ARTICLE IN PRESS

Journal of Crystal Growth 275 (2005) 157–166 www.elsevier.com/locate/jcrysgro

Feasibility of III–V on-silicon strain relaxed substrates M. Kostrzewaa,b,, G. Greneta, P. Regrenya, J.L. Leclercqa, P. Perreaub, E. Jalaguierb, L. Di Ciocciob, G. Hollingera b

a Ecole Centrale de Lyon, LEOM, (UMR CNRS 5512), 69134 Ecully, Cedex, France CEA-DRT-LETI/DIHS-CEA/GRE, 17, rue des Martyrs, 38054 Grenoble, Cedex 9, France

Available online 8 December 2004

Abstract In this work we have investigated the feasibility of using ultrathin III–V films stuck on silicon as seed layers for subsequent epitaxial growths. The sticking is done by a thick viscous layer, which is assumed to act as an accommodating layer allowing the elastic relaxation of the initially strained III–V film. Two kinds of viscous layers have been employed: the Apiezon wax for an academic study and the borophosphorosilicate glass (BPSG) for an actual technological study. From the academic investigation, viz, with the Apiezon wax, we have learned how undulation and in-plane expansion compete as elastic relaxation processes. In fact, a final uniform and flat seed-layer morphology is hard to achieve and, when so, only for very small samples. In the actual investigation, viz, with the BPSG, wafer bonding techniques have been successfully used to transfer a 20 nm thick InAs0.25P0.75 film onto a Si host substrate. Then, a post-process Rapid Thermal Annealing (RTA) has been performed to lower the BPSG viscosity and allow the 0.8%-strained film to relax. These two preliminary studies have clearly shown the feasibility of the approach even if the sticking-interface stability has to be greatly improved from a chemical point of view before considering any practical application in optoelectronics. r 2004 Elsevier B.V. All rights reserved. PACS: 81.05.Ea; 62.20.Fe; 46.32.+x Keywords: A1. Substrates; A3. Molecular beam epitaxy; B1. Glasses; B2. Semiconducting III–V materials

1. Introduction In recent years, wafer-bonding techniques have aroused a great interest in the scientific community Corresponding author. CEA-DRT-LETI/DIHS-CEA/ GRE, 17, rue des Martyrs, 38054 Grenoble, Cedex 9, France. Fax: +33 4 38 78 24 34. E-mail address: [email protected] (M. Kostrzewa).

for the integration of III–V optoelectronics devices onto silicon substrates. In fact, the direct heteroepitaxy of III–V semi-conductors on silicon is out of the question because of differences in lattice parameters and thermal expansion coefficients. Besides, these wafer-bonding techniques could be a very promising short-cut to solve other problems of lattice-mismatched growth in heteroepitaxy.

0022-0248/$ - see front matter r 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.jcrysgro.2004.10.080

ARTICLE IN PRESS 158

M. Kostrzewa et al. / Journal of Crystal Growth 275 (2005) 157–166

Fig. 1. Plastic-compliant substrate technology: (a) Epitaxial growth of the seed layer; (b) Sticking on a host substrate; (c) Back-etching of the initial substrate; (d) Growth of an overlayer.

In standard heteroepitaxy growths, lattice-mismatched epilayers when grown beyond a critical thickness are subject to plastic relaxation harmful to their optoelectronic properties. They suffer this structural deterioration as they are thinner than the substrate on which they are grown. Therefore, one can imagine an upturned scheme in which the substrate is thinner than the epilayer. In this case, the stress relaxation is expected to induce dislocations into the substrate rather than in the epilayer [1]. However, on the one hand, because of the extremely challenging mechanical handling of such large and ultra-thin seed substrates, most of the time, the seed substrate must be stuck onto an appropriate host substrate. But, on the other hand, the threading dislocations must be free to emerge from the seed substrate on its backside. That is why a special interface between the seed substrate and its host substrate has to be engineered. Various schemes have been suggested such as twisted/tilted bonding [2–4] or patterned interfaces [5] but generally, the threading dislocations cannot be completely confined into the seed substrate and some spoil the optical properties of the overgrown active layer. This pioneer approach is summarised in Fig. 1 but more complete accounts can be found in Refs. [6–8].

Another line of attack could be to look for elastic relaxation instead of plastic relaxation. For instance, attempts have been made at using freestanding ultra-thin substrates [9,10]. Investigating along this line, our group has implemented a new approach we named ‘‘paramorphic’’, which is sketched in Fig. 2. The initial heterostructure (Fig. 2a) consists of the following sequence: a pseudomorphic epilayer, a sacrificial layer, an etch-stop buffer layer, and finally the substrate. The sacrificial layer and the etch-stop buffer layer are lattice-matched with the substrate. The pseudomorphic epilayer is lattice-mismatched but its thickness is smaller than the critical thickness for plastic relaxation. Therefore, no threading dislocations are expected in it. Then this initial heterostructure is patterned (Fig. 2b) and the pseudomorphic epilayer freed from the substrate by the lateral selective under-etching of the sacrificial layer (Fig. 2c). In the etching solution, the pseudomorphic epilayer relaxes its compressive strain by a lateral in-plane expansion that begins at the film edges and slowly progresses to the sample centre at the same speed as the lateral etching itself. Using this technique, our group succeeded in growing a 2 mm-thick In0.65Ga0.35As layer without any dislocations using a InAs0.25P0.75

ARTICLE IN PRESS M. Kostrzewa et al. / Journal of Crystal Growth 275 (2005) 157–166

159

Fig. 2. Paramorphic substrate technology: (a) Epitaxial growth of an heterostructure comprising the seed layer plus a sacrificial layer; (b) Patterning and lateral etching leading to the seed layer elastic relaxation; (c) Drying and sticking onto the initial substrate; (d) Growth of an overlayer.

layer grown on InP (0 0 1) as seed layer [11,12]. The high quality of the growth was checked by recording the photoluminescence (PL) spectra of a 7 nm thick InAs quantum well between In0.65Ga0.35As barriers. However, in the paramorphic approach, the residual strain and the selectivity of the etching solution limit the reachable sample diameter to around 300 mm. Another idea could be to combine the two previous approaches by sticking an ultra thin membrane onto a host substrate via a thick viscous layer which is supposed to be the compliant layer allowing the elastic relaxation of the initially strained III–V film [13]. Because of its low viscosity, the sticking layer holds the thin substrate in position but also allows its elastic relaxation as the etching solution does in the paramorphic approach (Fig. 3b). However, several differences should be noted. First, the viscosity of a glass (around 108 P) [14] is far greater than that of an acid solution (around 2
102 P for HCl concentrated acid). Second, unlike lateral under-etching, which induces a strictly

edge-driven relaxation, the back etching mainly leads to a surface-driven relaxation. There results a strong competition between an in-plane expansion induced by the edges and a buckling induced by the surface. The result is thus strongly dependant on the distance from edges and surface, respectively. Finally it should be noticed that even if the sticking itself is usually done at low temperature (Fig. 3b), the stickinglayer viscosity can be lowered by an annealing process either before (‘‘paramorphic approach’’) or during (‘‘compliant approach’’) the overlayer growth. The key steps are summarised in Fig. 3. In this study, the first stage was aimed at evaluating the balance between in-plane expansion and buckling for an ultrathin pseudomorphic In0.65Ga0.35As layer bonded onto a Si host substrate via a thick Apiezon wax layer. The second stage consisted in studying the behaviour of strained InAs0.25P0.75 (20 nm)/In0.65Ga0.35As (20 nm) bilayers stuck onto a silicon substrate via a thick borophosphorosilicate glass (BPSG). Post

ARTICLE IN PRESS 160

M. Kostrzewa et al. / Journal of Crystal Growth 275 (2005) 157–166

Fig. 3. Elastic-compliant substrate technology: (a) Epitaxial growth of the seed layer; (b) Sticking on a host substrate via a viscous layer; (c) Back-etching of the initial substrate leading to the seed layer elastic relaxation; (d) Growth of an overlayer.

process rapid thermal annealing (RTA) was used to decrease the BPSG viscosity. All the heterostructures were grown on epiready semi-insulating 2-inch InP (0 0 1) substrates from InPact SA using a Riber 2300 solid-sourcemolecular-beam-epitaxy (SSMBE) reactor equipped with high capacity P and As cracking cells and standard In and Ga solid sources. The structures thus prepared were stuck onto a thick silicon host substrate via an amorphous material either Apiezon wax or BPSG.

2. Apiezon wax as viscous sticking layer The first system studied was an ultra thin pseudomorphic In0.65Ga0.35As layer (0.8% com-

pressively stressed on InP) stuck onto a Si host substrate via a thick Apiezon wax layer, a material of very low viscosity (fusion point 85 1C). This material allows the thin film elastic relaxation to be time monitored even at room temperature. As schematised in Fig. 4, the original heterostructure consisted of a 500 nm thick In0.53Ga0.47As layer (acting both as etch-stop and buffer layer), of a 200 nm thick InP layer (etch-stop layer) and of a 30 nm thick In0.65Ga0.35As pseudomorphic film whose elastic relaxation we were interested in. An additional capping InP layer protected the In0.65Ga0.35As layer during the selective back etching of the sacrificial In0.53Ga0.47As layer. This heterostructure was then patterned (Fig. 4b) using standard photolithography process into mesas ranging from 100
100 mm2 up to 600
600 mm2

ARTICLE IN PRESS M. Kostrzewa et al. / Journal of Crystal Growth 275 (2005) 157–166

161

In0.65Ga0.35As (30 nm)

In0.65Ga0.35As (30 nm) InP (200 nm)

InP (200 nm)

In0.53Ga0.47As (500 nm)

In0.53Ga0.47As (500 nm)

InP substrate

InP substrate

(a)

(b)

InP substrate In0.53Ga0.47As (500 nm)

Camera Apiezon Wax

InP (200 nm) 20 µm Silicon substrate

(c)

Silicon substrate

(d)

Fig. 4. Key technological steps: (a) Heterostructure growth by MBE; (b) Patterning into mesas; (c) Sticking onto a Si host substrate via Apiezon W layer; (d) Back-etching of the sacrificial substrate and of the etch-stop layer.

with edges orientated along the cleaving orientations, i.e. f1 1 0g and f1  1 0g: It was then stuck upside down (Fig. 4c) onto a silicon host substrate previously covered with a thick (20 mm) wax layer melted at 100 1C. The bonding was achieved by a cooling down to room temperature. The last technological step (Fig. 4d) was the selective back-etching of the initial InP substrate, of the In0.53Ga0.47As sacrificial layer and finally of the protecting InP layer. Typical Nomarski optical images obtained after 9 min at room temperature are shown in Fig. 5. It is clear from this figure that the mesa lateral dimensions are a critical issue. The 100
100 mm2 (Fig. 5a) mesas seem almost completely relaxed through in-plane expansion whereas the relaxation of the 150
150 mm2 mesas is clearly not so complete (Fig. 5b): the corners look (Fig. 5c) relaxed by in-plane relaxation—like the small mesa—whereas other areas show undulations periodically rumpling the film (Fig. 5d). These undulations are perpendicular to the edges when close enough to them but oriented along the f1 1 0g directions at the mesa centre. In our case, the critical dimension for in-plane expansion is roughly 100
100 mm2. The average undulation periodicity and amplitude (half the feature height)

measured by Atomic Force Microscopy (AFM) at the centre of a large mesa are close to 4.8 and 0.09 mm, respectively. We have also prepared a 30 nm thick In0.41Ga0.59As epilayer tensilely strained ðDa=a ¼ 0:85%Þ on InP substrate using the same technological preparation. In the tensile case, wrinkling, buckling or any kind of undulations are not allowed as relaxation phenomena. The only authorised way to relax internal stress is in-plane shrinking. The associated in-plane mass displacement creates f1 1 0g—oriented cracks on the surface as shown in Fig. 6. The crack amount depends on the membrane dimensions. Once again there is a critical dimension for total elastic relaxation. The critical dimension is slightly less than the 100 mm of the compressive case because first the lattice mismatch is greater in the tensile case (0.85%) than in the compressive one (0.80%) and second because no other relaxation modes are available. In the tensile case, the crack direction is ruled by the zinc blende cleaving directions, i.e. f1 1 0g and not by the elastic anisotropy along f1 0 0g as previously observed [15]. We concluded from this preliminary study that uniform and flat in-plane elastic relaxation is

ARTICLE IN PRESS 162

M. Kostrzewa et al. / Journal of Crystal Growth 275 (2005) 157–166

Fig. 5. Compressive case–In0.65Ga0.35As–Nomarski views (a) of a 100
100 mm2 mesa showing almost complete elastic relaxation, (b) of a 150
150 mm2 mesa showing incomplete elastic relaxation, (c) of a corner of view (b) and (d) of the centre of view (b).

compatible with the use of such layers as seed layers for subsequent epitaxial overgrowths.

3. Borophosphorosilicate glass (BPSG) as viscous sticking layer

Fig. 6. Tensile case–In0.41Ga0.59As–Nomarski views. There is a slight misalignment between the mesas edges and the crystallographic directions.

achievable only for very small areas otherwise buckling is the main elastic process. However, we believe that the resulting morphology in terms of undulation periodicity and amplitude could be

The next step was thus to swap the Apiezon wax for a stabilised BPSG (SiO2: 83.27%, B2O3: 14.11%, P2O5: 2.62%) which is potentially compatible with all the III–V semiconductor growth requirements. However, we are faced with a dilemma. On the one hand, the temperature in a MBE chamber cannot be increased beyond 550 1C without running the risk of degrading the III–V semiconductor heterostructure. On the other hand, a viscosity around 108 P is required for our task which would theoretically imply heating BPSG up to 900 1C [14]. However, the experimental BPSG flowing on a patterned silicon surface demonstrates that in fact, the temperature can be lowered down to the 600–700 1C range. For this reason, we

ARTICLE IN PRESS M. Kostrzewa et al. / Journal of Crystal Growth 275 (2005) 157–166

163

InP substrate / InGaAs LM 0.5 µm / InP 0.2 µm In 0.65Ga0.35As (20 nm) SiO2

(a)

In As0.25P0.75 (20 nm)

In 0.65Ga0.35As (20 nm) SiO2

In As0.25P0.75 (20 nm)

BPSG (1300 nm)

BPSG (1300 nm)

Silicon substrate

Silicon substrate

(b)

overlayer SiO2

In As0.25P0.75 (20 nm)

SiO2

BPSG (1300 nm)

(c)

Silicon substrate

In As0.25P0.75 (20 nm) BPSG (1300 nm)

(d)

Silicon substrate

Fig. 7. Key technological steps: (a) Initial heterostructure grown by MBE sticked onto a Si host substrate via SiO2 plus BPSG; (b) Back-etching of the initial substrate and the etch-stop layers; (c) Back-etching of the protective InGaAs layer; (d) Growth of an overlayer.

chose a bilayer, viz, InAs0.25P0.75 (20 nm) plus In0.65Ga0.35As (20 nm) instead of a homogeneous In0.65Ga0.35As layer (40 nm). The real seed layer was the InAs0.25P0.75 (20 nm) layer because it is much easier to grow an overlayer on InAsP than on a GaInAs [16,17]. But this InAsP seed layer required a protection—the In0.65Ga0.35As layer— because an annealing was scheduled. With this bilayer configuration, the annealing needed to lower the BPSG viscosity can be done either at 500 1C under arsine pressure in our MBE chamber or up to 630 1C by RTA. This RTA temperature of 630 1C appeared as a good compromise for obtaining a viscosity low enough for an observable elastic relaxation without deteriorating the III–V heterostructure too much. Finally, let us mention some additional details about the adhesion process when PBSG is used as viscous sticking layer. The BPSG is deposited at 390 1C (PECVD) and then stabilised at high temperature (900 1C) first to drive out the moisture and second to make the layer denser. In fact, the wafer-to-wafer spontaneous bonding is of the hydrophilic type between the two carefully cleaned and hydrated surfaces. A 5–10 nm thin SiO2 has to be deposited using Electron Cyclotron Resonance (ECR) plasmas on the III–V surface to help with the adhesion process. The bonding itself is realised at room temperature but then annealed at 200 1C

to reinforce bonding between the silanol groups [18]. Such a bonding interface (SiO2/BPSG) is susceptible to desorb trapped atoms and molecules when submitted to higher temperatures, especially in an ultra high vacuum chamber [19]. Fig. 7 gives details about the whole process. Fig. 8 presents the morphology after step b of Fig. 7, i.e. before the chemical etching of protecting In0.65Ga0.35As layer but after a RTA at 630 1C for 5 min. The elastic relaxation mode appears similar to the one obtained with the Apiezon wax. However, the undulation periodicity is shorter, viz 1.8 mm (instead of 4.8 mm) with almost the same undulation amplitude 0.1 mm (instead of 0.09 mm) indicating a not so complete elastic relaxation probably due to a higher viscosity. Afterwards, the protecting In0.65Ga0.35As layer was removed by a chemical back etching and the 20 nm thick InAs0.25P0.75 was ready to be used as seed substrate. At this point, let us just mention some previous findings concerning overgrowth performed on a similar system—thin InP films stuck onto a host Si wafer via silicon dioxide, (InP (0 0 1)/SiO2/Si). We have shown that thick layers—either latticematched (InP) or lattice-mismatched materials (InAs0.25P0.75 and the In0.65Ga0.35As) can be grown on such substrates with almost the same structural and optical properties as those grown on

ARTICLE IN PRESS 164

M. Kostrzewa et al. / Journal of Crystal Growth 275 (2005) 157–166

Fig. 8. Nomarski views of InAs0.25P0.75/In0.65Ga0.35As compressive strained bilayer relaxed on BPSG viscous layer by RTA at 630 1C. (a) Large view showing the surface deterioration; (b) detail of the center of mesa.

standard bulk InP (0 0 1) substrates [17]. Moreover, the PL measurement for a 6 nm thick InAs0.25P0.75 strained quantum well structure convinced us that such InP (0 0 1)/SiO2/Si substrates can be used as reliable substrates even for optoelectronic applications. With the above-mentioned results in mind we grew 250 nm thick InAs0.25P0.75 overlayers on a 20 nm thick InAs0.25P0.75 seed layers themselves stuck onto BPSG/Si substrate. These InAs0.25P0.75 seed layers were either elastically relaxed by RTA or not. Fig. 9 show optical evidence—in both cases—of degradation: blisters and holes for the largest mesa and roughness for the smallest ones. We attributed this phenomenon to a chemical instability at the bonding interface InAsP/SiO2/ BPSG. Then, the desorbed species migrate along the easiest path, that is to say, either along the interface to the sample edges or through the seed layer to the surface depending on which they are closest to. Note that even if the RTA temperature (630 1C) was higher than the growth temperature (480 1C), the seed layer was not damaged (Fig. 8) at this stage because the seed layer was still capped—and thus thicker (40 nm instead of 20 nm)—and it was not under ultra high vacuum. This makes all the difference. In the case of an overgrowth on a non annealed InAsP seed layer (Fig. 9a), the smallest mesa (upto 200
200 mm2)

are obviously less damaged than the largest mesa, the centre of the latter being completely destroyed by interface bubbling. If now the overgrowth is carried out on an annealed seed layer (Fig. 9b), the situation is notably improved for the largest mesa (thanks to the interface stabilisation provided by RTA) but slightly worsened for the smallest ones (owing to the seed layer initial buckling). Finally, for a 250 nm thick InAs0.25P0.75 layer grown on a non annealed seed layer, we detected on the PL spectrum at 77 K a peak positioned at 1127.5 nm (full width at half maximum—FWHM 70 nm) comparable to the one at 1160 nm (FWHM 40 nm) obtained for a similar overlayer but grown on standard InP substrate. Even if it is doubtful that this shift towards higher energies can be attributed to a compliant effect, nevertheless, we can conclude that growing III–V alloys on a seed layer stuck onto Si host substrate via BPSG is feasible and, what is more, this could be achieved with the quality required for optoelectronic applications.

4. Discussion and concluding remarks Recently Sridhar et al. [20] have proposed a model for the elastic relaxation of a laterallyinfinite compressively-stressed film stuck onto a rigid substrate by a glass-like layer. Using linear

ARTICLE IN PRESS M. Kostrzewa et al. / Journal of Crystal Growth 275 (2005) 157–166

165

Fig. 9. Nomarski views after the growth of a 250 nm thick InAs0.25P0.75 layer on InAs0.25P0.75 seed layer. (a) Seed layer without annealing before growth; (b) seed layer with annealing (2 min, 630 1C) before growth.

stability analysis and small amplitude approximation, they demonstrate that undulations appear with a growth rate depending on viscosity but a periodicity depending only on the film latticemismatch and on the relative thickness of the layers. Let us assume that in our case the viscous layer is infinitely thick (1 mm) compared to the thin layer (30 or 40 nm). In this case, the undulation wavelength is given by lm ¼ phf =½jjð1 þ nÞ 1=2 where hf is the film thickness, n the Poisson ratio (n0.33), and  the mismatch ð ¼ 0:008Þ: On the other hand, the undulation amplitude Am cannot exceed the full length of the completely relaxed film, that is to say, for a period lm ð1 þ Þ: Using this equation, we found that theoretically the wavelength should have been lm ¼ 0:91 mm and Am ¼ 48 nm for a 30 nm thick 0.8%-strained In0.65Ga0.35As layer, and lm ¼ 1:25 mm and Am ¼ 33 nm for a 40 nm thick 0.8%-strained In0.65Ga0.35As/InAs0.25P0.75 bilayer. The experimental findings are thus significantly different from the expected ones even if they have the same magnitude. In our opinion, this discrepancy is

mainly due to the disregard for edge driven relaxation mode [21]. As a matter of fact, edges would have allowed the peripheral mesa parts to relax at a speed determined by the sticking layer viscosity. As a result, the undulation wavelength and amplitude would have become dependent on the sticking layer viscosity with the corollary that the lower the viscosity, the higher the edge-driven effect is expected on the relaxation mode. The experimental results seem to support this assumption as the disagreement is greater for the Apiezon wax case (viscosity 102 P) then for BPSG (viscosity 108 P). The next crucial issue is the use of such undulated thin films as seed layers to overgrowth materials, knowing that it seems quite impossible to engineer large membranes without any undulations. As mentioned above, the undulation wavelength is large and the amplitude rather small. Therefore, the resultant local curvature will produce only a relatively small strain-field modulation on surface. Probably too weak to induce perceptible effects on a subsequent lattice-matched

ARTICLE IN PRESS 166

M. Kostrzewa et al. / Journal of Crystal Growth 275 (2005) 157–166

overgrowth. It is certain that if lattice-matched binary III–V compounds ever existed, their overgrowth would probably flatten the seed layer surface and a high quality material would be obtained. But there are no such III–V materials which are lattice-matched with such seed substrates: only ternary or quaternary III–V alloys can actually be overgrown. Therefore, the main difficulty will be to limit as much as possible the alloy phase separation (an effect strongly responsive to local curvature) occurring for all ternary or quaternary III–V alloys. Both In0.65Ga0.35As and InAs0.25P0.75 can be grown lattice matched to InAs0.25P0.75 but InAs0.25P0.75 being a V-group ternary is less likely to amplify pre-existing roughness and thus more likely to smoothen the surface than In0.65Ga0.35As is, but on the other hand, its chemical instability at high temperature is a real handicap for BPSG flowing. To conclude these two preliminary studies have clearly shown the feasibility of the approach but also its technical limitations, especially when dealing with III–V semiconductors. In fact, the stability of all the materials and interfaces involved in the process have to be greatly improved from a chemical point of view before considering any practical application in optoelectronics.

Acknowledgements This paper is a part of a study dedicated to ‘‘compliant substrates for heteroepitaxy’’ partially supported by the ‘‘Re´gion Rhoˆne-Alpes’’ under Contracts 00815050 and 00815165. The authors thank A. Danescu and N. Mokni (ECL-LTDSCNRS 5513) for helpful discussions. References [1] Y.H. Lo, Appl. Phys. Lett. 59 (1991) 2311.

[2] F.E. Ejeckam, C.L. Chua, Z.H. Zhu, Y.H. Lo, M. Hong, R. Bhat, Appl. Phys. Lett. 67 (1995) 3963. [3] M.L. Seaford, P.J. Hesse, D.H. Tomich, K.G. Eying, J. Electron. Mater. 28 (1998) 878. [4] S. Rohart, G. Grenet, C. Priester, Appl. Surf. Sci. 188 (2002) 193. [5] C. Carter-Coman, R. Bicknell-Tassius, R.G. Benz, A.S. Brown, N.M. Jokerst, Appl. Phys. Lett. 70 (1997) 1754. [6] G. Ka¨stner, U¨. Go¨sele, J. Appl. Phys. 88 (2000) 4048. [7] A.S. Brown, W.A. Doolittle, Appl. Surf. Sci. 166 (2000) 392. [8] A. Bourret, Appl. Surf. Sci. 164 (2000) 3. [9] C.L. Chua, W.Y. Hsu, C.H. Liu, G. Christenson, Y.H. Lo, Appl. Phys. Lett. 64 (1994) 3640. [10] A.M. Jones, J.L. Jewell, J.C. Mabon, E.E. Reuter, S.G. Bishop, S.D. Roh, J.J. Coleman, Appl. Phys. Lett. 74 (1999) 1000. [11] J-F. Damlencourt, J-L. Leclercq, M. Gendry, P. Regreny, G. Hollinger, Appl. Phys. Lett. 75 (1999) 3638. [12] J-F. Damlencourt, J-L. Leclercq, M. Gendry, M. Garrigues, N. Aberkane, G. Hollinger, Jpn. J. Appl. Phys. 38 (1999) L996. [13] K.D. Hobbart, F.J. Kub, M. Fatemi, M.E. Twigg, P.E. Thompson, T.S. Kuan, C.K. Inoki, J. Electron. Mater. 29 (2000) 987. [14] N.P. Bansal, R.H. Doremus, Handbook of Glass Properties, Academic Press, New York, 1986. [15] N. Mokni, F. Sidoroff, A. Danescu, M. Kostrzewa, G. Grenet, 6e`me Congre`s de Me´canique, Tanger, Maroc, 2003. [16] M. Boudaa, Thesis, Ecole Centrale de Lyon, 2003. [17] M. Kostrzewa, P. Regreny, M.P. Besland, J.L. Leclercq, G. Grenet, P. Rojo-Romeo, G. Hollinger, E. Jalaguier, P. Perreau, H. Moriceau, O. Marty, IPRM 2003, Santa Barbara, Etats Unis, 2003. [18] E. Jalaguier, L. Di Cioccio, M. Kostrzewa, M. Zussy, P. Perreau, S. Pocas, N. Kernevez, C. Seassal, P. Regreny, P. Rojo-Romeo, M.P. Besland, J-L. Leclerc, G. Grenet, G. Hollinger, O. Marty, J.M. Chatelanaz, X. Wallaert, ICAST 2004, France, 2004. [19] R. Stengl, T. Tan, U. Gosele, Jpn. J. Appl. Phys. 28 (1989) 1735. [20] N. Sridhar, D.J. Srolovitz, Z. Suo, Appl. Phys. Lett. 78 (2001) 2482. [21] H. Yin, R. Huang, K.D. Hobart, Z. Suo, T.S. Kuan, C.K. Inoki, S.R. Shieh, T.S. Duffy, F.J. Kub, J.C. Sturm, J. Appl. Phys. 91 (2002) 9716; H. Yin, R. Huang, K.D. Hobart, Z. Suo, T.S. Kuan, C.K. Inoki, S.R. Shieh, T.S. Duffy, F.J. Kub, J.C. Sturm, J. Appl. Phys. 94 (2003) 6875.