Home
Search
Collections
Journals
About
Contact us
My IOPscience
Locally measuring the adhesion of InP directly bonded on sub-100 nm patterned Si
This content has been downloaded from IOPscience. Please scroll down to see the full text. 2016 Nanotechnology 27 115707 (http://iopscience.iop.org/0957-4484/27/11/115707) View the table of contents for this issue, or go to the journal homepage for more
Download details: IP Address: 137.189.171.235 This content was downloaded on 18/02/2016 at 02:25
Please note that terms and conditions apply.
Nanotechnology Nanotechnology 27 (2016) 115707 (9pp)
doi:10.1088/0957-4484/27/11/115707
Locally measuring the adhesion of InP directly bonded on sub-100 nm patterned Si K Pantzas1,2, E Le Bourhis2, G Patriarche1, D Troadec3, G Beaudoin1, A Itawi1, I Sagnes1 and A Talneau1 1
Laboratoire de Photonique et des Nanostructures, CNRS UPR20, Route de Nozay, F-91460, Marcoussis, France 2 Institut P’, CNRS—Université de Poitiers—ENSMA—UPR 3346, SP2MI—Téléport 2 Bd Marie Pierre Curie, B.P. 30179, F-86962, Futuroscope Chasseneuil CEDEX, France 3 Institut d’Electronique, de Microélectronique et de Nanotechnologie, CNRS UMR8520, F-59652 Villeneuve Dascq, France E-mail: [email protected] Received 6 July 2015, revised 11 January 2016 Accepted for publication 21 January 2016 Published 16 February 2016 Abstract
A nano-scale analogue to the double cantilever experiment that combines instrumented nanoindentation and atomic force microscopy is used to precisely and locally measure the adhesion of InP bonded on sub-100 nm patterned Si using oxide-free or oxide-mediated bonding. Surfacebonding energies of 0.548 and 0.628 J m−2, respectively, are reported. These energies correspond in turn to 51% and 57% of the surface bonding energy measured in unpatterned regions on the same samples, i.e. the proportion of unetched Si surface in the patterned areas. The results show that bonding on patterned surfaces can be as robust as on unpatterned surfaces, provided care is taken with the post-patterning surface preparation process and, therefore, open the path towards innovative designs that include patterns embedded in the Si guiding layer of hybrid III-V/Si photonic integrated circuits. Keywords: direct bonding, photonic integration, instrumented nano-indentation, atomic force microscopy, scanning transmission electron microscopy (Some figures may appear in colour only in the online journal) 1. Introduction
unpatterned area can provide one with feedback useful for improving the patterning , post-patterning surface preparation, or bonding processes. Ultimately, such measurements can allow one to propose optimal designs for advanced embedded optical functions in hybrid III-V/Si PICs. The strength of adhesion between two materials is expressed in terms of a surface bonding energy, i.e. the energy per unit area to adiabatically and reversibly separate the two materials. A measure of the surface bonding energy can be obtained by deforming the materials until the interface yields. A variety of mechanical tests have been proposed over the years to measure this surface bonding energy. In the field of direct bonding of semiconductor materials, the doublecantilever beam (DCB) experiment is the most widely spread among these methods [4–8]. In the DCB experiment, a thin blade is inserted between two bonded wafers, propagating an interfacial debonding crack. Within the limits of linear
Direct bonding of III-V semiconductors on Si has recently become a promising technology for the fabrication of hybrid photonic integrated circuits (PICs) [1–3]. In such hybrid PICs, active optical functions (amplification, emission) are ensured by the direct-gap III-V semiconductor and passive functions (guiding, switching) by the Si guiding layer. One particularly attractive feature of hybrid III-V/Si platforms is the possibility to densely integrate a variety of advanced optical functions in the guiding layer using sub-100 nm patterns in the Si. Nano-patterning the Si surface may, however, weaken the adhesion of the III-V semiconductor to Si, potentially making the hybrid bonded stack subject to debonding in subsequent processing steps, such as cleaving. In this context, quantitatively assessing the adhesion of the III-V to Si in a patterned area and comparing it against the adhesion in a nearby 0957-4484/16/115707+09$33.00
1
© 2016 IOP Publishing Ltd Printed in the UK
Nanotechnology 27 (2016) 115707
K Pantzas et al
elasticity, the debonding crack length at fixed displacement provides a measure of the surface bonding energy. While this approach is well-suited to wafer-scale measurements of the surface bonding energy, it is less relevant if the Si surface is locally nano-patterned, as the DCB experiment provides an average measurement over a large area of the sample. Indeed, in typical experiments, the crack propagates over several tens of millimeters [7]. Such debonding crack lengths are much larger than individual components on a PIC, that are typically of the scale of a few hundred micrometers. Furthermore, this version of the DCB experiment requires the preparation of a dedicated sample that cannot be used for further processing after the measurement. A more localized measurement of adhesion that will preserve the hydbrid structure for ulterior processing steps would, therefore, be preferable. One such method was proposed in a recent letter [9]. In the letter, the InP membrane was shown to buckle during instrumented nano-indentation experiments. As a result, blisters form on the sides of the indent. An estimate of the surface bonding energy G can be obtained from the debonding-crack opening δ and debonding-crack length L of the blisters using the following equation: G=
3 ESi hSi3 ⎞ 3d 2 ⎛ EInP hInP ⎜ ⎟, 3 4 8L ⎝ EInP hInP + ESi hSi3 ⎠
where the perfectly sharp, the ratio Ap Au is none other than the filling factor of the pattern. Regardless of the ideality of the patterns, equation (2) indicates that surface bonding energy in a patterned region will be lower than in the unpatterned case. Given the relationship between the surface bonding energy and the debonding-crack length in equation (1), the blister is expected to propagate farther away from the center of the indent, and may therefore exceed typical sizes of STEM lamellae. Finally, the preparation of the lamella may relax strain in the blister. Hence, the ensuing measurement of the surface bonding energy may not accurately reflect the strength of bonding prior to the preparation of the lamella. To address these issues, atomic force microscopy (AFM) is proposed as an alternative to STEM to measure δ and L and deduce G in the present contribution.
2. Experiment For the purposes of the present investigation, Si samples were patterned with square arrays of holes. The holes had a diameter of 76 nm and a period of 146 nm, i.e. a filling factor of 21.3%. The patterns were obtained using standard e-beam lithography and chlorine-based ICP etching, conditions in which the Si immediately surrounding the patterns was rendered amorphous 25 nm from the edge of the pattern, as observed in STEM images of cross-sections of the patterns. The patterns were arranged in bands that were 25 μm wide and 500 μm long. The bands were placed 100 μm apart, providing sufficient space to isolate measurements on patterned regions from measurements on unpatterned regions. 10 mm by 5 mm InP samples were then bonded to the Si samples, a first one using bonding mediated through a 5 nm thin silicon-dioxide layer obtained through thermal oxidation of the Si surface [10] (oxide-mediated bonding), and a second one without any intermediate layer (oxide-free bonding) [11]. After both bonding experiments, the InP substrate and etchstop layers were chemically removed, leaving only a 365 nm thick InP membrane on the Si substrate. The InP membranes were then deformed in a Nanohardness tester from CSM Switzerland using a Berkovich diamond tip. The calibration procedure suggested by Oliver and Pharr [12] was used to correct for the load-frame compliance of the apparatus and the imperfections of the shape of the indenter tip. An indentation load of 10 mN was applied at the surface of the membranes. All AFM images were acquired using a Veeco Dimension V atomic force microscope. All images are 512 pixel by 512 pixel, for an image range of 6 μm by 6 μm. Therefore the pixel step is 15 nm. The precision along the Z direction is 1 Å, however, it is not possible to determine the blister height with that precision, as the RMS roughness of the chemically revealed surface of InP is 1 nm. The blister height hb and blister length Lb are determined from AFM profiles using the boundary conditions of the DCB problem [13]. The first
(1 )
where hInP (hSi ) and EInP (ESi ) represent the thickness and the Young modulus of the InP membrane (Si substrate), respectively. The membrane thickness hInP , debonding-crack opening δ, and debonding-crack length L were measured from cross-sections of the indented zone in a scanning transmission microscope (STEM). The blisters were shown to not extend farther than 2 μm from the center of the indent for small indentation loads [9]. This type of mechanical test is, therefore, much better suited for localized measurements of the surface bonding energy in III-V/Si hybrid bonded stacks. Precisely determining L from STEM cross-sections that are typically thinner than the patterns is, nonetheless, a challenge. Indeed, an inherent imprecision in the measurement arises from the fact that the debonding-crack tip may be located in the interstice between two adjacent patterns that are perpendicular to the lamella. During the fabrication of the lamella, the precise location of the crack tip may, therefore, be masked by the patterns. Quantitatively the imprecision would be close to the pattern period. In the case sub-100 nm patterns this may result in an error of as much as 10% the measurement of the debonding-crack length L, that in turn can give an error of as much as 40% in the value of G. Furthermore, blisters are expected to be larger in the case of InP bonded to patterned Si. Indeed, if the bonding is uniform across the patterned and unpatterned regions, the bonding energy measured in the patterned region Gp is linked to the bonding energy Gu in the unpatterned region through: Gp = Gu
Ap Au
,
(2 )
where Ap is the surface available for bonding after patterning, and Au is the total surface under the blister. In the ideal case, 2
Nanotechnology 27 (2016) 115707
K Pantzas et al
indents. Scanning transmission electron microscopy experiments were performed in an aberration-corrected JEOL 2200FS microscope, operating at 200 kV with a probe current of 150 pA, and a probe size of 0.12 nm at the full width at half maximum. The convergence half-angle of the probe was 30 mrad and the detection inner and outer half-angles for the HAADF-STEM images were 100 mrad and 170 mrad respectively. The cross-sections were imaged along the á110ñ zone axis. To accurately determine the debonding-crack length and debonding-crack opening, 4096 pixel by 4096 pixel STEM images were acquired at a magnification of 120000. The acquisition time was 80 s, during which no drift was observed in the image. In these conditions, the pixel step is 3 Å. The debonding-crack opening δ and debonding-crack length L are measured as follows: δ is taken as the opening between InP and Si at the last visible, indenter-induced defect in InP in a BF-STEM image such as the one in figure 2(a) (see inset a-1 for defect). Starting from this location, one can follow the opening until it disappears at the location a identified as the crack tip (inset a-2). As an example, profiles extracted from the BF-STEM image in figure 2(a) are given in in figures 2(b)–(d). The three plots show the STEM-intensity profile at the location where δ is measured, at the crack tip and beyond the crack tip, respectively. The distance between the profile where δ is measured and the last profile where the signal from the opening can be measured is taken as L.
Figure 1. The three plots correspond to (bottom) the Z profile of a
blister, (middle) the first, and (top) second derivative of Z with respect to X. These three plots allow one to determine hb and Lb. More details on the method are given in the text.
condition gives Z (X = X0 ) = hb at X0 such that: dZ (X = X 0) ¹ 0, dX
(3 )
d2Z (X = X 0) = 0. dX 2
(4 )
3. Results and discussion
The second condition gives L b = Xb - X0 , where Xb satisfies: dZ (X = Xb) = 0, dX
(5 )
d2Z (X = Xb) = 0. dX 2
(6 )
Nomarski images of the indents in the unpatterned and patterned regions of the oxide-free and oxide-mediated samples are shown in figures 3(a) and (b), respectively. As explained above, the blisters are larger in the patterned regions than in the unpatterned regions, although the size of the indents remains the same. The area of the blisters was measured from these Nomarski images for about ten indentations in each one of the four cases—InP bonded oxide-mediated or oxide-free to patterned or unpatterned Si. In all four cases, the standard deviation on the measured area was less than 5%, indicating that there is little dispersion in the energy measurements. Indeed, given that the debonding crack is approximately a quarter of the blister diameter, a 5% dispersion in the blister area amounts to about 1% dispersion in the surface bonding energy. AFM images of the blisters surrounding the indents are shown in figure 4. The four images correspond to the four configurations considered here: (a) InP bonded oxide-free to bare Si, (b) InP bonded oxide-free to patterned Si, (c) InP bonded oxide-mediated to bare Si, and, finally, (d) InP bonded oxide-mediated to patterned Si. In all four cases, the images reveal that what in figure 3 appears to be a single blister surrounding the indent, is in fact three distinct blisters. Each of the blisters is next to a facet of the Berkovich tip and has a distinct height hb and lateral extent Lb. The presence of these three distinct blisters is closely related to the mechanism that is responsible for the buckling. As explained in previous
The method is illustrated graphically in figure 1. Three stacked plots are displayed in the figure. The bottom plot is the Z profile extracted from the AFM image of the blister of InP bonded oxide-mediated to unpatterned Si. The middle and top plots are that of the first and second derivatives of Z with respect to X, respectively. Vertical dashed lines show the locations where the X0 and Xb are located.The precision with which X0 and Xb is related to the step between points in the curves of the derivatives, i.e. the lateral step of the AFM image. The points are linearly interpolated, resulting in an error of half the AFM step in the determines of X0 and Xb, i.e. 8 nm. Typical debonding crack lengths L = Xb - X0 in the samples studied here are greater than 800 nm. The overall error on the debonding crack length can therefore be determined to within a percent. The lamellae used for STEM were prepared using focused ion beam (FIB) etching and ion milling. Prior to FIB etching, the samples were coated with a 50 nm thick carbon layer, followed by a 100 nm thick silicon nitride layer, to preserve the sample surface and blister shape. The crosssections were carefully positioned to cross the center of the 3
Nanotechnology 27 (2016) 115707
K Pantzas et al
Figure 2. (a) BF-STEM image used to measure δ and L for InP bonded oxide-free to unpatterned Si. The image is annotated to show d , L , hb ,
and Lb. The inset shows the last indenter-induced defect visible in InP. This defect marks the boundary between the plastic and elastic region in InP. (b)–(d) Profiles showing the intensity of the bright-field signal. The profiles are taken along the á001ñ direction, i.e. perpendicular to the bonding interface. The first (b) is at taken at the location where δ is measured. The second, (c), shows the last profile where the opening between InP and Si is visible. The location of this last profile identifies the debonding-crack tip. The last profile, (d), shows the bright-field intensity beyond the crack tip.
Figure 3. Nomarski images of indents in the InP membranes bonded to the Si substrates using (a) oxide-free and (b) oxide-mediated bonding.
Blisters surrounding the indents can be seen in both cases. The blisters are larger in the regions where the Si is patterned.
work [9, 14], the InP membrane buckles under the effect of shear stress accumulating in the membrane during the indentation experiment. The shear stress itself is induced by torque from the edge component of the dislocations that the indenter tip introduces as it penetrates the InP membrane and that do not cross the InP/Si boundary. In InP, dislocations slip along {111} planes and, therefore, the shear stress is projected along the á110ñ direction. Hence, the smaller the angle θ between the normal to one of the facets and the á110ñ direction, the higher the shear stress becomes, propagating the debonding crack length L farther away from the indenter tip. The blister will reach its equilibrium height and length when sufficient energy is released from the debonding crack to accommodate the shear stress applied by a given facet of the indenter. Assuming that the bonding is uniform in the
immediate area surrounding the indented zone, the blister height and length couples from the three blisters surrounding the indent will give the same energy G. Prior to computing G using equation (1), however, one needs to establish two relationships: the first between the blister height hb and the debonding crack opening δ, and the second between the lateral extent of the blister Lb and the debonding crack length L. Cross-sections of the blisters shown in figure 4 were prepared and observed in a scanning transmission electron microscope. Low-magnification bright-field STEM micrographs of the four cross-sections are shown in figure 5. The cross-sections reveal two distinct regions in the InP membrane around the indent. In the first, the InP membrane is plastically deformed by the indenter. In the second, the membrane is intact and elastically debonded from the Si 4
Nanotechnology 27 (2016) 115707
K Pantzas et al
Figure 4. AFM images (6×6 μm) of debonding blisters in the case of InP bonded (a) oxide-free on bare Si, (b) oxide-free to patterned Si, (c)
oxide-mediated to bare Si, and (d) oxide-mediated to patterned Si. The height scale for all images is 55 nm.
and oxide-mediated bonding, δ and hb are in excellent agreement, indicating that the hb can be used in equation (1) instead of δ. The lateral extent of the blister Lb is, however, 70%–80% longer than the debonding crack L. There are two possible explanations for this discrepancy. The first is that elastic strain in the InP membrane is partially relaxed during the fabrication of the lamella. The blister hb and δ do not differ significantly, however, indicating that no relaxation occurs during the fabrication process. Furthermore, the AFM profile of the InP membrane prior to the fabrication of the lamella matches well the profile of the surface of InP extracted from the STEM images (for an example see figure 6(a)). The second possibility is that the InP membrane is elastically deformed beyond the crack tip. In this case, the difference between Lb and L accounts for the range over which the InP membrane is elastically deformed beyond the crack tip. This second hypothesis is supported by the
substrate. It is this second region that fulfills the conditions necessary to apply equation (1): close to the indent, the InP membrane is maintained suspended at height δ by the plastically deformed portion of the membrane, any stresses between the two being relaxed through the dislocations of the latter. Far from the indent, the InP membrane is firmly bonded to the Si substrate, as revealed in the atomically resolved images of the InP/Si shown in the insets. Between these two extremes, the InP membrane is elastically deformed. The debonding crack propagates only along the InP/Si interface (InP/SiO2 interface in the oxide-mediated bonding case). To establish the relationship between the embedded crack and the blister observed on the surface of InP, the values of (d , L ), measured from STEM images, are compared against those of (d , L ), measured from AFM profiles at the location where the lamellae were prepared in the unpatterned cases. The measured values are summarized in table 1. In both oxide-free 5
Nanotechnology 27 (2016) 115707
K Pantzas et al
Figure 5. Cross-sections of the four indents observed in AFM in figure 5. Insets show the InP/Si interface beyond the debonding-crack tip, where InP is firmly bonded to the underlying Si substrate.
to model it and propose corrections to equation (1). Among those models, the semi-empirical formula of Gillis and Gillman [15]:
Table 1. Measured values for the crack opening (δ), crack length (L),
blister height (hb), and blister lateral extent (Lb), for InP bonded oxide-free to unpatterned Si (OFU), oxide-free to patterned Si (OFP), oxide-mediated to unpatterned Si (OMU), and oxidemediated to patterned Si (OMP). While the measured blister heights and crack are in good agreement, the debonding crack length is found to be 40% shorter than the lateral extent of the blister, on average. This discrepancy is explained in the text.
OFU OFP OMU OMP Precision
δ nm
L μm
hb nm
Lb μm
DL nm
45.4 — 38.5 — ±0.3
0.9632 — 0.8761 — ±0.3
45.1 36 38.2 35.3 ±1
1.627 1.677 1.585 1.625 ±15
664
G=
2 3 + FSi2 ESi hSi3 EInP hInP 3d 2 FInP , 3 3 4 8L (FInP EInP hInP + FSi3 ESi hSi3 )2
(7 )
where Fi = 1 + 3c
h2 hi 3 + (1 + ni ) i2 , L 5 L
(8 )
has been known to yield reliable results in the case of Silicon h2 3 to Silicon bonding [7]. In equation (8), the term 5 (1 + n ) Li2 accounts for shear stress perpendicular to the direction of propagation of the debonding crack, and the term 3c hLi accounts for the rotation of the built-in section that is due to elastic deformation of the DCB beyond the crack tip. In this second term, c is a constant that has been found through experimental adjustment to be equal to 0.63 [13]. Based on this model and given that hSi hInP , a first order approximation of the difference between L and Lb is 3chInP , or 654 nm, a value that is in reasonable agreement with the
709 ±15
observation of deformation fringes that extend beyond the crack tip in the BF-STEM image presented in figure 6(b). Elastic deformation of the beams beyond the crack tip in the DCB experiment is well-established in the literature, and a variety of models have been developed over the years [15–17] 6
Nanotechnology 27 (2016) 115707
K Pantzas et al
Figure 6. (a) Superposition of the AFM and STEM profiles from the surface of InP bonded oxide-mediated to unpatterned Si. The two
profiles are in excellent agreement, indicating that there is no relaxation of the InP membrane during the fabrication of the STEM lamella. (b) BF-STEM of an InP membrane bonded oxide-free to Si. Deformation fringes that extend ∼600 nm beyond the crack tip are visible in this image. The fringes are visible in this image and not others throughout the article as this lamella was prepared specifically in conditions to minimize amorphization in the sidewalls and is thinner than the other lamellae.
measured values of DL in table 1. Given these observations, we assume L to be equal to L b - 3chInP . It is now possible to compute G using equation (7). The results are reported in table 2. Values of the surface bonding energy computed for the same values of δ and L and using equation (1) are also reported in the table. In the case of bonding to unpatterned Si, the values reported for the Maszara model are higher than those reported previously in the literature for direct bonding of InP to Si and using the same model [6, 8, 9]. These values are all higher than the fracture energy of InP [6]—1.5 J m−2, although there is no evidence of fractures or plastic deformation in the InP membrane. The values of G obtained using the model of Gillis and Gilman are, therefore, considered to give a more accurate estimate of the adherence of InP to Si. Using this model, the surface bonding energy for both oxide-free and oxide-mediated bonding are ∼1 J m−2. While this value is already sufficient to ensure a mechanically robust bonding—indeed, values of this order of magnitude are similar to the cohesion energy of bulk iono-covalent crystals [13]—there is still room for improving the bonding process. Such an improvement is typically achieved by annealing the bonded stack at higher temperatures [5–8]. High-temperature anneals may, however, be undesirable from the technological point of view, in particular in the context of CMOS-compatible integration. Table 2 also reports the ratio Gp Gu of the surface bonding energy on patterned Si to that on bare Si. This ratio is 51% in the case of oxide-free bonding and 57% in the case of oxide-free bonding. These two percentages correspond to the actual surface of Si available for bonding. In both cases, the ratio is lower than the reciprocal of the filling factor of the patterns—78.7 %. This discrepancy is related to the fact that the edges of the patterns are rounded off during the patterning process. Bonding only occurs in the middle between patterns,
Table 2. Surface bonding energies deduced for InP bonded oxide-
free to unpatterned Si (OFU), oxide-free to patterned Si (OFP), oxide-mediated to unpatterned Si (OMU), and oxide-mediated to patterned Si (OMP). Values computed both using the semi-empirical formula Gillis and Gilman [15] and analytic expression by Maszara [4] are reported in the table. As the InP membrane is elastically deformed beyond the crack tip, the values given by the semiempirical formula of Gillis and Gilman are expected to give a more accurate estimate of the adhesion of InP to Si. The table also reports the ratio Gp Gu of the surface bonding energy on patterned Si to that on bare Si.
OFU OFP OMU OMP Precision
GGG J m−2
GM J m−2
1.07 0.548 1.04 0.628 ±0.04
2.22 1.13 2.36 1.34 ±0.09
Gp Gu %
51 57 ±1
where the surface of Si has not been affected by the patterning process and can still be in contact with InP. STEM images reveal that this region correlates well with the portion of Si that has not been rendered amorphous during the patterning process. These observations are summarized in figure 7. The slight difference between oxide-free and oxide-mediated bonding—6%—can be attributed to the fact that the thermal silicon dioxide layer used in the case of oxide-mediated bonding slightly planarizes the Si surface. These results show that to obtain more robust bonding on the patterned areas, one needs to produce perfectly sharp emerging edge holes. Such an optimization could be achieved using a pulsed-plasma ICP etching process, such as the one proposed in the literature for improved etched profiles of silicon wires [18]. 7
Nanotechnology 27 (2016) 115707
K Pantzas et al
Figure 7. (a) AFM image of the Si surface after patterning (color scale 0–85 nm). (b) The same image filtered to show only heights from
83–85 nm, corresponding to the actual bonded surface. (c) AFM profile along the dashed red line in (a) showing the curvature of the Si surface (red line) and the portion of the surface that InP adheres to (green dashed fill). (d) BF-STEM image at InP/Si interface (oxide-free case). (e) The same image overlayed with colors corresponding to (red) InP, (green) crystalline Si, and (blue) amorphous Si. The overlays were obtained by combining EDX and diffraction data from the image in (d)—color online.
8
Nanotechnology 27 (2016) 115707
K Pantzas et al
4. Conclusion
References
In conclusion, the method presented in the present contribution that combines instrumented nano-indentation with AFM has been shown to provide precise, highly localized measurements of the surface bonding energy of InP to sub100 nm patterned Si. Using this method, one receives feedback on both the quality of the bonding process and to what extent this process has been affected by the patterning of the underlying Si. This feedback allows one to identify and individually optimize the conditions for the patterning and bonding processes and design optimal experiments that produce mechanically robust InP/Si hybrid stacks for PICs. The underlying mechanism responsible for the debonding is tied to mechanical properties that InP shares with other III-V semiconductors. The method presented here can, therefore, be adapted to other technologically relevant III-V/Si hybrid bonded stacks, such as GaAs/Si [19]. Furthermore, measurements can be carried out at several locations across a wafer, allowing one to collect statistics, compare the results to large-area measurements like the ones from the DCB experiment, and evaluate the homogeneity of the bonding process. The precision of the measurement, while already relatively high, can be further improved by reducing the area scanned in AFM to only the elastically debonded area and increasing the sampling rate. Furthermore, more precise modeling of the debonded surface may also help refine the models discussed here and ultimately yield more precise measurements.
[1] Tong Q Y, Gan Q, Hudson G, Fountain G and Enquist P 2004 Appl. Phys. Lett. 84 732–4 [2] Jain S R, Sysak N N, Kurczveil G and Bowers J E 2011 Opt. Express 19 13692–9 [3] Mechet P, Verstuyft S, de Vries T, Spuesens T, Regreny P, Thourhout D V, Roelkens G a and Morthier G 2013 Opt. Express 21 19339–52 [4] Maszara P, Goetz G, Caviglia A and Mckitterick J B 1988 J. Appl. Phys. 64 4943–50 [5] Tong Q and Gösele U 1999 Semiconductor Wafer Bonding: Science and Technology (The ECS Series of Texts and Monographs) (New York: Wiley) [6] Pasquariello D, Camacho M, Hjort K, Hjort K, Dosza L and Szentpali B 2001 Mater. Sci. Eng. B 80 164–7 [7] Fournel F, Continni L, Morales C, Da Fonseca J, Moriceau H, Rieutord F, Barthelemy A and Radu I 2012 J. Appl. Phys. 111 104907 [8] Pasquariello D and Hjort K 2000 J. Electrochem. Soc. 147 2343–6 [9] Pantzas K, Patriarche G, Le Bourhis E, Troadec D, Itawi A, Beaudoin G, Sagnes I and Talneau A 2013 Appl. Phys. Lett. 103 081901 [10] Itawi A, Pantzas K, Sagnes I, Patriarche G and Talneau A 2014 J. Vac. Sci. Technol. B 32 021201 [11] Talneau A, Roblin C, Itawi A, Mauguin O, Largeau L, Beaudouin G, Sagnes I, Patriarche G, Pang C and Benisty H 2013 Appl. Phys. Lett. 102 212101 [12] Oliver W C and Pharr G M 1992 J. Mater. Res. 7 1564–83 [13] Maugis D 1997 Contact, Adhesion and Rupture of Elastic Solids (Berlin: Springer) [14] Pantzas K, Le Bourhis E, Patriarche G, Itawi A, Beaudoin G, Sagnes I and Talneau A 2014 Eur. Phys. J. Appl. Phys. 65 20702 [15] Gillis P and Gilman J 1964 J. Appl. Phys. 35 647–58 [16] Kanninen M 1973 Int. J. Fract. 9 83–92 [17] Foote R and Buchwald V 1985 Int. J. Fract. 29 125–34 [18] Moritz H, Darnon M, Cunge G and Joubert O 2015 J. Vac. Sci. Technol. B 33 032203 [19] Tanabe K, Watanabe K and Arakawa Y 2012 Sci. Rep. 2 1–6
Acknowledgments The authors would like to gratefully acknowledge funding from the CNRS RENATECH network and the Agence Nationale de la Recherche P2N 2012 grant COHEDIO.
9
Search
Collections
Journals
About
Contact us
My IOPscience
Locally measuring the adhesion of InP directly bonded on sub-100 nm patterned Si
This content has been downloaded from IOPscience. Please scroll down to see the full text. 2016 Nanotechnology 27 115707 (http://iopscience.iop.org/0957-4484/27/11/115707) View the table of contents for this issue, or go to the journal homepage for more
Download details: IP Address: 137.189.171.235 This content was downloaded on 18/02/2016 at 02:25
Please note that terms and conditions apply.
Nanotechnology Nanotechnology 27 (2016) 115707 (9pp)
doi:10.1088/0957-4484/27/11/115707
Locally measuring the adhesion of InP directly bonded on sub-100 nm patterned Si K Pantzas1,2, E Le Bourhis2, G Patriarche1, D Troadec3, G Beaudoin1, A Itawi1, I Sagnes1 and A Talneau1 1
Laboratoire de Photonique et des Nanostructures, CNRS UPR20, Route de Nozay, F-91460, Marcoussis, France 2 Institut P’, CNRS—Université de Poitiers—ENSMA—UPR 3346, SP2MI—Téléport 2 Bd Marie Pierre Curie, B.P. 30179, F-86962, Futuroscope Chasseneuil CEDEX, France 3 Institut d’Electronique, de Microélectronique et de Nanotechnologie, CNRS UMR8520, F-59652 Villeneuve Dascq, France E-mail: [email protected] Received 6 July 2015, revised 11 January 2016 Accepted for publication 21 January 2016 Published 16 February 2016 Abstract
A nano-scale analogue to the double cantilever experiment that combines instrumented nanoindentation and atomic force microscopy is used to precisely and locally measure the adhesion of InP bonded on sub-100 nm patterned Si using oxide-free or oxide-mediated bonding. Surfacebonding energies of 0.548 and 0.628 J m−2, respectively, are reported. These energies correspond in turn to 51% and 57% of the surface bonding energy measured in unpatterned regions on the same samples, i.e. the proportion of unetched Si surface in the patterned areas. The results show that bonding on patterned surfaces can be as robust as on unpatterned surfaces, provided care is taken with the post-patterning surface preparation process and, therefore, open the path towards innovative designs that include patterns embedded in the Si guiding layer of hybrid III-V/Si photonic integrated circuits. Keywords: direct bonding, photonic integration, instrumented nano-indentation, atomic force microscopy, scanning transmission electron microscopy (Some figures may appear in colour only in the online journal) 1. Introduction
unpatterned area can provide one with feedback useful for improving the patterning , post-patterning surface preparation, or bonding processes. Ultimately, such measurements can allow one to propose optimal designs for advanced embedded optical functions in hybrid III-V/Si PICs. The strength of adhesion between two materials is expressed in terms of a surface bonding energy, i.e. the energy per unit area to adiabatically and reversibly separate the two materials. A measure of the surface bonding energy can be obtained by deforming the materials until the interface yields. A variety of mechanical tests have been proposed over the years to measure this surface bonding energy. In the field of direct bonding of semiconductor materials, the doublecantilever beam (DCB) experiment is the most widely spread among these methods [4–8]. In the DCB experiment, a thin blade is inserted between two bonded wafers, propagating an interfacial debonding crack. Within the limits of linear
Direct bonding of III-V semiconductors on Si has recently become a promising technology for the fabrication of hybrid photonic integrated circuits (PICs) [1–3]. In such hybrid PICs, active optical functions (amplification, emission) are ensured by the direct-gap III-V semiconductor and passive functions (guiding, switching) by the Si guiding layer. One particularly attractive feature of hybrid III-V/Si platforms is the possibility to densely integrate a variety of advanced optical functions in the guiding layer using sub-100 nm patterns in the Si. Nano-patterning the Si surface may, however, weaken the adhesion of the III-V semiconductor to Si, potentially making the hybrid bonded stack subject to debonding in subsequent processing steps, such as cleaving. In this context, quantitatively assessing the adhesion of the III-V to Si in a patterned area and comparing it against the adhesion in a nearby 0957-4484/16/115707+09$33.00
1
© 2016 IOP Publishing Ltd Printed in the UK
Nanotechnology 27 (2016) 115707
K Pantzas et al
elasticity, the debonding crack length at fixed displacement provides a measure of the surface bonding energy. While this approach is well-suited to wafer-scale measurements of the surface bonding energy, it is less relevant if the Si surface is locally nano-patterned, as the DCB experiment provides an average measurement over a large area of the sample. Indeed, in typical experiments, the crack propagates over several tens of millimeters [7]. Such debonding crack lengths are much larger than individual components on a PIC, that are typically of the scale of a few hundred micrometers. Furthermore, this version of the DCB experiment requires the preparation of a dedicated sample that cannot be used for further processing after the measurement. A more localized measurement of adhesion that will preserve the hydbrid structure for ulterior processing steps would, therefore, be preferable. One such method was proposed in a recent letter [9]. In the letter, the InP membrane was shown to buckle during instrumented nano-indentation experiments. As a result, blisters form on the sides of the indent. An estimate of the surface bonding energy G can be obtained from the debonding-crack opening δ and debonding-crack length L of the blisters using the following equation: G=
3 ESi hSi3 ⎞ 3d 2 ⎛ EInP hInP ⎜ ⎟, 3 4 8L ⎝ EInP hInP + ESi hSi3 ⎠
where the perfectly sharp, the ratio Ap Au is none other than the filling factor of the pattern. Regardless of the ideality of the patterns, equation (2) indicates that surface bonding energy in a patterned region will be lower than in the unpatterned case. Given the relationship between the surface bonding energy and the debonding-crack length in equation (1), the blister is expected to propagate farther away from the center of the indent, and may therefore exceed typical sizes of STEM lamellae. Finally, the preparation of the lamella may relax strain in the blister. Hence, the ensuing measurement of the surface bonding energy may not accurately reflect the strength of bonding prior to the preparation of the lamella. To address these issues, atomic force microscopy (AFM) is proposed as an alternative to STEM to measure δ and L and deduce G in the present contribution.
2. Experiment For the purposes of the present investigation, Si samples were patterned with square arrays of holes. The holes had a diameter of 76 nm and a period of 146 nm, i.e. a filling factor of 21.3%. The patterns were obtained using standard e-beam lithography and chlorine-based ICP etching, conditions in which the Si immediately surrounding the patterns was rendered amorphous 25 nm from the edge of the pattern, as observed in STEM images of cross-sections of the patterns. The patterns were arranged in bands that were 25 μm wide and 500 μm long. The bands were placed 100 μm apart, providing sufficient space to isolate measurements on patterned regions from measurements on unpatterned regions. 10 mm by 5 mm InP samples were then bonded to the Si samples, a first one using bonding mediated through a 5 nm thin silicon-dioxide layer obtained through thermal oxidation of the Si surface [10] (oxide-mediated bonding), and a second one without any intermediate layer (oxide-free bonding) [11]. After both bonding experiments, the InP substrate and etchstop layers were chemically removed, leaving only a 365 nm thick InP membrane on the Si substrate. The InP membranes were then deformed in a Nanohardness tester from CSM Switzerland using a Berkovich diamond tip. The calibration procedure suggested by Oliver and Pharr [12] was used to correct for the load-frame compliance of the apparatus and the imperfections of the shape of the indenter tip. An indentation load of 10 mN was applied at the surface of the membranes. All AFM images were acquired using a Veeco Dimension V atomic force microscope. All images are 512 pixel by 512 pixel, for an image range of 6 μm by 6 μm. Therefore the pixel step is 15 nm. The precision along the Z direction is 1 Å, however, it is not possible to determine the blister height with that precision, as the RMS roughness of the chemically revealed surface of InP is 1 nm. The blister height hb and blister length Lb are determined from AFM profiles using the boundary conditions of the DCB problem [13]. The first
(1 )
where hInP (hSi ) and EInP (ESi ) represent the thickness and the Young modulus of the InP membrane (Si substrate), respectively. The membrane thickness hInP , debonding-crack opening δ, and debonding-crack length L were measured from cross-sections of the indented zone in a scanning transmission microscope (STEM). The blisters were shown to not extend farther than 2 μm from the center of the indent for small indentation loads [9]. This type of mechanical test is, therefore, much better suited for localized measurements of the surface bonding energy in III-V/Si hybrid bonded stacks. Precisely determining L from STEM cross-sections that are typically thinner than the patterns is, nonetheless, a challenge. Indeed, an inherent imprecision in the measurement arises from the fact that the debonding-crack tip may be located in the interstice between two adjacent patterns that are perpendicular to the lamella. During the fabrication of the lamella, the precise location of the crack tip may, therefore, be masked by the patterns. Quantitatively the imprecision would be close to the pattern period. In the case sub-100 nm patterns this may result in an error of as much as 10% the measurement of the debonding-crack length L, that in turn can give an error of as much as 40% in the value of G. Furthermore, blisters are expected to be larger in the case of InP bonded to patterned Si. Indeed, if the bonding is uniform across the patterned and unpatterned regions, the bonding energy measured in the patterned region Gp is linked to the bonding energy Gu in the unpatterned region through: Gp = Gu
Ap Au
,
(2 )
where Ap is the surface available for bonding after patterning, and Au is the total surface under the blister. In the ideal case, 2
Nanotechnology 27 (2016) 115707
K Pantzas et al
indents. Scanning transmission electron microscopy experiments were performed in an aberration-corrected JEOL 2200FS microscope, operating at 200 kV with a probe current of 150 pA, and a probe size of 0.12 nm at the full width at half maximum. The convergence half-angle of the probe was 30 mrad and the detection inner and outer half-angles for the HAADF-STEM images were 100 mrad and 170 mrad respectively. The cross-sections were imaged along the á110ñ zone axis. To accurately determine the debonding-crack length and debonding-crack opening, 4096 pixel by 4096 pixel STEM images were acquired at a magnification of 120000. The acquisition time was 80 s, during which no drift was observed in the image. In these conditions, the pixel step is 3 Å. The debonding-crack opening δ and debonding-crack length L are measured as follows: δ is taken as the opening between InP and Si at the last visible, indenter-induced defect in InP in a BF-STEM image such as the one in figure 2(a) (see inset a-1 for defect). Starting from this location, one can follow the opening until it disappears at the location a identified as the crack tip (inset a-2). As an example, profiles extracted from the BF-STEM image in figure 2(a) are given in in figures 2(b)–(d). The three plots show the STEM-intensity profile at the location where δ is measured, at the crack tip and beyond the crack tip, respectively. The distance between the profile where δ is measured and the last profile where the signal from the opening can be measured is taken as L.
Figure 1. The three plots correspond to (bottom) the Z profile of a
blister, (middle) the first, and (top) second derivative of Z with respect to X. These three plots allow one to determine hb and Lb. More details on the method are given in the text.
condition gives Z (X = X0 ) = hb at X0 such that: dZ (X = X 0) ¹ 0, dX
(3 )
d2Z (X = X 0) = 0. dX 2
(4 )
3. Results and discussion
The second condition gives L b = Xb - X0 , where Xb satisfies: dZ (X = Xb) = 0, dX
(5 )
d2Z (X = Xb) = 0. dX 2
(6 )
Nomarski images of the indents in the unpatterned and patterned regions of the oxide-free and oxide-mediated samples are shown in figures 3(a) and (b), respectively. As explained above, the blisters are larger in the patterned regions than in the unpatterned regions, although the size of the indents remains the same. The area of the blisters was measured from these Nomarski images for about ten indentations in each one of the four cases—InP bonded oxide-mediated or oxide-free to patterned or unpatterned Si. In all four cases, the standard deviation on the measured area was less than 5%, indicating that there is little dispersion in the energy measurements. Indeed, given that the debonding crack is approximately a quarter of the blister diameter, a 5% dispersion in the blister area amounts to about 1% dispersion in the surface bonding energy. AFM images of the blisters surrounding the indents are shown in figure 4. The four images correspond to the four configurations considered here: (a) InP bonded oxide-free to bare Si, (b) InP bonded oxide-free to patterned Si, (c) InP bonded oxide-mediated to bare Si, and, finally, (d) InP bonded oxide-mediated to patterned Si. In all four cases, the images reveal that what in figure 3 appears to be a single blister surrounding the indent, is in fact three distinct blisters. Each of the blisters is next to a facet of the Berkovich tip and has a distinct height hb and lateral extent Lb. The presence of these three distinct blisters is closely related to the mechanism that is responsible for the buckling. As explained in previous
The method is illustrated graphically in figure 1. Three stacked plots are displayed in the figure. The bottom plot is the Z profile extracted from the AFM image of the blister of InP bonded oxide-mediated to unpatterned Si. The middle and top plots are that of the first and second derivatives of Z with respect to X, respectively. Vertical dashed lines show the locations where the X0 and Xb are located.The precision with which X0 and Xb is related to the step between points in the curves of the derivatives, i.e. the lateral step of the AFM image. The points are linearly interpolated, resulting in an error of half the AFM step in the determines of X0 and Xb, i.e. 8 nm. Typical debonding crack lengths L = Xb - X0 in the samples studied here are greater than 800 nm. The overall error on the debonding crack length can therefore be determined to within a percent. The lamellae used for STEM were prepared using focused ion beam (FIB) etching and ion milling. Prior to FIB etching, the samples were coated with a 50 nm thick carbon layer, followed by a 100 nm thick silicon nitride layer, to preserve the sample surface and blister shape. The crosssections were carefully positioned to cross the center of the 3
Nanotechnology 27 (2016) 115707
K Pantzas et al
Figure 2. (a) BF-STEM image used to measure δ and L for InP bonded oxide-free to unpatterned Si. The image is annotated to show d , L , hb ,
and Lb. The inset shows the last indenter-induced defect visible in InP. This defect marks the boundary between the plastic and elastic region in InP. (b)–(d) Profiles showing the intensity of the bright-field signal. The profiles are taken along the á001ñ direction, i.e. perpendicular to the bonding interface. The first (b) is at taken at the location where δ is measured. The second, (c), shows the last profile where the opening between InP and Si is visible. The location of this last profile identifies the debonding-crack tip. The last profile, (d), shows the bright-field intensity beyond the crack tip.
Figure 3. Nomarski images of indents in the InP membranes bonded to the Si substrates using (a) oxide-free and (b) oxide-mediated bonding.
Blisters surrounding the indents can be seen in both cases. The blisters are larger in the regions where the Si is patterned.
work [9, 14], the InP membrane buckles under the effect of shear stress accumulating in the membrane during the indentation experiment. The shear stress itself is induced by torque from the edge component of the dislocations that the indenter tip introduces as it penetrates the InP membrane and that do not cross the InP/Si boundary. In InP, dislocations slip along {111} planes and, therefore, the shear stress is projected along the á110ñ direction. Hence, the smaller the angle θ between the normal to one of the facets and the á110ñ direction, the higher the shear stress becomes, propagating the debonding crack length L farther away from the indenter tip. The blister will reach its equilibrium height and length when sufficient energy is released from the debonding crack to accommodate the shear stress applied by a given facet of the indenter. Assuming that the bonding is uniform in the
immediate area surrounding the indented zone, the blister height and length couples from the three blisters surrounding the indent will give the same energy G. Prior to computing G using equation (1), however, one needs to establish two relationships: the first between the blister height hb and the debonding crack opening δ, and the second between the lateral extent of the blister Lb and the debonding crack length L. Cross-sections of the blisters shown in figure 4 were prepared and observed in a scanning transmission electron microscope. Low-magnification bright-field STEM micrographs of the four cross-sections are shown in figure 5. The cross-sections reveal two distinct regions in the InP membrane around the indent. In the first, the InP membrane is plastically deformed by the indenter. In the second, the membrane is intact and elastically debonded from the Si 4
Nanotechnology 27 (2016) 115707
K Pantzas et al
Figure 4. AFM images (6×6 μm) of debonding blisters in the case of InP bonded (a) oxide-free on bare Si, (b) oxide-free to patterned Si, (c)
oxide-mediated to bare Si, and (d) oxide-mediated to patterned Si. The height scale for all images is 55 nm.
and oxide-mediated bonding, δ and hb are in excellent agreement, indicating that the hb can be used in equation (1) instead of δ. The lateral extent of the blister Lb is, however, 70%–80% longer than the debonding crack L. There are two possible explanations for this discrepancy. The first is that elastic strain in the InP membrane is partially relaxed during the fabrication of the lamella. The blister hb and δ do not differ significantly, however, indicating that no relaxation occurs during the fabrication process. Furthermore, the AFM profile of the InP membrane prior to the fabrication of the lamella matches well the profile of the surface of InP extracted from the STEM images (for an example see figure 6(a)). The second possibility is that the InP membrane is elastically deformed beyond the crack tip. In this case, the difference between Lb and L accounts for the range over which the InP membrane is elastically deformed beyond the crack tip. This second hypothesis is supported by the
substrate. It is this second region that fulfills the conditions necessary to apply equation (1): close to the indent, the InP membrane is maintained suspended at height δ by the plastically deformed portion of the membrane, any stresses between the two being relaxed through the dislocations of the latter. Far from the indent, the InP membrane is firmly bonded to the Si substrate, as revealed in the atomically resolved images of the InP/Si shown in the insets. Between these two extremes, the InP membrane is elastically deformed. The debonding crack propagates only along the InP/Si interface (InP/SiO2 interface in the oxide-mediated bonding case). To establish the relationship between the embedded crack and the blister observed on the surface of InP, the values of (d , L ), measured from STEM images, are compared against those of (d , L ), measured from AFM profiles at the location where the lamellae were prepared in the unpatterned cases. The measured values are summarized in table 1. In both oxide-free 5
Nanotechnology 27 (2016) 115707
K Pantzas et al
Figure 5. Cross-sections of the four indents observed in AFM in figure 5. Insets show the InP/Si interface beyond the debonding-crack tip, where InP is firmly bonded to the underlying Si substrate.
to model it and propose corrections to equation (1). Among those models, the semi-empirical formula of Gillis and Gillman [15]:
Table 1. Measured values for the crack opening (δ), crack length (L),
blister height (hb), and blister lateral extent (Lb), for InP bonded oxide-free to unpatterned Si (OFU), oxide-free to patterned Si (OFP), oxide-mediated to unpatterned Si (OMU), and oxidemediated to patterned Si (OMP). While the measured blister heights and crack are in good agreement, the debonding crack length is found to be 40% shorter than the lateral extent of the blister, on average. This discrepancy is explained in the text.
OFU OFP OMU OMP Precision
δ nm
L μm
hb nm
Lb μm
DL nm
45.4 — 38.5 — ±0.3
0.9632 — 0.8761 — ±0.3
45.1 36 38.2 35.3 ±1
1.627 1.677 1.585 1.625 ±15
664
G=
2 3 + FSi2 ESi hSi3 EInP hInP 3d 2 FInP , 3 3 4 8L (FInP EInP hInP + FSi3 ESi hSi3 )2
(7 )
where Fi = 1 + 3c
h2 hi 3 + (1 + ni ) i2 , L 5 L
(8 )
has been known to yield reliable results in the case of Silicon h2 3 to Silicon bonding [7]. In equation (8), the term 5 (1 + n ) Li2 accounts for shear stress perpendicular to the direction of propagation of the debonding crack, and the term 3c hLi accounts for the rotation of the built-in section that is due to elastic deformation of the DCB beyond the crack tip. In this second term, c is a constant that has been found through experimental adjustment to be equal to 0.63 [13]. Based on this model and given that hSi hInP , a first order approximation of the difference between L and Lb is 3chInP , or 654 nm, a value that is in reasonable agreement with the
709 ±15
observation of deformation fringes that extend beyond the crack tip in the BF-STEM image presented in figure 6(b). Elastic deformation of the beams beyond the crack tip in the DCB experiment is well-established in the literature, and a variety of models have been developed over the years [15–17] 6
Nanotechnology 27 (2016) 115707
K Pantzas et al
Figure 6. (a) Superposition of the AFM and STEM profiles from the surface of InP bonded oxide-mediated to unpatterned Si. The two
profiles are in excellent agreement, indicating that there is no relaxation of the InP membrane during the fabrication of the STEM lamella. (b) BF-STEM of an InP membrane bonded oxide-free to Si. Deformation fringes that extend ∼600 nm beyond the crack tip are visible in this image. The fringes are visible in this image and not others throughout the article as this lamella was prepared specifically in conditions to minimize amorphization in the sidewalls and is thinner than the other lamellae.
measured values of DL in table 1. Given these observations, we assume L to be equal to L b - 3chInP . It is now possible to compute G using equation (7). The results are reported in table 2. Values of the surface bonding energy computed for the same values of δ and L and using equation (1) are also reported in the table. In the case of bonding to unpatterned Si, the values reported for the Maszara model are higher than those reported previously in the literature for direct bonding of InP to Si and using the same model [6, 8, 9]. These values are all higher than the fracture energy of InP [6]—1.5 J m−2, although there is no evidence of fractures or plastic deformation in the InP membrane. The values of G obtained using the model of Gillis and Gilman are, therefore, considered to give a more accurate estimate of the adherence of InP to Si. Using this model, the surface bonding energy for both oxide-free and oxide-mediated bonding are ∼1 J m−2. While this value is already sufficient to ensure a mechanically robust bonding—indeed, values of this order of magnitude are similar to the cohesion energy of bulk iono-covalent crystals [13]—there is still room for improving the bonding process. Such an improvement is typically achieved by annealing the bonded stack at higher temperatures [5–8]. High-temperature anneals may, however, be undesirable from the technological point of view, in particular in the context of CMOS-compatible integration. Table 2 also reports the ratio Gp Gu of the surface bonding energy on patterned Si to that on bare Si. This ratio is 51% in the case of oxide-free bonding and 57% in the case of oxide-free bonding. These two percentages correspond to the actual surface of Si available for bonding. In both cases, the ratio is lower than the reciprocal of the filling factor of the patterns—78.7 %. This discrepancy is related to the fact that the edges of the patterns are rounded off during the patterning process. Bonding only occurs in the middle between patterns,
Table 2. Surface bonding energies deduced for InP bonded oxide-
free to unpatterned Si (OFU), oxide-free to patterned Si (OFP), oxide-mediated to unpatterned Si (OMU), and oxide-mediated to patterned Si (OMP). Values computed both using the semi-empirical formula Gillis and Gilman [15] and analytic expression by Maszara [4] are reported in the table. As the InP membrane is elastically deformed beyond the crack tip, the values given by the semiempirical formula of Gillis and Gilman are expected to give a more accurate estimate of the adhesion of InP to Si. The table also reports the ratio Gp Gu of the surface bonding energy on patterned Si to that on bare Si.
OFU OFP OMU OMP Precision
GGG J m−2
GM J m−2
1.07 0.548 1.04 0.628 ±0.04
2.22 1.13 2.36 1.34 ±0.09
Gp Gu %
51 57 ±1
where the surface of Si has not been affected by the patterning process and can still be in contact with InP. STEM images reveal that this region correlates well with the portion of Si that has not been rendered amorphous during the patterning process. These observations are summarized in figure 7. The slight difference between oxide-free and oxide-mediated bonding—6%—can be attributed to the fact that the thermal silicon dioxide layer used in the case of oxide-mediated bonding slightly planarizes the Si surface. These results show that to obtain more robust bonding on the patterned areas, one needs to produce perfectly sharp emerging edge holes. Such an optimization could be achieved using a pulsed-plasma ICP etching process, such as the one proposed in the literature for improved etched profiles of silicon wires [18]. 7
Nanotechnology 27 (2016) 115707
K Pantzas et al
Figure 7. (a) AFM image of the Si surface after patterning (color scale 0–85 nm). (b) The same image filtered to show only heights from
83–85 nm, corresponding to the actual bonded surface. (c) AFM profile along the dashed red line in (a) showing the curvature of the Si surface (red line) and the portion of the surface that InP adheres to (green dashed fill). (d) BF-STEM image at InP/Si interface (oxide-free case). (e) The same image overlayed with colors corresponding to (red) InP, (green) crystalline Si, and (blue) amorphous Si. The overlays were obtained by combining EDX and diffraction data from the image in (d)—color online.
8
Nanotechnology 27 (2016) 115707
K Pantzas et al
4. Conclusion
References
In conclusion, the method presented in the present contribution that combines instrumented nano-indentation with AFM has been shown to provide precise, highly localized measurements of the surface bonding energy of InP to sub100 nm patterned Si. Using this method, one receives feedback on both the quality of the bonding process and to what extent this process has been affected by the patterning of the underlying Si. This feedback allows one to identify and individually optimize the conditions for the patterning and bonding processes and design optimal experiments that produce mechanically robust InP/Si hybrid stacks for PICs. The underlying mechanism responsible for the debonding is tied to mechanical properties that InP shares with other III-V semiconductors. The method presented here can, therefore, be adapted to other technologically relevant III-V/Si hybrid bonded stacks, such as GaAs/Si [19]. Furthermore, measurements can be carried out at several locations across a wafer, allowing one to collect statistics, compare the results to large-area measurements like the ones from the DCB experiment, and evaluate the homogeneity of the bonding process. The precision of the measurement, while already relatively high, can be further improved by reducing the area scanned in AFM to only the elastically debonded area and increasing the sampling rate. Furthermore, more precise modeling of the debonded surface may also help refine the models discussed here and ultimately yield more precise measurements.
[1] Tong Q Y, Gan Q, Hudson G, Fountain G and Enquist P 2004 Appl. Phys. Lett. 84 732–4 [2] Jain S R, Sysak N N, Kurczveil G and Bowers J E 2011 Opt. Express 19 13692–9 [3] Mechet P, Verstuyft S, de Vries T, Spuesens T, Regreny P, Thourhout D V, Roelkens G a and Morthier G 2013 Opt. Express 21 19339–52 [4] Maszara P, Goetz G, Caviglia A and Mckitterick J B 1988 J. Appl. Phys. 64 4943–50 [5] Tong Q and Gösele U 1999 Semiconductor Wafer Bonding: Science and Technology (The ECS Series of Texts and Monographs) (New York: Wiley) [6] Pasquariello D, Camacho M, Hjort K, Hjort K, Dosza L and Szentpali B 2001 Mater. Sci. Eng. B 80 164–7 [7] Fournel F, Continni L, Morales C, Da Fonseca J, Moriceau H, Rieutord F, Barthelemy A and Radu I 2012 J. Appl. Phys. 111 104907 [8] Pasquariello D and Hjort K 2000 J. Electrochem. Soc. 147 2343–6 [9] Pantzas K, Patriarche G, Le Bourhis E, Troadec D, Itawi A, Beaudoin G, Sagnes I and Talneau A 2013 Appl. Phys. Lett. 103 081901 [10] Itawi A, Pantzas K, Sagnes I, Patriarche G and Talneau A 2014 J. Vac. Sci. Technol. B 32 021201 [11] Talneau A, Roblin C, Itawi A, Mauguin O, Largeau L, Beaudouin G, Sagnes I, Patriarche G, Pang C and Benisty H 2013 Appl. Phys. Lett. 102 212101 [12] Oliver W C and Pharr G M 1992 J. Mater. Res. 7 1564–83 [13] Maugis D 1997 Contact, Adhesion and Rupture of Elastic Solids (Berlin: Springer) [14] Pantzas K, Le Bourhis E, Patriarche G, Itawi A, Beaudoin G, Sagnes I and Talneau A 2014 Eur. Phys. J. Appl. Phys. 65 20702 [15] Gillis P and Gilman J 1964 J. Appl. Phys. 35 647–58 [16] Kanninen M 1973 Int. J. Fract. 9 83–92 [17] Foote R and Buchwald V 1985 Int. J. Fract. 29 125–34 [18] Moritz H, Darnon M, Cunge G and Joubert O 2015 J. Vac. Sci. Technol. B 33 032203 [19] Tanabe K, Watanabe K and Arakawa Y 2012 Sci. Rep. 2 1–6
Acknowledgments The authors would like to gratefully acknowledge funding from the CNRS RENATECH network and the Agence Nationale de la Recherche P2N 2012 grant COHEDIO.
9
