Sensors and Actuators B 101 (2004) 39–46
An electronic nose based on solid state sensor arrays for
low-cost indoor air quality monitoring applications
S. Zampolli∗ , I. Elmi, F. Ahmed1 , M. Passini, G.C. Cardinali, S. Nicoletti, L. Dori
CNR-IMM Sezione di Bologna, Via P. Gobetti 101, 40129 Bologna, Italy
Received 9 October 2003; received in revised form 9 October 2003; accepted 11 February 2004
Available online 28 March 2004
Abstract
The occurrence of illnesses related with poor ventilation has driven an increasing attention towards indoor air quality monitoring. In
buildings equipped with climate control systems, the diseases related to the air quality can be significantly reduced if smart intervening
procedures, aiming to control the concentration of pollutants in the indoor air, can be implemented in the heating, ventilation air conditioning
unit. When reliable information about both the indoor and outdoor air quality is made available, the climate control system can provide
the most appropriate amount of ventilation, ensuring safe and comfortable living conditions.
In this paper, a dedicated, miniaturized, low-cost electronic nose based on state-of-the-art metal oxide sensors and signal processing
techniques was developed. The proposed device is targeted to the quantification of carbon monoxide and nitrogen dioxide in mixtures with
relative humidity and volatile organic compounds by using an optimized gas sensor array and highly effective pattern recognition techniques.
The electronic nose was tested in an environment reproducing real operating conditions. Exploiting the unique response patterns of the
different sensors in the array and the capability of a simple fuzzy-logic system it was possible to identify and discriminate concentrations
as low as 20 ppb for NO2 and 5 ppm for CO in the test gas environment, allowing to reach the necessary sensitivity towards the target
pollutants together with the selectivity towards the typical interfering gas species.
© 2004 Elsevier B.V. All rights reserved.
Keywords: Indoor air quality; Gas sensors; Electronic nose; Fuzzy logic
1. Introduction
The increasing interest in indoor air quality (IAQ) monitoring is mainly due to the growing incidence of a new class
of diseases, identified as building-related illnesses (BRI) and
sick building syndromes (SBS), arising from long-term occupancy of confined living spaces, like office buildings or
homes/apartments. The occurrence of these diseases is related with the presence of physical, biological and/or chemical contaminants inside the building. The effects on the
human health due to indoor pollution include pathologies
for which the etiologic agent has been precisely identified
and pathologies not directly linked with a specific physical,
chemical or biological specie, since the symptoms arise from
the concomitant effect of several pollutants which are all
present in the ambient, e.g. from low air quality [1]. Chem-
∗ Corresponding author. Tel.: +39-051-639-9109;
fax: +39-051-639-9216.
E-mail address: zampolli@bo.imm.cnr.it (S. Zampolli).
1 On leave from: Department of Physics, Jahangirnagar University,
Savar, Dhaka 1342, Bangladesh.
0925-4005/$ – see front matter © 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.snb.2004.02.024
ical contaminants, and in particular gaseous compounds,
were found to be among the main responsible ones for both
the BRI and the SBS diseases, since in buildings with inadequate ventilation they can accumulate with time.
Typically, we can find several hundreds of indoor chemical contaminants, including by-products of the combustion
(NO2 , SO2 , CO, etc.), cigarette smoke, particulate matter,
mineral fibers and a number of volatile organic compounds.
In spite of the very low concentrations, some of these compounds are extremely toxic, like NO2 or CO; some other, like
benzene and formaldehyde, were proved to be carcinogenic.
Therefore, the monitoring of the air quality is of paramount
importance to keep safe and healthy conditions.
In buildings equipped with heating ventilation air conditioning (HVAC) systems, the pollutants are diluted by ventilation, which is generally operated on the basis of fixed
duty cycles. This approach does not necessarily ensure an
improvement of the indoor air quality, since in many cities
“fresh” outdoor air can contain many pollutant species at
concentrations higher than the threshold values. Furthermore, an uncontrolled preventive increase of the ventilation
results in an increment of the overall energy consumption,
40
S. Zampolli et al. / Sensors and Actuators B 101 (2004) 39–46
Table 1
Typical indoor air quality contaminants
Aldehydes
Ozone
Nitrogen dioxide
Carbon dioxide
Particulate matter
Radon
Carbon monoxide
Sulfur dioxide
Total VOC
Formaldehyde
Water vapor
Lead
especially when the temperature gap between indoor and
outdoor air is non-negligible [2].
Several attempts to implement demand-controlled ventilation have been reported in literature. Many of them rely on
the quantification of CO2 , used as a tracer of human occupancy in confined living spaces or as surrogate of inhabitant
generated pollution. However, this approach is inadequate
to monitor IAQ, since several toxic compounds released by
building materials and furnishings or generated by human
activities are also present in the indoor environment. Since
each compound has a different impact on human health,
it is important to monitor their concentrations individually.
Table 1 reports a list of the most common indoor contaminants found in confined living spaces, which can be used as
tracers for the IAQ [1,3,4].
For this reason, the availability of reliable, low-cost sensors suitable to monitor both the indoor and outdoor air
quality would allow the implementation of smart HVAC intervening procedures, considering not only the typical information about temperature and relative humidity but also the
concentration of a number of compounds used as air quality tracers. This way, by driving the ventilation systems on
demand to keep the IAQ under control, it would be possible
to maintain acceptable healthy and comfortable conditions
while minimizing the overall power consumption. Being the
information about the quality, the humidity and the temperature available for the inside and the outside air, the climate
control system would intake fresh air, recycle the indoor air
or operate active scrubbers, depending on the most favorable
energetic conditions.
Some low-cost tools for IAQ analyses are nowadays commercially available, but they are not suitable to provide re-
liable information, since the various pollutants are not precisely identified and quantified. These devices usually give
indications about the overall IAQ, without any estimation of
the concentration of each pollutant. In fact, only the detailed
information about the concentration of the single gas would
allow an optimized control of the HVAC system, including
the activation of scrubbers to catalytically convert some pollutants, like, e.g. volatile organic compounds, in CO2 and
water vapor.
In this work, an electronic nose (e-nose) based on a solid
state gas sensor array for the identification and the quantification of two typical indoor air quality tracers is presented. The
simplicity of the proposed approach, which uses stand-alone
gas sensors, simple driving electronics and fuzzy-logic pattern recognition algorithms, aims to realize a low-cost tool
suitable to monitor some of the compounds of interest for
IAQ. The availability of this type of devices is fundamental
for a capillary analysis of the pollution level inside buildings equipped with forced ventilation and air climate control units. To validate the approach, only CO and NO2
were taken into consideration as target compounds, while
some VOCs and the humidity were considered as interfering
species.
2. Experimental
The aim of this work is to develop a reliable tool for IAQ
monitoring using a very simple system architecture based
on a stand-alone gas sensor array, driving electronics and
suitable pattern recognition algorithms, without any fluidic
components (valves, filters, pumps, etc.). Furthermore, the
system has been designed to integrate additional sub-units
able to monitor other parameters relevant for IAQ. To validate this approach, the number of compounds considered
within this work was limited to two gases chosen among
those reliably detectable by metal oxide solid state gas sensors. Table 2 reports these gases together with the target concentration ranges considered for effective IAQ monitoring
and the relevant indoor threshold values as defined in [3]. In
spite of their relevance for indoor comfort, some other compounds, like carbon dioxide (CO2 ), were not considered here
because their detection is problematic with metal oxide gas
sensors. However, other sensing devices capable to detect
these compounds are nowadays commercially available and
they could be easily integrated into a modular architecture.
Table 2
Some gas species considered as IAQ tracers and their target concentration ranges
Compound
Concentration range
Indoor threshold values (8 h exposure)
CO
NO2
VOC (benzene, toluene, m-xylene)
Water vapor (RH)
5–30 ppm
20–200 ppb
60–600 ppb
5–95%
9 ppm
53 ppba
n.a.b
No threshold
a
b
US-EPA suggested value for 24 h exposure outdoor threshold. No indications given for indoor exposure.
No standards have been set for VOCs in non-industrial indoor settings.
S. Zampolli et al. / Sensors and Actuators B 101 (2004) 39–46
Fig. 1. Schematic overview of the considered e-nose architecture.
The system developed for this work consists of an array
of thin-film metal-oxide gas sensors, driven by ad hoc electronic circuits suitable to power and measure each device
independently. The gas sensors were operated under different conditions and they provided distinct response patterns.
A commercial RH sensor monitored the relative humidity,
which interferes in both CO and NO2 detection. The discrimination between the different target gases and their quantification were achieved using state-of-the-art data processing
and pattern recognition algorithms, which rely on the distinct response patterns provided by the sensors. In fact, the
low selectivity of the solid state gas sensors has been compensated by the combination of the response patterns, allowing the quantification of the single gas species. In this
work, a fuzzy-logic system was used for pattern recognition. A similar approach, based on a multi-sensor array and
using an artificial neural net (ANN) for pattern recognition,
has been proposed in [5].
A schematic drawing of the unit is shown in Fig. 1. This
simple architecture is targeted for the detection and quantification of CO and NO2 , but it does not allow a precise quantification of volatile organic compounds (VOC), because of
the lack of selectivity in metal oxide semiconductor gas sensors. We are currently addressing this problem following an
approach based on a gas-chromatographic architecture realized by MEMS technologies and developed for the detection
and quantification of VOC [6], which will be described in
a future paper. Both the dedicated gas-chromatographic architecture and the simple e-nose presented in this work are
conceived to be part of a modular system for innovative IAQ
monitoring applications.
2.1. Metal oxide resistive gas sensors
The solid state gas sensors used for the detection of the air
quality tracers are based on thin films of metal oxide semi-
41
conductor materials [7,8] deposited onto state-of-the-art micromachined hotplate arrays integrated in a single 5 mm ×
5 mm Si chip [9]. These hotplates consist of four suspended
200 nm thick Si3 N4 membranes integrating a buried platinum heater and passivated by a 800 nm thick SiO2 layer.
The very low thermal mass of the suspended structure allows
to reduce the power consumption with respect to commercial devices (only 60 mW to operate the sensor at 400 ◦ C),
as well as to heat up and cool down the sensing element very
fast, which is necessary for fast pulsed temperature (FPT)
mode sensor operation [10].
On top of these hotplates, we deposited two types of gas
sensing materials by means of different deposition techniques, namely tin oxide (SnO2 ) by rheotaxial growth and
thermal oxidation (RGTO) [7] and tungsten trioxide (WO3 )
by pulsed laser ablation (PLA) [8]. The morphology of the
thin films achieved through the optimization of the deposition techniques leads to very high sensitivity towards the
considered gas species, allowing the detection of the gas
concentrations down to the range required for IAQ.
The lack of selectivity of resistive gas sensors was partially compensated by the deposition of catalyst materials as
well as very thin SiO2 caps above the sensing layer. The response towards the single gas species was further enhanced
through the use of different operating temperatures and by
operating the sensors in FPT mode [10]. Table 3 shows
some of the optimal combinations of materials and operating conditions and the corresponding target gas species. The
optimization of both the sensing material and its operating
conditions results in excellent sensor responses towards CO
and NO2 . On the other hand, the low selectivity of metal oxide semiconductor gas sensors was only partially overcome
by the material optimization, since interference phenomena
by water vapor and VOC still affected the detection, as will
be discussed later.
2.2. Characterization protocol
The characterization was performed with a specifically
designed characterization system, where up to eight pollutants could be simultaneously injected into a Pyrex measurement chamber. The large inner volume (32 l) of the chamber
and the applied flow rate (500 sccm) were chosen to replicate as close as possible the operating conditions found in
a room equipped with a HVAC system, where every hour
a complete room volume of air is pumped-in. During the
test, mixtures of all pollutants at three different concentrations were injected into the characterization chamber. The
Table 3
Sensing materials and operating conditions used to detect the different gas species
Sensing layer material
Catalyst
Operation mode
Operating temperature (◦ C)
Target gas species
SnO2
WO3
SnO2 + SiO2
SnO2
Au
Au
Au
–
FPT
DC
DC
FPT
375
250
400
325
NO2
NO2
CO
CO
42
S. Zampolli et al. / Sensors and Actuators B 101 (2004) 39–46
gas distribution system delivers synthetic air, humidified air
and the target pollutants from certified gas cylinders in balanced quantities to get the final concentration of each compound. The system is completely controlled by a PC running
a specifically developed software package, which allows the
recursive execution of automatic sequences lasting several
weeks. The sensor arrays are driven and controlled by a
custom-made electronic interface connected to a further PC,
which allows the simultaneous and completely independent
acquisition of up to 16 sensors. The sensing layer resistivity,
i.e. the sensor response, was sampled by applying a 1.2 V
bias to the sensing layer and measuring the flowing current.
Finally, a standard GC/MS is connected to the measurement
chamber, performing spot-wise quality control tests of the
gas mixture inside the chamber.
The set of data used for the analysis presented in this work
consisted of 2000 patterns acquired within the last 45 days
of a 60-day sensor characterization run. The data relative to
the first 15 days were neglected, since the sensors used in
this characterization were brand-new and had undergone no
stabilizing burn-in procedures, thus showing a considerable
drift of the conductivity during the first operation period.
2.3. Pattern recognition techniques
The characterization data specifically collected during
long-term measurement campaigns lasting more than 2
months, were then used to train and test a fuzzy system,
which provided an estimation of the initial concentration of
each target compound in the gas mixture.
Operating the sensors under optimized conditions for
the detection of a given compound does not make them
fully selective, but they do provide unique response patterns, which can be used to classify and quantify the single
pollutants inside complex gas mixtures, as found in real
applications. To evaluate concentration of each compound
in a complex gas mixture, we used a neuro-fuzzy system
provided by the adaptive fuzzy modeler (AFM) developed
by ST-Microelectronics [11]. AFM is a software tool for the
automatic fuzzy model generation of a system starting from
its input–output data sampling. It implements neural training
algorithms for the successive optimization of both the rules
and the sets of a simple Sugeno-type zero-order fuzzy system. The automatic rules identification is carried out, thanks
to an unsupervised learning on a winner-take-all fuzzy associative memory neural network, while the optimization
of the fuzzy sets parameters is carried out by a supervised
learning on a multilayer backward-propagation fuzzy associative memory neural network [11]. As fuzzy intersection
operators (inference methods), AFM allows to select either
the minimum or the product operator. The AFM package
is supplied for feasibility studies of ST-Microelectronics’
fuzzy-logic ASICS (W.A.R.P. 1 + 2, ST52x301), which implement all the processing capabilities of the fuzzy systems
simulated in the AFM. Therefore, the processing electronics
necessary for the realization of the e-noses studied in this
work are already commercially available. Further details on
fuzzy logic can be found in [12].
The AFM was used to simulate two fuzzy systems having
three input variables and one output prediction each, being
the input variables the output from the CO, NO2 and RH
sensors and the output predictions, the CO and the NO2
concentrations, respectively. In the case of the CO sensor,
five Gaussian fuzzy sets were used for the CO input and
three Gaussian sets for each of the interfering NO2 and RH
inputs. In the case of the NO2 sensor, five Gaussian fuzzy
sets were used for the NO2 input and three Gaussian sets for
each of the interfering CO and RH inputs. The system was
trained with the product inference operator for 50 learning
epochs and converged rapidly.
3. Results and discussion
3.1. Response of the gas sensor arrays
The optimization of the sensing layer materials and of
the operating conditions has allowed to reach the sensitivity
necessary for the reliable detection of CO down to concentrations of 5 ppm and of NO2 down to 20 ppb. To detect CO,
we used a SnO2 + SiO2 sensor with Au catalyst operated in
DC mode at 400 ◦ C, while the sensor targeted for NO2 was
a SnO2 sensor with Au catalyst operated in FPT mode with
a pulse temperature of 375 ◦ C, a duration of 1 s, and a period of 60 s. Other sensors were used during the extensive
characterization run, principally FPT mode operated SnO2
sensors for CO detection and WO3 -based sensors for NO2
detection, but the validation of the sensor response patterns
during the training of the fuzzy system have shown that these
combinations are less effective, and therefore only the sensor
responses of the first two sensor types will be considered.
Fig. 2a and b shows the typical response of two sensors
targeted for the detection of CO and NO2 to three different
concentrations of these pollutants. First, the sensor baseline
in synthetic air with RH = 50% is shown, up to T = 0 min.
Afterwards, three different concentrations of CO or NO2 are
consecutively injected and kept each one for 15 min. In this
case, the considered concentrations were 5, 15 and 30 ppm
for CO and 20, 100 and 200 ppb for NO2 . The plots of Fig. 2
emphasize the excellent response of the sensors. For the CO
sensor, the increase of the conductivity for the first concentration of 5 ppm amounts to approximately 400%, while
the NO2 sensor shows the typical effect of an oxidizing gas
specie on n-type semiconducting materials, with a variation
of the resistivity of approximately 300% for the 20 ppb concentration. While the CO sensor shows a good dynamic response over the entire concentration range, the NO2 sensor
saturates at the higher concentrations ([NO2 ] > 100 ppb).
However, in the range of the indoor threshold value [NO2 ] =
53 ppb, the sensor features good discrimination capabilities.
From these results it is important to evidence that, having
response variations three to four times the baseline value,
S. Zampolli et al. / Sensors and Actuators B 101 (2004) 39–46
43
Fig. 2. Sensing layer current at the different target gas concentrations for
the CO sensor (a) and the NO2 sensor (b) at RH = 50%.
Fig. 3. Sensing layer current evolution over 45 days of characterization
at the different concentrations of the target gases in mixtures with all
interfering species, for the CO sensor (a) and the NO2 sensor (b).
the detection limit of our sensors for both CO and NO2 is
well below the lowest concentration considered in this work,
ensuring a high degree of confidence for the concentration
predicted by the fuzzy system.
The plots in Fig. 2 do not give information about the
selectivity of the sensors, since the responses shown are
relative to the injections of a single pollutant at fixed RH
values. Furthermore, only a single measurement shift is
shown, and the possible drift of the baseline with time cannot be disclosed. To evaluate the overall sensor selectivity
and the evolution of the device response with time, we compared the conductance of the sensing layer element of both
the CO and NO2 sensors for the different concentrations of
the target pollutant with and without the other gases in the
mixtures. The results for a 45-day measurement period are
shown in Fig. 3a and b. More in detail, Fig. 3a shows the
currents of the CO sensor when 0, 5, 15 and 30 ppm of CO
are present in the measurement chamber, in any mixture
with NO2 and VOC at different concentrations. For simplicity, only the measurements acquired with RH = 30% are
shown. Conversely, Fig. 3b shows the currents of the NO2
sensor when 0, 20, 100 and 200 ppb of NO2 are injected with
RH = 30%, in any mixture with CO and VOC. From Fig. 3
it can be pointed out that, during 45 days of operation, the
sensors exhibit a good sensitivity towards the considered gas
species together with a certain sensor drift with time. Fig. 3
also show a spread of the single measurements, which is re-
lated with some cross-response to the other interfering compounds present in the gas mixture. A detailed analysis of the
sensor response shows that especially the relative humidity
and the NO2 concentration interfere in the CO detection,
while the NO2 sensor is affected primarily by interference
from humidity, only. As far as the VOCs are concerned, only
very small interference on the CO sensor can be observed,
but this cross-sensitivity can be neglected if compared to
the excellent sensor response towards CO. In fact, the optimized conditions for VOC detection were found to be quite
different from those used for CO and NO2 detection.
These results imply that the selectivity exhibited by our
sensors is still not enough to allow for a direct use of the
sensor output for the quantification of the gas concentration.
However, being the information acquired by a commercial
RH sensor included as input to the fuzzy system, the combination of the input patterns from the three sensors should
allow for the extraction of the concentrations of the two target gases, CO and NO2 .
Furthermore, if we consider the sensor response, defined
as (GGAS − G0 )/G0 or (RGAS − R0 )/R0 for oxidizing and
reducing gas species, respectively, this parameter is quite
stable for both the CO and the NO2 sensors, but the overall
sensor conductivity sensibly decreases with time. In a real
application, direct information about the baseline conductivity is not available unless a calibration or an air purifier
unit has been integrated into the system. If the target is a
44
S. Zampolli et al. / Sensors and Actuators B 101 (2004) 39–46
Fig. 4. Pre-processed sensor response spread as a function of the target gas
concentration, prior to fuzzy-logic pattern recognition, for the CO sensor
(a) and the NO2 sensor (b). The error bars are relative to the 2σ interval.
low-cost e-nose, the drift of the sensor conductivity with
time has to be analytically compensated before supplying
the sensor data to the fuzzy pattern recognition system, and
this is easily possible due to the smooth exponential decay
of the conductivity with the time, which can be disclosed
from Fig. 3.
However, the baseline drift compensation remains a critical issue, since the nature of the conductivity decay could
not be completely explained yet. In this concern, investigations about the nature of the drift are in progress, and first
results show a good reproducibility of the conductivity decay shape.
Fig. 4a and b resumes the quantification capability of the
single gas sensors, prior to the application of the fuzzy-logic
pattern recognition algorithms. The plots show the raw characteristics of the single sensor’s currents as a function of the
target gas concentration, for all the measurements acquired
within the 45 days of characterization. Fig. 4a is relative to
the CO sensor, Fig. 4b to the NO2 sensor, and the data is
already being compensated for the exponential drift in the
sensor current.
From the plots in Fig. 4, one can see that the sensors are
partially selective and show a clear trend of the sensing layer
current with the target gas concentration. However, the simple response of the single sensors is not sufficient for the precise quantification of the pollutants. The error bars reported
Fig. 5. Prediction of the target gas concentrations as a result of the multisensor response pattern combination performed by the fuzzy-logic algorithm: CO prediction (a) and NO2 prediction (b). The error bars are
relative to the 2σ interval.
in the plots represent the two standard deviation (2σ) interval of all the sensor output values at a fixed concentration.
As shown in the plots, the error bars relative to the single
concentrations overlap with the neighboring concentrations,
due to the cross-response to the interfering gas species. For
this reason, a pattern recognition technique must be applied,
in order to combine the responses of the different sensors,
allowing the prediction of the final target gas concentrations.
3.2. Fuzzy pattern recognition algorithm
The plots in Fig. 5a and b show the results of the prediction for both the CO and the NO2 concentrations obtained
applying the two fuzzy systems to the collected set of data.
From the 2000 response patterns, 1400 were randomly extracted and used for the training of the fuzzy system and the
remaining 600 were used for the validation of the recognition algorithm. The error bars are relative to the 2σ interval
of the various predictions for each concentration.
As can be seen from these figures, considering the simplicity of the proposed approach, the estimated values of
the concentrations are in excellent agreement with the real
pollutant concentrations injected into the mixture. Using
this approach, the three concentrations for CO can be easily discriminated, while for NO2 only the highest concentrations may overlap. This behavior can be explained by
S. Zampolli et al. / Sensors and Actuators B 101 (2004) 39–46
considering Fig. 2b, which shows the excellent sensor response towards the low concentrations and the significant
saturation effects between the 100 and 200 ppb concentrations. This means that the NO2 sensor is that sensitive
to already saturate at these sub-ppm concentrations. This
problem can be easily overcome by applying slightly different operating conditions for the NO2 sensor, such as a
higher operating temperature. This way the sensitivity at low
concentrations decreases, but the slope of the response at
concentrations higher then hundred ppb become more pronounced, allowing for better separation and discrimination
between each concentration. However, in our application we
focused more to the range below 100 ppb, where concentrations of NO2 as low as 20 ppb can be very well identified and
separated.
4. Conclusions
In this work, a simple e-nose architecture for indoor CO
and NO2 monitoring is presented. This instrument belongs
to a new class of low-cost e-noses for air quality monitoring to be integrated within the climate control unit, which
would allow a more efficient use of HVAC systems, keeping the IAQ under control while lowering the overall power
consumption.
The e-nose is based on an array of solid state metal oxide semiconductor gas sensors realized onto micromachined
hotplate arrays. The use of a micromachined substrate, the
optimization of the deposition techniques as well as the use
of the most appropriate operating parameters, allowed for
the realization of low-cost, low power consumption devices
with high sensitivity and enhanced selectivity.
These devices were tested in an environment reproducing real operating conditions and concentrations as low as
20 ppb of NO2 and 5 ppm of CO were continuously monitored for more than 45 days of extensive sensor characterization, which showed no significant degradation in the sensor
response. The target gas concentrations in the mixtures have
been precisely estimated exploiting the capability of a simple fuzzy-logic system used to combine the unique response
patterns of the various sensing elements and to extract the
most significant information from the sensor data. Using this
approach, it was possible to identify and discriminate the
presence of each pollutant and to estimate the composition
of the air mixture inside the test environment. The presented
results show the feasibility of low-cost e-noses developed to
detect CO and NO2 at concentrations lower than the IAQ
threshold values, suitable for the integration in HVAC systems.
45
and by the European Social Fund, Ministero del Lavoro e
delle Politiche Sociali, Consorzio Spinner, Regione Emilia
Romagna. F. Ahmed is at CNR-IMM Sezione di Bologna
under the “Abdus Salam ICTP Training and Research in
Italian Laboratories” program. The authors would like to
acknowledge the Regione Emilia Romagna, Assessorato
Territorio, Programmazione e Ambiente, Servizio Analisi
e Pianificazione Ambientale for the financial support. The
technical support of S. Guerri, P. Negrini, G. Pizzocchero,
M. Sanmartin and F. Tamarri for sample preparation is also
gratefully acknowledged.
References
[1] Air quality guidelines, World Health Organization. http://www.who.
int/.
[2] W.J. Fisk, A.T. De Almeida, Sensor-based demand-controlled ventilation: a review, Elsevier Energy Build. 29 (1998) 35–45.
[3] ASHRAE, Standard 62-1999, Ventilation for acceptable indoor
air quality, American Society of Heating, Refrigeration and Airconditioning Engineers, Atlanta, 1999. http://www.ashrae.org/.
[4] An Introduction to Indoor Air Quality, US Environmental Protection
Agency. http://www.epa.gov/iaq/ia-intro.html.
[5] H.E. Endres, W. Göttler, R. Hartinger, S. Drost, et al., A thin-film
SnO2 sensor system for simultaneous detection of CO and NO2 with
neural signal evaluation, Sens. Actuators B: Chem. 35–36 (1996)
353–357.
[6] J. Stürmann, S. Junge, W. Benecke, S. Zampolli, L. Dori, S. Nicoletti,
Application of a membrane reactor and an ultra-short packed gas
chromatographic column to optimize the gas selectivity for a resistive
thin film gas sensor, in: Proceedings of the 16th IEEE International
MEMS Conference, Kyoto, Japan, January 19–23, 2003.
[7] G. Sberveglieri, G. Faglia, S. Groppelli, P. Nelli, A. Camanzi, A
new technique for growing large surface area SnO2 thin film (RGTO
technique), Semicond. Sci. Technol. 5 (1990) 1231–1233.
[8] S. Nicoletti, S. Zampolli, I. Elmi, L. Dori, M. Severi, Use of different
sensing materials and deposition techniques for thin-film sensors to
increase sensitivity and selectivity, IEEE Sens. J. 3 (4) (2003) 454–
459.
[9] L. Dori, P. Maccagnani, G.C. Cardinali, M. Fiorini, I. Sayago,
S. Guerrri, R. Rizzoli, G. Sberveglieri, New materials and heater
geometry for high performance micromachined thermally insulated
structures in gas sensors applications, in: Proceedings of the EUROSENSORS XI Conference, Warsaw, Poland, September 21–24,
1997
[10] I. Elmi, S. Nicoletti, G.C. Cardinali, L. Dori, S. Zampolli, et al.,
Fast pulsed temperature mode operation: a viable way to improve
the sensitivity and the selectivity in metal oxide gas sensors, in:
Proceedings of the 14th European Conference on Solid-state Transducers, Copenhagen, Denmark, August 27–30, 2000.
[11] For details about AFM, see ST-Microelectronics Web: http://eu.st.
com/stonline/books/pdf/docs/6087.pdf.
[12] A. Zadeh, in: L.A. Zadeh, Da Ruan, C. Huang (Eds.), Fuzzy Sets and
Fuzzy Information-granulation Theory: Key Selected Papers, Beijing
Normal University Press, Beijing, 2000.
Biographies
Acknowledgements
This work was partially supported by the European Community FP5, Clean Air Project (No. NNE5-1999-00415)
S. Zampolli graduated in physics at the University of Bologna in 2000,
discussing a thesis on the application of Fuzzy Logic for detection of
buried landmines. He is a grant student at CNR-IMM Sezione di Bologna,
since April 2000, working primarily on gas sensor characterization and
data processing within the sensor and microsystem R&D program.
46
S. Zampolli et al. / Sensors and Actuators B 101 (2004) 39–46
I. Elmi graduated in physics at the University of Bologna in 1998,
discussing a thesis on characterization of Sensors for Benzene detection.
Since April 1999 he is working as grant student at CNR-IMM Sezione
di Bologna within the sensor and microsystem R&D program, primarily
developing a system for environmental monitoring and characterizing gas
sensors.
F. Ahmed received MSc degree in physics from Jahangirnagar University,
Bangladesh, in 1990, and the PhD in electrical engineering from the University of Osaka, Japan, in 1998. He is currently an Associate Professor
of Physics Department, Jahangirnagar University, Bangladesh. His current research interests are the structural and electrical characterization of
superconducting and semiconducting materials. He worked for 2 years at
the CNR-IMM in Bologna, Italy, under the Abdus Salam ICTP Training
and Research in Italian Laboratories programme.
M. Passini is part of the technical staff of CNR-IMM Sezione di Bologna
since 2002. Her primary activities include micromachined hotplate array
fabrication and GC/MS analyses.
G.C. Cardinali received a degree in electronic engineering from the
University of Bologna, Italy, in 1979. Since 1982 he has been work-
ing at the CNR-IMM Institute. His scientific interests are in the areas
of design, fabrication, and testing of electronic devices and microsystems. From 1996 onwards he has been involved in research projects
dealing with the implementation of systems for air quality monitoring based on micro-gas sensors. He is also dealing with the development of dedicated microelectronic processes suitable for integration into
microsystems.
S. Nicoletti received his PhD in physics from the Universitè J. Fourier
of Grenoble (France) in 1996 discussing a thesis on HTC Josephson junction devices. Since 1996 he is working at CNR-IMM Sezione
di Bologna (formerly LAMEL Institute) on the fabrication and the
characterization of gas sensors solid state devices, microsensors and
microsystems.
L. Dori joined the CNR-IMM Institute in 1969. In 1980 he received
the Master in Physics at the University of Bologna. With the IBM visiting program, he was 4 years at the IBM-Thomas J. Watson Research
Laboratory in Yorktown Heights (NY, USA) working on gate dielectric structures of DRAM or Flash memory devices. At the present time
he is working on sensor and microsystem R&D program at CNR-IMM
Institute.