**Physical Vapour Deposition Techniques for Producing Advanced Organic Chemical Sensors**

Michele Tonezzer1,2 and Gianluigi Maggioni3

*1Università di Trento, Dip. Ingegneria dei Materiali e delle Tecnologie Industriali, 2Laboratori Nazionali di Legnaro - Istituto Nazionale di Fisica Nucleare, 3Università di Padova, Dip. Fisica at LNL - Istituto Nazionale di Fisica Nucleare, Italy* 

## **1. Introduction**

In the organic chemical sensors field, the main focus to date has been on the molecular design of the receptor as a function of the analyte to be detected. Nevertheless chemical sensing requires an integrated approach, where both the molecular and the material properties of the sensing layer must be finely tuned to achieve the desired properties. Despite its great influence on the ultimate performances of the sensors, the material side has been largely neglected.

In this respect, chemical sensing of gases and vapours performed via thin solid films represents a particular challenge, as the desired recognition events need to operate at the gas-solid interface. Taking into account that the analyte recognition is mediated by the layer properties of the coated receptors, precise control and accurate characterization of these properties (e.g., thickness, permeability, and morphology) are critical for developing advanced sensing materials.

In this chapter, a description of several different real cases is carried out by the authors to highlight the significant advantages introduced into the sensing field by a bottom-up approach in which molecules purposely developed for detecting a specific analyte are deposited in well-controlled and designed architectures. The production methods, the properties and the sensing capabilities of novel advanced organic sensing materials grown by Physical Vapour Deposition (PVD) techniques are described and compared to the conventional ones. In particular the results concerning organic sensors developed with two main deposition techniques will be detailed and discussed, namely:


Specific cases related to the improvement of the sensing capabilities of different conventional organic sensors by using these two techniques are displayed. In particular the

Physical Vapour Deposition Techniques for Producing Advanced Organic Chemical Sensors 281

10 to 40W, −20 to −500V and 5.00±0.05 Pa (Ar) and 20.0±0.1 Pa (He), respectively. A water cooled quartz crystal thickness sensor, placed close to the sample holder 7 cm above the source, has been used for measuring in real time the deposition rate of the organic layer.

The apparatus used for the evaporation of the organic films consisted of a stainless steel vacuum chamber equipped with a turbomolecular pump capable of base pressure of 10-4 Pa (Maggioni et al., 1997). The sample holder is a copper circular plate with a central aperture where a quartz crystal thickness sensor is located. The heater units are three copper crucibles: each of them is equipped with a thermocouple and a heating resistance so that temperatures up to 500°C can be reached and maintained within ±1°C. During the

Thin porphyrin films were produced by spin coating technique. Porphyrin solutions (0.1

The physical and chemical features of the samples were analyzed by Scanning Electron Microscopy (SEM), Atomic Force Microscopy (AFM), Infrared analysis (FT-IR), and UV-

The surface morphology of the samples was investigated with a Philips XL-30 scanning electron microscope (SEM) at a working pressure of 10-3 Pa. SEM was also used for

The surface morphology of the organic samples was investigated in air by a non-contactmode atomic force microscope (AFM) model C-21 (Danish Micro Engineering), mounting a DualScope Probe Scanner 95-50. AFM measurements were also utilized to investigate the thickness of the samples by scratching a small area of the samples and acquiring an image of the borderline between the exposed substrate and the pristine film. The observations were

FT-IR spectra of the samples were recorded in the 4000-400 cm-1 range using a Jasco FT-IR 660 Plus spectrometer with a resolution of 16 cm-1. The sample cell and the interferometer

A Jasco V-570 dual-beam spectrophotometer was used to perform the UV-visible absorption

performed with a Si3N4 pyramidal tip with a curvature radius lower than 50 nm.

were evacuated to remove the adsorption peaks of water and atmospheric gases.

measurements in the 200-800 nm range, with a resolution of 2 nm.

**2.1.2 High Vacuum Evaporation (VE)** 

**2.1.3 Spin coating (SPIN)** 

visible analysis (UV-Vis).

deposition the substrates were kept at room temperature.

**2.2 Physical and chemical characterization** 

**2.2.1 Scanning Electron Microscopy (SEM)** 

**2.2.2 Atomic Force Microscopy (AFM)** 

**2.2.3 Infrared analysis (FT-IR)** 

**2.2.4 UV-Visible spectroscopy (UV-Vis)** 

wt.% of porphyrin powder in chloroform) were spun at 800 rpm for 30 s.

measuring the thickness of GDS samples through cross-section pictures.

chapter describes the production, the physical and chemical characterization and the sensing capabilities (by different transducing methods) of novel sensing materials based on three classes of macromolecular compounds: phthalocyanines, porphyrins and cavitands. Following the specific cases for each molecular class are reported:


In order to underline the role of the molecular architectures on the final sensing capabilities of the samples, the sensing responses are explained on the basis of their physical properties. Moreover the authors, in order to point out the role of the deposition techniques on the final sensing capabilities of the new developed samples, compare in several cases their sesing performances with those of analogous samples produced by conventional methods.

## **2. Experimental**

#### **2.1 Film deposition**

The organic chemical sensors were produced by Glow-Discharge-induced Sublimation (GDS), High Vacuum Evaporation (VE) and Spin Coating (SPIN).

#### **2.1.1 Glow-Discharge-induced Sublimation (GDS)**

The experimental setup used for the deposition of GDS films consists of a vacuum chamber evacuated by a turbomolecular pump to a base pressure of 10-4 Pa. The glow discharge is sustained by a 1-in. cylindrical magnetron sputtering source connected to a radio frequency power generator (600 W, 13.56 MHz) through a matching box. The organic powder is put on the surface of an aluminum target, placed on the sputtering source. The glow discharge feed gas is argon or helium (99.9999%), whose pressure inside the chamber is measured through a capacitance gauge. Typical values of rf power, target dc self-bias and working pressure are 10 to 40W, −20 to −500V and 5.00±0.05 Pa (Ar) and 20.0±0.1 Pa (He), respectively. A water cooled quartz crystal thickness sensor, placed close to the sample holder 7 cm above the source, has been used for measuring in real time the deposition rate of the organic layer.

## **2.1.2 High Vacuum Evaporation (VE)**

The apparatus used for the evaporation of the organic films consisted of a stainless steel vacuum chamber equipped with a turbomolecular pump capable of base pressure of 10-4 Pa (Maggioni et al., 1997). The sample holder is a copper circular plate with a central aperture where a quartz crystal thickness sensor is located. The heater units are three copper crucibles: each of them is equipped with a thermocouple and a heating resistance so that temperatures up to 500°C can be reached and maintained within ±1°C. During the deposition the substrates were kept at room temperature.

## **2.1.3 Spin coating (SPIN)**

280 Advances in Chemical Sensors

chapter describes the production, the physical and chemical characterization and the sensing capabilities (by different transducing methods) of novel sensing materials based on three classes of macromolecular compounds: phthalocyanines, porphyrins and cavitands.

• Phtahlocyanines. The physical and chemical features of copper (CuPc) and zinc (ZnPc) phthalocyanines thin solid films produced by GDS technique are reported. Their sensing capabilities towards several analytes (NO, NO2 and ethanol) are analysed by

• Porphyrins. Different porphyrin compounds have been deposited by VE and GDS and their physical properties and sensing capabilities are reported and discussed. As regards VE technique, three porphyrin compounds have been deposited: free- (H2TPP), cobalt- (CoTPP), and iron chloride- (Fe(TPP)Cl) 5,10,15,20 *meso*-tetraphenyl porphyrins. The chemical and morphological properties of VE samples have been analyzed by FT-IR and SEM images and their optical sensing capabilities towards different alcohol vapours (methanol, ethanol and iso-propanol) have been measured. As regards GDS technique, CoTPP and Fe(TPP)Cl samples have been produced and physically characterized. Moreover their optical responses towards ethanol vapours are discussed

• Cavitands. VE deposition technique has been used for growing a novel supramolecular mass sensing film based on an ultimate insoluble receptor specifically designed for detecting short chain alcohols: the tetraphosphonate Tiiii [H, CH3, Ph] cavitand. Nonspecific tetrathiophosphonate TSiiii [H, CH3, Ph] cavitand layers have been also deposited for comparison. The properties of deposited films were investigated by FT-IR analysis and their sensing capabilities were investigated by exposing Tiiii- and TSiiii-coated quartz crystal microbalances (QCMs) to ethanol in very low concentrations. Elovich kinetics were

also used to analyze the sorption process occurring onto the different samples.

performances with those of analogous samples produced by conventional methods.

(GDS), High Vacuum Evaporation (VE) and Spin Coating (SPIN).

**2.1.1 Glow-Discharge-induced Sublimation (GDS)** 

In order to underline the role of the molecular architectures on the final sensing capabilities of the samples, the sensing responses are explained on the basis of their physical properties. Moreover the authors, in order to point out the role of the deposition techniques on the final sensing capabilities of the new developed samples, compare in several cases their sesing

The organic chemical sensors were produced by Glow-Discharge-induced Sublimation

The experimental setup used for the deposition of GDS films consists of a vacuum chamber evacuated by a turbomolecular pump to a base pressure of 10-4 Pa. The glow discharge is sustained by a 1-in. cylindrical magnetron sputtering source connected to a radio frequency power generator (600 W, 13.56 MHz) through a matching box. The organic powder is put on the surface of an aluminum target, placed on the sputtering source. The glow discharge feed gas is argon or helium (99.9999%), whose pressure inside the chamber is measured through a capacitance gauge. Typical values of rf power, target dc self-bias and working pressure are

Following the specific cases for each molecular class are reported:

means of electrical and optical transducing methods.

on the basis of their physical properties.

**2. Experimental 2.1 Film deposition**  Thin porphyrin films were produced by spin coating technique. Porphyrin solutions (0.1 wt.% of porphyrin powder in chloroform) were spun at 800 rpm for 30 s.

## **2.2 Physical and chemical characterization**

The physical and chemical features of the samples were analyzed by Scanning Electron Microscopy (SEM), Atomic Force Microscopy (AFM), Infrared analysis (FT-IR), and UVvisible analysis (UV-Vis).

## **2.2.1 Scanning Electron Microscopy (SEM)**

The surface morphology of the samples was investigated with a Philips XL-30 scanning electron microscope (SEM) at a working pressure of 10-3 Pa. SEM was also used for measuring the thickness of GDS samples through cross-section pictures.

## **2.2.2 Atomic Force Microscopy (AFM)**

The surface morphology of the organic samples was investigated in air by a non-contactmode atomic force microscope (AFM) model C-21 (Danish Micro Engineering), mounting a DualScope Probe Scanner 95-50. AFM measurements were also utilized to investigate the thickness of the samples by scratching a small area of the samples and acquiring an image of the borderline between the exposed substrate and the pristine film. The observations were performed with a Si3N4 pyramidal tip with a curvature radius lower than 50 nm.

#### **2.2.3 Infrared analysis (FT-IR)**

FT-IR spectra of the samples were recorded in the 4000-400 cm-1 range using a Jasco FT-IR 660 Plus spectrometer with a resolution of 16 cm-1. The sample cell and the interferometer were evacuated to remove the adsorption peaks of water and atmospheric gases.

## **2.2.4 UV-Visible spectroscopy (UV-Vis)**

A Jasco V-570 dual-beam spectrophotometer was used to perform the UV-visible absorption measurements in the 200-800 nm range, with a resolution of 2 nm.

Physical Vapour Deposition Techniques for Producing Advanced Organic Chemical Sensors 283

The measurement system (Gaslab 20.1; IFAK, Magdeburg) is equipped with a flow chamber, containing four coated quartz crystals, a reference quartz crystal and a thermocouple. The chamber was thermostatted at 20 ± 0.1 °C. The QCM chamber is connected with two massflow controllers (Brooks 5850S) for controlling the flow rate of analyte mixture and pure nitrogen, respectively. During the analyses QCM frequency was measured versus time every 1 s. The ethanol vapour and NO gas were supplied by SAPIO Srl in gas cylinders with

Despite the prolonged studies and applications of metal phthalocyanines (MPcs) during the last forty years, the scientific research on these organic semiconductors is still vivid owing to their unique properties, such as thermal stability, chemical inertness, and biocompatibility. MPcs are thus interesting for several applications, including chemical sensing, photoconducting agents, photovoltaic cell elements, nonlinear optics, and electrocatalysis. In the gas sensing field, MPcs are mainly used as electrical gas sensors, because of the conductivity changes induced by the adsorption of oxidizing or reducing gases such as NOx, halogens, and ammonia (Snow & Barger, 1989). More recently, MPcs have also been successfully tested as opto-sensing materials for the detection of volatile organic compounds

The classical deposition methods of thin organic films such as spin coating, dip coating and the sol-gel method cannot be easily applied to the production of MPc coatings owing to the low solubility of these compounds in organic solvents. The most widely used deposition method for MPcs is then become the high vacuum evaporation (VE), which allows to avoid the drawbacks of the derivatization of MPcs aimed at improving their solubility. Plasmabased deposition methods have been also reported to succeed in depositing MPc films for gas sensing applications: plasma polymerized CuPc was deposited on piezoelectric quartz crystals (Kurosawa et al., 1990) and plasma was also used for the activation of CuPc

In this section the results obtained in the production of phthalocyanine-based sensing films by means of the novel Glow-Discharge-induced Sublimation (GDS) technique are reported. In particular the authors report the experimental data concerning the physical characterization and the sensing capabilities of two MPcs: copper phthalocyanine (CuPc) and zinc phthalocyanine (ZnPc). Figure 1 displays the structures of the two compounds.

Fig. 1. Structure of: (left) copper phthalocyanine (CuPc); (right) zinc phthalocyanine (ZnPc).

a certified concentration of 504 ppm and 150 ppm, respectively.

molecules during the evaporation process (Choi et al., 1996).

**3. Phthalocyanines** 

(VOCs) (Spadavecchia et al., 2006).

## **2.3 Sensing properties**

The sensing capabilities of the different sensing thin films were analysed by Electrical, Optical and Piezo-electrical (QCM) transducing methods.

## **2.3.1 Electrical sensors**

The electrical responses were measured on samples grown onto two interdigitated combs of 1301 gold electrodes (650 + 651) deposited on 30mm×30mm×1mm silica slides (Maggioni et al., 2008). The experimental setup used to evaluate the gas-sensing capabilities of the films is equipped with two mass flow controllers (MFCs): the former allows to control the flow rate of analytes and the latter, which controls the flow rate of pure nitrogen, is used to suitably dilute the gas mixture to the desired composition. After heating the samples up to 150 ◦C under a nitrogen flow, the electrical current at fixed voltage (1 V) was measured every 6 s after setting the flow of both controllers at the desired values. For all the measurements reported the total flow was 1000 sccm. The gas mixtures in the cylinders used for the measurements were N2 +NO2 (0.98±0.05) ppm and N2 +NO (98±3) ppm.

## **2.3.2 Optical sensors**

The analysis chamber for the optical measurements of the CuPc samples to ethanol was the same used for the electrical measurements (Maggioni et al., 2008). The changes of reflectance upon ethanol vapour exposure was measured in the wavelength range 250-800 nm.

FT-IR response of the CuPc films was measured using a N2+NO2 500 ppm mixture. After the evacuation of the sample cell, the spectra were collected before gas admittance, after 10 min of gas exposure (105 Pa total pressure) and then after 15 h of cell evacuation.

The optical sensing performances of the porphyrin samples were measured in a purposebuilt testing apparatus in which the sample chamber is placed inside a Jasco V-570 Spectrophotometer. The sample chamber is connected to two MFCs: the former (AERA FC-7800CD) allows to control the flow rate of the analyte mixture and the latter (AERA FC-7700CD) controls the flow rate of pure nitrogen. This apparatus allows to record both the whole 200-800 nm spectra and dynamic behaviours at fixed absorption wavelengths in different vapour concentration atmospheres.

## **2.3.3 Mass sensors (QCM)**

Sensing measurements were performed using AT-cut quartzes with a fundamental frequency of 10 MHz and a crystal diameter of 8 mm. QCM sensors are mass transducers where the frequency of oscillation, for small increases of mass, linearly changes according to the Sauerbrey equation

$$
\Delta f = k\_{\mathcal{q}} \Delta m
$$

where the Δ*f* is the frequency variation and Δ*m the* increase in mass. The quartz constant is experimentally estimated to be *k*q=-0.46 Hz ng-1. This value provides a nominal mass resolution of 1.6 ng Hz-1, considering a minimum reliable frequency measurement of 1 Hz. To control the amount of the deposited films, QCM frequency was monitored during the deposition process: a total frequency variation of Δ*f*=-20 ± 0.5 kHz was obtained for all samples produced.

The measurement system (Gaslab 20.1; IFAK, Magdeburg) is equipped with a flow chamber, containing four coated quartz crystals, a reference quartz crystal and a thermocouple. The chamber was thermostatted at 20 ± 0.1 °C. The QCM chamber is connected with two massflow controllers (Brooks 5850S) for controlling the flow rate of analyte mixture and pure nitrogen, respectively. During the analyses QCM frequency was measured versus time every 1 s. The ethanol vapour and NO gas were supplied by SAPIO Srl in gas cylinders with a certified concentration of 504 ppm and 150 ppm, respectively.

## **3. Phthalocyanines**

282 Advances in Chemical Sensors

The sensing capabilities of the different sensing thin films were analysed by Electrical,

The electrical responses were measured on samples grown onto two interdigitated combs of 1301 gold electrodes (650 + 651) deposited on 30mm×30mm×1mm silica slides (Maggioni et al., 2008). The experimental setup used to evaluate the gas-sensing capabilities of the films is equipped with two mass flow controllers (MFCs): the former allows to control the flow rate of analytes and the latter, which controls the flow rate of pure nitrogen, is used to suitably dilute the gas mixture to the desired composition. After heating the samples up to 150 ◦C under a nitrogen flow, the electrical current at fixed voltage (1 V) was measured every 6 s after setting the flow of both controllers at the desired values. For all the measurements reported the total flow was 1000 sccm. The gas mixtures in the cylinders used for the

The analysis chamber for the optical measurements of the CuPc samples to ethanol was the same used for the electrical measurements (Maggioni et al., 2008). The changes of reflectance

FT-IR response of the CuPc films was measured using a N2+NO2 500 ppm mixture. After the evacuation of the sample cell, the spectra were collected before gas admittance, after 10 min

The optical sensing performances of the porphyrin samples were measured in a purposebuilt testing apparatus in which the sample chamber is placed inside a Jasco V-570 Spectrophotometer. The sample chamber is connected to two MFCs: the former (AERA FC-7800CD) allows to control the flow rate of the analyte mixture and the latter (AERA FC-7700CD) controls the flow rate of pure nitrogen. This apparatus allows to record both the whole 200-800 nm spectra and dynamic behaviours at fixed absorption wavelengths in

Sensing measurements were performed using AT-cut quartzes with a fundamental frequency of 10 MHz and a crystal diameter of 8 mm. QCM sensors are mass transducers where the frequency of oscillation, for small increases of mass, linearly changes according to

*Δf=kqΔm*

where the Δ*f* is the frequency variation and Δ*m the* increase in mass. The quartz constant is experimentally estimated to be *k*q=-0.46 Hz ng-1. This value provides a nominal mass resolution of 1.6 ng Hz-1, considering a minimum reliable frequency measurement of 1 Hz. To control the amount of the deposited films, QCM frequency was monitored during the deposition process: a total frequency variation of Δ*f*=-20 ± 0.5 kHz was obtained for all

upon ethanol vapour exposure was measured in the wavelength range 250-800 nm.

of gas exposure (105 Pa total pressure) and then after 15 h of cell evacuation.

**2.3 Sensing properties** 

**2.3.1 Electrical sensors** 

**2.3.2 Optical sensors** 

different vapour concentration atmospheres.

**2.3.3 Mass sensors (QCM)** 

the Sauerbrey equation

samples produced.

Optical and Piezo-electrical (QCM) transducing methods.

measurements were N2 +NO2 (0.98±0.05) ppm and N2 +NO (98±3) ppm.

Despite the prolonged studies and applications of metal phthalocyanines (MPcs) during the last forty years, the scientific research on these organic semiconductors is still vivid owing to their unique properties, such as thermal stability, chemical inertness, and biocompatibility. MPcs are thus interesting for several applications, including chemical sensing, photoconducting agents, photovoltaic cell elements, nonlinear optics, and electrocatalysis. In the gas sensing field, MPcs are mainly used as electrical gas sensors, because of the conductivity changes induced by the adsorption of oxidizing or reducing gases such as NOx, halogens, and ammonia (Snow & Barger, 1989). More recently, MPcs have also been successfully tested as opto-sensing materials for the detection of volatile organic compounds (VOCs) (Spadavecchia et al., 2006).

The classical deposition methods of thin organic films such as spin coating, dip coating and the sol-gel method cannot be easily applied to the production of MPc coatings owing to the low solubility of these compounds in organic solvents. The most widely used deposition method for MPcs is then become the high vacuum evaporation (VE), which allows to avoid the drawbacks of the derivatization of MPcs aimed at improving their solubility. Plasmabased deposition methods have been also reported to succeed in depositing MPc films for gas sensing applications: plasma polymerized CuPc was deposited on piezoelectric quartz crystals (Kurosawa et al., 1990) and plasma was also used for the activation of CuPc molecules during the evaporation process (Choi et al., 1996).

In this section the results obtained in the production of phthalocyanine-based sensing films by means of the novel Glow-Discharge-induced Sublimation (GDS) technique are reported. In particular the authors report the experimental data concerning the physical characterization and the sensing capabilities of two MPcs: copper phthalocyanine (CuPc) and zinc phthalocyanine (ZnPc). Figure 1 displays the structures of the two compounds.

Fig. 1. Structure of: (left) copper phthalocyanine (CuPc); (right) zinc phthalocyanine (ZnPc).

Physical Vapour Deposition Techniques for Producing Advanced Organic Chemical Sensors 285

optically respond to alcohol vapours (Spadavecchia et al., 2006; Maggioni et al., 2007): Figure 4 shows the changes of optical reflectance of a CuPc sample to ethanol vapours at a wavelength in the Q band (λ=580nm). The responses are characterized by a fast signal increase: t50 response time (defined as the time taken for the signal intensity to reach 50% of its final saturated value) is 7 s; moreover, when the ethanol vapour stream is switched off, the original signal yield is almost completely recovered after few tens of seconds (t50= 12 s). Taking into account that the measurements are performed at room temperature, this

Fig. 3. (left) Electrical responses of a GDS CuPc film to NO2-containing mixtures in the subppm range. Fitting lines are reported for each curve. (right) Slopes of the electrical responses

The slope of the calibration curve of this sensor (Figure 4, right) decreases at concentrations higher than 7000 ppm, suggesting a progressive saturation of the sites available for the analyte molecules. As the optical signal for a concentration of 1500 ppm is about 20 and the corresponding noise is about ±1, the detection limit for the present experimental configuration is around 200 ppm. This limit can be easily improved by changing some experimental parameters such as the active film surface area. In order to investigate the adsorption processes involved in the response of the CuPc sample to ethanol, the Langmuir isotherm model has been adopted. The linear fit of the Langmuir adsorption (Figure 4, right,

red line) was done after neglecting the three data points at the higher concentrations.

Fig. 4. (left) Optical response of CuPc sample towards different ethanol vapour

towards ethanol at λ=580 nm.

concentrations at λ=580 nm. (right) Calibration curve and Langmuir plot of CuPc sample

evidence becomes particularly interesting.

of a GDS CuPc film as a function of NO2 concentration.

## **3.1 Copper phthalocyanine (CuPc) deposited by GDS technique**

## **3.1.1 Physical and chemical properties**

GDS technique allowed to produce CuPc films which feature much rougher morphologies than VE ones and very high porosity (Maggioni et al., 2005a).

Fig. 2. SEM micrographs of VE (left) and GDS (right) CuPc samples.

In fact, as shown in Figure 2, while the VE films are relatively flat and compact, the GDS films are characterized by a rough structure with big grains (2 *ì*m and more) and a high porosity inside and between the grains. Nitrogen physisorption measurements confirm the high porosity of the GDS films (Maggioni et al., 2005a). Specific Surface Area (SSA) values, calculated for the GDS films, are higher than those for the VE samples: in fact, the VE films feature 31 m2/g against 155 m2/g of the GDS films. The pore size distribution curve of the VE films features a mono-modal and sharp distribution centered at around 4 nm. By contrast, GDS films are characterized by a bimodal distribution with micropores (≈ 2 nm) and mesopores (> 30 nm). These structural characteristics of GDS films bring about their higher sensing capabilities with respect to the VE ones, owing to the increased interactive area between sensing receptors and analyte molecules.

## **3.1.2 Sensing properties**

The sensing capabilities of the CuPc samples have been tested by several transducing methods. It is well known that the adsorption of reducing or oxidizing gases such as NOx changes the electrical conductivity of CuPc, but both response time and recovery time are very long (in the range of hours). This problem can be solved by monitoring the change of the electrical current during the first tens of seconds after the gas exposure: this allows to shorten significantly the response times. Figure 3 reports that the slope of the curve current *versus* time is related to the gas concentration showing a linear relationship for low concentrations and a small divergence from linearity at higher concentrations (see Figure 3, right).

The detection limit of CuPc GDS sensor is around few tens of ppb for NO2 and few ppm for NO (Maggioni et al., 2008). A strong change of the optical properties of MPc films can occur when they are exposed to a specific gas/vapour, because the electronic transitions within their π-aromatic system and the π –π interactions between contiguous MPc molecules can be highly affected by the interaction with the gas molecules. MPcs have been shown to

GDS technique allowed to produce CuPc films which feature much rougher morphologies

In fact, as shown in Figure 2, while the VE films are relatively flat and compact, the GDS films are characterized by a rough structure with big grains (2 *ì*m and more) and a high porosity inside and between the grains. Nitrogen physisorption measurements confirm the high porosity of the GDS films (Maggioni et al., 2005a). Specific Surface Area (SSA) values, calculated for the GDS films, are higher than those for the VE samples: in fact, the VE films feature 31 m2/g against 155 m2/g of the GDS films. The pore size distribution curve of the VE films features a mono-modal and sharp distribution centered at around 4 nm. By contrast, GDS films are characterized by a bimodal distribution with micropores (≈ 2 nm) and mesopores (> 30 nm). These structural characteristics of GDS films bring about their higher sensing capabilities with respect to the VE ones, owing to the increased interactive

The sensing capabilities of the CuPc samples have been tested by several transducing methods. It is well known that the adsorption of reducing or oxidizing gases such as NOx changes the electrical conductivity of CuPc, but both response time and recovery time are very long (in the range of hours). This problem can be solved by monitoring the change of the electrical current during the first tens of seconds after the gas exposure: this allows to shorten significantly the response times. Figure 3 reports that the slope of the curve current *versus* time is related to the gas concentration showing a linear relationship for low concentrations and a small divergence from linearity at higher concentrations (see Figure 3,

The detection limit of CuPc GDS sensor is around few tens of ppb for NO2 and few ppm for NO (Maggioni et al., 2008). A strong change of the optical properties of MPc films can occur when they are exposed to a specific gas/vapour, because the electronic transitions within their π-aromatic system and the π –π interactions between contiguous MPc molecules can be highly affected by the interaction with the gas molecules. MPcs have been shown to

**3.1 Copper phthalocyanine (CuPc) deposited by GDS technique** 

Fig. 2. SEM micrographs of VE (left) and GDS (right) CuPc samples.

area between sensing receptors and analyte molecules.

**3.1.2 Sensing properties** 

right).

than VE ones and very high porosity (Maggioni et al., 2005a).

**3.1.1 Physical and chemical properties** 

optically respond to alcohol vapours (Spadavecchia et al., 2006; Maggioni et al., 2007): Figure 4 shows the changes of optical reflectance of a CuPc sample to ethanol vapours at a wavelength in the Q band (λ=580nm). The responses are characterized by a fast signal increase: t50 response time (defined as the time taken for the signal intensity to reach 50% of its final saturated value) is 7 s; moreover, when the ethanol vapour stream is switched off, the original signal yield is almost completely recovered after few tens of seconds (t50= 12 s). Taking into account that the measurements are performed at room temperature, this evidence becomes particularly interesting.

Fig. 3. (left) Electrical responses of a GDS CuPc film to NO2-containing mixtures in the subppm range. Fitting lines are reported for each curve. (right) Slopes of the electrical responses of a GDS CuPc film as a function of NO2 concentration.

The slope of the calibration curve of this sensor (Figure 4, right) decreases at concentrations higher than 7000 ppm, suggesting a progressive saturation of the sites available for the analyte molecules. As the optical signal for a concentration of 1500 ppm is about 20 and the corresponding noise is about ±1, the detection limit for the present experimental configuration is around 200 ppm. This limit can be easily improved by changing some experimental parameters such as the active film surface area. In order to investigate the adsorption processes involved in the response of the CuPc sample to ethanol, the Langmuir isotherm model has been adopted. The linear fit of the Langmuir adsorption (Figure 4, right, red line) was done after neglecting the three data points at the higher concentrations.

Fig. 4. (left) Optical response of CuPc sample towards different ethanol vapour concentrations at λ=580 nm. (right) Calibration curve and Langmuir plot of CuPc sample towards ethanol at λ=580 nm.

Physical Vapour Deposition Techniques for Producing Advanced Organic Chemical Sensors 287

minor and is mainly a broadening of the peaks in the range from 1500 to 1000 cm-1. The broadening is due to the disorder caused by the NO2 doping in the film structure. As to the interaction between NO2 and the central Cu atom, the Cu-O-NO vibration was reported at 1460 cm-1 (Sadaoka et al., 1982). In the spectrum of the NO2-doped film a broad peak between 1440 and 1510 cm-1 with a maximum at 1457 cm-1 is found, while in the other spectra only one peak at 1466 cm-1 appears. Since the peak at 1457 cm-1 is distinct enough from that at 1466 cm-1, it can be ascribed to a Cu-O-NO vibration thus suggesting a partial

Fig. 6. FT-IR spectra of GDS film exposed to N2+NO2 500 ppm mixture in the regions from 800 to 670 cm-1 (a) and from 1550 to 850 cm-1 (b). In each Figure: before NO2 exposure

The same test was also performed for the VE CuPc, but no change of the IR features of this sample was found. According to Sadaoka et al. (Sadaoka et al., 1982) this behaviour is due to the very low NO2 partial pressure used for these measurements (50 Pa against 105 Pa used in

AFM images of the surface of a ZnPc film deposited on silicon substrate are shown in Figure 7. In spite of the high smoothness of the substrate, the morphology of the ZnPc film is

Like for CuPc films, a comparison with VE ZnPc films (Iwatsu et al., 1980) points out that the surface of GDS ZnPc films is much more porous. The roughness profile of the film surface and the mean height are 77±3 nm and 510±70 nm, respectively (Maggioni et al., 2007). Nitrogen physisorption measurements highlighted microporous solids in which the size of the micropores, as derived from the slit-like Horvath–Kawazoe model (Horvath &

characterized by the presence of grains with lateral sizes of hundreds of nanometers.

(upper); sample in N2+NO2 (middle); after NO2 exposure (lower).

**3.2 Zinc phthalocyanine (ZnPc) deposited by GDS technique** 

**3.2.1 Physical and chemical properties** 

interaction between Cu and NO2.

the cited work).

The linear correlation is very good at low concentrations while a strong deviation is observed at increasing concentration. The good linear correlation obtained at low concentrations shows that the Langmuir adsorption model, even with its limited assumptions, provides a basic understanding of the ethanol/CuPc interaction process.

The better performances of the GDS CuPc films as compared to the VE ones are pointed out by their response to NO when used as quartz crystal microbalance (QCM) sensors Figure 5 shows the dynamic response of two CuPc-coated QCMs to 40 ppm of NO: the response is measured as a change of QCM resonance frequency upon adsorption of the analyte molecules.

The VE-coated QCM exhibits responses characterized by low intensity (Figure 5, left) and its response remains the same for all the tested concentrations in the range from 10 to 150 ppm (Figure 5, right). On the contrary, the response of GDS-coated QCM is well pronounced even down to 10 ppm of NO and it is related to the NO concentration: the relation is quite linear in the low concentration range, while above 60 ppm the response reaches a constant value, which corresponds to the saturation of the available NO recognition sites.

The NO detection limit in the present configuration is around few ppm, taking into account that the noise is around 1 Hz. Moreover, GDS samples are characterized by short response (t50 = 6 s) and recovery times (t50 = 9 s).

Fig. 5. (left) QCM response of GDS (black) and VE (red) CuPc samples in the presence of 40 ppm of NO. (right) Calibration curve of GDS sample in the NO range from 10 to 150 ppm

A transduction mechanism which can be also used for gas sensing involves the effects of analyte molecules on the IR features of the CuPc samples (Maggioni et al., 2005b). Figure 6 shows the FT-IR spectra of a GDS CuPc sample exposed to a N2+NO2 500 ppm atmosphere.

As can be seen, the NO2 adsorption gives rise to clear spectral changes and these changes are almost completely reversible. Considering the out-of-plane vibrations, which are the most sensitive to the chemical environment, the γ(C-H) peak at 726 cm-1 decreases in intensity and shifts to 730 cm-1. Moreover the relative intensity of the peak at 780 cm-1 slightly increases with respect to that of the peak at 773 cm-1. These changes indicate that the NO2 molecule strongly interacts with the benzene rings of the CuPc molecule, in agreement with the results of Sadaoka et al. (Sadaoka et al., 1982), which found that the sites for NO2 adsorption were the ligand � electron systems for CuPc and H2Pc and the central metal atoms for FePc and CoPc, respectively. Considering the in-plane-vibrations, the NO2 effect is

The linear correlation is very good at low concentrations while a strong deviation is observed at increasing concentration. The good linear correlation obtained at low concentrations shows that the Langmuir adsorption model, even with its limited assumptions, provides a basic understanding of the ethanol/CuPc interaction process.

The better performances of the GDS CuPc films as compared to the VE ones are pointed out by their response to NO when used as quartz crystal microbalance (QCM) sensors Figure 5 shows the dynamic response of two CuPc-coated QCMs to 40 ppm of NO: the response is measured

The VE-coated QCM exhibits responses characterized by low intensity (Figure 5, left) and its response remains the same for all the tested concentrations in the range from 10 to 150 ppm (Figure 5, right). On the contrary, the response of GDS-coated QCM is well pronounced even down to 10 ppm of NO and it is related to the NO concentration: the relation is quite linear in the low concentration range, while above 60 ppm the response reaches a constant

The NO detection limit in the present configuration is around few ppm, taking into account that the noise is around 1 Hz. Moreover, GDS samples are characterized by short response

Fig. 5. (left) QCM response of GDS (black) and VE (red) CuPc samples in the presence of 40 ppm of NO. (right) Calibration curve of GDS sample in the NO range from 10 to 150 ppm

A transduction mechanism which can be also used for gas sensing involves the effects of analyte molecules on the IR features of the CuPc samples (Maggioni et al., 2005b). Figure 6 shows the FT-IR spectra of a GDS CuPc sample exposed to a N2+NO2 500 ppm atmosphere. As can be seen, the NO2 adsorption gives rise to clear spectral changes and these changes are almost completely reversible. Considering the out-of-plane vibrations, which are the most sensitive to the chemical environment, the γ(C-H) peak at 726 cm-1 decreases in intensity and shifts to 730 cm-1. Moreover the relative intensity of the peak at 780 cm-1 slightly increases with respect to that of the peak at 773 cm-1. These changes indicate that the NO2 molecule strongly interacts with the benzene rings of the CuPc molecule, in agreement with the results of Sadaoka et al. (Sadaoka et al., 1982), which found that the sites for NO2 adsorption were the ligand � electron systems for CuPc and H2Pc and the central metal atoms for FePc and CoPc, respectively. Considering the in-plane-vibrations, the NO2 effect is

as a change of QCM resonance frequency upon adsorption of the analyte molecules.

value, which corresponds to the saturation of the available NO recognition sites.

(t50 = 6 s) and recovery times (t50 = 9 s).

minor and is mainly a broadening of the peaks in the range from 1500 to 1000 cm-1. The broadening is due to the disorder caused by the NO2 doping in the film structure. As to the interaction between NO2 and the central Cu atom, the Cu-O-NO vibration was reported at 1460 cm-1 (Sadaoka et al., 1982). In the spectrum of the NO2-doped film a broad peak between 1440 and 1510 cm-1 with a maximum at 1457 cm-1 is found, while in the other spectra only one peak at 1466 cm-1 appears. Since the peak at 1457 cm-1 is distinct enough from that at 1466 cm-1, it can be ascribed to a Cu-O-NO vibration thus suggesting a partial interaction between Cu and NO2.

Fig. 6. FT-IR spectra of GDS film exposed to N2+NO2 500 ppm mixture in the regions from 800 to 670 cm-1 (a) and from 1550 to 850 cm-1 (b). In each Figure: before NO2 exposure (upper); sample in N2+NO2 (middle); after NO2 exposure (lower).

The same test was also performed for the VE CuPc, but no change of the IR features of this sample was found. According to Sadaoka et al. (Sadaoka et al., 1982) this behaviour is due to the very low NO2 partial pressure used for these measurements (50 Pa against 105 Pa used in the cited work).

#### **3.2 Zinc phthalocyanine (ZnPc) deposited by GDS technique**

#### **3.2.1 Physical and chemical properties**

AFM images of the surface of a ZnPc film deposited on silicon substrate are shown in Figure 7. In spite of the high smoothness of the substrate, the morphology of the ZnPc film is characterized by the presence of grains with lateral sizes of hundreds of nanometers.

Like for CuPc films, a comparison with VE ZnPc films (Iwatsu et al., 1980) points out that the surface of GDS ZnPc films is much more porous. The roughness profile of the film surface and the mean height are 77±3 nm and 510±70 nm, respectively (Maggioni et al., 2007). Nitrogen physisorption measurements highlighted microporous solids in which the size of the micropores, as derived from the slit-like Horvath–Kawazoe model (Horvath &

Physical Vapour Deposition Techniques for Producing Advanced Organic Chemical Sensors 289

sensor is linear in the low concentration range, then it reaches a saturation value at higher concentrations. It has been shown that in the case of methanol and isopropanol the response keeps linear in the whole concentration range, although the concentrations are higher than those of ethanol (Maggioni et al., 2007). These differences in responses are due to the

Porphyrins, the "pigments of life", are perhaps the most important and widespread class of natural pigments. Nature provides a variety of tetrapyrrolic pigments with a similar ligand core, but differing in the metal centre and the side groups attached to the porphyrin rings. For instance, in heme the metal centre is an iron ion: in the case of chlorophyll it is

Tetraphenyl porphyrins have assumed a peculiar role in different fields of disciplines, ranging from molecular semiconductors to non-linear optics: in particular, their physical and chemical properties make these compounds good sensing materials for detecting

Fig. 9. Structure of: (left) free (H2TPP); (centre) cobalt, and (right) iron chloride (FeTPPCl)

In fact, porphyrins are stable compounds and their physical and chemical properties can be finely tuned by simple modifications of their basic molecular structure. Moreover porphyrins present optical absorption and fluorescence bands in the visible region related to the electronic transitions of the aromatic systems. The interactions of analytes with porphyrin thin films affect both the optically active transitions of the single molecule and the π–π interactions between macrocycles, giving rise to detectable changes of the optical absorption spectra. This property allowed to develop several opto-chemical sensing

In this section the cases of three 5,10,15,20 tetraphenyl porphyrins will be treated: free (H2TPP), cobalt (CoTPP), and iron chloride (FeTPPCl). Figure 9 displays the structures of

In order to be exploited as sensing materials, porphyrin compounds usually need to be deposited as solid films onto an appropriate substrate: a large number of chemical techniques (solvent casting, spin coating, Langmuir–Blodgett, electropolymerization, self assembled monolayers) have been studied for this purpose. Less attention has been paid to

different gaseous species, from NOx and HCl to VOCs-(Brunik et al., 1999).

different affinity between the ZnPc macromolecular ring and the vapour molecules.

**4. Porphyrins** 

magnesium and in vitamin B12 it is cobalt.

5,10,15,20 tetraphenyl porphyrins.

procedures (Rakow & Suslick, 2000).

**4.1 Porphyrins deposited by VE technique** 

these compounds.

Kawazoe, 1983), ranges between 1.1 and 1.4 nm. Specific surface area (SSA) of the sample is 58 m2/g: this value is much higher than that of VE ZnPc sample, which is less than 1 m2/g.

Fig. 7. AFM images of the ZnPc film: plane view (left) and 3D view (right) of the surface.

#### **3.2.2 Sensing properties**

Like for CuPc, the optical properties of ZnPc samples can be highly affected by the interaction with gas/vapour molecules. The exposure to ethanol, methanol and isopropanol (vapours changes the optical absorption in the spectral region 500–800 nm, where QI band (peaked at 680 nm) and QII band (peaked at 620 nm) of ZnPc are placed (Maggioni et al., 2007). Responses of GDS-deposited ZnPc films exposed to different ethanol concentrations are reported in Figure 8. The measurements were performed by recording the integral area (I) of the absorption spectrum in the selected wavelength region versus time.

Fig. 8. (left) Optical response of GDS ZnPc sample in the presence of different ethanol vapour concentrations in the spectral region 500-800 nm. (right) Calibration curve of GDS ZnPc sample in the ethanol range from 1.5x103 to 8x104 ppm.

The response is fast and reversible: at an ethanol concentration of 5000 ppm the response time t90 (defined as the time taken for the signal intensity to reach 90% of its final saturated value) is around 60 s. The calibration curve obtained from the dynamic curves is reported in Figure 8., right. This curve represents the response of the sensors (intended as the difference, ∆I, between the intensity in presence of the ethanol vapour, Igas, and in presence of dry air, Iair, i.e. ∆I = Igas −Iair) in the presence of different vapour concentrations. The response of the sensor is linear in the low concentration range, then it reaches a saturation value at higher concentrations. It has been shown that in the case of methanol and isopropanol the response keeps linear in the whole concentration range, although the concentrations are higher than those of ethanol (Maggioni et al., 2007). These differences in responses are due to the different affinity between the ZnPc macromolecular ring and the vapour molecules.

## **4. Porphyrins**

288 Advances in Chemical Sensors

Kawazoe, 1983), ranges between 1.1 and 1.4 nm. Specific surface area (SSA) of the sample is 58 m2/g: this value is much higher than that of VE ZnPc sample, which is less than 1 m2/g.

Fig. 7. AFM images of the ZnPc film: plane view (left) and 3D view (right) of the surface.

(I) of the absorption spectrum in the selected wavelength region versus time.

Fig. 8. (left) Optical response of GDS ZnPc sample in the presence of different ethanol vapour concentrations in the spectral region 500-800 nm. (right) Calibration curve of GDS

The response is fast and reversible: at an ethanol concentration of 5000 ppm the response time t90 (defined as the time taken for the signal intensity to reach 90% of its final saturated value) is around 60 s. The calibration curve obtained from the dynamic curves is reported in Figure 8., right. This curve represents the response of the sensors (intended as the difference, ∆I, between the intensity in presence of the ethanol vapour, Igas, and in presence of dry air, Iair, i.e. ∆I = Igas −Iair) in the presence of different vapour concentrations. The response of the

ZnPc sample in the ethanol range from 1.5x103 to 8x104 ppm.

Like for CuPc, the optical properties of ZnPc samples can be highly affected by the interaction with gas/vapour molecules. The exposure to ethanol, methanol and isopropanol (vapours changes the optical absorption in the spectral region 500–800 nm, where QI band (peaked at 680 nm) and QII band (peaked at 620 nm) of ZnPc are placed (Maggioni et al., 2007). Responses of GDS-deposited ZnPc films exposed to different ethanol concentrations are reported in Figure 8. The measurements were performed by recording the integral area

**3.2.2 Sensing properties** 

Porphyrins, the "pigments of life", are perhaps the most important and widespread class of natural pigments. Nature provides a variety of tetrapyrrolic pigments with a similar ligand core, but differing in the metal centre and the side groups attached to the porphyrin rings. For instance, in heme the metal centre is an iron ion: in the case of chlorophyll it is magnesium and in vitamin B12 it is cobalt.

Tetraphenyl porphyrins have assumed a peculiar role in different fields of disciplines, ranging from molecular semiconductors to non-linear optics: in particular, their physical and chemical properties make these compounds good sensing materials for detecting different gaseous species, from NOx and HCl to VOCs-(Brunik et al., 1999).

Fig. 9. Structure of: (left) free (H2TPP); (centre) cobalt, and (right) iron chloride (FeTPPCl) 5,10,15,20 tetraphenyl porphyrins.

In fact, porphyrins are stable compounds and their physical and chemical properties can be finely tuned by simple modifications of their basic molecular structure. Moreover porphyrins present optical absorption and fluorescence bands in the visible region related to the electronic transitions of the aromatic systems. The interactions of analytes with porphyrin thin films affect both the optically active transitions of the single molecule and the π–π interactions between macrocycles, giving rise to detectable changes of the optical absorption spectra. This property allowed to develop several opto-chemical sensing procedures (Rakow & Suslick, 2000).

In this section the cases of three 5,10,15,20 tetraphenyl porphyrins will be treated: free (H2TPP), cobalt (CoTPP), and iron chloride (FeTPPCl). Figure 9 displays the structures of these compounds.

#### **4.1 Porphyrins deposited by VE technique**

In order to be exploited as sensing materials, porphyrin compounds usually need to be deposited as solid films onto an appropriate substrate: a large number of chemical techniques (solvent casting, spin coating, Langmuir–Blodgett, electropolymerization, self assembled monolayers) have been studied for this purpose. Less attention has been paid to

Physical Vapour Deposition Techniques for Producing Advanced Organic Chemical Sensors 291

due to the presence of retained chloroform, as confirmed by their disappearance after a

The optical sensing capabilities of the produced porphyrin films have been tested by the following procedure: i) an optical absorption spectrum of the samples before analyte exposure was acquired in the 350–750 nm range under a nitrogen atmosphere (*A*o); ii) the samples underwent a thorough conditioning procedure with exposures to saturated vapours of the various analytes (i.e. MeOH, EtOH and 2-propanol); iii) another absorption

This procedure allows to calculate the percentage absorbance variation patterns Δ*A*% = ((*A*t – *A*o)/*A*o × 100) as a function of wavelength. This way it is possible to find, for each (sensing material-analyte) couple, the maximum optical absorbance change and the related wavelength. Figure 11 shows the case of VE and SPIN samples of H2TPP exposed to ethanol.

Fig. 11. Absorbance spectra before exposure (solid line) and after conditioning procedure in EtOH (dashed line) (a and b) and corresponding Δ*A*% patterns (A and B) of VE and SPIN

Figure 12 reports the intensities of the maximum optical absorbance change (Δ*A*%)MAX of different porphyrin films grown by SPIN and VE technique after the conditioning procedure. VE samples showed higher optical absorbance changes than the SPIN ones with all the tested organic vapours, pointing out greater interactions with the analyte molecules: this different behaviour can be ascribed to the production method. In fact,

combined purification treatment of 10 h at 100 °C in vacuum (*P*=10−3 Pa).

spectrum was acquired (*A*t) when a steady absorbance signal was obtained.

**4.1.2 Sensing properties** 

samples of H2TPP, respectively.

the use of physical deposition techniques in the production of sensing materials, in spite of their ability to significantly improve the recognition and sensing capabilities of the molecular precursors by growing highly competitive organic architectures.

The VE technique introduces several advantages in the production of porphyrin-based sensing thin films such as greater reproducibility, higher uniformity and stricter control of the film thickness in comparison with standard chemical techniques. Moreover vacuum evaporation, featuring the peculiarity to produce thin solid films without using any extraneous compound, allows to deposit samples characterized by a much higher purity than the common chemical techniques which need to use solvents. This aspect is particularly important in the gas sensing field because the unpredictable effects of the retained solvent on the final response of the sensing device are manifold, going from the occupation of adsorption sites to the interference in the analyte/material interaction.

In this section the chemical properties of VE thin films of H2TPP, CoTPP, and (Fe(TPP)Cl) are described. Moreover their optical sensing capabilities upon exposure to vapours of three different alcohols, namely methanol (MeOH), ethanol (EtOH) and isopropanol (2- propanol) are discussed and compared to the sensing capabilities of conventional spin coated (SPIN) films (Bernini et al., 2008).

#### **4.1.1 Chemical properties**

Infrared (FT-IR) spectrum of VE H2TPP film is reported in Figure 10 together with those of SPIN film and H2TPP starting powder.

Fig. 10. FT-IR spectra of powder (lower line), VE film (middle line) and SPIN film (upper line) of H2TPP.

The main results of this analysis are related to the molecular integrity and purity of the samples. In particular: i) VE samples are composed by integer porphyrin molecules, as demonstrated by the fact that FT-IR spectrum of the VE film shows all the characteristic peaks of this compound (Tonezzer et al., 2007b); ii) VE samples are characterized by high purity, as highlighted by the absence in the VE spectrum of any extraneous peak. On the contrary, the FT-IR spectrum of the SPIN sample shows two additional intense peaks at 2920 and 2854 cm−<sup>1</sup>

due to the presence of retained chloroform, as confirmed by their disappearance after a combined purification treatment of 10 h at 100 °C in vacuum (*P*=10−3 Pa).

#### **4.1.2 Sensing properties**

290 Advances in Chemical Sensors

the use of physical deposition techniques in the production of sensing materials, in spite of their ability to significantly improve the recognition and sensing capabilities of the

The VE technique introduces several advantages in the production of porphyrin-based sensing thin films such as greater reproducibility, higher uniformity and stricter control of the film thickness in comparison with standard chemical techniques. Moreover vacuum evaporation, featuring the peculiarity to produce thin solid films without using any extraneous compound, allows to deposit samples characterized by a much higher purity than the common chemical techniques which need to use solvents. This aspect is particularly important in the gas sensing field because the unpredictable effects of the retained solvent on the final response of the sensing device are manifold, going from the occupation of

In this section the chemical properties of VE thin films of H2TPP, CoTPP, and (Fe(TPP)Cl) are described. Moreover their optical sensing capabilities upon exposure to vapours of three different alcohols, namely methanol (MeOH), ethanol (EtOH) and isopropanol (2- propanol) are discussed and compared to the sensing capabilities of conventional spin coated (SPIN)

Infrared (FT-IR) spectrum of VE H2TPP film is reported in Figure 10 together with those of

Fig. 10. FT-IR spectra of powder (lower line), VE film (middle line) and SPIN film (upper

The main results of this analysis are related to the molecular integrity and purity of the samples. In particular: i) VE samples are composed by integer porphyrin molecules, as demonstrated by the fact that FT-IR spectrum of the VE film shows all the characteristic peaks of this compound (Tonezzer et al., 2007b); ii) VE samples are characterized by high purity, as highlighted by the absence in the VE spectrum of any extraneous peak. On the contrary, the FT-IR spectrum of the SPIN sample shows two additional intense peaks at 2920 and 2854 cm−<sup>1</sup>

molecular precursors by growing highly competitive organic architectures.

adsorption sites to the interference in the analyte/material interaction.

films (Bernini et al., 2008).

**4.1.1 Chemical properties** 

line) of H2TPP.

SPIN film and H2TPP starting powder.

The optical sensing capabilities of the produced porphyrin films have been tested by the following procedure: i) an optical absorption spectrum of the samples before analyte exposure was acquired in the 350–750 nm range under a nitrogen atmosphere (*A*o); ii) the samples underwent a thorough conditioning procedure with exposures to saturated vapours of the various analytes (i.e. MeOH, EtOH and 2-propanol); iii) another absorption spectrum was acquired (*A*t) when a steady absorbance signal was obtained.

This procedure allows to calculate the percentage absorbance variation patterns Δ*A*% = ((*A*t – *A*o)/*A*o × 100) as a function of wavelength. This way it is possible to find, for each (sensing material-analyte) couple, the maximum optical absorbance change and the related wavelength. Figure 11 shows the case of VE and SPIN samples of H2TPP exposed to ethanol.

Fig. 11. Absorbance spectra before exposure (solid line) and after conditioning procedure in EtOH (dashed line) (a and b) and corresponding Δ*A*% patterns (A and B) of VE and SPIN samples of H2TPP, respectively.

Figure 12 reports the intensities of the maximum optical absorbance change (Δ*A*%)MAX of different porphyrin films grown by SPIN and VE technique after the conditioning procedure. VE samples showed higher optical absorbance changes than the SPIN ones with all the tested organic vapours, pointing out greater interactions with the analyte molecules: this different behaviour can be ascribed to the production method. In fact,

Physical Vapour Deposition Techniques for Producing Advanced Organic Chemical Sensors 293

In this section the chemical properties of GDS films of CoTPP and Fe(TPP)Cl are reported. Moreover their optical sensing capabilities upon exposure to ethanol vapours are discussed

Figure 13 reports the visible absorption spectra of GDS and SPIN films of Fe(TPP)Cl, compared with the absorption spectrum of Fe(TPP)Cl compound dissolved in chloroform (CHCl3) (10−5 M). The presence of the characteristic Fe(TPP)Cl optical B and Q bands into the GDS and SPIN films clearly indicates the presence of integer molecules into both the solid films: this experimental evidence points out the absence of molecular iron decomplexation effects into the produced films and demonstrates the integrity of Fe(TPP)Cl

As optical properties of porphyrin compounds can be affected by changes in the aggregation state, UV-Vis spectra give also information about the film structures: in particular the main differences between GDS and SPIN spectra (Q bands of the samples show different redshifting and both the B and Q bands show different broadening) clearly indicate different molecular arrangements within the two differently deposited Fe(TPP)Cl films (Tonezzer et

Fig. 13. UV–Visible (UV–Vis) absorption spectra of Fe(TPP)Cl in 10−5 M CHCl3 solution (a)

In order to analyze the structures of differently deposited porphyrin films, several analyses were performed (Tonezzer, 2007c): among them SEM images give an important information about the samples morphology. Figure 14 shows SEM images of VE and GDS

SEM pictures point out the two very different morphologies, which depend on the different deposition techniques. In particular VE sample shows surfaces characterized by flat morphology while GDS samples show a rough surface composed of micro-size

and compared to the sensing capabilities of VE and SPIN films.

**4.2.1 Physical and chemical properties** 

and Fe(TPP)Cl films: SPIN (b) and GDS (c).

molecules within GDS films.

al., 2009).

CoTPP films.

domains.

while vacuum evaporation yields thin impurity-free films, SPIN technique produces samples containing traces of chloroform which hinder the adsorption process of the alcohol vapour molecules.

Furthermore dynamic responses of the samples have been investigated by recording the absorbance changes of the samples at the wavelengths corresponding to the maximum absorbance variations upon exposure to several cycles in different vapour atmospheres and concentrations. Figure 12 reports the evolution of the Soret band intensity of the Fe(TPP)ClVE and Fe(TPP)ClSPIN as a function of time upon EtOH exposure (3000 ppm). Both response and recovery phases are conducted at room temperature without any thermal treatment.

The samples responses are characterized by an increase of the absorbance after the EtOH vapour stream is switched on, followed by a slow increase until a saturation value is reached. At 700 s, the EtOH vapour stream is switched off and a dry nitrogen flush is activated: the absorbance rapidly decreases and the original band intensity is recovered. The sensing response of VE sample is characterized by a higher intensity (about three times) and faster behavior than the SPIN one. In particular, VE sample shows *t*50 much shorter (16 s) than SPIN ones (25 s). The faster responses of VE samples can be also ascribed to their higher purity: in fact, this feature is expected to improve their reactivity towards analyte molecules.

Fig. 12. (left) Maxima intensities of optical absorbance changes (Δ*A*%)MAX of VE and SPIN samples of H2TPP, CoTPP and Fe(TPP)Cl owing to conditioning procedure in MeOH, EtOH and 2-propanol. (right) Absorbance variation of the Soret band (410 nm) of VE (•) and SPIN (○) films of Fe(TPP)Cl as a function of time during exposure to 3000 ppm of EtOH.

It can be concluded that vacuum evaporation, allowing to produce more competitive sensing thin films than conventional spin coating, can play an important role in the development of new sensing elements.

#### **4.2 Porphyrins deposited by GDS technique**

Like for phthalocyanines, Glow-Discharge-induced Sublimation (GDS) represents a new promising technique in the production of porphyrin-based sensing materials, allowing to obtain thin films with both high purity and very large surface area to volume ratio.

In this section the chemical properties of GDS films of CoTPP and Fe(TPP)Cl are reported. Moreover their optical sensing capabilities upon exposure to ethanol vapours are discussed and compared to the sensing capabilities of VE and SPIN films.

## **4.2.1 Physical and chemical properties**

292 Advances in Chemical Sensors

while vacuum evaporation yields thin impurity-free films, SPIN technique produces samples containing traces of chloroform which hinder the adsorption process of the

Furthermore dynamic responses of the samples have been investigated by recording the absorbance changes of the samples at the wavelengths corresponding to the maximum absorbance variations upon exposure to several cycles in different vapour atmospheres and concentrations. Figure 12 reports the evolution of the Soret band intensity of the Fe(TPP)ClVE and Fe(TPP)ClSPIN as a function of time upon EtOH exposure (3000 ppm). Both response and recovery phases are conducted at room temperature without any

The samples responses are characterized by an increase of the absorbance after the EtOH vapour stream is switched on, followed by a slow increase until a saturation value is reached. At 700 s, the EtOH vapour stream is switched off and a dry nitrogen flush is activated: the absorbance rapidly decreases and the original band intensity is recovered. The sensing response of VE sample is characterized by a higher intensity (about three times) and faster behavior than the SPIN one. In particular, VE sample shows *t*50 much shorter (16 s) than SPIN ones (25 s). The faster responses of VE samples can be also ascribed to their higher purity: in fact, this feature is expected to improve their reactivity towards analyte

Fig. 12. (left) Maxima intensities of optical absorbance changes (Δ*A*%)MAX of VE and SPIN samples of H2TPP, CoTPP and Fe(TPP)Cl owing to conditioning procedure in MeOH, EtOH and 2-propanol. (right) Absorbance variation of the Soret band (410 nm) of VE (•) and SPIN

It can be concluded that vacuum evaporation, allowing to produce more competitive sensing thin films than conventional spin coating, can play an important role in the

Like for phthalocyanines, Glow-Discharge-induced Sublimation (GDS) represents a new promising technique in the production of porphyrin-based sensing materials, allowing to

(○) films of Fe(TPP)Cl as a function of time during exposure to 3000 ppm of EtOH.

obtain thin films with both high purity and very large surface area to volume ratio.

development of new sensing elements.

**4.2 Porphyrins deposited by GDS technique** 

alcohol vapour molecules.

thermal treatment.

molecules.

Figure 13 reports the visible absorption spectra of GDS and SPIN films of Fe(TPP)Cl, compared with the absorption spectrum of Fe(TPP)Cl compound dissolved in chloroform (CHCl3) (10−5 M). The presence of the characteristic Fe(TPP)Cl optical B and Q bands into the GDS and SPIN films clearly indicates the presence of integer molecules into both the solid films: this experimental evidence points out the absence of molecular iron decomplexation effects into the produced films and demonstrates the integrity of Fe(TPP)Cl molecules within GDS films.

As optical properties of porphyrin compounds can be affected by changes in the aggregation state, UV-Vis spectra give also information about the film structures: in particular the main differences between GDS and SPIN spectra (Q bands of the samples show different redshifting and both the B and Q bands show different broadening) clearly indicate different molecular arrangements within the two differently deposited Fe(TPP)Cl films (Tonezzer et al., 2009).

Fig. 13. UV–Visible (UV–Vis) absorption spectra of Fe(TPP)Cl in 10−5 M CHCl3 solution (a) and Fe(TPP)Cl films: SPIN (b) and GDS (c).

In order to analyze the structures of differently deposited porphyrin films, several analyses were performed (Tonezzer, 2007c): among them SEM images give an important information about the samples morphology. Figure 14 shows SEM images of VE and GDS CoTPP films.

SEM pictures point out the two very different morphologies, which depend on the different deposition techniques. In particular VE sample shows surfaces characterized by flat morphology while GDS samples show a rough surface composed of micro-size domains.

Physical Vapour Deposition Techniques for Producing Advanced Organic Chemical Sensors 295

Figure 16 shows the sensing responses of Fe(TPP)Cl samples towards EtOH: in particular Figure 16, left depicts the dynamic response of GDS- and SPIN-deposited Fe(TPP)Cl thin films upon exposure to 2300 ppm of EtOH and Figure 16, right reports the calibration curves of the samples. Fe(TPP)Cl responses are reported as the signal-to-noise ratio (SNR). SNR, expressed as the ratio between the signal intensity (I) and the signal noise (N), represents in fact a powerful dimensionless parameter which, indicating how the signal is higher than the noise by which it is affected, releases the sensing responses from the transducing method

Fig. 16, left highlights the higher and faster response of GDS sample with respect to the SPIN one. In particular GDS film shows a SNR value of 33 which is 15 times higher than that

Fig. 16, right shows the plots of the response magnitude versus EtOH concentration for GDS and SPIN samples: GDS samples are characterized by optical responses much more intense than SPIN ones at all the tested concentrations. Moreover, the more leaning slope of the calibration curve of GDS sample in comparison with that of SPIN one highlights the higher

The optical behaviours of all the CoTPP and Fe(TPP)Cl samples presented in this section are characterized by a complete and fast recovery: at this regard it is worth to note that both CoTPP and Fe(TPP)Cl optical responses have been measured at room temperature. In fact, many authors have reported the requirement to heat porphyrin samples during the recovery phase, in order to obtain complete recovery in reasonable times (Pedrosa et al., 2002). Nevertheless if, on one hand, heating procedure gives the advantages to improve the recovery degree and speed, on the other hand it induces thermal stresses into the samples

Fig. 16. (left) Single exposure/recovery cycle for GDS (a) and SPIN (b) films of Fe(TPP)Cl exposed to 2300ppm EtOH (T = 20°C). (right) Calibration curves of SPIN and GDS films of

These results-point out the fundamental role of the deposition technique used for the growth of sensing materials characterized by suited molecular structures and consequently

Fe(TPP)Cl upon exposures to EtOH atmospheres over the range 1800–4500 ppm.

of SPIN one (SNR = 2.2) and t90 value shorter (43 s) than SPIN one (63 s).

used to examine the sample sensing capabilities.

sensitivity of the GDS sample than the SPIN one.

which can result in film degradation.

by very competitive sensing capabilities.

Fig. 14. SEM images of VE (left) and GDS (right) films of CoTPP (magnification 2200x).

This unconventional rough morphologies are of paramount importance in sensing field: in fact they increase the surface interaction area between film and analyte molecules improving significantly the sensing capabilities of the final material.

## **4.2.2 Sensing properties**

The optical sensing capabilities of the CoTPP and Fe(TPP)Cl thin films produced by GDS technique upon exposure to ethanol vapours were tested by using the same procedure already adopted for VE samples (see section 4.1.2).

Fig. 15. Concentration dependence of SPIN (a), VE (b) and GDS (c) CoTPP films upon exposures to EtOH atmospheres over the concentration range 1100–4500 ppm.

Figure 15 shows the responses of different-CoTPP samples (GDS, VE and SPIN) upon exposure to EtOH in the range from 1100 to 4500 ppm. The responses increase with increasing the analyte concentration for all the samples. The response of the GDS sample is really fast, shows complete recovery and is much more intense than those of both SPIN (approximately 15 times) and VE (more than 10 times) samples at all the tested concentrations. This much higher response intensity can be attributed to its high surface morphology, as evidenced by the SEM images reported in Figure 14 (Tonezzer, 2007a).

Fig. 14. SEM images of VE (left) and GDS (right) films of CoTPP (magnification 2200x).

significantly the sensing capabilities of the final material.

already adopted for VE samples (see section 4.1.2).

**4.2.2 Sensing properties** 

This unconventional rough morphologies are of paramount importance in sensing field: in fact they increase the surface interaction area between film and analyte molecules improving

The optical sensing capabilities of the CoTPP and Fe(TPP)Cl thin films produced by GDS technique upon exposure to ethanol vapours were tested by using the same procedure

Fig. 15. Concentration dependence of SPIN (a), VE (b) and GDS (c) CoTPP films upon exposures to EtOH atmospheres over the concentration range 1100–4500 ppm.

Figure 15 shows the responses of different-CoTPP samples (GDS, VE and SPIN) upon exposure to EtOH in the range from 1100 to 4500 ppm. The responses increase with increasing the analyte concentration for all the samples. The response of the GDS sample is really fast, shows complete recovery and is much more intense than those of both SPIN (approximately 15 times) and VE (more than 10 times) samples at all the tested concentrations. This much higher response intensity can be attributed to its high surface morphology, as evidenced by the SEM images reported in Figure 14 (Tonezzer, 2007a).

Figure 16 shows the sensing responses of Fe(TPP)Cl samples towards EtOH: in particular Figure 16, left depicts the dynamic response of GDS- and SPIN-deposited Fe(TPP)Cl thin films upon exposure to 2300 ppm of EtOH and Figure 16, right reports the calibration curves of the samples. Fe(TPP)Cl responses are reported as the signal-to-noise ratio (SNR). SNR, expressed as the ratio between the signal intensity (I) and the signal noise (N), represents in fact a powerful dimensionless parameter which, indicating how the signal is higher than the noise by which it is affected, releases the sensing responses from the transducing method used to examine the sample sensing capabilities.

Fig. 16, left highlights the higher and faster response of GDS sample with respect to the SPIN one. In particular GDS film shows a SNR value of 33 which is 15 times higher than that of SPIN one (SNR = 2.2) and t90 value shorter (43 s) than SPIN one (63 s).

Fig. 16, right shows the plots of the response magnitude versus EtOH concentration for GDS and SPIN samples: GDS samples are characterized by optical responses much more intense than SPIN ones at all the tested concentrations. Moreover, the more leaning slope of the calibration curve of GDS sample in comparison with that of SPIN one highlights the higher sensitivity of the GDS sample than the SPIN one.

The optical behaviours of all the CoTPP and Fe(TPP)Cl samples presented in this section are characterized by a complete and fast recovery: at this regard it is worth to note that both CoTPP and Fe(TPP)Cl optical responses have been measured at room temperature. In fact, many authors have reported the requirement to heat porphyrin samples during the recovery phase, in order to obtain complete recovery in reasonable times (Pedrosa et al., 2002). Nevertheless if, on one hand, heating procedure gives the advantages to improve the recovery degree and speed, on the other hand it induces thermal stresses into the samples which can result in film degradation.

Fig. 16. (left) Single exposure/recovery cycle for GDS (a) and SPIN (b) films of Fe(TPP)Cl exposed to 2300ppm EtOH (T = 20°C). (right) Calibration curves of SPIN and GDS films of Fe(TPP)Cl upon exposures to EtOH atmospheres over the range 1800–4500 ppm.

These results-point out the fundamental role of the deposition technique used for the growth of sensing materials characterized by suited molecular structures and consequently by very competitive sensing capabilities.

Physical Vapour Deposition Techniques for Producing Advanced Organic Chemical Sensors 297

previously described ones (Melegari et al., 2008) are: (i) the four alkyl chains at the lower rim were removed, to minimize nonspecific interactions; (ii) four methyl groups were introduced in the apical positions to deepen the cavity and increase the strength of CH-*π*

In order to be exploited as sensing materials, cavitands are usually deposited as thin solid films by wet deposition techniques; in particular, spin-coating method is commonly employed (Feresenbet et al., 2004). Unfortunately wet deposition methods suffer from several drawbacks such as inhomogeneous surface morphology and uncontrollable thickness. Moreover, insoluble compounds cannot be deposited by such solution techniques. To date, the most common method of depositing insoluble compounds by solution techniques consists of decorating them with peripheral alkyl chains to improve their solubility: nevertheless this molecular derivatization introduces dispersion interactions that dilute the specific analyte response, thus significantly decreasing sensor selectivity. As alternatives, Langmuir-Blodgett and Langmuir-Schafer deposition methods provide ordered and reproducible monolayers: however, when a bulk response is required by the sensing layer (like in QCM sensors), deposition of thicker permeable

In this section we report the employment of the VE technique for the deposition of Tiiii[H, CH3, Ph] cavitand which represents the ultimate receptor for supramolecular mass sensing of short chain alcohols. VE technique is also used for producing TSiiii[H, CH3, Ph]-sensing coatings for comparison. In the case of the Tiiii and TSiiii cavitands, the absence of alkyl chains at the lower rim implies a different film deposition method with respect to the standard ones (spray- and –spin coating) owing to the high insolubility of

VE technique overcomes solubility problems, allowing for the direct formation of films from solid materials. The lack of residual solvent during the deposition process assures the formation of high purity films: this represents a basic requirement in the gas sensing field, because of the unpredictable effects of the retained solvent on the final response of the sensor, including occupation of adsorption sites and interference in analyte/material interactions. Moreover, VE technique guarantees good reproducibility, high uniformity, and homogeneity and provides accurate control over both the growth rate and the final

In order to investigate the purity of the VE cavitand films and to exclude the presence of impurities derived from cavitand decomposition, infrared spectra of Tiiii and TSiiii films have been collected and compared with the spectra of the respective powders (Figure 18). Tiiii and TSiiii samples are characterized by similar infrared spectra due to the close resemblance of their molecular structures showing some differences in certain positions. Both Tiiii and TSiiii samples show all the characteristic peaks of the respective powders, indicating the absence of damaged molecules in both the VE samples, within the detection

interactions.

coatings is mandatory.

the two cavitands.

thickness of the samples.

**5.1.1 Physical and chemical properties** 

**5.1 Cavitands deposited by VE technique** 

## **5. Cavitands**

Molecular recognition of gases is an emerging area of chemistry (Rudkevich, 2004). Cavitands, together with cyclodextrins and calixarenes, are the most studied receptors for gas/vapor sensing because of their outstanding host-guest properties, which are tunable for recognizing different classes of analytes (Cram & Cram, 1994). Among them phosphonate cavitands represent a third-generation emerging class of synthetic receptors whose molecular recognition properties toward alcohols and water were thoroughly demonstrated at the molecular scale by several analytical techniques such as ESI-MS and X-ray crystallography (Melegari et. al., 2008).

Fig. 17. Structures (above) and 3D CPK models (below) of tetraphosphonate Tiiii[H,CH3,Ph] and tetrathiophosphonate TSiiii[H,CH3,Ph] cavitands.

The compounds described in this section are two novel phosphonate cavitands devoted to the detection of short chain alcohols synthetized by Dalcanale and co-workers (Biavardi et al., 2009): tetraphosphonate (Tiiii) and tetrathiophosphonate (TSiiii). Figure 17 shows the structures of the two compounds: they present an open, conformationally rigid cavity, delimited by four inward oriented P=O/P=S bridging groups at the upper rim. The structure of these cavitands is born out by previous studies on the sensing properties of tetraphosphonate cavitands toward short chain alcohols. Substitution of the four P=O groups with the P=S moieties completely prevents complexation by eliminating H-bonding interactions between the cavitand and the analyte. The structural similarity of the two cavitands allows for a valid comparison of the influence of molecular recognition on sensing performance. The structural variations of these novel compounds with respect to previously described ones (Melegari et al., 2008) are: (i) the four alkyl chains at the lower rim were removed, to minimize nonspecific interactions; (ii) four methyl groups were introduced in the apical positions to deepen the cavity and increase the strength of CH-*π* interactions.

## **5.1 Cavitands deposited by VE technique**

296 Advances in Chemical Sensors

Molecular recognition of gases is an emerging area of chemistry (Rudkevich, 2004). Cavitands, together with cyclodextrins and calixarenes, are the most studied receptors for gas/vapor sensing because of their outstanding host-guest properties, which are tunable for recognizing different classes of analytes (Cram & Cram, 1994). Among them phosphonate cavitands represent a third-generation emerging class of synthetic receptors whose molecular recognition properties toward alcohols and water were thoroughly demonstrated at the molecular scale by several analytical techniques such as ESI-MS and X-ray

Fig. 17. Structures (above) and 3D CPK models (below) of tetraphosphonate Tiiii[H,CH3,Ph]

The compounds described in this section are two novel phosphonate cavitands devoted to the detection of short chain alcohols synthetized by Dalcanale and co-workers (Biavardi et al., 2009): tetraphosphonate (Tiiii) and tetrathiophosphonate (TSiiii). Figure 17 shows the structures of the two compounds: they present an open, conformationally rigid cavity, delimited by four inward oriented P=O/P=S bridging groups at the upper rim. The structure of these cavitands is born out by previous studies on the sensing properties of tetraphosphonate cavitands toward short chain alcohols. Substitution of the four P=O groups with the P=S moieties completely prevents complexation by eliminating H-bonding interactions between the cavitand and the analyte. The structural similarity of the two cavitands allows for a valid comparison of the influence of molecular recognition on sensing performance. The structural variations of these novel compounds with respect to

and tetrathiophosphonate TSiiii[H,CH3,Ph] cavitands.

**5. Cavitands** 

crystallography (Melegari et. al., 2008).

In order to be exploited as sensing materials, cavitands are usually deposited as thin solid films by wet deposition techniques; in particular, spin-coating method is commonly employed (Feresenbet et al., 2004). Unfortunately wet deposition methods suffer from several drawbacks such as inhomogeneous surface morphology and uncontrollable thickness. Moreover, insoluble compounds cannot be deposited by such solution techniques. To date, the most common method of depositing insoluble compounds by solution techniques consists of decorating them with peripheral alkyl chains to improve their solubility: nevertheless this molecular derivatization introduces dispersion interactions that dilute the specific analyte response, thus significantly decreasing sensor selectivity. As alternatives, Langmuir-Blodgett and Langmuir-Schafer deposition methods provide ordered and reproducible monolayers: however, when a bulk response is required by the sensing layer (like in QCM sensors), deposition of thicker permeable coatings is mandatory.

In this section we report the employment of the VE technique for the deposition of Tiiii[H, CH3, Ph] cavitand which represents the ultimate receptor for supramolecular mass sensing of short chain alcohols. VE technique is also used for producing TSiiii[H, CH3, Ph]-sensing coatings for comparison. In the case of the Tiiii and TSiiii cavitands, the absence of alkyl chains at the lower rim implies a different film deposition method with respect to the standard ones (spray- and –spin coating) owing to the high insolubility of the two cavitands.

VE technique overcomes solubility problems, allowing for the direct formation of films from solid materials. The lack of residual solvent during the deposition process assures the formation of high purity films: this represents a basic requirement in the gas sensing field, because of the unpredictable effects of the retained solvent on the final response of the sensor, including occupation of adsorption sites and interference in analyte/material interactions. Moreover, VE technique guarantees good reproducibility, high uniformity, and homogeneity and provides accurate control over both the growth rate and the final thickness of the samples.

#### **5.1.1 Physical and chemical properties**

In order to investigate the purity of the VE cavitand films and to exclude the presence of impurities derived from cavitand decomposition, infrared spectra of Tiiii and TSiiii films have been collected and compared with the spectra of the respective powders (Figure 18).

Tiiii and TSiiii samples are characterized by similar infrared spectra due to the close resemblance of their molecular structures showing some differences in certain positions. Both Tiiii and TSiiii samples show all the characteristic peaks of the respective powders, indicating the absence of damaged molecules in both the VE samples, within the detection

Physical Vapour Deposition Techniques for Producing Advanced Organic Chemical Sensors 299

nonspecific dispersion interactions between the organic layer and EtOH and the large difference in the sensor responses can be attributed to the specific binding of EtOH by the

Fig. 19. (left) Change of the resonance frequency of VE Tiiii and TSiiii samples as a function of time during exposure to 25 ppm EtOH (T = 20 °C). (right) Change of the resonance frequency of SPIN Tiiii[C11H23, H, Ph] and VE Tiiii [H, CH3, Ph] samples at increasing

The superior sensing performances of VE Tiiii[H, CH3, Ph] sensor have been demonstrated by comparing its responses to those of the spin-coated long chain analogue Tiiii[C11H23, H, Ph], the best performer to date (Melegari, 2008). Figure 19, right shows that the combination of Tiiii[H,CH3, Ph] receptor with VE deposition doubles the EtOH sensor sensitivity in the 5- 50 ppm range in which nonspecific extracavity adsorption becomes significant. This result is remarkable, as it was obtained without altering the receptor site responsible for the alcohol

To investigate the interaction between EtOH and the VE films, the recovery phase of the samples was analyzed by Elovich kinetics. According to this model, the surface uncovering

> ( ) ( ) *<sup>1</sup> <sup>Θ</sup> t = ln t + K β*

where *β* and *K* are constants. This model is based on the assumption that the desorption probability of an analyte molecule during surface purging decreases exponentially as a function of the number of analyte molecules already desorbed. Thus, by assuming that the change of resonance frequency (Δ*F*) is related only to the interaction between EtOH and cavitand coatings, the value of Θ should be proportional to Δ*F*. Hence, by plotting Δ*F* as a

Figure 20 reports the plots of Δ*F* vs ln(*t*) for the recovery phases of the VE Tiiii and TSiiii

Θ during the recovery phase as a function of time is given by the following formula:

Tiiii cavitand (Tonezzer et al., 2008).

amount of EtOH.

complexation at the gas-solid interface.

function of ln(*t*), a linear relationship should be obtained.

layers, exposed to 5 and 100 ppm EtOH, respectively.

limits of the employed technique. Moreover, the lack of any additional peaks in the FT-IR spectra excludes the presence of extraneous compounds, demonstrating the high purity of the samples (Tonezzer et al., 2008).

Fig. 18. FT-IR spectra of vacuum evaporated (VE) films of Tiiii (A) and TSiiii (B) cavitands. The spectra of the corresponding starting powders pressed in KBr pellet (KBr) are also reported as reference.

The integrity of the cavitand molecules into the sublimated films as well as the purity of the samples demonstrate the viability of the VE technique for the deposition of thermally stable molecular receptors. AFM measurements of the surface of the Tiiii and TSiiii films point out their high uniformity and homogeneity. Moreover, the similar thickness of the two films (350 nm for Tiiii and 355 nm for TSiiii), which were deposited with the same process parameters, highlights the high reproducibility of the VE deposition process (Tonezzer, 2008).

#### **5.1.2 Sensing properties**

Sensing measurements were performed by exposing cavitand-coated QCMs to different EtOH concentrations (ranging from 5 to 200 ppm) and monitoring the shift of the QCM fundamental resonance frequency induced by the mass change as a function of time. Figure 19, left shows the responses of Tiiii- and TSiiii- coated QCM sensors exposed to 25 ppm of EtOH. The most significant result is the high difference in response intensity between Tiiii (Δ*F* = -60 Hz) and TSiiii (Δ*F* = -7.5 Hz) mainly due to the following two key factors: (i) the preorganized cavity which provides a free volume available for the analyte, pivotal for effective H-bonding; (ii) the presence of synergistic CH-*π* interactions with the *π*-basic cavity and the energetically equivalent H-bonding options between the P=O groups at the upper rim and the analyte (specific interactions). The resulting mode of interaction between ethanol and the cavity shows how the ethanol chain fits into the cavity with its methyl residue (CH-*π* interactions), while the OH moiety undergoes H-bonding interactions with the P=O groups. The effective contribution of the H-bonds is absent in the TSiiii cavitand, because the P=S group is much less polarized than the P=O one and, consequently, it is ineffective as H-bond acceptor. Therefore the observed TSiiii responses are due to

limits of the employed technique. Moreover, the lack of any additional peaks in the FT-IR spectra excludes the presence of extraneous compounds, demonstrating the high purity of

Fig. 18. FT-IR spectra of vacuum evaporated (VE) films of Tiiii (A) and TSiiii (B) cavitands. The spectra of the corresponding starting powders pressed in KBr pellet (KBr) are also

The integrity of the cavitand molecules into the sublimated films as well as the purity of the samples demonstrate the viability of the VE technique for the deposition of thermally stable molecular receptors. AFM measurements of the surface of the Tiiii and TSiiii films point out their high uniformity and homogeneity. Moreover, the similar thickness of the two films (350 nm for Tiiii and 355 nm for TSiiii), which were deposited with the same process parameters, highlights the high reproducibility of the VE deposition process (Tonezzer,

Sensing measurements were performed by exposing cavitand-coated QCMs to different EtOH concentrations (ranging from 5 to 200 ppm) and monitoring the shift of the QCM fundamental resonance frequency induced by the mass change as a function of time. Figure 19, left shows the responses of Tiiii- and TSiiii- coated QCM sensors exposed to 25 ppm of EtOH. The most significant result is the high difference in response intensity between Tiiii (Δ*F* = -60 Hz) and TSiiii (Δ*F* = -7.5 Hz) mainly due to the following two key factors: (i) the preorganized cavity which provides a free volume available for the analyte, pivotal for effective H-bonding; (ii) the presence of synergistic CH-*π* interactions with the *π*-basic cavity and the energetically equivalent H-bonding options between the P=O groups at the upper rim and the analyte (specific interactions). The resulting mode of interaction between ethanol and the cavity shows how the ethanol chain fits into the cavity with its methyl residue (CH-*π* interactions), while the OH moiety undergoes H-bonding interactions with the P=O groups. The effective contribution of the H-bonds is absent in the TSiiii cavitand, because the P=S group is much less polarized than the P=O one and, consequently, it is ineffective as H-bond acceptor. Therefore the observed TSiiii responses are due to

the samples (Tonezzer et al., 2008).

reported as reference.

**5.1.2 Sensing properties** 

2008).

nonspecific dispersion interactions between the organic layer and EtOH and the large difference in the sensor responses can be attributed to the specific binding of EtOH by the Tiiii cavitand (Tonezzer et al., 2008).

Fig. 19. (left) Change of the resonance frequency of VE Tiiii and TSiiii samples as a function of time during exposure to 25 ppm EtOH (T = 20 °C). (right) Change of the resonance frequency of SPIN Tiiii[C11H23, H, Ph] and VE Tiiii [H, CH3, Ph] samples at increasing amount of EtOH.

The superior sensing performances of VE Tiiii[H, CH3, Ph] sensor have been demonstrated by comparing its responses to those of the spin-coated long chain analogue Tiiii[C11H23, H, Ph], the best performer to date (Melegari, 2008). Figure 19, right shows that the combination of Tiiii[H,CH3, Ph] receptor with VE deposition doubles the EtOH sensor sensitivity in the 5- 50 ppm range in which nonspecific extracavity adsorption becomes significant. This result is remarkable, as it was obtained without altering the receptor site responsible for the alcohol complexation at the gas-solid interface.

To investigate the interaction between EtOH and the VE films, the recovery phase of the samples was analyzed by Elovich kinetics. According to this model, the surface uncovering Θ during the recovery phase as a function of time is given by the following formula:

$$
\Theta(t) = \frac{1}{\beta} \ln(t) + K
$$

where *β* and *K* are constants. This model is based on the assumption that the desorption probability of an analyte molecule during surface purging decreases exponentially as a function of the number of analyte molecules already desorbed. Thus, by assuming that the change of resonance frequency (Δ*F*) is related only to the interaction between EtOH and cavitand coatings, the value of Θ should be proportional to Δ*F*. Hence, by plotting Δ*F* as a function of ln(*t*), a linear relationship should be obtained.

Figure 20 reports the plots of Δ*F* vs ln(*t*) for the recovery phases of the VE Tiiii and TSiiii layers, exposed to 5 and 100 ppm EtOH, respectively.

Physical Vapour Deposition Techniques for Producing Advanced Organic Chemical Sensors 301

• Porphyrins. Porphyrin thin films were grown both by high vacuum evaporation (VE) and Glow-Discharge-induced Sublimation (GDS). In particular, three different porphyrin-based films were produced by VE technique: free (H2TPP), cobalt (CoTPP), and iron chloride (Fe(TPP)Cl) 5,10,15,20 *meso*-tetraphenyl porphyrins. Their chemical properties and optical sensing capabilities towards different alcohol vapours (methanol, ethanol and isopropanol) have been analyzed and compared to those of conventional spin coated (SPIN) ones. FT-IR spectra pointed out that VE technique allows to produce samples characterized by a much higher purity than the SPIN ones and the study of the optical responses highlighted that their higher purity provide them with much larger sensitivity and much faster response times than the SPIN

CoTPP and Fe(TPP)Cl films have been also produced by GDS technique: their chemical and physical features have been analyzed by FT-IR, UV-Vis and SEM analyses and their optical sensing capabilities towards ethanol have been measured. The chemical and physical characterization highlighted that GDS porphyrin films have a high purity, analogously to the VE samples, and feature an extremely large surface area to volume ratio, like the GDS phthalocyanine films. Optical sensing measurements confirmed that GDS technique allows to produce very competitive sensing films. In fact, both GDS CoTPP and Fe(TPP)Cl films feature much more intense response (up to 10 times) and much larger sensitivity with respect to conventional SPIN samples, high speed responses (t50 and t90 respectively of 10 s and 1 min approximately) and fast and

• Cavitands. VE deposition has been used for the first time in order to produce sensing thin films based on Tiiii[H, CH3, Ph] cavitand, an ultimate receptor for supramolecular sensing. The characterization of the Tiiii films highlighted that VE technique is particularly suitable for the deposition of supramolecular compounds: in fact, VE allows to grow thin films starting from compounds characterized by low solubility (a quite common feature of highly specific recognition molecules). As regard the sensing capabilities, Tiiii films have been tested towards very low concentration of ethanol vapours (5-50 ppm) demonstrating that VE technique provides a significant improvement of the performances of cavitand receptor as sensitive material in mass sensors. Moreover Elovich kinetics, used to elucidate the sorption processes occurring onto the layers, highlighted the high specificity of the Tiiii sensors, which reduces the

Chemical sensing requires an integrated approach, where both the molecular and the materials properties of the sensing layer must be finely tuned to achieve the desired

and isopropanol) showed stable and reproducible signals.

sensing elements.

complete recovery.

incidence of aspecific adsorption.

The sensing capabilities of CuPc samples have been investigated towards different analytes and by different transducing methods. As electrical sensors CuPc samples exhibit detection limits down to 0.1 ppm for NO2 (the standard attention level set by the European Union) and 10 ppm for NO and response times shorter than 30 s for 0.98 ppm NO2 and 98 ppm NO. CuPc films have been also tested as optical sensors towards ethanol vapours within 1500 - 34000 ppm concentration range highlighting high-speed responses (t50 = 7 s and t90 = 12 s) and fast recovery (t50 =12 s and t10 =38 s). It has been shown that CuPc-coated QCMs can be used for NO detection down to few ppm. ZnPc films, tested as optical sensors towards different alcohol vapours (methanol, ethanol

Fig. 20. Elovich recovery kinetics for vacuum-evaporated Tiiii and TSiiii films after exposing to 5 and 100 ppm EtOH, respectively.

As can be observed, the trends of Tiiii- and TSiiii-coated QCMs are characterized by different behaviors. The Tiiii behavior is approximately linear over the first 10 s, deviating from linearity at higher times. This indicates a change in activation energy dependent on surface coverage, supporting a two-step interaction process: (i) a weak interaction between analyte molecules and nonspecific sites and (ii) a stronger specific interaction due to cavity inclusion. In particular, the linear region represents the fast process when EtOH molecules are removed from the nonspecific sites of the Tiiii coating, while the nonlinear region represents slow EtOH release from the specific sites of the layer. By contrast, the TSiiiicoated QCM shows a completely linear behavior, indicating an interaction process dominated totally by nonspecific dispersion interactions.

These findings clearly indicate that TSiiii films are characterized mainly by nonspecific extracavity adsorption, while Tiiii layers feature intracavity complexation of EtOH.

## **6. Conclusions**

In this chapter the authors exemplified the fundamental role played by the film deposition technique in the attainment of both specific physical properties and final detecting capabilities of organic sensing thin films. In this respect two physical vapour deposition techniques used in the production of sensing materials have been discussed: High Vacuum Evaporation (VE) and Glow-Discharge-induced Sublimation (GDS),the latter being a novel patented plasma-based deposition technique.

The results obtained for three classes of different organic and metallo-organic compounds, i.e. phthalocyanines, porphyrins, and cavitands, are here summarized.

• Phthalocyanines. Two phthalocyanine compounds, namely copper (CuPc) and zinc (ZnPc) phthalocyanines, have been deposited by GDS technique and thoroughly characterized. Morphological measurements, performed by SEM and AFM techniques, point out that both ZnPc and CuPc thin films have a very high surface roughness, much higher than that found for similar films deposited by conventional methods. Moreover both GDS samples feature peculiar microporous molecular architectures, as evidenced by nitrogen physisorption measurements.

Fig. 20. Elovich recovery kinetics for vacuum-evaporated Tiiii and TSiiii films after exposing

As can be observed, the trends of Tiiii- and TSiiii-coated QCMs are characterized by different behaviors. The Tiiii behavior is approximately linear over the first 10 s, deviating from linearity at higher times. This indicates a change in activation energy dependent on surface coverage, supporting a two-step interaction process: (i) a weak interaction between analyte molecules and nonspecific sites and (ii) a stronger specific interaction due to cavity inclusion. In particular, the linear region represents the fast process when EtOH molecules are removed from the nonspecific sites of the Tiiii coating, while the nonlinear region represents slow EtOH release from the specific sites of the layer. By contrast, the TSiiiicoated QCM shows a completely linear behavior, indicating an interaction process

These findings clearly indicate that TSiiii films are characterized mainly by nonspecific

In this chapter the authors exemplified the fundamental role played by the film deposition technique in the attainment of both specific physical properties and final detecting capabilities of organic sensing thin films. In this respect two physical vapour deposition techniques used in the production of sensing materials have been discussed: High Vacuum Evaporation (VE) and Glow-Discharge-induced Sublimation (GDS),the latter being a novel

The results obtained for three classes of different organic and metallo-organic compounds,

• Phthalocyanines. Two phthalocyanine compounds, namely copper (CuPc) and zinc (ZnPc) phthalocyanines, have been deposited by GDS technique and thoroughly characterized. Morphological measurements, performed by SEM and AFM techniques, point out that both ZnPc and CuPc thin films have a very high surface roughness, much higher than that found for similar films deposited by conventional methods. Moreover both GDS samples feature peculiar microporous molecular architectures, as

i.e. phthalocyanines, porphyrins, and cavitands, are here summarized.

evidenced by nitrogen physisorption measurements.

extracavity adsorption, while Tiiii layers feature intracavity complexation of EtOH.

to 5 and 100 ppm EtOH, respectively.

**6. Conclusions** 

dominated totally by nonspecific dispersion interactions.

patented plasma-based deposition technique.

The sensing capabilities of CuPc samples have been investigated towards different analytes and by different transducing methods. As electrical sensors CuPc samples exhibit detection limits down to 0.1 ppm for NO2 (the standard attention level set by the European Union) and 10 ppm for NO and response times shorter than 30 s for 0.98 ppm NO2 and 98 ppm NO. CuPc films have been also tested as optical sensors towards ethanol vapours within 1500 - 34000 ppm concentration range highlighting high-speed responses (t50 = 7 s and t90 = 12 s) and fast recovery (t50 =12 s and t10 =38 s). It has been shown that CuPc-coated QCMs can be used for NO detection down to few ppm. ZnPc films, tested as optical sensors towards different alcohol vapours (methanol, ethanol and isopropanol) showed stable and reproducible signals.

• Porphyrins. Porphyrin thin films were grown both by high vacuum evaporation (VE) and Glow-Discharge-induced Sublimation (GDS). In particular, three different porphyrin-based films were produced by VE technique: free (H2TPP), cobalt (CoTPP), and iron chloride (Fe(TPP)Cl) 5,10,15,20 *meso*-tetraphenyl porphyrins. Their chemical properties and optical sensing capabilities towards different alcohol vapours (methanol, ethanol and isopropanol) have been analyzed and compared to those of conventional spin coated (SPIN) ones. FT-IR spectra pointed out that VE technique allows to produce samples characterized by a much higher purity than the SPIN ones and the study of the optical responses highlighted that their higher purity provide them with much larger sensitivity and much faster response times than the SPIN sensing elements.

CoTPP and Fe(TPP)Cl films have been also produced by GDS technique: their chemical and physical features have been analyzed by FT-IR, UV-Vis and SEM analyses and their optical sensing capabilities towards ethanol have been measured. The chemical and physical characterization highlighted that GDS porphyrin films have a high purity, analogously to the VE samples, and feature an extremely large surface area to volume ratio, like the GDS phthalocyanine films. Optical sensing measurements confirmed that GDS technique allows to produce very competitive sensing films. In fact, both GDS CoTPP and Fe(TPP)Cl films feature much more intense response (up to 10 times) and much larger sensitivity with respect to conventional SPIN samples, high speed responses (t50 and t90 respectively of 10 s and 1 min approximately) and fast and complete recovery.

• Cavitands. VE deposition has been used for the first time in order to produce sensing thin films based on Tiiii[H, CH3, Ph] cavitand, an ultimate receptor for supramolecular sensing. The characterization of the Tiiii films highlighted that VE technique is particularly suitable for the deposition of supramolecular compounds: in fact, VE allows to grow thin films starting from compounds characterized by low solubility (a quite common feature of highly specific recognition molecules). As regard the sensing capabilities, Tiiii films have been tested towards very low concentration of ethanol vapours (5-50 ppm) demonstrating that VE technique provides a significant improvement of the performances of cavitand receptor as sensitive material in mass sensors. Moreover Elovich kinetics, used to elucidate the sorption processes occurring onto the layers, highlighted the high specificity of the Tiiii sensors, which reduces the incidence of aspecific adsorption.

Chemical sensing requires an integrated approach, where both the molecular and the materials properties of the sensing layer must be finely tuned to achieve the desired

Physical Vapour Deposition Techniques for Producing Advanced Organic Chemical Sensors 303

Iwatsu, F., Kobayashi, T. & Uyeda, N. (1980). Solvent effects on crystal growth and transformation of zinc phthalocyanine, *J. Phys. Chem.*, Vol. 84, pp. 3223–3230. Li, B.; Xu, Y. & Choi, J. (1996). Applying Machine Learning Techniques, *Proceedings of ASME* 

Maggioni, G., Carturan, S., Boscarino, D., Della Mea, G. & Pieri U. (1997). Polyimide and

Maggioni, G., Quaranta, A., Carturan, S., Patelli, A., Tonezzer, M., Ceccato, R. & Della Mea,

Maggioni, G., Manera, M. G., Spadavecchia, J., Tonezzer, M., Carturan, S., Quaranta, A., de

Melegari, M., Suman, M., Pirondini, L., Moiani, D., Massera, C., Ugozzoli, F., Kalenius, E.,

Rakow, N.A. & Suslick, K.S. (2000). A colorimetric sensor array for odour visualization

Rudkevich, D.M., (2004). Emerging Supramolecular Chemistry of Gases, *Angewandte Chemie* 

Sadaoka, Y., Sakai, Y., Yamazoe, N. & Shiyama, T. (1982). Effect of adsorbed oxidative gases

Siegwart, R. (2001). Indirect Manipulation of a Sphere on a Flat Disk Using Force

Snow, A.W. & Barger, W.R. (1989) *Phthalocyanines: properties and applications* in C.C. Leznoff, A.B.P. Lever (Ed.), Vol. 1, pp. 341-392, VCH Publishers, Inc., New York, U.S.A. Spadavecchia, J., Ciccarella, G., Valli, L. & Rella, R. (2006). A novel multisensing optical

Tonezzer, M., Maggioni, G., Quaranta, A., Carturan, S. & Della Mea, G. (2007). Optical

Tonezzer, M., Quaranta, A., Maggioni, G., Carturan, S. & Della Mea, G. (2007). Optical

volatile organic compounds. *Sens. Actuators B,* Vol. 127, pp. 150-156. Maggioni, G., Carturan, S., Tonezzer, M., Quaranta, A. & Della Mea, G. (2008). Plasma-

substituted porphyrin, *J. Mater. Chem*., Vol. 12, pp. 2659–2664.

organic compounds, *Sens. Actuators B*, Vol. 113, pp. 516–525

sublimation, *Sens. Actuators B,* Vol. 122, pp. 613-619

*Actuators B*, Vol. 122, pp. 620-626

23-3, Phoenix, Arizona, USA, May 17-22, 2010

Sublimation, *Chem. Mater.*, Vol. 17, No. 7, pp. 1895-1904

responses, *Sens. Actuators B*, Vol. 131, pp. 496-503

*Nature*, Vol. 406 (6797), pp. 710-713

50, No. 6, pp. 457-462.

*- International Edition*, Vol 43 (5), pp. 558-571.

(December 2009), pp. 12-16, ISSN 1729-8806

*2010 4th International Conference on Energy Sustainability*, pp. 14-17, ISBN 842-6508-

platinum containing polyimide thin films obtained by vapour deposition polymerization; effects of thermal treatments, *Mater. Lett.*, Vol. 32, pp. 147-150. Maggioni, G., Quaranta, A., Carturan, S., Patelli, A., Tonezzer, M., Ceccato, R. & Della Mea,

G. (2005). Deposition of Copper Phthalocyanine Films by Glow-Discharge-induced

G. (2005). Deposition of Copper Phthalocyanine Films by Glow-Discharge-induced Sublimation for Gas Sensing Applications, *Surf. Coat. Technol.*, Vol. 200, pp. 476-480

Julián Fernández, C., Rella, R., Siciliano, P., Della Mea, G., Vasanelli, L. & Mazzoldi, P. (2007). Optical response of plasma-deposited zinc phthalocyanine films to

deposited copper phthalocyanine: A single gas-sensing material with multiple

Vainiotalo, P., Mulatier, J.C., Dutasta, J.-P. & Dalcanale, E. (2008). Supramolecular sensing with phosphonate cavitands, *Chem. Eur. J.*, Vol 14 (19), pp. 5772-5779 Pedrosa, J.M. , Dooling, C.M., Richardson, T.H., Hyde, R.K., Hunter, C.A., Martin, M.T. &

Camacho, L. (2002). The optical gas-sensing properties of an asymmetrically

on electrical properties of evaporated films of phthalocyanine. *Denki Kagaku*, Vol.

Information. *International Journal of Advanced Robotic Systems,* Vol.6, No.4,

approach based on a single phthalocyanine thin films to monitoring volatile

sensing properties of CoTPP thin films deposited by glow-discharge-induced

sensing responses of tetraphenyl porphyrins toward alcohol vapours: A comparison between vacuum evaporated and spin-coated thin films, *Sens.* 

properties. In this respect, thin film chemical sensors represents a particular challenge, taking into account that the analyte recognition is mediated by the layer properties of the coated receptors.

In this chapter, the authors describes different real cases in which significant advantages are introduced into the sensing field by a bottom-up approach in which recognition molecules are deposited in well-controlled and designed architectures by novel and accurate deposition techniques. The integrated approach used by the authors, where both the molecular and the materials properties of the sensing layers have been finely tuned, demonstrated to produce organic and metallo-organic sensors with improved performances over the existing ones.

This approach can be easily extended to many different classes of organic receptors, opening the way for the rational design of sensor materials tailored for the analytes to be detected.

## **7. Acknowledgment**

The authors thank the research group of Prof. Dalcanale of Industrial Chemistry at the Department of Organic and Industrial Chemistry of Parma University (Italy) for providing them Tiiii and TSiiiii cavitands. M.T. thanks personally Prof. Dalcanale for the helpful discussions on supramolecular receptors. The research leading to these results has received funding by Fondazione Cassa di Risparmio di Trento e Rovereto (CARITRO) within GREEN project and from the European Community's Seventh Framework Programme FP7/2007- 2013 under grant Agreement *Marie Curie 7th Framework Program - PCOFUND-GA-2008- 226070, acronomy "Progetto Trentino"* within PHOTOFUTURE project*.*

#### **8. References**


properties. In this respect, thin film chemical sensors represents a particular challenge, taking into account that the analyte recognition is mediated by the layer properties of the

In this chapter, the authors describes different real cases in which significant advantages are introduced into the sensing field by a bottom-up approach in which recognition molecules are deposited in well-controlled and designed architectures by novel and accurate deposition techniques. The integrated approach used by the authors, where both the molecular and the materials properties of the sensing layers have been finely tuned, demonstrated to produce organic and metallo-organic sensors with improved performances

This approach can be easily extended to many different classes of organic receptors, opening the way for the rational design of sensor materials tailored for the analytes to be detected.

The authors thank the research group of Prof. Dalcanale of Industrial Chemistry at the Department of Organic and Industrial Chemistry of Parma University (Italy) for providing them Tiiii and TSiiiii cavitands. M.T. thanks personally Prof. Dalcanale for the helpful discussions on supramolecular receptors. The research leading to these results has received funding by Fondazione Cassa di Risparmio di Trento e Rovereto (CARITRO) within GREEN project and from the European Community's Seventh Framework Programme FP7/2007- 2013 under grant Agreement *Marie Curie 7th Framework Program - PCOFUND-GA-2008-*

Arai, T. & Kragic, D. (1999). Variability of Wind and Wind Power, In: *Wind Power,* S.M. Muyeen, (Ed.), 289-321, Scyio, ISBN 978-953-7619-81-7, Vukovar, Croatia Bernini, R., Tonezzer, M., Mottola, F., Zeni, L., Quaranta, A., Maggioni, G., Carturan, S. &

cavitand-functionalized silicon surface, *JACS*, Vol. 131 (21), pp. 7447-7455. Brunik, J.A.J., Di Natale, C., Campo dall'Orto, V., Macagnano, A., Angelaccio, A., Sgarlata,

Cram, D. J. & Cram, J. M. (1994). *Container Molecules and Their Guests*, The Royal Society of

Feresenbet, E.B., Dalcanale, E., Dulcey, C. & Shenoy, D.K. (2004) Optical sensing of the

Horvath, G. & Kawazoe, K. (1983). Method for the calculation of effective pore size distribution in molecular sieve carbon, *J. Chem. Eng. Jpn.,* Vol. 16, pp. 470–475.

Della Mea, G. (2007). Volatile organic compounds detection using porphyrinbasedmetal-cladding leaky waveguides, *Sens. Actuators B*, Vol. 127, pp. 231-236 Biavardi, E., Favazza, M., Motta, A., Fragalà, I.L., Massera, C., Prodi, L., Montalti, M.,

Melegari, M., Condorelli, G.G. & Dalcanale, E. (2009). Molecular recognition on a

A., Hurst, J., Rezzano, I., Mascini, M. & D'Amico, A. (1999). Porphyrin thin films coated quartz microalances prepared by electropolymerization techniques, *Thin* 

selective interaction of aromatic vapors with cavitands *Sensors and Actuators B,* Vol.

*226070, acronomy "Progetto Trentino"* within PHOTOFUTURE project*.*

*Solid Films,* Vol. 354, pp. 245–250.

Chemistry, Cambridge, U.K.

97 pp. 211–220.

coated receptors.

over the existing ones.

**7. Acknowledgment** 

**8. References** 


**0**

**14**

*Italy*

**Drift Correction Methods for Gas Chemical**

<sup>2</sup>*SENSOR CNR-IDASC, Brescia, University of Brescia, Dept. of Chemistry and Physics*

The human sense of smell is a valuable tool in many areas of industry such as perfumery, food and drink production, clinical diagnosis, health and safety, environmental monitoring and process control (Gobbi et al., 2010; Vezzoli et al., 2008). Artificial olfaction mimics human olfaction by using arrays of gas chemical sensors combined with pattern recognition (PaRC) systems (Pearce et al., 2003). When a volatile compound comes into contact with the surface of the array, a set of physical changes modifies the properties of the material from which each sensor is composed. This perturbation can be measured, digitalized and used as a feature for the specific compound. A preliminary training phase collecting samples from known volatile compounds is used to train a selected PaRC algorithm in order to map each concentration of gas to the responses from the sensor array. The trained model is then used for identification during later measurements. The classification rate of the PaRC system determines the final

Gas sensor arrays represent a potentially low-cost and fast alternative to conventional analytical instruments such as gas chromatographs. However, successful applications of gas sensor arrays are still largely limited to specialized laboratories (Pardo & Sberveglieri, 2004). Lack of stability over time and the high cost of recalibration are factors which still limit the widespread adoption of artificial olfaction systems in real industrial setups (Padilla et al.,

The sensor drift consists of small and non-deterministic temporal variations of the sensor response when it is exposed to the same analytes under identical conditions (Holmberg et al., 1997). The main result is that the sensor's selectivity and sensitivity decrease. The gas sensor drift changes the way samples distribute in the data space, thus limiting the ability to operate over long periods. PaRC models become useless after a period of time, in some cases weeks or a few months. After that time the artificial olfaction system must be completely re-calibrated

It is still impossible to fabricate chemical sensors without drift. In fact, drift phenomena afflict almost all kinds of sensors (Chen & Chan, 2008; Owens & Wong, 2009; Polster et al., 2009).

**1. Introduction**

2010).

performance of the electronic olfaction system.

to ensure valid predictions (Aliwell et al., 2001).

**Sensors in Artificial Olfaction Systems:**

**Techniques and Challenges**

<sup>1</sup>*Department of Control and Computer Engineering*

S. Di Carlo1 and M. Falasconi2

*for Engineering and Materials, Brescia*

*Politecnico di Torino, Torino*


## **Drift Correction Methods for Gas Chemical Sensors in Artificial Olfaction Systems: Techniques and Challenges**

S. Di Carlo1 and M. Falasconi2

<sup>1</sup>*Department of Control and Computer Engineering Politecnico di Torino, Torino* <sup>2</sup>*SENSOR CNR-IDASC, Brescia, University of Brescia, Dept. of Chemistry and Physics for Engineering and Materials, Brescia Italy*

#### **1. Introduction**

304 Advances in Chemical Sensors

Tonezzer, M., (2007). *Production, characterization and testing of solid organic and metallo-organic* 

Tonezzer, M., Melegari, M., Maggioni, Milan, R., Della Mea, G. & Dalcanale, E. (2008).

Interactions in Ethanol Detection, *Chem. Mater.*, Vol. 20, pp*.* 6535–6542 Tonezzer, M., Maggioni, G., Quaranta, A., Carturan, S. & Della Mea, G. (2009). Growth,

Vacuum-Evaporated Cavitand Sensors: Dissecting Specific from Nonspecific

characterization and sensing capabilities of 5,10,15,20-meso-tetraphenyl iron (III) porphyrin chloride films obtained by means of a novel plasma-based deposition

*films for optical gas sensors*, University of Trento, Trento, Italy

technique, *Sens. Actuators B*, Vol. 136, pp. 290-296

The human sense of smell is a valuable tool in many areas of industry such as perfumery, food and drink production, clinical diagnosis, health and safety, environmental monitoring and process control (Gobbi et al., 2010; Vezzoli et al., 2008). Artificial olfaction mimics human olfaction by using arrays of gas chemical sensors combined with pattern recognition (PaRC) systems (Pearce et al., 2003). When a volatile compound comes into contact with the surface of the array, a set of physical changes modifies the properties of the material from which each sensor is composed. This perturbation can be measured, digitalized and used as a feature for the specific compound. A preliminary training phase collecting samples from known volatile compounds is used to train a selected PaRC algorithm in order to map each concentration of gas to the responses from the sensor array. The trained model is then used for identification during later measurements. The classification rate of the PaRC system determines the final performance of the electronic olfaction system.

Gas sensor arrays represent a potentially low-cost and fast alternative to conventional analytical instruments such as gas chromatographs. However, successful applications of gas sensor arrays are still largely limited to specialized laboratories (Pardo & Sberveglieri, 2004). Lack of stability over time and the high cost of recalibration are factors which still limit the widespread adoption of artificial olfaction systems in real industrial setups (Padilla et al., 2010).

The sensor drift consists of small and non-deterministic temporal variations of the sensor response when it is exposed to the same analytes under identical conditions (Holmberg et al., 1997). The main result is that the sensor's selectivity and sensitivity decrease. The gas sensor drift changes the way samples distribute in the data space, thus limiting the ability to operate over long periods. PaRC models become useless after a period of time, in some cases weeks or a few months. After that time the artificial olfaction system must be completely re-calibrated to ensure valid predictions (Aliwell et al., 2001).

It is still impossible to fabricate chemical sensors without drift. In fact, drift phenomena afflict almost all kinds of sensors (Chen & Chan, 2008; Owens & Wong, 2009; Polster et al., 2009).

chromatography, mass spectroscopy and ion mobility spectrometry. This trend has also narrowed the gap between traditional ENs – used as a black box – and classical analytical

307

Although the constant improvements in micro fabrication techniques and the rapid development of new nano fabrication techniques have allowed the production of functional micro and nanoscale chemical sensing devices with finer sensitivity (Comini & Sberveglieri, 2010) and selectivity (Haupt & Mosbach, 2000), signal repeatability over time still remains the real challenge in the chemical sensor field for all types of sensors. In order to have a reliable instrument it is of great importance that individual sensor signals are stable and reproducible. In practice, it is not worthy spending several weeks for training an EN system in a particular application if, as a consequence of changes in the sensor response, the system can only be used

The problem of chemical sensor stability over time is known as "sensor drift". It consists of (more or less) small and non-deterministic temporal variations of the sensor response when it is exposed to the same analytes under identical conditions (Holmberg et al., 1997). This is generally attributed to sensors aging (Sharma et al., 2001) or thermo-mechanical degradation (Mielle, 1996), but it can also be influenced by a variety of sources including environmental factors (Di Natale et al., 2002b; Ionescu et al., 2000). The main result is that sensors selectivity and sensitivity slowly decrease with time. The physical causes of this phenomenon are technology dependent and are strictly correlated to the sensing and transduction mechanisms.

Among the different technologies used to fabricate chemical sensors, which in general lack enough information about the physical causes of the sensor drift, the physical meaning of the drift phenomenon in conductometric metal oxide sensors (MOX) has been deeply investigated in the past. In fact, signal drift is known to be a severe problem for these devices which have widespread commercial diffusion. A typical example is the sintered tin dioxide Taguchi Gas Sensor (TGS), a n-type semiconductor solid state device marketed by Figaro Engineering Inc.<sup>2</sup>

The sensing properties of TGS are based on the electronic (n-type) conductivity of tin dioxide (*SnO*2) (Goepel & Schierbaum, 1995). The device consists of small *SnO*<sup>2</sup> grains which are in contact with each other. The sensing effect is due to an electronic depletion layer at the surface of the grains. The depletion layer is generated when oxygen is adsorbed, thus trapping electrons from the oxide. This induces an increased resistance at the grain surfaces. When the current passes from one grain to another it has to cross these depletion layers (Schottky barriers) which thus determine the sensor's resistance. The sensing effect, i.e., the response to reducing or oxidizing gases, is therefore determined by the adsorption of these compounds and the subsequent trapping (or donation) of electrons by the adsorbed species. This modifies the space charge potential thus changing the sensor's conductivity. Similar working models also apply to thick- and thin-film semiconductor MOX gas sensors, including for instance *SnO*<sup>2</sup> <sup>−</sup> *RGTO*<sup>3</sup> gas sensors (Sberveglieri, 1992). The main reasons for the MOX sensor

<sup>3</sup> Rheotaxial Growth and Thermal Oxidation (RGTO) technique was developed in the early 90s at the Sensor Lab on purpose of growing metal oxide thin films. The films obtained with such technique show a structure characterized by polycrystalline agglomerates uniformly distributed and connected by necks. This highly porous structure leads to a large surface area well suited for gas absorption

techniques which aim to quantify individual volatile components.

**2.1 Drift phenomenon in metal oxide semiconductor gas sensors**

since 1968 and widely applied for detection of oxidizing and reducing gases.

for a few days before recalibration is required.

Drift Correction Methods for Gas Chemical Sensors in Artificial Olfaction Systems: Techniques and Challenges

<sup>2</sup> http://www.figaro.co.jp

Sensor drift must be therefore detected and compensated to achieve reliable measurements from a sensor array.

Algorithms to mitigate the negative effect of gas sensor drift are not new in the field; the first attempt to tackle this problem dates back to the early 90s (Pearce et al., 2003). Nevertheless, the study of sensor drift is still a challenging task for the chemical sensor community (Padilla et al., 2010; Pearce et al., 2003). Solutions proposed in the literature can be grouped into four main categories: (i) periodic calibration, (ii) attuning methods, (iii) filtering of drift components and (iv) adaptive models. In this chapter the authors introduce the main challenges faced when developing drift correction techniques and will propose a deep overview of state-of-the-art methodologies that have been proposed in the scientific literature trying to underlying pros and cons of these techniques and focusing on challenges still open and waiting for solutions.

## **2. Gas chemical sensors and the drift phenomenon**

A chemical sensor is a device that transforms chemical information, ranging from the concentration of a specific component to the total composition analysis, into an analytically useful signal (International Union of Pure and Applied Chemistry, 1991). In the gas chemical sensing, the input signal is the concentration of one or more gaseous species while the output signal depends on the transduction mechanism that is usually a variation of some physical properties of the sensing element such as: conductivity (or other electrical magnitudes), oscillation frequency (mass), temperature, electrochemical potentials, surface work function or optical properties (Janata, 2009).

Artificial olfaction systems, usually referred to as Electronic Noses (ENs) (Pearce et al., 2003), are machines designed for detecting and discriminating among complex odors using an array of broadly-tuned (non-specific) gas chemical sensors typically belonging to the above mentioned categories. An odor stimulus generates a characteristic fingerprint from the sensor array. Patterns from known samples can be used to construct a database (training set) and train a pattern recognition system so that unknown odor samples (test set) can be subsequently identified.

Attempts to measure odors with electronic instruments were made in the early 60s, but the "modern era" of artificial olfaction began in 1982 with the work of Persaud and Dodd (Persaud & Dodd, 1982), who used a small array of gas-sensitive metal-oxide devices to classify odors. The expression "electronic nose" appeared for the first time in 1987 (Shurmer et al., 1987), with its current definition given in the same year by Gardner (Gardner, 1987). Commercial instruments became available in the early 90s – first in Europe immediately followed by the U.S. – with pioneering machines developed by Alpha Mos1 and Aromascan. ENs take their inspiration from the working mechanism of biological olfaction. However, current technologies based on chemical sensors are still far away from the capability of biological systems mainly because of their (still) poor selectivity and sensitivity with respect to biological receptors, and more importantly their lack of stability.

In recent years, classical chemical sensor technologies were complemented by new emerging technologies (Röck et al., 2008). In particular, machine olfaction has benefited from developments in several fields ranging from optical technologies developed by the telecommunications industry to the improvements in analytical chemistry such as: gas

<sup>1</sup> http://www.alpha-mos.com/

2 Chemical Sensors

Sensor drift must be therefore detected and compensated to achieve reliable measurements

Algorithms to mitigate the negative effect of gas sensor drift are not new in the field; the first attempt to tackle this problem dates back to the early 90s (Pearce et al., 2003). Nevertheless, the study of sensor drift is still a challenging task for the chemical sensor community (Padilla et al., 2010; Pearce et al., 2003). Solutions proposed in the literature can be grouped into four main categories: (i) periodic calibration, (ii) attuning methods, (iii) filtering of drift components and (iv) adaptive models. In this chapter the authors introduce the main challenges faced when developing drift correction techniques and will propose a deep overview of state-of-the-art methodologies that have been proposed in the scientific literature trying to underlying pros and cons of these techniques and focusing on challenges still open

A chemical sensor is a device that transforms chemical information, ranging from the concentration of a specific component to the total composition analysis, into an analytically useful signal (International Union of Pure and Applied Chemistry, 1991). In the gas chemical sensing, the input signal is the concentration of one or more gaseous species while the output signal depends on the transduction mechanism that is usually a variation of some physical properties of the sensing element such as: conductivity (or other electrical magnitudes), oscillation frequency (mass), temperature, electrochemical potentials, surface work function

Artificial olfaction systems, usually referred to as Electronic Noses (ENs) (Pearce et al., 2003), are machines designed for detecting and discriminating among complex odors using an array of broadly-tuned (non-specific) gas chemical sensors typically belonging to the above mentioned categories. An odor stimulus generates a characteristic fingerprint from the sensor array. Patterns from known samples can be used to construct a database (training set) and train a pattern recognition system so that unknown odor samples (test set) can be

Attempts to measure odors with electronic instruments were made in the early 60s, but the "modern era" of artificial olfaction began in 1982 with the work of Persaud and Dodd (Persaud & Dodd, 1982), who used a small array of gas-sensitive metal-oxide devices to classify odors. The expression "electronic nose" appeared for the first time in 1987 (Shurmer et al., 1987), with its current definition given in the same year by Gardner (Gardner, 1987). Commercial instruments became available in the early 90s – first in Europe immediately followed by the U.S. – with pioneering machines developed by Alpha Mos1 and Aromascan. ENs take their inspiration from the working mechanism of biological olfaction. However, current technologies based on chemical sensors are still far away from the capability of biological systems mainly because of their (still) poor selectivity and sensitivity with respect to biological

In recent years, classical chemical sensor technologies were complemented by new emerging technologies (Röck et al., 2008). In particular, machine olfaction has benefited from developments in several fields ranging from optical technologies developed by the telecommunications industry to the improvements in analytical chemistry such as: gas

from a sensor array.

and waiting for solutions.

or optical properties (Janata, 2009).

subsequently identified.

<sup>1</sup> http://www.alpha-mos.com/

**2. Gas chemical sensors and the drift phenomenon**

receptors, and more importantly their lack of stability.

chromatography, mass spectroscopy and ion mobility spectrometry. This trend has also narrowed the gap between traditional ENs – used as a black box – and classical analytical techniques which aim to quantify individual volatile components.

Although the constant improvements in micro fabrication techniques and the rapid development of new nano fabrication techniques have allowed the production of functional micro and nanoscale chemical sensing devices with finer sensitivity (Comini & Sberveglieri, 2010) and selectivity (Haupt & Mosbach, 2000), signal repeatability over time still remains the real challenge in the chemical sensor field for all types of sensors. In order to have a reliable instrument it is of great importance that individual sensor signals are stable and reproducible. In practice, it is not worthy spending several weeks for training an EN system in a particular application if, as a consequence of changes in the sensor response, the system can only be used for a few days before recalibration is required.

The problem of chemical sensor stability over time is known as "sensor drift". It consists of (more or less) small and non-deterministic temporal variations of the sensor response when it is exposed to the same analytes under identical conditions (Holmberg et al., 1997). This is generally attributed to sensors aging (Sharma et al., 2001) or thermo-mechanical degradation (Mielle, 1996), but it can also be influenced by a variety of sources including environmental factors (Di Natale et al., 2002b; Ionescu et al., 2000). The main result is that sensors selectivity and sensitivity slowly decrease with time. The physical causes of this phenomenon are technology dependent and are strictly correlated to the sensing and transduction mechanisms.

#### **2.1 Drift phenomenon in metal oxide semiconductor gas sensors**

Among the different technologies used to fabricate chemical sensors, which in general lack enough information about the physical causes of the sensor drift, the physical meaning of the drift phenomenon in conductometric metal oxide sensors (MOX) has been deeply investigated in the past. In fact, signal drift is known to be a severe problem for these devices which have widespread commercial diffusion. A typical example is the sintered tin dioxide Taguchi Gas Sensor (TGS), a n-type semiconductor solid state device marketed by Figaro Engineering Inc.<sup>2</sup> since 1968 and widely applied for detection of oxidizing and reducing gases.

The sensing properties of TGS are based on the electronic (n-type) conductivity of tin dioxide (*SnO*2) (Goepel & Schierbaum, 1995). The device consists of small *SnO*<sup>2</sup> grains which are in contact with each other. The sensing effect is due to an electronic depletion layer at the surface of the grains. The depletion layer is generated when oxygen is adsorbed, thus trapping electrons from the oxide. This induces an increased resistance at the grain surfaces. When the current passes from one grain to another it has to cross these depletion layers (Schottky barriers) which thus determine the sensor's resistance. The sensing effect, i.e., the response to reducing or oxidizing gases, is therefore determined by the adsorption of these compounds and the subsequent trapping (or donation) of electrons by the adsorbed species. This modifies the space charge potential thus changing the sensor's conductivity. Similar working models also apply to thick- and thin-film semiconductor MOX gas sensors, including for instance *SnO*<sup>2</sup> <sup>−</sup> *RGTO*<sup>3</sup> gas sensors (Sberveglieri, 1992). The main reasons for the MOX sensor

<sup>2</sup> http://www.figaro.co.jp

<sup>3</sup> Rheotaxial Growth and Thermal Oxidation (RGTO) technique was developed in the early 90s at the Sensor Lab on purpose of growing metal oxide thin films. The films obtained with such technique show a structure characterized by polycrystalline agglomerates uniformly distributed and connected by necks. This highly porous structure leads to a large surface area well suited for gas absorption

Fig. 1. *Rair* and *Rgas* for a *SnO*<sup>2</sup> (*Au* doped) thin film vs. the aging time; the sensor is operated at 400degC. (a): 500 ppm *CO* in 30% *RH*; (b): 5000 ppm *CH*<sup>4</sup> in 30% *RH*

accelerates the degradation rate of the above deposited *Pt* layer;

correlation of these results show that the long term stability is mostly determined by the instability of platinum inter-digitized contacts (IDCs). Two concurrent effects were observed

309

• first, the erosion of the IDC, which appears thinner after aging, and the formation of *Pt* agglomerates. *Pt* agglomeration was observed only on the IDCs deposited over the *SnO*<sup>2</sup> − *RGTO* layer, while the *Pt* structures deposited over the alumina substrate did not exhibit agglomeration. From this, we might argue that the larger RGTO roughness

(reproduced from (Nelli et al., 2000))

Drift Correction Methods for Gas Chemical Sensors in Artificial Olfaction Systems: Techniques and Challenges

(Fig. 2):

drift has been attributed to different structural and morphological variations of the sensor as discussed in the following subsections.

#### **2.1.1 Chemical diffusion of oxygen vacancies**

Usually two chemical effects are considered as potential sources of drift in MOX sensors: a) the resistivity change induced by chemisorption of water on the sensor (Schierbaum et al., 1991), and b) the chemical diffusion of oxygen vacancies that was investigated by means of relaxation experiments on tin dioxide single crystals (Kamp et al., 2001).

Kamp et al. have proven that for TGS sensors the chemical diffusion of oxygen is fast enough to cause severe drift effects due to stoichiometry changes. This effect was in turn attributed to two main origins:


#### **2.1.2 Physical changes of the MOX sensing elements**

Scanning electron microscopy investigation of thin-film *SnO*<sup>2</sup> sensors integrated on CMOS micro-machined hotplates (Sharma et al., 2001) indicated that the sensing thin-film cracks after long operation cycles. Indeed, the thermal stress in the micro-hotplate induces cracks during a large number of heating and cooling cycles. Cracks in the thin-film can be considered a major physical cause of the sensor drift. Authors also suggest that *Cu* doping could be a manner to suppress (or strongly reduce) the film cracking.

Similar results are mentioned in other works, e.g., investigating the variation of electrical response to *CO* and *CH*<sup>4</sup> induced by the continuous thin-film *SnO*<sup>2</sup> sensor operation at 400degC for six months. A slowly varying decreasing of response to both baseline air and gases was observed (Fig. 1) and attributed to the coalescence of grains due to their intrinsic poor degree of crystallinity. In fact a significant improvement on the stability of the sensor platforms can be achieved using single crystalline materials such as novel quasi-one-dimensional MOX nanostructures (Comini & Sberveglieri, 2010).

#### **2.1.3 Degradation of the electric contacts**

Besides the preparation of stable grain structures, the proper choice of electric contact geometries (shape and thickness) is particularly important in the design of reliable conductometric MOX sensors.

Recently, thin-film *SnO*<sup>2</sup> − *RGTO* sensors have been investigated at SENSOR lab by scanning electron microscopy including the study of morphology by secondary electron detection and back scattered electrons which allow for compositional information (unpublished results). Electrical tests, performed in parallel, predicted a drift time constant of about 85 days . The 4 Chemical Sensors

drift has been attributed to different structural and morphological variations of the sensor

Usually two chemical effects are considered as potential sources of drift in MOX sensors: a) the resistivity change induced by chemisorption of water on the sensor (Schierbaum et al., 1991), and b) the chemical diffusion of oxygen vacancies that was investigated by means of

Kamp et al. have proven that for TGS sensors the chemical diffusion of oxygen is fast enough to cause severe drift effects due to stoichiometry changes. This effect was in turn attributed to

• the chemical diffusion of oxygen in *SnO*<sup>2</sup> consists of the simultaneous transport of oxygen vacancies and conduction electrons that induces conductivity changes in the bulk and/or space charge layer of *SnO*<sup>2</sup> grains. Changes in oxygen vacancy concentration over the sensor working time influences the space charge and the overall conductivity which

• the diffusion of oxygen vacancies can be induced by the space charge electric field alone (field induced migration). For the case of the TGS this process is induced by a signal change, via changes of the space charge potential. Since oxygen vacancies are usually majority carriers, the result of the redistribution of oxygen vacancies is a severe

Scanning electron microscopy investigation of thin-film *SnO*<sup>2</sup> sensors integrated on CMOS micro-machined hotplates (Sharma et al., 2001) indicated that the sensing thin-film cracks after long operation cycles. Indeed, the thermal stress in the micro-hotplate induces cracks during a large number of heating and cooling cycles. Cracks in the thin-film can be considered a major physical cause of the sensor drift. Authors also suggest that *Cu* doping could be a manner to

Similar results are mentioned in other works, e.g., investigating the variation of electrical response to *CO* and *CH*<sup>4</sup> induced by the continuous thin-film *SnO*<sup>2</sup> sensor operation at 400degC for six months. A slowly varying decreasing of response to both baseline air and gases was observed (Fig. 1) and attributed to the coalescence of grains due to their intrinsic poor degree of crystallinity. In fact a significant improvement on the stability of the sensor platforms can be achieved using single crystalline materials such as novel

Besides the preparation of stable grain structures, the proper choice of electric contact geometries (shape and thickness) is particularly important in the design of reliable

Recently, thin-film *SnO*<sup>2</sup> − *RGTO* sensors have been investigated at SENSOR lab by scanning electron microscopy including the study of morphology by secondary electron detection and back scattered electrons which allow for compositional information (unpublished results). Electrical tests, performed in parallel, predicted a drift time constant of about 85 days . The

relaxation experiments on tin dioxide single crystals (Kamp et al., 2001).

modification of the properties of the space charge region itself.

quasi-one-dimensional MOX nanostructures (Comini & Sberveglieri, 2010).

**2.1.2 Physical changes of the MOX sensing elements**

suppress (or strongly reduce) the film cracking.

**2.1.3 Degradation of the electric contacts**

conductometric MOX sensors.

as discussed in the following subsections.

reflects on the sensor baseline drift;

two main origins:

**2.1.1 Chemical diffusion of oxygen vacancies**

Fig. 1. *Rair* and *Rgas* for a *SnO*<sup>2</sup> (*Au* doped) thin film vs. the aging time; the sensor is operated at 400degC. (a): 500 ppm *CO* in 30% *RH*; (b): 5000 ppm *CH*<sup>4</sup> in 30% *RH* (reproduced from (Nelli et al., 2000))

correlation of these results show that the long term stability is mostly determined by the instability of platinum inter-digitized contacts (IDCs). Two concurrent effects were observed (Fig. 2):

• first, the erosion of the IDC, which appears thinner after aging, and the formation of *Pt* agglomerates. *Pt* agglomeration was observed only on the IDCs deposited over the *SnO*<sup>2</sup> − *RGTO* layer, while the *Pt* structures deposited over the alumina substrate did not exhibit agglomeration. From this, we might argue that the larger RGTO roughness accelerates the degradation rate of the above deposited *Pt* layer;

binding substances in the samples, such as sulphur compounds (Pratt & Williams, 1997) or some acids Schaller et al. (2000) led to irreversible poisoning. When this occurs, the sensor should be replaced and there is no possibility to compensate the effect. On the contrary, slowly varying drift phenomena, which are frequently reported in the literature, can be coped with

Attempts to mitigate the negative effect of gas sensor drift are not new. A great deal of work has been directed towards the development of drift correction methods and algorithms, which tackle the problem from different perspectives, depending on the situation. Nevertheless, the study of sensor drift still remains a challenging task for the chemical sensor community always looking for novel improved solutions. Fig. 3 provides a rough classification into four main categories of the solutions proposed in the literature, which are presented hereafter.

> Drift correction methods

> > Attuning methods

Independent Component Corr.

Componen

Indepen

Orthogonal Signal Correction

Orthogo Cor

Adaptive methods

311

Neural networks (SOM, ART, …)

ne (SOM

A

Evolutionary Algorithms

Evolutio Algorit

proper "soft" methods that will be illustrated in the following sections.

Periodic calibration

PCA/PLS - Component Correction

Comp Corre PCA/

Comp Defla

(the so called "baseline") thus the name of *baseline manipulation*.

drift effects which are both present in the baseline and gas response:

Three basic transformations are common practice:

Component Deflation

Fig. 3. Taxonomy of drift correction methodologies published in the scientific literature

One of the simplest methods for drift compensation that has been proposed in the literature, which is also widely used as a pre-processing method, is the transformation of individual sensor signals (Gardner & Bartlett, 1999) based on the initial value of the transient response

1. *Differential*: subtracts the baseline of each sensor and then can help compensating additive

*s*ˆ(*t*) = *y* (*t*) − *y* (0) = [*x* (*t*) + *δ*] − [*x* (0) + *δ*] = *x* (*t*) − *x* (0) (1)

Multiplicative

Co Multipl

manipul Drift Correction

Sensor signal preprocessing

Baseline manipulation

Baseli

Frequency domain filtering

**3.1 Sensor signal preprocessing 3.1.1 Baseline manipulation**

Freque domain fi

**3. Drift counteraction methods: a taxonomy and review**

Drift Correction Methods for Gas Chemical Sensors in Artificial Olfaction Systems: Techniques and Challenges

• second, the formation of *Pt* agglomerates could be promoted by electro-migration phenomena. This second hypothesis has been confirmed observing that the formation of *Pt* agglomerates is more evident in the area in which the section of the electrodes is thinner, i.e., where the current density is bigger. Increasing the IDC thickness from 500nm to 900nm the drift time constant was increased to about 130 days.

Fig. 2. Comparison between as prepared (left) and 120 days aged sensor (right) by SEM- EDX analysis. Aged sensor shows a larger transparency of the IDC (compare upper left and right images) and the formation of Pt agglomerates (compare bottom left with bottom right). Legend: SE=Secondary Electrons; BSE=Back Scattered Electrons (Courtesy of Dr. A. Ponzoni and Dr. M.Ferroni, University of Brescia and CNR-IDASC)

### **2.2 Overall considerations**

In summary, even if we considered a well defined class of deeply investigated sensors such as MOX sensors, the drift phenomenology is still not totally understood since real polycrystalline samples show a variety of aspects and a high degree of complexity. Depending on the specific technology (e.g., TGS or thin-film) one aspect can be dominant over the others, but often concomitant causes are present.

Irreversible changes of the sensor response might also occur, one of the most common reasons being the sensor surface poisoning (Ruhland et al., 1998). This arises when the sensor is exposed to a gas (e.g., a corrosive acid) which strongly binds or interacts with the sensing material leading to a deep change in its physicochemical properties. Indeed, some authors have reported the occurrence of sensor faults during long measurement runs in monitoring odors on a landfill site (Romain & Nicolas, 2010). In other cases, the presence of strongly 6 Chemical Sensors

• second, the formation of *Pt* agglomerates could be promoted by electro-migration phenomena. This second hypothesis has been confirmed observing that the formation of *Pt* agglomerates is more evident in the area in which the section of the electrodes is thinner, i.e., where the current density is bigger. Increasing the IDC thickness from 500nm

Fig. 2. Comparison between as prepared (left) and 120 days aged sensor (right) by SEM- EDX analysis. Aged sensor shows a larger transparency of the IDC (compare upper left and right images) and the formation of Pt agglomerates (compare bottom left with bottom right). Legend: SE=Secondary Electrons; BSE=Back Scattered Electrons (Courtesy of Dr. A. Ponzoni

In summary, even if we considered a well defined class of deeply investigated sensors such as MOX sensors, the drift phenomenology is still not totally understood since real polycrystalline samples show a variety of aspects and a high degree of complexity. Depending on the specific technology (e.g., TGS or thin-film) one aspect can be dominant over the others, but often

Irreversible changes of the sensor response might also occur, one of the most common reasons being the sensor surface poisoning (Ruhland et al., 1998). This arises when the sensor is exposed to a gas (e.g., a corrosive acid) which strongly binds or interacts with the sensing material leading to a deep change in its physicochemical properties. Indeed, some authors have reported the occurrence of sensor faults during long measurement runs in monitoring odors on a landfill site (Romain & Nicolas, 2010). In other cases, the presence of strongly

and Dr. M.Ferroni, University of Brescia and CNR-IDASC)

**2.2 Overall considerations**

concomitant causes are present.

to 900nm the drift time constant was increased to about 130 days.

binding substances in the samples, such as sulphur compounds (Pratt & Williams, 1997) or some acids Schaller et al. (2000) led to irreversible poisoning. When this occurs, the sensor should be replaced and there is no possibility to compensate the effect. On the contrary, slowly varying drift phenomena, which are frequently reported in the literature, can be coped with proper "soft" methods that will be illustrated in the following sections.

## **3. Drift counteraction methods: a taxonomy and review**

Attempts to mitigate the negative effect of gas sensor drift are not new. A great deal of work has been directed towards the development of drift correction methods and algorithms, which tackle the problem from different perspectives, depending on the situation. Nevertheless, the study of sensor drift still remains a challenging task for the chemical sensor community always looking for novel improved solutions. Fig. 3 provides a rough classification into four main categories of the solutions proposed in the literature, which are presented hereafter.

Fig. 3. Taxonomy of drift correction methodologies published in the scientific literature

#### **3.1 Sensor signal preprocessing**

#### **3.1.1 Baseline manipulation**

One of the simplest methods for drift compensation that has been proposed in the literature, which is also widely used as a pre-processing method, is the transformation of individual sensor signals (Gardner & Bartlett, 1999) based on the initial value of the transient response (the so called "baseline") thus the name of *baseline manipulation*. Three basic transformations are common practice:

1. *Differential*: subtracts the baseline of each sensor and then can help compensating additive drift effects which are both present in the baseline and gas response:

$$\mathbf{x}'(t) = \mathbf{y}'(t) - \mathbf{y}'(0) = [\mathbf{x}(t) + \delta] - [\mathbf{x}(0) + \delta] = \mathbf{x}(t) - \mathbf{x}'(0) \tag{1}$$

convolution with the signal gives the high-frequency components. These components are

313

The multilevel wavelet decomposition and signal reconstruction is performed by the Mallat algorithm. It consists of iteratively applying high-pass and low-pass filters on the vector of approximation coefficients, obtaining a sequence of increasingly smoothed and halved versions of the original signal. Once a specific wavelet has been chosen (e.g. Daubechies

The DWT decomposition level is fixed once analysis in the frequency domain has been carried out to single out the frequency domain (cut-off frequency) of the drifting trend. The approximation coefficients associated with the lowest frequencies which have drift contamination are then discarded and the wavelet reconstruction of the corrected signal is

Different drift correction methods are based on the estimation of the drift effect on the system to be later removed. Drift effect estimation can be made, for instance, by measuring the change in the sensor responses to one (or more) reference gas, which is measured with some intervals along the experiment. This strategy can be applied in a univariate way (sensor-by-sensor) or in a multivariate way by removing the directions of dispersion of the reference data in the

Univariate calibration is a straightforward method in which a reference value, i.e., the response to a reference gas, is used and all subsequent sensor readings are individually corrected to it. Fryder et al. (Fryder et al., 1995) and Haugen et al. (Haugen et al., 2000) proposed to model temporal variations of the system with a multiplicative drift correction (MDC) factor obtained by measuring the calibrant and then to apply the same correction to

In particular, Haugen et al. propose a re-calibration method performed in two steps: within a single measurement sequence to compensate for short terms trends and between measurement sequences to compensate for long term fluctuations. This strategy provides very good results and it is currently in use in commercial electronic noses. However, assumptions are too tight: MDC makes a supposition that the drift is multiplicative, which means that the perturbation is proportional to the signal level. Furthermore, it assumes that the relationship between the individual sensor response to the reference gas and the response to the test gas

One of the first attempts of performing robust drift correction by multivariate methods was proposed by Artursson et al. (Artursson et al., 2000) under the name of Component Correction (CC). Two correction methods, one based on Principal Component Analysis (PCA) and one

Measurements performed with arrays of chemical sensors contain several redundant information since, in general, the different variables are collinear. PCA (Wold et al., 1987) is a common compression method used to efficiently represent this information. With PCA, the dominating variability in the measurement space can be captured by means of two matrices:

orthonormal functions), the pair of low-pass and high-pass filters are defined.

called approximation coefficients.

**3.2 Periodic calibration**

**3.2.1 Multiplicative drift correction**

feature space.

the actual samples.

has to be strictly linear.

**3.2.2 Multivariate component correction**

based on Partial Least Square (PLS) are proposed in the paper.

computed by using the remaining coefficients.

Drift Correction Methods for Gas Chemical Sensors in Artificial Olfaction Systems: Techniques and Challenges

where *s*ˆ is the transformed (corrected) response, *y* is the measured response, *x* is the ideal sensor response without drift and *δ* is the drift contribution which is assumed to be constant and uniform.

2. *Relative*: divides by the baseline and might correct for (constant and uniform) multiplicative drift effects:

$$\mathfrak{F}\left(t\right) = \frac{\mathfrak{Y}\left(t\right)}{\mathfrak{Y}\left(0\right)} = \frac{\mathfrak{x}\left(t\right) + \delta\mathfrak{x}\left(t\right)}{\mathfrak{x}\left(0\right) + \delta\mathfrak{x}\left(0\right)} = \frac{\mathfrak{x}\left(t\right)\left(1+\delta\right)}{\mathfrak{x}\left(0\right)\left(1+\delta\right)} = \frac{\mathfrak{x}\left(t\right)}{\mathfrak{x}\left(0\right)}\tag{2}$$

3. *Fractional*: a combination of the previous two that works for multiplicative drift and has the advantage of providing dimensionless measurements and normalized sensor responses.

$$\text{s}^{\circ}(t) = \frac{\text{y}(t) - \text{y}(0)}{\text{y}(0)} = \frac{\text{x}(t) \left(1 + \delta\right) - \text{x}(0) \left(1 + \delta\right)}{\text{x}(0) \left(1 + \delta\right)} = \frac{\text{x}(t) - \text{x}(0)}{\text{x}(0)}\tag{3}$$

The first two transformations are too specific because in real applications the drift is generally not additive neither multiplicative, thus they are not able to correct the drift effect while they are (usually the second one) used to simply "normalize" the sensor response. The last manipulation obviously does not work if additive drift is present. Conversely it can amplify the noise in the measurements because the drift term, which is typically small, remains at the denominator thus degrading the quality of the sample. It therefore provides again poor correction against drift effects. More advanced preprocessing methods have been tried out based on sensor signal processing in the frequency domain more than in the time domain.

#### **3.1.2 Filtering**

Counteracting methods based on filtering strategies focus on the application of signal preprocessing techniques to filter out portions of the signal containing drift contaminations. Drift typically occurs in a different frequency domain with respect to interesting signals, being in general a slower process. Therefore, proper transformations of sensor signals from time to frequency domain and a careful removal of the lowest frequency components can filter the drift out.

Feature extraction techniques based on Discrete Wavelet Transform (DWT) can be a powerful tool to remove the drift contamination in the low-frequency behavior of the sensor responses (Hui et al., 2003; Llobet et al., 2002; Zuppa et al., 2003). A moving median filter and Fourier band-pass filters are some examples applied to removing either high-frequency fluctuations (such as noise, spikes) or low-frequency changes such as drift. In comparison to these filters, DWT technique provides a flexible analysis of the signal at different resolutions by applying iteratively high-pass and low pass filters. DWT technique allows to remove the selected low-frequency components easily and in such a way that the signal is not distorted.

The DWT provides a multiresolution signal decomposition of sensor response: it analyzes the signal at different frequency bands with different resolutions by successively projecting it down onto two basis of functions, which are obtained by applying shift and scaling operations to two prototype functions called the "scaling function" and the "wavelet function", respectively. The scaling function is associated with low-pass filter and the convolution between the signal and the scaling functions gives the low-frequency components of the signal. Conversely, the wavelet function is associated with high-pass filter and its convolution with the signal gives the high-frequency components. These components are called approximation coefficients.

The multilevel wavelet decomposition and signal reconstruction is performed by the Mallat algorithm. It consists of iteratively applying high-pass and low-pass filters on the vector of approximation coefficients, obtaining a sequence of increasingly smoothed and halved versions of the original signal. Once a specific wavelet has been chosen (e.g. Daubechies orthonormal functions), the pair of low-pass and high-pass filters are defined.

The DWT decomposition level is fixed once analysis in the frequency domain has been carried out to single out the frequency domain (cut-off frequency) of the drifting trend. The approximation coefficients associated with the lowest frequencies which have drift contamination are then discarded and the wavelet reconstruction of the corrected signal is computed by using the remaining coefficients.

#### **3.2 Periodic calibration**

8 Chemical Sensors

2. *Relative*: divides by the baseline and might correct for (constant and uniform)

3. *Fractional*: a combination of the previous two that works for multiplicative drift and has the advantage of providing dimensionless measurements and normalized sensor responses.

*<sup>y</sup>* (0) <sup>=</sup> *<sup>x</sup>* (*t*) (<sup>1</sup> <sup>+</sup> *<sup>δ</sup>*) <sup>−</sup> *<sup>x</sup>* (0) (<sup>1</sup> <sup>+</sup> *<sup>δ</sup>*)

The first two transformations are too specific because in real applications the drift is generally not additive neither multiplicative, thus they are not able to correct the drift effect while they are (usually the second one) used to simply "normalize" the sensor response. The last manipulation obviously does not work if additive drift is present. Conversely it can amplify the noise in the measurements because the drift term, which is typically small, remains at the denominator thus degrading the quality of the sample. It therefore provides again poor correction against drift effects. More advanced preprocessing methods have been tried out based on sensor signal processing in the frequency domain more than in the time domain.

Counteracting methods based on filtering strategies focus on the application of signal preprocessing techniques to filter out portions of the signal containing drift contaminations. Drift typically occurs in a different frequency domain with respect to interesting signals, being in general a slower process. Therefore, proper transformations of sensor signals from time to frequency domain and a careful removal of the lowest frequency components can filter the

Feature extraction techniques based on Discrete Wavelet Transform (DWT) can be a powerful tool to remove the drift contamination in the low-frequency behavior of the sensor responses (Hui et al., 2003; Llobet et al., 2002; Zuppa et al., 2003). A moving median filter and Fourier band-pass filters are some examples applied to removing either high-frequency fluctuations (such as noise, spikes) or low-frequency changes such as drift. In comparison to these filters, DWT technique provides a flexible analysis of the signal at different resolutions by applying iteratively high-pass and low pass filters. DWT technique allows to remove the selected

The DWT provides a multiresolution signal decomposition of sensor response: it analyzes the signal at different frequency bands with different resolutions by successively projecting it down onto two basis of functions, which are obtained by applying shift and scaling operations to two prototype functions called the "scaling function" and the "wavelet function", respectively. The scaling function is associated with low-pass filter and the convolution between the signal and the scaling functions gives the low-frequency components of the signal. Conversely, the wavelet function is associated with high-pass filter and its

low-frequency components easily and in such a way that the signal is not distorted.

*<sup>x</sup>* (0) <sup>+</sup> *<sup>δ</sup><sup>x</sup>* (0) <sup>=</sup> *<sup>x</sup>* (*t*) (<sup>1</sup> <sup>+</sup> *<sup>δ</sup>*)

*<sup>x</sup>* (0) (<sup>1</sup> <sup>+</sup> *<sup>δ</sup>*) <sup>=</sup> *<sup>x</sup>* (*t*)

*<sup>x</sup>* (0) (<sup>1</sup> <sup>+</sup> *<sup>δ</sup>*) <sup>=</sup> *<sup>x</sup>* (*t*) <sup>−</sup> *<sup>x</sup>* (0)

*<sup>x</sup>* (0) (2)

*<sup>x</sup>* (0) (3)

*<sup>y</sup>* (0) <sup>=</sup> *<sup>x</sup>* (*t*) <sup>+</sup> *<sup>δ</sup><sup>x</sup>* (*t*)

be constant and uniform.

multiplicative drift effects:

**3.1.2 Filtering**

drift out.

*<sup>s</sup>*ˆ(*t*) <sup>=</sup> *<sup>y</sup>* (*t*)

*<sup>s</sup>*ˆ(*t*) <sup>=</sup> *<sup>y</sup>* (*t*) <sup>−</sup> *<sup>y</sup>* (0)

where *s*ˆ is the transformed (corrected) response, *y* is the measured response, *x* is the ideal sensor response without drift and *δ* is the drift contribution which is assumed to

> Different drift correction methods are based on the estimation of the drift effect on the system to be later removed. Drift effect estimation can be made, for instance, by measuring the change in the sensor responses to one (or more) reference gas, which is measured with some intervals along the experiment. This strategy can be applied in a univariate way (sensor-by-sensor) or in a multivariate way by removing the directions of dispersion of the reference data in the feature space.

#### **3.2.1 Multiplicative drift correction**

Univariate calibration is a straightforward method in which a reference value, i.e., the response to a reference gas, is used and all subsequent sensor readings are individually corrected to it. Fryder et al. (Fryder et al., 1995) and Haugen et al. (Haugen et al., 2000) proposed to model temporal variations of the system with a multiplicative drift correction (MDC) factor obtained by measuring the calibrant and then to apply the same correction to the actual samples.

In particular, Haugen et al. propose a re-calibration method performed in two steps: within a single measurement sequence to compensate for short terms trends and between measurement sequences to compensate for long term fluctuations. This strategy provides very good results and it is currently in use in commercial electronic noses. However, assumptions are too tight: MDC makes a supposition that the drift is multiplicative, which means that the perturbation is proportional to the signal level. Furthermore, it assumes that the relationship between the individual sensor response to the reference gas and the response to the test gas has to be strictly linear.

#### **3.2.2 Multivariate component correction**

One of the first attempts of performing robust drift correction by multivariate methods was proposed by Artursson et al. (Artursson et al., 2000) under the name of Component Correction (CC). Two correction methods, one based on Principal Component Analysis (PCA) and one based on Partial Least Square (PLS) are proposed in the paper.

Measurements performed with arrays of chemical sensors contain several redundant information since, in general, the different variables are collinear. PCA (Wold et al., 1987) is a common compression method used to efficiently represent this information. With PCA, the dominating variability in the measurement space can be captured by means of two matrices:

**3.2.3 Multivariate component deflation**

Drift Correction Methods for Gas Chemical Sensors in Artificial Olfaction Systems: Techniques and Challenges

**Y** (by means of regression/deflation).

This can be formally expressed as:

and **X***pred*. This is formally expressed as:

**3.3.1 Independent component correction**

have been explored.

**3.3 Attuning methods**

Another attempt to perform multivariate drift correction has been proposed in Gutierrez-Osuna (2000). The overall idea is to introduce a set of variables defined by a vector **Y** whose variance can be attributed to drift or interferents. Examples of these variables can be the response to a wash/reference gas that is usually performed prior to each measurement, time stamps, temperature, pressure, humidity, etc. The basic idea is to measure the sensor/array response to an odor **X** and remove the variance in **X** that can be explained by the variables in

The approach basically applies Canonical Correlation Analysis (CCA) or Partial Least Squares (PLS) to find two linear projections **Y**� = **a** · **Y** and **X**� = **b** · **X** that are maximally correlated.

**Y**� and **X**� are in fact low-dimensional projections that summarize the linear dependencies between **Y** and **X**. At this point Ordinary Least Squares (OLS) (Dillon & Goldstein, 1984) can be used to find a regression model **X***pred* = **w** · **X**� able to minimize the difference between **X**

The OLS prediction vector (**X***pred*) in fact contains the variance of the odor vector **X** that can be explained by **X**� and indirectly by **Y** since **X**� and **Y**� are correlated. At this point it is enough

In summary, techniques based on recalibration of the system using a reference gas can give very good results, but care must be taken in the selection of such gas. It must be representative of all classes being measured, since they are supposed to drift in similar ways. It should also be stable along the time, available and easy to measure (Salit & Turk, 1998). Fitting all these constraints is in general complex and expensive, therefore methods overcoming this limitation

Attuning methods try to perform component correction without resorting to the use of calibration samples, but trying to deduce drift components directly from the training data. They can provide significant improvements in the classification rate over a fixed time period, and may also make possible to obtain real responses to be used in gas quantitative analysis.

Di Natale et al. approached the problem of sensor drift with an attuning method considering also disturbances derived from the measurement environment (Di Natale et al., 2002a). In correction methods such as CC (Artursson et al., 2000) that are based on PCA, the computed principal components are mutually uncorrelated. However, this condition is not enough to guarantee that relevant signals are completely separated by disturbances (non-correlation does not necessarily implies statistical independence) and often a principal component carries information on both signal and disturbances. To tackle with this limitation Di Natale et al.

**X** − **w** · **X**�

**w** = *argmin*

to deflate **X** and use the residual **X***corr* as a drift-corrected vector:

{**a**, **b**} = *argmax* [*corr* (**aY**, **bX**)] (8)

**X***corr* = **X** − **X***pred* (10)

(9)

315


Within the new space defined by **P**, dimensions (referred to as components) are ordered by decreasing variability in the input measurements. CC uses PCA in conjunction with the reference gas technique. If the sensor responses to a certain reference gas contain a significant amount of drift, the first component identified by the PCA analysis on these measurements, which is the one that describes the maximum variability, will likely define the direction of the drift. This is motivated by the fact that the sensors are always exposed to the same gas and thus they are expected to provide always the same response with the exception of some random noise. This direction is defined by a loading vector **p** obtained as the first column of the loading matrix **P**. Projecting (multiplying) the sample **X** on this loading vector gives a score vector **t** representing the drift:

$$\mathbf{t} = \mathbf{X} \cdot \mathbf{p} \tag{4}$$

Drift correction can be then implemented by subtracting from the original data the bilinear expression **<sup>t</sup>** · **<sup>p</sup>***<sup>T</sup>* which represent an approximation of the drift component in the original sample space (i.e., all other directions are preserved and the variance that distinguishes and separates classes of samples in the data space is preserved), thus obtaining a corrected sample :

$$\mathbf{X}\_{corr} = \mathbf{X} - \mathbf{t} \cdot \mathbf{p}^T \tag{5}$$

Removing one component is usually enough whenever we are facing drift effects caused by aging of the sensors. However, if non-linear drift effects are observed (e.g., caused by both aging effects and chemical background) more than one component can be subtracted.

A similar correction strategy can be also obtained by using a regression model. Since the drift caused by aging effects has a preferred direction in the measurement space, it should be possible to describe this change as a function of time. The Partial Least Square (PLS) regression model (Wold et al., 1984) is able to infer the dependence between two set of variables (the sample matrix **X** and a matrix **Y** representing the time in our specific case) by using a set of orthogonal score vectors. Artursson et al. propose to compute a weight vector **w** and a loading vector **p** according to the PLS model on a set of measurements of a reference gas (as in the case of PCA). These vectors are first used to compute the drift component **t** according to the PLS regression model:

$$\mathbf{t} = \mathbf{X} \cdot \mathbf{w} \left(\mathbf{p}^T \mathbf{w}\right)^{-1} \tag{6}$$

Then, similarly to (5) the corrected sample is obtained as the residual after the first component has been subtracted from the original data:

$$\mathbf{X}\_{corr} = \mathbf{X} - \mathbf{t} \cdot \mathbf{p}^T \tag{7}$$

Again several components can be removed by repeating equations (6) and (7).

#### **3.2.3 Multivariate component deflation**

10 Chemical Sensors

1. the loading matrix **P** that represents a new coordinate system onto which the measurement

2. the score matrix **T** that represents the coordinate of the sample in the space represented by

Within the new space defined by **P**, dimensions (referred to as components) are ordered by decreasing variability in the input measurements. CC uses PCA in conjunction with the reference gas technique. If the sensor responses to a certain reference gas contain a significant amount of drift, the first component identified by the PCA analysis on these measurements, which is the one that describes the maximum variability, will likely define the direction of the drift. This is motivated by the fact that the sensors are always exposed to the same gas and thus they are expected to provide always the same response with the exception of some random noise. This direction is defined by a loading vector **p** obtained as the first column of the loading matrix **P**. Projecting (multiplying) the sample **X** on this loading vector gives a

Drift correction can be then implemented by subtracting from the original data the bilinear expression **<sup>t</sup>** · **<sup>p</sup>***<sup>T</sup>* which represent an approximation of the drift component in the original sample space (i.e., all other directions are preserved and the variance that distinguishes and separates classes of samples in the data space is preserved), thus obtaining a corrected sample

Removing one component is usually enough whenever we are facing drift effects caused by aging of the sensors. However, if non-linear drift effects are observed (e.g., caused by both

A similar correction strategy can be also obtained by using a regression model. Since the drift caused by aging effects has a preferred direction in the measurement space, it should be possible to describe this change as a function of time. The Partial Least Square (PLS) regression model (Wold et al., 1984) is able to infer the dependence between two set of variables (the sample matrix **X** and a matrix **Y** representing the time in our specific case) by using a set of orthogonal score vectors. Artursson et al. propose to compute a weight vector **w** and a loading vector **p** according to the PLS model on a set of measurements of a reference gas (as in the case of PCA). These vectors are first used to compute the drift component **t** according

aging effects and chemical background) more than one component can be subtracted.

**t** = **X** · **w**

Again several components can be removed by repeating equations (6) and (7).

 **p***T***w** −<sup>1</sup>

Then, similarly to (5) the corrected sample is obtained as the residual after the first component

**t** = **X** · **p** (4)

**<sup>X</sup>***corr* <sup>=</sup> **<sup>X</sup>** <sup>−</sup> **<sup>t</sup>** · **<sup>p</sup>***<sup>T</sup>* (5)

**<sup>X</sup>***corr* <sup>=</sup> **<sup>X</sup>** <sup>−</sup> **<sup>t</sup>** · **<sup>p</sup>***<sup>T</sup>* (7)

(6)

vector **X** must be projected, and

score vector **t** representing the drift:

to the PLS regression model:

has been subtracted from the original data:

:

the columns of **P**.

Another attempt to perform multivariate drift correction has been proposed in Gutierrez-Osuna (2000). The overall idea is to introduce a set of variables defined by a vector **Y** whose variance can be attributed to drift or interferents. Examples of these variables can be the response to a wash/reference gas that is usually performed prior to each measurement, time stamps, temperature, pressure, humidity, etc. The basic idea is to measure the sensor/array response to an odor **X** and remove the variance in **X** that can be explained by the variables in **Y** (by means of regression/deflation).

The approach basically applies Canonical Correlation Analysis (CCA) or Partial Least Squares (PLS) to find two linear projections **Y**� = **a** · **Y** and **X**� = **b** · **X** that are maximally correlated. This can be formally expressed as:

$$\{\mathbf{a}, \mathbf{b}\} = \operatorname\*{argmax}\left[\operatorname{corr}\left(\mathbf{a}\mathbf{Y}, \mathbf{b}\mathbf{X}\right)\right] \tag{8}$$

**Y**� and **X**� are in fact low-dimensional projections that summarize the linear dependencies between **Y** and **X**. At this point Ordinary Least Squares (OLS) (Dillon & Goldstein, 1984) can be used to find a regression model **X***pred* = **w** · **X**� able to minimize the difference between **X** and **X***pred*. This is formally expressed as:

$$\mathbf{w} = \operatorname\*{argmin}{\left[\mathbf{X} - \mathbf{w} \cdot \mathbf{X}^{\prime}\right]} \tag{9}$$

The OLS prediction vector (**X***pred*) in fact contains the variance of the odor vector **X** that can be explained by **X**� and indirectly by **Y** since **X**� and **Y**� are correlated. At this point it is enough to deflate **X** and use the residual **X***corr* as a drift-corrected vector:

$$\mathbf{X}\_{corr} = \mathbf{X} - \mathbf{X}\_{pred} \tag{10}$$

In summary, techniques based on recalibration of the system using a reference gas can give very good results, but care must be taken in the selection of such gas. It must be representative of all classes being measured, since they are supposed to drift in similar ways. It should also be stable along the time, available and easy to measure (Salit & Turk, 1998). Fitting all these constraints is in general complex and expensive, therefore methods overcoming this limitation have been explored.

#### **3.3 Attuning methods**

Attuning methods try to perform component correction without resorting to the use of calibration samples, but trying to deduce drift components directly from the training data. They can provide significant improvements in the classification rate over a fixed time period, and may also make possible to obtain real responses to be used in gas quantitative analysis.

#### **3.3.1 Independent component correction**

Di Natale et al. approached the problem of sensor drift with an attuning method considering also disturbances derived from the measurement environment (Di Natale et al., 2002a). In correction methods such as CC (Artursson et al., 2000) that are based on PCA, the computed principal components are mutually uncorrelated. However, this condition is not enough to guarantee that relevant signals are completely separated by disturbances (non-correlation does not necessarily implies statistical independence) and often a principal component carries information on both signal and disturbances. To tackle with this limitation Di Natale et al.

317

 !!

" !\$!

!!#!

attempt of developing an adaptive drift correction method dates back to the late 90's (Marco et al., 1998; Vlachos et al., 1997), only few publications have been proposed in the literature,

The first adaptive approaches for drift correction have been developed resorting to artificial neural networks. Neural networks are an important tool for building PaRC systems with several interesting features. First, neural networks are data driven self-adaptive methods in that they can adjust themselves to the data without any explicit specification of functional or distributional form for the underlying model. Second, they are universal functional approximations in that neural networks can approximate any function with arbitrary accuracy (Cybenko, 1989; Devijver & Kittler, 1982). Since any classification procedure seeks a functional relationship between the group membership and the attributes of the object, accurate identification of this underlying function is doubtlessly important. Third, neural networks are nonlinear models, which make them flexible in modeling real world complex relationships such as those presented by gas chemical sensors. Finally, neural networks are able to estimate the posterior probabilities, which provide the basis for establishing classification rule and

Several neural network based drift correction methods exploit Kohonen Self Organizing Maps (SOMs) (Kohonen, 1990). (Davide et al., 1994; Marco et al., 1998; Zuppa et al., 2004) use a single SOM common to all classes while (Distante et al., 2002) proposes to use a separate SOM for

!!!#!

**-**

**3.4.1 Neural networks**

Fig. 4. Orthogonal signal correction overall idea

and this problem still remains a challenge for the sensor community.

performing statistical analysis (Richard & Lippmann, 1991).

Drift Correction Methods for Gas Chemical Sensors in Artificial Olfaction Systems: Techniques and Challenges

propose to exploit the Independent Component Analysis (ICA) (Cornon, 1994) as a technique to separate a data matrix into a series of components each independent from the others. In this case, independence means that the information carried by each component cannot be inferred from the others, i.e., the joint probability of independent quantities is obtained as the product of the probability of each item.

ICA is applied to EN data to preserve only those components correlated with the sample features relevant to the application. In fact, ICA is computed on the training set (in this case no reference gas is required). Contrarily to Gaussian based models such as PCA and PLS, in ICA it is not possible to determine the variance carried by each component, and it is therefore not possible to choose the component to eliminate based on this information. In fact it is not possible to establish an order among the different components.

The way the authors propose to select components to eliminate is a supervised method: the independent components that mostly correlate with the objective of the measurement are chosen, while those more correlated with the disturbances are discarded. This solution provides good results especially in removing drift effect due to external causes (e.g., temperature, pressure) that can be monitored during experiments and used as variables to select the components to discard. However, whenever the causes of the drift are not completely known, selecting the components to discard may result complex or in some cases not possible.

#### **3.3.2 Orthogonal Signal Correction**

Recently, Padilla et al. (Padilla et al., 2010) proposed a very interesting drift attuning method based on Orthogonal Signal Correction (OSC). OSC is a signal processing technique first introduced by Wold et al. in (Wold et al., 1998) for NIR spectra correction.

OSC analyzes a set of sensor-array data **X** that represent a set of independent variables, and a concentration vector, or class label vector **C** that represent a set of dependent variables. The main idea is to remove variance of **X** which is not correlated to the variables in **C**. This is achieved by constraining the deflation of non-relevant information of **X** so that only information orthogonal to **X** must be removed (Fig. 4). The condition of orthogonality therefore assures that the signal correction process removes as little information as possible form the original data. Several different variants of the OSC algorithm have been presented in the literature. Padilla et al. applied the Wise implementation of the OSC algorithm4.

Even if OSC proved to be one of the most effective techniques for drift correction it does not completely solve the problem. One of the main drawbacks is the need for a set of training data containing a significant amount of drift making it possible to precisely identify the set of orthogonal components to be rejected. This may not be possible in industrial setups where training data are usually collected over a short period of time. Moreover, the introduction of new analytes to the recognition library represents a major problem since rejected components might be necessary to robustly identify these new classes.

#### **3.4 Adaptive methods**

Adaptive methods for drift correction try to adapt the PaRC model by taking into account pattern changes due to drift effects. The main benefit of the adaptation is an increasing time validity of the PaRC model which in turn reduces the request for calibration. Even if the first

<sup>4</sup> http://www.SIGE vector.com/MATLAB/OSC.HTML

12 Chemical Sensors

propose to exploit the Independent Component Analysis (ICA) (Cornon, 1994) as a technique to separate a data matrix into a series of components each independent from the others. In this case, independence means that the information carried by each component cannot be inferred from the others, i.e., the joint probability of independent quantities is obtained as the product

ICA is applied to EN data to preserve only those components correlated with the sample features relevant to the application. In fact, ICA is computed on the training set (in this case no reference gas is required). Contrarily to Gaussian based models such as PCA and PLS, in ICA it is not possible to determine the variance carried by each component, and it is therefore not possible to choose the component to eliminate based on this information. In fact it is not

The way the authors propose to select components to eliminate is a supervised method: the independent components that mostly correlate with the objective of the measurement are chosen, while those more correlated with the disturbances are discarded. This solution provides good results especially in removing drift effect due to external causes (e.g., temperature, pressure) that can be monitored during experiments and used as variables to select the components to discard. However, whenever the causes of the drift are not completely known, selecting the components to discard may result complex or in some cases

Recently, Padilla et al. (Padilla et al., 2010) proposed a very interesting drift attuning method based on Orthogonal Signal Correction (OSC). OSC is a signal processing technique first

OSC analyzes a set of sensor-array data **X** that represent a set of independent variables, and a concentration vector, or class label vector **C** that represent a set of dependent variables. The main idea is to remove variance of **X** which is not correlated to the variables in **C**. This is achieved by constraining the deflation of non-relevant information of **X** so that only information orthogonal to **X** must be removed (Fig. 4). The condition of orthogonality therefore assures that the signal correction process removes as little information as possible form the original data. Several different variants of the OSC algorithm have been presented in the literature. Padilla et al. applied the Wise implementation of the OSC algorithm4. Even if OSC proved to be one of the most effective techniques for drift correction it does not completely solve the problem. One of the main drawbacks is the need for a set of training data containing a significant amount of drift making it possible to precisely identify the set of orthogonal components to be rejected. This may not be possible in industrial setups where training data are usually collected over a short period of time. Moreover, the introduction of new analytes to the recognition library represents a major problem since rejected components

Adaptive methods for drift correction try to adapt the PaRC model by taking into account pattern changes due to drift effects. The main benefit of the adaptation is an increasing time validity of the PaRC model which in turn reduces the request for calibration. Even if the first

introduced by Wold et al. in (Wold et al., 1998) for NIR spectra correction.

might be necessary to robustly identify these new classes.

<sup>4</sup> http://www.SIGE vector.com/MATLAB/OSC.HTML

possible to establish an order among the different components.

of the probability of each item.

**3.3.2 Orthogonal Signal Correction**

not possible.

**3.4 Adaptive methods**

Fig. 4. Orthogonal signal correction overall idea

attempt of developing an adaptive drift correction method dates back to the late 90's (Marco et al., 1998; Vlachos et al., 1997), only few publications have been proposed in the literature, and this problem still remains a challenge for the sensor community.

#### **3.4.1 Neural networks**

The first adaptive approaches for drift correction have been developed resorting to artificial neural networks. Neural networks are an important tool for building PaRC systems with several interesting features. First, neural networks are data driven self-adaptive methods in that they can adjust themselves to the data without any explicit specification of functional or distributional form for the underlying model. Second, they are universal functional approximations in that neural networks can approximate any function with arbitrary accuracy (Cybenko, 1989; Devijver & Kittler, 1982). Since any classification procedure seeks a functional relationship between the group membership and the attributes of the object, accurate identification of this underlying function is doubtlessly important. Third, neural networks are nonlinear models, which make them flexible in modeling real world complex relationships such as those presented by gas chemical sensors. Finally, neural networks are able to estimate the posterior probabilities, which provide the basis for establishing classification rule and performing statistical analysis (Richard & Lippmann, 1991).

Several neural network based drift correction methods exploit Kohonen Self Organizing Maps (SOMs) (Kohonen, 1990). (Davide et al., 1994; Marco et al., 1998; Zuppa et al., 2004) use a single SOM common to all classes while (Distante et al., 2002) proposes to use a separate SOM for

**X***corr* = **X** + **X** · **C** (11)

319

Although this assumption is a limit for previous drift counteractions that do not allow for adaptation, it is a good approximation in this case since the correction factor is not a fixed quantity but it is continuously adapted to follow the variation imposed by the drift. The hypothesis of linearity is therefore assumed only within a restricted time window (or number

The correction algorithm can be coupled with any selected PaRC model (e.g., SVM, K-NN, Random Forests, etc.) and elaborates groups of consecutive measurements, denoted as

1. The initial correction factor, immediately after training of the PaRC model, is set to the null

(a) Correct each sample of the window using the current correction factor according to

(c) Use the corrected samples, and the classification results in an evolutionary process able to adapt the correction factor to the changes observed in the current window (see later

The adaptation of the correction factor exploits a Covariance Matrix Adaptation Evolutionary Strategy (CMA-ES) which is an optimization method first proposed by Hansen, Ostermeier, and Gawelczyk (Hansen et al., 1995) in mid 90s. It basically iteratively searches for a definite matrix which is able to minimize a given objective function. The approach is best suited for difficult non-linear, non-convex, and non-separable problems, of at least moderate dimensionality. In this specific case, the problem tackled by the CMA-ES is finding the best correction factor that makes possible to obtain similar distributions between corrected samples of a window compared to the samples used to train the PaRC model. Several metrics can be

In (Di Carlo et al., 2011), under the hypothesis of Gaussian distribution of samples around the related centroids, the CMA-ES is exploited to identify the best correction factor able to move each sample toward the centroid of the related class, thus compensating for drift effects that tend to move classes in the features space. The concept of proximity to the related centroid is introduced by testing the system using several types of objective functions that compute the

This method introduces a set of important improvements compared to previous adaptive

• it can work in cooperation with any PaRC algorithm thus allowing for the selection of the

• together with sample classifications it also provides corrected measures thus allowing also

of measurements) whose size can be adapted in order to respect this constraint.

2. For each window of samples the following operations must be performed:

(d) Correct each sample again using the updated correction factor; (e) Classify the new corrected samples and provide the obtained results.

distance of a sample from the related centroid with different metrics.

best classification model depending on the specific application;

windows, according to the following steps:

Drift Correction Methods for Gas Chemical Sensors in Artificial Olfaction Systems: Techniques and Challenges

(b) Classify each corrected sample;

equation (11);

exploited to identify similarity.

methods based on neural networks:

for gas quantitative analysis.

• it is more robust to discontinuity in the data;

matrix since no correction is required at this time;

for an explanation of how this process works);

each odor. A slightly different approach is proposed in (Vlachos et al., 1997) that exploits a different network architecture, namely the Adaptive Resonance Theory (ART) neural network (Carpenter et al., 1991) that allows for new classes to be created.

Regardless the type and the specific architecture of the neural network all proposed methods achieve drift correction by exploiting the inner way neural networks work. As common in pattern recognition, a preliminary training phase is used to train the neural network in the identification of samples from the sensor array. In this phase a set of samples is provided to the network which learns similarities according to different learning rules (Bishop, 1995). Each new sample slightly changes the way the nodes of the network behave. Both supervised learning in which training samples are already labeled into a set of classes and unsupervised learning in which training samples are not labeled and the set of classes is inferred by the network itself is possible. When training has converged, the net has learned the characteristics of the input patterns and can be used for classification. In this stage the learning capability of the network is usually disabled. The basic idea to allow for drift compensation is to maintain a certain learning rate also during the normal use of the network in order to learn changes of the input patterns due to drift effects. The learning rate must be kept to a low level in order to avoid over-fitting of the model.

Although neural networks represented the first attempt of implementing adaptive drift correction methods they have several drawbacks:


#### **3.4.2 Evolutionary algorithms**

Recently, a new adaptive drift correction method based on the use of evolutionary algorithms has been presented by Di Carlo et al. (Di Carlo et al., 2010; 2011). The overall idea is to exploit the learning capabilities of evolutionary algorithms to compute a multiplicative correction factor **C** used to correct incoming samples. Under the hypothesis that, in the very short term, the variation imposed by the drift can be considered linear in time, the paper proposes to apply the correction exploiting a linear transformation:

14 Chemical Sensors

each odor. A slightly different approach is proposed in (Vlachos et al., 1997) that exploits a different network architecture, namely the Adaptive Resonance Theory (ART) neural network

Regardless the type and the specific architecture of the neural network all proposed methods achieve drift correction by exploiting the inner way neural networks work. As common in pattern recognition, a preliminary training phase is used to train the neural network in the identification of samples from the sensor array. In this phase a set of samples is provided to the network which learns similarities according to different learning rules (Bishop, 1995). Each new sample slightly changes the way the nodes of the network behave. Both supervised learning in which training samples are already labeled into a set of classes and unsupervised learning in which training samples are not labeled and the set of classes is inferred by the network itself is possible. When training has converged, the net has learned the characteristics of the input patterns and can be used for classification. In this stage the learning capability of the network is usually disabled. The basic idea to allow for drift compensation is to maintain a certain learning rate also during the normal use of the network in order to learn changes of the input patterns due to drift effects. The learning rate must be kept to a low level in order to

Although neural networks represented the first attempt of implementing adaptive drift

• Drift correction is possible only for slow phenomena. A discontinuity in response between consecutive exposures (regardless of the time interval between the exposures) would

• Selecting the appropriate learning rate to keep during normal operation is complex and may strongly impact the correction capability. To the best of our knowledge no automatic

• The adaptive model is rather complex and typically requires a high number of training

• Several classification algorithms have been presented in the literature other than neural networks (e.g., SVM, K-NN, Random Forests, etc.). The literature shows that, depending on the specific set of data, some methods may perform better than others. Having a very tight integration between the correction method and the classification system prevents

• Finally, drift correction methods based on neural networks are mainly limited to gas classification applications. Whenever both classification and gas quantitative analysis are required, it would be difficult for current adaptive methods to be applied to obtain reliable

Recently, a new adaptive drift correction method based on the use of evolutionary algorithms has been presented by Di Carlo et al. (Di Carlo et al., 2010; 2011). The overall idea is to exploit the learning capabilities of evolutionary algorithms to compute a multiplicative correction factor **C** used to correct incoming samples. Under the hypothesis that, in the very short term, the variation imposed by the drift can be considered linear in time, the paper proposes to

immediately invalidate the classification model and would prevent adaptation.

methods have been proposed so far for efficiently tuning this parameter.

from exploiting the best PaRC model for the specific problem.

gas concentration measurements (Hui et al., 2003).

apply the correction exploiting a linear transformation:

(Carpenter et al., 1991) that allows for new classes to be created.

avoid over-fitting of the model.

**3.4.2 Evolutionary algorithms**

samples.

correction methods they have several drawbacks:

$$\mathbf{X}\_{corr} = \mathbf{X} + \mathbf{X} \cdot \mathbf{C} \tag{11}$$

Although this assumption is a limit for previous drift counteractions that do not allow for adaptation, it is a good approximation in this case since the correction factor is not a fixed quantity but it is continuously adapted to follow the variation imposed by the drift. The hypothesis of linearity is therefore assumed only within a restricted time window (or number of measurements) whose size can be adapted in order to respect this constraint.

The correction algorithm can be coupled with any selected PaRC model (e.g., SVM, K-NN, Random Forests, etc.) and elaborates groups of consecutive measurements, denoted as windows, according to the following steps:

	- (a) Correct each sample of the window using the current correction factor according to equation (11);
	- (b) Classify each corrected sample;
	- (c) Use the corrected samples, and the classification results in an evolutionary process able to adapt the correction factor to the changes observed in the current window (see later for an explanation of how this process works);
	- (d) Correct each sample again using the updated correction factor;
	- (e) Classify the new corrected samples and provide the obtained results.

The adaptation of the correction factor exploits a Covariance Matrix Adaptation Evolutionary Strategy (CMA-ES) which is an optimization method first proposed by Hansen, Ostermeier, and Gawelczyk (Hansen et al., 1995) in mid 90s. It basically iteratively searches for a definite matrix which is able to minimize a given objective function. The approach is best suited for difficult non-linear, non-convex, and non-separable problems, of at least moderate dimensionality. In this specific case, the problem tackled by the CMA-ES is finding the best correction factor that makes possible to obtain similar distributions between corrected samples of a window compared to the samples used to train the PaRC model. Several metrics can be exploited to identify similarity.

In (Di Carlo et al., 2011), under the hypothesis of Gaussian distribution of samples around the related centroids, the CMA-ES is exploited to identify the best correction factor able to move each sample toward the centroid of the related class, thus compensating for drift effects that tend to move classes in the features space. The concept of proximity to the related centroid is introduced by testing the system using several types of objective functions that compute the distance of a sample from the related centroid with different metrics.

This method introduces a set of important improvements compared to previous adaptive methods based on neural networks:


the sensors (CC method) remains linear, there is no problem for both methods in handling

321

Calibration is certainly the most time-intensive method for drift correction since it requires system retraining, additional measurements and labour. Hence, it must be used sparingly. Moreover, while this approach is quite simple to implement for physical sensors, where the quantity to be measured is exactly known, chemical sensors pose a series of challenging problems. Indeed, in chemical sensing, the choice of the calibrant strongly depends on the specific application especially when the sensing device is composed of a considerable number of cross-correlated sensors (Haugen et al., 2000; Hines et al., 1999). This leads to loss of generalization and lack of standardization which, on the contrary, is an important requirement

Several attuning methods, which deduce drift components directly from the training data, have been proposed to perform drift correction without resorting to the use of calibration. Padilla et al. (Padilla et al., 2010) compared OSC and CC. According to this work, OSC outperforms PCA-CC for a limited period of time while on the long term the advantage was not clear. More complex OSC models (with higher number of components) resulted in a better correction of variance for shorter times, but they degraded faster than simpler OSC. It was also observed that both OSC and PCA-CC methods are relatively robust regarding small calibration set sizes and perform rather well with a reduced calibration set. OSC results showed a higher variance than PCA-CC, but this was attributed to the number of components chosen in the models. On the other hand, PCA-CC needed a smaller training set and a single chemical species. This advantage may turn into disadvantage if the reference class is not

Therefore, OSC seems to be a promising approach: it is time tested in chemometrics, uses multivariate and calibrant information, and it is simple to implement and interpret. However, although attuning methods represent a step forward in the definition of an efficient solution to the drift correction problem, the main limitation is that they do no contain provisions for updating the model and thus may ultimately be invalidated by time evolving drift effects. Conversely, they can dramatically fail when the drift direction changes. This might happen if one sensor gets stressed. In this case one must resolve to use adaptive methods. Adaptive methods represent an important step forward in tackling the problem of sensor drift in artificial olfaction systems. Until now they have been not sufficiently investigated and contrasted with other methods in order to definitively assess their superior capabilities. In fact, current solutions still present some limitations that prevent their widespread application. In particular, they require equiprobable and frequent sampling of all classes to avoid that a single class drifts too much making it unrecognizable. Moreover they strongly rely on correct identification from the PaRC model to take track of how different classes change. Local errors in the classification may easily reduce the capability of adaptation, thus reducing the effectiveness of the correction system. Further research efforts are required in this direction.

In this chapter, after a brief overview of drift phenomenology in gas chemical sensors, the main challenges faced when developing drift correction techniques have been presented and discussed. A deep review of state-of-the-art methodologies proposed in the scientific

literature has been illustrated with a rational taxonomy of the various approaches.

non-linearities.

Drift Correction Methods for Gas Chemical Sensors in Artificial Olfaction Systems: Techniques and Challenges

for industrial systems.

properly chosen.

**5. Conclusions and future research directions**

#### **4. Comparisons and discussions**

Only partial comparisons and limited considerations can be done based on literature reports. In fact, due to lack of shared sensor dataset and public software codes, only few comparison works that partially contrast different approaches and report the pros and cons of each solution have been published (Hui et al., 2003; Padilla et al., 2010; Romain & Nicolas, 2010; Sisk & Lewis, 2005). Every paper tends to show good results related to the newly proposed approach. However, in many cases, measurements are collected during few days, thus neglecting long term drift effects, or drift was simulated.

Signal preprocessing techniques seem attractive for their simplicity, but cannot be considered an effective drift correction approach for chemical sensors. Baseline manipulation works well only in special cases that fit the assumptions of the employed feature transformation. Filtering methods in the frequency domain appear more promising. (Zuppa et al., 2007) compared the DWT filter with the relative baseline manipulation technique. PCA shows that the DWT approach is superior in terms of cluster dispersion with clear improvements in samples discrimination. The DWT method has been often combined to adaptive drift correction strategies based on neural networks. Nevertheless, frequency decomposition remains mostly useful for diagnostics and analysis because strong dependence of results from user defined parameters (e.g. the cut-off frequency).

According to Sisk and Lewis (Sisk & Lewis, 2005) using a single calibrant or a set of calibrants is perhaps the only robust method to mitigate drift effects even in the presence of sensor drift over an extremely long period of time. Sisk and Lewis implemented a simple linear sensor-by-sensor calibration scheme which proved to be effective at restoring the classification performance of difficult binary separation tasks affected by drift. In many other cases this univariate approach failed to work. Thus, one relevant open point is the tradeoff between model simplicity and robustness.

Univariate calibration methods, like MDC, are first-order approximations that do not exploit sensor cross-correlations. Empirically they provided good performance. However, they have several intrinsic limitations. The most important limitation is that the MDC method assumes that the relative percentage (increase or decrease) of the sensor response due to the drift is the same for the reference samples and test samples. This can be true for sample chemical compositions and concentrations very close to the reference gas but not otherwise. Therefore the amount of drift may be different for different samples, i.e., the drift for the reference gas and some samples is not same. The consequence of this is that the sensor drift does not have to be the multiplicative factor estimated from the reference gas, which the MDC method assumes.

As pointed out by Artursson et al. (Artursson et al., 2000) multivariate techniques, for instance CC, better perform both in terms of drift correction performance and flexibility. The reader may notice that both the CC method and the MDC method are linear, but they differ in what is assumed to be linear. For the CC method, drift is assumed to mainly influence the sensors along a straight line, described by the drift in the reference gas. In the univariate MDC method it is assumed that the relative change in each sensor is the same for the reference gas and all test gases. There exist cases where the drift is linear for one of the methods but not for the other.

Both MDC and CC methods suffer from the same disadvantages when it comes to handling non-linearities under their respective restrictions. However, as long as the relationship between the reference gas and the test gas (for the MDC method) or the relationship between 16 Chemical Sensors

Only partial comparisons and limited considerations can be done based on literature reports. In fact, due to lack of shared sensor dataset and public software codes, only few comparison works that partially contrast different approaches and report the pros and cons of each solution have been published (Hui et al., 2003; Padilla et al., 2010; Romain & Nicolas, 2010; Sisk & Lewis, 2005). Every paper tends to show good results related to the newly proposed approach. However, in many cases, measurements are collected during few days, thus

Signal preprocessing techniques seem attractive for their simplicity, but cannot be considered an effective drift correction approach for chemical sensors. Baseline manipulation works well only in special cases that fit the assumptions of the employed feature transformation. Filtering methods in the frequency domain appear more promising. (Zuppa et al., 2007) compared the DWT filter with the relative baseline manipulation technique. PCA shows that the DWT approach is superior in terms of cluster dispersion with clear improvements in samples discrimination. The DWT method has been often combined to adaptive drift correction strategies based on neural networks. Nevertheless, frequency decomposition remains mostly useful for diagnostics and analysis because strong dependence of results from user defined

According to Sisk and Lewis (Sisk & Lewis, 2005) using a single calibrant or a set of calibrants is perhaps the only robust method to mitigate drift effects even in the presence of sensor drift over an extremely long period of time. Sisk and Lewis implemented a simple linear sensor-by-sensor calibration scheme which proved to be effective at restoring the classification performance of difficult binary separation tasks affected by drift. In many other cases this univariate approach failed to work. Thus, one relevant open point is the tradeoff between

Univariate calibration methods, like MDC, are first-order approximations that do not exploit sensor cross-correlations. Empirically they provided good performance. However, they have several intrinsic limitations. The most important limitation is that the MDC method assumes that the relative percentage (increase or decrease) of the sensor response due to the drift is the same for the reference samples and test samples. This can be true for sample chemical compositions and concentrations very close to the reference gas but not otherwise. Therefore the amount of drift may be different for different samples, i.e., the drift for the reference gas and some samples is not same. The consequence of this is that the sensor drift does not have to be the multiplicative factor estimated from the reference gas, which the MDC method

As pointed out by Artursson et al. (Artursson et al., 2000) multivariate techniques, for instance CC, better perform both in terms of drift correction performance and flexibility. The reader may notice that both the CC method and the MDC method are linear, but they differ in what is assumed to be linear. For the CC method, drift is assumed to mainly influence the sensors along a straight line, described by the drift in the reference gas. In the univariate MDC method it is assumed that the relative change in each sensor is the same for the reference gas and all test gases. There exist cases where the drift is linear for one of the methods but not for the

Both MDC and CC methods suffer from the same disadvantages when it comes to handling non-linearities under their respective restrictions. However, as long as the relationship between the reference gas and the test gas (for the MDC method) or the relationship between

**4. Comparisons and discussions**

parameters (e.g. the cut-off frequency).

model simplicity and robustness.

assumes.

other.

neglecting long term drift effects, or drift was simulated.

the sensors (CC method) remains linear, there is no problem for both methods in handling non-linearities.

Calibration is certainly the most time-intensive method for drift correction since it requires system retraining, additional measurements and labour. Hence, it must be used sparingly. Moreover, while this approach is quite simple to implement for physical sensors, where the quantity to be measured is exactly known, chemical sensors pose a series of challenging problems. Indeed, in chemical sensing, the choice of the calibrant strongly depends on the specific application especially when the sensing device is composed of a considerable number of cross-correlated sensors (Haugen et al., 2000; Hines et al., 1999). This leads to loss of generalization and lack of standardization which, on the contrary, is an important requirement for industrial systems.

Several attuning methods, which deduce drift components directly from the training data, have been proposed to perform drift correction without resorting to the use of calibration. Padilla et al. (Padilla et al., 2010) compared OSC and CC. According to this work, OSC outperforms PCA-CC for a limited period of time while on the long term the advantage was not clear. More complex OSC models (with higher number of components) resulted in a better correction of variance for shorter times, but they degraded faster than simpler OSC. It was also observed that both OSC and PCA-CC methods are relatively robust regarding small calibration set sizes and perform rather well with a reduced calibration set. OSC results showed a higher variance than PCA-CC, but this was attributed to the number of components chosen in the models. On the other hand, PCA-CC needed a smaller training set and a single chemical species. This advantage may turn into disadvantage if the reference class is not properly chosen.

Therefore, OSC seems to be a promising approach: it is time tested in chemometrics, uses multivariate and calibrant information, and it is simple to implement and interpret. However, although attuning methods represent a step forward in the definition of an efficient solution to the drift correction problem, the main limitation is that they do no contain provisions for updating the model and thus may ultimately be invalidated by time evolving drift effects. Conversely, they can dramatically fail when the drift direction changes. This might happen if one sensor gets stressed. In this case one must resolve to use adaptive methods. Adaptive methods represent an important step forward in tackling the problem of sensor drift in artificial olfaction systems. Until now they have been not sufficiently investigated and contrasted with other methods in order to definitively assess their superior capabilities. In fact, current solutions still present some limitations that prevent their widespread application. In particular, they require equiprobable and frequent sampling of all classes to avoid that a single class drifts too much making it unrecognizable. Moreover they strongly rely on correct identification from the PaRC model to take track of how different classes change. Local errors in the classification may easily reduce the capability of adaptation, thus reducing the effectiveness of the correction system. Further research efforts are required in this direction.

#### **5. Conclusions and future research directions**

In this chapter, after a brief overview of drift phenomenology in gas chemical sensors, the main challenges faced when developing drift correction techniques have been presented and discussed. A deep review of state-of-the-art methodologies proposed in the scientific literature has been illustrated with a rational taxonomy of the various approaches.

Chen, D. & Chan, P. (2008). An intelligent isfet sensory system with temperature and drift compensation for long-term monitoring, *Sensors Journal, IEEE* 8(12): 1948–1959. Comini, E. & Sberveglieri, G. (2010). Metal oxide nanowires as chemical sensors, *Materials*

323

Cornon, P. (1994). Independent component analysis-a new concept?, *Signal Process* 36: 287. Cybenko, G. (1989). Approximation by superpositions of a sigmoidal function, *Mathematics of*

Davide, F., Di Natale, C. & D'Amico, A. (1994). Self-organizing multisensor systems for

Devijver, P. & Kittler, J. (1982). *Pattern recognition: A statistical approach*, Prentice/Hall

Di Carlo, S., Falasconi, M., Sánchez, E., Scionti, A., Squillero, G. & Tonda, A. (2010). Exploiting

Di Carlo, S., Falasconi, M., Sanchez, E., Scionti, A., Squillero, G. & Tonda, A. (2011).

Di Natale, C., Martinelli, E. & D'Amico, A. (2002a). Counteraction of environmental

Di Natale, C., Martinelli, E. & D'Amico, A. (2002b). Counteraction of environmental

Dillon, W. & Goldstein, M. (1984). *Multivariate analysis: Methods and applications*, New York

Distante, C., P., S. & Persaud, K. (2002). Dynamic cluster recognition with multiple

Fryder, M., Holmberg, M., Winquist, F. & Lündstrom, I. (1995). A calibration technique for an

Gardner, J. & Bartlett, P. (1999). *Electronic Noses. Principles and Applications*, Oxford University

Gardner, J. W. (1987). Pattern recognition in the warwick electronic nose, *Proc. of the 8th Int.*

Gobbi, E., Falasconi, M., Concina, I., Mantero, G., Bianchi, F., Mattarozzi, M., Musci, M. &

Goepel, W. & Schierbaum, K. (1995). Sno2 sensors: current status and future prospects, *Sensors*

Gutierrez-Osuna, R. (2000). Drift reduction for metal-oxide sensor arrays using canonical

Hansen, N., Ostermeier, A. & Gawelczyk, A. (1995). On the adaptation of arbitrary normal

juices: An emerging diagnostic tool, *Food Control* 21(10): 1374–1382.

Sberveglieri, G. (2010). Electronic nose and alicyclobacillus spp. spoilage of fruit

correlation regression and partial least squares, *Proceedings of the 7th International Symposium On Olfaction and Electronic Nose*, Institute of Physics Publishing,

mutation distributions in evolution strategies, *The Generating Set Adaptation. Proc. 6th*

self-organising maps, *Pattern Analysis and Applications* 5: 306–315.

electronic nose, *Proc. of Eurosensors IX*, pp. 683–686.

*Congress of European Chemoreception Res. Organisation*.

drift compensation, *Pattern Recognition Letters* 32(13): 1594–1603.

odour classification: Internal categorization, adaptation, and drift rejection, *Sensors*

evolution for an adaptive drift-robust classifier in chemical sensing, *Applications of*

Increasing pattern recognition accuracy for chemical sensing by evolutionary based

disturbances of electronic nose data by independent component analys, *Sensors and*

disturbances of electronic nose data by independent component analysis, *Sensors and*

*Today* 13(7-8): 36–44.

Drift Correction Methods for Gas Chemical Sensors in Artificial Olfaction Systems: Techniques and Challenges

International.

Wiley.

pp. 147–152.

*Control, Signals, and Systems* 2(4): 303–314.

*and Actuators B* 18-19: 244–250.

*Evolutionary Computation* pp. 412–421.

*Actuators B: Chemical* 82(2-3): 158–165.

*Actuators B: Chemical* 82(2-3): 158–165.

*and Actuators B: Chemical* 26(1-3): 1–12.

*Internat. Conf. Genetic Algorithms*, pp. 312–317.

Press, New York, NY, USA.

Drift correction is perhaps one the most relevant challenges in the field of chemical sensors. Indeed, in spite of constant improvements in micro/nano fabrication techniques that allowed the production of sensing devices with superior stability, it is still impossible to fabricate chemical sensors without drift.

Much work has been done in the last fifteen years to develop adapted techniques and robust algorithms. In spite further research is still required to compare them on some benchmark data set. A complete comparison of performance should take into account several parameters, for instance: the type of sensors and likely the use of hybrid arrays; the presence/absence of a drift model for the given sensors; the short and long term drift behavior; the size of the data set (feature space dimension, number of measurements); the problem complexity (number of classes, degree of overlap, type of chemicals/odors).

The problem of data correction in presence of simultaneous sources of drift, other than sensor drift, should be also investigated since it is often the case in practical situations. To this, one idea could be combining semi-supervised methods able to learn the actual source of drift, which might clearly change with the measured samples, with adaptive drift correction strategies that can account for the continuous drift direction change in the feature space.

Finally, many algorithms of drift correction could be adapted to the problem of sensor failure and subsequent replacement, which has received little attention until now but represents a relevant problem in the long term. Indeed, only few works can be found in the literature aiming at detection and correction of sensor faults, like for example Tomic et al. (Tomic et al., 2004) that studied calibration transfer techniques like MDC and CC related to sensor replacement.

The current trend in the field of artificial olfaction is to enlarge both the size of data sets and the dimensionality of the sensor arrays, by building for instance huge micro-machined sensor arrays made of many thousands sensors. In the near future, this shall definitively originate challenging issues for data analysis requiring more powerful drift correction algorithms able to handle large volume of data as well as high-dimensional features with acceptable time and storage complexities. Very large sensor arrays will also pose problems connected to variables redundancy. Therefore, a further relevant issue is the selection of appropriate and meaningful features to combine with the drift correctors, which can greatly reduce the burden of subsequent designs of classification/regression systems.

#### **6. References**


Bishop, C. M. (1995). *Neural Networks for Pattern Recognition*, Oxford Univ. Press.

Carpenter, G. A., Grossberg, S. & Reynolds, J. H. (1991). Artmap: Supervised real-time learning and classification of nonstationary data by a self-organizing neural network, *Neural Networks* 4(5): 565–588.

18 Chemical Sensors

Drift correction is perhaps one the most relevant challenges in the field of chemical sensors. Indeed, in spite of constant improvements in micro/nano fabrication techniques that allowed the production of sensing devices with superior stability, it is still impossible to fabricate

Much work has been done in the last fifteen years to develop adapted techniques and robust algorithms. In spite further research is still required to compare them on some benchmark data set. A complete comparison of performance should take into account several parameters, for instance: the type of sensors and likely the use of hybrid arrays; the presence/absence of a drift model for the given sensors; the short and long term drift behavior; the size of the data set (feature space dimension, number of measurements); the problem complexity (number of

The problem of data correction in presence of simultaneous sources of drift, other than sensor drift, should be also investigated since it is often the case in practical situations. To this, one idea could be combining semi-supervised methods able to learn the actual source of drift, which might clearly change with the measured samples, with adaptive drift correction strategies that can account for the continuous drift direction change in the feature space. Finally, many algorithms of drift correction could be adapted to the problem of sensor failure and subsequent replacement, which has received little attention until now but represents a relevant problem in the long term. Indeed, only few works can be found in the literature aiming at detection and correction of sensor faults, like for example Tomic et al. (Tomic et al., 2004) that studied calibration transfer techniques like MDC and CC related to sensor

The current trend in the field of artificial olfaction is to enlarge both the size of data sets and the dimensionality of the sensor arrays, by building for instance huge micro-machined sensor arrays made of many thousands sensors. In the near future, this shall definitively originate challenging issues for data analysis requiring more powerful drift correction algorithms able to handle large volume of data as well as high-dimensional features with acceptable time and storage complexities. Very large sensor arrays will also pose problems connected to variables redundancy. Therefore, a further relevant issue is the selection of appropriate and meaningful features to combine with the drift correctors, which can greatly reduce the burden

Aliwell, S. R., Halsall, J. F., Pratt, K. F. E., O'Sullivan, J., Jones, R. L., Cox, R. A., Utembe, S. R.,

Artursson, T., Eklov, T., Lundström, I., Mårtensson, P., Sjöström, M. & Holmberg, M.

Carpenter, G. A., Grossberg, S. & Reynolds, J. H. (1991). Artmap: Supervised real-time

*Chemometrics,Special Issue: Proceedings of the SSC6* 14(5-6): 711–723. Bishop, C. M. (1995). *Neural Networks for Pattern Recognition*, Oxford Univ. Press.

Hansford, G. M. & Williams, D. E. (2001). Ozone sensors based on wo3: a model for sensor drift and a measurement correction method, *Measurement Science & Technology*

(2000). Drift correction for gas sensors using multivariate methods, *Journal of*

learning and classification of nonstationary data by a self-organizing neural network,

chemical sensors without drift.

replacement.

**6. References**

12(6): 684–690.

*Neural Networks* 4(5): 565–588.

classes, degree of overlap, type of chemicals/odors).

of subsequent designs of classification/regression systems.


Polster, A., Fabian, M. & Villinger, H. (2009). Effective resolution and drift of paroscientific

Pratt, K. & Williams, D. (1997). Self diagnostic gas sensitive resistors in sour gas applications,

Richard, M. & Lippmann, R. (1991). Neural network classifiers estimate bayesian a posteriori

Röck, F., Barsan, N. & Weimar, U. (2008). Electronic nose: current status and future trends,

Romain, A. & Nicolas, J. (2010). Long term stability of metal oxide-based gas sensors for

Salit, M. & Turk, G. (1998). A drift correction procedure, *Analytical Chemistry* 70(15): 3184–3190. Sberveglieri, G. (1992). Classical and novel techniques for the preparation of sno2 thin-film

Schaller, E., Bosset, J. & Escher, F. (2000). Instability of conducting polymer sensors in an

Schierbaum, K., Weimar, U., Goepel, W. & Kowalkowski, R. (1991). Conductance, work

Sharma, R. K., Chan, P. C. H., Tang, Z., Yan, G., Hsing, I.-M. & Sin, J. K. O. (2001). Investigation

Shurmer, H., Fard, A., Barker, J., Bartlett, P., Dodd, G. & Hayat, U. (1987). Development of an

Sisk, B. C. & Lewis, N. S. (2005). Comparison of analytical methods and calibration methods

Tomic, O., Eklov, T., Kvaal, K. & Haugena, J. (2004). Recalibration of a gas-sensor array system

Vezzoli, M., Ponzoni, A., Pardo, M., Falasconi, M., Faglia, G. & Sberveglieri, G. (2008).

Vlachos, D., Fragoulis, D. & Avaritsiotis, J. (1997). An adaptive neural network topology for

Wold, S., Albano, C., Dunn, W., Edlund, U., Esbensen, K., Geladi, P., Helberg, S., Johansson, E.,

Wold, S., Antti, H., Lindgren, F. & Öhman, J. (1998). Orthogonal signal correction of near-infrared spectra, *Chemometrics and Intelligent Laboratory Systems* 44: 175–185.

gas sensors, *Sensors and Actuators B: Chemical* 6(1-3): 239–247.

sensor devices, *Sensors and Actuators B: Chemical* 81(1): 9–16.

vapor detector, *Sensors and Actuators B: Chemical* 104(2): 249–268.

related to sensor replacement, *Analytica Chimica Acta* 512: 199–206.

electronic nose, *Physics in Technology* 18(4): 170–176.

e-nose environmental applications: An overview, *Sensors and Actuators B: Chemical*

"uller, G. (1998). Gas-kinetic interactions of nitrous oxides with sno2 surfaces, *Sensors*

function and catalytic activity of sno2-based gas sensors, *Sensors and Actuators B:*

of stability and reliability of tin oxide thin-film for integrated micro-machined gas

for correction of detector response drift in arrays of carbon black-polymer composite

Exploratory data analysis for industrial safety application, *Sensors and Actuators B:*

degradation compensation of thin film tin oxide gas sensors, *Sensors and Actuators B:*

Lindberg, W. & Sjostrom, M. (1984). *Dependencies between variables: multiple regression (MR), partial least square modeling with latent variables (PLS) and PARC levels 3 and 4*,

*Sensors and Actuators B: Chemical* 45(2): 147–153.

probabilities, *Neural computation* 3(4): 461–483.

*and Actuators B: Chemical* 50(1): 85–94.

electronic nose system, *Analusis* 28(3): 217–227.

*Geosyst.* 10.

*Chem Rev.* 108(2): 705–725.

Drift Correction Methods for Gas Chemical Sensors in Artificial Olfaction Systems: Techniques and Challenges

146(2): 502–506. Ruhland, B., Becker, T. & M

*Chemical* 3(3): 205–214.

*Chemical* 131(1): 100–109.

*Chemical* 45(3): 223–228.

Reidel, pp. 61–74.

pressure sensors derived from long-term seafloor measurements, *Geochem. Geophys.*

325


20 Chemical Sensors

Haugen, J.-E., Tomic, O. & Kvaal, K. (2000). A calibration method for handling the temporal drift of solid state gas-sensors, *Analytica Chimica Acta* 407(1-2): 23–39. Haupt, K. & Mosbach, K. (2000). Molecularly imprinted polymers and their use in biomimetic

Hines, E., Llobet, E. & Gardner, J. (1999). Electronic noses: a review of signal processing techniques, *IEE Proceedings - Circuits, Devices and Systems* 146(6): 297–310.

Hui, D., Jun-Hua, L. & Zhong-Ru, S. (2003). Drift reduction of gas sensor by wavelet and principal component analysis, *Sensors and Actuators B: Chemical* 96(1-2): 354–363. International Union of Pure and Applied Chemistry (1991). Chemical sensors definitions and

Ionescu, R., Vancu, A. & Tomescu, A. (2000). Time-dependent humidity calibration for drift

Janata, J. (2009). *Principles of Chemical Sensors*, 2nd ed.2nd printing. edn, Springer Publishing

Kamp, B., Merkle, R. & Maier, J. (2001). Chemical diffusion of oxygen in tin dioxide, *Sensors*

Llobet, E., Brezmes, J., Ionescu, R., Vilanova, X., Al-Khalifa, S., Gardner, J. W., Barsan, N. &

Marco, S., Ortega, A., Pardo, A. & Samitier, J. (1998). Gas identification with tin oxide

Mielle, P. (1996). Managing dynamic thermal exchanges in commercial semiconduting gas

Nelli, P., Faglia, G., Sberveglieri, G., Cereda, E., Gabetta, G., Dieguez, A., Romano-Rodriguez,

Owens, W. B. & Wong, A. P. S. (2009). An improved calibration method for the drift of

Pardo, M. & Sberveglieri, G. (2004). Electronic olfactory systems based on metal oxide

Pearce, T. C., Shiffman, S. S., Nagle, H. T. & Gardner, J. W. (2003). *Handbook of machine olfaction*,

Persaud, K. & Dodd, G. (1982). Analysis of discrimination mechanisms in the mammalian

Kohonen, T. (1990). The self-organizing map, *Proceedings of the IEEE* 78(9): 1464–1480.

*Transactions on Instrumentation and Measurement* 47(1): 316–321.

and structural characterization, *Thin Solid Films* 371(1-2): 249–253.

*Deep-Sea Research Part I-Oceanographic Research Papers* 56(3): 450–457. Padilla, M., Perera, A., Montoliu, I., Chaudry, A., Persaud, K. & Marco, S. (2010). Drift

sensors, *Sensors and Actuators B: Chemical* 34(1-3): 533–538.

semiconductor sensor arrays, *MRS Bulletin* 29(10): 703–708.

olfactory system using a model nose, *Nature* 299: 352–355.

*and Intelligent Laboratory Systems* 100(1): 28–35.

"om, I. (1997). Drift counteraction in odour recognition applications: lifelong

corrections in electronic noses equipped with sno2 gas sensors, *Sensors and Actuators*

Correig, X. (2002). Wavelet transform and fuzzy artmap-based pattern recognition for fast gas identification using a micro-hotplate gas sensor, *Sensors and Actuators B:*

sensor array and self-organizing maps: adaptive correction of sensor drifts, *IEEE*

A. & Morante, J. R. (2000). The aging effect on sno2-au thin film sensors: electrical

the conductivity sensor on autonomous ctd profiling floats by theta–s climatology,

compensation of gas sensor array data by orthogonal signal correction, *Chemometrics*

Holmberg, M., Davide, F., Di Natale, C., D'Amico, A., Winquist, F. & Lundstr

calibration method, *Sensors and Actuators B: Chemical* 42(3): 185–194.

sensors, *Chemical Reviews* 100(7): 2495–2504.

classification, *Pure&App/. Chern.* 63(9).

*and Actuators B: Chemical* 77(1-2): 534–542.

*B: Chemical* 69(3): 283–286.

Company, Incorporated.

*Chemical* 83(1-3): 238–244.

Weinheim: Wiley-VHC.


**0**

**15**

*USA*

**Statistical Analysis of Chemical Sensor Data**

Chemical sensors measure and quantify substances via their associated chemical or physical response, thus providing data that can be analyzed to address a scientific question of interest (Eggins; 2002). Used in a variety of applications from monitoring to medicine, chemical sensors vary vastly by construction, style, format, size and dimension, and complexity. The common, underlying feature of these sensors lies in the associated data, which are abundant with technical and structural complexities, making statistical analysis a difficult task. These data further share a common need to be measured, analyzed, and interpreted properly so that

There are many image analysis algorithms available and amenable to a myriad of chemical analysis problems, thus potentially applicable to chemical sensor data problems in particular. By applying these tools to chemical sensor data, we can optimize and evaluate a chemical sensor's ability to perform its intended tasks. This chapter is designed to give an overview of the modern statistical algorithms that are commonly used when designing and analyzing chemical sensor experiments. Without focusing the discussion around a specific chemical sensor platform, our goal is to provide a general framework that will be applicable, to some

From the beginning to the end of an experiment, various statistical methods can be employed to improve or understand the current scientific analysis. We decompose a general experiment into several facets and provide the motivation for potential statistical methods that can be applied within each component. Section 2 describes the pre-processing techniques that are available for summarizing the low-level image or signal data so that the subsequent scientific questions can be properly addressed. In this section, we particularly focus on removing background noise in order to isolate the chemical sensor signal data, quantifying this data, and normalizing it so that the resulting data are scalable across conditions. Section 3 introduces the higher-level statistical approaches that are used to analyze the pre-processed data, and Section 4 describes the statistical computational tools available for use to perform the analyses. Finally, Section 5 demonstrates and motivates the significance of these methods via a chemical

sensor case study, and Section 6 concludes the chapter with discussion and summary.

For the analysis of chemical sensor data, many of the suggested means to resolve low-level analysis problems have been posed by computer scientists or engineers, with little statistical contribution or consideration, and they remain open problems because of significant

**1. Introduction**

the resulting inference is accurate.

degree, for all chemical sensor data.

**2. Pre-processing**

Jeffrey C. Miecznikowski1 and Kimberly F. Sellers2

<sup>1</sup>*SUNY University at Buffalo* <sup>2</sup>*Georgetown University*

