Infrared‐absorption spectroscopy of Si(100) and Si(111) surfaces after
`chemomechanical polishing
`G. J. Pietsch, Y. J. Chabal, and G. S. Higashi
`
`Citation: Journal of Applied Physics 78, 1650 (1995); doi: 10.1063/1.360721
`View online: http://dx.doi.org/10.1063/1.360721
`View Table of Contents: http://scitation.aip.org/content/aip/journal/jap/78/3?ver=pdfcov
`Published by the AIP Publishing
`
`Articles you may be interested in
`Interaction of hydrogen-terminated Si(100), (110), and (111) surfaces with hydrogen plasma investigated
`by in situ real-time infrared absorption spectroscopy
`J. Vac. Sci. Technol. A 21, 25 (2003); 10.1116/1.1524146
`
`Characterization of SiC whiskers through infrared‐absorption spectroscopy
`J. Appl. Phys. 73, 8506 (1993); 10.1063/1.353403
`
`Infrared spectroscopy of Si(111) and Si(100) surfaces after HF treatment: Hydrogen termination and
`surface morphology
`J. Vac. Sci. Technol. A 7, 2104 (1989); 10.1116/1.575980
`
`The surface state of Si (100) and (111) wafers after treatment with hydrofluoric acid
`AIP Conf. Proc. 167, 329 (1988); 10.1063/1.37163
`
`Infrared spectroscopy of Si(111) surfaces after HF treatment: Hydrogen termination and surface
`morphology
`Appl. Phys. Lett. 53, 998 (1988); 10.1063/1.100053
`
`
`
` Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. IP: 104.129.194.129 On: Tue, 01 Mar 2016
`23:12:15
`
`Raytheon2033-0001
`
`Sony Corp. v. Raytheon Co.
`IPR2015-01201
`
`

`
`) surfaces
`
`of Si(1 00) and Si(lll
`
`spectroscopy
`Infrared-absorption
`after che~omechanical
`polishing
`G. J. Pietsch,‘) Y. J. ChabaLb) and G. S. Higashi@
`AT&T Bell Laboratories, Murray Hill, New Jersey 07974
`(Received 3 February 1995; accepted for publication 13 April 1995)
`is studied by
`(CMP)
`The mechanism of silicon stock removal
`in chemomechanical polishing
`characterizing
`surface chemical
`species with
`infrared-absorption measurements and
`the
`corresponding degree of hydrophobic@ with contact angle measurements immediately after CMP.
`Surface properties and stock removal rates are found to depend strongly on the pH of the silica
`slurry used in this “syton polishing”
`technique. At the peak of the removal rate [pH-11
`for both
`Si(100) and Si(1 ll)],
`the surfaces have the highest hydrophobicity and the highest hydrogen
`coverage. Si(ll1) has an ideal monohydride termination, while Si(100) is characterized by a variety
`of hydrides
`(mono-, di-, and trihydrides), suggesting different morphologies
`for
`the surfaces:
`atomically
`flat domains on Si( 111) and rougher areas on Si( 100). Away from the optimum slurry pH
`(at lower stock removal rates), a higher concentration of hydroxyl groups is observed, increasing the
`surface hydrophilicity. At all pH, some oxidation occurs beneath the H-terminated Si surface, as
`evidenced by a characteristic frequency shift of oxygen-backbonded hydrides. The mechanisms of
`stock removal are considered in view of these observations for the different ranges of slurry pH. In
`particular, at the highest removal rates, an interplay of surface oxidation,
`removal of oxidized
`silicon, and subsequent H termination
`is suggested. Based on the spectroscopic characterization of
`surface morphologies,
`the relevance of CMP to prepare atomically smooth silicon surfaces is
`Institute of Physics.
`discussed. 0
`I995 American
`
`1. INTRODUCTION
`Since
`the
`late 196Os, chemomechanical polishing*
`(CMP) has been used to prepare smooth, defect-free starting
`surfaces for the patterning of microelectronic devices.’ CMP
`uses an alkaline suspension of colloidal silica (“syton”) and
`combines a mechanical grinding action with a chemical etch-
`ing action. It is the last and, thus, most critical step in a series
`of surface-shaping processes3 (slice cutting from the crystal
`ingot; grinding/lapping; wet-chemical etching
`for
`the re-
`moval of the surface layers damaged by the preceding me
`chanically abrasive steps) designed to obtain
`the surface
`smoothness and integrity required by the state-of-the-art pat-
`terning technologies;4 CMP is also the surface planarization
`process of choice to produce the extremely
`low local and
`total
`thickness variations
`that the new
`low-depth-of-field
`photolithography
`techniques demand; CMP yields surface
`morphologies appropriate for further deposition of thin “ac-
`tive” epitaxial
`layers; finally, CMP is used to prepare thin-
`film Si devices
`for silicon-on-insulator
`(SOI) structures5
`formed by wafer direct bonding. In all these important appli-
`cations chemomechanical polishing
`is a precise, well-
`controllable, and cost-effective
`technology to create silicon
`surfaces with superior flatness (see upper portion of Fig. 1)
`and specific properties required by technology.
`Most recently, CMP has become even more important as
`a method to remove excess interlevel dielectric SiO, over the
`connection studs
`in very
`large scale integrated
`(VLSI)
`
`%resent address: FB Physik, Philipps-Uaiversit%t Marburg, 35032 Marburg,
`Germany.
`b)Electronic mail: yves@physics.att.com
`‘)Present address: AT&T Microelectronics, 9333 South John Young Park-
`way, Orlando, FL 32819.
`
`interlayer
`multilayer chips after device patterning to allow
`connection and to produce a planar surface for the next level
`of wiring6 (see lower portion of Fig. 1). Here, CMP forms a
`low-cost and low-damage alternative to plasma etch-backs
`with an even better resulting surface planarity. In this post-
`patterning application, silicon dioxide is removed during pol-
`ishing (lower portion of Fig. l), in contrast to the prepattern-
`ing applications listed above where silicon is removed (see
`upper portion of Fig. 1). In the future, it is expected that
`CMP will become even more
`important, with shrinking
`device dimensions and demands for defect-free, ultrathin
`(-50 A) gate oxides.
`the
`Technology has motivated extensive studies of
`chemomechanical polishing process. Properties such as stock
`removal
`rate, macroscopic
`removal homogeneity, and
`selectivityZV7 have been studied in great detail. However, it is
`becoming more urgent to understand the microscopic
`re-
`moval mechanisms and surface processes as the device prop-
`erties begin to depend on the atomic scale properties (rough-
`ness, etc.). The molecular origin of the observed hydro-
`phobicity of silicon surfaces immediately after CMP, for in-
`stance, has only recently been addressed. Schnegg et al. have
`suggested that the hydrophobicity was predominantly due to
`hydrogen surface termination.879 Such a hypothesis has just
`been verified for the case of Si(ll1) surfaces’0 and there is a
`clear need to generalize the study to the technologically more
`important Si(100) surfaces.
`In this article we use infrared (IR) absorption spectros-
`copy, in addition
`to stock removal rates and contact angle
`measurements,
`to study
`the surface properties of both
`Si(100) and Si(ll1)
`after chemomechanical polishing. This
`work is an extension of the more limited study on Si(ll1)
`surfaces,” and attempts to uncover
`the chemical etching
`
`1650
`
`J. Appl. Phys. 78 (3), 1 August 1995
`
`0021-8979/95/76(3)11650l9/$6.00
`
`Q 1995 American Institute of Physics
`
` Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. IP: 104.129.194.129 On: Tue, 01 Mar 2016
`23:12:15
`
`Raytheon2033-0002
`
`

`
`surf. level
`after CMP
`
`f
`
`I
`
`1
`
`after CMP
`and metalit.
`
`FIG. 1. Schematics of various applications of chemomechanical polishing:
`(top) silicon surface planarization prior to device patterning; (bottom) re-
`moval of interlevel silicon dioxide for interlayer connection and planariza-
`tion of the oxide surface for the next layer of wiring.
`
`the concentra-
`
`mechanisms relevant to CMP by monitoring
`tion of various species at the surface.
`Infrared spectroscopy is particularly well suited to detect
`surface species such as Si-H, Si-O-H, and water,” expected
`as a result of colloidal silica polishing chemistry. Using a
`multiple
`internal reflection (MIR) geometry,‘” good sensitiv-
`ity can be achieved for vibrational modes higher than 1450
`this optical technique does not require
`cm-‘. Furthermore,
`introduction
`into a vacuum chamber as does electron energy
`loss spectroscopy,‘3”4 making
`it much more convenient
`to
`investigate
`the silicon surfaces immediately after CMP. In
`Sec. II we describe
`the experimental configuration and
`sample preparation.
`In Sec. III we summarize
`the depen-
`dence of the stock removal rate and contact angle (degree of
`hydrophobicity) on the solution pH.
`In the discussion (Sec.
`IV), we relate the observations to atomic processes at sur-
`faces and propose microscopic models for the various sur-
`face reactions taking place in CMP.
`
`II. EXPERIMENT
`for chemomechanical polishing experi-
`The samples
`ments are prepared from n-type Si(ll1)
`and Si(100) wafers
`with a moderate resistivity
`(p-150 Q cm) and a well-
`oriented cut bolar angle of cut A@, versus (111) and (100) is
`IAq<O.Ol”]
`to provide a low initial step density. Rectangular
`plates (38X 15X0.45 mm3) are cut and their short end bev-
`eled at 45” to let the IR radiation in and out (see bottom part
`
`pressure
`
`45”-beveled
`Si samje
`
`slurry feed
`hing
`pad
`
`/otating
`polishing plate
`
`v
`
`&-beveled sample
`
`the rectangular, beveled
`FIG. 2. (Top) experimental setup for polishing
`samples used for
`the IR measurements: The vacuum chuck holds the
`samples that are moved back and forth over the rotating polishing pad,
`(bottom) schematic of the beveled sample showing the IR beam path (with
`multiple internal reflections) and definitions of the polarizations (s polariza-
`tion is strictly parallel to the surface).
`
`of Fig. 2). Once in the plate, the IR beam is totally internally
`reflected, with a total of -75
`reflections
`{both sides are
`probed). This
`internal
`reflection geometry gives
`roughly
`equal sensitivity
`to all components of the measured surface
`(E, , E, , and E, are equal to within 10%). The
`absorption”
`sharp edges formed by the 45” bevels are slightly rounded to
`avoid scratching of the polished surface by chipped-off sili-
`con during polishing. The Fourier transform infrared (FTIR)
`spectrometer is a Nicolet 6OSX, operated with a 1 cm-’
`reso-
`lution and equipped with a Nx-purged
`load
`lock and a
`N,-cooled
`InSb detector.
`Prior
`to polishing
`the samples are thoroughly cleaned
`using repeated cycles of chemical oxidation,
`the RCA-SC1
`standard clean producing hydrophilic surfaces’6V’7 (10 min in
`NH40H:H,0,:Hz0=1:1:5
`at 80 “C), and a HF dip in dilute
`HF, resulting
`in hydrophobic surfaces.17 This alternating
`growth17 and removal of a thin hydrophilic wet-chemical ox-
`ide results
`in hydrocarbon-, metal-, and particle-free sur-
`faces.17-19 The process ends with a chemical oxidation, leav-
`ing a thin oxide layer on the surface prior to chemomechani-
`cal polishing.
`The solution used for chemomechanical polishing con-
`sists of dilute suspension of colloidal silica (10% syton)2*4
`(*pH-
`11). Potassium hydroxide (KOH) and/or phosphoric
`acid (H,PO,) can be added to adjust the pH.
`For
`the chemomechanical polishing,
`the samples are
`held at their circumference by a Teflon gasket on a small
`vacuum chuck. This arrangement makes it possible to polish
`both surfaces, with a fast turnover and minimum contamina-
`tion of the side just polished (see upper portion of Fig. 2).
`The vacuum chuck can pivot freely about a tip which pro-
`
`J. Appl. Phys., Vol. 78, No. 3, 1 August 1995
`
`Pietsch, Chabal, and Higashi
`
`1651
`
` Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. IP: 104.129.194.129 On: Tue, 01 Mar 2016
`23:12:15
`
`Raytheon2033-0003
`
`

`
`Si(lil):-
`Si(iOO):---
`
`g100
`-iz
`
`=
`v
`2
`
`80
`
`60
`
`oL.“‘a’a”‘*‘*“’
`0
`2
`4
`6
`8
`IO 12 14 16
`POLISHING SLURRY pH
`
`FIG. 3. Stock removal rate during chemomechanical polishing as a function
`of slung pH: The open circles and solid squares are for the Si(100) and
`Si( 111) surfaces, respectively.
`
`absolute oxygen concentration on Si(lOO), for instance, has
`been established, varying from -70”
`for no detectable sur-
`face oxygen to -25”
`for 4 at. %.‘O In our experiments, the
`contact angles measured immediately after chemomechanical
`polishing show a dependence on the polishing slurry pH that
`strongly resembles that of the stock removal rate (Fig. 3)
`both for Si(100) and Si( 111). It peaks at pH-I 1 and drops
`for acid and highly alkaline pH values.
`The magnitude of the effects is reversed between Si( 111)
`and Si(100). While the highest removal rates are observed
`for Si(lOO), the highest degree of hydrophobicity is recorded
`for Si(lll).
`The contact angle dependence on pH
`is much
`less pronounced for Si(100) samples (30”-55”)
`than for
`Si(ll1) samples (30”-70”). The most extreme contact angles
`achievable by wet-chemical means, i.e., those of the most
`
`. I - I - I . *N&l
`
`. I * I
`
`90”
`5 80”
`
`
`
`
`
`g :;;I :y;;y* , , RCA;-SC,B ,I .
`
`
`
`vides an adjustable polishing pressure (-lo4 Pa was used for
`compatibility with commercial polishing procedures”*4). Dur-
`ing polishing the tip moves the chuck with the sample peri-
`odically back and forth over the rotating polishing plate car-
`rying
`the polishing pad, thus providing a uniform stock
`removal of the sample. The resilient, poromeric polishing
`pad made from black polyurethane velours (pad size -100
`cm”) rotates about three times a second resulting in a polish-
`ing speed -100 cm/s. The polishing slurry is fed constantly
`via a wick onto the rotating pad at a rate of about 0.2 ml/s,
`providing a thin homogeneous film between the pad and the
`sample surface without allowing the slurry to wet-and
`thus
`corrode-the
`optically smooth beveled faces. Each side of
`the sample is polished for -5 min and subsequently repol-
`ished for an additional <l min to limit
`the risk of contami-
`nation of the polished surfaces by exposure to the ambient
`during the polish of the other side. Within a few minutes
`after polishing, the samples are inserted into the HZ-purged
`loadlock and transferred to the IR bench; the IR beam is
`adjusted and focused onto the beveled edges in a couple of
`minutes and data are then collected.
`The IR spectra obtained on polished samples are ratioed
`to spectra of either RCA-treated (oxidized samples) to study
`the Si-H stretch
`region, or NH&etched
`samples (H-
`terminated surfaces) to study the hydrocarbon and hydroxyl
`regions. These reference spectra are taken immediately be-
`fore or after the spectra of polished wafers so as to eliminate
`bulk absorption and to limit artifacts in the spectra base line.
`Since measurements in the O-H stretch absorption area (V
`=3000-4000 cm-‘) are very sensitive to residual water ad-
`sorbed on the windows, mirror surfaces, and detector, as well
`as the homogeneity of N, purge, the reference sample is
`mounted together on the same sample holder so that the IR
`beam can probe alternatively
`the chemomechanically pol-
`ished and reference samples by a rapid, remotely controlled
`translation.
`
`III. RESULTS AND INTERPRETATION
`A. Removal
`rate and degree of hydrophobicity
`While the polishing speed, pressure, and syton concen-
`tration are important parameters for the mechanical polish-
`ing, the pH of the polishing slurry is the most important
`parameter for
`the chemical polishing. Figure 3 shows the
`strong dependence on the slurry pH of the resulting stock
`removal rate during CMP for both Si(111) and Si(100). The
`main observation is the peak in removal rate at pH- 11 for
`both Si(100) and Si(lll), with a sharp drop at higher pH. At
`lower pH,
`the rate drops but displays a small increase for
`slightly acid slurry pH @H-4). Another important observa-
`tion is the consistently higher removal rates on Si( loo), par-
`ticularly at the peaks of the curves.
`Figure 4 shows the results of contact angle measure-
`ments (using 5 ,~l water droplets) of chemomechanically pol-
`ished Si(ll1) and Si( 100). Contact angle measurements” are
`an indirect probe of surface oxygen with submonolayer sen-
`sitivity, which has been calibrated by x-ray photoelectron
`spectroscopy (XI’S) measurements carried out on the same
`surfaces.20 A linear dependence of the contact angle on the
`
`0
`
`10 12 14
`8
`6
`4
`2
`POLISHING SLURRY ,uH
`
`FIG. 4. Surface contact angles (obtained with a 5 ~1 water droplet) for
`chemomechanically polished Si(lll)
`(solid circles) and Si(100) (open
`circles) as a function of polishing slurry pH. The reference points are the
`most hydrophobic Si(ll1) surface (solid square labeled NH4F to reflect the
`preparation conditions), an intermediate surface obtained by HF etching of
`Si(100) (labeled cont. HF), and the most hydrophilic surfaces [solid and
`open squares for Si( 111) and Si(lOO), respectively, labeled RCA-SCl].
`
`1652
`
`J. Appl. Phys., Vol. 78, No. 3, 1 August 1995
`
`Pietsch, Chabal, and Higashi
`
` Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. IP: 104.129.194.129 On: Tue, 01 Mar 2016
`23:12:15
`
`Raytheon2033-0004
`
`

`
`hydrogen-
`[“ideally”
`surface preparable
`hydrophobic
`terminated Si(ll1)
`surface obtained after etching with 40%
`NH,F solution,“r labeled NHJF] and those of the most hy-
`drophilic surfaces (after wet-chemical growth of an ultrathin
`hydrophilic
`oxide by
`the RCA-SC1 Hz02:NH,0H:Hz0
`process, l6 labeled RCA-SCl) are given as a reference. The
`measured contact angles obtained after CMP are always well
`within
`these values, indicating
`that the surfaces are never
`completely H terminated nor oxidized at any given pH dur-
`ing the CMP process.
`
`B. Infrared-absorption data
`1. Hydrogen termination
`Figure 5 shows infrared-absorption spectra of chemome-
`chanically polished Si(ll1)
`(upper portion of Fig. 5) and
`Si(100)
`(lower portion)
`in the range of
`the Si-H stretch
`vibrations”“5”8.22-24 (2000-2200 cm-‘). The spectra of
`both Si(ll1) and Si(lO0) are characterized by a strong ab-
`sorption band around 2100 cm-‘,
`indicating that a substan-
`tial part of the surface is covered with hydrogen after chemo-
`mechanical polishing @H-11). Similar spectra taken for
`different slurry pH confirm
`that the highest H coverage is
`correlated with the highest hydrophobicity, The shape of the
`absorption bands is different on Si(ll1) and Si(lOO), as de-
`scribed below.
`The most prominent feature on Si( 111) immediately after
`CMP at pH- 11 is a single sharp absorption
`line at
`v-2082.4 cm-*
`(upper portion of Fig. 5), strictly polarized
`perpendicular to the surface.” This is the spectral signature
`(polarization and frequency) of
`ideally monohydride-ter-
`minated Si(111),21*B obtained,
`for
`instance, by etching
`Si(ll1)
`in a 40% ammonium
`fluoride solution. While
`the
`CMP and NH,F-etched surfaces have a similar spectrum,
`important differences exist, as shown in the center portion of
`Fig. 5: The surface etched in ammonium fluoride has a mode
`at 2083.8 cm-’
`instead of 2082.4 cm-‘, a 1.1 cm-’
`linewidth
`instead of 2.1 cm-‘, and twice the integrated area. The lower
`intensity of the StH mode on the CMP surface indicates that
`the hydrogen coverage is roughly half of a monolayer.
`Changes in effective surface dielectric constant (due to the
`presence of other species such as OH) cannot account for the
`different intensity ratio. The broader mode with a 1.4 cm-’
`downshift is consistent with the presence of small H/Si(ll
`l)-
`(1X 1) domains. In the absence of chemical shifts arising
`from neighboring impurities such as OH, the average size of
`such domains is estimated at 8 StH units.2”-28 This estimate
`would change if part of the shift arises from chemical shift
`and cannot therefore be taken as an accurate measure. The
`qualitative picture of small, predominantly 1 monohydride-
`terminated Si(ll1) domains separated by steps is coafirmed
`by the observation of a “defect mode” at v=2072 cm-’
`(arrow in upper portion of Fig. 5), assigned to the asymmet-
`ric stretch vibration of coupled monohydrides associated
`with a step along the (0 11) direction. This corresponds to the
`predominant step type expected on Si(ll1)
`(miscuttoward
`[~1~])~W9”0
`The Si(100) surface after CMP exhibits a much broader
`spectrum (bottom of Fig. 5) which cannot be assigned in as
`
`2000
`
`2050
`
`2100
`
`2150
`
`2200
`
`FIG. 5. IR-absorption spectra (silicon hydrogen stretch region) of chemo-
`mechanically polished Si(ll1)
`(top, slurry pH= 12.5) and Si(100) (bottom,
`slurry pH=12),
`referenced to RCA-cleaned surfaces. The center panel
`shows the spectrum of the same Si(ll1) surface as in the top panel, but
`etched instead in a 40% solution of ammonium fluoride.
`
`much detail. The featureless absorption ranges from 2050 to
`2150 cm-* . This region encompasses the frequencies of vari-
`ous monohydride structures (2070-2090 cm-‘), dihydride
`structures (2090-2120 cm-‘), and the modes of the trihy-
`dride (2120-2150 cm-‘). These assignments are based on
`exhaustive studies, including
`isotopic mixture experiments,
`of h-etched
`silicon surfaces and ab
`cluster cal-
`initio
`culations.2’*223*-34 Since the spectrum peaks at -2110 cm-‘,
`
`J. Appl. Phys., Vol. 78, No. 3, 1 August 1995
`
`Pietsch, Chabal, and Higashi
`
`1653
`
` Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. IP: 104.129.194.129 On: Tue, 01 Mar 2016
`23:12:15
`
`Raytheon2033-0005
`
`

`
`‘
`
`we conclude that the Si(100) surface is predominantly cov-
`ered by dihydrides. The surface, however,
`is atomically
`rough as evidenced by the presence of other hydrides and the
`lack of a strong polarization dependence. Comparison to the
`broad spectrum measured on the atomically rough HF-etched
`Si( 1 11),33 which has the same overall width but peaks around
`2085 cm-‘, highlights the ability of IR spectra to differenti-
`ate a predominant monohydride
`from a predominant dihy-
`dride termination on rough surfaces.
`Although the Si(100) spectrum is broad, we can rule out
`the predominance of an ideally dihydride-terminated
`flat
`Si(100). Such dihydride
`termination has been observed in
`UHV
`to coexist with monohydride
`to
`form a 3X 1
`periodicity.35 The dihydride modes are at 2091 and 2103
`cm-’
`for the symmetric and asymmetric stretch, respectively,
`with a strong polarization dependence:“’ The symmetric
`stretch is perpendicular to the surface and the asymmetric
`stretch parallel to the surface. The peak position and polar-
`ization dependence of the spectrum at the bottom of Fig. 5
`are inconsistent with such structure. On HF-etched Si(100)
`[i.e., atomically rough, H-terminated Si(100) surfaces], the
`position of the dihydride symmetric/asymmetric modes has
`been recorded at 210412115 cm-’
`for one type of dihydride
`and at 2108/2121 cm-’
`for another.32 These peak positions
`are much more consistent with the CMP data. Finally, the
`overall
`intensity of the broad absorption band is stronger
`than expected for an atomically smooth surface,31 providing
`further support for an atomically rough surface. The overall
`strength, however, is lower than that measured on the micro-
`scopically rough HF-etched Si( 100),22,33 suggesting that the
`surface is not completely terminated by hydrogen.
`
`2. Other species
`In this subsection we concentrate on the chemical spe-
`cies that make up the rest of the imperfectly H-terminated
`CMP Si surfaces. Figure 6 shows the full range of the spectra
`from 2000 to 4000 cm-’
`for Si(100) after CMP, obtained by
`using
`the spectrum of
`the NH4F-etched,
`ideally mono-
`hydride-terminated Si(ll1) as a reference. This reference sur-
`face is chosen because its spectrum is remarkably simple
`(strong band at 2083.8 cm-’ polarized perpendicular to the
`surface). In particular, the hydrocarbon and hydroxyl con-
`tamination is negligible on this surface. By switching quickly
`between this reference sample and the CMP sample, contri-
`butions from mirrors and windows are effectively eliminated.
`Using this reference, a sharp “negative” absorption peak at
`2083.8 cm-’
`(labeled A) appears in all CMP spectra taken
`with p polarization. The spectra associated with Si(100) are
`shown in the main figure, while those of the Si(ll1)
`are
`shown as insets. The upper figure portion summarizes the IR
`spectra of Si(ll1) and Si(100) polished at the optimum pol-
`ishing slurry pH of -11 while the lower figure portion sum-
`marizes the spectra obtained at _pH values where significantly
`lower polishing removal rate and more hydrophilic surfaces
`are obtained.
`first,
`Several absorption features can be distinguished:
`the features labeled B, a broad absorption centered around
`2110 cm-’
`for Si(100) and a sharp absorption at 2082.5
`cm-’
`for Si(ll1). These are assigned to the predominance of
`
`1654
`
`J. Appl. Phys., Vol. 78, No. 3, 1 August 1995
`
`qii$ijG$ijr
`
`s-pol
`
`LWoo
`2000
`
`2200
`WAVENUMBER
`
`[cm-‘]
`
`2
`
`0
`
`WAVENUMBER
`
`[cm-l]
`
`FIG. 6. IR-absorption spectra of chemomechanically polished Si(100) (main
`figures) and Si(ll1)
`(insets), referenced to Si(ll1) surfaces etched in a 40%
`solution of NH.,F. The sharp “negative” feature, labeled A, corresponds to
`the Si-H stretch absorption of this reference surface. The features labeled B
`and C correspond to SiH, vibration and O,SiH,
`(oxygen backbonded hy-
`drides), respectively. The dashed arrows around label C in the insets point to
`different types of oxygen backbonded hydrides. D and E correspond to the
`fingerprint regions of hydrocarbons and hydroxyls, respectively.
`
`dihydride termination on the Si(100) and to ideal monohy-
`dride termination on Si(lll)
`as discussed above. Second,
`there is a weak absorption, labeled C at -2250 cm-’ with
`contributions
`in both s and p polarizations for Si(100) and
`predominantly in p polarization for Si(ll1). Next, the region
`labeled D corresponds to the hydrocarbon absorption. The
`negative peaks indicate that slightly more hydrocarbons have
`adsorbed on the reference surface than on the sample surface
`during
`transport. Finally,
`the broad feature labeled E at
`-3100 cm-’
`is indicative of hydroxyl termination or possi-
`bly water adsorption. This region has already been discussed
`for the Si(ll1)
`surface,” so that it is not presented here.
`
`Pietsch, Chabal, and Higashi
`
`e
`
` Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. IP: 104.129.194.129 On: Tue, 01 Mar 2016
`23:12:15
`
`Raytheon2033-0006
`
`

`
`Since the presence of hydrocarbon contamination’3 (fea-
`ture D) may have important implications
`for processing, we
`review our measurement method here. By ratioing the spec-
`tra
`to
`the spectrum of
`the
`ideally monohydride-ter-
`minated Si(lll),
`the large contribution
`from hydrocarbons
`adsorbed on mirrors, windows, and detector is made negli-
`gible. However, the reference surface itself does adsorb hy-
`drocarbons after processing, as evidenced by the “negative”
`absorption peaks. We therefore cannot obtain an absolute
`measure of hydrocarbon contamination but a relative one to
`compare different CMP conditions. The expected features
`due to CH, in this IR region are the asymmetric “methyl”
`stretch mode, -CtiH3,
`at 2970 cm-‘,
`indicating
`the pres-
`ence of CH,, and the CH,-related modes at 2930 cm-’ and
`2855 cm-”
`(asymmetric, -CFtH2, and symmetric, -C=tH2,
`“ethylene” stretch, respectively). Thus, it is evident that both
`CHs and CH, groups are present. The main conclusion, how-
`ever, is that the overall hydrocarbon contamination is mini-
`mal, as evidenced by the fact that these features are some-
`times positive, sometimes negative.
`The absorption features around 2250 cm-’ are interest-
`ing because they provide information on the oxidation of the
`silicon during CMP A single broad, unpolarized feature at
`2250 cm-’
`is observed on CMP Si(lOO), while sharper, po-
`larized features are noted at 2113, 2250, 2285, and possibly
`2162 cm-’
`on CMP Si(ll1). Early model calculations,36
`electron energy-loss spectroscopy (EELS) studies in UHV of
`H adsorption on oxygen-exposed silicon,37 and IR studies of
`the wet-chemica138-40 or UV/ozone-promoted41’42 oxidation
`of H-terminated silicon have assigned these features to sili-
`con hydrides with oxygen backbonding: 2250 cm-i corre-
`sponds to the stretch frequency of both dihydride and mono-
`hydride backbonded to two oxygens atoms, i.e., dihydrides,
`Hz+Si=(O-),,
`and monohydrides, H+Si=[(O-)2,(Si=j],
`both at -2250 cm-r . The mode at 2283 cm-’ corresponds to
`the stretch mode of monohydride backbonded to three oxy-
`gen atoms, H+--Si=[(O-),I,
`while
`the modes at 2113 and
`2162 cm-’ are assigned to mono and dihydrides backbonded
`to a single oxygen atom, i.e., HtSi=[(O-),,(Si=)J
`and
`HztSi=[(O-)r
`,(Si=),],
`respectively.37 For Si( loo), the ob-
`served band points to the dominance of hydrides backbonded
`to two oxygen atoms, although it is not possible to rule out
`hydrides backbonded to one oxygen atom because of the
`broad band centered at 2110 cm-“. For the Si( 11 l), a variety
`of oxidation states can be seen, as indicated with the solid
`and dashed arrows in the inset of Fig. 6.
`None of these oxygen-backbonded hydride features are
`observed on
`freshly HF-etched silicon surfaces.21’2Z3234
`They therefore are produced during or after the chemome-
`chanical polishing. If the oxygen is produced during the pol-
`ishing, then it is suggestive that a partial oxidation may play
`a role in the removal mechanism during CMP. There is, how-
`ever, no measurable change in the oxygen-backbonded hy-
`dride features with different slurry pH (i.e., removal rate), as
`seeninFig.6[pH=ll
`and>14forSi(lOO),andpH=ll
`and
`7 for Si( 1 ll)].
`It is therefore not possible to show that partial
`oxidation plays a role. The data do not rule out such a
`mechanism, because it is believed that a partial oxidation is
`the rate-limiting process in Si etching when- HF is present,21
`
`with a very fast removal of the oxide and subsequent H ter-
`mination. As a result, the interpretation of the data remains
`ambiguous.
`In the region labeled E, the broad absorption feature ap-
`pearing in some spectra is characteristic of hydroxyl or water
`species.34 It has been observed on hydrophilic Si surfaces or
`for OH or H20 in or on silica surfaces.43 It is difficult
`to
`measure reliably because of adsorption of water on the mir-
`rors, windows, and detector. Once again, the sandwich tech-
`nique, not used in earlier studies,‘e is crucial to obtain reli-
`able data. On Si(lOO), a band at -3100
`cm-”
`is only
`observed at high pH, e.g. >14 (or also for low pH<7; not
`shown here). This frequency is lower
`than most forms of
`hydroxyls at surfaces: Isolated OH obtained on annealed
`glasses peaks at -3750 cm-‘,43 while water on hydrophilic
`oxide layers shows a maximum around 3400 cm-‘.a4 Shifts
`to
`lower
`frequencies are possible
`for H-bonded OH
`groups, ‘3,34 however, the position of the absorption band is
`also consistent with the presence of ammonium fluoride. If
`this band observed at very high pH (and also very low pH
`C7; not shown here) is correlated to oxide formation
`then
`the findings of contact angle measurements2’ can be ex-
`plained in a consistent manner.
`
`IV. DISCUSSION
`
`To arrive at a microscopic model for the removal mecha-
`nism, we first summarize the observations. The main conclu-
`sion from the above studies is that the surfaces, obtained by
`CMP at a pH yielding
`the highest removal rates, are very
`similar
`to
`those obtained
`by
`etching
`in
`buffered
`
`HF21*22*25*29-34 or hot water with dissolved oxygen.3Qv44*45 The
`Si( 111) surfaces have flat, -monohydride-terminated
`(111)
`planes separated mostly by [21 l]-type atomic steps; Si(100)
`surfaces are atomically rough with a predominance of dihy-
`dride termination. The stock removal rate is also higher for
`Si(100) than for Si(ll
`l), just as observed in chemical etch-
`ing. These simple observations point to a dominant chemical
`action of CMP under usual conditions, that is, after any thin
`oxide is mechanically removed, at a slurry pH around 11.
`Next, there is evidence that OH- plays an active role in
`the removal mechanism, since the removal rate increases
`with pH, i.e., with OH- concentration. Furthermore, the con-
`centration of OH- decreases during CMP’”
`if the syton is not
`pH stabilized by buffering
`the slurry pH. The depletion of
`OH- and its detection at higher pH suggest that it partici-
`pates directl