PHYSICAL REVIEW B 78, 012102 共2008兲

First-principles study on doping and phase stability of HfO2

Choong-Ki Lee,1 Eunae Cho,1 Hyo-Sug Lee,2 Cheol Seong Hwang,3 and Seungwu Han1,*
1Department
of Physics, Ewha Womans University, Seoul 120-750, Korea
2Samsung
Advanced Institute of Technology, Suwon 400-600, Korea
3Department of Materials Science and Engineering and Inter-University Semiconductor Research Center, Seoul National University,

Seoul 151-742, Korea

共Received 11 March 2008; published 3 July 2008兲
Based on density functional methods, relative stabilities between monoclinic, tetragonal, and cubic phases of
HfO2 with cation dopants or oxygen vacancies are investigated. It is found that dopants such as Si, Ge, Sn, P,
Al or Ti with ionic radii smaller than Hf stabilize the tetragonal phase but destabilize the cubic phase. In
contrast, larger dopants such as Y, Gd or Sc favor the cubic phase. The ionized oxygen vacancies compensating
trivalent dopants greatly stabilize both cubic and tetragonal phases. Microscopic explanations on the results are
also given. The metastable phase favored by each dopant is in good agreement with experimental data. Our
results can serve as a useful guide in selecting dopants to stabilize a specific phase.

DOI: 10.1103/PhysRevB.78.012102 PACS number共s兲: [Link], [Link], [Link]

The continuous downscaling of complementary metal- similar to the case of ZrO2 共Ref. 14兲 in some parts. In addi-
oxide semiconductor 共CMOS兲 devices with performance en- tion, theoretical studies are very rare.15
hancement has been enabled by reducing the thickness of In this Brief Report, we theoretically investigate the effect
gate insulators. As the thickness of SiO2, the traditional gate of dopants on the relative stability between m-, t- and
dielectric, is reduced to a few nanometers, leakage currents c-HfO2. We employ first-principles methods based on the
due to the quantum tunneling have increased greatly. The density functional theory. For the computation of total ener-
introduction of an insulator with a dielectric constant 共k兲 gies and structural optimizations, we use VASP.16 Supercells
higher than for SiO2 can solve the leakage problem as it including 96 atomic sites are used for all structures. These
allows for increasing the physical thickness of the gate insu- supercells are obtained by expanding structures in Fig. 1
lator. Among various oxides explored to date, hafnia 共HfO2兲 twice along each axis. The energy cutoff for the plane-wave
and its family materials are considered to be the most prom- basis is chosen to be 500 eV and k-points are sampled on
ising as a high-k gate oxide since they satisfy various tech- 2 ⫻ 2 ⫻ 2 uniform grids. Gaussian broadening is used with a
nical requirements. width of 0.05 eV to smear the density of states. The
There are several polymorphs in HfO2 such as monoclinic exchange-correlation interactions between electrons are de-
共m兲, tetragonal 共t兲 or cubic 共c兲 phases 共see Fig. 1兲. The ortho- scribed by the generalized gradient approximation 共GGA兲
rhombic phase is also observed at high pressure conditions.1 共Ref. 17兲 and projector-augmented wave 共PAW兲 potentials
While m-HfO2 is stable at ambient conditions, the material are used for the description of ion-electron interactions.18
undergoes a phase transition to the tetragonal phase at 2000 The atomic positions and cell parameters are relaxed until
K and to the cubic phase at 2900 K. The calculated dielectric the atomic forces and stress tensors are reduced to within
constants vary widely from phase to phase, and the average 0.02 eV/Å and 1 kbar, respectively. The optimized structural
dielectric constants are much higher for cubic and tetragonal parameters are given in Table I, and the agreements with
phases.2,3 Thin films of HfO2 共typical thickness of ⬃5 nm兲 other work and experimental data are within 2%.
deposited in CMOS devices are usually amorphous due to As a dopant, we choose Si, P, Ge, Al, Y, Ti, Zr, Gd, and
the low growth temperature 共⬃300 ° C兲 and they crystallize Sc, and replace one of Hf atoms with the dopant. 共All Hf
into the monoclinic phase upon annealing at temperatures sites are equivalent.兲 This corresponds to a doping concen-
above 700 ° C. The two structures are similar in local bond- tration of 3.125%. In addition, we also consider oxygen va-
ing configurations and have dielectric constants significantly cancies since they are introduced to compensate trivalent
lower than those of cubic or tetragonal phases.4 This suggests dopants 共such as Al, Y, Gd or Sc兲. As the vacancy site loses
that the dielectric constant of the HfO2 thin film can be in- electrons in this case, the doubly ionized vacancy 共VO 2+
兲, as
creased without changing materials if the relative stability well as the neutral one 共VO兲, is studied.
0
among polymorphs can be controlled in a way to make a
high-temperature phase to be most stable at growth condi- (a) (b) (c)
tions. Recently, a series of experiments have demonstrated
that such a phase control can be achieved by cation doping.
For example, the doping with Si,5 Al 共Refs. 6 and 7兲 or Zr
共Ref. 8兲 atoms leads to thin films with a significant portion of
t-HfO2. It was also observed that c-HfO2 is stabilized by
using Y,9–11 Gd,12 Sc 共Ref. 13兲 or Dy 共Ref. 13兲 dopants. Hf O
Despite the accumulating data on the phase control of HfO2
based on the doping method, the microscopic origin has not FIG. 1. 共Color online兲 The structure of HfO2 in 共a兲 monoclinic,
been elaborated much although the mechanism would be 共b兲 cubic, and 共c兲 tetragonal phases.

1098-0121/2008/78共1兲/012102共4兲 012102-1 ©2008 The American Physical Society

BRIEF REPORTS PHYSICAL REVIEW B 78, 012102 共2008兲

TABLE I. The structural parameters of c-, t-, and m-HfO2. Only monoclinic tetragonal cubic
nontrivial parameters are shown. (a)

Phase This work Other theorya Experimentb

Monoclinic a = 5.12 Å, a = 5.22 Å, a = 5.12 Å,

b = 5.20 Å, b = 5.29 Å, b = 5.17 Å,
c = 5.28 Å, c = 5.35 Å, c = 5.30 Å, (b)
␤ = 99.7° ␤ = 99.7° ␤ = 99.2°
Cubic a = 5.05 Å a = 5.15 Å a = 5.08 Å
Tetragonal a = 5.06 Å, a = 5.17 Å, a = 5.15 Å,
c = 5.20 Å c = 5.22 Å c = 5.29 Å
aReference 1. Hf O
b
References 19–21.
FIG. 2. 共Color online兲 Relaxation patterns around 共a兲 Si dopants
The main computational results are presented in Table II. or 共b兲 ionized oxygen vacancies. The dopant and vacancy sites are
The effective ionic radii22 are also shown for the purpose of denoted by arrows.
analysis. The relative energies 共⌬E⬘s兲 of t- and c-HfO2 with
respect to that of m-HfO2 are calculated per supercell. The site and the second-nearest O atoms relax outward 关see Fig.
energies of crystalline phases without defects or dopants fol- 2共a兲兴. The resulting local structure is reminiscent of SiO2
low the order of m ⬍ t ⬍ c, in agreement with a previous where four O atoms form a tetrahedron with the Si atom at
the center. Furthermore, the relaxed Si-O bond length of
calculation.1 From Table II, one can see that there are two
1.69 Å and O-Si-O bond angles of 105.4° or 117.9° are
types of dopants. The “type-I” dopant represented by Si, Ge,
close to 1.63 Å and 108.9° – 110.5° in the quartz structure of
Sn, P, Al, and Ti decreases ⌬Etetra but increases ⌬Ecubic. In SiO2, which were obtained with the same computational
“type-II” dopants such as Y, Gd or Sc, ⌬Ecubic is reduced setup. Consequently, the local environment around the Si
further than ⌬Etetra although overall magnitudes of the atom is similar to that in the quartz, explaining why the Si
changes are smaller than for type-I dopants. This means that dopant strongly favors t-HfO2. In c-HfO2, eight O atoms
type-II dopants tend to stabilize the cubic phase. surrounding the Si dopant relax inward symmetrically and
First, we explain the pronounced effect of Si doping.15 the resulting Si-O bond length is reduced from 2.19 Å to
The coordination number of the cation in t-HfO2 is eight, the 2.08 Å. However, this is much longer than the ideal Si-O
same as in c-HfO2, and the Hf-O bond lengths are equally bonding distance. To examine the possibility of a spontane-
distributed at 2.07 and 2.37 Å. As a result of structural re- ous symmetry lowering in the Si-doped c-HfO2, we initially
laxations, four O atoms nearest to Si are pulled to the dopant displace four O atoms in the tetrahedral configuration toward
TABLE II. The relative energies of tetragonal 共⌬Etetra兲 and cu- the dopant and perform the structural relaxation. It is found
bic phases 共⌬Ecubic兲 with respect to that of the monoclinic phase. that the structure eventually transforms to m-HfO2 indicating
The energies in parentheses show the difference from the value for that c-HfO2 becomes unstable upon Si doping. In the crys-
the perfect lattice. The ionic radii are also shown with coordination talline m-HfO2, on the other hand, the coordination number
numbers indicated within parentheses. The ionic radii for the Hf of the cation is seven with bond lengths ranging over
atom are 0.76 and 0.83 Å for seven- and eightfold coordination, 2.02– 2.19 Å. The relaxed structure with the Si dopant
respectively. shows that five Si-O bonds are shortened by ⬃0.3 Å while
two Si-O bonds are elongated by 0.2– 0.7 Å. The reduction
Dopant Ionic radius ⌬Etetra ⌬Ecubic of the coordination number is similar to t-HfO2, but bond
共Å兲 共eV/supercell兲 共eV/supercell兲 lengths and angles are less ideal.
The above discussion also applies to other type-I dopants.
Perfect 5.47 8.09 As can be seen in Table II, the ionic radii of all type-I dop-
Si 0.26共4兲 3.90共−1.57兲 10.38共+2.29兲 ants are significantly smaller than those of Hf atoms
Ge 0.39共4兲 4.55共−0.92兲 9.21共+1.12兲 共0.76– 0.83 Å兲. Therefore, type-I dopants tend to shorten the
Sn 0.55共4兲 5.31共−0.16兲 8.29共+0.20兲 dopant-oxygen bonds. In the tetragonal phase, four Hf-O
P 0.17共4兲 5.00共−0.47兲 10.90共+2.81兲 bonds are already shorter than the others, thereby facilitating
Al 0.39共4兲 5.01共−0.46兲 8.77共+0.68兲 the formation of stable dopant-oxygen bonds with relatively
small lattice distortions. On the other hand, the ionic radii of
Ti 0.42共4兲 5.32共−0.15兲 8.59共+0.50兲
type-II dopants are larger than those of Hf atoms, which
Y 0.96共7兲 5.40共−0.07兲 7.54共−0.55兲 requires the elongation of the bonds with nearby O atoms.
Gd 1.00共7兲 5.24共−0.23兲 7.38共−0.71兲 Compared to the cubic phase, monoclinic and tetragonal
Sc 0.87共8兲 5.68共+0.21兲 7.95共−0.14兲 phases contain several short Hf-O bonds with a length be-
Zr 共50%兲 0.84共8兲 4.74共−0.73兲 6.91共−1.18兲 tween 2.0 and 2.1 Å. Therefore, the strain energy arising
VO0
共fourfold site兲 5.16共−0.31兲 7.75共−0.34兲 from the size mismatch should be more significant in m- and
VO2+
共threefold site兲 3.30共−2.17兲 4.42共−3.67兲 t-HfO2 and, hence, the cubic phase is favored by the over-
sized type-II dopants.

012102-2
BRIEF REPORTS PHYSICAL REVIEW B 78, 012102 共2008兲

200
TABLE III. The predicted stable phases and the minimum dop-

DOS (states/eV/supercell)
ing concentrations required for a metastable phase to be more stable 150 (a) cubic
monoclinic
than the monoclinic phase. The experimental data are also shown tetragonal

for comparison. 100

50
Dopant Stable phase Minimum doping Experimenta
concentration 共%兲 0
-4 -2 0 2 4
Energy (eV)
Si Tetragonal 10.9 Tetragonal, 5%
200

DOS (states/eV/supercell)
Ge Tetragonal 18.6 N/A
Al Tetragonal 11.1 Tetragonal 150
(b) cubic
monoclinic
tetragonal
Y Cubic 10.6 Cubic, 4%–6.5% 100
Sc Cubic 12.8 Cubic, 10%
50
Gd Cubic 9.9 Cubic, 10%–15%
a 0
References 5–7 and 9–13. -4 -2 0 2 4
Energy (eV)

Next, we discuss the effects of oxygen vacancies. It is

200

DOS (states/eV/supercell)
well established that Y atoms in ZrO2 introduce oxygen va-
cancies to compensate their charge states.14 A similar process 150
(c) cubic
monoclinic
tetragonal
would occur in HfO2: Being trivalent, two Y atoms substi-
100
tuting Hf sites results in a deficiency of two electrons, and
this is compensated by the introduction of an oxygen va- 50

cancy. In m-HfO2, there are two types of oxygen, threefold 0

or fourfold, while all oxygen atoms are equivalent in t- and -4 -2 0 2 4
Energy (eV)
c-HfO2. It is known that the neutral oxygen vacancy in
m-HfO2 is more stable at the fourfold site than at the three- FIG. 3. 共Color online兲 The DOS of c-, t-, and m-HfO2 in the
fold one, and vice versa for the charged vacancy.23 As shown presence of 共a兲 Si, 共b兲 Ti, and 共c兲 Y dopants. The Fermi level is set
in Table II, the neutral oxygen vacancy 共VO 0
兲 only slightly to zero.
affects the relative stability. However, when the vacancy is
doubly ionized, ⌬Etetra and ⌬Ecubic are significantly reduced.
be 4.02 eV and 5.90 eV, respectively, the alloying with Zr
This can be explained in terms of the coordination number of
atoms lowers ⌬Etetra and ⌬Ecubic compared to pure HfO2. In
Hf atoms; Hf atoms favor a sevenfold coordination as in
fact, by exploring Hf1−xZrxO2 sold solutions over a wide
m-HfO2. The charged oxygen vacancy in t- and c-HfO2 leads
range of x, it is found that the following linear relation holds
to the sevenfold coordination of Hf atoms around the defect
to a good precision:
site, and this is the primary reason for the stabilization of
these phases. 共When the vacancy is neutral, the coordination ⌬E共Hf1−xZrxO2兲 = 共1 − x兲⌬E共HfO2兲 + x⌬E共ZrO2兲. 共1兲
number is effectively unchanged because defect states are
occupied.兲 In addition to this, we note that the structural This implies that the Zr doping does not change the order of
relaxation plays a role. As shown in Fig. 2共b兲, large atomic relative stability at any doping concentration. In polycrystal-
relaxations are noticeable around charged vacancies com- line thin films, the surface energy may help stabilize the
pared to neutral ones 共not shown兲. Due to the effectively tetragonal phase.8,24
positive charge 共+2兲 centered at the vacancy site, the anions Based on the above results, we theoretically guess which
relax toward the vacancy, while cations are repelled. This metastable phase becomes more stable than the monoclinic
essentially accounts for the dielectric screening of charged phase at a certain minimum doping concentration. As a first
defects by the lattice. From the inspection of the bonding approximation, we neglect any interaction between defects
distribution, it is found that the inward motion of oxygen and/or dopants, and ⌬E at an arbitrary concentration is esti-
atoms is noticeably suppressed in m-HfO2, which is attrib- mated by a linear extrapolation of the results in Table II.
uted to relatively longer O-O bonding distances in this phase. 共However, the undersized dopants can attract the oxygen
As a consequence, t- and c-HfO2 are further stabilized by the vacancy.14兲 It is also assumed that charged defects are com-
charged vacancy, contributing to the reduction of ⌬Etetra and pletely compensated by oxygen vacancies. For instance, if
⌬Ecubic. there are two Y dopants and one oxygen vacancy within the
Like HfO2, zirconia 共ZrO2兲 has three metastable phases of supercell, the doping concentration corresponds to 6.25%
t-, c-, and m-ZrO2. In our calculations, the doping with one and ⌬Ecubic共2Y + VO 2+
兲 = 8.09− 2 ⫻ 0.55− 3.67= 3.32 eV. This
Zr atom in each supercell did not give rise to noticeable assumption is tested by actually introducing the defect set
changes in the relative stability. When Zr atoms randomly into the supercell. The computed ⌬Ecubic is 3.70 eV which
replace 50% of Hf atoms, however, ⌬Etetra and ⌬Ecubic agrees reasonably with the simple estimation above. The re-
change considerably 共see Table II兲. This is closely related to sults for selected dopants are given in Table III and they are
the phase stability among ZrO2 polymorphs. Since the total also compared with experimental data when available. It is
energies of t- and c-ZrO2 relative to m-ZrO2 are calculated to found that the estimations based on the first-principles results

012102-3
BRIEF REPORTS PHYSICAL REVIEW B 78, 012102 共2008兲

agree well with the experiments. However, the minimum 3共c兲兴. This is understandable because the valence electrons of
doping concentrations are larger than the experimental val- the Y atom are close in energy to those of the Hf atom as is
ues in many cases. This might be related to the defect-dopant evidenced in similar energy gaps of Y2O3 and HfO2. From
interactions, which need further investigations. the study on DOS’s for other dopants, we find that DOS’s for
For type-II dopants of Y, Sc, and Gd, one can also define trivalent dopants such as Al, Gd or Sc are similar to that of Y;
minimum doping concentrations, nc1 and nc2, at which the there is no ingap state associated with the dopant and the
cubic phase becomes more stable than the tetragonal phase
共⌬Etetra = ⌬Ecubic兲, and the tetragonal phase becomes lower in Fermi level lies below the valence edge. It is also observed
energy than the monoclinic phase 共⌬Etetra = 0兲, respectively. that DOS’s for tetrahedrally bonded dopants such as Ge, Sn
The values of nc1共nc2兲 are 6.7共14.8兲, 7.4共19.5兲, and or P in c-HfO2 show defect levels within the energy gap,
6.7共13.0兲% for Y, Sc, and Gd, respectively. In the case of Al which is similar to the case of Si. The defect levels are un-
doping, the cubic phase becomes more stable than the mono- occupied for Ge and Sn while they are singly occupied for P.
clinic phase at a doping concentration of 21.9% but is always In summary, we investigated the energetics of HfO2 poly-
unstable against the tetragonal phase. For Si- and Ge-doped morphs in the presence of various dopants and oxygen va-
HfO2, the cubic phase is the highest in energy at any con- cancies. The tetragonal phase was favored by the type-I dop-
centration of the dopant. ants 共Si, Ge, Sn, Ti, P, and Al兲 because ionic radii are smaller
In Fig. 3, the density of states 共DOS兲 is presented for the than for Hf atoms and the formation of short bonds are fa-
representative cases of Si, Ti and Y dopants. For the Si dop- cilitated in t-HfO2. On the other hand, the type-II dopants 共Y,
ant in Fig. 3共a兲, gap states are found only in c-HfO2. The Sc, and Gd兲 with atomic radii larger than that of the Hf atom
inspection of the wave functions indicates that they are lo- prefer c-HfO2. It was also found that the ionized oxygen
calized around the Si atom with s-like symmetry. The corre- vacancy plays a crucial role in the stabilization of metastable
sponding state does not exist for t- and m-HfO2. Therefore, phases. The stable phases estimated by the first-principles
Si dopants in t-HfO2 would not degrade much of leakage results were in good agreement with recent experimental
behaviors since they do not create trap sites. In the case of Ti data. Our results can serve as a useful guide in selecting
doping in Fig. 3共b兲, gap states appear in all phases. These dopants to stabilize a specific phase.
levels are derived from Ti-d orbitals with degeneracy lifted
by the crystal field. For instance, the Ti d-levels in c-HfO2 This work was supported by the System IC2010 program
are split into eg and t2g states 共with eg states lower in energy兲 and the National Program for 0.1 Terabit NVM Devices. The
due to the crystal field with cubic symmetry. On the other computations were carried out at KISTI 共Grant No. KSC-
hand, the Y dopant does not introduce any gap state 关see Fig. 2007-S00-1006兲.

*Corresponding author; hansw@[Link] 12 Y. B. Losovyj, I. Ketsman, A. Sokolov, K. D. Belashchnko, P. A.

1
J. Kang, E.-C. Lee, and K. J. Chang, Phys. Rev. B 68, 054106 Dowben, J. Tang, and Z. Wang, Appl. Phys. Lett. 91, 132908
共2003兲. 共2007兲.
2 X. Zhao and D. Vanderbilt, Phys. Rev. B 65, 233106 共2002兲. 13 C. Adelmann, V. Sriramkumar, V. Van Elshocht, P. Lehnen, T.
3
G.-M. Rignanese, X. Gonze, G. Jun, K. Cho, and A. Pasquarello, Conard, and S. De Gendt, Appl. Phys. Lett. 91, 162902 共2007兲.
Phys. Rev. B 69, 184301 共2004兲. 14 P. Li, I.-W. Chen, and J. E. Penner-Hahn, J. Am. Ceram. Soc. 77,
4 X. Zhao, D. Ceresoli, and D. Vanderbilt, Phys. Rev. B 71,
118 共1994兲.
085107 共2005兲. 15 D. Fischer and A. Kersch, Appl. Phys. Lett. 92, 012908 共2008兲.
5
K. Tomida, K. Kita, and A. Toriumi, Appl. Phys. Lett. 89, 16
G. Kresse and J. Hafner, Phys. Rev. B 47, 558共R兲 共1993兲; 49,
142902 共2006兲. 14251 共1994兲.
6
P. K. Park and S.-W. Kang, Appl. Phys. Lett. 89, 192905 共2006兲. 17
J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77,
7 Y. Yang, W. J. Zhu, T. P. Ma, and S. Stemmer, J. Appl. Phys. 95,
3865 共1996兲.
3772 共2004兲. 18
P. E. Blöchl, Phys. Rev. B 50, 17953 共1994兲.
8 D. H. Triyoso, R. I. Hedge, J. K. Schaeffer, D. Roan, P. J. Tobin, 19 J. Wang, H. P. Li, and R. Stevens, J. Mater. Sci. 27, 5397 共1992兲.
20
S. B. Samavedam, B. E. White, Jr., R. Gregory, and X.-D. Wang, D. M. Adams, S. Leonard, D. R. Russel, and R. J. Cemik, J.
Appl. Phys. Lett. 88, 222901 共2006兲. Phys. Chem. Solids 52, 1181 共1991兲.
9 21
S.-G. Lim, S. Kriventsov, T. N. Jackson, J. H. Haeni, D. G. D. W. Stacy, J. K. Johnstone, and D. R. Wilder, J. Am. Ceram.
Schlom, A. M. Balbashov, R. Uecker, P. Reiche, J. L. Feeouf, Soc. 55, 482 共1972兲.
and G. Lucovsky, J. Appl. Phys. 91, 4500 共2002兲. 22
R. D. Shannon and C. T. Prewitt, Acta Crystallogr., Sect. B:
10
K. Kita, K. Kyuno, and A. Toriumi, Appl. Phys. Lett. 86, Struct. Crystallogr. Cryst. Chem. 25, 925 共1969兲.
102906 共2005兲. 23 A. S. Foster, F. Lopez Gejo, A. L. Shluger, and R. M. Nieminen,
11 E. Rauwel, C. Dubourdieu, B. Holländer, N. Rochat, F. Ducro-
Phys. Rev. B 65, 174117 共2002兲.
24 S. K. Kim and C. S. Hwang, Electrochem. Solid-State Lett. 11,
quet, M. D. Rossell, G. Van Tendeloo, and B. Pelissier, Appl.
Phys. Lett. 89, 012902 共2006兲. G9 共2008兲.

012102-4