You are on page 1of 4

Oxygen vacancies in ZnO

Anderson Janotti, and Chris G. Van de Walle

Citation: Appl. Phys. Lett. 87, 122102 (2005); doi: 10.1063/1.2053360


View online: https://doi.org/10.1063/1.2053360
View Table of Contents: http://aip.scitation.org/toc/apl/87/12
Published by the American Institute of Physics

Articles you may be interested in


A comprehensive review of ZnO materials and devices
Journal of Applied Physics 98, 041301 (2005); 10.1063/1.1992666

Correlation between photoluminescence and oxygen vacancies in ZnO phosphors


Applied Physics Letters 68, 403 (1996); 10.1063/1.116699

Defects in ZnO
Journal of Applied Physics 106, 071101 (2009); 10.1063/1.3216464

Mechanisms behind green photoluminescence in ZnO phosphor powders


Journal of Applied Physics 79, 7983 (1996); 10.1063/1.362349

On the optical band gap of zinc oxide


Journal of Applied Physics 83, 5447 (1998); 10.1063/1.367375

Green luminescent center in undoped zinc oxide films deposited on silicon substrates
Applied Physics Letters 79, 943 (2001); 10.1063/1.1394173
APPLIED PHYSICS LETTERS 87, 122102 共2005兲

Oxygen vacancies in ZnO


Anderson Janotti and Chris G. Van de Walle
Materials Department, University of California, Santa Barbara, California 93106-5050
共Received 22 March 2005; accepted 28 July 2005; published online 14 September 2005兲
The electronic properties of ZnO have traditionally been explained by invoking intrinsic defects. In
particular, the frequently observed unintentional n-type conductivity has often been attributed to
oxygen vacancies. We report first-principles calculations showing that the oxygen vacancy VO is not
a shallow donor, but has a deep ␧共2 + / 0兲 level at ⬃1.0 eV below the conduction band. The negative-
U behavior that causes the 1 + charge state to be unstable is associated with large local lattice
relaxations. We present a detailed configuration coordinate diagram, which allows us to provide a
detailed interpretation of recently reported ODEPR 共optically detected electron paramagnetic
resonance兲 measurements 关L. S. Vlasenko and G. D. Watkins, Phys. Rev. B 71, 125210 共2005兲兴.
© 2005 American Institute of Physics. 关DOI: 10.1063/1.2053360兴

Zinc oxide is a wide-band-gap semiconductor with nu- calculations are carried out at the theoretical equilibrium lat-
merous applications, for instance in varistors, phosphors, and tice constants,11 which is essential in order to avoid spurious
sensors. Recent progress in crystal growth has also opened effects when calculating relaxations. We derive the formation
up prospects for optoelectronic devices. Such applications energy E f , which determines the likelihood for a defect to
are currently hampered, however, by the lack of control over form during growth. The formation energy of a native defect
conductivity. ZnO is typically n-type, and high-resistivity D in charge state q is defined as12
material has been difficult to obtain; p-type doping has been
reported,1,2 but reliability and reproducibility are serious is-
sues. Understanding the source of unintentional conductivity E f 共Dq兲 = Etot共Dq兲 − Etot共ZnO兲 + nZn␮Zn + nO␮O + qEF . 共1兲
is a crucial step towards control of doping. The accepted
wisdom is that the unintentional n-type conductivity is due to Etot共Dq兲 is the total energy of the supercell containing the
native point defects.3 Oxygen vacancies have been most defect D in the charge state q, and Etot共ZnO兲 is the total
widely discussed, but most of the arguments have been based energy of the ZnO perfect crystal in the same supercell. The
on circumstantial evidence, in the absence of unambiguous Fermi energy EF 共which will be referenced to the top of the
experimental observations. valence band Ev兲 enters because the defect can occur in vari-
Recent computational studies4–7 have cast doubt on the ous charge states, and the electrons that are removed from
categorization of the oxygen vacancy 共VO兲 as a shallow do- the defect need to be placed in their reservoir, with energy
nor. However, the reported electronic levels display wide dis- EF. ␮Zn and ␮O are the chemical potentials of Zn and O, i.e.,
crepancies. In this Letter we present a comprehensive theo- the energies of the reservoirs with which Zn and O atoms are
retical study which not only provides more reliable exchanged. For displaying our results below, we choose Zn-
electronic levels for the defect, but also addresses the unusu- rich conditions 共␮Zn = ␮Zn共solid兲兲, the most extreme case for
ally large atomic relaxations that occur in its various charge favoring formation of oxygen vacancies.
states. These result in the 1 + charge state being thermody- From the formation energies E f 共Dq兲 we can calculate
namically unstable, i.e., VO behaves as a ”negative-U” cen- transition levels ␧共q / q⬘兲.12 Since the local density approxi-
ter. Since the 1 + charge state is paramagnetic 共with VO +
con- mation 共LDA兲 severely underestimates band gaps, a correc-
taining an unpaired electron兲 this instability has important tion is necessary to compare the calculated ␧共q / q⬘兲 with ex-
repercussions for observations of VO based on electron para- periment. We base the correction on calculations performed
magnetic resonance 共EPR兲: no EPR signal should be ob- using the LDA on the one hand, and the “LDA+ U”
served in equilibrium, and an excitation process 共illumina- approach13 on the other hand. The idea of the LDA+ U
tion兲 is necessary to produce the paramagnetic state. Our method is to separate electrons into two subsystems: local-
calculated formation energies indicate that VO is too high in ized d electrons for which the strong Coulomb repulsion U
energy to play any significant role in as-grown n-type mate- should be taken into account via a Hubbard-like term in a
rial. However, it is expected to act as a compensating center model Hamiltonian, and delocalized electrons which can be
in p-type ZnO and it can also be formed during irradiation described by an orbital-independent one-electron potential.
and ion implantation. Our calculated configuration coordi- The LDA+ U approach lowers the energy of the Zn d states,
nate diagram allows a consistent interpretation of recent ex- thus reducing their repulsion with the O p states at the top of
periments by Vlasenko and Watkins8 which have provided the valence band, and increasing the band gap.11 The LDA
the most detailed information to date about EPR and optical + U thus provides a partial but physically important correc-
transitions associated with VO. tion of the band gap, the remaining correction arising from
Our investigation is based on state-of-the-art first- the intrinsic LDA error 共see Table I兲. However, we can use
principles calculations using density-functional theory, the LDA and LDA+ U results to extrapolate to the experi-
pseudopotentials, and plane-wave expansions of wave func- mental gap. Indeed, the shift of the transition level is deter-
tions and potentials.9,10 A supercell geometry with 96 atoms mined by its relative amount of valence-band versus
in the cell was employed, and a 2 ⫻ 2 ⫻ 2 mesh of special conduction-band character, and this information is contained
k-points for integrations over the Brillouin zone. All of our in the shift of ␧共q / q⬘兲 between LDA and LDA+ U. Table I

0003-6951/2005/87共12兲/122102/3/$22.50 87, 122102-1 © 2005 American Institute of Physics


122102-2 A. Janotti and C. G. Van de Walle Appl. Phys. Lett. 87, 122102 共2005兲

TABLE I. Calculated band gaps and transition levels ␧共q / q⬘兲 for oxygen
vacancies in ZnO 共in eV兲. LDA, LDA+ U, and extrapolated values are
listed.

Eg ␧共2 + / + 兲 ␧共+ / 0兲 ␧共2 + / 0兲


LDA 0.81 1.30 0.16 0.73
LDA+ U 1.50 1.73 0.64 1.18
Ext. 3.43 2.90 1.94 2.42

shows the LDA and LDA+ U results as well as the extrapo-


lated ␧共q / q⬘兲, which is given by
FIG. 2. 共Color online兲 Ball and stick representation of the local atomic
relaxations around the oxygen vacancy in the 0, 1+, and 2 + charge states.
␧共q/q⬘兲LDA+U − ␧共q/q⬘兲LDA The position of the a1 state is also shown for the equilibrium configuration
␧共q/q⬘兲 = 共Eexp LDA+U
g − Eg 兲 of each charge state.
ELDA+U
g − ELDA
g

+ ␧共q/q⬘兲LDA+U , 共2兲 and VO 2+


for all values of EF. This is related to the fact that the
where ELDA and ELDA+U are the band gaps given by the LDA ␧共2 + / + 兲 transition level is located higher in the gap than the
g g
and LDA+ U approximations, and Eexp is the experimental ␧共+ / 0兲 level, an unusual ordering based on the expected
g
gap.14 Table I shows that the transition levels exhibit signifi- Coulomb repulsion that occurs when electrons are added to a
cant upward shifts when the band gap is corrected. This is to defect. Since the 1 + charge state is not stable in equilibrium,
be expected, since the VO defect states are composed of dan- any EPR measurements claiming to identify VO without pro-
gling bonds on the neighboring Zn atoms, which exhibit viding any excitation should be treated with suspicion. In
mainly conduction-band character. The extrapolated values fact, those papers systematically claim to identify VO based
for the transition levels are also very close to results obtained on a line with g ⬇ 1.96. In reality, VO corresponds to a line
using self-interaction-corrected pseudopotentials,15 enhanc- with g ⬇ 1.99, and the reports16,17 that correctly make this
ing confidence in the approach. identification have in common that 共1兲 electron irradition
Figure 1共a兲 shows the formation energy of the oxygen was used to produce the defect 共since, in agreement with our
vacancy VO as a function of Fermi level EF for the three calculations, the starting n-type material contains only a neg-
possible charge states VO 0
, VO+
, and VO2+
, corrected according ligible concentration of VO兲 and 共2兲 optical excitation was
to the procedure described above. Under p-type conditions, found to be necessary to observe the EPR signal, in agree-
+
where EF is near the top of the valence band, the oxygen ment with the calculated metastability of VO .
vacancy is stable in the VO 2+
charge state and therefore acts as The effective correlation energy U 共no relation to
a compensating center with low formation energy. Under the parameter in the LDA+ U approach above兲 is defined
n-type conditions, however, where EF is near the bottom of as the difference between the relevant transition levels, i.e.,
the conduction band, the oxygen vacancy is in the neutral U = ␧共+ / 0兲 − ␧共2 + / + 兲 = E f 共VO2+兲 + E f 共VO0兲 − 2E f 共VO+兲. Our
charge state. The calculated 共2 + / 0兲 transition level is located calculations yield ␧共+ / 0兲 = 1.9 eV and ␧共2 + / + 兲 = 2.9 eV,
deep in the band gap, at ⬃1.0 eV below the bottom of the and, hence, U = −1.0 eV. The negative value reflects the fact
conduction band, indicating that VO will not provide elec- that atomic relaxations play a major role in stabilizing par-
trons to the conduction band at room temperature. In addi- ticular charge states. Our calculations indeed show that the
tion, the formation energy of VO under n-type conditions is lattice relaxations around VO are large and very different in
0
very high 共⬃3.5 eV兲, indicating that it is unlikely to form different charge states. For VO , the four Zn nearest neighbors
even at high temperatures during growth. We conclude that are displaced inward by 12% of the equilibrium Zn–O bond
+ 2+
oxygen vacancies are not the source of unintentional n-type length, whereas for VO and VO the displacements are out-
conductivity. ward by +2% and +23%, as shown in Fig. 2. Note that the
Figure 1共a兲 also shows that the 1+ charge state is not symmetry of the wurtzite structure does not impose that the
+ 0 relaxations of all four Zn atoms are identical; however, we
stable: the formation energy of VO is higher than that of VO
find that to a good approximation the Zn atom along the c
axis relaxes by the same amount as the other three Zn neigh-
bors 共to within 0.01Å兲, and the magnitude of this breathing
relaxation is very similar to that obtained from calculations
for the zinc-blende structure.
The origin of these large lattice relaxations lies in the
electronic structure of VO. The removal of an oxygen atom
from a perfect ZnO crystal leaves four Zn dangling bonds,
each contributing 1 / 2 electron to a neutral vacancy. The in-
teraction of these dangling bonds results in a completely
symmetric a1 state 共which lies in the band gap兲 and three
almost degenerate higher energy states 共resonant in the con-
0
duction band兲. For VO , the a1 state is occupied by two elec-
FIG. 1. 共a兲 Formation energies vs Fermi level for oxygen vacancies in ZnO trons, and its energy is lowered when the four Zn atoms
in the 0, 1+, and 2 + charge states, under Zn-rich conditions, calculated using
the procedure described in the text. 共b兲 Position of ␧共2 + / 0兲 with respect to
approach each other, with the gain in electronic energy ex-
the top of the valence band Ev in ZnO compared to that in ZnSe, taking into ceeding the cost to stretch the Zn–O bonds surrounding the
account the natural valence-band offset. vacancy. The resulting a1 state 共identified by the Kohn–Sham
122102-3 A. Janotti and C. G. Van de Walle Appl. Phys. Lett. 87, 122102 共2005兲

+
EM0 + VO → EM+ + VO
0
, 共3兲
corresponding to signals in a broad photoluminescence 共PL兲
band which is centered at 600 nm 共2.07 eV兲 and has a zero-
phonon line around 500 nm 共2.48 eV兲 . The authors propose
two possible models. In the first, the 共+ / 0兲 transition level is
placed at 0.9 eV above the top of the valence band, and the
luminescence results from the transfer of an electron from an
effective-mass donor 共EM兲 to the vacancy, i.e., the spin-
dependent process is the luminescence process itself. In the
second model, the 共+ / 0兲 transition level is placed at 2.48 eV
FIG. 3. Calculated configuration coordinate diagram for VO0, VO+, and VO2+. In
above the top of the valence band and the luminescence re-
共a兲 the Fermi energy is located at the conduction-band minimum Ec and in
共b兲, at the valence-band maximum Ev. The arrows indicate light absorption sults from hole capture by the neutral vacancy.
Ea and light emission Ee. Our computational results strongly favor the second
model. Indeed, the cc diagram of Fig. 3共b兲 indicates a zero-
level in our calculated band structures兲 lies near the top of phonon line transition from VO 0 +
to VO at ⬃2.0 eV, with an
the valence band 共see Fig. 2兲. In the VO +
configuration, the a1 emission band centered at ⬃1.5 eV. Furthermore, Fig. 3共a兲
state is occupied by one electron, and the electronic energy shows that the spin-dependent electron-capture process in
gain is too small to overcome the strain energy; the four Zn Eq. 共3兲 is predicted to produce luminescence at ⬃0.7 eV
atoms are displaced slightly outward, moving the a1 state and, indeed, the PL spectra in Ref. 8 show a weak broad
near the middle of the gap. In the VO 2+
configuration, the a1 band centered around 1500 nm 共0.8 eV兲.
state is unoccupied and the four Zn atoms strongly relax In summary, we have shown that the oxygen vacancy is
outward, assuming almost planar configurations with respect a deep donor in ZnO with the 共2 + / 0兲 transition level at
to their oxygen neighbors. This relaxation strengthens the ⬃1.0 eV below the bottom of the conduction band. There-
Zn–O bonds, and the empty a1 state lies near the bottom of fore, VO is not responsible for the unintentional n-type con-
the conduction band. This large relaxation significantly re- ductivity often observed in as-grown ZnO, although it may
2+ + act as a compensating center in p-type ZnO. Our calculated
duces the formation energy of VO with respect to VO , mak-
ing the oxygen vacancy a negative-U center. configuration coordinate diagram provides a consistent inter-
One may wonder why the transition levels associated pretation of recent ODEPR measurements.8
with the anion vacancy in ZnO lie so much higher in the The authors would like to express our gratitude to Pro-
band gap than the levels associated with, for instance, the fessor G. D. Watkins for making Ref. 8 available to us prior
selenium vacancy in ZnSe,18 since in both cases the elec- to publication and for many enlightening discussions. This
tronic states have mainly Zn dangling bond character. In fact, work was funded by AFOSR under Grant No. F49620-02-1-
the positioning of the levels is consistent provided one takes 1163 through a subcontract from the Palo Alto Research
the alignment of band structures on an absolute energy scale Center Incorporated, and made use of MRL Central Facilities
into account 关see Fig. 1共b兲兴. The large difference in energetic supported by the NSF MRSEC Program under award No.
position of the anion p states that determine the valence-band DMR00-80034, and the CNSI Computing Facility under
maximum 共O 2p versus Se 4p兲 causes a large valence-band NSF grant No. CHE-0321360.
offset between ZnSe and ZnO.19 Given this large discontinu-
1
ity, it should be no surprise that Zn-related states occur at D. C. Look, B. Claflin, Ya. I. Alivov, and S. J. Park, Phys. Status Solidi A
very different relative positions in the band gap. 201, 2203 共2004兲.
2
D. C. Look and B. Claflin, Phys. Status Solidi B 241, 624 共2004兲.
The large difference in the local lattice relaxations of 3
D. C. Look, J. W. Hemsky, and J. R. Sizelove, Phys. Rev. Lett. 82, 2552
2+ + 0
VO , VO , and VO leads to large Stokes shifts in the optical 共1999兲, and references therein.
4
spectra according to the Franck-Condon principle. Figure 3 A. F. Kohan, G. Ceder, D. Morgan, and C. G. Van de Walle, Phys. Rev. B
shows our calculated configuration coordinate 共cc兲 diagram, 5
61, 15019 共2000兲.
obtained by calculating the formation energy in the different S. B. Zhang, S.-H. Wei, and A. Zunger, Phys. Rev. B 63, 075205 共2001兲.
6
E.-C. Lee, Y.-S. Kim, Y.-G. Jin, and K. J. Chang, Phys. Rev. B 64, 085120
charge states for a series of atomic positions across the range 共2001兲.
of relaxations. Figure 3共a兲 shows that the peak in the light 7
F. Oba et al., J. Appl. Phys. 90, 824 共2001兲.
excitation spectrum of VO 0
→ VO+
is predicted to occur at 8
L. S. Vlasenko and G. D. Watkins, Phys. Rev. B 71 125210 共2005兲.
9
⬃2.0 eV, in reasonable agreement with the measured value P. Hohenberg and W. Kohn, Phys. Rev. 136, B864 共1964兲; W. Kohn and
L. J. Sham, Phys. Rev. 140, A1133 共1965兲.
of 2.3 eV 共530 nm兲.17,20 The calculated emission is predicted 10
G. Kresse and J. Furthmüller, Phys. Rev. B 54, 11169 共1996兲; G. Kresse
to occur at 0.7 eV, corresponding to a Stokes shift as large as and J. Furthmüller, Comput. Mater. Sci. 6, 15 共1996兲.
1.3 eV. Excitation of the center to the +1 charge state may 11
Anderson Janotti, D. Segev, and C. G. Van de Walle 共unpublished兲.
12
lead to persistent photoconductivity. The calculated thermal C. G. Van de Walle and J. Neugebauer, J. Appl. Phys. 95, 3851 共2004兲.
13
barrier to escape out of the +1 charge state back to the V. I. Anisimov et al., Phys. Rev. B 48, 16929 共1993兲; A. I. Liechtenstein,
V. I. Anisimov, and J. Zaanen, ibid. 52, R5467 共1995兲.
ground state is 0.30 eV. Therefore persistent photoconductiv- 14
O. Madelung, editor, Semiconductors - Basic Data, 2nd revised edition
ity 共as well as a persistence of the VO +
EPR signal after the 共Springer, Berlin, 1996兲.
exciting light is turned off兲 is expected to be observed only at 15
C. G. Van de Walle 共unpublished兲.
16
low temperatures. J. M. Smith, W. E. Vehse, Phys. Lett. 31A, 147 共1970兲.
17
Finally, we turn to an interpretation of recent experi- C. Gonzales, D. Galland, and A. Hervé, Phys. Status Solidi B 72, 309
共1975兲.
ments by Vlasenko and Watkins using optical detection of 18
A. Garcia and J. E. Northrup, Phys. Rev. Lett. 74, 1131共1995兲.
electron paramagnetic resonance 共ODEPR兲 in electron irra- 19
C. G. Van de Walle and J. Neugebauer, Nature 共London兲 423, 626 共2003兲.
diated ZnO.8 They observe a spin-dependent process: 20
D. Galland and A. Hervé, Solid State Commun. 14, 953 共1974兲.

You might also like