Introduction
Very few of the gas sensors currently on the market or predicted to enter the
gas sensing market soon are operable without additional heating elements. A
well-known exception with widespread application is the electrochemical
sensing principle. In contrast, solid-state devices whose detection principle
relies mostly on catalysed molecule–surface reactions with specific target
gases are typically operated with an internal heating element, capable of
heating the sensor up to several hundred ∘C. Figure 1a–e show
schematics of different selected types of sensors, including semiconducting
sensors, pellistors, field-effect devices, Schottky diodes and capacitance
sensors. The number of sensing technologies here is limited, as a high
technology readiness level is taken as a prerequisite. The devices shown in
Fig. 1a–d operate at elevated sensing surface temperatures. One device
capable of non-heated operation is the capacitance-type humidity sensor. It
is based on absorption of water molecules within a polymeric network and the
resulting changes to the dielectric coefficient (Farahani et al., 2014). The
liquid-state electrochemical sensing principle, well known for
decades (Bakker and Telting-Diaz, 2002) and typically applied in sensors for
portable applications, is sketched in Fig. 1f for comparison. Such devices
are available for a broad variety of electrochemically accessible molecules,
such as carbon monoxide. In this case, CO is oxidized at the working
electrode (WE, i.e. sensing electrode) which is biased at low or zero
voltages with respect to the reference electrode and is typically composed of
graphite and a catalytically active metal. The half-cell reactions include
oxidation (CO+H2O⟶WECO2+2H++2e-) and reduction at the counter
electrode CE (12O2+2H++2e-⟶CEH2O) and hence involve
H2O and H+ ions. It is thus contrasted with the high-temperature
CO-to-CO2 oxidation that occurs at the surfaces of solid-state sensors.
They involve other elementary reaction steps (Bârsan and Weimar, 2002).
For electrochemical sensors, the enclosure and outer and internal membranes,
chosen with respect to permeation properties for CO, O2 and H2O,
maintain the H+ conducting phase aqueously for the lifetime. This is
evidently unnecessary and impossible for heated solid-state sensors.
Schematic and comparative representation of various sensor
configurations. FET, MOS, ECS and PEO denote field-effect transistor,
metal-oxide semiconductor, electrochemical sensor and plasma-electrolytic
oxidation, respectively.
Our conceptual guideline for selecting appropriate materials for the
non-heated solid-state device shown in Fig. 1g was to combine an oxide with
H+ ion surface mobility and water absorption capability with
catalytically active metals, dispersed in the graphite top electrode. In this
device, both the sensitive electrode and the oxide are accessible to gases.
For this purpose, we have developed a sensor preparation technique based on
plasma electrolytic oxidation of titanium (Schierbaum and El Achhab, 2011)
and a printing method for the top, catalytically active electrode, which
consists of graphite and nanosized platinum. Thus, the sensor element has a
platinum–graphite/TiO2/Ti solid-state structure.
In this study we demonstrate that platinum–graphite/TiO2/Ti devices are
capable of sensing CO and H2 at room temperature without the need for
permanently operated heaters. In a first set of experiments, we show that
graphite–platinum layers show sufficient catalytic activity to enable
hydrogen sensing with calorimetric devices. Second, we report on
room-temperature humidity sensors, based on graphite/TiO2/Ti sensors,
which interact with water vapour reversibly and on a surprisingly short
timescale. Here, an impedance measurement is used to gain the sensor signal
from the graphite/TiO2/Ti device. Third, we apply the amperometic
principle (i.e. a current measurement at a constant voltage between the Pt-C
top electrode and the Ti substrate) to demonstrate CO detection in humid air
at room temperature with such devices.
Experiments
We used titanium foils (99.6 %, 125 µm thick, Chempur) and a
laser cutting process to yield frames with small pieces of Ti
(3.5 × 3.5 mm2) as shown in Fig. 2a. For batch fabrication of
sensors, we applied an automatized plasma-electrolytic oxidation by which the
titanium oxide is formed under potential control and current
limitation (Schierbaum and El Achhab, 2011). The process led to excellent
adhesion and stability of the oxide layer. For the present work the current
limit was 200 mA. Two different maximum voltages were applied, 150 V for
highly porous oxide layers and 120 V for less porous oxide layers. Details
of the preparation procedure and resulting microstructures are given in El
Achhab et al. (2014). Figure 2b shows the frame after
oxidation. The upper electrodes were produced with a special printing
technology (Hermetic Universal 6–12, Tampoprint), which uses refined
catalytically active inks. Figure 2c shows the imprinted frame. For humidity
sensors, the ink consists of pure graphite. For H2 and CO sensors, we
add Pt paste (Chempur 900487) to achieve a low at. % level of platinum. The single sensors are obtained from the
frame by cutting the small holding strips with a sharp blade. By doing so the
oxide layer is removed locally from the edges and the Ti substrate is
electrically accessible for the following bonding process. Electrical
connection between substrates as well as upper electrode and socket pins or
PCB pads was achieved with an epoxy-die bond technology (Westbond 7200CR
system D) with silver epoxy glue (Loctite 3880). Figure 2d shows the final
sensor, mounted on a TO39 socket (Schott Singapore, Electronic Packaging).
Key steps in the sensor fabrication process: (a) the
titanium frame with 35 samples structured using a laser cutter,
(b) after plasma-electrolytic oxidation of a titanium foil,
(c) after printing the top electrodes and (d) after
assembly of a single specimen, cut out from the processed frame, on a
TO39-socket via dye-bonding. For calorimetric devices, the Pt–C/TiO2/Ti
sample is glued to the active thermistor of the Wheatstone configuration.
Figure 3 displays key parts of the test chamber, sensor and device
electronics. Parts exposed to test gases were fabricated by means of
stereolithography (Accura XTREME, 3-D Systems), all other parts by laser
sintering. The electronic circuits and their layouts were developed in our
department. The printed circuit boards were fabricated externally in
two-layer technology with 6 mil minimum gap size.
Test chamber with 1/4′′ tube gas inlet and outlet, sealing parts
and device electronics with sensors mounted on a TO39 socket. The blue arrow
indicates the RJ11 socket for one-wire bus connection. For internal
communication, an I2C bus was used.
A homemade LabVIEW code (LV 2014, National Instruments) was used for device
control and data collection. A DS9490 USB adapter (Maxim Integrated) provided
the connection between the one-wire network and the computer (Windows 7,
64-bit), making use of the TMEX application programming interface and
corresponding drivers and libraries (see Application Note 1740, Maxim
Integrated). The communication between the one-wire bus and the device
electronics was established via a one-wire switch (DS2413, Maxim Integrated),
which formed the SDA and SCL lines of the I2C bus, or – in the later
stages of this project – via a one-wire slave-to-I2C master bridge
(DS28E17, Maxim Integrated). They transmit and receive the I2C signals
to and from the different components such as I/O port drivers, amplifiers,
16-bit ΔΣ analogue-to-digital converters (ADC), digital
rheostats and potentiometers and various analogue front-end parts. For
amplifiers, converters, reference voltage and current sources, integrated
circuits with excellent thermal drift behaviour and ultra-low noise were
chosen in tiny packages. EEPROMs were implemented to store adjustments of
devices and calibration data. The one-wire bus, in which all I2C
components can also be addressed individually via the unique 64-bit number
stored in the one-wire-to-I2C bridge, runs at normal speed (100 kHz) or
overdrive speed (400 kHz for the DS2413, in which the SDA data and SCL clock
signals are created at its two I/O outputs, i.e. by “bit banging”: Janssen,
2009). For calorimetry, a Wheatstone bridge arrangement, comprising
thermistors (Murata NCP18WB473-D03RB), precision resistors and a 1024-stage
digital rheostat for bridge balancing, was used in connection with a
dual-precision voltage source and a 16-bit ΔΣ-ADC with
programmable gain G (1 V V-1 to 256 V V-1) and
reference input, set at half the level of the bridge excitation voltage
Vbridge. A full differential amplifier stage was used so the
converter could swing from the negative to positive tail within an input
voltage range determined by the programmed gain. For amperometry, a
programmable voltage source (-9.75 to +9.75 V with 8-bit resolution) and
a high-side current amplifier with a zero-drift instrumentation amplifier
with selectable gain (PGA 281, Texas Instruments) was applied. For impedance
measurements, a 12-bit impedance converter (AD 5933, Analog Devices) was used
together with a constant power source, set up with a power and current
monitor (MAX 4210, Maxim Integrated) and controlled with a 10-bit
digital-to-analogue converter. This circuit can maintain the sensor
temperature slightly above the water condensation point.
Results and discussion
Microstructure of the upper electrode
Studies with electron microscopy (SEM), energy dispersive X-ray spectroscopy
(EDX), spatially resolved Raman X-ray diffraction and photoemission
spectroscopy were previously performed to determine the relevant details of
the microstructure of the titanium oxide layer (El Achhab et
al., 2014). In summary, the microstructure was
confirmed as consisting of a highly porous oxide which facilitates
penetration of the graphite–platinum ink and explains the good adhesion of
the top electrode.
Our EDX and SEM studies revealed that the nanosized Pt particles tend to
agglomerate in the graphite matrices. Figure 4 displays typical SEM images
at different magnifications. The bright features consist of Pt, as confirmed
with EDX. The high-resolution image shows that even the smallest Pt
particles seem to consist of very small crystallites, thus creating a rough
surface. The agglomerates are homogeneously distributed over the plate-like
graphite matrix. Such a distribution requires a careful mixing procedure and
ultrasonic treatment of the ink prior to application.
Scanning electron microscope image of the Pt–C top electrode at
magnifications of (a) 1000 and (b) 50 000. The insets show the C (red) and Pt (green) elemental
distributions of the central Pt agglomerate, as determined with EDX.
The high-resolution image of a single agglomerate of nanosized Pt particles
in Fig. 4b suggests good electrical interconnectivity between Pt and the
surrounding graphite plates. Since the upper electrode also exhibits
porosity, a large three-phase boundary is formed which is accessible to the
ambient gas and the sensed molecules.
Calorimetric detection of H2 with platinum–graphite/TiO2/Ti
sensors
For our first example, we chose a hydrogen sensor that uses catalytic
turnover and the generated heat of oxidation (Cakabay et al., 2015). This
detection principle has been known for a long time (Peinecke and Mohr, 1999)
and is e.g. used in pellistors (Krawczyk and Namiesnik, 2003) and recently in
calorimetric thermoelectric gas sensors (Park et al., 2014). These heated
devices operate at elevated temperatures above 200 ∘C. In contrast,
the hydrogen sensor presented in this study does not require permanent
heating and operates at ambient temperature. Maintaining the optimum activity
of the Pt catalyst is essential to avoid strong heating of the sensor when
exposed to LEL volume fractions of hydrogen in air. To achieve this,
5 at. % Pt content was chosen for the graphite ink. In Cakabay et
al. (2015) we studied the catalytic properties of the Pt particles, used in
the ink, by means of an isothermally operated micro-calorimeter in a wide
temperature range between 1 and 157 ∘C. We showed that the overall
reaction rate below approximately 30 ∘C is determined by the
activation energy of the H2-to-water oxidation over platinum, whereas
pore and film diffusion is rate-limiting above this temperature. To make use
of the calorimetric operation, a Wheatstone bridge arrangement with two
thermistors (R0=47 kΩ at T0=25 ∘C, tolerance
0.5 %) was used in this study. Unlike microcalorimetry, the bridge
arrangement is, however, operated without temperature regulation. The sensor
foil was assembled on top of an NTC thermistor by means of a heat-conducting
adhesive (TBS20S, Electrolube), while the second one remained as a reference
unaffected by hydrogen exposure. We used a symmetrical Wheatstone bridge
arrangement for our experiments with precision resistors and a 10-bit digital
rheostat with 20 kΩ. The arrangement is shown in Fig. 5 along with
calculated voltages Vbal as a function of the rheostat resistance
RRheo.
Vbal=R4R1+R4-R3+RRheo-1+R5-1-1R3+RRheo-1+R5-1-1+R2⋅Vbridge
In our case the bridge voltage Vbridge equals 4.096 V and is
generated with a precision reference source. The temperature dependence of
the thermistors' resistances is given by R2=R1=R0⋅expBT-1-T0-1 with B=4030 K-1 ± 0.5 % at T0 (Steinhart and Hart, 1968). The
curves intercept at zero out-of-balance voltage. Hence, the lowest possible
influence of ambient temperature changes is expected for a fully balanced
bridge. Experimental data are shown for 21 ∘C operation of the
bridge; the small difference voltage ΔVbal results from
additional voltage offset contributions in the electronics. Hence, the
rheostat must be adjusted at a slightly different offset than calculated
(compare Vbalexp and Vbaltheor
in Fig. 5). The deviation may have different physical origins. Fabrication
tolerances of the thermistors lead to differences in their resistances and
B values. Similarly, thermovoltages produced in various components and
soldering joints of the PCB may contribute as well. Since all these
contributions are difficult to localize and compensate for (e.g. by using
split resistors to reduce thermovoltage effects) under the constraints of a
manageable design effort, we used the temperature sensor implemented in the
ΔΣ-ADC and re-adjusted the rheostat in the Wheatstone bridge
through software control periodically. Thus, influences of ambient
temperature changes can be further reduced.
Calculated out-of-balance voltage Vbal as a function of
the resistance of the 10-bit digital rheostat (as given by the expression
RRheo=Offset×20kΩ/210) for
different temperatures. R1 and R2 denote the active and reference
thermistor, respectively. Experimental data of Vbal vs. Offset
at 21 ∘C are represented by dots and indicate a constant difference
ΔVbal with respect to the theoretical curve (black).
Further details are given in the text.
Hydrogen exposure and the resulting heat of oxidation, produced at the sensor
foil, led to temperature differences between the active and reference
thermistors. This in turns yields a pronounced out-of-balance voltage. The
ΔΣ analogue-to-digital converter used to determine the
temperature difference features a special zero-baseline drift technology that
guarantees a constant baseline over time. Because water vapour produces no
thermal effect at the catalyst, the baseline is also stable upon changes in
humidity. To reduce the inherent thermal cross-sensitivity of such a
thermistor-based sensor design, appropriate shielding against airflow and
infrared radiation is, however, essential.
Figure 6 displays a typical transient of the out-of-balance voltage at
23 ∘C, derived at an amplification of 8 V V-1 of the ΔΣ-ADC. The curve shows the response towards H2 exposures in a
range from 0.5 to 1.5 vol %, i.e. well below the low-explosive limit of
4.0 vol % for hydrogen in air in dry and moist air of increasing
humidity (10 to 50 % RH). Short rise and decay times within seconds are
achievable. Sensitivity becomes smaller at increasing humidity mainly because
of the reduced hydrogen-to-water oxidation rate. Long-term exposure
experiments showed that steady-state signals are maintained. They resulted
from stationary temperature differences between the active and reference
thermistor, given by the different heat source and sink contributions of our
setup. Up-to-the-air wiring of thermistors is found to provide larger
signals, but soldering on a printed circuit board is a technologically more
relevant approach for a practical implementation of the calorimetric
detection principle. Tuning the thermal decoupling of both thermistors while
maintaining their correlated dependency on ambient temperature changes and
the dynamics upon temperature changes are essential for future achievements.
In summary, the data in Fig. 6 confirm, however, the applicability of
room-temperature calorimetric operation at relevant concentrations of
hydrogen below LEL. Moreover, the calibration curve (out-of-balance voltage
vs. H2 volume fraction) follows a linear relationship in a range between
0.5 and 1.5 vol % at constant temperatures. Calibration curves, however,
depend on the ambient air temperature because the hydrogen oxidation is a
thermally activated reaction. Also, pore diffusion may come into play at
elevated temperatures (compare results for pure Pt in Cakabay et al., 2015).
Compensation of such effects could be achieved, in principle, by taking into
account data from the on-board temperature sensor in the ΔΣ-ADC.
Typical transient of signals (out-of-balance voltage
Vbal) for hydrogen detection in dry air with
Pt–graphite/TiO2/T sensors at 21 ∘C between 0.5 and
1.5 vol % (increase per step is 0.2 vol %) and at different
humidities from 0 RH to 50 % RH.
We also implemented an SMD resistor underneath the Ti foil to briefly heat
the sensor and thermally “refresh” the catalyst for different
applications. According to our practical tests in a laboratory context, the
typical period for such a treatment is 24 h with a duration of 5 min.
This procedure, which requires additional power of 500 mW from the USB port,
is controlled by software and tracked in the EEPROM.
Impedance metric detection of humidity with graphite/TiO2/Ti
sensors
Most resistive-type humidity sensors use interdigitated electrodes and a
humidity sensitive coating which is either made from electrolytic conductive
polymers such as salts and acids or a doped ceramic sensing film (compare
Fig. 1e). Typically, the resistance change follows an inverse exponential
relationship with humidity and varies between approximately 1 kΩ
and 100 MΩ between 20 and 90 RH % and response times in the
range of 10 to 30 s. The commonly applied configuration of capacitive
humidity sensors is a sandwiched structure with a dielectric polymer film
deposited between two electrodes. For a comprehensive review, see Farahani et
al. (2014).
PEO sensors with a pure graphite electrode were used as humidity sensors. In
contrast to the hydrogen sensor tests, which were performed in a gas flow
system, we used a completely different procedure for tests of humidity
effects on sensor signals. The sensor was quickly introduced between
Erlenmeyer flasks of 200 mL volume with constant relative humidity, adjusted
by using aqueous saturated salt solutions. We used NaOH (7 % RH),
Mg(NO3)2 ⋅ 6 H2O (33 % RH),
MgCl2 ⋅ 6 H2O (53 % RH), NaCl (75 % RH), and
K2SO4 (97 % RH) (Pennig, 2013). A rubber stopper, through
which the electrical connections ran, enclosed the interior of the vessel
from the surrounding atmosphere. It should be noted that the relative
humidity over the salt solution does only slightly vary with temperature,
while the water partial pressure shows the same dependency on temperature
than the absolute water saturation pressure does. Figure 7 shows a typical
result of frequency-dependent phase angles of the impedance, measured in a
frequency scan between 3 and 51.8 kHz and calculated on-line from the real
(Re) and imaginary (Im) data registers of the impedance converter chip. It is
implemented in the sensor electronics that has the same form factor and
digital communication as the hydrogen sensor. The converter has an integrated
digital synthesizer that produces a sine wave with an adjustable frequency
with a maximum peak-to-peak voltage of 2 Vp-p and a DC bias of 1.48 V.
Phase angles ϕ=tan-1Im/Re were referenced
to ϕ0=tan-1Im/Re that was measured with
a 100,0 kΩ resistor. Values φ-φ0=0∘
hence represent an resistor-type behaviour of graphite/TiO2/Ti for the
specific frequency and humidity, while φ-φ0=-90∘ is associated with a capacitor-type behaviour. It is clear from
an inspection of Fig. 7 that graphite/TiO2/Ti behaves almost like a pure
capacitance in dry air in the low-frequency region and almost like a pure
resistor in humid air at 97 % RH. The phase angle difference is
associated with the phasor in the Im(Z̃) vs. Re(Z̃)
plane and its change is hence related to the dielectric property of the
porous TiO2 layer and how it is affected by water absorption.
Typical frequency-dependent phase angles φ-φ0
of the impedance of graphite/TiO2/Ti sensors for different humidities
between 3 and 51.8 kHz. Excitation voltages equal 2 Vp-p. Values of
φ-φ0 are given with respect to phase angles φ0
of a 100.0 kΩ resistor that is used instead of the sensor prior to
the exposure experiment. The experiments were performed at T=22 ∘C. The dotted line refers to a fixed frequency of 5 kHz at
which transients of φ-φ0 were recorded.
The calibration curve of sensor signals φ-φ0 as a
function of relative humidity at constant frequency can be determined from
Fig. 7. Here values φ-φ0 are taken at constant frequency
of 5 kHz (dotted line). The result is shown in Fig. 8. The calibration curve
was fitted with a polynomial curve of type ϕ-ϕ0=a+b×%RH+c%RH2 (solid line)
with a coefficient of determination R2 of 0.98 (a= -81,53∘,b= 1, 10 and c=-0,0028/∘). The inserted image displays the
transient of the signal during a humidity from 75 % RH to 33 % RH
in comparison with a HDC1000 humidity sensor which was integrated in our
electronics. We did not calibrate the HDC 1000 and recorded the raw data from
the internal registers of this device. The experiment demonstrated a faster
response time of the graphite/TiO2/Ti sensor (denoted as PEO in the
image) in comparison with the HDC1000. It should be noted that the entire
equilibration times of both sensors are identical (over 24 h) and are
determined by re-adjusting the humidity over the salt solution, which is
disturbed by inserting the sensors in the flask. Also, the limited control of
room temperature over time led to a rather large error bar of the water
partial pressure as seen in Fig. 8.
Sensor signals φ-φ0 of graphite/TiO2/Ti as
a function of relative humidity and water partial pressure pwater
at T=22 ∘C. Measurement was made at a constant frequency of
5 kHz of the excitation AC voltage (experimental data: points; fit curve:
solid lines). The horizontal error bars show the calculated change of the
water partial pressure pwater for a temperature change of
±3 ∘C. Note that corresponding changes in the relative humidity
are below ±1 % RH over the saturated salt solutions in the flask. The
inserted image shows the transient of sensor signals of graphite/TiO2/Ti
in comparison with a commercial HDC1000 humidity sensor (Texas Instruments)
for a humidity change from 75 % to 33 %. At t0, sensors were
quickly extracted from a flask with 75 % RH and inserted into a flask
with 33 % RH.
Unlike the HDC1000 sensor, we noticed that the PEO sensor followed the
humidity changes in breathed air during inhalation and exhalation, also with
good time resolution. We ascribe the dynamics of the graphite/TiO2/Ti
sensor to the porosity of the titanium dioxide and permeation of water
molecules through the graphite top electrode.
A constant power source, digitally controllable, drove an SMD resistor
underneath the sensor. This made it possible to adjust the operation
temperature at different values between room temperature and 43 ∘C
while the sensor remained in the glass vessel of a specific constant relative
humidity. It should be noted that the absolute humidity (e.g. expressed by
the water partial pressure, compare the top scale in Fig. 7) did not change
by heating the sensor. We found that phase angles φ-φ0
did not alter when the temperature was increased from room temperature to
43 ∘C. These findings suggest that the graphite/TiO2/Ti sensor
measures the absolute humidity rather than the relative humidity. The physics
and chemistry of the water interaction with titanium oxide layers, produced
by plasma-electrochemical oxidation, have been reported in Cakabay et al. (2016) and
may serve as a starting point for an advanced understanding of humidity
sensing with graphite/TiO2/Ti structures.
Amperometric detection of CO with Pt–graphite/TiO2/Ti
Sensors
In a further set of experiments, we investigated the capability of PEO
sensors for CO detection at room temperature. Figure 9a displays a typical
result from an amperometric PEO sensor, operated at a bias voltage of 10 V
at the Pt–C top electrode, referenced to the Ti substrate. For this
measurement, a small digital electronics system with adjustable precision
voltage and a programmable gain amplifier was used. It measured the voltage
decay over a precision 10.0Ω resistor caused by the current I. It
is obvious from the comparison of the data for 20 % RH, 30 % RH, and
40 % RH, that the presence of H2O increased CO detection
sensitivity. We determine sensor signals of 0.43 µA at 20 % RH,
1.18 µA at 30 % RH, and 3.17 µA at 40 % RH in
pure air, whereas the signals increase by a factor of 2.51 (9.66) at 20 %
RH, 4.74 (28.12) at 30 % RH, and 8.17 (66.37) at 40 % RH in the
presence of 50 ppm (1000 ppm) CO. The findings confirm the applicability of
the solid-state sensor, revealed in Fig. 1g. Note that the maximum voltage,
applied during the plasma-electrolytic oxidation, and hence the porosity,
influence sensitivity. While the result shown in Fig. 7 was obtained for very
porous titanium oxide layers (150 V, compare El Achhab et al.,
2014), less porous layers can be applied to monitor
even small concentrations of CO well below the threshold limit value of
30 ppm in air.
(a) Typical transients of signals of
Pt–graphite/150 V–TiO2/Ti CO sensors at different humidities for CO
volume fractions between 50 and 1000 ppm. The bias voltage is 10 V. The
titanium oxide layer was prepared at a maximum voltage of 150 V.
(b) Signal (here: current after 30 min exposure time) as a function
of concentration. The error is mainly determined by variations of humidity in
the flow set-up.
Typical transients of signals of Pt–graphite/120 V-TiO2/Ti CO
sensors at 20, 30, 40, and 50 RH% for CO volume fractions between 2.5 and
15 ppm. The bias voltage is 2 V.
Figure 10 shows results in humid air (20 to 50 % RH) for a sensor with a
titanium oxide layer prepared at 120 V. Here, a lower bias voltage of 2 V
was applied to keep the current through the device below 20 mA. The baseline
shift due to a humidity change from 20 to 40 % RH corresponds to
approximately 10 ppm CO. At constant humidity, the lowest detection limits
are potentially in the range of 1 ppm. Furthermore, we found that the
transient I in the recovery regime after the CO exposure period shows an
“undershooting” effect in which the current falls temporarily below the
initial current I0 prior to the CO exposure. It takes a period of
approximately 20 min at a gas flow rate of 20 mL s-1 until the
current again approaches I0.
Apart from the obvious potential that PEO sensors provide for
room-temperature detection of CO, the influence of humidity is an interesting
feature. At certain levels of relative humidity, surface hydroxyl groups and
multi-layer H2O adsorption are known features of water chemistry on
TiO2 surfaces at room temperature, depending on surface crystallography
and defects (Huang et al., 2014). Furthermore, porosity favours water
absorption because of the thermodynamically driven capillary effect. The
diffusion of hydrogen is an intrinsic property of TiO2 and has been
recently clarified with scanning tunneling microscopy (Zhang et al., 2006).
We believe that CO-to-CO2 conversion, which has also been experimentally
confirmed from the rising CO2 concentration in the downward airflow
behind the sensor, can be associated with the presence of surface hydroxyl
groups and H+ ion surface diffusion over TiO2 surfaces. This should
be the subject of further studies. From a practical point of view, the most
interesting approach would be to stabilize the relative humidity in the
vicinity of the sensing element, as is done in liquid-state electrochemical
sensors.
It should be noted that Pt–graphite/TiO2/Ti sensors for CO are affected
by hydrogen. The interaction of H2 with Pt/TiO2/Ti has been
previously studied in detail (see the reference of Schierbaum and El Achhab,
2011). Also, oxidizing gases like NO2 interact with
Pt–graphite/TiO2/Ti. The results will be the subject of another
publication.