Si partially relaxed heterostructures

Materials Science and Engineering B91– 92 (2002) 437– 440 www.elsevier.com/locate/mseb

Misfit dislocation and threading dislocation distributions in InGaAs and GeSi/Si partially relaxed heterostructures C. Ferrari a,*, G. Rossetto b, E.A. Fitzgerald c a

C.N.R. Maspec Institute, Parco Area delle Scienze 37 /A, 43010 Parma, Italy b C.N.R. Ictima, Corso Stati Uniti 4, 35127 Pado6a, Italy c Massachusetts Institute of Technology, Massachusetts A6enue, Cambridge, MA 02139, USA

Abstract Single and compositionally graded InGaAs/GaAs and SiGe/Ge partially relaxed heterostructures of equivalent lattice mismatch and thickness have been characterized by high resolution X-ray diffraction and by X-ray double crystal topography. X-ray topography evidences a rapid decrease of the average length of misfit dislocation segments with the increase of the density of misfit dislocations both in InGaAs and GeSi/Si structures. For an equivalent amount of strain release, SiGe/Si heterostructures exhibit shorter misfit dislocation lengths with respect to InGaAs/GaAs heterostructures. Compositionally graded InGaAs/GaAs heterostructures show much longer misfit dislocation segments with respect to equivalent single heterostructures, thus confirming the effectiveness of composition grading to reduce the dislocation interaction. The threading dislocation density evaluated from the lattice mismatch and the average length of the misfit dislocations on the basis of a simple formula increases more rapidly as a function of strain release in GeSi/Si heterostructures with respect to InGaAs/GaAs ones. © 2002 Elsevier Science B.V. All rights reserved. Keywords: X-ray topography; InGaAs/GaAs and GeSi/Si heterostructures; Misfit dislocation distribution; Threading dislocations

1. Introduction InGaAs/GaAs and GeSi/Si heterostructures are of particular interest among lattice mismatch based heterostrustures. The GeSi/Si and the InGaAs/GaAs systems are largely used for electronic and opto-electronic applications, and have lattice mismatch values ranging from 4 to 7%. The accommodation of lattice mismatch occurs through the formation of misfit (MD) and threading dislocations. Threading dislocations are detrimental for the reliability of devices because they cross the whole thickness of the structure. It is therefore of particular interest to develop non-destructive characterization techniques which permit to understand the mechanism of defect formation. Despite the limited resolution of X-ray topography technique due to the width of dislocation images in heterostructures with a high dislocation density, a com* Corresponding author. Tel.: + 39-0521-269-222; fax: +39-0521269-206. E-mail address: [email protected] (C. Ferrari).

parison of topographs taken on GeSi/Si and InGaAs heterostructures have evidenced a different dislocation behaviour in these systems [1]. In the present work GeSi/Si and InGaAs/GaAs heterostructures with constant or graded composition are analysed using the same topographic method. 2. Experimental The SiGe/Si samples wafers were grown at 650 °C in a ultra high vacuum CVD system with a background pressure of 10 − 10 –10 − 9 Torr. The pressure during growth was 25× 10 − 3 Torr. InGaAs films were grown by MOVPE in a standard AIXTRON (AIX200) reactor at a growth temperature of 580 °C, a III/V ratio of 250 and a total pressure of 20 mbar. Group V source materials were PH3 and AsH3. The metalorganic precursors for Ga and In were Me3Ga and Me3In, respectively, and Pd-purified H3 was used as a carrier gas. Prior to the growth of the InGaAs layer a 500 nm thick GaAs buffer was deposited. The measured growth velocity for the InGaAs films was 0.35 nm s − 1 on the average.

0921-5107/02/$ - see front matter © 2002 Elsevier Science B.V. All rights reserved. PII: S 0 9 2 1 - 5 1 0 7 ( 0 1 ) 0 0 9 9 4 - 1

C. Ferrari et al. / Materials Science and Engineering B91–92 (2002) 437–440

438

Table 1 Summary of the sample characteristics and of the results of X-ray diffraction and topography measurements Sample

Thickness (nm)

Composition x

Type

Parallel mismatch Average MD length [110] (mm)

Average MD length [1−10] (mm)

440 439 UHV36 UHV35 UHV34 A B C

100 240 68 246 550 1330+1200 1790+1200 2000+1200

12.4% (In) 12% (In) 17% (Ge) 18.2 (Ge) 23% (Ge)

InGaAs/GaAs InGaAs/GaAs GeSi/Si GeSi/Si GeSi/Si See caption See caption See caption

6×10−4 5.3×10−3 0 4×10−4 6.3×10−3 1.5×10−2 1.75×10−2 2.4×10−2

30 3.5 0.5 1 0.1 5 9 6

30 2 0.5 1 0.1 3 6 2

TD density (cm−2)

4×104 4×106 8×105 1.2×108 7.5×106 4.6×106 1.2×107

For UHV36 MDs have been observed only at the wafer edges or in part of the wafer containing inclusions. Samples A, B and C are made by an InGaAs composition graded buffer with maximum In content of 0.2, 0.27 and 0.3 respectively and a final InGaAs layer with constant composition and thickness 1.2 mm.

The composition graded InGaAs/GaAs heterostructures have been grown on semi-insulating GaAs wafers (001) oriented by solid source Molecular Beam Epitaxy (MBE) using As4 beams with a composition gradient of 0.15 mm − 1. The growing surface temperature was kept at 500 °C during the growth. The As4 beam equivalent pressure (BEP) was 9.5×10 − 6 Torr, while the As4/Ga BEP ratio was 18. Other details of the growth procedure are reported by Bosacchi et al. [2]. Si CuKa (422) and Ge (620) curvable monochromators have been used for double crystal X-ray topographic characterization of GeSi/Si and InGaAs/GaAs structures respectively [3]. The X-ray topographs were taken fixing the angle of incidence of the X-ray beam at the maximum of the substrate Bragg peak. Taking into account that MD images as large as few mms are usually detected in X-ray topography, dislocations lying at the interface between layer and substrate or up to a few mm above it can be seen through their strain field acting on the substrate lattice. Using a Philips 5 crystal high resolution diffractometer both the two [110] in plane and the perpendicular lattice mismatches were determined using the CuKa (004) symmetric and the (335) asymmetric reflections [4]. From lattice mismatches the Ge and In content have been calculated assuming the Vegard law with an estimated accuracy of 9 0.2% due to the uncertainty of the X-ray diffraction peak center. The results are summarized in Table 1.

[0− 11] direction respectively, whereas MD lines crossing almost completely the wafer are seen in sample 440 of Table 1.

3.2. GeSi/Si single heterostructures In Fig. 2 a detail of the X-ray topography of the 68 nm thick Ge0.17Si0.83/Si sample showing a part of the sample close to the wafer rim is reported. The figure shows 500 mm long MD segments starting from the

Fig. 1. CuKa Ge 620 X-ray double crystal topograph of a 240 nm thick, 12% In InGaAs/GaAs single heterostructure corresponding to sample 439 of Table 1. The arrow represents the projection of the 620 scattering vector.

3. Results and discussion

3.1. InGaAs/GaAs single heterostructures Fig. 1 reports the X-ray topograph of the In0.12Ga0.88As/GaAs single layers of 240 nm thick of Table 1. Many ending segments are visible with approximate MD lengths of 2 and 3.5 mm along the [011] and

Fig. 2. CuKa Si 422 X-ray double crystal topograph of a 68 nm thick, 17% Ge GeSi/Si single heterostructure corresponding to sample UHV36 of Table 1. The arrow represents the projection of the 422 scattering vector.

C. Ferrari et al. / Materials Science and Engineering B91–92 (2002) 437–440

Fig. 3. CuKa Si 422 X-ray double crystal topograph of a 200 nm thick, 18.2% Ge GeSi/Si single heterostructure corresponding to sample UHV35 of Table 1. The arrow represents the projection of the 422 scattering vector.

Fig. 4. CuKa Si 422 X-ray double crystal topograph of a 480 nm thick, 23% Ge GeSi/Si single heterostructure corresponding to sample UHV34 of Table 1. The arrow represents the projection of the 422 scattering vector.

439

rmed by the presence of MD segments all having the same length. In Fig. 3 a detail of the topograph performed on the 200-nm thick GeSi/Si sample is shown, which evidences many ending MD segments. According to the observation of the large fluctuations of local MD density, the presence of segments of different width and contrast can be explained assuming that each visible segment corresponds to a bunch of MDs. In samples with dislocation densities larger than of 103 cm − 1 Samavedam et al. [6] explained the decrease of MD length with increasing strain release by the blocking mechanism of dislocation glide (Freund [7]) induced by the formation of MD pileups. This is consistent with the observation that all the visible MD segments in Fig. 3 end in correspondence with a perpendicular MD line. The changes in the slip direction of the MD segments seen in Fig. 3 can be attributed to a cross slip (see for instance G. Lacey et al. [8]). By comparison with the InGaAs sample 439, shorter MD segments are present in GeSi/Si heterostructures in agreement with the three order of magnitude lower dislocation velocity in Si based structures with respect to GaAs based ones [9]. In Fig. 4 a topograph of sample UHV34 with a parallel mismatch of 6.3× 10 − 3 is shown. Event though the determination of the MD length can be only indicative, the MD network now appears as made of much shorter segments.

3.3. Composition graded InGaAs/GaAs heterostructures In Fig. 5 the double crystal X-ray topograph of the composition graded heterostructure corresponding to sample C is reported. The final In content reaches the 30% and the final lattice mismatch is much higher than the other single heterostructures. From Fig. 5 much longer segments are seen, in comparison with single heterostructures having similar lattice mismatch. Due to the structure thickness it is not possible to attribute the contrast only to the presence of dislocations, since the layer morphology can affect the contrast. Nevertheless this confirms the reduction of misfit dislocation interaction in composition graded heterostructures [10].

3.4. E6aluation of threading dislocation density Fig. 5. CuKa Ge 620 X-ray double crystal topograph of the InGaAs composition graded sample C of Table 1.

edge of the wafer or from defects in the film, probably growth inclusions. It is worth to note that in dislocation free Si substrates dislocations are nucleated at the (001) edges of the wafer because of the higher Schmid factor for the formation and glide of half loops [5]. The kinematical limitation of the strain release is also confi-

The only knowledge of the parallel lattice mismatch Dd¦/d does not permit the determination of the threading dislocation density zTD in partially relaxed heterostructures. Taking into account that each dislocation must reach the surface or the edges of the crystal, zTD can be connected to the mismatch through the average MD segment length Žl by (Fitzgerald [11]):



zTD : 4zMD

1 1 − l L



(1)

440

C. Ferrari et al. / Materials Science and Engineering B91–92 (2002) 437–440

where L is the sample size. The previous formula states that the threading dislocation density rapidly decreases by increasing the average MD length Žl. The linear MD density zMD can easily be calculated by the measured lattice mismatch Dd¦/d by: Dd¦ d zMD = bedge

(2)

where for 60° MDs the Burgers vector component is bedge =a/ 8, a being the lattice parameter. Although qualitative in some samples, the estimation of the average MD length Žl from the topographs of Figs. 1–5 permits an evaluation of TD density in a completely non-destructive way. The results are reported in Table 1. Transmission electron microscopy cross sections micrographs of composition graded samples [12] indicate a much lower threading dislocation density than reported in Table 1. Possible reasons for such a discrepancy are: (a) the presence of MD segments much longer than those visible in the topographs; and (b) a large number of MD cross slip events, which leave MD segments at 90°. Although in the present investigation cross slip events only represent a small part of the visible MD interactions, this point deserves a particular interest because of the observation of large variation of lattice tilts in composition graded heterostructures. Lattice tilts between heterolayer and substrate are due to a MD network with non random distribution of the possible Burgers vectors of the MDs [13]. Natali et al. [14] propose the mechanism of cross slip as responsible for the formation of an epilayer to substrate concave lattice curvature up to several degrees/cm in composition graded heterostructures.

4. -Conclusions X-ray topography technique is normally used to investigate sample with a low density of structural defects due to the relatively large size of defect image given by X-ray diffraction. Nevertheless in partially relaxed structures with MD density up to several 104 cm − 1, the rather large fluctuation of local MD density produces a contrast in X-ray topographs which can be used to study the MD distribution. Thanks to that particular features, X-ray topography has been used to study InGaAs and GeSi/Si samples with lattice mismatch values Dd/d up to 104.

The analysis of the X-ray topographs evidenced that the increase of the strain release leads to the formation of much shorter MD segments, the main responsible for such mechanism being the blocking mechanism of MDs. The shorter average MD length for comparable MD linear dislocation densities in GeSi/Si heterostructures with respect to InGaAs/GaAs ones are due to the lower glide mobility of dislocations, as confirmed by the topograph of a metastable GeSi/Si heterostructure beyond the critical thickness. As a result the threading dislocation densities evaluated by the strain release and the average length of the MD segments increase more rapidly with the strain release in GeSi/Si heterostructures with respect to equivalent InGaAs/GaAs ones. In comparison InGaAs/GaAs composition graded heterostructures show much longer MD segments thus confirming the reduction of dislocation interaction in such systems.

References [1] C. Ferrari, S. Gennari, G. Salviati, M. Natali, A.V. Drigo, G. Rossetto, M.T. Currie, E.A. Fitzgerald, in: E. Fitzgerald (Ed.), Lattice Mismatch and Heterovalent Thin Film Epitaxy, The Minerals, Metals and Materials Society, Warrendale, Pennsylvania, 1999, p. 39. [2] A. Bosacchi, A.C. De Riccardis, P. Frigeri, S. Franchi, C. Ferrari, S. Gennari, L. Lazzarini, L. Nasi, G. Salviati, A.V. Drigo, F. Romanato, J. Crystal Growth 175/176 (1997) 1009. [3] B. Jenichen, R. Ko¨ hler, W. Mo¨ hling, Phys. Stat. Sol. 89 (1985) 79. [4] C. Bocchi, C. Ferrari, P. Franzosi, A. Bosacchi, S. Franchi, J. Crystal Growth 132 (1993) 427. [5] J. Matsui, Appl. Surf. Sci. 50 (1991) 1. [6] S. Samavedam, E.A. Fitzgerald, J. Appl. Phys. 81 (1997) 3108. [7] L.B. Freund, J. Appl. Phys. 68 (1990) 2073. [8] G. Lacey, C.R. Whitehouse, P.J. Parbrook, A.G. Cullis, A.M. Keir, P. Mo¨ ck, A.D. Jonson, G.W. Smith, G.F. Clark, B.K. Tanner, T. Martin, B. Lunn, J.H.C. Hogg, M.T. Emeny, B. Murphy, S. Bennet, Appl. Surf. Sci. 123/124 (1998) 718. [9] I. Yonenaga, K. Sumino, J. Appl. Phys. 65 (1989) 85. [10] J. Tersoff, Appl. Phys. Lett. 62 (1992) 693. [11] E.A. Fitzgerald, Mater. Sci. Rep. 7 (1991) 87. [12] L. Lazzarini, C. Ferrari, S. Gennari, A. Bosacchi, S. Franchi, M. Berti, A.V. Drigo, F. Romanato, G. Salviati, Inst. Phys. Conf. Ser. 157 (1997) 149. [13] C. Ferrari, S. Gennari, S. Franchi, L. Lazzarini, M. Natali, F. Romanato, A.V. Drigo, J. Baruchel, J. Crystal Growth 205 (1999) 474. [14] M. Natali, F. Romanato, E. Napolitan, D. De Salvador, A.V. Drigo, Phys. Rev. B 62 (2000) 11054.