Monolithic CMOS multi-transducer gas sensor microsystem for organic and inorganic analytes

Sensors and Actuators B 126 (2007) 431–440

Monolithic CMOS multi-transducer gas sensor microsystem for organic and inorganic analytes Y. Li, C. Vancura, D. Barrettino, M. Graf, C. Hagleitner, A. Kummer, M. Zimmermann, K.-U. Kirstein, A. Hierlemann ∗ ETH Zurich, Physical Electronics Laboratory, Wolfgang-Pauli-Strasse 16, 8093 Zurich, Switzerland Received 12 December 2006; received in revised form 23 March 2007; accepted 24 March 2007 Available online 5 April 2007

Abstract A monolithically integrated multi-transducer microsystem to detect organic and inorganic gases is presented. The system comprises two polymerbased sensor arrays based on capacitive and gravimetric transducers, a metal-oxide-based sensor array, the respective driving and signal processing electronics and a digital communication interface (see the first figure). The chip has been fabricated in industrial 0.8-␮m, complementary-metaloxide-semiconductor (CMOS) technology with subsequent post-CMOS micromachining. The simultaneous detection of organic and inorganic target analytes with the single chip multi-transducer system has been demonstrated. The system is very flexible and can provide different information of interest: the capacitive sensors can, e.g., act as humidity sensors to deal with the cross-sensitivity of the metal-oxide-based sensors to water, or the capacitive sensors can be coated with differently thick polymer layers to detect organic volatiles even in a background of water. The multi-transducer approach provides a wealth of information that can be used to improve the system discrimination capability and performance in gas detection. © 2007 Elsevier B.V. All rights reserved. Keywords: CMOS; Multi-transducer system; Gas sensor; Cantilever; Capacitor; Microhotplate; Polymer; Metal oxide

1. Introduction Gas sensors are widely employed for a variety of applications, such as environmental monitoring and air quality control [1–7]. In recent years, hand-held devices for gas detection are getting more and more popular. This entails increasing research activities to develop gas sensors featuring small size, low power consumption and low costs. The monolithic integration of CMOS gas sensors is a promising approach that has been fueled by the rapid development in integrated-circuit and MEMS technology [7–12]. The aim in utilizing microfabrication techniques and, in particular, CMOS technology for realizing chemical sensors is to devise more intelligent, more autonomous, more integrated, and more reliable gas sensor systems at low costs in a generic approach. Since the sensor market is strongly fragmented, i.e., there exists a large variety of applications with different needs and sensor requirements, a modular approach or “toolbox strategy”

∗

Corresponding author. E-mail address: [email protected] (A. Hierlemann).

0925-4005/$ – see front matter © 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.snb.2007.03.035

relying on a platform technology was identified as the most promising attempt to achieve major progress [13,14]. Once the platform technology has been chosen, the components of the toolbox such as transducers, sensor modules, and circuit modules can be developed, some of which afterwards can be assembled into a customized system that meets the respective applications needs. A multitude of development activities are necessary to obtain all the modules needed for such a CMOS “toolbox”: (a) the design and miniaturization of transducers and directly related electronic components (potentiostats, heaters, amplifiers, etc.), (b) the development of digital-to-analog and analog-to-digital conversion units, interface and communication units, (c) the development of additional and auxiliary functions, which are pivotal for the system performance (e.g., temperature control, temperature sensors, humidity sensors), and (d) the development of dedicated microsystem packaging solutions, which are suitable for chemical or gas analysis [13,14]. It is important to note that the package has to be thought of already in the initial conception phase of a microsystem, since the design and architecture of a microsystem heavily depend on the envisaged packaging concept, as will become evident later in this paper (see system description and layout).

432

Y. Li et al. / Sensors and Actuators B 126 (2007) 431–440

Fig. 1. Micrograph of the monolithic multi-transducer chip.

The main disadvantages of a monolithic CMOS–MEMS solution include the restriction to CMOS-compatible materials and the limited choice of micromachining processes. However, the use of CMOS–MEMS offers, on the other hand, unprecedented advantages over hybrid designs, especially with regard to signal quality, device performance, increased functionality and available standard packaging solutions. These advantages clearly outweigh the drawbacks and limitations. In the case of well-established physical sensors such as acceleration and pressure sensors, a trend towards monolithic solutions can be identified for large production volumes and severe cost restrictions [15–18]. The monolithic integration of sensor and circuitry allows for on-chip control and monitoring of the mechanical functions and data processing. The reduction of the number of electrical connections through the use of standard interface units on chip significantly contributes to the reduction of the overall system costs and improves its reliability. At the same time, the processing of the sensor signals at the signal source helps to improve the sensor signal quality, as the influence of external interferences can be reduced.

The selectivity of individual gas sensors or systems still poses a major problem. Many sensitive materials, such as metal oxides or polymers, respond to a variety of inorganic gases or volatile organic compounds (VOCs). The use of a set of identical transducers coated with different materials along with software tools, such as multi-component analysis algorithms [19–25], or the integration of different transducers, which respond to distinct analyte properties, with dedicated circuitry in a microsystem [26–29] can help to overcome the problems associated with poor selectivity and drift of individual gas or liquid-phase sensors. In the following a monolithically integrated, CMOS-based multi-transducer microsystem to detect organic and inorganic gases will be presented. 2. System description The monolithic multi-transducer chip (Fig. 1) comprises two types of polymer-coated transducers (two cantilevers and two measuring capacitors), which are predominantly sensitive to organic volatiles, two microhotplates, which respond preferentially to inorganic gases, the needed driving electronics, the

Fig. 2. Schematic representation of the system architecture.

Y. Li et al. / Sensors and Actuators B 126 (2007) 431–440

readout electronics, an on-chip biasing, and a digital communication interface. The different transducers with the associated circuitry units will be detailed below. The cantilevers and the capacitive sensors rely on bulk physisorption of organic volatiles in polymers, which is strongly temperature-dependent. Therefore, a temperature sensor relying on the linear temperature dependency of the base-emitter-voltage of the vertical bipolar transistor, which is available in the CMOS process, has been implemented to monitor the chip temperature. A block diagram of the system and its integrated signal processing capabilities is shown in Fig. 2. The analog part of the system includes the sensor driving circuitry, the sensor signal read-out circuitry and provides bias voltages and bias currents to set the operating points of the operational amplifiers and filter stages. The digital part of the system includes a configuration register to control the system, which enables to select a sensor element from the array, and to set the hotplate temperatures and the cantilever feedback parameters. Additionally, the digital part hosts the communication interface, an I2 C bus-interface, which features a robust communication protocol and requires only two connection lines so that the overall number of connection lines (wire bonds) is very low. 2.1. Resonant cantilever and feedback circuitry The resonant cantilevers are 150 ␮m long and 140 ␮m wide, exhibit a quality factor of approximately 1000 in air at 400 kHz resonance frequency and feature electromagnetic excitation and piezoresistive read-out. The electromagnetic excitation is based on the Lorentz force and requires a small permanent magnet in the package underneath the cantilever; for details, see Fig. 6 and Refs. [30,31]. The vibration of the magnetically excited cantilever is detected by a set of four stress-sensitive MOS transistors (two transistor gates parallel to the cantilever axis, which are severely deformed, and two gate regions perpendicular to the cantilever axis, which are hardly deformed) in a Wheatstone bridge configuration located at the cantilever base, which are biased in a linear region. Compared to another widely used excitation method, namely thermal excitation, this method features the advantage of low power dissipation (1.3 mW), which leads to a significantly lower temperature (temperature increase only 1–2 ◦ C above ambient temperature) on the cantilever as well as in the sensitive polymer layer. Since, the quantity of absorbed analyte in the polymer is inversely proportional to the temperature, a reduction of the power dissipation on the cantilever yields an enhanced chemical sensitivity of the device [31]. Upon analyte absorption in the chemically sensitive polymer on the silicon cantilever, the oscillating mass increases. As a consequence the resonance frequency of the system decreases. This causes a negative frequency shift, f [Hz]. The sensitivity S of a polymer-coated cantilever is given by: S=

f = GCant hKc MA cA

(1)

Here, f denotes the mechanical resonance frequency of the cantilever and cA is the analyte concentration in the gas phase. Eq.

433

Fig. 3. Block diagram of the cantilever feedback circuitry.

(1) includes a summary term for the mechanical properties of the cantilever, GCant ; see also [32]. The sensitivity is proportional to the polymer layer thickness, h (for h < 5 ␮m) and the partition coefficient, Kc , as well as to the molecular mass of the absorbed analyte, MA . Swelling effects and analyte-induced changes in the elastic modulus of the polymer have been neglected since only very low analyte concentrations have been applied [32]. The cantilever mechanical resonator and the associated feedback circuitry form an oscillator. The high quality factor of the cantilever in the range of 1000 in air relaxes the specifications for the feedback circuitry, since the cantilever acts as a band-pass filter with an extremely narrow pass band. However, amplitude condition and phase condition still have to be met to achieve a stable oscillation, which is the most important issue in designing cantilever oscillator circuits. The block diagram (see Fig. 3) illustrates the architecture of the feedback circuitry and the comparator, which converts the sinusoidal oscillation signal into a square wave that serves as the input of the digital counter. The variable-gain amplifier [33], which is based on a differential difference amplifier (DDA) [34], provides a tunable gain between 30 and 45 dB by changing one of the bias currents. The possibility to adjust the loop gain is very important because the variation in cantilever properties, such as resonance frequency, quality factor, and peak gain, can reach a maximum of 10% as a consequence of the fabrication spread. An all-pass filter that acts as a phase shifter adjusts the total loop phase. Due to the high resonator Q-factor, the system can benefit from the natural steep slope of the phase response so that a phase tuning with a step size of 10◦ is enough to achieve the required frequency stability of 0.1 Hz. Moreover, the narrow pass-band filter function of the cantilever helps to remove noise so that an additional band-pass filter noise reduction is not necessary here. The oscillation amplitude is regulated by the nonlinear transconductance to achieve a stable operation, the gain of which is a function of the input signal amplitude. At a defined input level the gain decreases with increasing input signal amplitude, which entails that the amplitude of oscillation builds up to the point, at which this nonlinear block decreases the loop gain to unity. The nonlinear transconductance is followed by a class-AB buffer to drive the low-resistance coil. The frequency readout circuitry consists of the comparator and a 24-bit digital counter included in the digital part of the system. An analog multiplexer allows for sharing the feedback loop and the frequency read-out mod-

434

Y. Li et al. / Sensors and Actuators B 126 (2007) 431–440

ule for the two cantilevers, which reduces chip area and power consumption. The respective cantilever is activated by connecting it to the feedback circuitry through the multiplexer, then, the oscillation frequency is measured. This process is sequentially applied to the two cantilevers. The design is a trade-off between intended low chip area and power consumption and the achievable sensor response time. 2.2. Capacitive sensor array and readout circuitry The capacitive sensor is based on two sets of interdigitated electrode structures, which correspond to the two plates of a standard capacitor (Fig. 4). The sensor monitors changes in the dielectric coefficient of the polymer upon analyte absorption. The capacitors are fabricated using exclusively layers and materials available in a standard CMOS process. One of the electrodes is made from the first CMOS metal layer, and the other is realized as a stack of the first and second metal layer. The dimensions of the capacitor are 814 ␮m × 824 ␮m, and it includes 128 finger pairs. The electrode width and spacing are 1.6 ␮m. The nominal capacitance of the interdigitated capacitor is a few picoFarad, whereas capacitance changes upon analyte absorption are in the range of a few attoFarad. Thus, a dedicated on-chip measurement configuration and specific signal conditioning circuitry is needed. The sensor response is read out as a differential signal between a passivated reference and a polymer-coated sensing capacitor. A digital output signal is then generated by comparing the minute loading currents of both capacitors using a fully differential second-order Sigma–Deltamodulator circuitry [35]. The modulator provides a pulse density modulated output that can be decimated by using a frequency counter. Thus, the output signal is a frequency change, which is linearly proportional to the capacitance change upon analyte absorption in the polymer (see Eq. (2) below). For thin polymer layers the swelling of the polymer upon analyte absorption always results in a capacitance increase regardless of the dielectric constant of the absorbed analyte. This is due to the increased polymer/analyte volume within the field line region exhibiting a larger dielectric constant than that of the substituted air [36,37]. Thin polymer layers include layer

thicknesses of less than half the periodicity of the electrodes. On the other hand, the capacitance change for a polymer layer thickness larger than half the periodicity of the electrodes is determined by the ratio of the dielectric constants of analyte and polymer. If the dielectric constant of the polymer is lower than that of the analyte, the capacitance will be increased, and, if the polymer dielectric constant is larger than that of the analyte, the capacitance will be decreased. This effect has been previously detailed and supported by simulations [36–38]. For thick polymer layers the sensitivity, S, is the change in capacitance, C, in dependence of the change in the analyte concentration, cA , as given by: S=

C = Gcap Kc ε cA

(2)

where Gcap includes the capacitor geometry. The partition coefficient, Kc , includes the polymer/analyte interactions, and ε is the change in the dielectric properties of the polymeric matrix upon analyte absorption. More details on capacitive sensing and Eq. (2) can be found in Refs. [35–38]. 2.3. Microhotplate sensor and circuitry Microhotplates usually feature metal-oxide-based coatings that have to be operated at temperatures between 200 and 400 ◦ C [39–42]. Resistance changes of the sensitive material upon gas exposure produce the sensor signal. The microhotplate is located on a micromachined membrane to thermally isolate it from the rest of the chip. Each hotplate features a resistive polysilicon ring-heater, which provides symmetric heat generation and dissipation [43]. The polysilicon temperature sensor on the membrane is used to monitor the hotplate temperature and provides the feedback signal for the temperature control loop [43,44]. Two microhotplate-based sensors and the necessary driving and signal-conditioning circuitry are integrated on the chip. They feature platinum electrodes and are coated with a SnO2 sensitive layer, which is operated at temperatures between 200 and 350 ◦ C. The on-chip temperature controller regulates the temperature of the membrane up to 350 ◦ C with an accuracy of ±2 ◦ C, whereat

Fig. 4. Block diagram of the capacitive sensor readout scheme.

Y. Li et al. / Sensors and Actuators B 126 (2007) 431–440

435

Fig. 5. Block diagram of the microhotplate temperature control and resistance readout circuitry.

the deviations at room temperature are relatively small and deviations of up to 2% occur in the range between 280 and 350 ◦ C due to the nonlinearity of the polysilicon temperature resistor. The two hotplates can be independently heated and controlled to achieve the best possible sensitivity to a variety of analytes, i.e., two temperature controllers are implemented on the chip. Besides sufficient control accuracy the controller should consume as little chip area as possible and may not interfere with other integrated transducers or circuitry units. The analog proportional temperature controller that has been implemented on the chip using an operational amplifier meets these requirements. The control voltage for setting the hotplate temperature can be programmed through the digital interface and on-chip 10-bitdigital analog converters as shown in Fig. 5. The signal of the polysilicon temperature sensor can be also read out via the digital interface so that the hotplate temperature can be continuously monitored. A more precise temperature control can be realized using an additional control loop. The resistance of the SnO2 sensitive layer and the gas-induced resistance changes can vary over a wide range between 1 k and 10 M (four orders of magnitude). Therefore, the resistance is measured using a logarithmic converter (Fig. 5), which is implemented with a voltage-to-current converter and a pair of diode-connected vertical PNP transistors. The sensor signal is multiplexed to the input of a single 10-bit analog-to-digital converter to save chip area and to reduce the power consumption. 3. Experimental 3.1. System fabrication During completion of the industrial CMOS process sequence, the standard CMOS-passivation in the microhotplate area is already opened to establish contact between the platinum electrode metallization and the CMOS aluminum. A photolithography step is used to define the size of the Pt electrodes, then 50 nm Ti/W and, afterwards, 100 nm Pt are sputtered onto the wafer through a shadow mask to ensure locally defined metal deposition. The electrodes are then fabricated using a lift-off process. The n-well-membrane of the mass-sensitive cantilevers and the thermally insulated island structure of the metal-oxide based

sensors are released simultaneously. This is done by anisotropic silicon etching with KOH (potassium hydroxide) from the back of the wafer with an electrochemical etch stop technique that stops at the n-well of the CMOS process. Then, the cantilevers are released by two front-side reactive-ion-etching (RIE) steps, which are used to release the cantilevers, i.e., to remove the dielectric layers of the CMOS process and the fraction of the silicon n-well, which does not form a part of the cantilever. The resulting cantilevers are 8.7 ␮m thick and are composed of the dielectric layers (silicon oxide 2.2 ␮m, silicon nitride 1.0 ␮m) of the CMOS process on top of the silicon n-well layer (5.5 ␮m). They are rather stiff and exhibit a force constant of 800 N/m. The wafers are then diced using a protective foil over the microstructures. 3.2. Sensor packaging and coating A package including a small permanent magnet has been developed in order to perform chemical measurements with the electromagnetically actuated cantilever sensors. The permanent magnet was placed underneath the chip, so that the polymercoated front side of the cantilever can be exposed to different analytes. A standard ceramic dual-in-line package was modified in a way that the bottom part was replaced with an aluminum block. After placing the permanent magnet into a cavity in this aluminum block, the chip has been glued on the aluminum block with the cantilever right on top of the rare-earth magnet. A crosssection is shown in Fig. 6. An important aspect of the packaging design includes the selection of a magnetic material that generates a strong magnetic field and the placement of this magnet as close as possible to the cantilevers. Other issues that have been considered in designing the system floor plan (sensor-package co-design, see Fig. 7a) concern the gas flow direction over heated (hotplates) and nonheated (capacitor, cantilever) transducers and the openings for gas exposure. Here the gas flow is first over the nonheated area and then the heated area to avoid that temperature fluctuations influence or upset the polymer sorption processes. Moreover, all sensors and transducers are placed on one side of the chip, whereas the electronics are placed on the opposite side so that the electronics can be covered and are not exposed to analytes.

436

Y. Li et al. / Sensors and Actuators B 126 (2007) 431–440

3.3. Gas manifold

Fig. 6. Cross-sectional schematic of the chip system and the permanent magnet in the package.

The electrical interconnections have been made by wire bonding. To protect the bond wires and the circuitry, the packaged chip has been then partially covered with a glob-top epoxy encapsulant as illustrated in Fig. 7b. First, a dam of low-viscosity epoxy has been placed by means of a dispenser along metal line features on the chip, and, then, the circuitry part of the chip has been covered using a higher-viscosity epoxy. Partial coverage of the chips with epoxy provides a good protection of the electronics, while still enabling free access of the analyte gas to the sensitive area of the sensors. After completion of the post-CMOS micromachining steps and after packaging, the chips and transducers have been coated with the sensitive layers. Polymers are widely used as a sensitive layer for the detection of VOCs. Standard polymers such as the slightly polar poly(etherurethane) (PEUT) and the nonpolar poly(dimethylsiloxane) (PDMS), both available from Fluka, Buchs, Switzerland, have been used here and have been deposited on the cantilevers and capacitive sensors by spraycoating using an airbrush method and shadow masks. For the microhotplates, the nanocrystalline SnO2 doped with 0.2 wt% Pd was deposited onto the hotplates using a drop-coating method [45,46]. The minute Pd-content entails a large sensitivity to carbon monoxide.

For gas tests, the CMOS chips were mounted on dual-in-line packages and then loaded into the measurement chamber of a computer-controlled gas manifold featuring a cross-over flow architecture. This cross-over flow architecture has two input gas lines, one supplying a pure carrier gas and the other supplying a carrier gas with defined doses of the volatile analyte, and two output gas lines, one leading to the measurement chamber and the other leading directly to the exhaust. This architecture offers the advantage that both input flows and both output flows are continuously flowing and the build-up time of a certain analyte concentration does not influence the dynamic sensor responses. The overall gas volume between the valve and the sensors was approximately 1.6 ml, which entails a time span of approximately 0.5 s after switching the valve until the gas reaches the sensors at the applied flow rate of 200 ml/min. The analyte vapors were generated from specifically developed temperaturecontrolled (T = 223–293 K) vaporizers using synthetic air as a carrier gas, and then diluted as desired using computer-driven mass-flow controllers. The internal volume of these vaporizers, which distribute the liquid over a large-area packed-bed type support to maximize the surface-to-volume ratio, was dramatically smaller than that of typical gas-washing bottles (bubblers) [47]. The vapor-phase concentrations at the respective temperatures were calculated following the Antoine equation [48]. A photoacoustic detector (infrared light for excitation, 1314 Photoacoustic Multi-gas Monitor, Innova Airtec Systems, Denmark) is used as an independent reference to assess the actual analyte gas-phase concentrations. The sensor measurements were performed in a thermo-regulated chamber at a temperature of 303 K. Both gas streams (pure carrier gas and carrier gas with analyte) were thermostabilized at the measurement chamber temperature before entering the chamber. The response time of the sensors at the given polymer thickness (2–4 ␮m) is on the order of a few seconds. Typical experiments consisted of alternating exposures to pure synthetic air and analyte-loaded synthetic air. Exposure times of 10–15 min to analyte-loaded gas (to reach thermody-

Fig. 7. (a) System layout considerations (system-package co-design): magnet and magnetic field for the cantilevers; gas and thermal flow, and gas exposure opening. (b) Micrograph of the packaged system chip.

Y. Li et al. / Sensors and Actuators B 126 (2007) 431–440

437

Fig. 8. Sensor responses of the cantilevers (bottom part) and capacitive sensors (upper part) upon exposure to various concentrations of ethanol and toluene. The analyte concentrations included 500–2500 ppm, up and down. The cantilever has been coated with a PDMS layer of 0.5 ␮m thickness. The capacitive sensor has been coated with a 1.4-␮m-thick PEUT-layer.

namic equilibrium) were followed by 10–15 min purging the chamber with pure synthetic air. The selected analytes included standard organic solvents and used as purchased from Fluka, Buchs, Switzerland without further purification (n-octane, toluene, ethanol). The inorganic cases like carbon monoxide were dosed from gas cylinders of defined analyte gas concentration. 3.4. Measurement results As already mentioned in the introduction, the system is very versatile and can provide a wealth of information that can be used for a certain detection problem at hand. In the following we will try to give an idea on how the system can effectively be used in different application scenarios. The first example is the detection of simple organic volatiles: ethanol and toluene. The two cantilevers are coated with different polymers, the nonpolar PDMS and the slightly polar PEUT. The capacitors featuring interdigitated electrodes include one reference capacitor protected by a passivation layer and two sensing capacitors; see Fig. 4. The two sensing capacitors are coated with PEUT layers of different thickness, 1.4 ␮m and 3 ␮m. By connecting two of these three capacitors to the input of the Sigma–Delta modulator, a differential measurement can be realized. For the organic-volatile-detection task, we connect the thick-layer PEUT capacitor C1 and the reference capacitor Cref to the Sigma–Delta converter. Some of the sensor results are displayed in Fig. 8a–d. The frequency responses of the PDMScoated cantilever (0.5 ␮m thickness) upon exposure to various concentrations of toluene and ethanol are shown in Fig. 8c and d. Since, the molecular weight of toluene is larger than that of ethanol, the sensor signal upon exposure to toluene is higher than upon exposure to ethanol. However, it is not possible to distinguish low concentrations of toluene from high concentra-

tions of ethanol. Here, the results of the capacitive sensor will help. The frequency responses of the thick-layer capacitive sensor upon exposure to these two analytes are shown in Fig. 8a and b. Ethanol with a dielectric constant of 24.3, which is larger than that of the polymer PEUT (4.8), causes a capacitance increase and, hence, a positive frequency shift, whereas toluene with a dielectric constant of 2.36 causes a capacitance decrease and a negative frequency shift. Hence, toluene and ethanol can be easily differentiated using the multi-transducer chip. Of course, the signals of the microhotplates (not shown) can be additionally used as input for pattern recognition or multicomponent analysis tools. A second example concerns the detection of carbon monoxide (CO) on a background of changing humidity. For this scenario, we use the microhotplates and the capacitive sensor, which acts in this case as a humidity sensor. The microhotplate covered with the Pd-doped nanocrystalline SnO2 (0.2 wt% Pd) was heated to 275 ◦ C, and the sensors were exposed to different analyte concentrations. Metal-oxide based sensors are highly sensitive to inorganic gases such as CO, but exhibit a significant crosssensitivity to humidity [39,41]. The measurements were carried out with varying relative humidity as shown in Fig. 9. The respective SnO2 sensor response amplitudes increase with increasing humidity, and the sensor baseline also shifts. Fig. 9b shows the signal as recorded from the capacitive sensor at the same time. The CO is too volatile to be enriched in a polymeric layer so that the capacitive sensor exclusively monitors the changing humidity. The co-integration of a capacitive sensor, which is highly sensitive to humidity (water has a dielectric coefficient as high as 78), allows for taking into account the humidity influence so that the cross-sensitivity of the hotplate to humidity can be compensated for in the subsequent data processing procedure. Humidity will, due to its high dielectric coefficient, also have a major impact on any capacitive organic-volatile mea-

438

Y. Li et al. / Sensors and Actuators B 126 (2007) 431–440

Fig. 9. Sensor responses upon dosage of different CO concentrations at different humidity levels (10, 20 and 40% relative humidity): (a) microhotplate responses; (b) capacitive sensor responses. The hotplate has been coated with nanocrystalline SnO2 containing 0.2 wt% Pd, the operating temperature was 275 ◦ C. The capacitive sensor has been coated with a 1.4-␮m-thick PEUT layer.

Fig. 10. Frequency shifts of capacitive sensors exposed to relative humidity (10, 20, 30, 40 and 50%), toluene (400, 800, 1200 and 1600 ppm), and n-octane (150, 300, 450 and 600 ppm). (a) The left curve displays the difference signal of one sensor capacitor and the reference capacitor. (b) The right curve shows the differential signal of two sensor capacitors coated with the same polymer at different thickness.

surement. Therefore, we want to demonstrate in a third example how organic volatiles can be measured even on a background of humidity or changing relative humidity. The capacitive sensor exhibits high sensitivity to humidity. However, its sensitivity towards analytes with a high dielectric constant shows a sensitivity maximum at a relatively low polymer layer thickness around 1–1.5 ␮m, whereas for analytes with lower dielectric constant this sensitivity maximum varies considerably out to greater layer thickness (3–3.5 ␮m, for details see [36]). Hence, the signal difference of two capacitors with different layer thicknesses in the range of 1.5–5 ␮m is almost insensitive to water but retains sensitivity to low-dielectric-constant analytes like toluene or n-octane [36]. This is evident from Fig. 10, which shows the results as measured in the standard configuration with C1 and the reference capacitor Cref connected to the Sigma–Delta converter (Fig. 4): large positive humidity signals and comparatively lowlevel negative organic volatile signals. In Fig. 10b, the signals achieved with C1 and C2 connected to the Sigma–Delta converter (Fig. 4) are displayed. The humidity signal amplitudes are drastically reduced in the differential signal whilst the two organic volatiles still show distinct and clear capacitive signals. Finally we will briefly discuss the reproducibility of the results and the devices. The device-to-device repeatability concerning the CMOS fabrication and micromachining is very good. Within a wafer run, the devices are almost identical, the cantilever frequencies, e.g., vary from 380 to 405 kHz, with the center frequencies being at 395 kHz, i.e., the production

spread is ± 3%. Similar considerations hold for the polysilicon temperature sensor and the microhotplates. Nevertheless a device-by-device calibration is necessary as it is common also for, e.g., commercially available humidity sensors. The main contribution to variations in the sensor characteristics is due to the sensitive layers. The reproducibility and long-term stability is, in general, better for the polymer-based sensors as compared to the metal oxide sensors. For both sensitive materials, the trialto trial reproducibility of the same device within a week shows a maximum variation of 5%. For the polymer-based sensors, longterm measurements evidenced a signal variation of ±10% over a year, while for metal-oxide-based sensors, the initial material resistance can widely vary, and signal variations up to ±30% over a year have been observed. 4. Conclusion In summary a very versatile and easy-to-use single chip system has been presented that can be applied to a multitude of detection tasks and scenarios, only a few of which have been demonstrated here. The simultaneous detection of organic and inorganic gases, even in more complex mixtures, can be achieved by the combination of different polymer-based and metal-oxidebased sensors. The multi-transducer system provides a wealth of information that can be used to improve gas identification and quantification. The end-user can access the system via a LabviewTM interface and can program all relevant parameters,

Y. Li et al. / Sensors and Actuators B 126 (2007) 431–440

i.e., the complexity of the system is shifted to the system development level and is not really visible for the end-user. The low-power design of the system enables its application in handheld devices. Acknowledgments The authors thank Professor Henry Baltes for sharing laboratory resources and for his ongoing interest in their work. The coating of the microhotplates with the sensitive metal-oxide layers was performed by AppliedSensor GmbH, Reutlingen, Germany, Dr. Stefan Raible and Dr. J¨urgen Kappler. This project has been financially supported by the Commission of Technology and Innovation, Bern, Switzerland under contract number KTI 5670.1. References [1] W. G¨opel, J. Hesse, J.N. Zemel (Eds.), Sensors: A Comprehensive Survey, vol. 2/3, Chemical and Biochemical Sensors, VCH-Verlagsgesellschaft, Weinheim, 1991; H. Baltes, W. G¨opel, J. Hesse (Eds.), Sensors Updates, vol. 1–10, VCH/Wiley, Weinheim, 1996–2002. [2] J. Janata, Principles of Chemical Sensors, Plenum Press, New York, 1989. [3] J. Janata, M. Josowicz, P. Vanysek, M.D. DeVaney, Chemical sensors, Anal. Chem. 70 (12) (1998) 179–208. [4] E. Kress-Rodgers (Ed.), Handbook of Biosensors and Electronic Noses, CRC Press Inc., Boca Raton, FL, 1997. [5] M.J. Madou, S.R. Morrison, Chemical Sensing with Solid State Devices, Academic Press, Boston, 1989. [6] J.W. Gardner, P.N. Bartlett, Electronic Noses, Oxford University Press, 1999. [7] J.W. Gardner, Microsensors, Wiley, Chichester, 1994. [8] M. Gad-el-Hak, The MEMS Handbook, CRC Press, Boca Raton, 2002. [9] G.T.A. Kovacs, Micromachined Transducers, MCB McGraw-Hill, New York, 1998. [10] M. Elwenspoek, H. Hansen, Silicon Micromachining, Cambridge University Press, Cambridge, 1998. [11] M. Madou, Fundamentals of Microfabrication, CRC Press, Boca Raton, FL, 1997. [12] O. Brand, G.K. Fedder (Eds.), CMOS–MEMS, VCH–Wiley, Weinheim, 2005. [13] A. Hierlemann, Integrated Chemical Microsensor Systems in CMOSTechnology, Springer, Berlin, 2005. [14] A. Hierlemann, O. Brand, C. Hagleitner, H. Baltes, Microfabrication techniques for chemical/biosensors, Proc. IEEE 91 (6) (2003) 839–863. [15] C. Hierold, Intelligent CMOS Sensors, in: Proceedings of the IEEE 13th Annual International Conference on Micro Electro Mechanical Systems 2000 (MEMS 2000), Miyazaki, Japan, 2000, pp. 1–6. [16] Infineon AG online: http://www.infineon.com. [17] G. Bitko, A. McNeil, R. Frank, Improving the MEMS Pressure Sensor, Sensors Magazine 2000. Web-based journal: http://archives.sensorsmag.com/articles/0700/62/index.htm. [18] Robert Bosch GmbH online: http://www.bosch.com. [19] G.C. Osbourn, J.W. Bartholomew, A.J. Ricco, G.C. Frye, Visual-empirical region-of-influence pattern recognition applied to chemical microsensor array selection and chemical analysis, Acc. Chem. Res. 31 (1998) 297–305. [20] D.L. Massart, B.G.M. Vandeginste, S.N. Deming, Y. Michotte, L. Kaufman, Chemometrics: A Textbook, Elsevier, Amsterdam, 1988. [21] A. Hierlemann, M. Schweizer-Berberich, U. Weimar, G. Kraus, A. Pfau, W. G¨opel, in: H. Baltes, W. G¨opel, J. Hesse (Eds.), Pattern Recognition and Multi-component Analysis, Sensors Update: VCH, 1996, pp. 119–180. [22] R.G.E. Brereton, Multivariate Pattern Recognition in Chemometrics, Elsevier, Amsterdam, 1992.

439

[23] T.C. Pearce, S.S. Schiffman, H.T. Nagle, J.W. Gardner, Handbook of Machine Olfaction, Wiley–VCH, Weinheim, 2003. [24] S. Semancik, R.E. Cavicchi, M.C. Wheeler, J.E. Tiffany, G.E. Poirier, R.M. Walton, J.S. Suehle, B. Panchapakesan, D.L. DeVoe, Microhotplate platforms for chemical sensor research, Sens. Actuators B 77 (2001) 579–591. [25] A.J. Ricco, R.M. Crooks, G.C. Osbourn, Surface-acoustic-wave chemical sensor arrays: new chemically sensitive interfaces combined with novel cluster analysis to detect volatile organic compounds and mixtures, Acc. Chem. Res. 31 (1998) 289–296. [26] E. Lauwers, J. Suls, W. Gumbrecht, D. Maes, G. Gielen, W. Sansen, A CMOS multiparameter biochemical microsensor with temperature control and signal interfacing, IEEE J. Solid-State Circuits 36 (2001) 2030–2038. [27] C. Hagleitner, A. Hierlemann, D. Lange, A. Kummer, N. Kerness, O. Brand, H. Baltes, Smart Single-chip Gas Sensor Microsystem, Nature 414 (2001) 293–296. [28] A. Witvrouw, F. Van Steenkiste, D. Maes, L. Haspeslagh, P. Van Gerwen, P. De Moor, S. Sedky, C. Van Hoof, A.C. de Vries, A. Verbist, A. De Caussemaeker, B. Parmentier, K. Baert, Why CMOS-integrated transducers? A review, Microsyst. Technol. 6 (2000) 192–199. [29] W. Gumbrecht, D. Peters, W. Schelter, W. Erhardt, J. Henke, J. Steil, U. Sykora, Integrated pO2 , pCO2 , pH Sensor System for Online Blood Monitoring, Sens. Actuators B 18/19 (1994) 704–708. [30] D. Lange, C. Hagleitner, C. Herzog, O. Brand, H. Baltes, Electromagnetic actuation and MOS-transistor sensing for CMOS-integrated micromechanical resonators, Sens. Actuators A 103 (2003) 150–155. [31] C. Vancura, M. R¨uegg, Y. Li, C. Hagleitner, A. Hierlemann, Magnetically actuated complementary metal oxide semiconductor resonant cantilever gas sensor systems, Anal. Chem. 77 (2005) 2690–2699. [32] D. Lange, C. Hagleitner, A. Hierlemann, O. Brand, H. Baltes, Complementary metal oxide semiconductor cantilever arrays on a single chip: mass-sensitive detection of volatile organic compounds, Anal. Chem. 74 (2002) 3084–3095. [33] A.J. Gano, J.E. Franca, Fully differential variable gain instrumentation amplifier based on a fully differential DDA topology, in: Proceedings of the ICECS’99, the 6th IEEE International Conference on Electronics, Circuits and Systems, vol. 2, Pafos, Cyprus, 1999, pp. 781–784. [34] E. S¨ackinger, W. Guggenb¨uhl, A versatile building block: the CMOS differential difference amplifier., IEEE J. Solid-State Circuits 22 (2) (1987) 287–294. [35] S. Kawahito, A. Koll, C. Hagleitner, H. Baltes, Y. Tadokoro, Delta–Sigma modulation sensor interface circuits with improved conversion gain for capacitive readout chemical sensor, Trans. IEE Japan 119-E (1999) 138–142. [36] A.M. Kummer, A. Hierlemann, H. Baltes, Tuning sensitivity and selectivity of complementary metal oxide semiconductor-based capacitive chemical microsensors, Anal. Chem. 76 (2004) 2470–2477. [37] A.M. Kummer, A. Hierlemann, Configurable electrodes for capacitive-type sensors and chemical sensors, IEEE Sens. J. 6 (2006) 3–10. [38] F.P. Steiner, A. Hierlemann, C. Cornila, G. Noetzel, M. B¨achtold, J.G. Korvink, W. G¨opel, H. Baltes, Polymer Coated Capacitive Microintegrated Gas Sensor, Digest Tech. Papers, in: Proceedings of the 8th Inter. Conf. Solid-State Sensors and Actuators and Eurosensors IX, 1995, pp. 814–817 (IEEE Cat. No. 95TH8173). [39] N. Barsan, M. Schweizer-Berberich, W. G¨opel, Fundamental practical aspects in the design of nanoscaled SnO2 gas sensors: a status report, Fresenius J. Anal. Chem. 365 (4) (1999) 287–304. [40] H. Geistlinger, Electron theory of thin-film gas sensors, Sens. Actuators B 17 (1993) 47–60. [41] G. Heiland, D. Kohl, Physical and chemical aspects of oxide semiconductor gas sensors, in: T. Seiyama (Ed.), Chemical Sensor Technology, vol. 1, Elsevier, Amsterdam, 1988, pp. 15–38. [42] D. Kohl, The role of noble-metals in the chemistry of solid-state gas sensors, Sens. Actuators B 1 (1990) 158–165. [43] M. Graf, D. Barrettino, S. Taschini, C. Hagleitner, A. Hierlemann, H. Baltes, Metal oxide-based monolithic complementary metal oxide semiconductor gas sensor microsystem, Anal. Chem. 76 (2004) 4437–4445. [44] D. Barrettino, M. Graf, M. Zimmermann, C. Hagleitner, A. Hierlemann, H. Baltes, A smart single-chip microhotplate-based gas sensor system

440

[45]

[46]

[47]

[48]

Y. Li et al. / Sensors and Actuators B 126 (2007) 431–440 in CMOS-technology, Analog Integr. Circuits Signal Process. 39 (2004) 275–287. J. Kappler, Characterisation of high-performance SnO2 gas-sensors for CO detection by in situ techniques, Ph.D. Thesis, Shaker-Verlag, Germany, 2001. D. Briand, A. Krauss, B. van der Schoot, U. Weimar, N. Barsan, W. G¨opel, N.F. de Rooji, Design and fabrication of high-temperature microhotplates for drop-coated gas sensors, Sens. Actuators B 68 (2000) 223–233. K. Bodenh¨ofer, A. Hierlemann, R. Schlunk, W. G¨opel, New method of vaporizing volatile organics for gas tests, Sens. Actuators B 45 (1997) 259–264. J.A. Riddick, W.B. Bunger, T.K. Sakano, Organic Solvents, 4th ed., Wiley/Interscience, New York, NY, USA, 1986.

Biographies Yue Li received her MS and PhD in electrical engineering from ETH Zurich in 2001 and 2005, respectively. Her research interests included analog integrated circuit design, and the design of microsensor systems. Currently, she is working on circuit designs for image sensing systems. Cyril Vancura studied physics at the University of Kaiserslautern, Germany and the Swiss Federal Institute of Technology (ETH) in Zurich, Switzerland and received his diploma from the ETH Zurich in 2001. After his diploma, he joined the Physical Electronics Laboratory at ETH as a PhD student and received his PhD in 2005. Since 2006, he is working at the Research and Technology Center of Robert Bosch LLC, Palo Alto, CA. His research interests include microstructures for chemical and biological sensing. Diego Barrettino received the diploma in electronic engineering from the University of Buenos Aires, Argentina, in 1997 and the PhD degree in electrical engineering from the Swiss Federal Institute of Technology (ETH) Zurich, in 2004. From 2004 to 2005, he was a postdoctoral research associate in the Microscale Life Sciences Center, Department of Electrical Engineering, University of Washington, Seattle, USA. From 2005 to 2006, he was a tenure-track assistant professor in the Department of Electrical Engineering at the University of Hawaii, Honolulu, USA. He joined the Integrated Systems Laboratory, Swiss Federal Institute of Technology (EPF) Lausanne as senior research scientist in September 2006 where he is working on lab-on-a-chip microsystems for cancer research. Markus Graf received the degree in physics from the University of Konstanz, Konstanz, Germany, in 1999. He was a research assistant at the Department of Micro-and Nanotechnology (MIC), Lyngby, Denmark, and received the PhD from the Swiss Federal Institute of Technology, Zurich, Switzerland, in 2004.

Since 2005, he is with Sensirion AG, Staefa, Switzerland, where he heads the R&D group for humidity sensors. His research is focused on CMOS-based sensor systems and related microfabrication technologies. Christoph Hagleitner obtained a diploma degree and a PhD degree in Electrical Engineering from the Swiss Federal Institute of Technology (ETH), Zurich in 1997 and 2002, respectively. During his PhD work, he specialized in interface circuitry and system aspects of CMOS integrated micro- and nanosystems. After receiving the PhD degree in 2002 with a thesis on a CMOS single-chip gas detection system he headed the circuit-design group of the Physical Electronics Laboratory at the ETH Zurich. In 2003, he joined the IBM Zurich Research Laboratory in Ruschlikon, Switzerland, where he works on the analog-frontend design and system aspects of a novel probe storage device. Adrian Kummer graduated 1999 in physics at ETH Z¨urich, Switzerland. He received his PhD in applied physics, also from ETH Zurich in 2004. The topic of his thesis was capacitive chemical microsensors for the detection of volatile organic compounds. Currently, he is employed in the Research & Technology Department of Kistler Instrumente AG, Winterthur, Switzerland. His present research includes pressure sensors and optical sensors for harsh environments, especially for extreme temperatures and pressure ranges. Martin Zimmermann received the diploma degree in electrical engineering from the University of Applied Sciences, Rapperswil, Switzerland, 1996. From 1996 to 2005, he was involved in the development of mixed-signal circuitry for CMOS sensor systems at the Physical Electronics Laboratory of ETH Zurich, Switzerland. In 2006, he joined the Camille Bauer AG, Wohlen, Switzerland, where his current activities are focused on capacitive angular displacement sensors. Kay-Uwe Kirstein received the diploma in electrical engineering from the University of Technology Hamburg-Harburg (TUHH), Germany, in 1997 and the PhD degree from the University of Duisburg, Germany, in 2001. He was a Research Associate with the Fraunhofer Institut of Microelectronic Circuits and Systems in Dresden and worked in the analog circuit design group at Micronas GmbH, Freiburg, Germany. Afterwards he has been team leader of the circuit design group at the Physical Electronics Laboratory, ETH Zurich and is now with Miromico AG, Zurich, Switzerland. Andreas Hierlemann received his diploma in chemistry in 1992 and the PhD degree in physical chemistry in 1996 from the University of T¨ubingen, Germany. After working as a Postdoc at Texas A&M University, College Station, TX (1997), and Sandia National Laboratories, Albuquerque, NM (1998), he is currently associate professor at the Physical Electronics Laboratory at ETH Zurich in Switzerland. The focus of his research activities is on CMOS-based microsensors and on interfacing CMOS electronics with electrogenic cells.