phys. stat. sol. (c) 0, No.

1, 537541 (2002)

Correlation between Emission Spectra and Defect Position in InGaN-Based Light Emitting Devices
F. Hitzel1), S. Lahmann, U. Rossow, and A. Hangleiter
Institute of Technical Physics, Technical University of Braunschweig, Mendelssohnstr. 2, 38106 Braunschweig, Germany (Received July 22, 2002; accepted October 1, 2002) PACS: 68.37.Uv; 68.55.Ln; 85.30.De GaN/InGaN/AlGaN devices grown on sapphire and SiC substrates have a high defect density due to a large lattice mismatch between substrate and active structure. Different models exist why defects are ineffective as recombination centres so that it is possible to manufacture high brightness InGaN-based LED devices despite the high defect density. Using a scanning near field optical microscope (SNOM) we were able to look at the emission spectra of GaInN/GaN multiple quantum wells in the vicinity of defects and to see the change of emission wavelength and light intensity around defect positions.

Introduction The large band gap of the GaN material system makes it the favourite material for manufacturing high brightness LED devices that emit in the UV to green wavelength range [1, 2]. Because currently the only affordable substrate materials are sapphire and SiC, nearly all available GaN devices have a very high defect density due to a large lattice mismatch between substrate and device structure. It is well known that defects in pure GaN or AlGaN behave as nonradiative recombination centres, but in InGaN quantum wells defects can become rather ineffective. The physical mechanism which stands behind this is not really known [3]. There are several possibilities that can be thought of: Carriers could be kept in randomly distributed potential minima within the quantum well, possibly caused by the inhomogeneity of In composition [4, 5]. A second possibility is that the potential landscape is correlated with the defect positions, so that potential minima are formed somewhere between defects and keep carriers away from them. Another possibility would be that the atomic structure of the defects self changes. Experimental To look at the nature of the potential landscape, we use a scanning near field optical microscope (SNOM), which is capable of taking a complete optical spectrum at each point of a scanned surface. For finding the positions of single defects, we utilize the existence of so called V-defects [6, 7], which decorate the defect positions so that for an atomic force microscope defects become visible on the surface. Because the SNOM is based on an AFM-like principle both a topography image showing the defect positions and the optical emission spectra are obtained at the same time. Our SNOM has a spatial resolution of about 100 nm, depending solely on the tip. We use a cooled CCD camera as detector that can take up to 20 optical spectra per second. This results in a short image scanning time and reduces drift effects of any kind. The SNOM can be evacuated and cooled down for low temperature measurements. As tips we use tube-etched UV fibres with a SiO2/Al two-layer coating. The SNOM is operated
1

) Corresponding author; e-mail: f.hitzel@tu-bs.de

# 2002 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

1610-1634/02/00012-0537 $ 17.50.50/0

538

F. Hitzel et al: Correlation between Emission Spectra and Defect Position

in illumination-collection mode, i.e. the sample is excited through the fibre and the PL signal is collected with the fibre, too. For the calculation of peak wavelength maps a low pass filter is applied to each optical spectrum. The actual peak position is found via linear regression at the peak maximum. Because each emission peak consists of 100 to 500 data points, using the right filter and regression procedures enables us to detect even wavelength movements that are beyond the digital resolution of our ccd camera. Our samples where InGaN-MQW-structures consisting of an undoped GaN nucleation layer, an undoped GaN buffer, 2, 4 or 10 InGaN quantum wells with GaN barriers, an AlGaN barrier and a GaN top layer. The quantum wells have an approximate width of 3 nm. The samples were grown with low pressure metalorganic vapour phase epitaxy (MOVPE) on sapphire substrate (horizontal AIX 200 RF reactor). Results Figure 1 shows a 4 4 micrometer scan at room temperature of sample BS 213 which has 10 quantum wells. The topography image shows a large number of black dots, which indicate the positions of defects. Counting these dots results in a defect 2 . Looking at the optical intensity image we see a quite density of about 1 109 cm inhomogeneous emission amplitude. The difference between bright and dark regions are about 40% of the maximum emission amplitude. We can see bright regions that have larger sizes than the defect distances as well as spot like regions which are in the size of defect distances. Looking at the peak wavelength map one can see that in some areas with particularly large defects a strong red-shift occurs (indicated by arrows), but most of the emission changes seem to be independent of defects. Figure 2 shows a 2 2 micrometer scan at room temperature of sample BS 214 that has 4 quantum wells. In the topography only a few dark spots are visible now. Counting 2 these spots leads to a defect density of about 6 108 cm , which is a about 24 times lower than defect densities determined by wet chemical etching on samples with similar buffer layer. Therefore it could be possible that not all defects have been decorated by
+ 3 nm

- 3 nm

emissi on intens .

380 5 20 wavelength / nm

Fig. 1 (online colour). The images show a 4 4 micrometer scan of sample BS 213 with 10 quantum wells. From left to right: topography information, integrated optical intensity, peak wavelength map (centre wavelength 440 nm), the diagram shows a filtered optical emission spectrum; lines in the optical images indicate defect positions

phys. stat. sol. (c) 0, No. 1 (2002)

539

Fig. 2 (online colour). The images show a 2 2 micrometer scan of sample BS 214 with 4 quantum wells. From left to right: topography information, integrated optical intensity, peak wavelength map (centre wavelength 440 nm), the diagram shows a filtered optical emission spectrum; lines in the optical images indicate defect positions

V-defects. Looking at the optical intensity images we can see a clear correlation between the dark spots in topography image and the optical emission intensity: At each defect position we get a dark spot in emission intensity. The intensity change between the bright and dark regions is about 50% of maximums emission intensity. Looking at the peak wavelength map one can see that the emission tends to be red-shifted at some defect positions. Emission intensity and wavelength changes that happen at a larger distance scale are here only of minor importance. The 2 quantum well sample (BS 267) was studied at a temperature of about 50 K. Figure 3 shows that integrated intensity image, the peakwavelength image, and a multiplication of both. As expected the width of the emission peak has become smaller. Unfortunately, the surface of this sample does not exhibit any defect-related pits, which can be used for correlation with near-field images. Nevertheless, the intensity image also gives a good indication for defect positions, since dark spots should represent nonradiative centres there. In contrast to the room temperature images, this low-temperature image shows short-wavelength emission in dim areas (close to defects) and longwavelength emission in bright areas. This can be seen well in the multiplication of intensity and peak wavelength image, where blue regions nearly have disappeared and red regions are brightest. Both the room temperature and the low temperature SNOM images show a clear correlation with the defect structure of the samples. As far as intensity images are concerned, this is no surprise at least for room temperature images, since defects do represent nonradiative centres. The larger scale room-temperature images obtained for the 10 QW sample show a strong red-shift at some larger defects but a fairly large scale wavelength variation otherwise. The better resolved room-temperature images for the 4 QW sample show a

540

F. Hitzel et al: Correlation between Emission Spectra and Defect Position

- 2 nm

emission intens.
380 520 wavelength / nm
Fig. 3 (online colour). The images show a 4 4 micrometer scan of sample BS 267 with 2 quantum wells at 50 K. From left to right: integrated optical intensity, peak wavelength map (centre wavelength 435 nm), multiplication of left two images, the diagram shows a filtered optical emission spectrum

clear correlation between topographic features due to defects and wavelength shifts. Surprisingly, however, bright areas between defects appear blue-shifted while dim areas close to defects appear red-shifted. This is somehow in contradiction to a fairly high room-temperature quantum efficiency (around 10%) of this sample, since one would expect that carriers diffuse towards lower energy (red-shifted) regions, which are close to the nonradiative defects. A possible origin of this red-shift could be the spiral growth of InGaN close to some defect types [8]. The low temperature images, where the emission linewidth is narrower and carriers are less diffusive, actually reveal the opposite: Here we can say that the emission is blue-shifted close to defects while it is red-shifted in between. If this represents a general scenario, it would provide an explanation for the unexpectedly large emission efficiency of high-quality InGaN/GaN quantum wells: If defects are surrounded by an area of larger bandgap this would provide a barrier suppressing nonradiative recombination. It is likely that the formation of such barriers is closely related to InGaN growth. Possible mechanisms include a reduced In incorporation in the vicinity of defects, a reduced quantum well width, or locally enhanced strain. Besides these correlated local bandgap variations, there are certainly also random alloy fluctuations. However, even if we are unable to resolve such fluctuations, they would still provide a random distribution of emission wavelength, if they would dominate over the correlated changes discussed earlier. Conclusion Our near-field images both at room temperature and at low temperature show a clear correlation between the defect structure and emission intensity and wavelength. At least at low temperature, areas close to defects appear blue-shifted. This implies that carriers may be kept away from defects by local potential variations which are correlated with the defect structure and which arise during sample growth.

phys. stat. sol. (c) 0, No. 1 (2002)

541

References
[1] [2] [3] [4] [5] [6] [7] [8] S. Nakamura, M. Senoh, N. Iwasa, S. Nagahama, Appl. Phys. Lett. 67, 1868 (1995). S. Nakamura, IEEE J. Sel. Top. Quantum Electron. 3, 435 (1997). S. D. Lester, F. A. Ponce, M. G. Craford, D. A. Steigerwald, Appl. Phys. Lett. 66, 1249 (1995). X. Zhang, D. H. Rich, J. T. Kobayashi, N. P. Kobayashi, P. D. Dapkus, Appl. Phys. Lett. 73, 1432 (1998). T. Someya, Y. Arakawa, Jpn. J. Appl. Phys. 38, L 1216 (1999). Y. Chen, T. Takeuchi, H. Amano, I. Akasaki, N. Yamada, Y. Kaneko, S. Y. Wang, Appl. Phys. Lett 72, 710 (1998). X. H. Wu, C. R. Elsass, A. Abare, M. Mack, S. Keller, P. M. Petroff, S. P. DenBaars, J. S. Speck, S. J. Rosner, Appl. Phys. Lett. 72, 692 (1998). S. Keller, U. K. Mishra, S. P. Denbaars, W. Seifert, Jpn. J. Appl. Phys. 37 (4B), L431 (1998).