J O U R N A L O F M AT E R I A L S S C I E N C E : M AT E R I A L S I N E L E C T RO N I C S 1 4 ( 2 0 0 3 ) 4 3 1 ± 4 3 5

Piezoelectric ZnO ®lms by r.f. sputtering

J. MOLARIUS, J. KAITILA, T. PENSALA, M. YLILAMMI
VTT Microelectronics, P.O. Box 1208, FIN-02044 VTT Espoo, Finland
E-mail: Jyrki.Molarius@vtt.®.

Piezoelectric zinc oxide ®lms are used in microelectromechanical systems (MEMS)

applications, where they can be used in sensors to detect, e.g., pressure or acceleration.
Beside sensors, ZnO ®lms are applied in activation devices, where force is needed.
Conductive-doped zinc oxide (most often with aluminum) is also used in optoelectronics.
Piezoelectric ®lms including AlN and ZnO are more dif®cult to produce than the
corresponding conductive materials. In order to achieve good piezoelectricity in ZnO ®lms,
they have to possess high purity, a (0 0 1) orientation (ZnO has hexagonal crystal structure),
high resistivity, and ®ne columnar microstructure perpendicular to the substrate. We have
used r.f. magnetron (13.56 MHz) sputtering from a ZnO target in an oxygen atmosphere to
achieve the piezoelectric ZnO. The aim of this work has been to develop an r.f. sputtering
process for ZnO to achieve highly piezoelectric thin ®lms. As a test vehicle to measure the
piezoelectricity of the ZnO ®lms we have fabricated resonators and passband ®lters in the
1±2 GHz range using standard microelectronics photolithography, deposition, and etching
techniques on 100-mm diameter Corning glass or silicon wafers. The in¯uence of the
sputtering-process parameters on the ®lm properties has been studied by X-ray diffraction,
scanning electron microscopy, atomic force microscopy, and electrical measurements. In
this study, the effects of the process parameters on the ®nal material properties of the ZnO
®lm are discussed in detail.
# 2003 Kluwer Academic Publishers

1. Introduction wafer chamber. R.f. magnetron sputtering was from a

Piezoelectric zinc oxide ®lms can be used in surface zinc oxide target with the purity of 99.999% in a pure
acoustic wave (SAW) or thin-®lm bulk acoustic wave oxygen atmosphere. The oxygen gas purity was
resonator (FBAR) ®lter technology in mobile commu- 99.9999%. The loadlocked cluster-type sputtering
nications. There is also possible applications in MEMS system was pumped with oil-free turbodrag and
both in sensors and actuators. In all of these applications diaphragm pumps to below 5610 5 Pa before sample
the piezoelectric quality of the ®lm is of utmost processing.
importance, since it determines the performance of the Resonators and ®lters were solidly mounted resonator
®nal product. The zinc oxide ®lm properties depend on (SMR)-type [2] fabricated on quarter-wavelength acous-
the growth method and parameters. Furthermore, to tical mirrors, consisting of alternating layers of high and
evaluate the measured properties they have to be tried in low acoustic impedance, two pairs of W±SiO2 or three
real applications. Therefore, in this work we have pairs of Mo±SiO2 . Tungsten has the highest known
developed, measured, and tested the r.f. sputtered acoustic impedance and the impedance ratio Zhigh : Zlow
piezoelectric ZnO ®lms in actual r.f. ®lter devices. for W±SiO2 pair is 7.7 : 1. This ensures good acoustic
isolation with only two layer pairs. In the case of Mo±
SiO2 the ratio is 4.8 : 1 and consequently one needs the
2. Experimental extra pair. Metals were sputter deposited by a d.c.
We have used 100 mm glass (Corning 7059) and (1 0 0)- magnetron and plasma-enhanced chemical vapor deposi-
oriented silicon wafers. The latter were used for tion (PECVD) was used for SiO2 . Our resonator and ®lter
structural characterization of the deposited zinc oxide fabrication is explained in more detail in Kaitila et al. [1].
®lms. Glass wafers were chosen as substrates for Phases and orientation of the ZnO ®lms were
processing resonators and ®lters to eliminate the parasitic characterized with X-ray diffraction (XRD, Y±2Y)
effects associated with the semiconducting silicon [1]. using a Philips PW 1830 diffractometer. Film mor-
Films were sputtered in a cluster tool (Von Ardenne CS phology was studied using a digital scanning electron
730 S) from 200 mm diameter round targets. In this microscope (SEM, Leo 1560). Surface roughness and
system, metal ®lms are deposited in a multitarget topology were measured by atomic force microscopy
chamber with d.c. magnetrons, but to minimize cross (AFM, Digital Instruments Dimension 3100). For
contamination, ZnO is sputtered in a dedicated single- electrical characterization of the completed devices an
0957±4522 # 2003 Kluwer Academic Publishers 431
Figure 1 Target potential as a function of oxygen ¯ow rate. Target
power is 500 W.

Agilent Technologies 8753 network analyzer was

utilized.

3. Results and discussion

In reactive sputtering there are often hysteresis effects,
which can change the deposition rates dramatically (see, Figure 2 ZnO growth rate as function of power on the sputtering target
for example, Sa® [3]). Therefore, we have measured the in (a) only power is varied, and in (b) also other processing parameters
have been varied.
target potential as a function of the oxygen ¯ow rate.
(Although physically it is the chamber pressure we are
be kept below 1 mm h 1 …*18 nm min 1 † to achieve
dealing with, in this sputtering system it is directly
high-quality ZnO [5]. However, there are also contra-
controlled by gas ¯ow (mass ¯ow controllers) and as
dicting reports [6].
pressure measurement during sputtering is not as precise,
The most important parameter of the quality of
we decided to plot it this way.) As can be seen in Fig. 1
piezoelectric ZnO is to have a strong preferred
there is a change in the slope of the curve at about
orientation of the ®lm. For FBAR devices operating in
36 sccm ¯ow rate. The coef®cient of the slope up to
the longitudinal wave mode this is (0 0 1). In Fig. 3, the
36 sccm is 0:80 V sccm 1 and above it 0:52 V sccm 1 .
effect of reduced sputtering power and therefore reduced
This is because of target oxidization ( poisoning). The
growth rate can be seen as diminishing the unwanted
effect is relatively small, because we do not use a metal
(1 0 0) and (1 0 1) X-ray peaks. This seems to support a
Zn target, but already have a ZnO target. Although we
model based on the growth-rate limitation. The idea is
use an oxide target, oxygen is also needed in the
the same as in epitaxial growth; namely that if there is
atmosphere, otherwise depletion of oxygen in the ZnO
insuf®cient time for the atoms to move to their
®lm will cause conductivity (Jeong and Park [4]) and
energetically favored position (in this case positions in
consequently poor piezoelectricity.
ZnO resulting in the (0 0 1) orientation) then crystals of
To see how signi®cant this target ``poisoning'' is in the
all directions will grow, resulting in non-oriented ®lm
actual deposition, the growth rate of the ZnO ®lm was
with poor piezoelectric properties.
measured as only the target power was changed. This
Already in 1982 Anderson et al. [8] reported that too
gives a linear dependence (Fig. 2(a)). When other
process parameters are varied the trend of the growth
rate as a function of the target power remains linear (see
Fig. 2(b)). Target power is limited by the low thermal
conductivity of the insulating ZnO target, which only
permits us to use up to 600 W …*2 W cm 2 † power. As
can be seen in Fig. 2(a) and (b) the growth rate is quite
low, and this, combined with the rather thick ®lms
needed for resonators (see layer stack in Fig. 7), results in
long sputtering times; e.g., for a ®lter at 1 GHz frequency
the sputtering time for ZnO is about two hours. In the
research laboratory one can live with low growth rates, if
the ®lm quality is good, but for manufacturing this might
be a problem. Fortunately, when scaling to higher
frequencies in the near future the piezoelectric layers
Figure 3 XRD measurement of the ZnO ®lms, with target power of (a)
get thinner, e.g., below 500 nm of ZnO at 5 GHz. From 450 W and (b) 300 W. (The standard powder ZnO peak positions are
the point of view of piezolayer quality, there have been marked by dashed lines [7]. Note the square root scale on the ordinate
many reports in the literature that the growth rate should axis.)

432
 (a)

(b)

Figure 4 AFM pictures …3 mm 6 3 mm† of (a) low-quality ZnO, sample X9#23 (vertical scale 300 nm), (b) high-quality piezoelectric ZnO ®lm,
sample 2W#66 (vertical scale 100 nm).

433
 observed keff . Knowing the exact resonator stack allows
one to calculate the material coupling coef®cient kmat .
Out of our two exemplary samples X9#23 did not
show any piezoelectric resonance, whereas resonators on
sample 2W#66 had keff between 0.23 and 0.24. These
values translate to material coupling coef®cients kmat of
0.25±0.26, which compare favorably with the bulk value
of 0.282 reported for ZnO [11] and also previously
published results for thin ®lms [12]. Most importantly,
these values are high enough to realize working ®lters for
the targeted application.
As already pointed out, in FBARs the thickness of the
piezoelectric ®lm is inversely related to the resonant
Figure 5 ZnO ®lm thickness uniformity in two directions across the frequency. This is clearly seen in Fig. 5: the center of the
wafer expressed as the resonance-frequency distribution. …P ˆ 550 W, wafer has a lower resonance frequency compared to the
T ˆ 150  C, bias ˆ 20 V, O2 ˆ 45 sccm, 2.0 Pa†. edges. (Here, the wafer ¯at is marked as South, when
looking on the wafer device side up; the other points of
strong a magnetic ®eld will adversely affect the ZnO ®lm the compass follow from there.) This means that the ®lm
quality. We did not have the means to change the is thicker at the center of the wafer and we can only get to
magnetic ®eld itself, and therefore we changed the target the right frequency for ®lters in a small part of the wafer.
to substrate distance (TSD) from 65 mm initially ®rst to This rather unfortunate fact follows from the sputtering
85 mm, then to 95, and ®nally to 105 mm. The increase of chamber con®guration changes, when we increased the
TSD also affects the energy of the particles striking the target to substrate distance from the optimum (65 mm)
substrate/growing ®lm and the hitting angles on the for the ®lm thickness uniformity to the optimum of ®lm
substrate will become shallower. No matter what the quality (105 mm).
exact physical reason might be, the results were striking: The highly oriented nature of the zinc oxide ®lm can
Both the (0 0 1) orientation of the piezoelectric ZnO
increased (by XRD) and the amount of off-oriented
crystals diminished (by AFM) with increasing distance.
In Fig. 4(a) the surface topography of sample X9#23 is
shown in an AFM picture. A TSD of 65 mm results in
large randomly oriented crystals. The ZnO ®lm on
sample 2W#66 was grown with a TSD of 105 mm. The
ZnO grain size is smaller, the ®lm is smoother, and even
the six-fold symmetry of the hexagonal columnar
crystals can be seen in Fig. 4(b). This is in agreement
with Anderson et al. [8] and Sharma et al. [9], although
they did not have AFM measurements to back up their
conclusions.
The real tests for the piezoelectric quality of the ZnO
®lms are the electrical measurements from fabricated
devices. The most important ®gures of merit for a
resonator are the effective coupling coef®cient keff and
the Q values. The effective coupling coef®cient is
de®ned as [10]

2 fp2 fs2
keff ˆ
fp2
where fp and fs are the measured parallel and series
resonance frequencies, respectively.
The Q values can be calculated by differentiating the
phase
 
o qj
Qˆ  
2 qo
which is to be evaluated at the series and parallel
resonance frequencies. From these two ®gures the
effective coupling coef®cient keff best describes the
piezoelectric quality of the ZnO ®lm. However, this does
not directly relate to the capability of the piezoelectric
Figure 6 Cross section of a FBAR resonator, (a) whole structure; glass
material itself to transform mechanical energy to substrate, mirror layers and resonator and (b) close up of ZnO
electrical energy, and vice versa. This is because other piezoelectric layer between bottom and top electrodes, sample T2#9.
layers in the device stack also have an effect on the (The scale line is 2000 nm long in (a) and 1000 nm long in (b).)

434
 quality of the piezoelectric ZnO ®lm in real devices. The
layer stack on the FBARs and ®lters is depicted in Fig. 7.
The sample 2W#69 was sputtered with the same
sputtering parameters …P ˆ 550 W, T ˆ 200  C,
bias ˆ 30 V, O2 ˆ 40 sccm, 1.7 Pa† as sample 2W#66
(Fig. 4(b)). This ladder-type ®lter could ful®l the EGSM
speci®cations without any performance enhancement by
external components (Fig. 8). The results are reported in
detail elsewhere by Kaitila et al. [1].

4. Conclusions
We have investigated the in¯uence of the sputtering
process parameters on the ZnO ®lm properties by XRD,
SEM, AFM, and electrical measurements. As a result of
this work a r.f. magnetron sputtering process was
developed to deposit high-quality piezoelectric ZnO
®lms. Films were highly insulating, had the desired
(0 0 1) orientation by XRD, and columnar microstructure,
by SEM and AFM. The ®nal quality of the piezoelectric
Figure 7 Layer stack for FBAR resonators and ®lters at 1 GHz. ZnO ®lms was tested by fabrication of EGSM bandpass
®lters. We could ful®l the EGSM speci®cations without
using any external components in the ®lter.

Acknowledgments
The authors want to thank Helena Pohjonen, Juha EllaÈ,
and Ilkka Suni for fruitful discussions and encourage-
ment during the course of this work. This work was
funded by Nokia Mobile Phones Ltd.

References
1. J. KAITILA, M. YLILAMMI, J. MOLARIUS, J. ELLA
È and T.
Figure 8 Electrical measurements of a ®lter structure ful®lling 1 GHz M A K KO N E N , IEEE Ultrason. Sympo. (2001) 223.
EGSM speci®cations (shown in red) without any external components. 2. K . L A K I N , G . K L I N E and K . M C G A R RO N , IEEE Trans.
Microw. Theory 43 (1995) 2933.
also be seen in the cross-sectional SEM (Fig. 6). The 3. I . S A F I , Surf. Coat. Technol. 127 (2000) 203.

ZnO ®lm clearly has a columnar microstructure, which 4. W. - J . J E O N G and G . - C . PA R K , Sol. Energy Mater. Cells 65
(2001) 37.
was also revealed by AFM in Fig. 4(b). This is in 5. C . R . A I TA , IEEE Ultrason. Symp. (1980) 483.
agreement with Igasaki et al. [13]. Both photographs, 6. F. S . H I C K E R N E L L , ibid. (1980), 785.
Fig. 6(a) and (b), are taken from an actual device, where 7. JCPDS-ICDD (the Joint Committee on Powder Diffraction
ZnO is grown on top of an acoustic quarter-wavelength Standards-International Centre for Diffraction Data) ®le no. 36-
mirror. The design thickness of these layers for 1.8 GHz 1451.
8. A . C . A N D E R S O N and D . E . O AT E S , IEEE Ultrason. Symp.
resonators are; starting from the glass substrate, Mo (1982) 329.
848 nm (6 3), SiO2 810 nm (6 3), bottom electrode Mo 9. P. S H A R M A , A . M A N S I G H , K . S R E E N I VA S , Appl. Phys. Lett.
300 nm, and ZnO 1100 nm. The top aluminum electrode 80 (2002) 553.
thickness is tuned according to the ZnO layer (usually 10. IEEE Standard on Piezoelectricity, ANSI/IEEE Std 176-1987.
200±300 nm). The top aluminum electrode is deformed 11. H . J A F F E and D . A . B E R L I N CO U R T, Proc. IEEE 53 (1965)
1372.
during sample cleaving for SEM. The glass wafer and 12. C . W. S E A B U R Y, J . T. C H E U N G , P. H . KO B R I N , R . A D D I S O N
SiO2 show amorphous structure, whereas the columnar and D . P. H AV E N S , IEEE Trans. Ultrason., Ferroelect., Freq.
microstructure of molybdenum in the mirrors (Fig. 6(a)) Control 42 (1995) 339.
and in the bottom electrode (Fig. 6(b)) is clearly 13. Y. I G A S A K I , T. N A I T O , K . M U R A K A M I and W. TO M O D A ,

revealed. Appl. Surf. Sci. 169±170 (2001) 512.

We used EGSM (extended GSM) ± the European
standard for second generation 1 GHz mobile commu- Received 10 June
nication ± ®lters as test vehicles to check the absolute and accepted 11 December 2002

435