IPRM 2003

Santa Barbara 12-16 May 2003

FEASIBILITY OF STRAIN RELAXED InAsP and InGaAs COMPLIANT SUBSTRATES M. Kostrzewa1 , G. Grenet1 , P. Regreny1, J. L. Leclercq1 , N. Mokni2, A. Danescu2, F. Sidoroff2 , E. Jalaguier3 , P. Perreau3 , H. Moriceau3 and G. Hollinger1 1

Ecole Centrale de Lyon, LEOM, (UMR CNRS 5512), 69134 Ecully cedex, France Ecole Centrale de Lyon, LTDS, (UMR CNRS 5513), 69134 Ecully cedex, France 3 CEA-DRT-LETI/DTS-CEA/GRE, 38054 Grenoble cedex 9, France

2

Abstract With a view to investigating the feasibility of using ultrathin films as compliant substrates, we present some preliminary results concerning InAs0.25 P0.75 (0.8% compressively - stressed on InP) film stuck onto a Si host substrate via borophosphorosilicate glass (BPSG). In an attempt to study relaxation mechanisms without any limitation on material viscosity, we also report on how a pseudomorphic In0.65Ga0.35 As layers (0.8% compressively-stressed on InP) elastically relaxes when stuck onto a Si host substrate via a thick Apiezon wax layer. It appears that uniform and flat in-plane elastic relaxation is actually possible only for samples of small areas. For larger samples, there is a competition between undulating and sliding as stress relaxation processes. However, we think that the resulting morphology in terms of undulation periodicity and amplitude is compatible with the use of such layers as seed layers for subsequent epitaxial overgrowths. I. Introduction A widely used method for preparing (by epitaxial growth technique) layers lattice-mismatched with their substrate is to insert a lattice-graded buffer layer between them. This process is known as the metamorphic approach. However, even if most of the stress is plastically relaxed in such a buffer layer through misfit dislocations, threading dislocations are not completely eliminated and limit the electrical and optical quality of the epitaxial layers thus obtained. It is why the concept of compliant substrates [1] was proposed by Lo in 1991 and has been developed since then by many groups. The key idea is the following: If the substrate is thin enough when compared with the overgrown epitaxial layer, the stress energy relaxation will induce defects in the former rather than in the latter. An alternate but closely related solution, is the paramorphic approach [2, 3]. In this case, the substrate is a thin pseudomorphic layer that is elastically relaxed before being used as a seed layer for the subsequent growth of a thick epitaxial layer. In both compliant and paramorphic approaches, the mechanical handling of ultra-thin membranes is a significant technological difficulty. Actually, these membranes have to be stuck onto a host substrate. One way to achieve this objective is to introduce a viscous layer such as borophosphorosilicate glass (BPSG) between the thin seed film and its host substrate, (usually a Si wafer). Such a viscous layer is believed to favour reversible elastic relaxation.

A few groups [4-6] have already studied the stress relaxation of semiconductor layers (mostly SiGe) using BPSG glass as a viscous layer (mostly by means of annealing process). They found that in-plane relaxation is hard to achieve and that undulations appear resulting from a compromise between the film elastic relaxation and the elastic stress associated with local curvatures. Our final goal is to use BPSG as viscous layer but there are difficulties in achieving a viscosity low enough to obtain elastic relaxation at the relatively low temperature required for III-V semiconductor growth. Therefore, as a preliminary step, we used Apiezon wax in order to evaluate the elastic relaxation without such a limitation on viscosity requirement. For that reason, this paper reports on results concerning InAs0.25 P0.75 and In 0.53 Ga 0.47As (both 0.8% compressivelystressed on InP) films stuck onto a Si host substrate via BPSG and via Apiezon wax, respectively. II. Experimental results The long-term objective of this work is to use waferbonding techniques to achieve compliant substrates with III-V semiconductors as seed layers. With this end in mind, we chose as relevant a 30nm thick InAs 0.25P0.75 film bonded onto Si host substrate via stabilized BPSG (SiO2 :83.27%, B2 O3 :14.11%, P2 O5 :2.62%). In our case, the initial semiconductor heterostructure is grown by Solid Source Molecular Beam Epitaxy (SSMBE) on an epi-ready InP(001) substrate. As indicated in Fig. 1, this heterostructure consists

IPRM 2003 of a 500nm thick In 0.53 Ga 0.47As layer (acting both as etch-stop and buffer layer) and of a 30nm thick InAs 0.25 P0.75 compressively-stressed (0.8%) film whose elastic relaxation we are currently interested in. The thickness of this layer is smaller than the critical thickness for plastic relaxation; therefore, no threading dislocation formation is expected in it. The heterostructure thus obtained is stuck upside down onto a silicon host substrate previously covered with a thick BPSG layer. Note that in order to reinforce the bonding, a 6-15nm thick silicon dioxide layer is deposited just before the bonding on III-V epitaxial heterostructure. The next technological step is the selective back-etching of the initial sacrificial InP substrate and of the In 0.53Ga 0.47 As layer. Finally, the heterostructure is patterned using the standard photolithography process, into mesas with edges orientated along the crystallographic orientations [110] and [1-10]. The BPSG viscosity is then progressively decreased via an annealing process.

(a) SiO2 (60-150Å) InAs 0.25P0.75 (300Å)

Santa Barbara 12-16 May 2003 that there is no possible comparison between standard temperatures (for either annealing or growth) usable for III-V alloys (480-520°C) and those usable for IV-IV alloys (750800°C). Obviously, the task requires more work before becoming operational. In particular, work is in progress to increase the B2 O3 concentration in BPSG in order to reduce its viscosity. Besides, to cope with the substantial chemical instability of the InAs0.25 P0.75 , an actual solution could be to protect it during the annealing by In 0.65 Ga 0.35As thin layer, which can be stabilised at 700°C under arsine pressure. Faced with this experimental difficulty, we decided to study what the elastic relaxation would be by simulating it using a very low viscosity material. The system under consideration is now an ultrathin pseudomorphic In 0.65 Ga 0.35As layer (0.8% compressively stressed on InP) stuck onto a Si host substrate not via BPSG but via a thick Apiezon wax layer. The Apiezon wax is a material of very low viscosity, which makes it possible to time-monitor the thin film morphological modifications at room temperature. As can be seen in Fig. 2. the technological method to transfer InGaAs strained layer onto a silicon substrate is analogous to the one described in Fig. 1. The only significant difference is

In0.65Ga 0.35As (5000Å) InP substrate

(a) In 0.65Ga 0.35As (300 Å)

(b)

InP (2000 Å) InP substrate

In0.53Ga 0.47As (5000 Å)

In0.65Ga 0.35As (5000Å) InAs 0.25P0.75 (300Å) BPSG layer

InP substrate

SiO2 (60-150Å)

(b)

Si host substrate

InP substrate

(c) InAs 0.25P0.75 (300Å)

Black wax

In0.53G a 0.47As (5000 Å)

BPSG layer

InP (2000 Å)

Si host substrate

In 0.65Ga

0.35As

(300 Å)

(20µm)

Fig. 1. Key technological steps (a) Heterostructure growth by MBE plus SiO2 layer (b) Bonding onto Si host substrate via BPSG (c) Back-etching of the sacrificial substrate and etch-stop layer We have evaluated that a viscosity around 108 Poise is required to obtain a BPSG viscosity low enough for our task. We have succeeded in preparing 2 inches InAsP/BPSG/Si using BPSG with bore concentration as high as 4.8%. For this BPSG composition, this would theoretically imply heating the samples up to 900°C. But in the course of a study by scanning microscopy the BPSG flowing on patterned silicon, it was found that in fact the potential temperature range extends over [700-900°C]. On the other hand, it has not been possible so far to increase the annealing temperature over 500°C in our MBE chamber without degrading the InAs0.25P0.75 layer. This is not sufficient yet to observe compliant effects. In this regard, note

Si host substrate

Camera

(c)

In 0.65Ga 0.35As (300 Å) Black wax Si host substrate

Fig. 2. Key technological steps (a) MBE growth and patterning into mesas (b) Bonding onto Si host substrate via preheated wax (c) Back-etching of the sacrificial substrate, etch-stop layer and last sacrificial layer that the patterning is done before and not after the sticking. There is also a slight change in the heterostructure which now

(b) [1 1 0]

[1 1 0]

(a)

20µm 20µm Fig. 3 Nomarski views of two In 0.65Ga 0.35 As mesas (a) 100x100µm mesa showing rather complete elastic relaxation (b) 150x150µm mesa showing incomplete elastic relaxation The In 0.65 Ga 0.35As film elastic relaxation was then timemonitored by a camera. Typical Nomarski optical images recorded after 9minutes at room temperature are shown in Fig. 3. It is clear from this figure that the mesa lateral dimensions are a critical issue. The 100x100µm2 (Fig. 3a) mesa seems almost completely relaxed by in-plane sliding whereas the relaxation of the 150x150µm2 mesa is clearly of another kind (Fig. 3b). Its corners look very much like the small mesa but other areas show undulations. The undulations periodically

b)

a)

20µm

20µm

Santa Barbara 12-16 May 2003 rumple the film perpendicularly to the edges when close enough to them but are mostly oriented along the {100} directions at the mesa centres. In order to have a closer look at these undulations, Atomic Force Microscopy (AFM) images were performed at the centre of a large mesa. The measurements were carried out on the film surface and on the underlying bare wax surface (after the thin In 0.65 Ga 0.35As film removal by chemical etching). Typical images are reported in Fig. 4 The morphologies on the thin film surface and on the wax surface reflect the same general trends but the features are a little more precise and detailed on the wax surface than on the film surface. Indeed, the imprint on the wax is directly moulded by the film stress relaxation while the morphology on the film surface is subject to modifications if the film comes unstuck. However, the undulation periodicity is almost the same, i.e., around 4.8µm on both the In 0.65Ga 0.35 As film surface and the bare wax surface, but the undulation amplitude (half the feature height) is slightly different, i.e. 0.090µm for the former and 0.065µm for the latter. Finally, in Fig. 5, we report the wavelength evolution on the heterostructure surface while the InP layer on top of the In 0.65 Ga 0.35As film is removed step-by-step. This wavelength gradually decreases as the InP layer thickness decreases but only for InP layer thickness less than 50nm. For thickness in the 50 -200nm range, the heterostructure is undulated but with a large wavelength (~16µm). 20 18

Undulation wavelength (µm)

IPRM 2003 consists of a 500nm thick In 0.53Ga 0.47As layer acting both as etch-stop and buffer layer, a 200nm thick InP layer and finally the relevant 30nm thick In 0.65Ga 0.35As compressively-stressed (0.8%) film. The additional InP layer role is to protect the In 0.65 Ga 0.35As layer during the selective back etching of the sacrificial In 0.53Ga 0.47As layer.

16 14 12 10 8 6 4 2

d)

c)

0

50

100

150

200

InP thickness (nm)

Fig. 5. Experimental undulation wavelength (dots) as a function of the InP film thickness on top of the In 0.65 Ga 0.35 As film. The line is just for visual help 5µm

5µm

Fig. 4. AFM images on (a,c) In 0.65 Ga 0.35As film surface (b,d) bare wax surface Some typical profile lines are shown below the corresponding images

III. Discussion Recently, using linear stability analysis and small amplitude approximation, Sridhar et al. [4] have shown that when a laterally-infinite compressively-stressed film (with mismatch ε, thickness h f , and Poisson ratio ν) is stuck onto a rigid substrate via a glass-like layer (with thickness hg and viscosity η) undulations appear as the result of a compromise between the film stress relaxation and the elastic stresses

IPRM 2003 associated with its local curvature. Taking a sinusoidal profile A(t) sin(kx) for the vertical displacement as a function of the horizontal x-axis and time t, they show that the undulation amplitude A(t) is fully governed by: A(t)=A(0) exp(α t) where the growth rate α has the following analytical expression: α = [sinh(2xg )-2xg ][βxf -xf3 ]/ [1+cosh(2xg )+2xf2 ] with xg =hg k , xf =hf k and β = 12ε (1+ν). As a result, there are two important wave numbers, namely, k c and km . The former k c corresponds to the wave number k beyond which α is negative and thus the undulation amplitude A(t) decays (stable regime). That is to say, for k ranging from 0 to k c the system presents undulations (unstable regime). This stability/instability threshold is given by k c = [h f2 β ] -1/2 The second wave number k m corresponds to the greatest growth rate and thus to the fastest developing undulation. Neither the critical wave number k c nor the optimum buckling wave number k m depend on the glass-like layer viscosity. Note that when h g >>h f , the optimum buckling wave number k m reaches its smallest value given by km = [3hf2 /β ] -1/2 and also that the corresponding asymptotic amplitude Am (t=∞) is given by [2hf 2/3] 1/2 . Using h g >>hf and t=∞ , a wavelength λm =2 π/km of 0.91µm and an amplitude Am of 0.025µm are thus theoretically expected for a 30nm thick In 0.65Ga 0.35As film (mismatch ε =0.008, ν = 0.3349). An experimental wavelength λ equal to 4.8µm implies a wave number k smaller than km indicating that extra relaxation has occurred. As a matter of fact, in this approach, a λm equal to 4.8µm would correspond to ε =0.0003, i.e. an almost completely relaxed mesa. At this point, it is worth noting that mesas cannot be considered laterally-infinite films as assumed in the Sridhar et al model. Indubitably, a sliding process starting from their edges can reach their centre part. Even if pure in-plane relaxation can hardly be fully achieved for large mesas, it can significantly lessen their stress energy leading to undulations with larger wavelength and smaller amplitude. We could thus define an experimental critical size below which the in-plane sliding is sufficient to produce an undulation weak enough for a subsequent overgrowth. In this case, a practical experimental solution would be a patterning of the initial paramorphic substrate into mesas with these sizes. Let us now analyse the experimental results reported in Fig. 5. concerning the elastic relaxation of an initial heterostructure made of a 30nm thick compressively stressed In 0.65 Ga 0.35As film plus the unstressed InP layer. As expected the wavelength increases and the amplitude decreases, when the InP layer thickness increases, just because the In 0.65 Ga 0.35As layer is less and less free to elastically relax. Besides, there are still undulations even for InP layer thickness as large as 200nm. This clearly indicates that, unlike

Santa Barbara 12-16 May 2003 what happens when using standard substrates, relaxation still goes on when using such a seed film as substrate. Even if the removal step-by-step of the InP layer gives some insights, it should be noted that in an actual process, the rational method would be to use such a paramorphic approach as an intermediate step before a new compressively stressed overgrowth. In this case, if elastically relaxed such a heterostructure will have a concave curvature whereas InP on In 0.65 Ga 0.35As layer has a convex curvature The most crucial issue to be discussed concerns the use of such undulated thin films as seed layers to overgrowth materials since it appears that it will be quite impossible to achieve elastically relaxed membranes without any undulations. The question is thus whether or not an epitaxial growth could be possible on such an undulating material. As mentioned above, the undulation wavelength is large and the amplitude extremely small. Therefore, the resulting local curvature will produce a rather minor surface strain field modulation probably too weak to induce perceptible effects on a subsequent lattice-matched overgrowth. It is certain that if a lattice-matched binary III-V compound ever existed, its overgrowth would make it possible to flatten the seed layer surface and thus to grow a high quality material. But it does not exist and only ternary or quaternary III-V alloys can actually be overgrown. Therefore, since III-V alloys are nonmiscible alloys, the main difficulty will be to limit the alloy phase separation, which is an effect strongly responsive to local curvature. For a lattice-mismatched overgrowth (compliant approach), the problem is even more complex because such a heterostructure will bend if stuck via a thick viscous layer. But on the other hand, this curvature will tend to relax the structure and thus to increase the undulation wavelength at the surface of the overgrowing film. In conclusion, this preliminary study corroborates the presumption that a paramorphic membrane could be used as seed layer for subsequent overgrowth. As a matter of fact, the inevitable undulations arising from elastic relaxation can lead to acceptable curvature for an epitaxy lattice-matched or mismatched overgrowth. To improve the latter point, the paramorphic membrane may eventually be patterned into mesas in order to facilitate additional in-plane relaxation. Acknowledgements This paper is a part of a study dedicated to “compliant substrates for heteroepitaxy” partially supported by the “Région Rhône-Alpes” under contracts 00815050 and 00815165. References [1] Y.H. Lo, Appl. Phys. Lett. 59, 2311 (1991) [2] J-F. Damlencourt and al.75, 3638, (1999) [3] J.F. Damlencourt and al. IPRM’00, WilliamsburgVirginia-USA, 14-18 Mai 2000 [4] N. Sridhar and al. Appl.Phys. Lett.78, 2482 (2001). [5] H. Yin and al., J. Appl. Phys. 91, 9716 (2002) [6] J. Liang and al. , Acta Materialia 50, 2933 (2002)