materials science
`
`A study of titanium nitride diffusion
`barriers between aluminium and silicon
`by X-ray absorption spectroscopy: the Si,
`Ti and N results
`
`Y.F. Hu,ab* T.K. Sham,b Z. Zou,cd G.Q. Xu,c L. Chan,d
`B.W. Yates,a G.M. Bancroftb
`
`aCanadian Synchrotron Radiation Facility, University of
`Wisconsin-Madison, 3731 Schneider Dr. Stoughton WI
`53589, USA, bDepartment of Chemistry, The University of
`Western Ontario London Ontario N6A 5B7, Canada,
`cDepartment of Chemistry Faculty of Science National
`University of Singapore 10 Kent Ridge 119260, Singapore,
`dChartered Semiconductor Manufacturing Pte. Ltd., 60
`Woodlands Industrial Park D, Street 2, 738406, Singapore.
`Email:yhu@julian.uwo.ca
`
`We report a multi-element, multi-edge and multi-detection mode
`X-ray photoabsorption study of a series of Al/TiNx/Si(100) thin
`films as a function of the TiNx film thickness (100Å-500Å) and
`of the annealing temperature (400 (cid:176)C-600 °C). The Si K- and L-
`edge results show that Si does not diffuse to the surface for all the
`films. The high resolution Ti L-edge and N K-edge spectra show
`that the TiNx layer undergoes a dramatic chemical reaction with
`the gradual increase in the annealing temperature. This chemical
`reaction stabilizes at 560 °C at which the TiNx film is known to
`fail to act as an effective diffusion barrier between Al and Si.
`
`Keywords: XANES, diffusion barrier, titanium nitride,
`oxidation reaction
`
`1.
`
`Introduction
`
`Titanium nitride (TiNx) thin films have long been used as an
`effective diffusion barrier between Al and Si in semiconductor
`technology because of theirs high thermal and chemical stability
`and low electrical resistivity (Wittmer, 1980). These properties
`allow TiNx to withstand the repeated thermal cycles in multi-step
`processes of integrated circuit devices. It is generally known that
`TiN barrier thin films are vulnerable to breakdown when
`annealed to temperatures above 550 °C due to the Al and Si inter-
`diffusion.
`
`techniques, such as X-ray photoelectron
`Many
`spectroscopy (XPS) (Prieto & Kirby, 1995), Auger electron
`spectroscopy (AES) (Tompkins, 1991; Kottke et al., 1991),
`electron energy-loss spectroscopy (EELS) (Walker et al., 1997),
`transmission electron microscopy (TEM) and electron dispersive
`spectroscopy (EDS) (Lee et al., 1999), and X-ray absorption
`spectroscopy (XAS) (Soriano, et al., 1993), have been used to
`study the thermal properties of TiNx films and to understand the
`failure mechanism of the TiNx barrier. In this work, we study a
`series of Al/TiNx/Si(100) thin films as a function of the TiNx film
`thickness (100Å-500Å) and of the annealing temperature (400
`(cid:176)C-600 °C), using multi-element, multi-edge and multi-detection
`mode X-ray photoabsorption spectroscopy. Major advantages of
`this technique for studying these sandwich samples are three
`folds: (1) By measuring XANES (X-ray Absorption Near Edge
`Structures) in both the surface sensitive total electron yield and
`the interface and bulk sensitive fluorescence yield, it provides
`
`depth information of a multi-layer material. (2) By studying
`XANES at different edges of different elements containing in a
`multi-layer material,
`it provides
`chemical
`composition
`information. (3) It is a non-destructive technique so there is little
`sample modification during the measurement, which is a common
`problem for the other conventional tools, such as AES. A review
`of this technique can be found elsewhere (Sham, et al., 2000).
`We have recently reported the XANES studies of these samples
`from the Al perspective (Zou, et al., 1999), we will focus on the
`Si, Ti, and N results in the present work.
`
`2. Experimental
`
`All the samples were prepared at the Chartered Semiconductor
`Manufacturing of Singapore. A Si(100) wafer was used as the
`substrate. Two sets of samples were studied. In the temperature
`series, a 300Å TiNx film was deposited onto the clean Si(100)
`substrate by sputtering a Ti target under nitrogen atmosphere.
`This TiNx layer was then capped by a 400Å Al film in an UHV
`chamber (using a Al target containing 5% Cu). Samples were
`then annealed for one hour under nitrogen atmosphere to a
`desired temperature (400 °C, 450 °C, 500 °C, 560 °C and 600
`°C). In the thickness series, all the samples were prepared in the
`same routine except that the thickness of the TiNx film varied
`from 100Å, 200Å, 300Å, 400Å to 500Å and all the samples were
`annealed to 560 °C for one hour.
`XANES measurements were performed at the Canadian
`Synchrotron Radiation Facility (CSRF) at the Synchrotron
`Radiation Center (SRC), University of Wisconsin-Madison. Si
`K-edge results were obtained using
`the Double Crystal
`Monochromator beamline, Si L-edge results were obtained using
`the grasshopper beamline, and the Ti L-edge and N K-edge
`results were obtained using the Canadian Spherical Grating
`Monochromator beamline. The photon energy was calibrated
`using a standard TiO2 (rutile), clean Si/SiO2 samples and N2 gas.
`XANES spectra were recorded in both the total electron yield
`(TEY) and the fluorescence yield (FY) modes. The TEY mode
`was measured by directly monitoring the sample current, it is
`surface and near surface sensitive with an estimated sampling
`depth of eg, 50Å at the Si L-edge (Kasrai, et al., 1996). The FY
`mode was measured using a detector comprised of two channel
`plates, it is generally sensitive to the bulk with an estimated
`sampling depth of eg, a few thousand angstrom at the Si K-edge
`(Kasrai, et al., 1996). All measurements were normalized to the
`incident flux. All reported spectra were recorded at the normal
`incidence angle.
`
`3. Results and Discussion
`
`3.1 Si results.
`
`It has been known that Si has some solid solubility in Al (Hanson
`& Anderko, 1958), and the Si content has been determined in the
`Al layer when annealed to high temperatures (Ting & Wittmer,
`1983). Our Al results showed that while the Al film oxidizes in
`the ambient, about 80% of the Al in the Al/TiNx/Si(100) system
`remains metallic after high temperature annealing, and there was
`no Si contribution (eg, formation of the Al silicide) observed in
`the Al K and L-edge XANES spectra (Zou, et al., 1999). Fig. 1
`shows the Si K-edge XANES for a series of Al/TiNx/Si(100)
`samples, with the thickness of the TiNx film varied from 100Å to
`500Å. Also shown in Fig. 1 are reference spectra of a clean
`Si(100) and of a TiSi2 sample. FY spectra of all these samples
`
`860 # 2001 International Union of Crystallography (cid:15) Printed in Great Britain ± all rights reserved
`
`J. Synchrotron Rad. (2001). 8, 860±862
`
`NVIDIA Corp.
`Exhibit 1108
`Page 001
`
`

`
`materials science
`
`mode of these samples and as a result, there is no significant
`spectral difference when the total thickness of the layer on top of
`Si is 500Å or 900Å. Second, there is a noticeable difference
`between Al/TiNx/Si(100) XANES and that of the Si(100) in the
`silicon oxide region (around 1850 eV). We attribute this
`difference to the formation of interface states (SiOx, other than
`SiO2) when Al/TiNx/Si(100) samples were annealed to 560 (cid:176)C.
`
`3.2 Ti L-edge results
`
`Fig. 2 presents the Ti L-edge TEY and FY spectra of three 400Å
`Al/300Å TiNx/Si(100) samples (as deposited and annealed to 450
`(cid:176)C and 560
`(cid:176)C, respectively) and of TiO 2 (rutile) and of a
`standard TiNx sample. For the as-deposited sample, both TEY
`and FY spectra show two peaks (around 458.5 and 464 eV photon
`energies) corresponding to transitions from the Ti 2p3/2 and 2p1/2
`initial states to vacant d orbitals. In previous XAS studies of
`TiNx films, only TEY mode was used and the oxide contribution
`at the surface was always present (see the TEY spectrum of the
`TiNx in Fig. 2, and see eg., Soriano et al., 1993, Esaka, et al.,
`1997). This is the first time that a clean TiNx spectrum is
`reported in both TEY and FY modes. This also confirms that the
`TiNx layer in our as-deposited sample was protected by the Al
`layer from oxidation before annealing.
`
`Figure 2
`Ti L-edge XANES of a series of Al/TiNx/Si(100) samples, as a function
`of the annealing temperature. The thickness of the Al and the TiNx layer
`is 400Å and 300 Å, respectively. Reference spectra of TiO2 (rutile) and a
`TiNx sample are also shown. Please note that the fluorescence yield
`spectrum of TiO2 single crystal is distorted due to self-absorption.
`
`Upon annealing, the TiNx layer in the Al/TiNx/Si(100)
`sample is slowly oxidized. The TEY spectrum of the 450 (cid:176)C
`annealed sample is almost identical to that of the ambient TiNx
`sample. At 560 (cid:176)C annealing, TiO 2 feature dominates the TEY
`spectrum, showing the spin-orbit and crystal field resolved fine
`structures, just like that of a true TiO2 spectrum (Fig.2 and van
`der Lann, 1990). In contrast to spectra measured in the TEY
`mode, FY spectra of Al/TiN/Si(100) samples exhibit no TiO2
`feature, even when the sample was annealed to 600 (cid:176)C (spectrum
`not shown). It is concluded that, based on the Ti L-edge results,
`the surface TiNx is oxidized to produce TiO2 during the annealing
`process, but the bulk TiNx layer remains intact.
`
`Figure 1
`Si K-edge XANES of a series of Al/TiNx/Si(100) samples, as a function
`of the thickness of TiNx layer. The thickness of the Al layer was 400Å
`for all the samples and all the samples were annealed to 560 (cid:176) C for one
`hour. Spectra of a clean Si(100) wafer and a TiSi2 are also shown.
`
`shown on the left side of Fig. 1 are all characteristic of a clean
`Si(100) substrate (Kasrai, et al., 1996). This is not surprising
`since the K shell fluorescence photons have a much larger escape
`depth than electrons (Henke et al., 1982) and the sampling depth
`of the Si K-edge in the FY mode is expected to be much larger
`than the thickness of the sample. Therefore, the contribution
`from the bulk Si, i.e., the Si(100) wafer, dominates in the FY
`mode at the Si K-edge of all the samples.
`On the other hand, the TEY mode at the Si K-edge was
`reported to have an estimated sampling depth of 700Å in a
`SiO2/Si(100) system (Kasrai, et al., 1996). The TEY mode
`allows us to probe the surface and interface Si content in these
`sandwich samples. TEY spectra of these Al/TiNx/Si(100)
`samples shown on the right side of Fig. 1 are all similar to that of
`the spectrum of a clean Si(100) but certainly different from that
`of the TiSi2. The resonance at ~1840 eV photon energy of the
`Si(100) wafer is due to the transition from the Si and the signal at
`~1846 eV is originated from transitions from the surface SiO2.
`Thus, these results indicate that there are interface states between
`Si and metal layer and these interface states are most likely Si
`oxides, but not of silicide origin.
`There are two other points worth noting about the TEY
`XANES of these samples. First, there is no obvious difference in
`the TEY spectra of these samples as the TiNx film thickness
`increases. This implies that there would be Si diffusion to the
`surface if the TEY sampling depth at Si K-edge were assumed to
`be ~700Å (Kasrai, et al., 1996). However, we could not detect
`any Si signal in the more surface sensitive Si L-edge XANES of
`these samples in both TEY and FY modes (spectra not shown),
`and there is no indication of the formation of the titanium silicide
`at the interface between the TiNx layer and the Si substrate. This
`titanium silicidation was commonly observed when the TiNx/Si
`system was annealed to high temperature (Kottke, et al., 1991).
`Furthermore, we did not observe the formation of Al and Si
`alloys in the surface and the bulk of these samples from the Al
`perspective (Zou et al., 1999). We think the effective sampling
`depth in the TEY mode at Si K-edge for these samples (with Al
`and TiNx on top of the Si) is probably greater than 700Å
`(estimated for a SiO2/Si system). Therefore, only the substrate Si
`(with less or no bulk Si contribution) was detected in the TEY
`
`J. Synchrotron Rad. (2001). 8, 860±862
`
`Y.F. Hu et al. 861
`
`NVIDIA Corp.
`Exhibit 1108
`Page 002
`
`

`
`interface and bulk of
`information for surface,
`structural
`complicated films. For the Al/TiNx/Si(100) films, our results
`showed that there is no inter-diffusion for Al and Si, even with
`high temperature annealing. When annealed to temperatures
`higher than 560 (cid:176)C, the TiN x layer is not an as effective diffusion
`barrier due to oxidation reactions.
`
`We thank staff at CSRF and SRC, especially Dr. K.H. Tan , for
`their technical assistance. Research performed at CSRF and
`UWO is supported by NSERC (Canada). SRC is supported by
`the US National Science Foundation Grant No. DMR-9531009.
`
`References
`
`Esaka, F., Furuya, K., Shimada, H., Imamura, M., Matsubayashi, N., Sata,
`H., Nishijima, Kawana, A., Ichimura, H. & Kikuchi, T. (1997) J. Vac.
`Sci. Technol. A 15, 2521-2528.
`Esaka, F., Shimada, H., Imamura, M., Matsubayashi, N., Kikuchi, T. &
`Furuya, K. (1998) J. Electron Spectrosc. Relat. Phenom. 88, 817-820.
`Hansen, M. & Anderko, A. (1958) Constitution of Binary Alloys, New
`York: McGraw-Hill.
`Henke, B.L., Lee, P., Tanaka, T.J., Shamabukuro, R.L. & Fujikawa, B.K.
`(1982) Atom. Data & Nucl. Data Tables 27, 1-144.
`Kasrai, M., Lennard, W.N., Brunner, R.W., Bancroft, G.M., Bardwell,
`J.A., & Tan, K.H. (1996) Appl. Surf. Sce. 99, 303-312.
`Kottke, M., Gregory, R., Pintchovski, F., Travis, E. & Tobin, P. J. (1991)
`J. Vac. Sci. Technol. B 9, 74-88.
`Laan, G. van der (1990) Phys. Rev. B 41, 12366-12368.
`Lee, H-J., Sinclair, R., Li, P. & Roberts, B. (1999) J. Appl. Lett. 86, 3096-
`3103.
`Pfluger, J., Find, J., Crecelius, G., Bohnen, K.P. & Winter, H. (1982)
`Solid State Comm. 44, 489-492.
`Prieto, P. & Kirby, R. E. (1995) J. Vac. Sci. Technol. A 13, 2819-2826.
`Sham, T.K., Naftel, S.J. & Coulthard, I. (2000) in Chemical Applications
`of Synchrotron Radiation, edited by T.K. Sham, Singapore: World
`Scientific, submitted.
`Soriano, L., Abbate, M., Fuggle, J.C., Prieto, P., Jimenez, C., Sanz, J.M.,
`Galan, L. &Hofmann, S. (1993) J. Vac. Sci. Technol. A 11, 47-51.
`Ting, C.Y. & Wittmer, M. (1983) J. Appl. Lett. 54, 937-943.
`Tompkins, H.G. (1991) J. Appl. Phys. 70, 3876-3880.
`Walker, C.G.H., Anderson, C.A., McKinley, A., Brown, N.M.D. & Joyce,
`A.M. (1997) Surf. Sci. 383, 248-260.
`Wittmer, M. (1980) Appl. Phys. Lett. 37, 540-542.
`Wittmer, M., Noser, J. & Melchior, H. (1981) J. Appl. Lett. 52, 6659-
`6664.
`Zou, Z., Hu, Y.F., Sham, T.K., Huang, H.H., Xu, G.Q., Seet, C.S. &
`Chan, L. (1999) J. Synchrotron Rad. 6, 524-525.
`
`materials science
`
`3.3 N K-edge results
`
`Fig. 3 illustrates the N K-edge TEY and FY spectra of the 400Å
`Al/300Å TiN/Si(100) samples as a function of annealing
`temperatures. FY spectra of all the samples are dominated by
`features due to the unreacted bulk TiNx (Pfluger, et al., 1982),
`with the gradual increase in the relative intensity of the 401 eV
`peak with the increase of the annealing temperature. This
`indicates that the bulk TiNx in these samples remains intact, from
`the N perspective, in agreement with findings based on the Ti L-
`edge results. The increase in the relative intensity of 401 eV peak
`is due to the oxidation reaction of the TiNx layer after the high
`temperature annealing (see below). We also note that FY spectra
`of 560 (cid:176)C and of 600 (cid:176)C annealed samples are virtually identical,
`implying the saturation of the oxidation reaction at 560 (cid:176)C.
`
`Figure 3
`N K-edge XANES of a series of 400Å Al/300Å TiNx/Si(100) samples, as
`a function of the annealing temperatures.
`
`When the 400Å Al/300Å TiNx/Si(100) sample was
`annealed to high temperatures, there is a more dramatic spectral
`change in the TEY spectra of this series of samples. There is a
`gradual decrease of the TiNx feature as the temperature increases.
`Oxidation of TiNx at 560 (cid:176)C and 600 (cid:176)C causes the TiN x feature
`to disappear almost completely from the TEY spectra. These
`spectra are dominated by a sharp peak at 401 eV photon energy
`which has been observed previously in TiNx films after similar
`thermal treatment (Soriano, et al., 1993, Prieto & Kirby, 1995,
`Esaka et al., 1997). This peak can be assigned to the contribution
`from the molecular N2, as a result of the TiNx oxidation. This
`(2TiNx+2O2fi2TiO 2+xN2)
`oxidation
`reaction
`is
`thermo-
`dynamically favorable (Wittmer, et al., 1981). The N2 molecule
`generated by this reaction was then trapped in the TiNx layer by
`physical and/or chemical absorption. The vibrational fine
`structure due to the interstitial N2 molecules, similar to that due to
`the N2 gas, has recently been resolved in the high resolution XAS
`spectrum of the thermally oxidized TiAlN film (Esaka, et al.,
`1998).
`
`4. Conclusion
`
`We demonstrated that the multi-element, multi-edge and multi-
`detection mode X-ray photoabsorption spectroscopy
`is a
`powerful, non-destructive technique in providing chemical and
`
`862 Received 31 July 2000 (cid:15) Accepted 22 November 2000
`
`J. Synchrotron Rad. (2001). 8, 860±862
`
`NVIDIA Corp.
`Exhibit 1108
`Page 003