modification of si(100)-surfaces by sf6 plasma etching

Cryst. Res. Technol. 35 2000 6–7 807–821

The effect of plasma pretreatments (reactive ion etching in SF6 and SF6/O2) on Si/Si wafer directbonding was investigated. Etching in SF6 caused a bonding behaviour generally known fromhydrophobic (HF etched) samples, whereas adding O2 to the feed gas caused the Si(100) surfaces tobecome hydrophilic and spontaneous bonding was achieved. The structure of the bonded interfaceswere analysed by high-resolution electron microscopy, ellipsometry, multiple internal reflectionspectroscopy, and secondary ion mass spectroscopy. All the plasma treatments result in an interfacestructure analogous to that known from bonded hydrophobic wafer pairs. The interface does notinvolve an additional layer such as an SiO2 or an amorphous Si layer but small one-dimensional defectsforming a disturbed layer about 1 to 2 nm thick.

Keywords: plasma activation, Si(100)-surfaces, wafer direct bonding

(Received May 3, 2000; Accepted July 1, 2000)

1. Introduction

Silicon wafer direct bonding is a technology promising for VLSI, semiconductor power andsensor device applications (see e.g. TONG and GÖSELE, HUNT et al.). For most applicationsan additional annealing step at higher temperatures is required after the initial room-temperature bonding in order to increase the bonding strength. There are, however, numerousapplications such as the bonding of processed wafers for sensor fabrication which allow onlyshort postannealings at relatively low temperatures such as 400°C or lower. Therefore, anactivation process of the surfaces is desirable which causes high interface energies already atrelatively low temperatures. One possibility for such an activation process is the preparationof interface layers of glass or TMOS (hydrolyzed tetramethoxysilane) causing an increase ofthe interfacial energy by a factor of 2 or more with respect to bonded hydrophilic wafer pairs(e.g.PLÖßL and KRÄUTER). In some cases, however, interface layers are not compatible withdevice requirements, or their preparation cannot be integrated into the device fabricationprocess. Another possibility is the modification of surface bonds by plasma processes.Analyses of the bonding behavior of silicon surfaces exposed to oxygen (DESMOND et al.,WIEGAND et al.), Ar or Ar/H2 plasma (TONG et al., TAKAGI et al.) showed an appreciableincrease in the bonding strength. Treatments in oxygen plasma cause the formation of a thinoxide layer at the bonding interface, whereas surfaces cleaned (etched) in an Ar/H2 plasmaare bondable only after a thermal desorption of hydrogen at temperatures between 400°C and600°C. Interfaces prepared by the latter procedure are atomically flat, showing nointermediate interface layers but the bonding strength after room temperature bonding isabout the same as for normal hydrophilic wafer bonding. Therefore other plasmapretreatments compatible with modern device processes are required. Fluorine compounds

M. REICHE, U. GÖSELE, M. WIEGAND

Max-Planck-Institut für Mikrostrukturphysik, Halle, Germany

Modification of Si(100)-Surfaces by SF6 Plasma Etching –Application to Wafer Direct Bonding

Dedicated to Prof. Dr. J. Heydenreich on the occasion of his 70th birthday

808 M. REICHE et al.: Modification of Si(100)-Surfaces

such as SF6, CF4, or CHF3 and their mixtures with O2, N2, or H2 are widely applied. All thesegases are characterized by a different selectivity of etching silicon or SiO2. Treatments influorocarbon plasmas, however, may cause the existence of residual layers in terms offluorocarbon and/or hydrocarbon compounds (e.g. MATHAD et al.) on the silicon surfacewhich effect the wafer bonding. Therefore, the present paper is concerned with theapplication of SF6 plasma pretreatments to wafer bonding.

2. Experimental

<100> oriented Czochralski-grown silicon wafers (diameter 4 in., p-type, ρ = 30 - 70 Ωcm)were used for the experiments. After standard RCA cleaning processes they were plasmaetched and subsequently bonded. For plasma etching (reactive ion etching, RIE) at 13.56MHz two different processes were used:a) etching in SF6 (100%) at a pressure of 25 Pa for 20 sec, a flow rate of 64 sccm SF6, and a

rf power of 220 W,b) etching in SF6 /18.9%O2 mixtures (SF6 : O2 = 60 : 14) at a pressure of 1.4 Pa for up to 160

sec and a rf power of 80 W.Process a) was carried out in a SENTECH system and process b) in an ALCATEL MCM200 system. After etching, the wafers were bonded using a microcleanroom set-up (STENGLet al.). Bonding under these conditions includes a water flushing of the surfaces before initialbonding. Furthermore, additional samples etched by process b) were also dry-bonded, i.e.without water flushing. After bonding, the samples were annealed for 1 hour at temperaturesfrom 100 to 1100°C. During annealing the energy required to separate the bonded surfaces(termed surface energy γ) was measured by the crack opening method in air (MASZARA etal.). The value of the surface energy is a good indicator for the bonding strength across theinterface. The generation of interface bubbles during annealing was recorded by subsequentinfrared inspection at room temperature.

In order to compare the interface structure of the plasma-etched and subsequently bondedwafer pairs to that of bonded hydrophilic and hydrophobic ones, respectively, furtherinvestigations were carried out for wafer pairs annealed at 1100°C for 5 hours. From suchpairs, samples were prepared for Fourier-transform infrared spectroscopic investigations(FT-IR), including also multiple internal reflection spectroscopy (MIRS), and for highresolution electron microscopy (HRTEM). The sample preparation for both investigations isdescribed elsewhere (REICHE et al. (1997)). After thinning one of the wafers of the bondedpairs down to about 2 µm analyses were carried out also by variable angle spectroscopicellipsometry (VASE), secondary ion mass spectroscopy (SIMS), and by plan-viewtransmission electron microscopy (TEM) in the diffraction contrast mode. The experimentalset-up used for SIMS is described by REICHE et al. (1996). Depth concentration profiles ofoxygen (16O), fluorine (19F, 47SiF), carbon (12C), hydrogen (1H), carbon-hydrogen (13CH), andsulphur (32S) were recorded through the bonded interface. For studying the modification ofthe Si(100) surface identic wafers were analyzed after plasma etching by VASE and atomicforce microscopy (AFM).

3. Results

3.1. Analysis of the plasma-etched Si(100) surface

Samples etched in SF6/O2 for 40 sec were used for analyzing the surface modification. AFMmeasurements were carried out in a Nanoscope IIIa equipped with a Dimension 5000 largesample stage. Due to the plasma process the surface roughness increased by more than afactor of 2 with respect to an otherwise identical RCA-cleaned wafer (Table 1).

Cryst. Res. Technol. 35 (2000) 6-7 809

a) 0 si_jell 0.5 mm

1 si_v4 52.57 Å2 sio2 22.461 Å

b)

Generated and Experimental Data

Wavelength in Å

3000 4000 5000 6000 7000 8000 9000

Tan(

Psi

)

0.0

0.2

0.4

0.6

0.8

Model FitExp E 60°Exp E 65°Exp E 70°Exp E 75°Exp E 80°

c)

Generated and Experimental Data

Wavelength in Å

3000 4000 5000 6000 7000 8000 9000

Cos

(Del

ta)

-1.0

-0.5

0.0

0.5

1.0

Model FitExp E 60°Exp E 65°Exp E 70°Exp E 75°Exp E 80°

d) Wavelength in Å

3000 4000 5000 6000 7000 8000 9000

Pse

udo

<n>

3.0

4.0

5.0

6.0

7.0

Data for Si_V4Si (data from Jellison)

Fig.1: Ellipsometric analysis of a Si(100) surface etched in a SF6/18.9%O2 plasma (RIE) for 40 sec. The model usedfor data fit (a), the experimentally generated data for the amplitude ratio Ψ (b) and the phase shift difference ∆ (c) aswell as the data used for refractive index n (d) and extinction coefficient k (e) for modeling the disturbed layer (Si-V). Variable angle spectroscopic ellipsometry (VASE) was applied.


An analogous roughness results from chemical etching in 1% HF solutions for 10 min(REICHE et al. (1994)). The roughness is associated with first indications of a granularstructure, which, however, is substantially less pronounced than for etching in a CF4/O2

plasma for the same time.

Rms (Rq) (nm) Ra (nm) Rmax (nm)RCA cleaned sample 0.128 0.102 1.176after etching for 40sec 0.268 0.214 2.116

Table 1: Results of roughness analysis on a SF6/18.9%O2 plasma etched Si(100) surface.

Data were collected from AFM measurements (analyzed area 1 x 1 µm2).

Fig. 2: Infrared transmission images of bondedwafer pairs pretreated by etching in an SF6

plasma (a,b) and an SF6/O2 plasma (c - h).a, b - Images of a wafer pair after room-temperature bonding (a) and annealing at1100°C for 5 hours (b).c-f - Images of a wafer pair after room-temperature bonding (c), and after annealingfor 1 hour at 200°C (d), 400°C (e), and 600°C(f). Bonding with a preceding water flush.g, h - Images of a wafer pair after room-temperature bonding without a preceedingwaterr flush and single-step annealing at1100°C for 5 hours.


Results of spectroscopic ellipsometry (VASE) are shown in fig. 1. For modelling the plasma-etched surface a 2-layer model is assumed on top of bulk silicon consisting of a disturbedsilicon layer, about 5.2 nm thick, and an oxide layer. Other common models such as theBruggeman effective-medium approximation (EMA layer) used for interpreting HF-etchedsurfaces (YAO et al.) fail. For the disturbed silicon layer (Si-V) the silicon data for refractiveindex n and extinction coefficient k are fitted. But as fig. 1 also shows the differences forboth are very small. The physical meaning of the disturbed layer is that it represents adamaged layer caused by plasma etching. This interpretation is supported by otherinvestigations demonstrating that plasma etching causes Si lattice defects (e.g. WU andMCLARTY). Furthermore, VASE measurements on surfaces analogously etched in other gasmixtures (CF4/O2, CHF3/C2F6/O2/He) are interpreted on the basis of the same model (REICHEet al. (2000). The thickness of the disturbed (damaged) layer and the data for n and k areslightly different indicating the different damage generation due to etching by differentgases. As shown by other authors (e.g. WU and MCLARTY), and as the investigations belowwill show the damage is annihilated by an appropriate postannealing.

3.2. Wafer bonding behavior, surface energy, and interface bubbles

The bonding of wafers etched in SF6(100%) shows numerous similarities to that ofhydrophobic wafers prepared by a HF-dip before bonding. There is no spontaneous bondingat room temperature (fig. 2a). This causes also low values of the surface energy γ, which wasmeasured to be 16.5 mJ/m2. These values for γ are about a factor of 10 lower than for bondedhydrophilic wafer pairs, and of the same order of magnitude as for bonded HF-treatedhydrophobic wafer pairs. The dependence of γ on subsequent heat treatments appears to bealso analogous to that of bonded hydrophobic wafer pairs. Only annealing at hightemperatures cause interface bond energies high enough for further applications.

In contrast, wafers treated in a SF6/18.9%O2 plasma bond spontaneously. All datameasured after bonding at room temperature show that γ is significantly higher than forbonded hydrophilic wafer pairs. Moreover, γ clearly increases if the wafers are etched for upto 40 sec in the SF6/O2 plasma. Further prolonging the plasma treatment causes γ to decreaseagain. The data also showed marked differences of γ between the bonding under dryconditions, and with a preceding water flush. After bonding under dry conditions at roomtemperature γ varies by a factor of about 2 for plasma treatments up to 40 sec, whereas aftermore extended treatments there is no more difference. The increased value of γ at roomtemperature (as compared to that of normal hydrophilic bonding) indicates an alteration ofthe surface structure, i. e. the treatment in the SF6/O2 plasma increases the number of activesites. Assuming the concept of OH-groups as the active sites for bonded hydrophilic wafers,

the differences in γ measured between bonding under dry conditions and with a precedingwater flush suggest that especially under room-temperature conditions the active sitesproduced by the plasma treatment are characterized by a higher affinity for bonding OH-groups on the surface, resulting in an increased surface energy γ during bonding. Thecomplexity of the mechanism is demonstrated by the dependence of γ on the annealingtemperature (fig. 3). For T < 400°C, plasma-treated wafer pairs bonded with a preceding

water flush yield values of γ similar to those of bonded hydrophilic wafers. After annealingsabove 400°C γ is higher than for bonded hydrophilic wafers (the high interface bond energydoes not allow measurements by the crack opening method for samples annealed at T ≥600°C, sometimes already after annealing at 400°C). Moreover, γ vaues measured afterannealing plasma-treated wafers bonded under dry conditions are substantially lower than fora bonded hydrophilic wafer pair (fig. 3b).


a)

b

Fig 3: Dependence of the surface energy γ on the temperature during additional annealing for 1 hour after bonding.Before bonding the samples were treated by RIE in a SF6/18.9%O2 plasma for different times. Data are given forbonding with (a) and without (b) a preceding water flush. Note that the high interface bond energy does not allowmeasurements by the crack opening method after annealing at T > 600°C.The figures show a bonded hydrophilic wafer pair as reference.


Fig. 4: Electron microscope images of the interfaces of bonded wafer pairs, pretreated by RIE in a SF6/100%)plasma for 20 sec (a) and in a SF6/18.9%O2 plasma for 40 sec (b) followed by a preceding water flush beforebonding. Fig. (c) shows the interface of a wafer pair pretreated in a SF6/18.9%O2 plasma for 40 sec and bondedwithout a preceding water flush. For comparison, the interface of a bonded hydrophobic wafer pair prepared byetching in a HF solution (1%) is shown in (d). After bonding all wafer pairs were annealed at 1100°C for 5 hours.


There is an analogous behavior also with respect to the interface defect generation (bubblegeneration) between SF6(100%) etched and hydrophobic wafers on the one hand, andSF6/18.9%O2 etched and hydrophilic wafers on the other hand. Figures 2c to 2f show infraredmicroscope images of the same wafer pair after room-temperature bonding and afterannealings at 200, 400, and 600°C, respectively, for 1 hour. Here, plasma etching in SF6/O2

and a preceding water flush before bonding were applied. The generation of bubbles seemsto be similar to that observed in bonded hydrophilic wafer pairs. Further annealing at highertemperatures reduces the bubble density. Interfaces without bubbles appear after single-stepannealing at high temperatures (figs. 2g, h). The same holds true also for SF6/O2 etched anddry-bonded wafer pairs as well as of samples etched in SF6(100%), only.

We conclude, that plasma etching in SF6(100%) causes Si(100) surfaces with a behavioranalogous to that of hydrophobic (HF-etched) ones, whereas an addition of oxygen to theetch gas changes the surface behavior analogous to that known from hydrophilic surfaces.

3.3. TEM, MIRS, and SIMS analyses of the bonded interface

Cross-sectional samples were prepared for HRTEM after annealing wafer pairs at 1100°C for5 hours. Typical images of the interface structure are shown in fig. 4. Etching in SF6(100%)(fig. 4a) and SF6/O2 (figs. 4b,c) causes an interface structure analogous to that after chemicaletching in HF solutions (fig. 4d). Moreover, there are no differences in the interface structureof SF6/O2-etched samples between wafer pairs bonded with a preceding water flush (fig. 4b)and without (fig. 4c). The oxide layer on the surface is always removed and both Si surfacesare directly bonded to each other. Under the experimental conditions used the oxide layer,however, is not completely removed. After annealing at 1100°C sometimes oxide islands ofoctahedral morphology are observed, similarly as was described for bonded hydrophobicwafer pairs prepared by HF-etching of the surfaces (REICHE et al. (1994)). Presently, it isunclear whether the oxide islands are remains of an incompletely removed surface oxidelayer or are the result of preferential oxygen precipitation at the bonding interface during theannealing procedure. The bonded interfaces are not atomically flat as in the case of etchingin an Ar/H2 plasma (TONG et al.). Instead, the interface represents a 1 to 2 nm thick defectivelayer. Most of the defects within the layer are of a rod-like morphology and orientedparallelly to 111 lattice planes. They cannot be observed in plan-view samples due to their1-dimensional extension. Rod-like defects were shown to consist of silicon self-interstitialsrather than of foreign phases composed of impurities detected by SIMS (REICHE et al.(1996)). The reason for such a defective interface layer is assumed to be the same for SF6 orSF6/O2 etched wafer pairs as described for HF-etched ones. For the latter, an accommodationprocess of two rough surfaces during annealing was suggested (REICHE et al. (1994)). It isgenerally known that plasma processes in SF6-containing gases are selective to etch siliconso that etched surfaces become rough. If these surfaces come into contact, processesanalogous to those of HF-etched surfaces can be asummed during annealing. The analogy ofthe structure of the interfaces revealed by HRTEM does not completely correspond to resultsof ellipsometric measurements (VASE). Analyzing the same samples by VASE afterthinning down one of the wafers by KOH etching proved that under the same assumption forthe KOH-etched surfaces, the interface cannot be modelled by a Bruggeman effective-medium approximation (EMA-layer) used for HF-treated samples (REICHE et al. (1994)).Here, the interface is modelled by an EMA-layer consisting of silicon and a few per cent ofvoids (about 27%), which physically means a lower optical density than applicable for bulksilicon. Models incorporating a defective layer which we used for interpreting freshlyplasma-etched surfaces also fail, suggesting that the defective layers are annihilated. Instead,the best fit of the experimental data of the plasma-treated and subsequently bonded samples


is achieved, if a thin SiO2 layer of about 0.1 nm in thickness, is assumed in the interface. Thevalue of 0.1 nm is too small even for a monolayer of SiO2 and thus has to be interpreted in adifferent way in terms of two approaches: 1. While VASE-data are collected from a largerarea than observable by HRTEM, the thin oxide layer is consistent with the results ofelectron microscopy: the layer may integrally represent the oxide islands. Then the defectlayer in the bonded interface itself does not cause its optical properties which are differentfrom that of bulk silicon. 2. The high concentration of impurities detected by SIMS suggestthat the interface layer, strongly enriched with these elements, has optical properties close tothose of SiO2.

Results of SIMS analyses of a bonded wafer pair annealed at 1100°C for 5 hours areshown in table 2. The wafer surfaces were etched in SF6/O2, water flushed and then bonded.For comparison, SIMS results of both bonded hydrophilic and hydrophobic wafer pairs,respectively, are summarized, too. The oxygen concentration (16O) at the interface of theSF6/O2 etched sample is almost equivalent to that of the interface of a bonded hydrophilicsample. At the same time, the fluorine concentration reaches values of about 3.1020 cm-3,which is more than 3 orders of magnitude higher than for bonded hydrophilic wafer pairs,and 2 orders of magnitude higher than for bonded hydrophobic samples. The almostequivalent concentrations obtained for 19F and 47SiF prove that the concentration values aredue to fluorine and are not effected by mass interferences with mO1Hn (with m= 16, 17, 18and n = 1, 2, 3) (REICHE et al. (1996)). Furthermore, also substantially higher hydrogenconcentrations are present at the bonded interface of the plasma-etched sample. Thehydrogen concentration (1H) is more than 4 orders of magnitude higher than measured forbonded hydrophobic or hydrophilic wafer pairs. Moreover, the concentrations of carbon (12C= 1.6.1020 cm-3) and sulphur (32S = 2.1020 cm-3) are also high at the interface of the plasma-etched sample. Note that also carbon-hydrogen compounds are present. It is remarkable thatall the impurities detected cause strong and especially sharp peaks at the interface. Thismeans that diffusion processes are of minor importance inspite of the high-temperatureannealing at 1100°C. This is especially remarkable for the case of hydrogen which candiffuse rapidly at 1100°C and thus has to exist in a strongly bond state at the bondinginterface.

Analyzed species #1 #2 #316O 2.0.1022 6.1.1020 1.1.1022

19F 3.0.1020 3.0.1018 9.3.1016

47SiF 5.0.1020 1.9.1018 - 1H 6.0.1021 2.2.1017 4.3.1017

12C 1.6.1020 n.o. n.o.32S 2.0.1020 n.o. n.o.

13CH 6.0.105 n.o. n.o.

Table 2: Results of SIMS measurements. Data (peak concentration at the bondedinterfaces) are given for #1 a wafer pair, with the surfaces etched in a SF6/18.9%O2 plasma and with apreceding water flush before bonding, #2 a bonded hydrophobic wafer pair with the surfaces dipped into 1% HF for 2 minand bonded without additional water flush, and #3 a bonded hydrophilic wafer pair with the surfaces only RCA cleaned and bondedwith a preceding water flush.After bonding all samples were annealed at 1100°C for 5 hours. All data are given asabsolute concentrations (at/cm

3), except for

13CH (counts/sec).


a)

b


c)

d)

Fig. 5: Polarized infrared spectra of the silicon-hydrogen stretching vibrations of bonded wafer pairs after annealingat 1100°C for 5 hours. The following pretreatments were used:a) Etching the surfaces in a SF

6(100%) plasma for 20 secs. and bonding with a preceding water flush.

b) Etching the surfaces in a SF6/18.9%O

2 plasma for 160 secs. and bonding with a preceding water flush.

c) Etching the surfaces in a SF6/18.9%O

2 plasma for 40 secs. and bonding with a preceding water flush.

d) Etching the surfaces in a SF6/18.9%O

2 plasma for 160 secs. and bonding without a preceding water flush.


Multiple internal reflection spectroscopy (MIRS) has repeatedly been used to investigate theinterface chemistry of bonded wafers (CHABAL et al., DUMAS et al., REICHE et al.(1997),WELDON et al.). In the present case MIRS revealed that a certain amount of hydrogen isbonded at the interface as Si-Hx. Typical examples of SF6(100%) and SF6/O2- etched samplesare shown in fig. 5 where a RCA-cleaned sample acts as reference. The sample etched in SF6

and water flushed before bonding is characterized by a dominant peak at 2091.8 cm-1,corresponding to the symmetric stretching mode (νss) of the coupled monohydride (Si-H).Further modes, however, at a very low intensity, appear at 2105.6 cm-1, 2116.5 cm-1, 2128.6cm-1, and 2141.0 cm-1 (fig. 5a). The first and second are related to the symmetric andantisymmetric stretching modes of the dihydride, whereas the third is caused by Si-H3 [16].At 2141 cm-1 the mode is probably related to the dihydride (constrained dihydride), too. Thebehaviour is analogous after extended etching in SF6/O2 and with a water flush beforebonding (fig. 5b). Here, the symmetric stretching mode of the coupled monohydride occuredat νss = 2088.1 cm-1, the symmetric and antisymmetric stretching modes of the dihydride, atνss = 2103.1 cm-1 and νas = 2115.4 cm-1 , respectively, and the mode of the trihydride at ν =2132.5 cm-1. Note that their positions are close to the modes observed at HF-etched Si(100)-surfaces. Short etching treatments in SF6/O2 plasma yield a similar spectrum, the position ofall the modes, however, are shifted to shorter wave numbers (fig. 5c). There is a mode at2077.5 cm-1, which might correlate with νas of the coupled monohydride. The modes at 2098and 2110.8 cm-1 may be correlated to the dihydride with a shift of ∆νss ≈ ∆νas ≈ 5 cm-1. Thesame shift, but to larger wave numbers, occurs for the mode of the trihydride (ν = 2137.5cm–1). Following this interpretation, the mode at 2124.3 cm-1 is related to the dihydride, too.Furthermore, as fig. 5d shows, plasma etching without subsequent water flush causes thesame Si-H vibration modes as for shorter etching times, however, with a water flush. Thisimplies that 1.) Si-H bonds on the plasma-etched surfaces are modified by the water flushing,and 2.) bonds on the surface are changed (modified) with increasing time of the plasmatreatment.

4. Discussion

Chemical sputtering is assumed to be the primary mechanism of ion-enhanced etching(d´AGOSTINO and FLAMM). Ion bombardment provides only the activation energy necessaryfor etching reactions on the substrate surface, where weakly bonded molecules are formedand (most of them) subsequently desorb into the gas phase. These reactions do not onlycause structural modifications of the surface, but also strong variations of its activity. Theconditions of the Si(100) surface may be changed from hydrophobic (by etching inSF6(100%)) up to more and more hydrophilic ones by adding oxygen to the feed gas. Etchingin SF6(100%) causes the gas phase to consist of F and SFx (1 ≤ x ≤ 5) formed by electronimpact dissociation (KOJIMA et al.). Their interaction with a silicon surface causes theformation of a thin SiFx layer (1 ≤ x ≤ 4) with a thickness of about 1 to 3 nm (CAMPO et al.).While the etch rate is directly related to the concentration of atomic fluorine in the gas phase,fluorine is assumed to be the main etchant. The fluorine atoms are transported through thegas phase and adsorbed on the SiFx layer. From there, they diffuse through the layer to reactwith Si to form new SiFx molecules, or they react with SiFx molecules in the layer to formSiFx+1 (the end product of the latter is SiF4, which desorbs into the gas phase). Therefore, afterSF6 etching, the Si(100) surface is covered with a thin fluorosilane (SiFx) layer. It is generallyknown that fluorosilanes are strongly hygroscopic so that a water flush before bondingresults in a reaction of the type

2SiF4 + 2H2O à SiO2 + SiF6

2- + 2H+ + 2HF . (1)


There may be analogous equations for reactions of other SiFx compounds (x ≤ 3). Hence,most of the SiFx is removed from the surface and open bonds are saturated with hydrogen,causing the hydrophobic behavior of the surface. The resulting SiO2 will be etched away bythe HF produced in analogy to the reaction occuring during chemically cleaning an oxidizedsilicon surface in HF via

SiO2 + 6HF = SiF6

2- + 2H+ + 2H2O. (2)

In both cases the bonds on the surface will preferentially be saturated with hydrogen. Thisexplains the analogous bonding behaviour of HF and SF6-etched surfaces.

If O2 is added to the feed gas besides F and SFx also sulphur oxyfluorides (SO2F2, SOF2,SOF4) occur in the discharge, with sulphoryl fluoride (SO2F2) being the most dominant(d´AGOSTINO and FLAMM, KOJIMA et al., TZENG and LIN). The formation of oxyfluorides isdue to reactions of oxygen atoms with SFx radicals. As a consequence, the concentration ofatomic fluorine increases, either by reaction sequences in the discharge such as (KOJIMA etal.)

O + SF3 -> SOF2 + F (3a)O+ SOF3 -> SO2F2 + F , (3b)

or by reducing the rates of the surface recombination reactions (d´AGOSTINO and FLAMM)

SFx + F -> SFx+1 1 ≤ x ≤ 5 . (4)

Changing the gas phase chemistry results also in a composition of the surface layer differentfrom that after etching in pure SF6. Depending on the O2 concentration on the feed gas thecomposition of the layer is that of fluorosiloxanes, i.e. SiOxFy (TZENG and LIN). For highoxygen concentrations it is close to SiOF2. It is known that, with water present,

fluorosiloxanes hydrolyze to silicic acid (SiO2⋅nH2O) so that, during the water flush beforebonding, reactions such as

SiOF2 + (n+1)H2O -> SiO2 ⋅nH2O + 2HF (5)

proceed at the Si surface. We speculate that the silicic acid leads to the formation of Si-Oand, more likely, Si-OH bonds at the surface, causing its hydrophilic behaviour. Thehydrophilicity is suggested to be increasing as the O2 concentration of the feed gas increases.This interpretation also explains the bondability of SF6/O2 etched surfaces without apreceding water flush. If the effect of moisture is neglected initial bonds may possibly beformed via Si - O - Si chains such as

Si O Si O SiSi

F F

F F

(6)

with a loose connection to the surface Si-atoms. Bonds to the Si surface are preferred viaoxygen, owing to a strong polarity induced by the bonds in the SiOF2 molecules.

The concept of surface activity and bonding after SF6 and SF6/O2 plasma etchingdiscussed above explains all the experimental results. As discussed before, it interprets thesimilarities between HF-etched and SF6-etched surfaces and their analogous bondingbehaviour. Moreover, also the hydrophilic behaviour of SF6/O2 etched surfaces and thedifferences in the bonding of such surfaces with and without preceding water flush areclearly shown. The weaker bonds of SOF2 on the Si surface atoms via oxygen are responsible


for the bond energy of the dry-bonded wafers at room temperature which is lower than forwafers bonded with a water flush, with Si-OH bonds initially generated at the interface. Afterannealing at higher temperatures the weak bonds are transformed into stable Si-O bonds,resulting in interface energies of dry-bonded wafers similar to those of hydrophilic-bondedones. Furthermore, the Si-H bonds at the interfaces detected by MIRS are primarily causedby surface reactions of SiF4 according to eq. (1) for SF6-etched samples, and, analogously,for HF-etched ones (according to eq. (2)). For SF6/O2-etched samples hydrogen may mainlyresult from the dissociation of the HF molecule

HF -> H+ + F- . (7)

The impurities at the bonded interfaces detected by SIMS have not yet been considered.Residual hydrocarbons may be another reason for Si-H bonds occuring at the interface. Thehigher concentration of hydrogen at the interface of SF6/O2-etched samples than ofhydrophilic ones may be correlated to the higher bonding strength due to the larger numberof Si-OH bonds. Furthermore, the higher fluorine concentration at the bonding interface ofplasma-etched samples relative to bonded hydrophobic ones suggests that dry bonding occurpartly via SOF2 as discussed above. In samples bonded with a prior water flush the higher Fconcentration at the interface may be due to an incomplete removing of these species, or to asurface concentration of fluorine generally higher after plasma etching than after a chemicaltreatment in HF solutions. Moreover, only speculations are possible with respect to the effectof sulphur. The concentration of sulphur at the interface of a bonded wafer pair wasdetermined to be 2⋅1020 cm-3. Like all the other impurities, sulphur is pinned to the interfaceand does not diffuse into the bulk during annealing (the sample investigated was annealed at1100°C for 5 hours). According to the HRTEM investigations the thickness of the interfaceis about 2 nm, yielding a sulphur concentration of about 4⋅1013 atoms/cm2 in it, correspondingto some hundredths of a monolayer. This may be the reason that sulphur was not detected onSF6 or SF6/O2 plasma-etched surfaces by other methods such as XPS. The detection limit forXPS measurements mostly used for such analyses is about a factor of 10 higher.

We finally mention also the possibility that the high concentration of hydrogen at thebonding interface of an SF6/O2 etched and subsequently water-flushed and bonded siliconwafers even after a 1100°C heat treatment might be associated with the correspondingly highsulphur concentration in terms of sulphur-oxygen-hydrogen complexes.

In conclusion, we have investigated the bonding of silicon wafers treated in SF6 or SF6/O2

plasma. The room temperature bonding strength, represented in terms of the surface energy γincreased by an SF6/O2 plasma treatment followed by a water flush and bonding.

Acknowledgment

The authors are grateful for the experimental support of Dr. K. Wandel (Sentech Instruments, Berlin)and Mr. D. Stolze (CIS Centrum für intelligente Sensorik GmbH, Erfurt).

References

CAMPO, A., CARDINAUD, CH., TURBAN, G.: J. Vac. Sci. Technol. B13 (1995) 235CHABAL, Y.J., HINES, M.A., AND FEIJOO, D.: J. Vac. Sci. Technol. A13 (1995) 1719D´AGOSTINO, R., AND FLAMM, D.L.: J. Appl. Phys. 52 (1981) 162DESMOND, C.A., HOBART, K., KUB, F., CAMPISI, G., AND WELDON, M.: in: Semiconductor Wafer

Bonding: science, Technology, and Applications IV (Eds.: U. Gösele, H. Baumgart, T. Abe, C.Hunt, and S. Iyer), The Electrochemical Society, Pennington, PV 97-36, p. 459 (1997)

DUMAS, P., CABAL, Y.J., AND JAKOB, P.: Surf. Sci. 269-270 (1992) 867


Kojima, M., Kato, H., Gatto, M.: J. Appl. Phys. 75 (1994) 7507HUNT, C.E., BAUMGART, H. AND GOESELE. U.: Semiconductor Wafer Bonding: Science, Technology,

and Applications V. Honolulu, Hawaii, The Electrochemical Society, Pennington, PV 99-35(1999)

YAO, H., WOOLLAM, J.A., ALTEROVITZ, S.A.: Appl. Phys. Lett. 62 (1993) 3324Maszara, W.P., Goetz, G., Cavilia, A., and McKitterick, J.B.: J. Appl. Phys. 64 (1988) 4943MATHAD, G.S., HESS, D.W., AND MEYYAPPAN, M.: Interface, Summer 1999, 34PLÖßL, A. AND KRÄUTER, G.: Materials Science and Engineering R25(1) (1999) 1REICHE, M. GÖSELE, U., AND TONG, Q.-Y.: in: Semiconductor Silicon 1994 (Eds. ), The Electrochem.

Soc., Pennington, PV 94-10, p. 408 (1994)REICHE, M., HOPFE, S., GÖSELE, U., AND STRUTZBERG, H.: Jpn. J. Appl. Phys. 35 (1996) 2102REICHE, M., HOPFE, S., GÖSELE, U., AND TONG, Q.-Y.: Mikrochim. Acta 125 (1997) 367REICHE, M., WIEGAND, M. AND GÖSELE, U.: Mikrochim. Acta (2000), in pressSTENGL, T., AHN, K.-Y., AND GÖSELE, U.: Jpn. J. Appl. Phys. 27 (1988) L2364TAKAGI, H., MAEDA, R., HOSODA, N., CHUNG, T.R., AND S UGA, T.: in: Semiconductor Wafer Bonding:

Science, Technology, and Applications V. (Eds: Hunt, C.E., Baumgart, H. and Goesele. U.: )The Electrochemical Society, Pennington, PV 99-35 (1999)

TONG, Q.-Y, LEE, T.-H., GÖSELE, U., REICHE, M., RAMM, J., AND BECK, E.: J Electrochem. Soc. 144(1)(1997) 384

TONG, Q.-Y., GÖSELE, U.: Semiconductor Wafer Bonding – Science and Technology, Wiley-Interscience, New York (1999)

TZENG, Y., LIN, T.H.: J. Electrochem. Soc. 134 (1987) 2304WELDON, M.K., MARSICO, V.E., CHABAL, Y.J., AGARWAL, A., EAGLESHAM, D.J., SAPJETA, J., BROWN,

W.L., JACOBSON,D.C., CAUDANO, S.B., CHRISTMAN, S.B., CHABAN, E.E.: J. Vac. Sci. Technol.B15 (1997) 1065)

WIEGAND, M., M. REICHE, AND U. GÖSELE: in: Semiconductor Wafer Bonding: Science, Technology,and Applications V. (Eds: Hunt, C.E., Baumgart, H. and Goesele. U.: ) The ElectrochemicalSociety, Pennington, PV 99-35 (1999)

Wu, W. and McLarty, P.K.: J. Vac. Sci. Technol. A13 (1995) 67

Contact information:

M. REICHE*, U. GÖSELE, M. WIEGAND

Max-Planck-Institut für MikrostrukturphysikWeinberg 206120 HalleGermany

*corresponding authore-mail: [email protected]

modification of si(100)-surfaces by sf6 plasma etching

Documents