Work Function Investigation in Advanced
Metal Gate-HfO2-SiO2 Systems with Bevel Structures
Atsushi Kuriyama1,3,4, Olivier Faynot, Laurent Brévard,
Amélie Tozzo, Laurence Clerc and Simon Deleonibus
1
CEA-LETI
17 rue des martyrs, 38054 Grenoble, France
atsushi.kuriyama@cea.fr, olivier.faynot@cea.fr
Jérôme Mitard1,2 and Vincent Vidal1,2
2
STMicroelectronics, Crolles 2
850 rue Jean Monnet, 38926 Crolles, France
Abstract— This paper presents for the first time the extraction
of chemical vapor deposition (CVD)-TiN and poly-Si work
functions on atomic layer deposition (ALD)-HfO2 and high
temperature SiO2 (HTO) for a wide range of EOT values. The
measurements were performed on bevel oxide structures. Our
results reveal that the work functions of both TiN and poly-Si
gates highly depend on the underlying dielectrics especially in
the case of TiN on HTO films likewise lower leakage currents
depending on dielectric stacks. It is notable that when TiN is
formed on the HTO film, its work function has two distinct
values depending on the HTO thickness; this indicates that Ti-
Si bonds strongly affect the work function variation. Both TiN
on HTO and poly-Si on HfO2 show the work function shifts to
about 4.3 eV, suggesting a pinning level on both structures.
I. INTRODUCTION
Metal gate technology as a replacement of polysilicon
gate has grown to be requisite for further device performance
as the dimensions of MOS transistors are scaled down.
However, integrating a metal gate material into the
conventional MOS transistor process involves several severe
difficulties. One of them is an undesirable instability of the
metal work function. As reported by several researches, this
phenomenon is attributed to the thermal instability followed
by post metallization annealing [1], metal deposition method
[2], metal orientation [3] or the Fermi pinning effect caused
by metal-induced gap state (MIGS) or chemical bonding [4-6,
10-12]. Concerning the Fermi level pinning effect, one can
find several contradictions among the articles, e.g., huge
differences between vacuum and effective work functions on
several metal gates on high-k dielectrics. By contrast, metal
gates on SiO2 showed consistent values. Furthermore it was
also reported that both n+- and p+-poly-Si gate showed stable
work function values on various dielectrics while metal gates
showed the strong dependence of the work functions on
1-4244-0301-4/06/$20.00 ©2006 IEEE.
Sorin Cristoloveanu
3
IMEP (UMR CNRS-INPG-UJF), ENSERG
BP257, 38016 Cedex 1, Grenoble, France
sorin@enserg.fr
Hiroshi Iwai
4
Frontier Collaborative Research Center
Tokyo Institute of Technology
4259 Nagatsuda, Midori-ku, Yokohama, Japan
several high-k dielectrics [4]. On the other hand, the
stronger Fermi level pinning effect on SiO2 than that on
HfO2 was presented in another article [5]. Moreover, the
strong Fermi level pinning effect on poly-Si, especially p+-
poly-Si on HfO2 was reported [6]. These reports indicate
that the gate work function could strongly depend on
extrinsic factors or process conditions over the intrinsic
Fermi pinning effect [4], resulting in unsettled work function
values. In addition, it is of importance to determine the work
functions from a wide range of EOT-flat band voltage (VFB)
plots for their accurate extractions; nevertheless in many
cases the work functions have been extracted from a few
plots and a narrow range of EOT. In this paper, we present
reliable TiN and poly-Si work functions determined from
wide and continuous EOT-VFB plots by using bevel
structures. We describe how the work function varies due to
a transition of the gate/dielectric interface. Our study also
provides a guide to understand where the pinning level is
located depending on the gate/dielectric interfaces.
II. DEVICE FABRICATION
Fig. 1 shows the device process flow for this study.
MOS capacitors were fabricated on 200 mm p-type silicon
substrates with damascene structures which facilitate a
patterning process for metal gates. All the substrates had 0.7
nm chemical oxides on the surfaces after surface cleaning.
40 nm HTO films were deposited at 730 °C. The HTO films
were etched down to a thickness range of 0-15 nm with bevel
structures by HF solution. The bevel oxide structure offers
the capacitors with a series of oxide thicknesses on the same
wafer, hence the reliable and precise extraction of work
function from numerous EOT-VFB plots. 3 nm ALD-HfO2
films were deposited after/prior to the HTO films to
demonstrate the work function dependence on gate/dielectric
interfaces. Post deposition anneal was carried out at 600 °C
109
for 15 minutes in N2 ambient. 10 nm CVD-TiN was used as
a metal gate material and deposited at 680°C on the stacked
dielectrics. 350 nm in-situ doped n+-poly-Si was used as a
contact metal. No post metallization anneal was performed
except for a final forming gas anneal at 425 °C for 30
minutes. Poly-Si gated capacitors stacked with the same
dielectrics were also fabricated as controls. Capacitors with
5 nm and 10 nm uniform HTO films on/under 3 nm HfO2
were fabricated instead of the bevel HTO to corroborate the
results of the capacitors with bevel oxide. All the samples
were characterized by C-V and I-V measurement. NCSU
CVC program [7] was used to extract EOT and VFB values.
These results were also verified by our extraction program.
Surface cleaning and chemical oxide formation (0.7 nm)
Bevel HTO (0~15 nm) ALD-HfO2 (3 nm) & anneal
ALD-HfO2 (3 nm) & anneal
CVD-TiN (10 nm)
In-situ doped n+-polysilicon (350 nm)
Forming gas anneal at 425 oC for 30 minutes
Figure 1. Device fabrication flow for 4 different gate/dielectric interfaces.
A. C-V Characteristics
Through the whole C-V measurement, no hysteresis was
observed on all capacitors and likewise no frequency
dependence between 1 kHz and 100 kHz was observed on all
capacitors shown in Fig. 2 and Fig. 3. Fig. 2 shows the C-V
characteristics of (a) TiN and (b) n+-poly-Si gated HfO2/SiO2
or SiO2/HfO2 stack capacitors with 5 nm HTO as SiO2 layers.
TiN gated capacitors show a VFB difference of 0.48 V
between two different dielectrics (Fig. 2(a)), whereas the VFB
difference in case of poly-Si gate is only 0.21 V (Fig. 2(b)).
Fig. 3 shows the comparison between TiN and poly-Si gated
capacitors with (a) gate/HfO2 stacks and (b) gate/SiO2 stacks
with 5 nm HTO layers. Although TiN and n+-poly-Si on
HfO2/SiO2 stacks exhibit a VFB difference of 0.51 V (Fig.
3(a)), these two gate materials on SiO2/HfO2 stacks show a
VFB difference of only 0.20 V (Fig. 3(b)). These C-V
characteristics suggest that the work function difference
Bevel HTO (0~15 nm) depends on gate/dielectric interfaces. The EOT difference
shown in Fig. 2 and Fig. 3 is discussed in another column.
0
WFTiN/HfO2 = 4.83 eV
-0.2 BevelOxide
HTO=5nm
HTO=10nm
-0.4
[V]

III. DEVICE CHARACTERISTICS AND DISCUSSION 10

13
Oxide charge density [1/cm ]
FB

As described above, 4 metal/dielectric interfaces, i.e., -0.6

BevelOxide
V

HTO=5nm

TiN/HfO2, TiN/SiO2, n+-poly-Si/HfO2 and n+-poly-Si/SiO2 HTO=10nm

were prepared for evaluating their work functions. Here, we

12
10

present the electrical characteristics of these devices. -0.8

1 1
TiN/HfO2/SiO2/Si Poly-Si/HfO2/SiO2/Si 11
10
Capacitance (µF/cm )
Capacitance (µF/cm )

TiN/SiO2/HfO2/Si Poly-Si/SiO2/HfO2/Si
2

0 2 4 6 8 10 12
2

0.8 HTO: 5 nm
0.8 HTO: 5 nm -1 EOT [nm]

TiN/HfO2 0 2 4 6 8 10 12
0.6 ∆VFB = 0.48 V 0.6 ∆VFB = 0.21 V
Poly-Si/HfO2 EOT [nm]
VFB=-0.43 V
0.4 0.4 VFB=-0.94 V Figure 4. EOT-VFB plot of TiN/HfO2/SiO2/Si stacks with bevel oxide. The
TiN/SiO2 Poly-Si/SiO2
capacitors with fixed SiO2 thickness of 5 nm and 10 nm instead of the bevel
0.2
oxide are also plotted. The inset shows EOT-oxide charge density plot.
0.2
VFB=-0.91 V VFB=-1.15 V
-0.2 10
13
Oxide charge density [1/cm ]
2

0 0
-2 -1 0 1 2 -2 -1 0 1 2 WFTiN/silicate = 4.54 eV
Voltage (V) Voltage (V)
-0.4
(a) (b) 10
12

Figure 2. C-V characteristics of HfO2/SiO2 and SiO2/HfO2 stack with (a)

CVD-TiN gate and (b) n+-polysilicon gate. -0.6 BevelOxide
[V]

HTO=5nm
1 1 HTO=10nm
TiN/HfO2/SiO2/Si TiN/SiO2/HfO2/Si 10
11
Capacitance (µF/cm )

Capacitance (µF/cm )

Poly-Si/HfO2/SiO2/Si Poly-Si/SiO2/HfO2/Si 0 2 4 6 8 10 12 14
2

FB

0.8 0.8 EOT [nm]

TiN/HfO2
HTO: 5 nm HTO: 5 nm -0.8
V

∆VFB = 0.51 V ∆VFB = 0.20 V

0.6 0.6 WFTiN/SiO2 = 4.36 eV
VFB=-0.43 V TiN/SiO2

0.4 0.4 -1 BevelOxide

Poly-Si/HfO2 HTO=5nm
Poly-Si/SiO2
0.2 0.2 VFB=-0.95 V HTO=10nm
VFB=-0.94 V VFB=-1.15 V -1.2
0
-2 -1 0 1 2
0
-2 -1 0 1 2
0 2 4 6 8 10 12 14
Voltage (V) Voltage (V) EOT [nm]
(a) (b) Figure 5. EOT-VFB plot of TiN/SiO2/HfO2/Si stacks with bevel oxide. The
Figure 3. C-V characteristics of CVD-TiN gate and n+-polysilicon gate with capacitors with fixed SiO2 thickness of 5 nm and 10 nm instead of the bevel
(a) HfO2/SiO2 stack and (b) SiO2/HfO2 stack. oxide are also plotted. The inset shows EOT-oxide charge density plot.

110
 -0.6 that the TiN work function highly depends on the dielectric
WFpoly-Si/HfO2 = 4.28 eV component at the interface, especially a presence of Si atoms.
BevelOxide
HTO=5nm
The equivalent oxide charge density shown in the inset of
-0.8 Fig. 5 depicts no abrupt change indicating a stable HfO2/Si
HTO=10nm
interface and also corroborating the explanation of the
rollover in Fig. 4. Fig. 6 shows EOT-VFB plots of n+-poly-
-1
[V]

10
12 Si/HfO2/SiO2/Si stacks with bevel oxides. The extracted
Oxide charge density [1/cm ]

work function is found to be 4.28 eV which is a higher value

2
FB

-1.2 than that of n+-poly-Si/SiO2/HfO2/Si (as low as 4.03 eV, not

V

10
11 shown). The rollover seen in Fig. 6 could also be explained
by the HfO2 film on the Si substrate through the chemical
-1.4 BevelOxide
HTO=5nm
oxide by intermixing, which leads to an increase in the
10
10
HTO=10nm
charge density as shown in the inset of Fig. 6.
0 2 4 6 8 10 12 14
-1.6 EOT [nm]

0 2 4 6 8 10 12 14 C. Equivalent Oxide Thickness (EOT)

EOT [nm] Table I and Table II report all the EOT values obtained
Figure 6. EOT-VFB plot of n+-poly-Si/HfO2/SiO2/Si stacks with bevel oxide. from the capacitors with HTO films of 5 nm and 10 nm.
The capacitors with fixed SiO2 thickness of 5 nm and 10 nm instead of the When comparing the EOT dependence on dielectric stack
bevel oxide are also plotted. The inset shows EOT-oxide charge density plot. structures with the same gate material, the structures of
gate/SiO2/HfO2/Si show larger EOT values than those of
B. Work Function gate/HfO2/SiO2/Si on both TiN and poly-Si as shown in
Fig. 4 shows the EOT-VFB plots of TiN/HfO2/SiO2/Si Table I, due to a growth of interfacial layers at HfO2/Si
stacks with bevel structures. The plots of the capacitors with interfaces, whereas chemical oxides as thick as 0.7 nm were
5 nm and 10 nm HTO are also given for comparison. This formed on Si substrates. Table II shows the comparison
figure clearly depicts a linear relation between EOT and VFB between TiN gate and poly-Si gate on the same dielectric
down to 2 nm, from which the work function of 4.83 eV is stack. The higher EOT values of 0.9 nm were found on
extrapolated. The plots of stacks with fixed HTO thickness poly-Si due to the poly-Si depletion effect. On the other
are on the extrapolated line without any inconsistency, so hand, higher EOT values as much as 1.2 nm were also
validating the extraction of the work function from the bevel measured between TiN gate and poly-Si gate on SiO2/HfO2
oxide structure. A rollover in VFB is observed when the EOT stacks. This result indicates the consumption of SiO2 layers
is less than 2 nm. Two possibilities are conceivable: one is by metal gates, confirming the results reported in [9].
that the HTO was totally etched off; as a result, the HfO2
formed on the HTO came on the Si substrate through 0.7 nm TABLE I. COMPARISON OF GATE DIELECTRIC STACKS UNDER
chemical oxide by intermixing and created an extra interface THE SAME GATE MATERIAL.
charge. The other possibility is caused by nitrogen diffusion
EOT EOT gap
onto the Si surface through HfO2 during TiN formation [8]. Gate/Dielectric1 Dielectric2/Si
[nm] [nm]
An abrupt increase in the equivalent oxide charge density for TiN/HfO2 SiO2 (5 nm)/Si 5.5
EOT values less than 2 nm extracted by [7], shown in the TiN/SiO2 (5 nm) HfO2/Si 6.9
1.4
inset of Fig. 4, supports these explanations. Fig. 5 shows the TiN/HfO2 SiO2 (10 nm)/Si 10.1
EOT-VFB plots of TiN/SiO2/HfO2/Si stacks with bevel oxides. 1.1
TiN/SiO2 (10 nm) HfO2/Si 11.2
In contrast to TiN/HfO2 case, the plots do not show a
Poly-Si/HfO2 SiO2 (5 nm)/Si 6.4
proportional relation throughout the EOT range. In a range 1.7
Poly-Si/SiO2 (5 nm) HfO2/Si 8.1
of EOT between 7-12 nm, the work function is found to be
4.36 eV and from a range between 5-7 nm, VFB changes Poly-Si/HfO2 SiO2 (10 nm)/Si 11.0
1.4
abruptly resulting in the plural extraction of work function. Poly-Si/SiO2 (10 nm) HfO2/Si 12.4
In a range between 2.5-5 nm, the work function is found to
be 4.54 eV. Moreover, in a range less than 2.5 nm, VFB TABLE II. COMPARISON OF METAL MATERIALS ON THE SAME GATE
DIELECTRIC STACK.
again changes abruptly and finally reaches almost the same
value as that of TiN/HfO2 case (cf. Fig. 4). One can find, as Gate/Dielectric1 Dielectric2/Si
EOT EOT gap
a result, three different characteristics due to two abrupt [nm] [nm]
changes in VFB. Based on these results, we speculate that TiN/HfO2 SiO2 (5 nm)/Si 5.5
0.9
there exist three interfaces in this structure: the first is the Poly-Si/HfO2 SiO2 (5 nm)/Si 6.4
TiN/SiO2 interface in a large EOT range (EOT=7-12 nm), TiN/HfO2 SiO2 (10 nm)/Si 10.1
0.9
hence no Hf atom is present at the interface; the second is the Poly-Si/HfO2 SiO2 (10 nm)/Si 11.0
TiN/Hf-silicate interface (EOT=2.5-5 nm) where both Hf TiN/SiO2 (5 nm) HfO2/Si 6.9
1.2
and Si atoms are present at the interface by intermixing; the Poly-Si/SiO2 (5 nm) HfO2/Si 8.1
third is the TiN/HfO2 interface where no Si atom is present TiN/SiO2 (10 nm) HfO2/Si 11.2
1.2
due to complete HTO etching. These considerations suggest Poly-Si/SiO2 (10 nm) HfO2/Si 12.4

111
D. EOT-J Characteristics resulting in the work function pinned at the more positively
Fig. 7 shows EOT-leakage current (Vg=VFB-1 [V]) plots biased side. Thus, Fermi pinning with the bonding theory
of HfO2/SiO2 and SiO2/HfO2 stacks with (a) TiN and (b) n+- explained in [11] is inconsistent with our result. The cause
poly-Si. Gate/SiO2/HfO2 stacks show leakage currents more for this inconsistency is now under investigation.
5
than one order of magnitude lower, indicating a growth of TiN/HfO2/SiO2/Si
TiN/SiO2(thin)/HfO2/Si
interfacial oxide. We note a slight difference in thermal TiN/SiO2(thick)/HfO2/Si

Work Function [eV]

4.8 n+-poly-Si/HfO2/SiO2/Si
budget due to the HTO deposition temperature higher than n+-poly-Si/SiO2/HfO2/Si
the annealing temperature for HfO2 films. Fig. 8 shows EOT- 4.6
leakage current (Vg=VFB-1 [V]) plots on TiN and n+-poly-Si Poly-Si gated

gate with (a) HfO2/SiO2 and (b) SiO2/HfO2 stacks. The 4.4
slight increase in leakage current observed on poly-Si gated
stacks indicates the thinner physical dielectric thickness for 4.2
TiN gated
the same EOT value caused by the poly-Si depletion effect.
10
-3 -3
10 4 Ti TiN n +- n +-
Leakage Current [A/cm ] @V - 1 V

TiN pol
Leakage Current [A/cm ] @V - 1 V

TiN/HfO2/SiO2/Si Poly-Si/HfO2/SiO2/Si / H f N /S iO /Si

O
pol
y-S y-S
O 2 (th 2 (th i/H i/Si
TiN/SiO2/HfO2/Si Poly-Si/SiO2/HfO2/Si 2 in ) ick fO O
) 2
FB

2
FB

-4 -4
10 10 Figure 9. Transition of work function with various metal/dielectric interfaces.
-5 -5
10 10
2

10
-6 ∆IG
(a) -6
10
(b) IV. CONCLUSION
∆IG The work functions of CVD-TiN and n+-poly-Si on
10
-7 -7
10 different gate dielectrics were investigated with the bevel
-8 -8
oxide structures. The reliable work function extraction was
10
0 2 4 6 8 10
10
0 2 4 6 8 10 demonstrated from a wide range of continuous EOT values
EOT[nm] EOT[nm]
enabled by the bevel oxides as well as from an excellent
Figure 7. EOT-J plot of (a) TiN/HfO2/SiO2/Si and TiN/SiO2/HfO2/Si, (b) oxide quality without any hysteresis. The TiN/SiO2/HfO2/Si
polysilicon/HfO2/SiO2/Si and polysilicon/SiO2/HfO2/Si with bevel oxide. stacks showed leakage currents one order of magnitude
-3 -3
10 10 lower for the same EOT value. The work functions of both
Leakage Current [A/cm ] @V - 1 V

Leakage Current [A/cm ] @V - 1 V

TiN/HfO2/SiO2/Si TiN/SiO2/HfO2/Si
Poly-Si/HfO2/SiO2/Si Poly-Si/SiO3/HfO2/Si
TiN and n+-poly-Si were found to depend on the underlying
FB

FB

-4 -4
10 10
dielectrics, especially TiN gate showed a larger shift than the
10
-5
10
-5 n+-poly-Si gate. TiN on HfO2 exhibited a work function
2

-6
(a) -6
(b) value of 4.83 eV, while TiN on SiO2/HfO2 showed two work
10 10 function values of 4.36 eV and 4.54 eV depending on the
10
-7
10
-7 SiO2 thickness. The poly-Si on HfO2 showed a work
function value of 4.28 eV which is consistent with several
10
-8
0 2 4 6 8 10
10
-8
0 2 4 6 8 10
reports. Both TiN/SiO2 and poly-Si/HfO2 stacks showed the
EOT[nm] EOT[nm] work function shifts toward about 4.3 eV due to the Fermi
Figure 8. EOT-J plot of (a) TiN/HfO2/SiO2/Si & polysilicon/HfO2/SiO2/Si, pinning effect most likely caused by Ti-Si and Si-Hf bonds.
(b) TiN/SiO2/HfO2/Si & polysilicon/SiO2/HfO2/Si with bevel oxide.

E. Discussion ACKNOWLEDGMENT
Fig. 9 summarizes the work functions extracted from this This work has been partly founded by the Crolles 2 Alliance
study. The work function of n+-poly-Si/HfO2 structure (STMicroelectronics, Philips, Freescale)-LETI joint program.
shifting to about 4.3 eV due to Hf-Si bonds is consistent with The authors thank LETI facilities for device processing.
other experimental and modeling studies [6, 10]. It should
be noted that the TiN/SiO2/HfO2 stack shows two work
REFERENCES
functions of 4.36 eV and 4.54 eV depending on the SiO2
thickness, while TiN/HfO2/SiO2 stack shows 4.83 eV. Here, [1] H.Y. Yu et al., IEEE Elec. Dev. Lett., Vol. 25, pp. 337-339, 2004.
we argue four facts: the first is that the Fermi level pinning [2] G. Lujan et al., Proc. ESSDERC, pp. 583-586, 2002.
effect occurs on both TiN and poly-Si gate depending on the [3] G-D. Park et al., Appl. Phys. Lett., Vol. 80, pp. 2514-2516, 2002.
metal/dielectric interfaces [11]; the second is that the Fermi [4] Y-C. Yeo, T-J. King and C. Hu, J. Appl. Phys. Vol. 92, pp.7266-7271, 2002.
level pinning effect on the n+-poly-Si is rather weak [5] C. Ren et al., IEEE Elec. Dev. Lett., Vol. 25, pp. 123-125, 2004.
compared to that on TiN [10]; the third is that the work [6] C.C. Hobbs et al., IEEE Trans. Elec. Dev., Vol. 51, pp. 971-977, 2004.
functions of both TiN/SiO2/HfO2 and n+-poly-Si/HfO2/SiO2 [7] J.R. Hauser and K. Ahmed, AIP Conf. Proc., pp. 235-239, 1998.
stacks converge to a value of about 4.3 eV; the fourth, based [8] J-H. Lee et al., IEDM Tech. Dig. pp. 323-326, 2003.
on the work function dependence on SiO2 thickness (Fig. 5), [9] T. Ushiki et al., IEEE Trans. Elec. Dev. Vol. 47, pp.2201-2207, 2000.
is that the TiN work function is likely to depend on the [10] K. Xiong et al., Appl. Phys. Lett. Vol. 86,012904, 2005.
amount of Si atoms at TiN/SiO2 interface [12]. However, [11] M.S. Joo et al., IEEE Elec. Dev. Lett., Vol. 25, pp. 716-718, 2004.
assuming extrinsic states between TiN and SiO2, Ti-Si bonds [12] M.S. Joo et al., IEEE Elec. Dev. Lett. Vol. 26, pp. 882-884, 2005.
should produce a negative dipole charge at the dielectric side,

112