36-himpsel-prb-1988-6084 - PHYSICAL REVIEW B VOLUME 38,...

Info iconThis preview shows pages 1–13. Sign up to view the full content.

View Full Document Right Arrow Icon
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Background image of page 2
Background image of page 3

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Background image of page 4
Background image of page 5

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Background image of page 6
Background image of page 7

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Background image of page 8
Background image of page 9

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Background image of page 10
Background image of page 11

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Background image of page 12
Background image of page 13
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: PHYSICAL REVIEW B VOLUME 38, NUMBER 9 15 SEPTEMBER 1988-11 Microscopic structure of the SiO/ Si interface F. J. Himpsel, F. R. McFeely, A. Taleb-Ibrahimi, and J. A. Yarmoff‘ IBM Research Division, Thomas J. Watson Research Center, PO. Box 218, Yorktown Heights, New York 10598 G. Hollinger Laboratoire d ’Electronique Automatique et Mesures E Iectriques, Ecole Centrale de Lyon, Bofte Postale No. 163, F—69I3I Ecully Cédex, France (Received 7 March 1988) The bonding of Si atoms at the SiOz/Si interface is determined via high-resolution core-level spectroscopy with use of synchrotron radiation. All four oxidation states of Si are resolved, and their distribution is measured for Si(lOO) and Si(lll) substrates. For oxides grown in pure 0;, the density of Si atoms in intermediate oxidation states is (1.5:0.5)>< 1015 cm‘z. This value is obtained by measuring the core-level intensity, the escape depth in Si and SiOz, and the relative Si 2p photo- ionization cross section for different oxidation states. From the density and distribution of intermediate-oxidation states, models of the interface structure are obtained. The interface is not abrupt, as evidenced by the high density of intermediate~oxidation states (about two monolayers of Si) and by their nonideal distribution. The finite width of the interface is explained by the bond- density mismatch between SiOz and Si. The electrical properties of the interface (band lineup, Fer- mi, and vacuum level) are determined. Annealing in H2 is found to influence the electrical parame— ters by removing the Pb centers that pin the Fermi level. The distribution of oxidation states is not affected. I. INTRODUCTION The SiOZ/Si interface has been the subject of intense study“4 because of its dominant role in silicon technolo- gy. The structure of this interface has been elusive despite many efforts to come up with models. Previous studies generally agree in identifying two distinct regions. The near interface consists of a few atomic layers con- taining Si atoms in intermediate oxidation states, i.e., Si1+ (SiZO), Si2+ (530), and Si3+ ($203). A second region extends about 30 A into the SiO2 overlayer. The Si02 in this layer is compressed because the density of Si atoms is higher for Si than for SiOZ. In this work, we will focus onto the near-interface region. Several structural models have been proposedS"ll for SiOz on Si(100), each predict- ing a characteristic distribution of oxidation states. Most of the models published to date assume an atomically abrupt interface. From our data we can exclude such abrupt models, since they cannot explain the large por- tion of Si}+ observed at the interface. A recent calcula- tion arrives at an extended-interface model for SiOz/Si(100) by minimizing the strain energy.11 This model is found to be in good agreement with our data. It is not unique, though. New models are proposed for SiOz/Si( 100) and SiOz/Siflll) based on the distribution and intensity of intermediate-oxidation states. These models are characterized by an extended interface, with protrusions of Si3+ reaching about 3 A into the SiOz overlayer. Many experimental techniques have been used previ- ously to determine the structure of the interface, its ex- tent and its roughness, e.g., transmission electron micros- 38 5,9,12— 14,15 copy 14 (TEM), scanning tunneling microscopy (STM), low-energy electron diffraction16 (LEED), posi- tron annihilation,17 ellipsometry,18 vibrational spectrosco- py,19’20 Rutherford backseattering21 (RBS), x-ray scatter- ing,22 field ion microprobe,23 Auger spectroscopy,”25 and x-ray photoelectron spectroscopy (XPS).26_43 The results range frogn atomically sharp to extended inter- faces of about 7 A width. These variations could be due to the different probing techniques or to the differences in sample preparation. We have ascertained that our ob- served interface properties hold for a wide class of oxides by covering a broad range of preparation conditions. A unique interface structure is found for the “ideal” SiOz/Si interface, i.e., an interface in thermal equilibrium (see Ref. 44) and free of impurities, such as hydrogen or OH. The preparation conditions play a minor role as long as one starts with clean and smooth Si surfaces. The experimental technique employed in this work is core-level spectroscopy (for a review of core—level spec- troscopy at Si surfaces, see Ref. 45). It has the advantage of resolving the oxidation states”42 of Si atoms in the in- terface layer via the chemical shift of the Si 2p core level. Therefore, a straightforward definition of the interface can be given by taking the layer that contains Si atoms in intermediate-oxidation states. The depth resolution is optimized by using tunable synchrotron radiation. At a photon energy of 130 eV, the escape depth of photoelec- torons from the Si 2p core level hasoa minimum of about 3 A in Si, compared with about 15 A in conventional XPS. The short escape depth makes it easy to detect a mono- layer of Si atoms at the interface. For a quantitative eval- uation of the core-level spectra one has to determine the 6084 ©1988 The American Physical Society 38 MICROSCOPIC STRUCTURE OF THE SiOz/Si INTERFACE escape depth in SiO2 as well as that in Si (SiO2 has about twice the escape depth of Si). Also, the difference in the photoionization cross section between the various oxida- tion states has to be taken into account (the higher oxida- tion states have about twice the cross section of elemental Si at hv: 130 eV due to a resonance effect). In this work we have determined these parameters for a range of pho- ton energies (120—400 eV), where there is very little infor- mation available. II. EXPERIMENT The data were taken with a spectrometer-mono- chromator setup at the National Synchrotron Light Source.45 The photoelectrons were collected within a cone of 86° full width centered around the sample nor- mal. The light was p polarized with an angle of incidence 64° from normal. The energy resolution of the electron analyzer was set between 0.1 and 0.2 eV, and the photon energy resolution varied from 0.2 eV at hv: 130 eV to 0.4 eV at hv=400 eV. Thin SiO2 films were obtained by oxidizing atomically clean Si surfaces in pure 02. Clean Si(111)7><7 and Si(100)2><1 surfaces were prepared by Ohmic heating of Si wafers to 1050 °C or by dipping in 10% HF and heat- ing to 850°C. Well-oriented wafers ( < y misorientation) were used to obtain the data shown here, but wafers with 3° misorientation gave similar results. Using an ultrahigh vacuum transfer system, the clean Si samples were transferred from the spectrometer (pressure < 10‘10 Torr) to a preparation chamber which could be filled with gases up to a pressure of 1 atm and subsequently pumped down to ultrahigh vacuum. During all oxygen exposures, the sample was cooled while still in oxygen. Thereby, the formation of holes in the SiO2 layer is prevented, which occurs when a thin SiO2 film is heated in vacuum (see Ref. 46). Vacuum-annealed films with pinholes exhibit a larger signal from bulk Si. The density of Si atoms in intermediate-oxidation states is underestimated when us- ing such data. The native oxide as well as oxides grown in H20 and by wet chemistry (HNO3, H202) have also been studied but in a less systematic way. Figure 1 illustrates the analysis of the data for a core- level spectrum from SiOZ/SiUOO). Oxidation states inter- mediate between Si and SiO2 are already visible in the raw data (top curve in Fig. 1). For a quantitative evalua- tion we apply three straightforward data-processing steps. The first is to subtract a secondary electron back- ground (dashed line in Fig. 1) from the raw data. This background curve is measured separately at a lO-eV— lower photon energy where the Si 217 core-level structure is shifted away. A small background remains at the low- kinetic—energy side of the spectrum due to energy losses from the core lines. This background is removed in a second step by subtracting a curve proportional to the in- tegral of the spectrum. Such a procedure is equivalent to approximating the loss function by a constant. After re- moving the total background (dotted line in Fig. 1) the spectrum is decomposed into the Si 2p1/2 and Si 2p3/2 6085 —7——T‘—i'_-l—~—_T_——i—_T‘__i_j Si 2p SIHOO) hl' : 130 eV Oxidized in 02 Photoemnssnon lntensny ( Arb. Units ) -107 #106 —105 —104 —103 —102 —101 —1OO —99 -98 Initial-State Energy (eV relative to EF) FIG. 1. Intermediate-oxidation states at the SiOz/SillOO) in- terface, identified by their Si 2p core-level shifts. The top curve represents the raw photoemission data for the Si 2pI ,u/z core levels. The bottom curve has the Si 2111 a line and the secondary electron background subtracted. All three intermediate- oxidation states are seen. For a truncated bulk structure only Si“ would be present since the SillOO) surface has two broken bonds per atom. spin-orbit partner lines. This decomposition is mathematically unique as long as the spin-orbit splitting and the intensity ratio are known. The splitting of 0.61 eV is an atomic property and practically independent of the chemical environment. A variation of the splitting between 0.59 and 0.61 eV has been observed by fitting the bulk Si lines. The 2p1 ,2 to 2P3,2 intensity ratio comes out in these fits to be equal to the statistical value of 1:2 within i2%. In all following figures the spin-orbit decomposed curves will be shown for clarity. III. CORE-LEVEL INTENSITIES, CROSS SECTIONS, AND ESCAPE DEPTHS In order to obtain the number of Si atoms in intermediate-oxidation states and to test structural mod— els it is necessary to quantitatively evaluate the core-level intensities. The photoemission intensity is determined by six quantities, i.e., the atomic photoionization cross sec- tions Us,— and 05102, the escape depths [Si and [$02, and the density of Si atoms nsi and nSiO2 in Si and SiOZ, re- 6086 spectively. Our experimental results for the relevant quantities are collected in Table I together with work at XPS energies.26’30’37'74 For Si in intermediate oxidation the results fall in between the values for Si and SiOZ as shown for the cross sections in Table II. The core-level intensity I Sioz for a Si02 overlayer of thickness d is obtained by integrating over the exponen- tial escape probability: d ISiOZ ~"5i0205iozf0 eXP( '2 /15102)dz ="51020510215102[1—CXP(*d/ISiozn- (1) The emission from the Si substrate is given by a similar expression except that the escape probability is multiplied by the attenuation factor exp( —d/15i02) of the overlayer: ISi~nSiUSiexp( ‘d/15102)f0wexp( —z’/lSi)dz’ ="SiUSilSiexp(—d/13ioz) . (2) Combining (1) and (2) we obtain, for the intensity ratio between the Si02 and Si core-level peaks, ISiOZ I0° [Si =I[6Xp(+d/ISiOZ)—'1] , where 1m = "$1020 $1021 $102 and F. J. HIMPSEL et al. 38 10 ="51051151 are the intensities for an infinitely thick SiO2 layer and for a bare Si substrate, respectively. The experimental in- tensity ratio 10,, /I0 is also given in Table I. It reverses between low and high photon energies. This reversal is due to a near cancellation of the factors contributing to I w /Io. One finds USiOZZO'Si and ISioz > [Si for the pho- ton energy range considered here, which is opposite to the density ratio n5i02<nsi. At low photon energies (hv: 130 eV) a resonance in O'Sioz tips the intensity bal- ance in favor of SiOz; at high photon energies the low density of Si atoms in SiOz dominates over the other fac- tors. To determine the escape depth in SiOZ we use a layer of known thickness d and solve Eq. (3a) with respect to I a : SIO2 [Sioz In The thickness is determined by ellipsometry on SiOZ films of 10—30 A thickness and checked with transmission electron microscopy (TEM). The result for the escape depth at hv=200 eV agrees with a previous determina- tion,25 which used the attenuation of Si 2p Auger elec- trons by a Si02 overlayer and calibrated the Si02 thick- ness by nuclear analysis of the oxygen content. For determining the escape depth in Si we use the in- tensity of shifted surface core levels at clean Si surfaces.47 15:02 10 — 1 15.1..+ . (3b) TABLE I. Experimental values of the parameters required for a quantitative description of Si 2p core-level intensities (see Sec. III). The kinetic energy of the photoelectrons is 104 eV (108 eV) smaller than the photon energy for Si (SiOz). The relation [Eq. (3a)] I w /Io=nSi0205i0215i02 /n5i05i15i allows a consistency check, which gives a typical accuracy of :t10%. The results for the photon energy range up to 400 eV are from this work, the XPS results are from Refs. 30, 37, and 74. Bulk atom density: HSiOZ=2.28X1022 CHI—3, n5i=5.00)<1022 CHI-3. N111 =7-8X 10” cm‘z, N1m=6.8><10” cm‘z. Surface atom density of Si (1 monolayer): w "$020 310215102 astoza a b a b hv (eV) 70— 05i 15102 (A) ISi(A) 120 1.8 1.7 1.8“ 8.5d 4.o= 130 2.1 2.2 2.2c 7.1d 3.3= 145 1.4 1.7 2.0c 6.3d 33° 200 0.84 0.77 1.3c 6.5,d 6.5f 5e 400 0.69 11d 10C 1254 0.823 1.1g 21,15 25k 13,8 23k 1487 0.823 1.18 26,8 37h 16,5 27,1] 26" 36-45,i 27k 3Only the adiabatic peak is considered (not the satellites). l’The escape depth is averaged over a 86° emission cone around the normal. cFrom absolute core-level intensities of monolayer oxide films (Ref. 57). dFrom the intensity ratio between oxide and substrate for samples of known thickness using Eq. (3b). °From the surface core-level intensity for Si(100)2>< 1 assuming half a monolayer with shifted core lev- els (asymmetric dimer model). Compare Eqs. (4) and (5) for the analysis. rFrom Derrien and Commandre (Ref. 25). 8From Hochella and Carim (Ref. 37). 1'From Hill et a1. (Ref. 30). iFrom G. Hollinger et al. (Ref. 74). kFrom Flitsch and Raider (Ref. 26). TABLE II. Energy positions, widths and cross sections for intermediate—oxidation states. The width quoted is the Gauss- ian FWHM due to inhomogeneous broadening. The instrumen- tal resolution is subtracted quadratically, and the lifetime broadening is accounted by a Lorentzian of 0.1 eV FWHM. Typical variations are i5% for the energy positions and i10% for the widths. Energy (eV) FWHM (eV) (7/05, (hv= 130 eV) Si0 0 0.28 l Si1+ ~O.95 0.44 1.02‘ Si“ —1.75" 0.58 1.1c Si” — 2.48b 0.66 1.7c Si4+ —3.9b 1.15 2.2c aFrom a comparison of surface core-level intensities for HIO/Si(lOO)2><l with clean Si(100)2><1 (compare Figs. 2 and 3). bThe Si4+ core level moves up for films thinner than 5 A due to a smaller valence—band ofl‘set and extra screening by the Si sub- strate. The Si3+ and Si2+ core levels move down for films thinner than 5 A due to the absence of dielectric screening by the SiOz overlayer. CFrom absolute core-level intensities of monolayer oxides films (for details see Ref. 57). For this purpose, we calculate the intensity from the outermost Si layer relative to the total core-level emis- sion. It is convenient to sum the intensities layer by layer rather than integrating over a continuum as in Eqs. (1) and (2). [In a continuum model one runs into difficulties in defining the boundaries between layers when the layers are not equally spaced, such as for Si(111). With equidis- tant layers the continuum and the discrete models give identical results, e.g., for Si(100).] Between layers one has an attenuation factor 4hk1=eXP( *dth/l )’ [=ISi’lSi02’lSi0X ’ (4) with an interlayer spacing dlmza/4= 1.36 A for Si(100) and two alternating interlayer spacings d,“ =a\/3/ 12 =0.784 A and d'm=3dm for Si(lll). The intensity from all Si layers can easily be summed up as a geometric series Efzoqfiklz l/(l—qhkl). The intensity of the sur- face layer relative to the total intensity is R100=(1—qioo) (5a) fora Si(100) surface, R1”=(l——q‘f11 for Si(lll) truncated between double layers (one broken bond), and Rin=(1—qi11)/(1+‘Iin) (5C) for Si(lll) truncated inside a double layer (three broken bonds). If one wants to apply a continuum model to the Si(lll) surface one has to use an average layer spacing '1'“=2dm in order to integrate over slabs of equal thickness. The resulting ratio 38 MICROSCOPIC STRUCTURE OF THE SiOz/Si INTERFACE 6087 Ri'11=(1—qi11) (5d) applies to both types of Si(lll) surfaces. This value differs by up to :f:lO% from Eqs. (5a) and (5b) at the minimum escape depth. Surface core-level data are shown in Fig. 2 for clean Si(100)2><1 and Si(lll)7><7. For Si(100)2><1 there is a surface core-level emission at 0.5 eV above the bulk line with the intensity ratio R =surface/(surface+bulk) =0.17 at hv=130 eV. The surface emission is assigned to half a layer of outer dimer atoms using the asymmetric dimer model.47_49 For Si(111)7>< 7 there is a surface line at 0.8 eV above the bulk line with the intensity ratio surface/(surface+bulk)=0.05 at hv=130 eV. It is as- signed to % of a monolayer of so-called rest atoms, which are negatively charged49 according to scanning tunneling spectroscopy50 and self-consistent calculations.51 Apply- ing Eqs. (4), (5a), and (5b) to these intensity ratios one ob- tains an escape depth of 3.3 A (3.2 A) from the dimer atoms (rest atoms), respectively. Well—defined Si core- level shifts are also found for adsorbate overlayers and in- terfaces, e.g., As/Si(111) with an intensity ratio R =surface/ (surface + bulk)=0. 40 (Ref. 52), and CaFZ/Siflll) with a surface/ (surface + bulk) ratio of 0.39—0.46 (see Fig. 3 and Refs. 53—55). [{sing Eq. (5°C) [Eq. (5b)] one obtains escape depths of 3.4 A (2.9—2.1 A) for As/Si(111) [Can/ Si(111)]. However, the analysis of Clean Si Surfaces Si(111)7x7 Si 293/2 hv = 130 eV ( Arb. Units ) Intensity Photoemission —3 -2 —1 0 1 2 hind-State Energy (eV relative to bulk Si2p3/2) FIG. 2. Core-level spectra from clean Si(100)2><1 and Si(lll)7><7 surfaces used for determining the escape depth in Si. The emission from shifted surface core levels correspond to g and § of a monolayer, respectively. Energy-loss features below the bulk lines are not included in the fit. 6088 the core-level intensities may be affected in this case by a change in cross section due to chemical bonding (see below). In summary, the escape depths for SiO2 and Si exhibit a minimum for photon energies around hv=140 eV. The escape depth for SiO2 is longer than that for Si, in agreement with theoretical estimates.56 In addition to the escape depth there is another factor that influences the core-level photoemission intensity, i.e., the photoionization cross section. Previously, it has been assumed that the cross section is identical for all oxida- tion states of Si at a given photon energy. It turns out that this is far from being true for the photon energy range in question. Near hv= 130 eV there is a shape res- onance in the cross section for the higher oxidation states which causes them to be enhanced by up to a factor of 2. This cross section resonance diminishes as one ap- proaches lower oxidation states (corresponding to lower core-level shifts). Therefore, we expect only minor cross section effects for the clean Si surface core levels, for As/Si(111) and for CaFZ/Si(111). Our results are summa- rized in Tables I and II. The cross section ratios are ob- tained by photon-energy-dependent measurements of the absolute core-level intensities of Si1+,Si2+,Si3+,Si4+. Thereby we use very thin films (monolayer or less) in or- der to minimize the effect of the photon-energy- dependent escape depth. The Si1+ cross section is re- ferred to the Si cross section by comparing the Sier peak at the HZO-exposed Si(100) surface with the surface core-level emission at the clean surface (see discussion of Figs. 2 and 3). A detailed account of the cross-section measurements will be published elsewhere.57 IV. NUMBER OF INTERFACE ATOMS An important quantity is the number of N S0 of Si atoms in intermediate-oxidation states (per unit area). In order to get a rough idea we compare in Fig. 3 core-level spectra of two reference structures with SiOZ/Si(1ll). The Ca-terminated Can/Si(1 l 1) interface”‘55 has about a monolayer of Si atoms bonding to Ca giving rise to a shifted core level located 0.4 eV above the bulk line. Some residual Si—F bonds show up on the low-energy side of the bulk line. For our purpose the HZO-exposed Si(100)2><1 surface is better suited since it takes a possi- ble change in cross section from elemental Si to oxidized Si into account. This surface exhibits a surface core level at 0.9 eV below the bulk line corresponding to 0.5 mono- layers (Ref. 58) of Si”. The intensity measured for satu- ration coverage (about 20 L) is Si1+/(Si0 + Si1+)=0.17. It has been assigned59 to 0.5 monolayers of OH bonding to Si. The other half of the Si surface atoms bond to H and exhibits a much smaller shift of 0.3 eV. They have been included in the bulk line. The Si1+ peak for SiOZ/Si(111) has an intensity ratio SilJr/(Si0 + Si”) =0.30 at hv= 130 eV, i.e., 1.8 times greater than for the HZO/Si(100) reference surface. From there we determine a coverage of 0.87 monolayers of Si1+ for SiOZ/Si(lll). Thereby the small change in the surface-to-bulk ratio due to the difference surface orientation has been taken into account.60 The coverage of the higher-oxidation states is more difficult to determine since their intensity is F. J. HIMPSEL et al. 38 Comparison with Reference Surfaces Sizpa/z hv = l3oeV CaF2 on Si(111) H20 on Si(100) 2 layers SiOx Photoemission Intensity (Normalized to Bulk Line) SiO2 on Si(1 1 1) Initial-State Energy (eV relative to bulk SIZpa/z) FIG. 3. Comparison of core-level spectra from SiOZ/Si(lll) with the Can/Si(111) interface and the HZO-exposed Si(100)2 X 1 surface. A density of two Si layers is derived for the Si atoms in intermediate-oxidation states by comparing with the HZO/Si(100) reference surface. enhanced relative to Si1+ at a photon energy of 130 eV. Two effects contribute to this enhancement. One is a res- onance in the cross section of Si3+ at this photon energy (see discussion in Sec. III and Table II), the other is an es- cape depth effect caused by the fact that Si3+ is farther away from the Si substrate than Si” (see Sec. V). In or- der to become free of these distortions it is helpful to use data at higher photon energies (h v=400 eV in our case) where the cross sections are nearly the same and the mean free path is significantly larger than the width of the interface. In this case the true distribution of oxida- tion states is seen. From the data for hv=400 eV in Table III we find that the total number of Si atoms in in- termediate oxidation states is 2.6 times that of Si”, i.e., 2.3 Si(lll) monolayers. This corresponds to 1.8)(1015 atoms/cmz. A somewhat smaller value of l.3>< 1015 atoms/cm2 (i.e., 1.9 Si(100) monolayers) is obtained for SiOz/Si(100) by comparing the (Si‘+ + Si2+ + Si3+)/Si° intensity ratios for Si(100) and Si(111) at hv=400 eV (see Table III). It is interesting to note that the SiOz/Si(100) interface also has a lower density of dangling bond de— fects (Pb centers61'62) than Si(lll). Thus, there seems to be a correlation between the density of intermediate— oxidation states and the density of Pb centers. The abso- lute number of Pb centers is 3 orders of magnitude lower, though. A similar correlation between the density of 38 MICROSCOPIC STRUCTURE OF THE SiOz/Si INTERFACE 6089 TABLE III. Relative intensities of intermediate-oxidation states for SiOz/Si(100) and SiOz/Si(lll). The intensities are obtained by a least-squares fit to the data with the energies and widths constrained to the values given in Table II. Typical variations between samples are i10% for 11+:12+:I3+, and up to 30% for I 0. In order to obtain the true distribution of oxidation states, the intensities have to be di- vided by the respective cross sections (given in Table II for h v: 130 eV). 13+ 12+ 11+ I0 hv (CV) Il++I2++13+ Il++12++13+ Il++12++13+ Il++12++13+ Si(lll) data 120 0.41 0.25 0.34 0.95 130 0.48 0.21 0.31 0.74 145 0.48 0.21 0.31 0.77 400 0.33 0.29 0.38 2.7 Si(100) data 130 0.48 0.28 0.24 1.1 400 0.37 0.35 0.28 3.6 intermediate-oxidation states and the electrically active interface states has been observed in the oxidation of SiGe alloys.63 There is an independent way to obtain the number of intermediate-oxidation states by using only data taken at hv=400 eV. At this photon energy the depth distribu- tion of intermediate-oxidation states does not matter. It can be modeled by a constant density nSiox over a width 8. Equation (3a) can be generalized to give the intensity ratio between intermediate-oxidation states and the Si substrate: "510x aSiOx ISiO)‘ [ ( 8/1 ) 1] (6) =—————————— ex + i — . ISi "$051151 p 50" ISiOx For 5 << [Sio , the last factor becomes 8/lSiOx. The num- ber of Si atoms in intermediate-oxidation states per unit area is then ISiO‘ 05‘ 1 (7) [Si aSiOX "Si Si . NSiOx = "saox 5 = The experimental ratio is [$0 /ISi=(I1++I2+ +I3+)/ I°=o.37 and 0.28 at hv=400 eV for Si(111) and Si(100), respectively (see Table III). Assuming OSi/ agio :1 at hv=400 eV and using ISi=10 A, n5i=5 X1022 cm“3 (Table I) one obtains for the number of Si atoms in inter- mediate oxidation states NSiO = 1.9 ><1O15 cm“2 on Si(111) [NSiox =1.4>< 1015 cm-2 on Si(100)]. These values are comparable with the results obtained in the preceding paragraph from comparison with reference spectra. The accuracy is less due to a relatively large uncertainty in the determination of [Si at hv=400 eV and due to the possible contribution of a satellite emission at about 1.5 eV below the bulk Si line.64 It is clear from the two in- dependent measurements that the density of intermediate-oxidation states at the SiOz/Si interface is about two silicon layers, i.e., twice of that expected for an atomically abrupt interface. V. DISTRIBUTION OF OXIDATION STATES In the Si 2p core-level spectra shown in Figs. 4 and 5 one can clearly identify three peaks between the Si and the SiOz lines, which correspond to the three possible intermediate-oxidation states Si‘+,Si2+,Si3+. One should 5 ix oxide h» = 130 eV Si(1001 Photoemission Intensity (Arb. Units) Si(111) —7 as ~5—4—3 —2 —1 0 1 2 3 Initial-State Energy (eV relative to buk Sl2p3/2) FIG. 4. Core-level spectra from ultrathin Si02 overlayers on Si(100) and Si(111) surfaces. The Si(l 11) substrate has less Si2+ and more SiH than the Si(100) substrate. Si“ is enhanced by a factor of 1.7 at this photon energy due to a cross»section reso- nance (see Table II). The films were grown in 2X 10'5 Torr 02 at 750 °C for 20 sec. 6090 Si 2p3/2 14 is oxide hy = 130 eV gown in dry 02 Photoemnssron Intensity ( Arb. Units ) Si(111) —8 ‘7 —6 —-5 —4 —3 —2 ——1 O 1 2 Initial-State Energy leV relative to bqu 8in3”) FIG. 5. Core-level spectra from intermediate SiOz overlayers on Si(100) and Si(lll). The distribution of oxidation states is similar to that for ultrathin layers. The films were grown in 0.1 Torr 02 at 850°C for 10 sec. keep in mind that these symbols refer to the oxidation state, not to the actual charge transfer. The charge transfer is only about half an electron per oxidation state.27’65 The energy positions and the widths of the core lines are given in Table II. Figures 4 and 5 compare Si 2p3/2 core-level spectra for SiO2 films of different thickness. The absolute intensity of intermediate— oxidation states goes down with increasing film thickness due to attenuation by the SiO2 overlayer. For films thick- er than about 30 A the interface signal decreases below the detection limit. This observation rules out the ex- istence of Si atoms in intermediate-oxidation states at the surface. The intensity distribution of intermediate- oxidation states, I 1+:IZ‘L:I3+, does not vary much with the oxide overlayer thickness, showing that this is a universal property of the SiOz/Si interface. Results ob- tained from least-squares fitting are given in Table 111. Also, the intensity of intermediate-oxidation states rela- tive to that of the Si substrate (1°) is given. It is general- ly independent of the layer thickness. However, it can be affected in both ways by poor sample preparation. F. J. HIMPSEL et al. 3 Pinholes due to vacuum annealing of thin SiO2 films cause the Si signal to increase. Rough starting surfaces, on the other hand, produce more intermediate-oxidation states. Several processing parameters have been varied in or— der to test the universal nature of SiOz/Si interface. Pres- sure and temperature variations (from 10—5 to 20 Torr and from 700 to 1100°C, respectively) do not affect the distribution of oxidation states significantly. This finding holds for a film thickness of up to 30 A which is the upper limit for our probing technique. Thicker films may be studied by back-etching with HF (see Refs. 2, 29, and 33), but at the risk of altering the stoichiometry and creating holes with H-terminated20’66‘67 Si. For very thin oxide films (8 A or less) the intensity of intermediate- oxidation states is found to decrease relative to the bulk line although their distribution remains unchanged. It is likely that these films are not continuous, giving rise to extra emission from clean Si through pinholes. Oxidation at room temperature leads to distributions with a larger proportion of the lower oxidation states (Si1+ is the strongest component). This holds for oxidation in dry ox- ygen39’4”68 as well as in oxidizing solutions“; (e.g., HNO3). These films have a structure different from the high-temperature oxides. The models in Refs. 9 and 10 predict exclusively Si1+ at the interface, in qualitative agreement with the room-temperature data. We will not discuss room-temperature oxidation in detail, as we are mainly concerned with the structure obtained at elevated temperature in thermal equilibrium.44 It has been shown for very thin oxides41 that the room-temperature distribu- tion converts to the universal high-temperature distribu- tion after annealing at 700°C in vacuum. Recently, thicker oxide films have been grown at room temperature on very flat molecular-beam epitaxy (MBE) Si(100) sub- strates,9’68 which exhibit a dominant Si1+ state.68 When these structures are heated in vacuum, the abundance of higher oxidation states increases. Very rough or defec- tive surfaces produce a larger density of intermediate- oxidation states. Such surfaces are produced by heating Si( 100) above 1100°C, or in the presence of impurities. They are characterized by weak surface core-level signals. We have also looked into the effect of H2 annealing on the distribution of intermediate-oxidation states (see Sec. VII). There is no noticeable difference. This is under- standable since the number of dangling bonds (Pb centers) that can be saturated by hydrogen reaches at most about 1012 cm‘2 (Ref. 61), Le, fl of a monolayer. Such small changes are not detectable with our technique. The pres- ence of H2, e.g., in steam oxidation, has been reported2ng to give rise to an abrupt interface where 6—17 % of the interface bonds are saturated by H. In this case the higher temperature may cause the strong Si—O—Si bridge bonds to break up (compare also the discussion in Sec. VIII). Annealing in atomic hydrogen has also been reported2 to cause incorporation of substantial amounts of H in the interface. This may be explained by the high reactivity of atomic versus molecular hydrogen. In the following we will concentrate on the pure SiOz/Si inter- face that is formed in the absence of hydrogen. In Figs. 4 and 5 the distribution of intermediate- oxidation states is shown for two crystallographic orien- tations of the substrate. All three intermediate-oxidation states are present for Si(lll) as well as for Si(100), but their intensity ratios are different. On Si(IOO) the Si”, Si“, and Si“ states appear in roughly equal proportion, on Si(lll) the Si2+ state is suppressed. This crystallo- graphic trend can be explained by the bond topology of the truncated bulk structure. A Si(l l 1) surface can have two types of crystallographic planes with one and three broken bonds per atom, respectively, which give rise to Si1+ and Si3+ when saturated by oxygen. The Si(lOO) surface has two broken bonds per atom, which give rise to Si“. The coexistence of all three oxidation states on Si(lOO) clearly indicates deviations from the ideal, atomi- cally abrupt interface. In order to pin down the structure of the interface it is helpful to obtain a clue about the depth distribution of various oxidation states. This can be done by changing the escape depth, either by changing the take-off angle of the photoelectrons (see Refs. 30, 31, 35, and 36) or by varying the photon energy (see Fig. 6; compare also Refs. 28 and 32). At a photon energy hv: 130 eV the outer portion of the interface is enhanced due to the small es- cape depth, whereas at hv=400 eV it is given the same weight as the deeper regions of the interface. A least- squares fit to the data (see Table III) shows that, indeed, ( l | ( Photon Energy Dependence SiO2 / Si(111) Si 2p3/2 hl' : 130 eV *- 145 eV ---- -- 120 eV ———- 400 eV ---- -- PhotoemISSIon Intensity (Normalized to 8qu Peak) —8 —7 —6 —5 —4 —3 ~2 —1 0 1 2 FIG. 6. Core-level spectra vs photon energy hv for 14 A ox- ide on Si(l 1 1). At hv= 130 eV there is a minimum in the escape depth as evidenced by the maximum in the SiOz/Si intensity ra- tio. The Si3+ contribution is enhanced at this photon energy relative to Si1+ and Si“. The enhancement is partly due to the fact that Si3+ is closer to the surface and partly due to a cross section resonance of Si3+ at this photon energy. 38 MICROSCOPIC STRUCTURE OF THE SiOz/Si INTERFACE 6091 the Si3+ intensity increases significantly relative to Si2+ and Si“L at hv= 130 eV. Therefore, one would want to conclude that the Si3+ atoms must be located farther out than Si2+ and Si”, as expected from the kinetics of the oxidation process. However, the enhancement of Si” at hv= 130 eV is largely caused by a cross—section resonance (see Sec. III). After dividing the intensities by the cross- section factors given in Table II one obtains relative in- tensities of 13+:12+:Il+=0.36:O.24:0.40 at hv=130 eV for Si(l l 1). This is already very close to the ratios 13+;12+;1‘+=o.33:0.29:0.38 at hv=400 eV, where cross section and escape depth effects are absent. Therefore, it is diflicult to extract quantitative depth information. Re- cent depth-profiling work35 uses angle-dependent XPS data, which are transformed into a depth distribution by an inverse Laplace transform. Si3+ is found to be 6—10 A farther away from the interface than Si“, somewhat more than expected from our structural models. We note that previous depth-profiling workzs‘33 gives information about a larger length scale. A small concentration of Si3+ (a fewopercent) is found to be distributed over a depth of 30 A or more inside the SiOZ film. VI. INTERFACE STRUCTURE A coarse look at the distribution of oxidation states gives two qualitative results about the structure of the in- terface. (i) The interface is not ideal, as evidenced by the coexistence of all three intermediate-oxidation states Si1+,Si2+,Si3+. With a truncated bulk structure one would obtain only Si2+ for Si(100), and Si1+ or Si3+ for Si(l l 1). (ii) However, there is a remnant of the crystallo- graphic dependence expected from a truncated bulk structure. Can these findings be made consistent? In or- der to come up with sensible models we have to consider the possible driving forces for the interface structure. A critical boundary condition is the density mismatch be- tween SiOZ and Si. The SiOZ lattice has a density of Si atoms that is 2.2 times lower than that in Si. Conse- quently, a SiO2 surface has only half as many open bonds as a Si surface. In the following we will discuss how this difference can be accommodated. First, the SiOz/Si(100) interface is considered. The truncated bulk structure can be connected to an amor- phous SiOz network as shown by Pantelides et al.7 Such a model has a full monolayer of Si atoms in the 2 + oxi- dation state. An epitaxial model by Herman et al.6 (Fig. 7) has the density difference between SiO2 and Si built in. This model uses a diamondlike structure of SiOz, which is obtained by straightening the Si—O—Si bonds in the B- cristobalite structure. The bond-length ratio between Si—O—Si and Si—Si is close to 1/2. Therefore, an epit- axial (Vim/Em 45° structure can be constructed on Si(100). In this case, the bond density across the interface changes by a factor of 2. As shown in Fig. 7, only half of the atoms at the Si(100) surface are connected to oxygen in SiOz. To absorb the remaining broken bonds one has to introduce impurities such as H, F, and OH at the in- terface. This situation may exist in high-temperature steam oxidation or in wet chemical oxidation. Alterna- tively, oxygen atoms may form double bonds with the 6092 F. J. HIMPSEL et a]. g SiO2 / Si(100) : Abrupt interface 8302/ Si(100) + Model A 3 + + /‘ o 1 / o 2 /‘ c t 3 t/ k N. 2° Y 00 O O O O 6 O [100] L IO‘IO] FIG. 7. Epitaxial model of the SiOz/SiHOO) interface after Herman et ct]. (Ref. 6). The mismatch in bond density at the in- terface prevents half of the broken bonds at the Si(100) surface from being connected with SiOz. This density mismatch presents the major constraint in constructing structural models of the interface. free Si surface atoms. The distribution of oxidation states expected at such interfaces is 0.5—1.0 monolayer of Si2+ and O. 5-0.0 monolayer of Si”. Thereby, Si atoms bond- ing to hydrogen are counted as Sio. An alternative way to take care of the density mismatch has been proposed re- cently by Ourmazd et 01.9 Only one of the two broken bonds of the Si surface atoms is connected to SiOz. The other is presumed to pair up with a neighboring Si atom in a 2X 1 structure, as on the clean Si(100) surface. This configuration matches tridymite, another crystalline modification of SiOZ. The interface consists of a full monolayer of Si1+ for this model. In a variation of this model, one may obtain Si2+ by inserting oxygen into the Si—Si dimer bond. A third possibility to fix the bond density mismatch has been proposed by Ohdomari et al.‘°’11 The Si(lOO) surface is terminated by Si(lll) facets with one broken bond per atom, resulting in about a monolayer of Si” at the interface. All these atomically abrupt interface models give only Si“ and Si1+ as intermediate-oxidation states.70 A look at the bond to- pology of the Si(IOO) surface shows that the common de— fects (e.g., adatoms, vacancies, steps) also produce only Si2+ and Si”, but not Si”. In order to explain the ob- served Si3+ peak we have to resort to extended-interface models. A solution to the density-mismatch problem is a grad- ed interface which is 2—3 atomic layers wide (see Fig. 8). Strain energy calculations show that a properly graded SiOz/Si(100) interface11 has a significantly lower strain energy than an abrupt one. A specific model has been proposed by Ohdomari et al.11 based on strain-energy minimization (Fig. 8, model A). We add a second struc- ture (Fig. 8, model B), which agrees equally as well with our data. A peculiar feature common to these two mod- els is the occurrence of Si3+ protrusions into the SiOz lay- /\/\/\/\/\/\ lllllll \./ \./ \./ \./ \./ \./ Sic2 / Si(100) 3+ 3 ModelB 1+ l+ l \I/ he. 2+ \I/ 2. i. )/ /\/'\/\/'\/\/\ Homlllllll \o/ \./ \./ \./ \o/ \./ L” [01]] l FIG. 8. Models of the SiOZ/Si(lOO) and SiOz/Si(lll) inter- faces that match the distribution of intermediate-oxidation states. Model A is similar to a structure proposed by Ohdomari et al. (Ref. 11) based on strain-energy minimization. These models are characterized by Si3+ protrusions into the SiOz over- layer. Open bonds are to be connected to a SiOz network via oxygen bridge bonds. The bond topology is given without tak— ing relaxation into account. er. The spacing of Si3+ protrusions can be estimated by fitting the observed [Si3+]/[Si2+] ratio at hv=400 eV (see Table 111). It comes out to be somewhat more than two Si lattice spacings. It is interesting to look at this spacing in terms of a misfit dislocation picture. An areal density mismatch of 1.7 is obtained from the bulk densities of SiO2 and Si given in Table I, i.e., there are 1.7 times as many Si atoms per unit area as Si02 molecules. One can match this ratio by introducing dislocations in a linear array, such as for the models shown in Fig. 8. To get the proper density ratio one needs 2—3 Si lattice spacings be— tween the dislocations (two Si atoms for one SiOz would give a density ratio of 2.0, three Si atoms for two Si02 would give a density ratio of 1.5). This is close to the ex- perimental result, thereby supporting the view of Si3+ protrusions as the cores of misfit dislocations. The SiOz/Si(1 l 1) interface can be modeled in the same spirit (Fig. 8, bottom). The reasoning for an extended in- terface is not as clear, as for Si(100). The bond topology of the Si(lll) surface is characterized by a double-layer structure, where layers with one broken bond alternate with layers with three broken bonds to give an average of two broken bonds per atom, as on the Si(IOO) surface. Taking the layer with one broken bond as interface layer 38 MICROSCOPIC STRUCTURE OF THE SiOz/Si INTERFACE 6093 would give about the right bond density for attaching SiOZ. Strain calculationslo’11 confirm this simple bond- counting argument by giving less strain energy for the abrupt (1 l 1) interface than for the abrupt (100) interface. The experimental distribution does indeed exhibit Si1+ as the strongest component. However, there is a large Si3+ component that would create a high bond density if the interface was abrupt. In addition, the total number of Si atoms in intermediate-oxidation states is about two layers [higher even than for Si(100)]. These features can only be accommodated by an extended interface. A highly simplified model is given in Fig. 8. It does not exhibit Si“, and the number of interface atoms is too small, but these deficiencies can easily be corrected by adding steps or other defects. VII. ELECTRICAL PROPERTIES Some of the parameters that determine the electrical properties of the SiOz/Si interface can be probed with photoelectron spectroscopy, e.g., the Fermi-level position in the gap, the valence-band offset, and the vacuum level. A summary of the results is given in Figs. 9 and 10 for sro2 / smom Si 203/2 hu=1306V 3) Annealed in vacuum Shlft of EF‘EVBM Photoemissuon lntensny ( Arb. Uruts ) 2) Annealed in H2 1 ) Grown I Si Si‘ + in dry 02 ' | Si ~107 ~106 —105 —104 —103 -—102 —101 —100 —99 —98 Initial-State Energy (eV relative to EF) FIG. 9. Efiect of H2 annealing (1 atm H2, 850°C, 10 sec) on the SiOz/SillOO) interface. A rigid shift of the whole spectrum is caused by the Fermi level moving from a pinned position to its bulk position (see Fig. 10). The distribution of oxidation states does not change, however, since the number of pinning defects is too small to be detected. Subsequent annealing in vac- uum (850°C, 1 min) restores the pure interface by driving off hy- drogen. hydrogen anneal dry oxygen 5_1 __________ __ vacuum level 4.8 —————————— —— SiOZ Si 0.2 _ . . . . . _ . . . . . . . . . . . . . . . . . . . . . . . . - . - ~ - - A - - - ~ - - - - -» O = Fermi level —0.6 —O.9 ~52 FIG. 10. Band diagram for the SiOz/Si(100) interface corre- sponding to the data in Fig. 9. After dry oxidation, the Fermi level is pinned at 0.6 eV above the valence—band maximum of Si, i.e., near the neutral point of the distribution of Pb centers (tak- en from Johnson et (11., Ref. 61). Annealing in H2 brings the Fermi level to its bulk position (shown for a 5><10l5~cm‘3 n- type doping). the SiOz/Si(100) interface. The position of the Fermi- level E F relative to the valence-band maximum EVBM of Si is obtained by a method described in Ref. 71. Essen- tially, the position of the bulk Si 2p line below E F is com- pared with that of a reference surface with known Fermi-level position. As such we use the CaFZ/SiUll) interface, where the Fermi level coincides with the valence-band maximum.54 The valence-band edge has a fixed distance from the Si 2p level, independent of the Fermi-level position. Thus, the difference in the Si 2p en- ergies reflects the difference in the valence-band maxima, measured relative to EF. The band offset is determined from a photoelectron spectrum of the valence-band re- gion (not shown). The conduction-band offset is obtained from the valence-band offset by adding the respective band gaps (1.1 eV for Si and 9.3 eV for Si02, see Ref. 72). The vacuum level is obtained from the low-energy cutofi‘ of the photoelectron spectrum. The electrical parameters of the SiOz/Si interface de— pend on the preparation method as shown in previous 6094 work.2’73 We focus on the effect of annealing in hydrogen. Such a treatment is known to reduce the density of inter- face states.2’61'62 Essentially, the hydrogen saturates free Si bonds (Pb centers) at the interface. Thereby the Fermi level becomes unpinned, and the band bending disap- pears, in agreement with our result. The Fermi-level movement shows up as an overall shift of the core-level spectrum (see Fig. 9). The shift reverses after annealing in vacuum, whereby the H is driven out. It is interesting to compare the observed pinning position of the Fermi level with the distribution of Pb centers as measured by Johnson et al.“ by deep-level transient spectroscopy by (DLTS) (Fig. 10). It coincides with the neutral point be- tween donorlike and acceptorlike centers. Such a pinning position is expected for a nearly intrinsic substrate. Another noteworthy point is that the band offset and the ionization energy (vacuum level minus valence-band max- imum) do not change significantly upon H2 annealing. This is understandable since these are intrinsic properties which are not affected by a small number of Pb centers. The barrier (conduction-band minimum minus Fermi lev— e] for n-type material) and the work function (vacuum level minus Fermi level) change, because the Fermi level moves relative to the band structure. The Fermi-level movement is given by the dipole that is formed between the electrons in the Pb centers and the ionized donors in the depletion region. VIII. COMPARISON WITH OTHER RESULTS The SiOZ/Si interface has been probed by many tech- niques. The interface widths reported in the literature range from abrupt to more than 7 A wide. These varia- tions may reflect different preparation conditions. How— ever, it is quite clear that various measurement tech— niques probe different aspects of the interface. Many methods, for example, cannot resolve the atomic struc- ture but see some average. Other techniques require in- trusive sample preparation methods, such as depth profiling by sputtering or etching. Such treatments alter the chemistry of the interface. First we discuss core-level spectroscopy results. They can be classified into two groups, i.e., XPSo measure— ments“"37 with a probing depth of about 30 A and syn— chrotron radiation resultsi’f“43 near the minimum of the escape depth (3-5 A). Our photon-energy-dependent measurements bridge the gap in probing depth. There is also a difference in the sample preparation. The surface- sensitive synchrotrop measurements have been performed on ultrathin (5—30 A) Si02 films grown in situ with pure 02 at low pressures (10’5 Torr). The XPS oresults have been obtained mainly from thick (100— 1000 A) SiOz films grown under device processing conditions (atmospheric pressure of oxygen or steam). These thick oxides are chemically thinned in order to see a signal from the inter- face. There exist a few discrepancies between the XPS and synchrotron results that need to be reconciled. An abrupt interface has been observed with XPS,29'33 i.e., one monolayer or less (83—94 % of a monolayer”) of Si atoms in intermediate-oxidation states. The missing fraction of F. J. HIMPSEL et al. 38 Si interface atoms has been assumed to bond to hydrogen, which comes in during steam oxidation or hydrogen an— nealing. Synchrotron measurements“;43 have reported an extended interface (:5 A wide) with about two layers of Si in intermediate-oxidation states. The easiest way to reconcile the results would be to assume that pure SiO2 and device oxides exhibit different interface structures. Using XPS, however, we have found no significant differences in the interface-to-bulk Si 2p intensity ratio between oxides grown in pure 02 at low pressure and un- der device processing conditions (see Ref. 74). Even chemical thinning (from 150 to 30 A) did not affect this ratio. Furthermore, our XPS data are similar to previous XPS results reported, e.g., by Grunthaner et al.29 The main differences appear to be due to the evaluation of the XPS data. In Ref. 29 an asymmetric line shape is used for bulk Si, which takes away intensity from Si1+ and, therefore, reduces the intensity of intermediate-oxidation states. In addition, a different formula is used in Ref. 29 to convert the intensity of intermediate oxidation states into the density of interface atoms. This conversion de- pends on the knowledge of the proper escape depth and density of SiOX. Techniques other than core-level spectroscopy will be touched upon briefly to point out which aspects of the SiOz/Si interface they probe. The field ion microprobe23 probes the Si/O ratio atom by atom. The interface is defined the same way as in core-level spectrosc0py, i.e., the region containing Si in intermediate-oxidation states. A width of 3—5 A is obtained23 which is comparable with our core level result. The twoolayers of interface atoms that we find corresponds to 3 Aoof pure Si or to 6 A of SiOz. An interface layer of 7:1:2 A SiOO.4 has been report— ed by ellipsometry.18 These measurements rule out an abrupt interface, in agreement with our findings. Ruther- ford backscatteringu is capable of determining the Si/O stoichiometry and the number of Si atoms displaced from lattice sites. Between 1.4 and 2.3 monolayers of Si atoms are found to be displaced from their lattice sites. Up to l monolayer of these can be attributed to an oxide with average stoichiometry SiO at the interface with 0.8—1.7 monolayers of displaced Si atoms remaining.21 The one monolayer of SiO is only about half as much as the num- ber of Si atoms in intermediate oxidation states that we find, but by assigning more of the displaced Si atoms as part of the interface layer one may be able to make the two results consistent. Transmission electron microscopy has been used extensively to determine the extent of the interface and to detect irregularities. The width of the in- terface is difficult to define in this case. Instead, various types of roughness parameterslz’13 have been used to characterize the interface. A similar situation holds for low-energy electron diffraction16 and scanning tunneling microscopy”’15 experiments where the oxide is etched away and the roughness of the remaining Si surface is measured. These roughness parameters cannot be com- pared directly with our results. Vibrational spectrosco- py19 shows that the bonding changes when the oxide cov- erage is on the order of a monolayer. Oxygen is found to be incorporated into bridge bonds at elevated tempera— tures, as assumed in our models. 38 MICROSCOPIC STRUCTURE OF THE SiOz/ Si INTERFACE IX. SUMMARY The aim of our work is to give a quantitative analysis of intermediate-oxidation states at the SiOz/Si interface and to derive structural models from it. In order to achieve this we measure the escape depth in Si and SiOz, and the photoionization cross section for various oxida- tion states of Si. From these calibration measurements the density of intermediate-oxidation states is deter- mined. It corresponds to about two layers of Si atoms with a somewhat higher density for a Si(111) substrate than for Si(100). This high density and the observation of a strong Si3+ component rule out most of the existing in- terface models, which exhibit an atomically abrupt inter- face. Extended—interface models are given that are con- sistent with our results. They feature characteristic pro- ‘Present address: Surface Science Division, National Bureau of Standards, Gaithersburg, MD 20899. 1The Physics of SiOz and its Interfaces, edited by Sokrates T. Pantelides (Pergamon, New York, 1978). 2F. J. Grunthaner and P. J. Grunthaner, Mater. Sci. Rep. 1, 65 (1986). 3See various articles in Philos. Mag. 55 (1987). 4Proceedings of the 173rd meeting of the Electrochemical So- ciety, Atlanta, Georigia, 1988, edited by C. R. Helms (unpub— lished). 5F. Rochet, S. Rigo, M. Froment, C. D’Anterroches, C. Maillot, H. Roulet, and G. Dufour, Adv. Phys. 35, 237 (1986). 6F. Herman, I. P. Batra, and R. V. Kasowski, in The Physics of Si02 and its Interfaces, Ref. 1, p. 333; F. Herman, and R. V. Kasowski, J. Vac. Sci. Technol. 19, 395 (1981). 7S. T. Pantelides and Marshall Long, in The Physics of Si02 and its Interfaces, Ref. 1, p. 339. 8T. Hattori, M. Muto, and T. Suzuki, in Proceedings of the 17th International Conference on the Physics of Semiconductors, edited by D. J. Chadi and W. Harrison (Springer-Verlag, New York, 1985), p. 229. 9A. Ourmazd, D. W. Taylor, J. A. Rentschler, and J. Bevk, Phys. Rev. Lett. 59, 213 (1987). l0I. Ohdomari, H. Akatsu, Y. Yamakoshi, and Koji Kishimoto, J. Non—Cryst. Solids 89, 239 (1987). 11I. Ohdomari, H. Akatsu, Y. Yamakoshi, and K. Kishimoto, J . Appl. Phys. 62, 3751 (1987). 12O. L. Krivanek and J. H. Mazur, App]. Phys. Lett. 37, 392 (1980). 13S. M. Goodnick, D. K. Ferry, C. W. Wilmsen, Z. Liliental, D. Fathy, and O. L. Krivanck, Phys. Rev. B 32, 8171 (1985). 14A. H. Carim, M. M. Dovek, C. F. Quate, R. Sinclair, and C. Vorst, Science 237, 630 (1987). 15W. J. Kaiser, L. D. Bel], M. H. Hecht, and F. J. Grunthaner, J. Vac. Sci. Techno]. A 6, 519 (1988). 1"’P. O. Hahn and M. Henzler, J. Vac. Sci. Technol. A 2, 574 (1984); M. Henzler and P. Marienhofl', J. Vac. Sci. Techno] B 2, 346 (1984). WB. Nielsen, K. G. Lynn, Y. C. Chen, and D. O. Welch, Appl. Phys. Lett. 51, 1022 (1987). 18D. E. Aspnes and J. B. Theeten-, J. Electrochem. Soc. 127, 1359 (1980). “H. Ibach, H. D. Bruchmann, and H. Wagner, Appl. Phys. A 6095 trusions of Si3+ into the SiOz overlayer, which can be viewed as the cores of misfit dislocations. The extended interface structure appears to be strain driven, which is easy to understand given the large misfit in bond density between Si and SiOz. ACKNOWLEDGEMENTS We acknowledge M. Liehr for providing calibrated ox- ide films and B. Robinson for ellipsometry measurements. Research was carried out in part at the National Syn- chrotron Light Source, Brookhaven National Laborato- ry, which is sponsored by the US. Department of Energy (Division of Materials Sciences and Division of Chemical Sciences of the Office of Basic Energy Sciences) under Contract No. DE-AC02-76CH0016. 29, 113 (1982). 20M. Grundner and H. Jacob, Appl. Phys. A 39, 73 (1986). 2‘R. Haight and L. C. Feldman, J. Appl. Phys. 53, 4884 (1982). 22R. A. Cowley and T. W. Ryan, J. Phys. D 20, 61 (1987); P. H. Fuoss, L. J. Norton, S. Brennan, and A. Fischer-Colbrie, Phys. Rev. Lett. 60, 600 (1988). 23A. Cerezo, C. R. M. Grovenor, and G. D. W. Smith, J. Mi- crosc. (Oxford) 141, 155 (1986); C. R. M. Grovenor, A. Cere— zo, and G. D. W. Smith, in Layered Structures, Epitaxy, and Interfaces, edited by J. M. Gibson and L. R. Dawson (MRS, Pittsburgh, 1985), p. 199; A. M. Stoneham, C. R. M. Gro— venor, and A. Cerezo, in Ref. 3, p. 201. 24C. R. Helms, Y. E. Strausser, W. E. Spicer, Appl. Phys. Lett. 33, 767 (1978); C. R. Helms, N. M. Johnson, S. A. Schwartz, and W. E. Spicer, J. Appl. Phys. 50, 5007 (1979). 25J. Derrien and M. Commandré, Surf. Sci. 118, 32 (1982). 26R. Flitsch and S. I. Raider, J. Vac. Sci. Technol. 12, 305 (1975). 27F. J. Grunthaner, P. J. Grunthaner, R. P. Vasquez, B. F. Lewis, J. Maserjian, and A. Madhukar, Phys. Rev. Lett. 43, 1683 (1979); J. Vac. Sci. Technol. 16, 1443 (1979). 2EM. H. Hecht, F. J. Grunthaner, P. Pianetta, L. I. Johanssonn, and I. Lindau, J. Vac. Sci. Technol. A 2, 584 (1984). 29F. J. Grunthaner, M. H. Hecht, and F. J. Grunthaner, J. Appl. Phys. 61, 629 (1987); M. H. Hecht, P. J. Grunthaner, and F. J. Grunthaner, in Proceedings of the 17th International Conference on the Physics of Semiconductors, edited by J. D. Chadi and W. A. Harrison (Springer-Verlag, New York, 1985), p. 217. 301. M. Hill, D. G. Royce, C. s. Fadley, L. F. Wagner, and F. J. Grunthaner, Chem. Phys. Lett. 44, 225 (1976). 31A. Ishizaka and S. lwata, Appl. Phys. Lett. 36, 71 (1980). 32T. Suzuki, M. Muto, M. Hara, T. Hattori, K. Yamabe, and H. Yamuauchi, in Proceedings of Extended Abstracts of the 16th International Conference on Soild State Devices and Materi- als, Kobe, 1984, p. 297 (unpublished). 33T. Hattori and T. Suzuki, Appl. Phys. Lett. 43, 470 (1983); T. Hattori and T. Suzuki, in Proceedings of Extended Abstracts of the 16th International Conference on Solid State Devices and Materials, Ref. 32, p. 497. 34J. Finster, D. Schulze, F. Bechstedt, and A. Meisel, Surf. Sci. 151/153, 1063 (1985). 35'1". D. Bussing and P. H. Holloway, J. Vac. Sci. Technol. A 3, 6096 1973 (1985); T. D. Bussing, P. H. Holloway, Y. X. Wang, J. F. Moulder, and J. S. Hammond (unpublished). 36J. Halbritter, J. Mater. Reson. 3, 506 (1988). 37M. F. Hochella, Jr. and A. H. Carim, Surf. Sci. Lett. 197, L260 (1988). 38C. M. Garner, 1. Lindau, C. Y. Su, P. Pianetta, and W. E. Spicer, Phys. Rev. B 19, 394-4 (1979). 39M. Tabe, T. T. Chiang, I. Lindau, and W. E. Spicer, Phys. Rev. B 34, 2706 (1986). 40A. Bianconi and R. S. Bauer, Surf. Sci. 99, 76 (1980). “G. Hollinger and F. J. Himpsel, J. Vac. Sci. Technol. A 1, 640 (1983); Phys. Rev. B 28, 3651 (1983). 42G. Hollinger and F. J. Himpsel, Appl. Phys. Lett. 44, 93 (1984). 43W. Braun and H. Kuhlenbeck, Surf. Sci. 180, 279 (1987). 44The pressure-versus-temperature equilibrium curve between a Si surface and 02 (H20) is given in F. W. Smith and G. Ghidi- ni, J. Electrochem. Soc. 129, 1300 (1982); 131, 2924 (1984). 45F. J. Himpsel, F. R. McFeely, J. F. Morar, A. Taleb-Ibrahimi, and J. A. Yarmoff, in Proceedings of the 1988 Enrico Fermi School on Photoemission and Absorption Spectroscopy of Solids and Interfaces with Synchrotron Radiation, Varenna, I 988 (North-Holland, Amsterdam, in press). 46R. Tromp, G. W. Rublofi‘, P. Balk, F. K. Le Goues, and E. J. van Loenen, Phys. Rev. Lett. 55, 2332 (1985). 47F. J. Himpsel, P. Heimann, T.-C. Chiang, and D. E. Eastman, Phys. Rev. Lett. 45, 1112 (1980); F. J. Himpsel, D. E. East- man, P. Heimann, B. Reihl, C. W. White, and D. M. Zehner, Phys. Rev. B 24, 1120 (1981). 4i“For recent theoretical work on Si(100) see P. Kriiger, A. Mazur, J. Pollmann, and G. Wolfgarten, Phys. Rev. Lett. 57, 1468 (1986). 49The other half-layer of inner dimer atoms has opposite charge transfer and can be located about 0.2 eV below the bulk line by least-squares fitting. STM pictures of Si(100) show both symmetric and asymmetric dimers (Ref. 50). In our assign- ment the symmetric dimers are in fact rapidly oscillating asymmetric dimers. The time scale of the photoemission pro- cess is faster than the oscillation frequency, whereas the time scale of the STM experiment is slower. The surface core-level emission from Si has also been studied by D. H. Rich, T. Miller, and T.-C. Chiang, Phys. Rev. Lett. 60, 357 (1987); Phys. Rev. B 37, 3124 (1988). The surface core-level intensi- ties (relative to the total intensity) obtained in this work [0.163 for Si(100) and 0.050 for Si(lll) at hv=150 eV] are very similar to ours [0.17 for Si(100) and 0.05 for Si(lll) at hv=l30 eV]. However, a difi'erent assignment is given by Rich et al. which doubles the number of surface atoms corre- sponding to a given core-level intensity. Essentially, such an assignment is not compatible with the surface core-level intensities on other model surfaces [the As/Si(111)1><l and Can/Si(111)1 X1 structures would give about two layers of Si with shifted core levels], and it is opposite to our knowledge about charge transfer at the Si(111)7><7 surface. Scanning tunneling spectroscopy (Ref. 50) and first-principles calculations (Ref. 51) find that electrons are transferred from adatoms to rest atoms to fill the dangling bond orbital of the rest atoms. Therefore, one would expect the rest atoms to ex- hibit the observed upwards core-level shift, and not the ada- toms as in the assignment by Rich et al. Since there are about twice as many adatoms in the unit cell as rest atoms (12 vs 7) there is a factor of 2 discrepancy between the assign- F. J. HIMPSEL et al. 38 ments. 50R. J. Hamers, R. M. Tromp, and J. E. Demuth, Phys. Rev. Lett. 56, 1972 (1986); Surf. Sci. 181, 346 (1987); Phys. Rev. B 34, 5343 (1986). 51J. E. Northrup, Phys. Rev. Lett. 57, 154 (1986). 52M. A. Olmstead, R. D. Bringans, R. I. G. Uhrberg, and R. Z. Bachrach, Phys. Rev. B 34, 6401 (1986). 53F. J. Himpsel, F. U. Hillebrecht, G. Hughes, J. L. Jordan, U. 0. Karlsson, F. R. McFeely, J. F. Morar, and D. Rieger, Appl. Phys. Lett. 48, 596 (1986). 54D. Rieger, F. J. Himpsel, U. 0. Karlsson, F. R. McFeely, J. F. Morar, and J. A. Yarmofi‘, Phys. Rev. B 34, 7295 (1986); F. J. Himpsel, U. 0. Karlsson, J. F. Morar, D. Rieger, and J. A. Yarmofl‘, Phys. Rev. Lett. 56, 1497 (1986); Mater. Res. Soc. Symp. Proc. 94, 181 (1987). 55M. A. Olmstead, R. I. G. Uhrberg, R. D. Bringans, and R. Z. Bachrach, Phys. Rev. B 35, 7526 (1987). 568. Tanuma, C. J. Powell, and D. R. Penn (unpublished). 57F. R. McFeer et al. (unpublished). 58D. Schmeisser, F. J. Himpsel, and G. Hollinger, Phys. Rev. B 27, 7813 (1983). 59H. Ibach, W. Wagner, and D. Bruchmann, Solid State Com- mun. 42, 457 (1982); E. M. Oellig, R. Butz, H. Wagner, and H. Ibach ibid. 51, 7 (1984); Y. Chabal, Phys. Rev. B 29, 3677 (1984). wThe surface to bulk+surface intensity ratio is 1.01 times greater for Si(lll) terminated by a single dangling bond than for Si(100) at hv=130 eV, see Eqs. (4), (5a), and (5b), and Table I. 61N. M. Johnson, D. K. Biegelsen, M. D. Moyer, and S. T. Chang, Appl. Phys. Lett. 43, 563 (1983). The distribution of Pl, centers is given for the SiOz/Sifl 1 1) interface, but a similar distribution is expected on Si(100). 62A. H. Edwards, Phys. Rev. B 36, 9638 (1987). 63F. K. Le Goues er al. (unpublished). 64K. Siegbahn, Philos. Trans. R. Soc. London A318, 3 (1986). 65L. Pauling, The Nature of the Chemical Bond (Cornell Univer— sity Press, Ithaca, New York, 1948). 66E. Yablonovitch, D. L. Allara, C. C. Chang, T. Gmitter, and T. B. Bright, Phys. Rev. Lett. 57, 149 (1986). 67B. S. Meyerson, F. J. Himpsel, and J. A. Yarmofl‘ (unpub- lished). 68]. F. Morar and J. Bevk (unpublished). 69G. Hughes, G. Hollinger, J. F. Morar, and F. J. Himpsel (un- published). 70The Si(100) surface terminated by (111) facets can exhibit Si3+ at the expense of creating a higher bond density at the Si sur- face. We can rule out such a model by observing that the dis- tribution of intermediate-oxidation states for Si(100) differs from that for Si(l 1 1). 71F. J. Himpsel, G. Hollinger, and R. A. Pollak, Phys. Rev. B 28, 7014 (1983). 722. A. Weinberg, G. W. Rublofi‘, and E. Bassous, Phys. Rev. B 19, 3107 (1979). A somewhat larger band gap of 9.7 eV has been reported for SiOz by V. J. Nithianandam and S. E. Schnatterly (unpublished). 73Z. A. Weinberg and A. Hartstein, J. Appl. Phys. 54, 2517 (1983). 74G. Hollinger, R. Saoudi, P. Ferret, M. Pitaval, in Proceedings of the 173rd meeting of the Electrochemical Society, Atlanta, Georgia, 1988 (unpublished). ...
View Full Document

This note was uploaded on 12/10/2009 for the course IF PFIS1200 taught by Professor Antonio during the Spring '09 term at Universidade Federal do Rio de Janeiro.

Page1 / 13

36-himpsel-prb-1988-6084 - PHYSICAL REVIEW B VOLUME 38,...

This preview shows document pages 1 - 13. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online