- Email: [email protected]
Investigations of titanium nitride as metal gate material, elaborated by metal organic atomic layer deposition using TDMAT and NH3
Microelectronic Engineering 82 (2005) 248–253 www.elsevier.com/locate/mee
Investigations of titanium nitride as metal gate material, elaborated by metal organic atomic layer deposition using TDMAT and NH3 F. Fillot a
a,b
, T. Morel a, S. Minoret a, I. Matko b, S. Maıˆtrejean B. Guillaumot c, B. Chenevier b, T. Billon a
a,*
CEA-DRT-LETI, CEA grenoble, 17 rue des martyrs, 38054 Grenoble Cedex 9, France b LMGP-UMR-CNRS 5628, 38402 St. Martin dÕHe`res, France c STMicroelectronics, 38926 Crolles Cedex, France Available online 29 August 2005
Abstract This study reports for the first time, the evaluation of the work function and thermal stability of TiN gate material for deep sub-micron CMOS, elaborated by using metal organic atomic layer deposition, from TDMAT and NH3 precursors. Composition, microstructure and electrical properties of atomic layer deposited TiN films are characterized by using combined analytical techniques. The TiN films exhibit suitable properties for nMOSFET requirement with an effective work function of 4.2 eV obtained on silicon oxide and a good stability up to 1050 C. The effective work function measured on high-k dielectric (HfO2) is found to be 4.3 eV and the stability upon high temperature annealing is less favorable. 2005 Published by Elsevier B.V. Keywords: Metal gate; TiN; Work function; ALD; TDMAT
1. Introduction As silicon devices are scaled below 45 nm, metal gate and advanced high-k materials are required to obtain less than 1 nm Equivalent Oxide Thickness *
Corresponding author. Tel.: +33 4 38789749; fax: +33 4 38783034. E-mail address: [email protected] (S. Maıˆtrejean). 0167-9317/$ - see front matter 2005 Published by Elsevier B.V. doi:10.1016/j.mee.2005.07.083
(EOT) [1]. Metal gate electrodes appear more promising than doped polysilicon gates, which present many restricting factors, like high resistivity, poly depletion, boron penetration and instability on high-k materials. The selection of new metal gate materials is a challenge because it is necessary to account for several critical points: the gate key parameters are the electrical properties (work function, resistivity), the chemical compatibility
F. Fillot et al. / Microelectronic Engineering 82 (2005) 248–253
of the gate material with the underlying dielectric, and the thermal stability of the MOS structure after standard CMOS process thermal step. Several metallic candidates exhibit attractive properties, like single metals [2], metal nitrides [3], conductive oxides [4] or more complex alloys [5]. Recently, Park et al. [6] have shown the benefits of a damage-free direct metal gate process using Atomic Layer Deposition (ALD) method for TiN deposition on SiO2 gate oxide. If ALD is employed to obtain TiN metal gate, the interface state density (Dit) is lower than the one detected in PVD (Physical Vapor Deposition) films or standard CVD (Chemical Vapor Deposition) films. Furthermore, the leakage currents are also reduced with ALD. However, Westlinder et al. [7] report strong work function instability of ALD-TiN from TiCl4/ NH3 with high thermal budget. Moreover, Moriwaki and Yamada [8] have shown that the amount of residual chlorine in TiN and the reliability degradation of the gate oxide are strongly correlated. For these reasons, others precursors than halide must be studied for TiN ALD. Metal Organic Atomic Layer Deposition (MOALD) of TiN using TetrakisDiMethylAmido Titanium (TDMAT) and ammoniac (NH3) has been already reported [9,10]. MOALD with these precursors has the advantage to be halide free. In this work, MOALD of TiN using TDMAT precursor is proposed to make a new possible metal gate material for deep sub-micron CMOS process technology. Studies of microstructure, chemical properties and work function of TiN films deposited on SiO2 or HfO2 are presented.
249
sition, the SiO2 substrates were cleaned by rapid thermal cleaning at 350 C for 25 s. The deposition process is made by applying a series of successive reactant pulses separated each other by purge. To achieve the TDMAT pulse, the titanium precursor was generated in an external bubbler. TDMAT is introduced into the chamber with a He carrier gas flow. NH3 was introduced in the chamber as ammonia pulse. Nitrogen (N2) and Helium (He) were introduced for the complete separation of the precursor and the reactant gas purging the deposition chamber. During deposition, substrate wafers were kept at 180 C and the pressure was kept constant at 5 Torr. Previous experiments have shown that in these conditions, the film growth rate is close to 2 mono layers per cycle [11]. 2.2. Characterization The composition and density of the TiN films were determined from Rutherford Backscattering Spectrometry (RBS). RBS used a 1 MeV He+ source with a backscattering angle of 135. Impurities incorporation and compositional profiles in thin films were measured by secondary ion mass spectroscopy (SIMS, Cameca IMS 5F Ion spectrometer). A Cs+ primary ion beam with an impact energy of 3 keV was used to analyze an area of 200 lm in diameter. For microstructure analysis, a Bruker diffractometer equipped with a Cu Ka source radiation was used. HRTEM images were obtained by using a JEOL 2010FX microscope. 2.3. Work function measurements
2. Experimental 2.1. Deposition TiN films were deposited by using a MOALD deposition technique in a MOCVD chamber (TxZ: Applied Materials Technology Inc., ENDURA5500) using a commercial 8 in. single wafer deposition tool. TiN thin films were deposited on 7 nm oxidized (1 0 0) silicon wafer by using TDMAT as the titanium precursor and ammonia (NH3, 500 sccm) as the reactant gas. Prior to depo-
To assess the TiN work function, simple Metal Oxide Semiconductor (MOS) capacitors were fabricated by using a single damascene process on silicon p-type (1 0 0) oriented silicon and boron implanted silicon substrate. After cleaning, a 600 nm thick oxide was thermally grown. A conventional photolithography process was then used to define a 0.0072 mm2 active region. Dry oxidation at 800 C was performed to grow the gate oxide (3 nm thick). On some of the capacitors, a 4 nm thick layer of HfO2 is deposited by
250
F. Fillot et al. / Microelectronic Engineering 82 (2005) 248–253
ALD from HfCl4/H2O precursors. Metallic TiN layers were then deposited by MOALD at 180 C. The typical thickness deposited is about 20 nm on the gate oxide. Shunt doped Si-poly was deposited at 580 C on gate material to obtain electrical contact and fill the cavity. Delay between TiN and Si-poly deposition is minimized in order to reduce TiN oxidation. A standard CMP step is finally applied on the capacitors. All samples were annealed in N2, H2 gas at 425 C for 30 min (Forming Gas anneal, FG) in order to reduce fixed oxide charges and saturate free silicon bounds. A Rapid Thermal Anneal (RTA) was then performed on selected samples in pure N2 between 600 and 1050 C for 30 s. The high frequency capacitance–voltage (C–V) characteristic was measured at 1 and 100 kHz frequencies. The flat band voltage (VFB) was estimated by calculating the flat band capacitance and was compared with the full consistent quantum modeling of MOS capacitor C–V characteristics [12]. In a MOS capacitor, the VFB is proportional to the gate material work function and to the amount of charges (interfacial charges or fixed charges) present in the dielectric. A simple extraction of VFB can give a reliable estimation of the work function of the metal. Assuming that the density of charge Nox, is located at the silicondielectric interface, the metal work function UM is given by the dependence of the flat band voltage (VFB) on the oxide thickness (EOT) by the following classical equation: V FB ¼ UM USi
N ox EOT; eox
3. Results 3.1. Microstructure X-ray diffraction patterns were recorded from 210 nm thick as-deposited TiN films (Fig. 1). The weak diffraction lines at 2h 36.7 and 42.6 can be indexed as 1 1 1 and 2 0 0 TiN (NaCl structure type). The intensity ratio indicates a slight (1 1 1) preferred orientation of the grains. The largely broadened profiles show that the film is nano crystallized. The average grain size, estimated from a Scherrer type analysis, is close to 2 nm. The asdeposited TiN microstructure can thus be described as an assembly of very small crystallites. The TEM cross-section image of Fig. 2 confirms
Si 004
TiN TiN 111 200
10
20
30
40
50
60
70
80
90
100
2 Théta angle (˚)
Fig. 1. XRD pattern of ALD-TiN as-deposited.
ð1Þ
where USi is the work function of the silicon substrate, eox the dielectric permittivity. Previous reports have shown that in TiN samples deposited on thermal SiO2, the interface charge is very small (<6 · 1010 cm2) [13]. This indicates the charge dependent term in Eq. (1) is much smaller than UM and USi and can thus be omitted. For work function measurements on HfO2, the interface charge for ALD HfO2 is close to 6 · 1011 cm2, the interface charge dependent term in Eq. (1) is roughly 20 · 103 eV and thus has to be taken into account.
Fig. 2. TEM cross-section of Si-poly/ALD-TiN/SiO2/Si stack annealed at 1050 C.
F. Fillot et al. / Microelectronic Engineering 82 (2005) 248–253
The composition of the TiN layer has been obtained by using RBS from a 210 nm thick TiN layer. The Ti and N contents are, respectively, 27% and 30%, i.e. a 1.1 atomic ratio. A 37% high oxygen level is also detected. Some carbon is also incorporated in the film: 6%. The density of the TiN estimated by RBS is 3.1 g cm3. The SIMS profile of Fig. 3, recorded from a 20 nm thick TiN sample, shows high surface oxidation of TiN, constant titanium, nitrogen and carbon in the volume of the layer. The variations of the Ti:N content ratio close to the TiN/SiO2 interface is given in Fig. 4. It is important to note that a Ti rich layer extending over 2 nm is clearly detected. At the maximum, the Ti:N ratio is 1.8. 3.3. Electrical characterization To analyze the work function temperature dependence, flat band voltages of MOS capacitors with p-type substrate were obtained at various annealing temperatures. Fig. 5 shows no significant variation of flat band voltage after annealing a high temperature when the dielectric used as gate oxide is SiO2. Concerning MOS capacitor with
SiO2
TiN
1,8 1,6 1,4
interface
3.2. Composition
2
stoechiométric ratio : Ti / N
that RTA annealing at 1050 C does not significantly alter the TiN film microstructure.
251
1,2 1 0,8 0,6 8
10
12
14
16
18
22
20
Depth of TiN (nm)
Fig. 4. Ti/N ratio evolution as a function of TiN film depth. The Ti-rich layer extends from 17 to 19.2 nm.
HfO2 gate oxide, the flat band voltage presents a good stability below 800 C (see Fig. 5). Both gate oxide EOTs are nearly constant up to 800 C (Fig. 6). At higher temperatures the EOT ˚ at increases: it is less important for SiO2 (4 A ˚ 1050 C) than for HfO2 (7 A at 1050 C). Finally, the work function variations in TiN/ SiO2 and TiN/HfO2 systems as a function of RTA temperature have been obtained by incorporating measured VFB and EOT values in Eq. (1) (see Table 1). In TiN/SiO2, UM is close to 4.2 eV and does not exhibit significant variations as the annealing temperature is varied up to 1050 C. In TiN/HfO2, the UM is slightly higher (4.3 eV) and 0
SiO2
1,E+06
Si
SiO2
TiN
HfO2
Flat band voltage (V)
-0,2
O
1,E+05
Intensity (c/s)
Ti N
1,E+04
C
1,E+03
*
-0,6
-0,8
H
Si
1,E+02
-0,4
-1 H
Ti
1,E+01 0
5
10
C 15
N 20
O 25
Depth (nm)
Fig. 3. SIMS profile of the TiN deposited on SiO2.
Si 30
0
200
400
600
800
1000
1200
RTA temperature (˚C)
Fig. 5. Flat band voltages of a poly Si/TiN/oxide/Si capacitor versus RTA annealing temperature (30 s, N2).
252
F. Fillot et al. / Microelectronic Engineering 82 (2005) 248–253 50
SiO2 HfO2
EOT (A)
40
30
20
10
0 0
200
400
600
800
1000
1200
RTA temperature(˚C)
Fig. 6. EOTs of a poly Si/TiN/oxide/Si capacitor versus RTA annealing temperature (30 s, N2).
presents a poor stability as the annealing temperature is increased and reaches 4.47 eV at 1050 C. 4. Discussion TiN films elaborated by MOALD exhibit nanocrystallized microstructure that is not significantly modified by high temperature annealing up to 1050 C. As no significant grain preferred orientation has been detected, no effect of work function anisotropy is expected for very small gate length. ALD self-limiting reaction between TDMAT and NH3 lead to lower carbon rate in the layer than standard pyrolysis of TDMAT [14]. FTIR measurements not reported here have shown the evidence of N–C2 and C–H3 bounds in the TiN layer [11]. During pathway of reaction, the product of reaction, dimethylamino molecules, are incorporated in the layer. This can explain the low density of TiN layer and oxygen absorption during air exposure. Oxygen in the layer may be responsible for the EOT increase after RTA annealing.
Usually, work function of TiN is found close to silicon mid gap. However, 4.2 eV was measured. The composition profile observable by SIMS shows a titanium interface rich when it is deposited on SiO2. As titanium work function is close to 4.2 eV, the titanium rich layer located at the TiN/SiO2 interface is possibly responsible for the of 4.2 eV low effective work function. This work function on SiO2 is stable with high temperature RTA. This is an essential condition to be integrated in a standard CMOS process. Concerning work function measurements on HfO2, the values extracted are higher than those obtained on silicon oxide. Moreover, work function increases with RTA temperature from 4.3 to 4.5 eV. We could remark that the effective work function on HfO2 is different than the effective work function on SiO2, and is pinned toward 4.5 eV, a value which corresponds to the HfO2 charge neutrality level of (/CNL(HfO2) = 4.5 eV [15]). This could either results from dipole formation by intrinsic or extrinsic states [16]. On one hand, intrinsic states or Metal Induced Gap States (MIGS) should only depend on physical properties of metal and dielectric. Thus, they may be independent of thermal treatment. On the other hand, extrinsic states are linked to chemical reaction and defects at the metal/dielectric interface and vary with thermal treatments. The effective work function variations as a function of thermal treatments suggest that both MIGS and extrinsic states exist at TiN/HfO2 interface. Considering the behavior difference between SiO2 and HfO2 flat band-voltage evolutions as a function of annealing temperature, we could suggest the following understanding. As silicon is more electronegative than titanium (vTi = 1.54, vSi = 1.90), the reaction of Ti with SiO2 is easily activated and chemical bounds can be achieved at low temperature. In this case,
Table 1 Work function of TiN/oxide as a function of RTA (30 s, N2) temperature (C) Annealing treatment
Only FG
RTA 600 + FG
RTA 700 + FG
RTA 800 + FG
RTA 900 + FG
RTA 1050 + FG
Effective work function (eV) on SiO2 Effective work function (eV) on HfO2
4.18 4.29
4.19 4.29
4.18 4.30
4.16 4.37
4.16 4.44
4.19 4.47
F. Fillot et al. / Microelectronic Engineering 82 (2005) 248–253
we assume that the extrinsic states are created during the low thermal treatment (forming gas anneal) and have no influence at higher thermal treatments. As hafnium is less electronegative than titanium (vHf = 1.3, vTi = 1.54), the reaction between Ti and HfO2 is not activated at low temperature and thus, extrinsic states are not created after Forming gas annealing. The formation of extrinsic states will be thermally activated only with higher thermal budget. In this hypothesis, extrinsic state formation is thought to control the charge transfer and dipole formation occurring between metal gate and gate oxide. It is clearly dependent on the RTA temperature. We can thus conclude that Fermi level pining can induce effective work function dependence of the gate dielectric. The observed dependence of work function on thermal treatments results likely from the variations of extrinsic states.
5. Conclusion MOALD TiN has been evaluated as metal gate material, by using combined material and electrical characterizations. Ti–N films elaborated at 180 C are nano crystallized and show no significant preferential crystalline orientation. A TiN phase – NaCl structure type – is identified. The high oxygen level detected in the layer is attributed to low density of the deposited material and its sensibility to oxygen absorption during air exposure. In the TiN films deposited on SiO2, a titanium rich layer has been detected at the interface. The existence of this layer may explain the low work function of TiN, 4.2 eV close to the Ti work function. This work function is stable with high temperature annealing. The work function measurements carried out on HfO2 are sensitively higher. The effective work function on HfO2 is clearly dependent on annealing temperature. This dependence may be attributed to formation of extrinsic states, which pin the Fermi level of metal gate at the metal oxide interface.
253
Acknowledgements This work has been carried out in the frame of the nanoCMOS European project. The authors gratefully acknowledge Y. Morand for stimulating discussions, P. Holliger, G. Rolland, F. Pierre, V. Vidal and Y. Campidelli for characterizations, and Applied Materials for financial support.
References [1] ITRS, International Technology Roadmap for Semiconductors. Available from:. [2] V. Narayanan, A. Callegari, F.R. McFeely, K. Nakamura, P. Jamison, S. Zafar, E. Cartier, A. Steegen, V. Ku, P. Nguyen, K. Milkove, C. Cabral, M. Gribelyuk, C. Wajda, Y. Kawano, D. Lacey, Y. Li, E. Sikorski, E. Duch, H. Ng, C. Wann, R. Jammy, M. Ieong, G. Shahidi, Symp VLSI: Tech. Dig. Tech. Pap. (2004) 192–193. [3] H.Y. Yu, H.F. Lim, J.H. Chen, M.F. Li, C. Zhu, C.H. Tung, A.Y. Du, W.D. Wang, D.Z. Chi, D.L. Kwong, IEEE Elec. Dev. Lett. 24 (4) (2003) 230–232. [4] K. Fro¨hlich, K. Husekova, D. Machajdik, J.C. Hooker, N. Perez, M. Fanciulli, S. Ferrari, C. Wiemer, A. Dimoulas, G. Vellianitis, F. Roozeboom, Mater. Sci. Eng. B 109 (2004) 117–121. [5] Y.S. Suh, G. Heuss, V. Misra, J. Vac. Sci. Technol. B 22 (1) (2004) 175–179. [6] D.G. Park, K.Y. Lim, H.J. Cho, T.H. Cha, I.S. Yeo, J.S. Roh, J.W. Park, Appl. Phys. Lett. 80 (14) (2002) 2514– 2516. [7] J. Westlinder, T. Schram, L. Pantisano, E. Cartier, A. Kerber, G.S. Lujan, J. Olsson, G. Groeseneken, IEEE Electron. Dev. Lett. 24 (9) (2003) 550–552. [8] M. Moriwaki, T. Yamada, Jpn. J. Appl. Phys. 40 (2001) 2679–2684. [9] H.K. Kim, J.Y. Kim, J.Y. Park, Y. Kim, Y.D. Kim, H. Jeon, J. Korean Phys. Soc. 41 (5) (2002) 739–744. [10] J.W. Elam, M. Schuisky, J.D. Ferguson, S.M. George, Thin Solid Films 436 (2003) 145–156. [11] F. Fillot, Thesis document of INPG, 2005. [12] R. Clerc, A.S. Spinelli, G. Ghibaudo, C. Leroux, G. Pananakakis, Microelectron. Reliab. 41 (7) (2001) 1027– 1030. [13] F. Fillot, S. Maıˆtrejean, T. Farjot, B. Guillaumot, B. Chenevier, G. Passemard, MRS Proc. Symp. 786 (2004) 259–265. [14] C. Marcadal, M. Eizenberg, A. Yoon, L. Chen, J. Electrochem. Soc. 149 (1) (2002) C52–C58. [15] A.A. Demkov, L.R.C. Fonseca, J. Tomfohr, O.F. Sankey, MRS Proc. Symp. 786 (2004) 29–34. [16] J. Robertson, J. Vac. Sci. Technol. B 18 (3) (2000) 1785– 1791.
Investigations of titanium nitride as metal gate material, elaborated by metal organic atomic layer deposition using TDMAT and NH3 F. Fillot a
a,b
, T. Morel a, S. Minoret a, I. Matko b, S. Maıˆtrejean B. Guillaumot c, B. Chenevier b, T. Billon a
a,*
CEA-DRT-LETI, CEA grenoble, 17 rue des martyrs, 38054 Grenoble Cedex 9, France b LMGP-UMR-CNRS 5628, 38402 St. Martin dÕHe`res, France c STMicroelectronics, 38926 Crolles Cedex, France Available online 29 August 2005
Abstract This study reports for the first time, the evaluation of the work function and thermal stability of TiN gate material for deep sub-micron CMOS, elaborated by using metal organic atomic layer deposition, from TDMAT and NH3 precursors. Composition, microstructure and electrical properties of atomic layer deposited TiN films are characterized by using combined analytical techniques. The TiN films exhibit suitable properties for nMOSFET requirement with an effective work function of 4.2 eV obtained on silicon oxide and a good stability up to 1050 C. The effective work function measured on high-k dielectric (HfO2) is found to be 4.3 eV and the stability upon high temperature annealing is less favorable. 2005 Published by Elsevier B.V. Keywords: Metal gate; TiN; Work function; ALD; TDMAT
1. Introduction As silicon devices are scaled below 45 nm, metal gate and advanced high-k materials are required to obtain less than 1 nm Equivalent Oxide Thickness *
Corresponding author. Tel.: +33 4 38789749; fax: +33 4 38783034. E-mail address: [email protected] (S. Maıˆtrejean). 0167-9317/$ - see front matter 2005 Published by Elsevier B.V. doi:10.1016/j.mee.2005.07.083
(EOT) [1]. Metal gate electrodes appear more promising than doped polysilicon gates, which present many restricting factors, like high resistivity, poly depletion, boron penetration and instability on high-k materials. The selection of new metal gate materials is a challenge because it is necessary to account for several critical points: the gate key parameters are the electrical properties (work function, resistivity), the chemical compatibility
F. Fillot et al. / Microelectronic Engineering 82 (2005) 248–253
of the gate material with the underlying dielectric, and the thermal stability of the MOS structure after standard CMOS process thermal step. Several metallic candidates exhibit attractive properties, like single metals [2], metal nitrides [3], conductive oxides [4] or more complex alloys [5]. Recently, Park et al. [6] have shown the benefits of a damage-free direct metal gate process using Atomic Layer Deposition (ALD) method for TiN deposition on SiO2 gate oxide. If ALD is employed to obtain TiN metal gate, the interface state density (Dit) is lower than the one detected in PVD (Physical Vapor Deposition) films or standard CVD (Chemical Vapor Deposition) films. Furthermore, the leakage currents are also reduced with ALD. However, Westlinder et al. [7] report strong work function instability of ALD-TiN from TiCl4/ NH3 with high thermal budget. Moreover, Moriwaki and Yamada [8] have shown that the amount of residual chlorine in TiN and the reliability degradation of the gate oxide are strongly correlated. For these reasons, others precursors than halide must be studied for TiN ALD. Metal Organic Atomic Layer Deposition (MOALD) of TiN using TetrakisDiMethylAmido Titanium (TDMAT) and ammoniac (NH3) has been already reported [9,10]. MOALD with these precursors has the advantage to be halide free. In this work, MOALD of TiN using TDMAT precursor is proposed to make a new possible metal gate material for deep sub-micron CMOS process technology. Studies of microstructure, chemical properties and work function of TiN films deposited on SiO2 or HfO2 are presented.
249
sition, the SiO2 substrates were cleaned by rapid thermal cleaning at 350 C for 25 s. The deposition process is made by applying a series of successive reactant pulses separated each other by purge. To achieve the TDMAT pulse, the titanium precursor was generated in an external bubbler. TDMAT is introduced into the chamber with a He carrier gas flow. NH3 was introduced in the chamber as ammonia pulse. Nitrogen (N2) and Helium (He) were introduced for the complete separation of the precursor and the reactant gas purging the deposition chamber. During deposition, substrate wafers were kept at 180 C and the pressure was kept constant at 5 Torr. Previous experiments have shown that in these conditions, the film growth rate is close to 2 mono layers per cycle [11]. 2.2. Characterization The composition and density of the TiN films were determined from Rutherford Backscattering Spectrometry (RBS). RBS used a 1 MeV He+ source with a backscattering angle of 135. Impurities incorporation and compositional profiles in thin films were measured by secondary ion mass spectroscopy (SIMS, Cameca IMS 5F Ion spectrometer). A Cs+ primary ion beam with an impact energy of 3 keV was used to analyze an area of 200 lm in diameter. For microstructure analysis, a Bruker diffractometer equipped with a Cu Ka source radiation was used. HRTEM images were obtained by using a JEOL 2010FX microscope. 2.3. Work function measurements
2. Experimental 2.1. Deposition TiN films were deposited by using a MOALD deposition technique in a MOCVD chamber (TxZ: Applied Materials Technology Inc., ENDURA5500) using a commercial 8 in. single wafer deposition tool. TiN thin films were deposited on 7 nm oxidized (1 0 0) silicon wafer by using TDMAT as the titanium precursor and ammonia (NH3, 500 sccm) as the reactant gas. Prior to depo-
To assess the TiN work function, simple Metal Oxide Semiconductor (MOS) capacitors were fabricated by using a single damascene process on silicon p-type (1 0 0) oriented silicon and boron implanted silicon substrate. After cleaning, a 600 nm thick oxide was thermally grown. A conventional photolithography process was then used to define a 0.0072 mm2 active region. Dry oxidation at 800 C was performed to grow the gate oxide (3 nm thick). On some of the capacitors, a 4 nm thick layer of HfO2 is deposited by
250
F. Fillot et al. / Microelectronic Engineering 82 (2005) 248–253
ALD from HfCl4/H2O precursors. Metallic TiN layers were then deposited by MOALD at 180 C. The typical thickness deposited is about 20 nm on the gate oxide. Shunt doped Si-poly was deposited at 580 C on gate material to obtain electrical contact and fill the cavity. Delay between TiN and Si-poly deposition is minimized in order to reduce TiN oxidation. A standard CMP step is finally applied on the capacitors. All samples were annealed in N2, H2 gas at 425 C for 30 min (Forming Gas anneal, FG) in order to reduce fixed oxide charges and saturate free silicon bounds. A Rapid Thermal Anneal (RTA) was then performed on selected samples in pure N2 between 600 and 1050 C for 30 s. The high frequency capacitance–voltage (C–V) characteristic was measured at 1 and 100 kHz frequencies. The flat band voltage (VFB) was estimated by calculating the flat band capacitance and was compared with the full consistent quantum modeling of MOS capacitor C–V characteristics [12]. In a MOS capacitor, the VFB is proportional to the gate material work function and to the amount of charges (interfacial charges or fixed charges) present in the dielectric. A simple extraction of VFB can give a reliable estimation of the work function of the metal. Assuming that the density of charge Nox, is located at the silicondielectric interface, the metal work function UM is given by the dependence of the flat band voltage (VFB) on the oxide thickness (EOT) by the following classical equation: V FB ¼ UM USi
N ox EOT; eox
3. Results 3.1. Microstructure X-ray diffraction patterns were recorded from 210 nm thick as-deposited TiN films (Fig. 1). The weak diffraction lines at 2h 36.7 and 42.6 can be indexed as 1 1 1 and 2 0 0 TiN (NaCl structure type). The intensity ratio indicates a slight (1 1 1) preferred orientation of the grains. The largely broadened profiles show that the film is nano crystallized. The average grain size, estimated from a Scherrer type analysis, is close to 2 nm. The asdeposited TiN microstructure can thus be described as an assembly of very small crystallites. The TEM cross-section image of Fig. 2 confirms
Si 004
TiN TiN 111 200
10
20
30
40
50
60
70
80
90
100
2 Théta angle (˚)
Fig. 1. XRD pattern of ALD-TiN as-deposited.
ð1Þ
where USi is the work function of the silicon substrate, eox the dielectric permittivity. Previous reports have shown that in TiN samples deposited on thermal SiO2, the interface charge is very small (<6 · 1010 cm2) [13]. This indicates the charge dependent term in Eq. (1) is much smaller than UM and USi and can thus be omitted. For work function measurements on HfO2, the interface charge for ALD HfO2 is close to 6 · 1011 cm2, the interface charge dependent term in Eq. (1) is roughly 20 · 103 eV and thus has to be taken into account.
Fig. 2. TEM cross-section of Si-poly/ALD-TiN/SiO2/Si stack annealed at 1050 C.
F. Fillot et al. / Microelectronic Engineering 82 (2005) 248–253
The composition of the TiN layer has been obtained by using RBS from a 210 nm thick TiN layer. The Ti and N contents are, respectively, 27% and 30%, i.e. a 1.1 atomic ratio. A 37% high oxygen level is also detected. Some carbon is also incorporated in the film: 6%. The density of the TiN estimated by RBS is 3.1 g cm3. The SIMS profile of Fig. 3, recorded from a 20 nm thick TiN sample, shows high surface oxidation of TiN, constant titanium, nitrogen and carbon in the volume of the layer. The variations of the Ti:N content ratio close to the TiN/SiO2 interface is given in Fig. 4. It is important to note that a Ti rich layer extending over 2 nm is clearly detected. At the maximum, the Ti:N ratio is 1.8. 3.3. Electrical characterization To analyze the work function temperature dependence, flat band voltages of MOS capacitors with p-type substrate were obtained at various annealing temperatures. Fig. 5 shows no significant variation of flat band voltage after annealing a high temperature when the dielectric used as gate oxide is SiO2. Concerning MOS capacitor with
SiO2
TiN
1,8 1,6 1,4
interface
3.2. Composition
2
stoechiométric ratio : Ti / N
that RTA annealing at 1050 C does not significantly alter the TiN film microstructure.
251
1,2 1 0,8 0,6 8
10
12
14
16
18
22
20
Depth of TiN (nm)
Fig. 4. Ti/N ratio evolution as a function of TiN film depth. The Ti-rich layer extends from 17 to 19.2 nm.
HfO2 gate oxide, the flat band voltage presents a good stability below 800 C (see Fig. 5). Both gate oxide EOTs are nearly constant up to 800 C (Fig. 6). At higher temperatures the EOT ˚ at increases: it is less important for SiO2 (4 A ˚ 1050 C) than for HfO2 (7 A at 1050 C). Finally, the work function variations in TiN/ SiO2 and TiN/HfO2 systems as a function of RTA temperature have been obtained by incorporating measured VFB and EOT values in Eq. (1) (see Table 1). In TiN/SiO2, UM is close to 4.2 eV and does not exhibit significant variations as the annealing temperature is varied up to 1050 C. In TiN/HfO2, the UM is slightly higher (4.3 eV) and 0
SiO2
1,E+06
Si
SiO2
TiN
HfO2
Flat band voltage (V)
-0,2
O
1,E+05
Intensity (c/s)
Ti N
1,E+04
C
1,E+03
*
-0,6
-0,8
H
Si
1,E+02
-0,4
-1 H
Ti
1,E+01 0
5
10
C 15
N 20
O 25
Depth (nm)
Fig. 3. SIMS profile of the TiN deposited on SiO2.
Si 30
0
200
400
600
800
1000
1200
RTA temperature (˚C)
Fig. 5. Flat band voltages of a poly Si/TiN/oxide/Si capacitor versus RTA annealing temperature (30 s, N2).
252
F. Fillot et al. / Microelectronic Engineering 82 (2005) 248–253 50
SiO2 HfO2
EOT (A)
40
30
20
10
0 0
200
400
600
800
1000
1200
RTA temperature(˚C)
Fig. 6. EOTs of a poly Si/TiN/oxide/Si capacitor versus RTA annealing temperature (30 s, N2).
presents a poor stability as the annealing temperature is increased and reaches 4.47 eV at 1050 C. 4. Discussion TiN films elaborated by MOALD exhibit nanocrystallized microstructure that is not significantly modified by high temperature annealing up to 1050 C. As no significant grain preferred orientation has been detected, no effect of work function anisotropy is expected for very small gate length. ALD self-limiting reaction between TDMAT and NH3 lead to lower carbon rate in the layer than standard pyrolysis of TDMAT [14]. FTIR measurements not reported here have shown the evidence of N–C2 and C–H3 bounds in the TiN layer [11]. During pathway of reaction, the product of reaction, dimethylamino molecules, are incorporated in the layer. This can explain the low density of TiN layer and oxygen absorption during air exposure. Oxygen in the layer may be responsible for the EOT increase after RTA annealing.
Usually, work function of TiN is found close to silicon mid gap. However, 4.2 eV was measured. The composition profile observable by SIMS shows a titanium interface rich when it is deposited on SiO2. As titanium work function is close to 4.2 eV, the titanium rich layer located at the TiN/SiO2 interface is possibly responsible for the of 4.2 eV low effective work function. This work function on SiO2 is stable with high temperature RTA. This is an essential condition to be integrated in a standard CMOS process. Concerning work function measurements on HfO2, the values extracted are higher than those obtained on silicon oxide. Moreover, work function increases with RTA temperature from 4.3 to 4.5 eV. We could remark that the effective work function on HfO2 is different than the effective work function on SiO2, and is pinned toward 4.5 eV, a value which corresponds to the HfO2 charge neutrality level of (/CNL(HfO2) = 4.5 eV [15]). This could either results from dipole formation by intrinsic or extrinsic states [16]. On one hand, intrinsic states or Metal Induced Gap States (MIGS) should only depend on physical properties of metal and dielectric. Thus, they may be independent of thermal treatment. On the other hand, extrinsic states are linked to chemical reaction and defects at the metal/dielectric interface and vary with thermal treatments. The effective work function variations as a function of thermal treatments suggest that both MIGS and extrinsic states exist at TiN/HfO2 interface. Considering the behavior difference between SiO2 and HfO2 flat band-voltage evolutions as a function of annealing temperature, we could suggest the following understanding. As silicon is more electronegative than titanium (vTi = 1.54, vSi = 1.90), the reaction of Ti with SiO2 is easily activated and chemical bounds can be achieved at low temperature. In this case,
Table 1 Work function of TiN/oxide as a function of RTA (30 s, N2) temperature (C) Annealing treatment
Only FG
RTA 600 + FG
RTA 700 + FG
RTA 800 + FG
RTA 900 + FG
RTA 1050 + FG
Effective work function (eV) on SiO2 Effective work function (eV) on HfO2
4.18 4.29
4.19 4.29
4.18 4.30
4.16 4.37
4.16 4.44
4.19 4.47
F. Fillot et al. / Microelectronic Engineering 82 (2005) 248–253
we assume that the extrinsic states are created during the low thermal treatment (forming gas anneal) and have no influence at higher thermal treatments. As hafnium is less electronegative than titanium (vHf = 1.3, vTi = 1.54), the reaction between Ti and HfO2 is not activated at low temperature and thus, extrinsic states are not created after Forming gas annealing. The formation of extrinsic states will be thermally activated only with higher thermal budget. In this hypothesis, extrinsic state formation is thought to control the charge transfer and dipole formation occurring between metal gate and gate oxide. It is clearly dependent on the RTA temperature. We can thus conclude that Fermi level pining can induce effective work function dependence of the gate dielectric. The observed dependence of work function on thermal treatments results likely from the variations of extrinsic states.
5. Conclusion MOALD TiN has been evaluated as metal gate material, by using combined material and electrical characterizations. Ti–N films elaborated at 180 C are nano crystallized and show no significant preferential crystalline orientation. A TiN phase – NaCl structure type – is identified. The high oxygen level detected in the layer is attributed to low density of the deposited material and its sensibility to oxygen absorption during air exposure. In the TiN films deposited on SiO2, a titanium rich layer has been detected at the interface. The existence of this layer may explain the low work function of TiN, 4.2 eV close to the Ti work function. This work function is stable with high temperature annealing. The work function measurements carried out on HfO2 are sensitively higher. The effective work function on HfO2 is clearly dependent on annealing temperature. This dependence may be attributed to formation of extrinsic states, which pin the Fermi level of metal gate at the metal oxide interface.
253
Acknowledgements This work has been carried out in the frame of the nanoCMOS European project. The authors gratefully acknowledge Y. Morand for stimulating discussions, P. Holliger, G. Rolland, F. Pierre, V. Vidal and Y. Campidelli for characterizations, and Applied Materials for financial support.
References [1] ITRS, International Technology Roadmap for Semiconductors. Available from: