by E. P. Gusev
`H.-C. Lu
`E. L. Garfunkel
`T. Gustafsson
`M. L. Green
`
`Growth and
`characterization
`of ultrathin
`nitrided silicon
`oxide films
`
`This paper reviews recent progress in
`understanding microstructural and growth-
`mechanistic aspects of ultrathin (<4 nm)
`oxynitride films for gate dielectric applications.
`Different techniques for characterizing these
`films are summarized. We discuss several
`nitridation methods, including thermal
`(oxy)nitridation in NO, N2O, and N2 as well as a
`variety of deposition methods. We show that a
`basic understanding of the gas-phase and
`thin-film oxygen and nitrogen incorporation
`chemistries facilitates the processing of
`layered oxynitride nanostructures with
`desirable electrical properties.
`
`1. Introduction
`While “pure” SiO2 films remained the principal material
`for gate dielectrics in MIS-based structures for more
`than three decades, the use of the traditional SiO2 gate
`dielectric has become questionable for sub-0.25-␮m ULSI
`devices [1–5]. Increasing problems with dopant (boron)
`penetration through ultrathin SiO2 layers and direct
`tunneling for ultrathin (⬍2 nm) oxide films dictate the
`search for and aggressive exploration of new materials for
`future gate dielectric applications with better diffusion
`barrier properties and higher dielectric constants [6, 7]. At
`
`this time, ultrathin silicon oxynitrides (SiOxNy or,
`more accurately, nitrogen-doped SiO2) are the leading
`candidates to replace pure SiO2 [8 –24]. Oxynitrides exhibit
`several properties superior to those of conventional
`thermal O2 oxides (SiO2), the more important being
`suppression of boron penetration from the poly-Si gate
`and enhanced reliability. Nitrogen also reduces hot-
`electron-induced degradation [25]. The dielectric constant
`of the oxynitride increases linearly with the percentage of
`nitrogen from ␧(SiO2) ⫽ 3.8 to ␧(Si3N4) ⫽ 7.8 [26, 27],
`though one should note that most SiOxNy films grown
`currently by thermal methods are lightly “doped” with N
`(⬍10 at.%) and therefore have a dielectric constant only
`slightly higher than that of pure SiO2.
`Recent publications suggest that the performance of
`CMOS-based devices depends on both the concentration
`and distribution of the nitrogen atoms incorporated into
`the gate dielectric [14, 16, 18, 28 –30]. For example,
`excessive nitrogen at the interface may reduce peak carrier
`mobility in the channel of MOSFETs and may allow boron
`accumulation in the oxide, which, in turn, may result in
`device instabilities [28]. The optimal nitrogen profile is
`determined by its specific application, although our
`incomplete understanding of the atomic-scale structural
`and electronic properties of dielectrics makes the desired
`structure an imperfectly defined goal. One possibility is
`an SiOxNy film with two nitrogen-enhanced layers: first,
`
`娀Copyright 1999 by International Business Machines Corporation. Copying in printed form for private use is permitted without payment of royalty provided that (1) each
`reproduction is done without alteration and (2) the Journal reference and IBM copyright notice are included on the first page. The title and abstract, but no other portions,
`of this paper may be copied or distributed royalty free without further permission by computer-based and other information-service systems. Permission to republish any other
`portion of this paper must be obtained from the Editor.
`
`265
`
`0018-8646/99/$5.00 © 1999 IBM
`
`IBM J. RES. DEVELOP. VOL. 43 NO. 3 MAY 1999
`
`E. P. GUSEV ET AL.
`
`TSMC 1120
`
`

`

`nitrogen at or near the Si/SiO2 interface to improve hot-
`electron immunity, and second, an even higher nitrogen
`concentration at the SiO2/polysilicon interface, as this is
`where it can best be used to minimize the penetration of
`boron from the heavily doped gate electrode [29]. The
`boron flux can be quite large (depending upon the thermal
`budget) and, therefore, higher nitrogen concentrations at
`the poly-Si interface are needed [28]. Although the actual
`“ideal” amounts of nitrogen required at each interface are
`not known, typical interfacial contents are of the order of
`(0.5–1) ⫻ 1015 cm⫺2 near each interface. The best methods
`to produce the desired ultrathin SiOxNy films are still
`under debate.
`Nitrogen may be incorporated into SiO2 using either
`thermal oxidation/annealing [8, 9, 12–14, 31–39] or
`chemical and physical deposition [15, 16, 18, 19, 40, 41]
`methods (Figure 1). Thermal nitridation of SiO2 in NO or
`N2O generally results in a relatively low concentration of
`nitrogen in the films, of the order of 1015 N/cm2 [20, 29,
`33, 42]. Since nitrogen content increases with temperature,
`thermal oxynitridation is typically performed at high
`temperatures, i.e., ⬎800⬚C [20, 29, 33, 42]. For more heavily
`N-doped SiOxNy films, other deposition methods, such as
`chemical vapor deposition (CVD) [40] with different
`precursors and its low-pressure (LP) and/or rapid thermal
`(RT) variants [18], jet vapor deposition (JVD) [43],
`atomic layer deposition (ALD) [44], or nitridation by
`energetic nitrogen particles (plasma, N atoms, or ions)
`[45–55], can be used. These nitridation methods can be
`performed at lower temperatures, ⬃300 – 400⬚C. However,
`low-temperature deposition methods may result in
`nonequilibrium films, and subsequent thermal processing
`steps are often required to improve film quality and
`minimize defects and induced damage [19, 56]. Because
`
`the thermodynamics [57, 58] of the SiOxNy system and
`the kinetics [12, 20, 29, 32, 38, 59 – 63] of nitrogen
`incorporation are rather complex, these different methods
`produce oxynitride films with different total nitrogen
`concentrations and depth distributions. From a scientific
`viewpoint, the addition of N into the Si–O system opens a
`number of questions concerning microstructure, defects,
`and growth mechanisms, issues which are still under
`intense debate even for the much simpler “pure” SiO2/Si
`system [64 –71].
`Characterizing the nitrogen distribution in ultrathin
`films with the required sub-nm accuracy is an analytical
`challenge. Conventional depth profiling approaches,
`such as SIMS (secondary ion mass spectroscopy) [9, 14,
`18, 34, 72] and HF etch-back methods [12, 33, 35] [in
`combination with XPS (X-ray photoelectron spectroscopy),
`NRA (nuclear reaction analysis), etc.], offer limited depth
`resolution, especially for ultrathin dielectrics. In addition,
`SIMS analysis is complicated by matrix effects [73], while
`HF etching may introduce nonuniform oxide removal
`(especially in the presence of local nitrogen-rich regions)
`and other deleterious chemical effects. We have recently
`demonstrated [62, 63, 74 –76] that high-resolution
`(⌬E/E ⬇ 0.1%) medium-energy ion scattering (MEIS)
`[77] is a useful technique for accurately obtaining the
`depth distribution profile of nitrogen in 1– 4-nm oxynitride
`films with sub-nm accuracy.
`In this paper, we review our recent progress in 1)
`characterizing nitrogen depth profiles (by MEIS
`[42, 74, 76]) and bonding (by XPS [78]) of ultrathin SiOxNy
`films, and 2) understanding the mechanisms of nitrogen
`incorporation into ultrathin oxide films [42, 62, 63, 79, 80].
`We demonstrate how this basic knowledge can be used to
`guide nitrogen nanoengineering technology, in particular to
`
`E. P. GUSEV ET AL.
`
`IBM J. RES. DEVELOP. VOL. 43 NO. 3 MAY 1999
`
`266
`
`

`

`produce layered nitrogen structures [81]. A significant part
`of the paper is devoted to thermal oxynitridation of Si in
`NO and N2O. Other nitridation methods are also briefly
`discussed.
`
`2. Thermodynamics of the Si–N–O system
`The bulk phase diagram of the Si–N–O system is shown
`in Figure 2. The diagram consists of four phases: Si, SiO2
`(cristobolite, tridymite), Si3N4, and Si2N2O [57, 58]. The
`three compound phases have similar structural units: SiO4
`tetrahedra for SiO2, SiN4 tetrahedra for Si3N4, and slightly
`distorted SiN3O tetrahedra for Si2N2O, implying that the
`phases can be converted from SiO2 to Si2N2O and finally
`to Si3N4 by replacing oxygen with nitrogen. However, the
`nitride (Si3N4) and the oxide (SiO2) phases never coexist
`in the bulk under equilibrium conditions. They are always
`separated by the oxynitride (Si2N2O), which is the only
`thermodynamically stable and crystalline form of silicon
`oxynitride.
`A puzzling question is why nitrogen atoms are
`incorporated at all in SiO2. According to thermodynamic
`equilibrium, nitrogen should not incorporate into an SiO2
`film that is grown on Si in almost any partial pressure of
`oxygen, i.e., ⬎10⫺17 atm, depending upon temperature
`[57, 58]. One can see from Figure 2 that the Si2N2O/SiO2
`phase boundary exists at about 10⫺18 atm for T ⫽ 1400 K.
`At any oxygen partial pressure greater than that, which
`surely exists in a furnace or rapid thermal processing
`(RTP) reactor or, for example, during N2O or NO
`decomposition at high temperatures, only SiO2 phases
`(crystalline or amorphous) should form. Therefore,
`nitrogen in bulk SiO2 is not thermodynamically stable.
`At least two reasons for the presence of nitrogen in the
`SiO2 film can be suggested. First, nitrogen atoms may
`simply be kinetically trapped at the reaction zone near the
`interface (i.e., the nitrogen is present in a nonequilibrium
`state, where the rate of the transition to equilibrium is
`slow and some N is trapped) or by structural defects in
`the SiO2 film [82]. The basic idea in this model is that
`nitrogen brought into the film during oxynitridation (as,
`for example, NO; see below) reacts only with Si–Si bonds
`at or near the interface, not with Si–O bonds in the bulk
`of the SiO2 overlayer. Alternatively, the nitrogen at the
`interface may indeed be thermodynamically stable due
`to the presence of free-energy terms that are not yet
`understood. For example, nitrogen may lower the
`interfacial strain known to exist at the SiO2/Si interface.
`This could explain why incorporated nitrogen (especially
`in N2O or NO oxides) is often associated with the Si/SiO2
`interface, consistent with a special, stabilizing role of the
`nitrogen at the interface. Even when nitrogen is implanted
`into Si, it tends to migrate to the Si/SiO2 interface during
`oxidation and be incorporated in the SiO2 [83, 84].
`Therefore, there is some evidence that the nitrogen
`
`plays a specific role at the interface, but also that it is not
`stable away from it. On the other hand, nonequilibrium
`techniques (for example, plasma nitridation) have
`yielded oxynitrides with much higher concentrations of
`incorporated nitrogen (including top-surface nitridation
`[49]) than thermal oxynitridation methods (e.g., in NO,
`N2O, N2). This fact seems to be supportive of the “kinetic
`trapping” concept. Finally, we note that the favorable
`thermodynamics of the SiO2 phase discussed above may be
`the reason why Si3N4 films (produced by JVD or other
`methods) always contain some oxygen; of course,
`contamination by H2O, O2, etc. with a high sticking
`coefficient would also result in mixed oxide–nitride
`phases.
`
`3. Diffusion-barrier properties of nitrided layers
`An important property of nitrogen in nitrided oxides is
`that it forms a barrier against the diffusion of boron.
`Concurrent with this, it also lowers the diffusion rates
`for oxygen, nitric oxide, and other dopants, significantly
`slowing the rate of further oxidation or nitridation [21, 32,
`34, 63, 82, 85–90]. For example, for a 2-nm oxynitride
`with ⬃1 ML (monolayer) of nitrogen (6.8 ⫻ 1014 N/cm2)
`located near the interface, the rate of continued oxidation
`at 900⬚C decreases by at least a factor of 4 relative to the
`pure oxide. A reasonable argument can be made that the
`decrease in film growth rate (due to nitrogen) results from
`
`267
`
`IBM J. RES. DEVELOP. VOL. 43 NO. 3 MAY 1999
`
`E. P. GUSEV ET AL.
`
`

`

`been suggested [91]. For a 1.5-nm SiO2 film, the diffusivity
`at 900⬚C would increase by a factor of 24 as compared
`with 10-nm oxides. According to this model, the role
`of nitrogen is that the N atoms compete with B for
`occupation of the defect sites. The model was criticized by
`Ellis and Buhrman [92], who argued that 1) Si–O–N–O–Si
`structures, which should form after N passivation, are not
`observed in XPS; and 2) according to percolation theory,
`to fit the experimental data the peroxy defects should have
`an unreasonably high concentration (⬃40% of the Si–O–Si
`bonds). Ellis and Buhrman developed a model in which
`boron diffuses substitutionally for Si atoms, and the role
`of the Si–N bond is to impede substitution for that Si
`atom [92]. The model was incorporated into a Monte
`Carlo simulation and showed good agreement with
`experimental data.
`Another interesting aspect of the diffusion barrier
`property of nitrided layers is that silicon interstitials
`generated during the film growth reaction at the interface
`are blocked from diffusing into the oxide [93]. This
`results in an enhanced flux of the interstitials into the Si
`substrate, which in turn yields an increased density of
`oxidation stacking faults and may also affect oxidation-
`induced diffusion [93]. Finally, we note that other
`chemistries may play an important role both in film
`growth processes and in the more technically important
`issue, electrical defects that occur in the final devices.
`Hydrogen, water, and various other species quite
`possibly exist at finite concentrations [94]. For example,
`both fluorine and hydrogen enhance the boron diffusion
`rate [91]. Hydrogen may also play an important role in the
`diffusion/reaction processes during (oxy)nitride formation
`[40].
`
`4. Materials characterization of ultrathin
`nitrided oxides
`Two major problems in measuring nitrogen in SiOxNy films
`are the following: 1) In many cases the concentration of
`nitrogen in the film is rather low (as an example, in earlier
`studies of NO oxynitrides [95] it was claimed that the films
`are essentially SiO2 because the nitrogen was undetectable
`at that time); and 2) the films in question are very thin
`(⬍5 nm), suggesting that sub-nm depth resolution is
`required to monitor nitrogen depth profiles. Over the past
`few years, several techniques, such as secondary ion mass
`spectroscopy (SIMS) [18, 34, 72, 73, 78], nuclear reaction
`analysis (NRA) [29, 32, 33, 38, 59, 80, 96, 97], medium-
`energy ion scattering (MEIS) [39, 42, 62, 63, 74 –76, 80],
`X-ray photoelectron spectroscopy (XPS) [12, 30, 34, 35,
`37, 48, 74, 78, 87, 98 –101], Auger electron spectroscopy
`(AES) [19], Fourier transform infrared spectroscopy
`(FTIR) [102], spectroscopic ellipsometry [103], and others
`have been utilized to study nitrogen concentrations,
`depth profiles, nitrogen bonding, and microstructure of
`
`a decreased rate of diffusion. The density of nitrides and
`oxynitrides is higher than that of the pure oxide [40];
`thus, the diffusivity of NO, O2, N2, or other nonreactive
`molecular species such as noble gases should be lower in
`the nitrogen-containing films. However, another property
`involving N bonds may be equally important: The lattice
`itself may become more rigid on an atomic level. The
`three bonds connected to each nitrogen (as in Si3N4) are
`more constrained than the two bonds of each O atom in
`SiO2, where the Si–O–Si bond angles can go from 120⬚ to
`180⬚ with little change in energy; this may also contribute
`to a decrease in the ability of the nitrided lattice to
`permit diffusion of atoms and small molecules. The latter
`argument is valid for either an interstitial diffusion mode
`(of molecules), or an exchange-hopping substitutional
`mode (of excess atoms).
`By analogy, both physical (e.g., density) and chemical
`arguments have been used to explain the important effect
`of the reduced penetration rate of boron from the poly-Si
`gate into nitrided oxides. (Boron penetration causes
`threshold voltage shifts and degrades reliability.) To
`explain the “stopping power” of the nitrogen in oxides, a
`model assuming boron diffusion via peroxy linkage defects
`(Si–O–O–Si bonds), whose concentration changes under
`different processing conditions and film thicknesses, has
`
`268
`
`E. P. GUSEV ET AL.
`
`IBM J. RES. DEVELOP. VOL. 43 NO. 3 MAY 1999
`
`

`

`Table 1
`Selected nuclear reactions for oxynitride
`characterization.
`
`Target
`element
`
`Nuclear
`reaction
`
`Energy
`(keV)
`
`Sensitivity
`(atoms/cm2)
`
`H
`
`D
`14N
`15N
`15N
`
`16O
`18O
`18O
`
`28Si
`29Si
`
`30Si
`
`H(15N, ␣␥)12C
`D(3He, p)4He
`
`14N(d, ␣0,1)12C
`15N(p, ␣)12C
`15N(p, ␣␥)12C
`
`16O(d, p)17O
`18O(p, ␣)15N
`18O(p, ␣)15N
`
`28Si(p, ␥)29P
`29Si(p, ␥)30P
`
`30Si(p, ␥)31P
`
`6420
`
`700
`
`1100
`
`1000
`
`429
`
`850
`
`730
`
`151
`
`371
`
`324
`417
`
`499
`
`⬃1014
`⬃1012
`⬃5 ⫻ 1013
`⬃1012
`⬃1013
`(estimated)
`⬃1014
`⬃1012
`⬃1013
`(estimated)
`
`⬃1013
`(estimated)
`⬃1013
`(estimated)
`
`oxynitrided films. Below we outline the features and the
`utility of these techniques for ultrathin oxynitride studies.
`SIMS, a standard technique used in the industry
`to monitor concentration profiles in semiconductor
`structures, was one of the first methods applied to the
`nitrogen depth distribution problem in oxynitrided films.
`The technique has a rather high sensitivity (of the order
`of 0.001 at.%), can be performed rapidly, and shows good
`long-term reproducibility [73]. As an example, Figure 3
`illustrates nitrogen depth profiles for ⬃5-nm oxynitride
`films grown by thermal oxynitridation of Si(100) in N2O
`(both in a furnace and in an RTP reactor), and in O2
`followed by NO [74]. One can see different depth
`profiles depending on the processing conditions. The
`NO-annealed film has the highest concentration of
`nitrogen incorporated near the SiOxNy/Si interface. The
`RTN2O (rapid thermal oxynitridation) film also shows
`nitrogen located near the interface, whereas the furnace-
`grown film has a broader nitrogen depth distribution. Both
`N2O oxynitrides have nitrogen concentrations lower than
`the NO-annealed film. The technique begins to reach the
`limits of depth resolution (estimated to be ⬃2–3 nm) for
`sub-5-nm films. Another more important complication for
`SIMS analysis are “matrix effects,” in which the sputtered
`nitrogen ion yield depends strongly on the local film
`chemistry around the nitrogen. For example, the (CsN⫹)
`ion yield from nitrogen in bulk Si is about six times
`smaller than in SiO2, while that for N near the interface is
`about three times that observed in the bulk of the SiO2
`film [73]. The use of CsN⫹ ions as the detected species
`
`seems to minimize the matrix effect, since for negative
`ions matrix effects are even more severe. In most
`cases, SIMS analysis is also complicated by surface
`contamination and initial sputtering effects which make
`measurements of the nitrogen in the top (⬃1 nm) surface
`layers not very meaningful. Finally, one more metrological
`aspect of the SIMS analysis is that the N concentration
`shown for the interface peak (see Figure 3) may not
`be accurate due to ion mixing. The areal density of
`the N peak should be used instead. Reference samples
`(especially with different nitrogen distributions) calibrated
`by other quantitative methods (NRA, MEIS, etc.) may
`be helpful for more accurate quantitative SIMS analysis.
`NRA analysis of oxynitrides is based on the detection
`of protons, ␣-particles, and ␥-rays generated in nuclear
`reactions of nitrogen, oxygen, silicon, and hydrogen
`(deuterium) induced by high-energy charged particles
`(Table 1) [29, 32, 33, 38, 59, 80, 96, 97]. The cross sections
`for these reactions can be determined in independent
`calibration experiments, and the signal-to-background
`ratio is in many cases quite favorable. Since the nuclear
`reaction rates are sensitive only to the number of nuclei in
`the films, the technique yields absolute concentrations of
`the species investigated. In particular, the technique allows
`one to determine the absolute concentration of 14N
`with an accuracy of ⬃7–10% and a detection limit of
`
`⬃5 ⫻ 1013 atoms/cm2 via the reaction of 14N(d, ␣0,1)12C
`induced by 1.1-MeV deuterons. Figure 4 shows the nitrogen
`content (which is an increasing function of temperature)
`in oxynitride films grown by rapid thermal oxynitridation
`in N2O [33]. The oxygen content in the films can also be
`measured by NRA, for example by the 16O(d, p)17O
`
`269
`
`IBM J. RES. DEVELOP. VOL. 43 NO. 3 MAY 1999
`
`E. P. GUSEV ET AL.
`
`

`

`approximately 1-nm scale under favorable conditions.
`Depth profiling can also be performed by an HF
`acid etch-back with subsequent NRA measurement
`of N remaining in the film [13]. One more useful
`application of NRA is based on its ability to measure
`hydrogen/deuterium [H(15N, ␣␥)12C at 429 keV and
`D(3He, p)4He at 700 keV] since H/D is believed to be
`important in both the nitridation mechanism and device
`performance [105, 106]. Finally, we note that nitrogen,
`oxygen, silicon, and hydrogen can also be monitored by
`the technique of elastic recoil detection (ERD) of primary
`MeV ions [86].
`MEIS, another ion beam technique, is based on
`the same principles of ion–solid interactions as is
`conventional Rutherford backscattering spectroscopy
`(RBS) [77]. Because of the lower ion energy used in
`MEIS (typically 100 keV, at which the energy loss of
`protons in a solid is maximal), and the use of a high-
`resolution toroidal electrostatic ion energy analyzer [107]
`(⌬E/E ⬇ 0.1%), almost-monolayer depth resolution can
`be obtained [108, 109]. We have demonstrated that a
`0.3– 0.5-nm accuracy (near the top surface or for very thin
`films) in the nitrogen depth profiles can be achieved [42, 74].
`To our knowledge, this is the best depth resolution
`for nitrogen in SiOxNy films achieved so far. However,
`due to the statistical nature of the ion energy loss in the
`solid (straggling effect) [110], the resolution decreases
`with increasing distance from the surface. MEIS allows one
`to monitor simultaneously both absolute concentrations
`and depth profiles of N, O, and Si in the film. The
`detection limit of nitrogen in SiO2 is about 2–3 ⫻ 1014
`atoms/cm2, depending on the film thickness and the width
`of the nitrogen distribution in the film. It is worth noting
`that, in opposition to other depth-profiling techniques
`(e.g., SIMS), both the sensitivity to nitrogen and the
`depth resolution increase with decreasing dielectric film
`thickness. Another strength of MEIS is mass sensitivity,
`which enables isotopic (18O and 16O) labeling experiments
`[111, 112]. Light elements (for example, hydrogen
`and boron) can also be detected in the elastic recoil
`configuration, as demonstrated by Copel and Tromp
`[113, 114].
`Figure 5(a) shows a typical MEIS spectrum for an
`SiO2 film annealed in NO. Three peaks are seen in this
`spectrum, corresponding to nitrogen, oxygen, and silicon.
`The lighter mass (nitrogen) yields a peak at lower energy,
`as given by classical two-body scattering kinematics [77].
`The areas under the peaks (after corrections for the
`known cross sections) are proportional to the total
`amounts of oxygen and nitrogen in the film [77].
`Obviously, the amount of nitrogen is much smaller
`than that of oxygen. The shape of the nitrogen peak is
`determined by the depth distribution of nitrogen in the
`film. Because of energy loss arising from electronic
`
`reaction at 850 keV. This allows one to calculate the
`thickness of the films (provided that film density is
`known). An additional advantage of the NRA technique is
`the detection of nitrogen (15N) and oxygen (18O) isotopes
`with the help of the following reactions: 15N(p, ␣)12C at
`1000 keV, 18O(p, ␣)15N at 730 keV, and the resonant
`reactions of 15N(p, ␣␥)12C at 429 keV and 18O(p, ␣)15N at
`151 keV. Isotopic labeling has been proven to be a useful
`method to study mechanistic aspects (including nitrogen
`and oxygen transport) of silicon (oxy)nitridation [38, 59,
`97, 104]. Owing to the narrow resonances (with widths of
`120 and 100 eV, respectively), the latter two reactions
`can be used for depth profiling of 15N and 18O on an
`
`270
`
`E. P. GUSEV ET AL.
`
`IBM J. RES. DEVELOP. VOL. 43 NO. 3 MAY 1999
`
`

`

`excitations, protons backscattered from nitrogen atoms
`closer to the Si–SiOxNy interface have lower energy than
`those scattered off the atoms near the surface. Accurate
`depth distributions [Figure 5(b)] of Si, O, and N are
`deduced from simulations of energy spectra, as discussed
`in [111]. More details about the MEIS setup, data
`acquisition and analysis, and the isotopic labeling
`technique can be found elsewhere [74, 108, 111, 112,
`115–117].
`The above ion beam methods are unfortunately
`not capable of examining local nitrogen bonding in
`(oxy)nitrided films. Photoemission (XPS) is a common
`technique used to help determine local bonding
`configurations of atoms on a surface or in a thin film
`[68, 118 –121]. XPS analysis is based on the fact that
`the energies of the electronic core levels are altered by
`the local electronic configurations. The observed energy
`levels shift with changing local chemical environment (the
`so-called chemical shifts). Early XPS studies were limited
`to the detection of nitrogen at the interface of SiO2
`annealed in nitrogen [87, 98]. More recent high-resolution
`(including synchrotron-based) photoemission studies of
`core levels of nitrogen (N 1s) and silicon (Si 2p) atoms
`were useful in understanding nitrogen bonding and depth
`distribution (with HF etch-back or variable photoelectron
`takeoff angle methods) [35, 37, 78, 99, 122]. Depth
`analysis in XPS is determined by the escape depth of the
`N 1s or Si 2p photoelectrons, which are of the order of
`2–3.5 nm for conventional AlK␣ or MgK␣ X-ray sources.
`N 1s spectra for NO and N2O oxynitrides are shown in
`Figure 6 [78]. The nitrogen spectrum for the N2O-grown
`film has much lower intensity than that of the NO
`oxynitride, which indicates that an NO source is more
`effective in terms of nitrogen incorporation into the
`dielectric film. A second feature to point out is that the
`spectra for both NO and N2O oxynitrides have similar
`shapes (see inset), suggesting that the local bonding
`configurations of nitrogen in the two films are similar.
`A more detailed analysis of the N 1s peak shape shows
`that the peak is asymmetric and consists of at least two
`components [35, 78, 99]. The lower binding energy (BE)
`component at ⬃397.6 eV (calibrated with respect to the
`Si 2p peak at 99.2 eV) is located closer to the interface,
`while the higher BE component at 398.3–399.0 eV (the
`thicker the film, the higher the value) corresponds to
`nitrogen atoms located further into the overlayer film. The
`lower BE peak can be assigned to N atoms bonded to
`three Si atoms, as follows from the observation of an N 1s
`peak at ⬃397.6 eV from a reference Si3N4 film [35, 99].
`The shift of the higher BE component originates from
`both electrostatic (charging, core hole screening) and
`structural effects. The structural effects may include
`different bond length and bond angles in the “bulk” of
`the film and near the interface, strain near the interface,
`
`and/or second-neighbor effects (i.e., bonding/atoms in
`the shell of second neighbors near the interface may be
`different from those in the middle of the oxynitride film)
`[100]. N–O bonds are unlikely to be present at high
`concentration in the NO and N2O oxynitrides because the
`chemical shift of ⬃1.5 eV (relative to the N–Si bond)
`calculated recently [100] for this configuration is larger
`than any experimentally observed ones. Recent work by
`R. Opila and J. Chang demonstrated that nitrogen profiles
`obtained from angular resolved XPS were consistent with
`MEIS depth profiles. They also showed that for oxynitride
`films processed by plasma, metastable nitrogen states may
`be observed with binding energies greater than 400 eV.
`Finally, we note that XPS is also useful in obtaining
`film thickness [123] (from the ratio of Si 2p peaks
`corresponding to the film and Si substrate) and
`interface structure [30, 35, 68, 118, 121, 124] (from the
`concentration of Si1⫹, Si2⫹, and Si3⫹ suboxide states
`observed in the Si 2p spectrum). Recent XPS studies of
`the suboxide states in N2O oxynitrides were interpreted
`as giving evidence for a lower defect density at the
`SiOxNy/Si interface with respect to the SiO2/Si interface
`[35]. However, an opposite effect (i.e., an increased
`concentration of the Si1⫹ state and an unchanged
`concentration of the Si1⫹ and Si2⫹) has also been reported
`
`271
`
`IBM J. RES. DEVELOP. VOL. 43 NO. 3 MAY 1999
`
`E. P. GUSEV ET AL.
`
`

`

`and models [103]. A structural two-layer model of N2O
`oxynitride films with an ⬃1.4 –1.6-nm-thick Si2N2O phase
`near the interface was suggested to explain a shift of the
`main peak in the FTIR spectrum as a function of the
`thickness of the films [102]. Optical second-harmonic
`spectroscopy has been used to study strain at the interface
`[125, 126]. Roughness at the SiOxNy/Si interface was
`studied by X-ray diffraction. It was found that, for N2O
`oxynitrides, the (RMS) roughness is smaller than for pure
`SiO2 films and decreases with temperature [33].
`
`5. Thermal (oxy)nitridation methods
`
`● Nitrogen incorporation into ultrathin dielectrics by NO
`processing
`Oxidation of silicon and annealing of SiO2 in nitrous
`(N2O) or nitric (NO) oxides are the leading hydrogen-
`free processing methods for making nitrided oxides by
`conventional thermal routes [20, 21, 29]. Oxynitridation in
`N2O is particularly attractive because 1) it allows one to
`incorporate what appears to be an appropriate amount of
`nitrogen near the SiOxNy/Si interface (typically ⬃5 ⫻ 1014
`atoms/cm2); and 2) its processing similarity to O2 permits
`N2O to replace oxygen in oxidation reactors/furnaces.
`However, among other factors, oxynitridation in N2O is
`complicated by the fast gas-phase decomposition of the
`molecule into N2, O2, NO, and O at typical oxidation
`temperatures, 800 –1100⬚C (see the subsection on gas-
`phase N2O decomposition which follows). NO is now
`believed to be responsible for nitrogen incorporation into
`the film [21, 29, 31, 37, 60, 127, 128], suggesting that
`understanding oxynitridation in NO is a necessary step
`before considering more complex gases such as N2O.
`If NO is the main species responsible for nitrogen
`incorporation into the film, oxynitridation in pure NO
`should be considered for ultrathin dielectrics, especially
`in processes in which thermal budget and film thickness
`issues are crucial. Compared to N2O, oxynitridation in NO
`results in more nitrogen incorporation at equivalent
`temperatures [14, 34, 37, 74, 82]. In addition, NO
`oxynitrides exhibit lower leakage currents and interface
`defect densities, as well as improved electrical stress
`properties [14, 34]. However, channel mobility may be
`reduced if the nitrogen concentration near the interface
`is too high.
`Figure 7 shows the total amounts of oxygen and
`nitrogen in ultrathin films on Si(100) after exposure in NO
`for an hour in the temperature range of 700 –1000⬚C [42].
`As the temperature increases, the total amounts of both
`nitrogen and oxygen increase. One should also note that
`the ratio of nitrogen to oxygen in the film increases with
`increasing temperature (increasing by ⬃50% from 700 to
`1000⬚C). This was also observed during the initial stage of
`the interaction of NO with Si(111) at much lower pressure
`
`[30]. Another electron spectroscopy, AES, can also be
`used to measure nitrogen content in SiOxNy films and has
`been used effectively in studies of plasma-nitrided samples
`[19]. Though the AES sensitivity to nitrogen is similar to
`that of XPS, the interpretation of spectral features is
`usually more difficult than in XPS because the Auger
`electron emission process is more complex than
`photoemission.
`Finally we discuss some other techniques for
`characterizing nitrogen and SiOxNy film microstructure.
`The amount of nitrogen can be crudely estimated
`from oxidation kinetics measurements. Since nitrogen
`significantly retards transport/reactions in (oxy)nitrided
`films [32, 34], this simple method can be used to monitor
`the presence of nitrogen in the film and to compare
`its relative value in different samples. However,
`the retardation rate depends not only on the N concentration
`but also on the depth distribution. A given amount of N
`evenly distributed in the film would produce a diffusion
`barrier quite different from one in which the nitrogen
`distribution is sharply peaked, making quantitative analysis
`difficult. Spectroscopic (immersion) ellipsometry results on
`N2O- and NO-grown oxynitrides show an agreement with
`SIMS measurements, though one should keep in mind that
`ellipsometric analysis is very dependent on parameters
`
`272
`
`E. P. GUSEV ET AL.
`
`IBM J. RES. DEVELOP. VOL. 43 NO. 3 MAY 1999
`
`

`

`(10⫺6 Torr) [129]. In other words, the film becomes more
`nitride-like at higher temperatures. The fact that the
`concentration of nitrogen increases with temperature has
`an important implication for the (oxy)nitridation of silicon
`in N2O. It was observed that in the case of N2O, a higher
`(oxy)nitridation temperature also gives rise to a higher
`nitrogen content in the film (Figure 4); this was attributed
`to a higher percentage of NO in the product stream
`resulting from the N2O gas-phase decomposition at higher
`temperatures. Our results with NO as the only reactive gas
`clearly show that the solid-state chemistry is also important
`in the understanding of the inc