Si heterostructures under thermal annealing

Si heterostructures under thermal annealing

ARTICLE IN PRESS Materials Science in Semiconductor Processing 8 (2005) 137–141 Misfit dislocation nucleation and multiplication in fully strained Si...

327KB Sizes 0 Downloads 58 Views

ARTICLE IN PRESS

Materials Science in Semiconductor Processing 8 (2005) 137–141

Misfit dislocation nucleation and multiplication in fully strained SiGe/Si heterostructures under thermal annealing M. Rzaeva,, F. Scha¨fflerb, V. Vdovinc, T. Yugovad a Lebedev Physical Institute RAS, 119991 Moscow, Russian Federation Institut fu¨r Halbleiter- und Festko¨rperphysik, Johannes-Kepler-Universita¨t, A-4040 Linz, Austria c Institute for Chemical Problems of Microelectronics, 119017 Moscow, Russian Federation d Institute of Rare Metals, 119017 Moscow, Russian Federation

b

Available online 30 October 2004

Abstract An evolution of dislocation structure formed in fully strained Si1xGex/Si(0 0 1) heterostructures during thermal annealing was studied. Heterostructures with Ge content x ¼ 0:15 and 0.30 were grown by MBE on low-temperature Si(400 1C) and SiGe(250 1C) buffer layers. The main attention was devoted to the initial stages of strain relaxation and to the role of intrinsic point defects in misfit dislocation nucleation. A mechanism is proposed for the misfit dislocation nucleation at heterogeneous sources placed within SiGe epitaxial layer. r 2004 Elsevier Ltd. All rights reserved. PACS: 68.55; 81.15.A; 68.55.L; 68.37.L Keywords: SiGe/Si heterostructures; Strain relaxation; Low-temperature epitaxial growth; Point defects; Misfit dislocations; Transmission electron microscopy

1. Introduction Fully strained heterostructures containing epitaxial layers enriched with intrinsic point defects are currently used for preparing SiGe/Si virtual substrates. Thermal annealing of such heterostructures leads to practically full strain relaxation even in the case of thin buffer layers (o200 nm) [1–4]. However, low threading dislocation density (p105 cm2) cannot always be obtained in these layers. This problem stimulates further investigations of the processes of misfit dislocation (MD) nucleation, propagation and multiplication.

Corresponding author. Tel.: +7 095 132 6686; fax: +7 095 135 7880. E-mail address: [email protected] (M. Rzaev).

A lot of mechanisms of MD generation have been suggested by now (see, for example, a review [5]). Most part of these mechanisms is based on the formation of spiral or Frank–Read sources at pre-existing threading dislocation in the epitaxial layer. Generation of such sources is commonly attributed to any points in the layer at which threading dislocations are pinned [6–8]. As a matter of fact these mechanisms describe the following stages of dislocation generation after dislocation nucleation itself. A few papers have been devoted to the study of the initial stages of strain relaxation, i.e. dislocation nucleation [9–11]. Different kinds of heterogeneous sources of dislocation nucleation have been observed: "diamond defects" (faulted dislocation loops with b ¼ a=6h114i; 20–200 nm) which act as multiply regenerative sources of 601-MDs with different b vectors [9]; small vacancy-type dislocation loops which are nucleated at

1369-8001/$ - see front matter r 2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.mssp.2004.09.027

ARTICLE IN PRESS 138

M. Rzaev et al. / Materials Science in Semiconductor Processing 8 (2005) 137–141

Ge-rich platelets (o1.5 nm) localized at the interface [10]; large dislocation loops (50–200 nm) which are formed due to an agglomeration of point defects related to the implantation of near-surface part of the layer with B+ ions but cannot propagate to form MDs [11]. Accordingly, further investigations of the processes of dislocation nucleation, propagation and multiplication are needed to be continued. An idea that intrinsic point defects (IPDs) contribute to dislocation nucleation seems to be reasonable. Moreover, an assumption has been put out that the presence of the low-temperature (LT) buffer layer in heterostructure changes the MD nucleation mechanism and/or rate of the dislocation nucleation [3]. In our previous work [12] we have studied the dislocation structure in multilayer Si0.7Ge0.3/Si heterostructures with LT-Si and LT-(Si+SiGe) buffer layers. It was found that in these heterostructures the MD generation occurred through mechanisms typical for SiGe/Si heterostructures grown at conventional temperatures (e.g., Perovic–Houghton mechanism [10]), whereas the rate of dislocation nucleation was much higher due to the high vacancy concentration near the interface. However, the dislocation structure formation occurred in these heterostructures during the epitaxial growth; thus, we had no chance to study the initial stages of strain relaxation. This work focuses on the kinetics of dislocation structure formation in the fully strained Si1xGex/Si heterostructures with LT-Si and LT-SiGe buffer layers during thermal annealing. The main attention is devoted to the MD nucleation at the initial stages of strain relaxation as well as the MD multiplication.

2. Experimental Heterostructures were grown in a RIBER–SIVA 45 MBE machine. Prior to epitaxial growth, the substrates were cleaned with a standard RCA process followed by annealing at 1035 1C for 15 min in a growth chamber to remove the native oxide. An undoped Si buffer layer was initially grown directly on the substrates at 750 1C. Two types of Si1xGex/Si(0 0 1) heterostructures with Ge content of 0.15 and 0.30 and layer thickness of 200 and 80 nm, respectively, were grown on the LTSi(400 1C) and LT-SiGe(250 1C) buffer layers with thickness of 50 nm. High-temperature SiGe layers were grown at a substrate temperature of 500 1C. Reference heterostructures were grown without LT layers. 5-nmthick Si-cap layer was grown on the surface of all heterostructures. Thermal annealing was carried out at 550, 600 and 650 1C for 3–10 min in hydrogen. Structural characterization of the samples was performed by preferential etching/Nomarski microscopy and transmission electron microscopy (TEM). Hereafter

the heterostructures studied will be referred as reference, LTSi and LTSiGe heterostructures.

3. Results and discussion Layer thicknesses of the SiGe alloys were chosen so that being about an order of magnitude greater than the appropriate critical thicknesses which are equal to 19.3 and 8.0 nm for x ¼ 0:15 and 0.30, as we calculated using Houghton’s approach [13]. Starting from the nominal layer thicknesses, we calculated energy of fully strained interface (Eee2h) [14]. For the reference and LTSi heterostructures, these values are equal to 1.52 J/m2 (x ¼ 0:15) and 2.4 J/m2 (x ¼ 0:30). For the LTSiGe heterostructures, these values are equal to 1.9 J/m2 (x ¼ 0:15) and 3.89 J/m2 (x ¼ 0:30) due to the enhanced thickness of SiGe layer. Chemical etching patterns show that all but two LTSiGe heterostructures have been grown fully strained. TEM examination of crosssectional samples (Fig. 1) shows that actual layer thicknesses slightly differ from the nominal ones. In as-grown Si/Si0.85Ge0.15/LT-Si0.85Ge0.15/Si-sub heterostructure, the measured values amount to 7 nm for Sicap layer, 210 and 80 nm for high-temperature and lowtemperature parts of the SiGe layer, respectively. LTSiGe layer possesses well-pronounced folded contrast in 220 dark-field images. 3.1. Evolution of dislocation structure during thermal annealing Using post-growth thermal annealing of the heterostructures, we studied different stages of dislocation structure evolution. Annealing conditions and experimental results obtained with the use of preferential chemical etching are summarized in Table 1. We measured linear density of morphological lines (NL) in cross-hatch revealed on the layer surface and density of dislocation etch pits (NS) revealed in the near-interface

Fig. 1. TEM cross-sectional image of as grown Si/Si0.85Ge0.15/ LT-Si0.85Ge0.15/Si-sub heterostructure.

ARTICLE IN PRESS M. Rzaev et al. / Materials Science in Semiconductor Processing 8 (2005) 137–141

139

Table 1 Annealing conditions and structural features of samples according to the chemical etching patterns Density*, NL (cm–1), NS (cm–2)

Annealing

x=0.30

Ref.

LT-Si

LT-SiGe

Ref.

LT-Si

LT-SiGe

3 min

NL NS NL

— — o1  101

— — o1  101

1.3  102 — 5.2  103

— — 1:0  103

— — o1:0  103

8.0  103 — 1.2  104

NS NL

— 2.8  101

— 4.4  101

1.0  107 7.0  103





6 min

1.6  103

p1  107 1.6  104

10 min

NS NL

— 6.1  101

— 1.3  102

1.5  107 1.0  104

2.0  105 1.9  103

NS NL NS NL NS

— 1.7  103 1.5  105 4.9  103 1.7  106

— 1.5  103 6.0  105 3.0  103 4.0  106

1.8  107 1.3  104 2.9  107 1.6  104 3.9  107

8.0  105 2.5  103 9.0  105 6.0  103 3.4  106

As grown 550 1C

x=0.15

600 1C

6 min

650 1C

6 min

3

1:7  10 — 3:1  103 — 4.2  103 2.3  106 5.7  103 5.7  106

2.6  107 1.7  104 2.6  107 1.9  104 3.9  107 1.9  104 5.5  107

NL—linear density of cross-hatch lines on the layer surface. NS—density of dislocation etch pits in the near-interface substrate region.

Fig. 2. A bunch of dislocation half-loops with the same b vector. TEM plan-view bright-field image of as grown Si/Si0.85Ge0.15/LT-Si0.85Ge0.15/Si-sub heterostructure.

substrate region. Cross-hatch patterns fairly well reflect the presence of MDs at the interface, however, the linear density of morphological lines is always less than the actual linear density of MDs. Dislocations in the substrate are formed as a result of MD multiplication at the interface through the modified Frank–Read mechanism [15]. Chemical etching patterns and results of TEM planview investigations allow us to distinguish three stages in evolution of dislocation structure. At the first stage, formation of initial 601-MDs is occurred in the form of separate long (about 100 mm) half-loops and bunches of closely spaced coplanar half-loops (Fig. 2). Stereomicroscopy showed that the half-loops in these bunches lie on the same {1 1 1} glide plane and each of them repulses the previous one into the substrate. Dislocation contrast analysis indicates that all half-loops in a bunch possess the same b vector. Some heterostructures with x ¼ 0:15 correspond to this stage (Table 1; the appropriate NL values are underlined by a single line). Since the displacement of MD lines into the substrate is relatively small one could not distinguish the dislocation etch pits revealed in the substrate.

At the second stage of dislocation structure formation, MDs propagate great distances (b100 mm) to form initial MD network consisting of the separate MDs and MD bunches. Some heterostructures with x ¼ 0:30 correspond to this stage (Table 1; the appropriate NL values are underlined by double line). In these samples, the dislocation etch pits revealed in the substrates cannot again be distinguished. Thus, we can characterize such MD networks as flat ones localized at the interface. In the third stage, MD multiplication occurs in the MD network. In appropriate heterostructures, dense threedimensional MD network is formed and numerous dislocation half-loops propagate from the interface downward into the substrate to a depth of about 1–3 mm. Such MD multiplication has happened in all heterostructures annealed at X600 1C. For annealing at 550 1C, MD multiplication has happened only in the heterostructures with LT-SiGe buffer layers. 3.2. Misfit dislocation nucleation Specific geometry of the dislocation half-loops in a bunch evidently implies that they were emitted by a heterogeneous regenerative dislocation source. We have found three types of structural defects in the SiGe layers, which could be placed both far away and close to MDs (Fig. 3): (i) small (o10 nm) dislocation loops (larger dislocation loop is shown in Fig. 3a); (ii) V-shaped dislocation half-loops with inclined segments reaching the layer surface (Fig. 3b); (iii) dislocation segments linked with one of the threading segment of MDs (Fig. 3c). It is interesting to note that such dislocation segments are always observed at the last half-loop in pileups. This fact indicates that they are involved in the MD generation. Stereomicroscopy gives us direct

ARTICLE IN PRESS 140

M. Rzaev et al. / Materials Science in Semiconductor Processing 8 (2005) 137–141

(a)

Fig. 4. Stereopair of micrographs showing a pileup produced from the V-shaped dislocation.

(b)

(c) Fig. 3. TEM plan-view micrographs of structural defects. (a) Circular dislocation loop in the volume of SiGe layer; (b) Vshaped dislocation half-loop with inclined segments reaching the layer surface; (c) Dislocation segment linked with one of the threading segment of short misfit dislocation; fragment of the image shown in Fig. 2; the sample is turned round the /1 1 0S direction parallel to the dislocation lines.

evidence that short fragment of the arm of V-shaped dislocation cross slip to form the next half-loop in the pileup (Fig. 4).

Obtained experimental results allow us to assume that MD generation is a multi-step process. At first, the incipient dislocation loops of vacancy type are nucleated at any interstitial platelets as suggested by Perovic and Houghton [10]. It is important to note that, as we observed, such loops can nucleate over the whole SiGe layer rather than at the interface. A further extension of the incipient loop leads to the formation of V-shaped half-loop when it slits up into two dislocation segments at the layer surface. However, in order to clarify this dislocation form, we need to continue the study of Vshaped half-loops. Near-surface fragment of any arm of V-shaped dislocation tends to get screw orientation and cross slip. After that this screw segment gradually rotates about a naturally formed pinning point to produce a free half-loop and reproduce the initial dislocation. A mechanism of the generation of such pileups has been described in detail by Washburn and Kvam [6]. In heterostructures with x ¼ 0:15; we observed different distribution of MDs at the interface. In reference and LTSi heterostructures annealed at 550 1C, the density of pileups was small (see Table 1) and predominant separate MDs propagated to great distances (4100 mm). While in LTSiGe heterostructures, the density of pileups was significantly greater, although their generation occurred during the epitaxial growth at lower temperature (500 1C). We suppose that this effect can be explained by a high rate of dislocation nucleation

ARTICLE IN PRESS M. Rzaev et al. / Materials Science in Semiconductor Processing 8 (2005) 137–141

in the latter heterostructure because vacancy concentration should be the greatest in the LT-SiGe buffer layer grown at 250 1C [16] rather than by greater energy of elastic deformation. 3.3. Misfit dislocation multiplication Experimental data summarized in Table 1 allow us to make some phenomenological conclusions on MD multiplication at the interface without taking specific mechanism into consideration. Firstly, MD multiplication does not obviously occur in the heterostructures in which the MD network is absent (x ¼ 0:15; first stage of dislocation structure formation). Thus, MD generation through the mechanisms suggested in Refs. [6,10] cannot lead to the strong plastic deformation of the substrates as it is usually observed in the case of well-developed three-dimensional MD networks, although it was stated by the authors. Secondly, MD multiplication does not occur even in the case of the presence of MD network at the interface until a needed MD density is achieved (x ¼ 0:30; second stage of dislocation structure formation). It is likely to explain the fact that intensive MD multiplication is only occurred in LTSiGe heterostructures during the annealing at 550 1C. Thirdly, in the comparable heterostructures with x ¼ 0:15 and 0.30, the value of density of dislocation etch pits (intensity of MD multiplication) changes proportionally to the energy Ee. The dislocation density in the substrate in the heterostructures with LT-SiGe buffer layers is about an order of magnitude higher than that in the other appropriate heterostructures. We believe that enhanced intensity of MD multiplication can be attributed to the higher elastic strains rather than the effect of IPDs.

4. Conclusions Fully strained Si1xGex/Si(0 0 1) heterostructures with Ge content x ¼ 0:15 and 0.30 have been grown at 500 1C by MBE. An effect of LT buffer layers on the dislocation structure formation during thermal annealing was studied. We show that, in general, the processes of MD nucleation, propagation and multiplication occur similarly in the heterostructures studied independently of the alloy composition and kind of buffer layer.

141

Intrinsic point defects related to the LT epitaxial growth influence mainly the rate of MD nucleation. The dislocation generation was found to occur in the form of bunches of dislocation half-loops emitted by heterogeneous regenerative dislocation sources. We have found that such pileups are produced as a result of acting a spiral source arising at the arms of V-shaped dislocation half-loops.

Acknowledgements This work was in part supported by the Russian Foundation of Basic Research (Grants 03-02-20007 BNTS, 02-02-16692) and INTAS (Grants 01-0194, 03-51-5015).

References [1] Hackbarth T, Herzog H-J, Zeuner M, Ho¨ck G, Fitzgerald EA, Bulsara M, Rosenblad C, von Ka¨nel H. Thin Solid Films 2000;369:148. [2] Bauer M, Lyutovich K, Oehme M, Kasper E, Herzog H-J, Ernst F. Thin Solid Films 2000;369:152. [3] Peng CS, Zhao ZY, Chen H, Li JH, Li YK, Guo LW, Dai DY, Huang Q, Zhou JM, Zhang YH, Sheng TT, Tung CH. Appl Phys Lett 1998;72:3160. [4] LeGoues FK, MRS Bulletin/April 1996;38. [5] Vdovin VI. Phys Stat Sol (a) 1999;171:239. [6] Washburn J, Kvam EP. Appl Phys Lett 1990;57:1637. [7] Tuppen CG, Gibbings CJ, Hockly M, Roberts SG. Appl Phys Lett 1990;56:54. [8] Beanland R. J Appl Phys 1992;72:4031. [9] Eaglesham DJ, Kvam EP, Maher DM, Humphreys CJ, Bean JC. Philos Mag A 1989;59:1059. [10] Perovic DD, Houghton DC. Inst Phys Conf Series 1995;146:117. [11] Stach EA, Hull R, Bean JC, Jones KS, Nejim A. Microsc Microanal 1998;4:294. [12] Vdovin VI, Mu¨hlberger M, Rzaev MM, Scha¨ffler F, Yugova TG. J Phys: Condens Matter 2002;14:13313. [13] Houghton DC. J Appl Phys 1991;70:2136. [14] Matthews JW. J Vac Sci Technol 1975;12:126. [15] LeGoues FK, Meyerson BS, Morar JF, Kirchner PD. J Appl Phys 1992;71:4230. [16] Szeles CS, Asoka-Kumar P, Lynn K, Gossmann GH- J, Unterwald FC, Boone T. Appl Phys Lett 1995;66:2855.