted. 5. M. Zhu, Y. Xia, W. Zhao, Y. Li, J. Senawiratne, T. Detchprohm, and C. Wetzel, J. Electron. Mater., submitted. 6. M. Zhu, T. Detchprohm, S. You, Y. Wang, Y. Xia, W. Zhao, Y. Li, J. Senawiratne, Z. Zhang, and C. Wetzel, Phys. Stat. Sol., submitted. 7. P. Cantu, F. Wu, P. Waltereit, S. Keller, A. E. Romanov, S. P. DenBaars, and J. S. Speck, J. Appl. Phy., 97, 103435 (2005). 8. D. M. Follstaedt, S. R. Lee, P. P. Provencio, A. A. Allerman, J. A. Floro, and M. H. Crawford, Appl. Phys. Lett. 87, 121112 (2005). 9. P. M. F. J. Costa, R. Datta, M. J. Kappers, M. E. Vickers, C. J. Humphreys, D. M. Graham, P. Dawson, M. J. Godfrey, E. J. Thrush, and J. T. Mullins, Phys. Stat. Sol. (a) 203, 1729 (2006). 10. H. K. Cho, J. Y. Lee, C. S. Kim, and G. M. Yang, J. Appl. Phy., 91, 1166 (2001). 11. W. Lu, D. B. Li, C. R. Li, F. Shen, and Z. Zhang, J. Appl. Phy., 95, 4362 (2004). 12. T. L. Song, J. Appl. Phy., 98, 084906 (2005). 13. S. Pereira, M. R. Correia, E. Pereira, C. Trager-Cowan, F. Sweeney, K. P. O�Donell, E. Alves, N. Franco, and A. D. Sequeira, Appl. Phys. Lett. 81, 1207 (2002).