Sensors and Actuators B 124 (2007) 24–29
WO
3
sensor response according to operating
temperature: Experiment and modeling
M. Bendahan
∗
,J.Gu
´
erin, R. Boulmani, K. Aguir
Laboratoire Mat´eriaux et Micro´electronique de Provence, L2MP-CNRS, Universit´e Paul C´ezanne,
Aix-Marseille III, Facult´e des Sciences et Techniques de St J´erˆome, France
Received 10 July 2006; received in revised form 22 November 2006; accepted 22 November 2006
Available online 20 December 2006
Abstract
WO
3
-based sensors are realized in the aim to detect ozone. The thin film of WO
3
is sputtered on a SiO
2
/Si substrate with Pt micro-electrodes. In a
previous work, the sensor response dependence on processing parameters has been studied. Now operating temperature of the sensor is investigated
and a theoretical model developed by our team confirms experimental measurements.
The interaction between the gas and the surface was modeled by Langmuir isotherm and the electrical resistivity was evaluated by solving the
transport equations.
© 2006 Elsevier B.V. All rights reserved.
Keywords: WO
3
; Gas sensor; Modeling
1. Introduction

Electrical properties of semiconductor oxides depend on the
composition of the surrounding gas atmosphere. The surface
conductivity of the sensor is modified by adsorption of gas
species and related space charge effects. In oxidizing atmo-
sphere, the oxide surface is covered by negatively charged
oxygen adsorbates and the adjacent space charge region is
electron-depleted: the oxide layer presents therefore a high resis-
tance. Under reducing conditions, the oxygen adsorbates are
removed by reaction with the reducing gas species and the elec-
trons are re-injected into the space charge layers: as a result, the
oxide layer resistance decreases.
Recently, gas sensing properties of simple binary metal
oxides, such as tin oxide (SnO
2
) and tungsten trioxide (WO
3
) [1]
have been tested for monitoring pollutant components of atmo-
sphere for improving quality of life and enhancing industrial
processes [2–4]. Tungsten oxide is an n-type metal oxide semi-
conductor with oxygen vacancies, which act as donors. Because
the electron density depends on the density of oxygen vacancies,
∗
Corresponding author. Tel.: +33 4 91288973; fax: +33 4 91288970.
E-mail address: (M. Bendahan).
the vacancies play a significant role in the detection mechanism
as in SnO
2
sensors [5].
Many techniques are being used for the fabrication of WO

3
films, including thermal evaporation [6,7], sol–gel [7] and sput-
tering [8–10].
Table 1 summarizes the responses (S=R
gas
/R
air
) of various
WO
3
-based ozone sensors and fabrication methods. For exam-
ple, Qu and Wlodarski [6] studied WO
3
ozone sensors deposited
on sapphire substrates by thermal evaporation. The working tem-
perature of the sensors was 573 K and the film thickness was
about 150 nm. Cantalini et al. [7] reported a study of WO
3
ozone
sensors realized on alumina substrates, by sol–gel, sputtering
and thermal evaporation techniques. The operating temperature
range was 473–673 K. The best ozone sensitivity is obtained with
the WO
3
sensors prepared using reactive magnetron sputtering,
with an operated temperature of 523 K [8], a film thickness of
about 40 nm and a grain size of 40 nm. These results show clearly
that the preparation method influences the sensor response. It is
now well known that the sensor response depends on physical
and chemical properties of sensitive films. In fact, morphology,

thickness, chemical composition, and microstructure of WO
3
thin films are very important parameters to obtained stable and
sensitive sensors. We have shown that sensor response depends
essentially on the grain size and film porosity [9]. These proper-
ties can be controlled during film deposition, using rf sputtering
0925-4005/$ – see front matter © 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.snb.2006.11.036
M. Bendahan et al. / Sensors and Actuators B 124 (2007) 24–29 25
Table 1
Gas sensor responses S vs. ozone concentration and different fabrication methods
Sensors Fabrication method Operating temperature (K) O
3
(ppb) S Reference
WO
3
Thermal evaporation 573 50 1.25 [6]
175 2.25
WO
3
Thermal evaporation 473 80 15 [7]
WO
3
Sol–gel 473 80 19 [7]
WO
3
rf sputtering 473 80 3 [7]
WO
3
rf magnetron sputtering 523 30 16 [8]

400 263
800 310
technique. The influence of processing parameters on the sensor
response, such as oxygen partial pressure in an Ar–O
2
gas mix-
ture used during the sputtering process [8] or self bias voltage
[10], has been studied by our team. The sputtering parameters
have been optimized to obtain sensors which exhibit the best
performance.
In the present work we report on electrical responses of WO
3
-
based sensors for ozone detection. WO
3
thin films are deposited
by rf reactive magnetron sputtering on a SiO
2
/Si substrate with
interdigitated platinum micro-electrodes (Fig. 1). Here, operat-
ing temperature of the sensor is investigated and the results are
compared with a theoretical model developed by our team.
2. Experimental
WO
3
thin films were prepared by reactive radio frequency
(13.56 MHz) magnetron sputtering, using a 99.9% pure tungsten
target. The vacuum chamber was evacuated to 5.0 × 10
−10
bar

by a turbo molecular pump. The films were sputtered in a reactive
atmosphere under an oxygen–argon mixture. Both argon and
oxygen flow were controlled by mass flow controllers. The total
gas flow was maintained constant at 10 sccm, keeping the total
pressure in the deposition chamber at 3.0 × 10
−3
mbar. Oxygen
content in the gas mixture, defined as the ratio of oxygen flow
to the total flow, was maintained at 50% [8].
As WO
3
layers are highly resistive, interdigitated electrodes
were used in order to reduce the sensor resistance. The distance
between the electrodes was 50 ␮m. They were obtained from a
sputtered Pt film, using photolithography and lift off processes.
The samples were kept in dry air and no conditioning step was
carried out before testing.
To investigate the ozone sensing properties of WO
3
films,
the sensors were introduced in a test chamber allowing the con-
trol of the sensor temperature under variable gas concentrations.
Dry air was used as a reference gas. Ozone gas was generated
by oxidizing oxygen using a pen-ray UV lamp (Stable Ozone
Generator UVP/185 nm). The intensity of the UV radiation was
varied by shifting a shutter around the lamp. The different ozone
concentrations are obtained in the range of 0.03–0.8 ppm with a
flow rate of dry air maintained at 30 l/h.
The operating temperature of the sensors was adjusted
between 423 and 673 K. The applied dc voltage was 50 mV

and the current was measured using a computerised HP4140B
source/pico-ammeter. The sensor response was defined as
S = G
air
/G
gas
, where G
air
and G
gas
are the conductance of the
sensor in air and in tested gas, respectively.
3. Sensor response versus operating temperature
3.1. Measurements
Fig. 2 illustrates a typical isothermal kinetic sensor response
at 523 K for various ozone concentrations. The response is plot-
ted versus time for ozone concentrations varying from 0.03 to
0.8 ppm. The recording cycle for each concentration is 6 min.
Sensitivity, stability, reversibility, reproducibility and response
Fig. 1. WO
3
sensor design.
26 M. Bendahan et al. / Sensors and Actuators B 124 (2007) 24–29
Fig. 2. Isothermal kinetic sensor response at 523 K to various ozone concentra-
tions.
time are very important parameters for evaluating sensor per-
formance. We can notice that the present sensor exhibits very
attractive performances: very high sensitivity to ozone concen-
tration at ppb levels, total reversibility, good reproducibility and
good stability of the baseline.

The sensor response in the presence of ozone can be inter-
preted by considering that the oxygen species interact with the
surface oxygen vacancies. Without ozone the density of surface
oxygen vacancies corresponds to an equilibrium established for
the oxygen partial pressure above the surface and to the con-
ductance in air. With ozone the oxygen species given by the
dissociative adsorption interact with the surface oxygen vacan-
cies according to
(O
3
)
g
→ 3O
ads
,
3O
ads
+ 6e
−
+ 3V
O
2+
→ 3O
lat
.
As a result, oxygen vacancies and the corresponding free elec-
trons are annihilated and the conductance decreases. In this
model six free electrons disappear when an ozone molecule
adsorbed on the WO
3

surface, so the electrical conductance will
be very sensitive to the ozone interaction with the sensing sur-
face. It is evident that the small grained films, which have a
large ratio of surface area to volume, will have a better sensitiv-
ity performance and the grain size appears as a very important
parameter for the sensitivity of undoped WO
3
thin films used as
a chemical sensing material.
In order to check the effect of operating temperature on the
sensor response, WO
3
sensors are maintained at fixed tempera-
tures from 423 to 673 K. Fig. 3 illustrates the response to 0.8 ppm
of ozone versus operating temperature for the sensor realized
with 50% O
2
in the oxygen–argon mixture during sputtering. It
shows a systematic increase of response with increasing operat-
ing temperature below 523 K, but reverse tendency is observed
above 523 K. We can also notice that the response and recov-
ery time decrease when temperature increases. The response
and recovery time are directly related to the adsorption and des-
orption activation energies, respectively. This can be explained
by considering the temperature dependence of the surface cov-
erage of chemisorbed species. At low temperature, there is
physisorption, but the rate of chemisorption is negligible. At
Fig. 3. WO
3
sensor response at different operating temperatures (800 ppb of

O
3
).
high temperature, the equilibrium chemisorption is possible but
the coverage decreases with increasing temperature because the
desorption rate rises faster than the adsorption rate. So, the cov-
erage of chemisorbed species shows a maximum with increasing
temperature [11].AtT = 423 K the gas desorption kinetic is slow,
which results in a high recovery time. The shortest times in
response and in recovery are obtained at 523 K. We can thus
conclude that the optimal operating temperature of the WO
3
thin film for ozone detection is about 523 K. This behaviour is
confirmed by the theoretical model developed in the following
section.
3.2. Theoretical model
In the last years, many authors have developed models for the
response of metal oxide gas sensors [12–14]. In these studies, the
sensitive layers are mainly tin oxide. The authors of these papers
have based their works on a potential barrier (Vs) theory model
[15]. In fact,according to this theory,conduction electrons can be
trapped by surface states driven by the energy difference between
the conduction band and surface states. The conductance of the
SnO
2
layer can then be expressed in a function of the potential
barrier (Vs) [14]. Nevertheless, this equation is available for a
porous layer and large grains with small contact regions (mean
diameter ∼1 ␮m). These models were developed to determine
the response of a temperature modulated sensor in the presence

of CO [12,13],NO
2
[13], and O
2
[12].
No model has been developed concerning tungsten oxide-
based sensors in the presence of ozone. So, in our work a
theoretical model has been developed to compute the WO
3
sen-
sor responses in the presence of different ozone concentrations
and for various working temperatures. The results are compared
with experimental results.
Interaction between a thin film and environment is modeled
by the Langmuir theory of the adsorption–desorption balance.
The analysis of the gas sensor operation using a semiconductor
metallic oxide thin layer can be simplified by considering the
effect of surrounding gases on the surface of the grains con-
stituting the sensitive layer on the one hand and the electronic
transport mechanisms in and between the grains on the other
hand.
M. Bendahan et al. / Sensors and Actuators B 124 (2007) 24–29 27
The surface of each grain is bathed by surrounding gas and
adsorption is uniform, leading to a radial establishment in elec-
tric field. In addition, because of the interdigitated structure of
the sensor electrodes, the applied electric field is axial. As tung-
sten trioxide is a wide gap semiconductor (2.7 eV), its electrical
conductivity is induced by the oxygen vacancies acting as n-type
impurities. Two different densities of vacancies associated with
two different donor levels are introduced to take into account the

two types of W–O bonds of the WO
3
crystalline structure. The
electrical charge of the transition zone that lies at the periphery
of the grains is induced by the environment of the layer.
The equations retained in the numerical model for the calcu-
lation of conduction are
• the Poisson’s equation related to the intrinsic potential:
ψ =
q
ε
(n − p − N
+
d
), (1)
where n and p are the densities of carriers described by the
Fermi–Dirac statistics;
• the continuity equations for the electrons and the holes:
Div J
n
= qU, Div J
p
=−qU, (2)
where U is the generation recombination rate derived from
the Shockley–Read–Hall model:
U =
np − n
2
i
τ

n
(p + n
i
) + τ
p
(n + n
i
)
; (3)
• the transport equations (drift diffusion model) for n and p:
J
n
=−nqμ
n
Grad ϕ
n
, J
p
=−pqμ
p
Grad ϕ
p
. (4)
These equations must be supplemented by a model of mobil-
ity:
μ(T ) = μ
0
300
T
. (5)

The complex adsorption mechanisms of ozone and atmospheric
oxygen by a surface can be schematized by the following four
chemical irreversible (or weakly reversible) reactions:
adsorption of oxygen : O
2
→ O
ads
+ O, (6)
adsorption of ozone : O
3
→ O
ads
+ O
2
, (7)
desorption of the sites : O
ads
→ O, (8)
molecular recombination : O + O → O
2
. (9)
Each reaction, thermally activated, is affected by each kinetic
constant:
k
i
= k
i0
exp

−

E
ai
kt

. (10)
When oxygen and ozone are simultaneously present, the equa-
tion of evolution is written:
d[N]
dt
= k
1
N
∗
[O
2
] + k
2
N
∗
[O
3
] − k
3
N, (11)
Table 2
Model parameters
N
max
(×10
13

cm
−2
) 0.8
ε
as
(eV) 0.0
k
cin
= k
30
/k
20
(×10
7
) 0.2
E
act
= E
a3
− E
a2
(eV) 0.8
k
ox
= k
10
/k
20
(×10
−8

) 0.1702
E
act
= E
a1
− E
a2
(eV) 0.1
N
d1
(×10
13
cm
−3
) 0.9
ε
d1
(eV) 0.672
N
d2
(×10
15
cm
−3
) 1.0
ε
d2
(eV) 0.959
N
max

: density of adsorption sites; ε
as
: acceptor level of adsorbed oxygen atoms;
k
cin
, k
ox
: kinetic constants associated to their activation energies E
act
; N
di
, ε
di
:
vacancies densities and donor levels used in the conduction model.
where N and N
*
= N
max
− N are the densities of occupied and free
adsorption sites, respectively, and [O
2
] and [O
3
] the oxygen and
ozone concentrations. This equation shows that in a stationary
state, the density of occupied sites is related to the concentrations
by the relation:
N =
N

max
(k
3
/(k
2
[
O
3
]
+ k
1
[
O
2
]
)) + 1
. (12)
The adsorbed atoms are partially ionized according to the reac-
tion:
αe
−
+ O
ads
→ αO
ads
−
+ (1 − α)O
ads
. (13)
The ionization rate α is deduced from the acceptor level ε

as
by
the Fermi–Dirac statistics [16]. So, it is possible to determine
the density of electrical charge on the surface of each grain.
The numerous parameters of the model could not be found in
literature. They were thus optimized mainly from thermoelectric
characterizations carried out in the laboratory. Table 2 gives the
main values of the parameters used in the simulations. We can
notice that the response of complete sensor (i.e. its variation of
resistivity) results from the composition of
• the adsorption mechanism with respect to oxygen and ozone;
• the modification of the electrical charge distribution in the
grains of the layer.
If the first process, described by the previous formulation of
N, is easy to be analysed, the second one cannot be modeled by
a simple analytical expression but it influences considerably the
total response.
Fig. 4 shows the theoretical evolution of the adsorption effi-
ciency σ =dθ
g
/d[O
3
] versus operating temperature in dry air
with various ozone concentrations (where θ
g
= N/N
max
is the
covering rate). The maximum efficiency of the adsorption pro-
cess with respect to ozone detection is obtained at the point

where the slope σ =dθ
g
/d[O
3
] is maximal. When ozone concen-
tration increases, this maximum occurs at higher temperatures
and its magnitude is smaller.
This can be explained by considering that at low tempera-
ture, the desorption is weak and the adsorption sites are almost
entirely saturated with oxygen; thus, a change in ozone concen-
tration does not produce significant effect. Conversely, at high
28 M. Bendahan et al. / Sensors and Actuators B 124 (2007) 24–29
Fig. 4. Adsorption efficiency in a mixture air/ozone vs. temperature for different
ozone concentrations (0.1, 0.3, and 1 ppm).
temperature, the desorption increases because of the high value
of the activation energy and the density of adsorbed species
becomes very weak; so, the sensitivity to ozone decreases.
Between these two opposite cases, for each ozone concentration
there is a temperature value for which the adsorption efficiency
is maximal. When the temperature rises, this optimum value
decreases.
We can also notice that the adsorption efficiency maximum
is shifted toward high temperatures with increasing the ozone
concentration. Indeed, when temperature increases, the desorp-
tion rate increases too. So, significant surface covering can be
reached only for higher ozone concentrations.
Fig. 5 shows the computed response of the sensor defined
by the ratio of resistances (or resistivities) in a dry air–ozone
mixture and dry air only: S = ρ
gas

/ρ
air
. Calculation is carried
out in two steps. First, the carrier densities and the electri-
cal potential of a set of two adjacent grains in thermodynamic
equilibrium surrounded by the gas mixture are computed using
Poisson equation. Then, a voltage is applied between the cen-
tre of the grains and the electrical current induced is calculated
using the transport equation. The resistivity is finally deduced.
We can then notice that there is an optimal operating temperature
which provides the highest response, as suggested by Fig. 4. The
simulation results are in good agreement with the experimental
measurements.
Fig. 5. Calculated sensor response in a mixture air/ozone vs. temperature for
different ozone concentrations (0.1, 0.3, and 1 ppm).
4. Conclusion
Tungsten oxide thin films are prepared by reactive magnetron
sputtering. A model of resistivity based on the existence of an
accumulation or a depletion layer induced by the surrounding
atmosphere has been elaborated and the simulations have been
compared to the experimental data. The interaction between the
gas and the surface was modeled by Langmuir isotherm and
the electrical resistivity was evaluated by solving the transport
equations.
We have shown that the sensor response to ozone depends on
the working temperature and that the adsorption efficiency in a
mixture air–ozone is also dependent on temperature. We can now
conclude that the variation of the sensor response with temper-
ature is linked to the temperature dependence of the adsorption
efficiency.

Acknowledgments
The authors gratefully acknowledge the fruitful collaboration
with many colleagues throughout this work. We want to mention
particularly the contribution by A. Combes (L2MP, Marseille)
for technical support.
References
[1] D.E. Williams, Semiconducting oxides as gas-sensitive resistors, Sens.
Actuators B 57 (1999) 1–16.
[2] X. Wang, N. Miura, N. Yamazoe, Study of WO
3
-based sensing materials
for NH
3
and NO detection, Sen. Actuators B 66 (2000) 74–76.
[3] D.S. Lee, S.D. Han, D.D. Lee, Nitrogen oxides-sensing characteristics of
WO
3
-based nanocrystalline thick film gas sensor, Sens. Actuators B 60
(1999) 57–63.
[4] D. Manno, A. Serra, M. Di Giulio, G. Micocci, A. Tepore, Physical and
structural characterization of tungsten oxide thin films for NO gas detection,
Thin Solid Films 324 (1998) 44–51.
[5] W. G
¨
opel, K.D. Schierbaum, SnO
2
sensors: current status and future
prospects, Sens. Actuators B 26–27 (1995) 1–12.
[6] W. Qu, W. Wlodarski, A thin film sensing element for ozone, humidity and
temperature, Sens. Actuators B 64 (2000) 42–48.

[7] C. Cantalini, W. Wlodarski, Y. Li, M. Passacantando, S. Santucci, E.
Comini, F. Faglia, G. Sberveglieri, Investigation on the O
3
sensitivity prop-
erties of WO
3
thin films prepared by sol–gel, thermal evaporation and rf
sputtering techniques, Sens. Actuators B 64 (2000) 182–188.
[8] M. Bendahan, R. Boulmani, J.L. Seguin, K. Aguir, Characterization of
ozone sensors based on WO
3
reactively sputtered films: influence of O
2
concentration in the sputtering gas, and working temperature, Sens. Actu-
ators B 100 (2004) 320–324.
[9] M. Gillet, K. Aguir, M. Bendahan, P. Mennini, Grain size effect in sputtered
tungsten trioxide thin films on the sensitivity to ozone, Thin Solid Films
484 (2005) 358–363.
[10] R. Boulmani, M. Bendahan, K. Aguir, Influence of rf sputtered parameters
on tungsten trioxide response sensors, in: Proceedings of the Eurosensors
XIX, Barcelona, Spain, September 2005.
[11] M.J. Madou,S. Roy Morrisson, Chemical Sensing withSolid State Devices,
chap3: Solid/Gas Interfaces, Academic Press, 1989, pp. 67–72.
[12] A. Fort, S. Rocchi, M.B. Serrano-Santos, M. Mugnaini, V. Vignoli, A.
Atrei, R. Spinicci, CO sensing with SnO
2
based thick film sensors: surface
state model for conductance responses during thermal-modulation, Sens.
Actuators B 116 (2006) 43–48.
[13] R. Ionescu, E. Llobet, S. Al-Khalifa, J.W. Gardner, X. Vilanova, J. Brezmes,

X. Correig, Response model for thermally modulated tin oxidebased micro-
hotplate gas sensors, Sens. Actuators B 95 (2003) 203–211.
M. Bendahan et al. / Sensors and Actuators B 124 (2007) 24–29 29
[14] J. Ding, T.J. McAvoy, R.E. Cavicchi, S. Semancik, Surface state trapping
models for SnO
2
-based microhotplate sensors, Sens. Actuators B 77 (2001)
597–613.
[15] S.R. Morrison, The Chemical Physics of Surfaces, 1st ed., Plenum press,
New York, NY, 1997.
[16] R.C. Jeager, F.H. Gaensslen, Simulation of impurity freezout through
numerical solution of Poisson’s equation and application to MOS device
behaviour, IEEE Trans. Electron Dev. 27 (1980) 914–920.
Biographies
M. Bendahan is a researcher at the Paul CEZANNE, Aix-Marseille III Univer-
sity (France). He is also lecturer in electronics at the Institute of Technology
of Marseille. He was awarded his PhD degree from the University of Aix-
Marseille III in 1996 with a thesis on shape memory alloys thin films. He is
specialized in thin films preparation and characterization for applications in
microsystems. Since 1997, he is interested in gas microsensors and he developed
a selective ammonia sensor based on CuBr mixed ionic conductor. He currently
works at Laboratoire Materiaux & Microelectronique de Provence (L2MP-
CNRS) Marseille (France), on WO
3
gas sensors and selectivity enhancement
strategies.
J. Gu
´
erin received his engineering diploma in electronics and radio-
communication at the Institut National Polytechnique of Grenoble (INPG) in

1972 and his PhD from the University of Aix-Marseille III (Paul Cezanne)
with a thesis on spatial silicon solar cells for observation satellites. After vari-
ous research and engineering developments (thermionic conversion, electronic
power devices, ), he joined the Sensors Group of the Laboratoire Materiaux
& Microelectronique de Provence (L2MP-CNRS) Marseille (France) in 2002.
Its principal research interests are now directed towards WO
3
gas sensors and
selectivity enhancement strategies, conduction and adsorption mechanisms and
modelling of sensor responses.
R. Boulmani obtained his PhD degree in physics and material science in the
L2MP laboratory at the Paul Cezanne Aix Marseille III University (France). His
research interest is the study and realization of microsensors based on tungsten
trioxide for the ozone detection.
K. Aguir is professor at Paul CEZANNE, Aix Marseille III University (France).
He was awarded his Doctorat d’Etat
`
es Science degree from Paul Sabatier Uni-
versity Toulouse (France) in 1987. He is currently head of Sensors Group at
Laboratoire Mat
´
eriaux & Micro
´
electronique (L2MP-CNRS) Marseille (France).
His principal research interests are directed towards microsystems, gas sensors
and selectivity enhancement strategies including multivariable analysis, noise
spectroscopy and modelling of sensor responses.