Applied Surface Science 72 (1993) 373-379 North-Holland
applied surface science
On the surface segregation of silicon in
Y. S a m s o n , J.L. R o u s s e t , G. B e r g e r e t , B. T a r d y , J.G. Bertolini * Institut de Recherches sur la Catalyse (CNRS), 2 Avenue Albert Einstein, F-69626 l.qlleurbanne Cedex, France
and G. L a r o z e Centre de Recherches des Carri~res (Rh~ne-Poulenc), BP 62, F-69192 Saint-Fons Cedex, France
Received 3 April 1993; accepted for publication 26 July 1993
Silicon segregates to a large extent to the surface of a Cu3Si alloystabilized at 700 K. The top layer contains as much as 70 at% Si, the second and third layers are also significantlyenriched in silicon. Such a phenomenon could largely account for the renewal of the surface layers of the silicide as the silicon is consumed during the direct synthesis (of methylchlorosilane by reaction with methyl chloride). As demonstrated by calculations of the segregation free enthalpy based on the broken-bond model, the difference in surface tension between Cu and Si is the main driving force involved in the observed segregation.
1. Introduction Commercial production of silicon polymers relies, as a first step, upon the synthesis of dimethyldichlorosilane ((CH3)2SiCI 2) by the direct reaction of methyl chloride (CH3CI) on silicon. That is the so-called direct synthesis, chiefly catalyzed by copper and performed in fluidized bed reactors between 550 and 620 K with a CH3C1 pressure of about 4 - 6 atm . Despite a huge amount of work since its invention by Rochow  in 1945, the direct synthesis is still poorly understood. Only recently, a few surface-science studies have been devoted to this major heterogeneous reaction. As previous works [3,4] have demonstrated the preponderant involvement of the r/-phase Cu3Si in the catalytic activity of the copper-silicon bed, authors were
* Contact author: Fax: (33) 72 44 53 99; tel.: (33) 72 44 53 09.
mainly concerned by the behavior of the Cu3Si alloy. Such studies [5-7] should be regarded as a first and necessary step toward a better understanding of the reaction mechanism. Frank and Falconer  have published an Auger spectroscopy study putting forward the segregation of silicon to the surface of the Cu3Si alloy. Unfortunately, a quantitative use of the low-energy Auger peaks (Cu MVV at 60 eV and Si LVV at 92 eV) is forbidden by the splitting of the silicon LVV peak upon alloying with copper [5,8]. As a result, the use of high-energy Auger peaks (Cu LMM at 920 eV and S i K L L at 1619 eV) leads to a poor surface sensitivity. In the present paper, the combined use of XPS and LEIS gives access to the ultimate surface composition and provides information about the in-depth extension of the surface segregation. In addition, we make an attempt at interpreting the observed surface segregation through a broken-bond model . Necessary attention has also been paid to bulk characterization by X-ray
0169-4332/93/$06.00 © 1993 - Elsevier Science Publishers B.V. All rights reserved
Y Samson et al. / On the surface segregation of silicon in Cu3Si
diffraction, mainly with respect to the crystallographic evolution of the material with temperature.
2. Experimental procedure The polycrystalline Cu3Si compound was supplied by Cristal Tec. It has been prepared by induction melting of the high-purity components in a cold crucible, under an inert atmosphere. This yielded a cylindrical bar of 9 mm in diameter. The nominal composition has been controlled by chemical analysis, giving 73.8 at% Cu and 26.2 at% Si. According to surface images obtained by scanning Auger electron spectroscopy (AES), the crystallite size can be roughly estimated to about 50 to 100 ~zm. For X-ray diffraction studies, a small part of the sample was in ground form. The experiments were performed on a goniometer (Siemens D500) equipped with a high-temperature camera using C u K a Ni filtered radiation. This home-made stainless-steel camera, closed by a water-cooled cylindrical Be window, is flowed with He. The program-controlled temperature can reach 1000 K. The detector was a position-sensitive proportional counter (Raytech) covering a 16° (20) angular range and the Diffrac-AT software (Socabim) was used for diffractogram processing. For surface-science studies, the samples were cut in discs of about 1 mm thickness and polished down to 1 /xm with diamond compounds. In UHV, the cleaning of the surface was achieved by repeated cycles of Ar+-ion bombardment at 3 keV and heating at 700 K. The heating temperature has been chosen in order to avoid a phase transition occurring just above 740 K. The main impurity detected during the cleaning process was oxygen. The low-energy ion spectroscopy (LEIS) and X-ray photoemission spectroscopy (XPS) studies were performed in an ESCALAB 200R machine from Fisons Instruments, with a working pressure of 3 × 10-10 mbar. LEIS analyses were done with 1 keV 4He+ ions, the scattering angle being 155°. The primary beam was fixed at 10 nA, and scanned over about 1 mm 2 during measurements.
Pure Si and pure Cu single crystals of (100) orientation were used as reference samples for the determination of the relative Si and Cu sensitivity factor S in our experimental conditions: Scu/Ssi = 9.2 (ratio of the respective LEIS peak areas divided by the appropriate surface atomic densities). XPS experiments were carried out using the non-monochromatized AI K a X-ray source of the dual anode (hv = 1486.6 eV). The photoelectron current was collected at various exit angles 0 with respect to the surface plane. Thus, the depth profile of the concentration was determined from the variation of the measured ratio of the silicon (Si 2p) and copper (Cu 3p) XPS signals versus the analysis angle. The aperture angle of the spectrometer was fixed at 15° during these measurements. AlES experiments were performed with a CMA spectrometer from Riber, using a primary beam energy of 2.5 keV and a 2 V peak-to-peak modulation amplitude.
3. Bulk characterization The phase composition of our sample has been checked between 300 and 1000 K by X-ray diffraction. At room temperature, the observed pattern is in close agreement with the one obtained by Weber  who prepared the coppersilicon alloy by reaction of molten copper chloride with a single crystal of silicon. However, it is not similar to the spectrum available in the JCPDS data base . In order to clearly establish the structure of our sample, we looked for the temperature of occurrence of the two phase transitions indexed in the concentration range of our sample (from r/" to ~7' and r/' to ~7). This search was carried out in the X-ray diffraction facility with an heating rate of 0.75 K/rain, under He flow. Diffractograms were recorded at 573, 673, 698, 723, 748, 773, 813, 853 and 973 K. At each of these temperatures of acquisition, two hours were allowed to ensure completeness of a possible phase transition. The first phase transition occurred between 748 and 773 K, leading to the "q' phase, which is
Y. Samson et aL / On the surface segregation of silicon in Cu3Si
very similar to -q". The second phase transition was observed between 853 and 973 K. These results are consistent with the commonly admitted phase diagram . Besides the changes related to the phase transitions, minor modifications of the diffraction pattern occurred between room temperature and 673 K. It was observed that some peaks shifted towards smaller angles while some others were better resolved. Surprisingly, the diffraction pattern observed at room temperature after further heating at 973 K and a slow cooling was the one previously obtained at 673 K and not the roomtemperature pattern firstly observed before heating. However, this first room-temperature pattern was again obtained after 48 h under ambient atmosphere. This is a good indication that the observed changes are due to bulk contamination of our sample by ambient atmosphere. These observations can be explained in the light of U H V experiments. Indeed, the first annealings (under about 10-10 mbar) induce a large oxidation of silicon in the surface layers of our sample, even if a clean surface (according to Auger measurements) has been obtained by Ar ÷ion bombardments. As the annealing can be very short (a few minutes), contamination by the residual atmosphere cannot account for the observed oxidation. As a result, we have to admit the presence of oxygen, mobile below 700 K, in the bulk of the copper-silicon alloy. This has painful implications for the U H V experiments because long annealings are needed to allow high-temperature observations of a clean surface. Furthermore, one could think that this phenomenon accounts for the surprisingly high oxygen concentrations sometimes cited in previous papers .
trial reaction temperature . The sample was equilibrated at 700 K during 5 min before analysis. It was verified that a longer annealing does not induce a further modification of the surface composition, thus ensuring we reached the equilibrium state at that temperature.
4.1. LEIS experiments Fig. 1 displays typical LEIS spectra of the Cu3Si surface, after ion bombardment (a) and annealing at 700 K (b). Two peaks are observed at 578 and 786 eV, corresponding to H e + ions backscattered by Si and Cu surface atoms, respectively. Both the absence of any other peaks (corresponding to C, O or S for instance) and the low background level at low energy attest to the cleanliness of the sample. By using the ratio of the sensitivities given above (Scu/Ssi = 9.2), the Si concentration in the
~ 600 0 400 2O0 0 450
600 650 700 750 Kinetlc energy/eV
1000 800 _
4. Surface characterization
Surface analysis has been performed mainly by LEIS and XPS. In order to characterize surface segregation, two different treatments have been applied to the sample before analysis: Ar + bombardment at 3 keV or annealing at 700 K. Thus, the annealed state provides a good image of what could be the copper-silicon surface at the indus-
uo 400 2O0 450
Fig. 1. LEIS spectra of the Cu3Si surface after (a) 3 keV Ar+-ion bombardment, (b) further annealing at 700 K. The primary ions were 4He+ with 1 keV energy. The incident current was 10 nA scanned over 1 mm2.
Y. Samson et aL / On the surface segregation of silicon in Cu3Si
first layer can then be deduced for the studied alloy assuming that: • the chemical environment does not affect the sensitivity factor of both considered elements, as it is generally admitted ; • the surface is sufficiently close-packed to assume that the second layer does not contribute to the signal. Moreover, as the spectrum was recorded within a few seconds and with a very low incident ion current density (10 nA scanned over 1 mm2), sputtering effects during the experiment are believed to be negligible. Indeed, a few successive spectra can be recorded without any changes. The silicon surface concentrations are the following: 54% after ion sputtering, 68% after annealing. Many experiments have been performed giving closely matched values, with a maximum discrepancy of 3%.
4.2. XPS experiments
Si Concentrot~on (°l.)
p r o f l t e
area Si 2p O'cu Tcu area Cu 3p asi Tsi
S[ Concentration (%)
40 50 60 (Degree)
XPS measurements performed at various angles 0 with respect to the surface allow a determination of the ratio R(O) defined as:
Depth profiLe (,~)
where cr and T are the Scofield photoemission cross section of the considered core levels and the transmission coefficient of the analyzer, respectively. The Si2pl/2+3/2 (Cu3p~/2+3/2) core-level photoemission peak areas were calculated after a non-linear background subtraction. For this, we used a Shirley-type background . In fig. 2 are shown the variations of R(O) with O for the Ar+-cleaned sample (a) and the annealed sample (b). One can clearly see that R(O) increases when O decreases, this indicates that the surface is silicon-enriched. Moreover, R(O) is significantly higher for the surface annealed at 700 K, indicating that the annealing induces silicon segregation to the surface. In order to get a quantitative concentration depth profile, we fitted our XPS data with a three-parameter model: the variables X 1, X 2 and
40 50 60 (Degree)
Fig. 2. R(O) ratio as defined by eq. (1), given by XPS data collected at various exit angles O (using the A1Ka line: hu = 1486.6 eV). In the insert are given the calculated in-depth profiles of the Si concentration. Depth is taken with respect to the surface. (a) After 3 keV Ar+-ion bombardment. (b) After annealing at 700 K.
X 3, which are the silicon concentrations in the first, second and third layers. The interlayer distances are supposed to be the same (d12 = d23 = 2 .~). The mean-free-path value has been chosen following the TPP2 calculation reported b~, Tanuma et al. , i.e. hcu = 16 ,~ and Asi = 26 A for about 1400 eV kinetic energy. Then, by assuming that the mean-free-path in the Cu3Si compound is a mean value weighted by the respective concentrations of Cu and Si, one obtains:
Y.. Samson et al. / On the surface segregation of silicon in Cu3Si
4.3. Auger experiments
Kinetic energy (eV)
Fig. 3. Low-energy region of AES spectra of Cu3Si after (a) 3 keV Ar÷-ion bombardment, (b) further annealing at 700 K. The primary energy was 2.5 keV and the applied modulation to the CMA was 2 V peak to peak.
Acu3si = 18 ,~ for both the Si 2p and Cu 3p lines, as they are very close in kinetic energy (1387 and 1413 eV, respectively). In fact, with a so large value of A, the analyzed depth is quite large even at the lowest exit angles and a small set of X 1, X 2 and X 3 values gives quite satisfactory fits. However, the maximum divergence between the calculated concentrations does not exceed 10%. Among the few acceptable depth profiles, those presented here (fig. 2) are the most consistent with the data obtained in the LEIS analysis. Results are summarized in the table 1.
The annealing of the sample at about 700 K induces a significant increase of the amplitude of the SiLVV(92 eV) and SiKLL(1619 eV) peaks with respect to the copper peaks (CuMVV(60 eV) and Cu LMM(920 eV)). This is a good indication of the silicon segregation to the surface, which is in agreement with data reported by Frank and Falconer . In the new chemical environment generated by annealing, the electronic state of the surface silicon differs notably from the bulk one. Indeed, the observed splitting of the SiLVV peak is strongly weakened after annealing as evidenced in fig. 3b compared to fig. 3a. This is easy to understand if one remembers that the splitting is due to hybridization of the silicon 3s-3p band with the d-band of copper [16,17]. As the silicon valence band is involved, this could have noticeable consequences for the silicon reactivity. Furthermore, photoemission experiments (performed with the use of synchrotron radiation provided by the super ACO ring at LURE), indicate clearly that the Si2p line shape moves away from the silicide one and comes closer to that of pure silicon one after heating .
4.4. Discussion According to the results displayed in table 1, both LEIS and XPS results show that the surface exhibits a significant silicon enrichment after ion bombardment at room temperature. However, one may ask if it is an artefact due to preferential sputtering of copper during ion bombardment. In favour of such a hypothesis, it must be remembered that copper distinguishes itself by a remarkably high sputtering yield (about five times more copper atoms by Ar impact than silicon
Table 1 Silicon concentrations in the top surface layers given by LEIS and XPS experiments performed at various exit angles
First layer Second layer Third layer
After 3 keV Ar+-ion bombardment (at% Si)
After annealing at 700 K (at% Si)
56 47 21
74 59 30
Y. Samson et al. / On the surface segregation of silicon in Cu3Si
atoms for the pure components at 3 keV ). Thus, preferential sputtering of copper on the Cu3Si surface would not be surprising. However, Falconer and coworkers seem convinced of the absence of preferential sputtering. To support their opinion, they cited a study of Auger emission induced on copper-silicon alloys by Ar+-ion bombardment [20,21]. Yet, the data do not appear conclusive. It seems that the ion-bombarded surface cannot be used to investigate the surface equilibrium at room temperature. Moreover, silicon shows a clear tendency to segregate with annealing at 700 K. As evidenced by XPS, the segregation extends below the top layer and the silicon concentration is rapidly decreasing after the second one. This implies that the ultimate surface is far more silicon-enriched than deduced from the AES experiments (either published  or performed by us). Indeed, AES does not separate the different contributions to the signal coming from the first atomic planes. In fact, the annealing leads to a surface composition remote from the bulk one at the temperatures involved in the direct synthesis (about 70 at% Si in the top layer instead of 25% in the bulk).
5. Surface composition: tentative comparison with predictions The surface segregation can be described in terms of an exchange reaction between bulk (Si b and Cu b) and surface (Si s and Cu~) atoms: Sib + Cus <--'Sis + CUb.
This yields an expression of the form: [Sis] [Sib] ( -AGseg ) Cus---[ ~ = [Cub----~ exp RT '
which relates the Si and Cu atom fractions in the surface and in the bulk, driven by the excess segregation enthalpy AGs~8. AGs~g can be tentatively estimated from thermodynamic values of the bulk alloy, by using the so-called broken-bond model to estimate pair interactions. It can be divided into contributions involving: (i) the difference in surface tensions of both elements,
(ii) some elastic strain energy contribution, (iii) the terms containing the excess mixing free enthalpy of the components . For the present system, Cu and Si are found to have similar radii, since Si is considered as having a quasi-metallic character (Rat(Si)= 1.32 A and R a t ( C u ) = 1.28 ~,). Consequently, the elastic strain energy term does not have to be considered, and the expression of dGseg can be written as follows: 2 AGmix aGsog = ( ~ s i - ~cu),O +
X[Ze(Xb--Xs) + Z v ( X b - 1/2)],
with AGmi x = AHmi x --
T AS x~,
where X b and Xs are the silicon concentration in the bulk and at the surface, respectively, Z is the number of nearest neighbours in the lattice, and Zf and Z v are the number of neighbours situated in the same plane and in the layer above (or below) that of the considered atom, respectively. As a result: Z = Z t + 2Z v. w represents the surface area occupied by the atoms, and y is the surface tension. For a high-index crystal surface, Tyson and Miller  proposed that to= 1.612N1/3V2/3 gives a reasonable value of the mean surface occupied per surface atom (N is Avogadro's number and V is the molar volume). For calculations, we will consider a V 2/3 value of 3.95 cm 2, i.e. a mean value between pure Cu and pure Si . The numerical values for y(0) (i.e. 3' at 0 K) are 1.85 and 1.29 J//m 2 for Cu and Si, respectively . With such numerical values, the AG 1 = ( T s i yc,)to term is negative and has a rather large amplitude: 30 kJ mo1-1. It is so that the Si component is expected to segregate to the surface. More difficult is a precise calculation of the second term of the segregation free enthalpy AGseg (eq. (4)). Firstly, to our knowledge, the heat of mixing AHm~ is not available for Cu3Si. However, one can roughly estimate that value by comparison with similar silicides. FeSi seems a good choice: as the few thermodynamic data
Y. Samson et al. / On the surface segregation of silicon in Cu 3Si
available for both FeSi and CuSi are very close, one could reasonably expect the A H and A S xs to be similar. For FeSi, A H ( T = 700 K) and A S x~ are - 9.6 kJ mol-1 and 1.864 J K - 1 m o l - l , respectively , which leads to a AGmi x value of - 8 . 3 kJ mo1-1 at 700 K according to eq. (5). Secondly, the structure of Cu3Si is complex : indeed, in this structure, each atom has eight nearest neighbours (situated at 2.45 + 0.02 .~) but six other atoms are located very near after (at about 2.86 + 0.02 ,~). Nevertheless, whatever the value of Z, 8 or 14, and the chosen value for Z v, 2 or 6, the second term in eq. (4) is calculated to contribute to less than 15% with respect to the A G 1 tenn. Consequently, the AGseg(T = 700 K) value is estimated to be - 3 0 + 5 kJ mo1-1 and the Si component is expected to segregate at the surface, as observed experimentally. In fact, with a so large value of AGseg, one could expect a quasi-pure silicon surface layer in the one-layer segregation model.
6. Conclusion The combined use of XPS and LEIS has provided a good image of the Cu3Si surface equilibrated at 700 K: • the surface is strongly enriched in silicon: about 70 at% Si in the top surface layer; • the silicon enrichment decreases rapidly in depth and the third layer composition is close to the bulk one. As evidenced by the performed calculations, the difference between the surface tensions of silicon and copper is the main driving force involved in the segregation process. As a result of the segregation, the surface silicon environment is far less metallic than for the nominal copper silicide: the electronic state of surface silicon differs significantly from that of the silicide and comes closer to the pure silicon one. Furthermore, the silicon propensity to segregate to the surface could allow the continuous replacement of the silicon consumed by reaction with the methyl chloride during the direct synthe-
sis, and thus partly account for the ability of copper silicides to catalyze the reaction.
References  R.J.H. Voorhoeve, in: Organohalosilanes: Precursors to Silicones (Elsevier, Amsterdam, 1967).  E.G. Rochow, J. Am. Chem. Soc. 67 (1945) 963.  R.J.H. Voorhoeve, J.A. Lips and J.C.V. Vlugter, J. Catal. 3 (1964) 414.  T.C. Frank, K.B. Kester and J.L Falconer, J. Catal. 91 (1985) 44.  T.C. Frank and J.L. Falconer, Appl. Surf. Sci. 14 (198283) 359.  W.F. Banholzer and M.C. Burrell, Surf. Sci. 176 (1986) 125.  T.C. Frank, K.B. Kester and J.L. Falconer, J. Catal. 95 (1985) 396.  A. Hirachi, A. Shimizu, M. Iwami, T. Narusawa and S. Komiya, Appl. Phys. Lett. 26 (1975) 57.  J.C. Bertolini, J.L. Rousset, P. Miegge, J. Massardier, B. Tardy, Y. Samson, B.C. Khanra and C. Creemers, Surf. Sci. 281 (1993) 102.  G. Weber, These de Doctorat d'Etat, Universit6 de Bourgogne (1988).  K.P. Mukherjee, J. Bandyopadhyaya and K.P. Gupta, Trans. Am. Inst. Min. Eng. 245 (1969) 2335.  R. Hultgren, P.A. Desai, D.T. Hawkins, M. Gleiser and J. Kelley, in: Selected Values of the Thermodynamic Properties of Binary Alloys (American Society for Metals, Metals Park, OH, 1973).  P. Bertrand, Analusis 14 (1986) 431.  P.M.A. Sherwood, in: Practical Surface Analysis, Eds. D. Briggs and M.P. Seah (Wiley, New York, 1983).  S. Tanuma, C.J. Powell and D.R. Penn, Surf. Interf. Anal. 17 (1991) 911.  A. Hiraki, S.C. Kim, W. Kammura and M. Iwami, Appl. Phys. Lett. 34 (1979) 194.  A. Taleb-Ibrahimi, V. Mercier, C.A. S6benne, D. Bolmont and P. Chen, Surf. Sci. 152/153 (1985) 1228.  Y. Samson, J.C. Bertolini, B. Tardy and G. Laroze, to be published.  G.P. Chambers and J. Fine, in: Practical Surface Analysis, Vol. 2, Eds. D. Briggs and M.P. Seah (Wiley, Chichester, 1992).  M. lwami, S.C. Kim, Y. Kataoka, T. Imura, A. Hiraki and F. Fujimoto, Jpn. J. Appl. Phys. 19 (1980) 1627.  A. Hiraki, S.C. Kim, T. Imura and M. Iwami, Jpn. J. Appl. Phys. 18 (1979) 1767.  W.C. Tyson and W.A. Miller, Surf. Sci. 62 (1977) 267.  F.R. de Boer, R. Boom, W.C.M. Mattens, A.R. Miedema and A.K. Niessen, Cohesion in Metals: Transition Metal Alloys (North-Holland, Amsterdam, 1988).  J.P. Solberg, Acta Cryst. A 34 (1978) 684.