Insights into the Mechanism of Gas Sensor Operation

Abstract

Since the development of the first models of gas detection on metal-oxide-based sensors much effort has been made to describe the mechanism responsible for gas sensing. Despite progress in recent years, a number of key issues remain the subject of controversy; for example, the disagreement between the results of electrophysical and spectroscopic characterization, as well as the lack of proven mechanistic description of surface reactions involved in gas sensing. In the present chapter the basics as well as the main problems and unresolved issues associated with the chemical aspects of gas sensing mechanism in chemiresistors based on semiconducting metal oxides are addressed.

“Sensors have a ‘life cycle’ consisting of preparation, activation, operation with deactivation and, possible, regeneration. Thus understanding the performance in terms of reaction and conductance mechanisms is only a part of the total understanding of a sensor.”

Dieter Kohl, Sensors and Actuators 1989, 18, 71.

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1 Chemiresistors: From Semiconductor Surfaces to Gas Detectors

Since the early 1920s numerous investigations have demonstrated the influence of the gas atmosphere on conductivity, free carrier mobility, surface potential, and work function on a number of semiconductors (see summary of early works in [1–13]). This led to the understanding that the surface of semiconductors is highly sensitive to chemical reactions and chemisorptive processes [3, 14–20] and resulted finally in the “theory of surface traps” (Brattain and Bardeen [21]), “boundary layer theory of chemisorption” [10, 22, 23] (Engell, Hauffe and Schottky) and “electron theory of chemisorption and catalysis on semiconductors” (Wolkenstein [5–7, 24]). They laid also the theoretical foundations for the subsequent development of metal-oxide-based gas sensors.

Although from this understanding to the use of semiconductors as gas sensors “was, in principle, a small step” [25], the idea of using the changes in conductivity of a semiconducting metal oxide for gas detection was not conceived until the middle of the 1950s. The earliest written evidence came in 1956, in the Diploma Thesis performed in Erlangen under supervision of Mollwo and Heiland and entitled “Oxygen detection in gases changes in the conductivity of a semiconductor (ZnO)” [26], the results discussed later in [1, 27]: “If one exposes a zinc oxide layer which has been given a previous heating at 500 K in a high vacuum to oxygen at a constant pressure, the conductivity falls very rapidly initially and more slowly later. If one then increases the oxygen pressure suddenly, the current of the conductivity exhibits a kink when plotted as a function of the time. In this change the slopes immediately before and immediately after the kink point are proportional to the partial pressure of oxygen. One can use this effect to relate a known and an unknown concentration of oxygen often even under conditions in which one has a mixture of gases…” (cited from Ref. [1]). In 1957, Heiland showed that the “well-conducting surface layer on zinc oxide crystals provides a new, very sensitive test for atomic hydrogen” [28] and Myasnikov demonstrated that ZnO films can be used as a highly-sensitive oxygen-analyzer [29]. Later he developed this “to the method of semicondutor probes”, which allows for “studying free radical processes” and for detecting “free active particles and to measure their concentration under stationary and non-stationary conditions in gases and liquids” [30]. However, the conditions under which ZnO was able to operate as a “sensing device” were far from the real ambient conditions (and, accordingly, from a practical application); the “sensitive” effects were observed: (i) in vacuum conditions, exposed to oxygen or hydrogen, (ii) after “activation” or “sensitization” of the surface by heating in H₂ and in UHV.

The practical use of metal-oxide-based gas sensors in normal ambient conditions was not considered until 1962, when Seiyama et al. reported that a ZnO film can be used as a detector of inflammable gases in air [31] (see also [32]), and Taguchi claimed that a sintered SnO₂ block can also work in the same way [33] (for the history of TGS (Taguchi Gas Sensor) sensors, see [34]). The latter approach became very successful, leading to the foundation of the first sensor company (Figaro Engineering Inc.), which established mass production and started selling the TGS sensors in 1968.

Since then, many different metal oxides have been investigated as sensing materials (see, for example, Ref. [35] for a comprehensive review), however, tin dioxide (SnO₂)—alone or “activated” with small quantities of noble metals/their oxides (Pd, Pt, Au)—has remained the most commonly used and the best-understood prototype material in commercial gas sensors [36] as well as in the basic studies of the gas sensing mechanism [35–43].

2 Characterization Methodology: From Prototype Surfaces to Operating Sensors

The detailed characterization of metal oxide sensors requires the “simultaneous measurement of the gas response and the determination of molecular adsorption properties for a better understanding of gas sensing mechanisms” [44]. This measurement can be done either on clean and well-defined surfaces in ultrahigh vacuum (UHV) conditions or at temperatures and pressures that mimic real sensor operating conditions (“in situ” [45]). Continuous progress has been made during the past few years for the latter strategy, i.e. toward the use of in situ and operando spectroscopic techniques.

The “crossing of interests” [46] and “bridges of physics and of chemistry across the semiconductor surface” [47] determined experimental methodology applied for the gas-semiconductor studies in general and gas sensing studies in particular in the course of the last 50 years.

The first systematic methodological approach (“design concept for chemical sensors”) in gas sensing-studies was explicitly formulated in 1985, in a series of papers entitled “Development of chemical sensors: empirical art or systematic research?” ([48–50], see also [51]]).

The underlying concept was that by “studying the surface of single crystals under well-defined conditions, one might try to achieve a better separation of parameters influencing the properties of gas sensors” [52]. The reactions were addressed by surface spectroscopic methods under ultra-high-vacuum (UHV) conditions on well-defined “prototype” structures while the sensor performance was tested under realistic measuring conditions on the structures of practical importance (“sensors”).

This “comparative approach” advanced the basic understanding of surface reactions and the corresponding conduction mechanism responsible for gas sensing. However, it showed also the limits of surface science in gas-sensing studies and led to the understanding that if spectroscopic and electrical data are not obtained simultaneously, they must be obtained (i) under the same conditions and (ii) on identical samples. A comprehensive description of surface reactions on SnO₂ published in 1989 resulted from simultaneous thermal desorption spectroscopy (TDS; i.e. reactive scattering of a molecular beam) and conductance measurements [52]. These measurements were applied to SnO₂ single crystals and thin evaporated films exposed to a certain dose of CH₃COOH, CO or CH₄ in UHV conditions while at the sensor operating temperature.

As an alternative to sensing studies on single crystals or thin films, sensing characterization studies have focused on a combination of electrical measurements with spectroscopic investigations of catalysis reactions on polycrystalline , high surface-area materials with the aim to “link semiconductor studies with catalytic studies” [9]. However, most of the studies were performed under conditions far from the real working conditions of sensors (for the summary of numerous studies on semiconducting metal oxides, see references [4, 13, 53]). Besides spectroscopic and catalytic (kinetic) investigations (SnO₂: kinetic studies of CO oxidation [54], IR spectroscopic studies of water, CO₂ and CO adsorption [55], (summarized in Ref. [56]), EPR investigations of oxygen adsorption, [57], (reviewed in references [58, 59])), the improvements were concentrated on devising systems and in situ cells for combined (i.e., performed under the same conditions on “identical” samples) and simultaneous electrical, catalytic and spectral investigations.

These activities, however, were overlooked by the sensor community at that time, as in situ electrical characterization of realistic (“polycrystalline”) samples, namely, the Hall effect measurements (1982 [60]), changes in work function (CPD) by the Kelvin method (1983 [61]), ac impedance spectroscopy (1991 [62, 63]), simultaneous work function change and conductance measurements (1991 [64]) were preferred for studying the mechanism of operating sensors [99].

Later, this approach was followed systematically in the number of works (reviewed in references [38, 65], recent works in references [66–70] and references therein) to elucidate a mechanism of gas detection on SnO₂-based sensors. Local electronic properties (e.g., the density of states in the region near the band gap) of a sensing material were determined by scanning tunnelling microscopy and spectroscopy (STM-STS) in vacuum conditions [71–73] or under N₂, CO and NO₂ (at room temperature) [74].

By the end of the 1990s, the spectroscopic techniques for gas-sensing studies were differentiated according to conditions under which they can be applied: those that may be applied “under in situ real operation conditions of the sensors” and those that may be applied “under ideal conditions far away from real practical world” [75]. This differentiation subsequently resulted in the systematic combination of phenomenological and spectroscopic measurement techniques under working conditions of sensors [38], and thus lead to the in situ and operando methodology.

Continuous progress has been made during the past few years for the latter strategy, that is, the use of in situ and operando spectroscopic techniques (see [76, 77]):

• In situ spectroscopy: spectroscopic characterization of sensing materials under operation conditions or conditions relevant to operation conditions; herein, the sensing performance of this material may be not characterized or may be characterized in a separate experiment,

• Operando spectroscopy: spectroscopic characterization of an active sensing element in real time and under operating conditions with the simultaneous read-out of the sensor activity and simultaneous monitoring of gas composition.

These definitions determine the boundary conditions under which an “operando” experiment is performed:

• 1. on a sensing element, which itself is a complex device and consists of several parts: in solid-state devices with an electrical response, for example, the sensing layer is deposited onto a substrate to which electrodes for an electrical read-out are attached (“transducer”); therefore the assessment of their interfaces is of paramount importance for understanding the overall sensing mechanism;

• 2. in real time: a sensor is devised to respond to the changes in the gas atmosphere as fast as possible; accordingly, it demands a fast spectroscopic response;

• 3. under operating conditions: these can vary from ambient conditions (RT and atmospheric pressure) to high temperatures and pressures;

• 4. with simultaneousread-out of sensor activity: the gas concentration to be measured is transduced by the sensor into an electrical or other convenient output, depending on the modus operandi of sensor (optical, mechanical, thermal, magnetic, electronic, or electrochemical) and the transducer technology;

• 5. with simultaneous monitoring of gas composition; on-line gas analysis in gas sensing plays a twofold role: (i) the output compositions and concentrations provide data about reaction products and possible reaction paths and (ii) the input concentration verifies the sensor input data (concentration of the component to be detected).

The operando methodology couples electrical (“phenomenological”) and spectroscopic techniques and, aims to correlate the sensor activity with the spectroscopic data obtained under the same conditions on the same sample (Fig. 1.1). In an ideal case, one would obtain four types of information: (i) gas-phase changes (and reaction products) from on-line gas analysis, (ii) species adsorbed on the surface, (iii) changes in the oxide surface and lattice, and (iv) sensor activity.

Fig. 1.1

Methodological approach for simultaneous spectroscopic and electrical (“phenomenological”) characterisation of metal-oxide-based gas sensors. Modified from ref. [76]

3 Mechanism of Gas Detection: Never Ending Story About Oxygen

Epigraph

Due to the electron affinity of oxygen, the electron can be transferred to the chemisorbed oxygen and, consequently, there will be no chemisorbed oxygen atoms, but ions, in the surface

K. Hauffe, Adv. Catal. 1955 , 7, 213– 257.

Since the development of the first models of gas detection on metal-oxide-based sensors [78, 79] much effort has been made to describe the mechanism responsible for gas sensing (see, for example, [80–82]). Despite progress in recent years, a number of key issues remain the subject of controversy; for example, the disagreement between electrophysical and spectroscopic investigations, as well as the lack of a proven mechanistic description of surface reactions involved in gas sensing.

Nowadays, the influence of the gas atmosphere on the electrical transport properties of semiconductors and, accordingly, the operation of metal-oxide-based gas sensors is currently described by the combination of two different models; they are the ionosorption and the reduction-reoxidation mechanisms (Table 1.1). The ionosorption model considers only the space-charge effects/changes of the electric surface potential that results from the “ionosorption” of gaseous molecules. The reduction-reoxidation model explains the sensing effects by changes in the oxygen stoichiometry, that is, by the variation of the amount of the (sub-) surface oxygen vacancies and their ionization. The latter involves explicitly the diffusion of oxygen (or oxygen vacancies) from/in the bulk of the sensing material.

Table 1.1 Gas sensing mechanism on SnO₂ according to ionosorption and oxygen vacancy models

4 Oxygen Ionosorption

The electrical conductivity and work function can be described as collective physical properties of semiconductors which are changed by an ionosorption process and are accessible to measurement. The key in the mechanistic description of gas sensing is “oxygen ionosorption” and reaction of reducing gases with ionosorbed oxygen ions.

The oxygen influence on the electrical conductivity and work function is very well documented. For SnO₂, for example, exposure of single crystals (Ref. [83] and refs therein), polycrystalline samples (porous films [84], powders [57], pressed bars[85]) as well as one dimensional nanostructures [86, 87] to oxygen leads to the (i) decrease in the electrical conductivity and in the concentration of conduction electron density (Hall effect measurements [57]), (ii) increase in the work function observed in UHV conditions (XPS/UPS [88]) and under atmospheric pressure (simultaneous Contact Potential Difference, CPD, and conductance measurements [84]). Similar effects have been also observed on TiO₂ and ZnO (see early publications on TiO₂ [89, 90] [91] and on ZnO [3, 92–95]).

The magnitude of the changes depends strongly on the oxide temperature (see for example [85] and [84]), particle size and pre-treatment (history). On high-surface area and reduced samples the changes are much higher in comparison to single crystals and oxidised samples. The reduced samples show activity at temperatures as low as room temperature (r.t.), for oxidised samples higher temperatures (>100 °C) are needed. This difference between oxidised and reduced samples is usually ignored by the “ionosorption theory”.

Because the detailed mechanism of oxygen adsorption cannot be derived directly from electrophysical investigations[96], the chemistry of adsorbed surface oxygen on SnO₂ was adapted from the “ionosorption model” [97–100]. It was assumed that the thermally stimulated processes of oxygen adsorption, dissociation and charge transfer involve only conduction electrons [4, 81]:

 

${\text{O}}_{2} ({\text{gas}}) \leftrightarrow {\text{O}}_{2} ({\text{ads}})$	Physisorption
${\text{O}}_{2} ({\text{ads}}) + {\text{e}}^{ - } ({\text{CB}}) \leftrightarrow {\text{O}}_{2}^{ - } ({\text{ads}})$	Ionosorption
${\text{O}}_{2}^{ - } ({\text{ads}}) + {\text{e}}^{ - } ({\text{CB}}) \leftrightarrow {\text{O}}_{2}^{2 - } ({\text{ads}}) \leftrightarrow 2{\text{O}}^{ - } ({\text{ads}})$	Ionosorption
${\text{O}}^{ - } ({\text{ads}}) + {\text{e}}^{ - } ({\text{CB}}) \leftrightarrow {\text{O}}^{2 - } ({\text{ads}})$	Ionosorption
${\text{O}}^{2 - } ({\text{ads}}) \leftrightarrow {\text{O}}^{2 - } (1{{\text{st}}} \;{\text{bulk}}\;{\text{layer}})$	Diffusion

The nature of the ionised oxygen species is assumed to depend on the adsorption temperature (Fig. 1.2). At low temperatures (150–200 °C) oxygen adsorbs on SnO₂ non-dissociatively in its molecular form (as charged O₂ ⁻_ads ions). At high temperatures (between 200 and 400 °C or even higher) it dissociates to atomic oxygen (as charged O ⁻_ads or O ²⁻_ads ions) [4, 37, 75, 80, 81, 98, 99, 101]. Neutral oxygen species such as physisorbed oxygen, O_{2, phys}, are assumed not to play any role in gas sensing. The same holds for the lattice oxygen ions, O ²⁻_lat , in bulk materials at temperatures not high enough for fast oxygen exchange reactions (see detailed discussion below).

Fig. 1.2

The simulated equilibrium coverages of the oxygen species. The transition from O₂ ⁻ to O⁻ is calculated to be around 700 K (intersection at 427 °C). Copyright Elsevier, reproduced with permission from Ref. [100]

At this point, a problem of semantics starts to bring additional confusion, especially in the operational use of the terms “charged” species and the “charge transfer” at the surface. In semiconductor physics, the charge transfer implies by definition the transfer of free charge carriers, that is, conduction electrons or holes. Accordingly, the species that influence the electrical conductivity are regarded as “charged” or “ionized”. They are represented by free oxygen ions. The species that do not influence the conductivity are regarded as “neutral”. They are represented by physisorbed oxygen molecules.

The phenomenological model describes the oxygen ionosorption on an n-type semiconductor as follows:

• ionosorbed oxygen species are formed due to the transfer of conduction electrons from the semiconductor;

• they can be regarded as free oxygen ions which are electrostatically stabilized in the vicinity of the surface;

• there are no other adsorbed oxygen species besides physisorbed oxygen and oxygen ions;

• physisorbed oxygen is electrically neutral and oxygen ions are electrically active (“charged”) species.

The simplified picture showing the influence of adsorption on surface conductivity and work function is as follows. An oxygen molecule becomes physisorbed at the surface. In the next step, an electron from the oxide’s conduction band is trapped at the adsorbed oxygen molecule. The adsorbed oxygen molecule and surface itself become negatively charged. The flow of electrons from the semiconductor into the chemisorbed layer, without any diffusion of ionic species at the same time, induces a space charge between the interior of the semiconductor and its surface. The negative surface charge is compensated by a positive charge and a space-charge layer forms below it. This positive space-charge layer has reduced electron densities as compared to the bulk and is called an “electron-depleted layer or a charge depleted layer”. As a result the energy band, pertaining to the surface, bends upwards with respect to the Fermi level. This causes the creation of barriers on the surface, (q∆V_S > 0), due to the increasing work function, (q∆V_S > 0), and decreasing surface conductance (G = Gexp(–q∆V_S/kT) (Fig. 1.3). The process of charge transfer continues until equilibrium is reached and a steady state is achieved. To prevent very high double-layer potentials, the total amount of the “charged” species is limited to 10⁻⁵–10⁻³ monolayer which corresponds approximately to 1 V of the surface potential V_S (this is the so-called Weisz limitation, see original [18] and discussion in [4]). Within the framework of this concept, the operation of SnO₂-based sensors is described as follows: oxygen adsorbs in a delocalized manner, trapping electrons from the conduction band and forming ions—“charged” molecular (O₂ ^{− ads)} and atomic (O⁻ads, O₂ ⁻ads) species—electrostatically stabilized at the surface in the vicinity of metal cations. This happens under real working conditions of sensors, between 100 and 450 °C, at atmospheric pressure, at 20.5 vol. % background oxygen.

Fig. 1.3

Band bending on an n-type semiconductor after ionosorption of oxygen. Work functions Φ of semiconductors contain three contributions; e.g. the energy difference between the Fermi level and conduction band in the bulk (E_C−E_F)_b, band bending qV_S (q denotes elementary charge) and electron affinity χ: Φ = (E_C − E_F)_b + qV_S + χ (due to the definition, V_S = E_C,S−E_C,B). For ionosorption the work function follows only the change in band bending (ΔΦ = qΔV_S). The z₀ denotes the depth of the depletion region; μ—the electrochemical potential; E_{V, B} and E_{V, S}—valence band edge in the bulk and at the surface, respectively; E _d,S —donor level at the surface; E _{C, B} and E _{C, S} —conductance band edge in the bulk and at the surface, respectively; E_F—Fermi level; O_{2 ,gas} is an oxygen molecule in the ambient atmosphere; O_2,phys—a physisorbed oxygen species; O ^−α_βS —a chemisorbed oxygen species (α = 1 and α = 2 for singly and doubly ionised forms, respectively; β = 1 and β = 2 for atomic and molecular forms, respectively)

Reducing gases, like CO, react with the oxygen ions (by either Eley–Rideal or Langmuir–Hinshelwood mechanism) freeing electrons that return to the conduction band.

The ionosorption theory explains also the increase in the sensing performance with decreasing crystal size. Firstly, the reactivity of nanomaterials is mainly determined by the so-called “smoothly scalable” size-dependent properties which are related to the fraction of atoms at the surface [102]. As the crystal size decreases, the surface-to-volume ratio increases proportionally with the inverse of the crystal size. The increase in the total surface-to-volume ratio with respect to the size decrease generates more “reactivity” due to a dominant surface-like behavior caused by an increased fraction of atoms at the surface [102]. Thus, all properties which depend on the surface-to-volume ratio change continuously and extrapolate rapidly at very low crystal sizes. As a consequence, nanoparticles with increased surface-to-volume ratio are expected to be more reactive and accordingly, more gas sensitive.

With decreasing crystal size there is also a transition from a partly to a completely charge depleted particle that can be observed, depending on the ratio between the crystal and the Debye screening length L _D (Λ in Fig. 1.4) (for calculation, see for example [37, 103]). For partly depleted particles, when surface reactions do not influence the conduction in the entire layer, the conduction process takes place in the bulk region. Formally, two resistances occur in parallel, one influenced by surface reactions and the other not; the conduction is parallel to the surface, and this explains the limited sensitivity [37, 39]. Fully depleted particles possess higher sensitivity as the charge depletion layer fully impacts the conduction channel within the nanoparticle, thus achieving better performance in gas exposure experiments [43].

Fig. 1.4

The mechanism controlling the conductivity change and its magnitude depends on the ratio between grain size (D) and Debye screening length (Λ). If D > 2Λ, the depletion of the surface between the grain boundaries controls the conductivity. In this case low response to the analyte is expected as only a small part of the semiconductor is affected by interaction with analyte. If D ≤ 2Λ, the whole grain depleted and changes in the surface oxygen concentration affects the whole semiconductor resulting in high response. Copyright Wiley–VCH, reproduced with permission from Ref. [43]

Summarizing, the atomic charged oxygen ion (O⁻ads) is assumed to be of particular importance in gas sensingbecause “the O⁻ion appears to be more reactive of the two possibilities and thus more sensitive to the presence of organic vapours or reducing agents…” [81]. Accordingly, “there are two important questions to resolve here: First, under what conditions does O⁻dominate over O ₂ ⁻? Second, what is the total surface charge as a function of … temperature and partial oxygen pressure?” [81] As a consequence, ambitious efforts have been made (i) to calculate the surface coverage by different types of ionosorbed oxygen [6, 104–106] (Fig. 1.2) and (ii) to correlate the overall conductance of the sensors with the chemical state of charged oxygen species at the surface [37, 107].

The contradiction arises when connecting the main statements of the ionosorption model to common chemical sense and spectroscopic findings.

A first example of this is to note that oxygen ionosorption should be reflected in equal changes in the work function and band bending, kTln(G₀/G) = qΔV_S = ΔΦ (see also Fig. 1.3). These values can be independently obtained, for example, in the simultaneous CPD (here ΔV_CPD = −ΔΦ = qΔV_S) and conductance measurements (here qΔV_S = kTln(G₀/G), see an example in [38].

However, even if one can measure formal evidence for the pure oxygen ionosorption (kTln(G₀/G) = ΔΦ, Fig. 1.5, the transient changes observed (qΔV_s = 0, ΔΦ = 0 after 50 h) reflects very slow surface processes. These slow changes are in sharp contrast to the fast charging expected at the oxide surface (see Ref. [108], charging takes less than 5 ms even at 250 K) and this discrepancy is not explained by “ionosorption theory”.

Fig. 1.5

a The contact potential difference (CPD = −ΔΦ) 1 and the resistance 2 have been recorded at different O₂ concentrations 3 on the nanocrystalline SnO₂ at 400 °C in dry nitrogen at atmospheric pressure (adapted from Ref. [84]). b Calculated from a work function change (ΔΦ) 1 and band bending (qΔV_S = kTln(G₀/G)) 2 changes. Copyright Wiley–VCH, reproduced with permission [145]

The results shown in Fig. 1.5 strongly suggest that at 400 °C all other species, besides ionic ones, can be regarded as being of secondary importance. However, at 200 °C a completely different behavior of the changes of the work function appear. This is illustrated by Fig. 1.6 where changes in work function, band bending and electronic affinity due to a pulse of 300 ppm oxygen are displayed for 200–400 °C, respectively. The most important difference is the strong decrease in electronic affinity at 200 °C. Such effects did not appear at 400 °C. As shown in Ref. [38], the changes in electronic affinity are connected with the formation or loss of dipolar species between adsorbate and adsorbent accompanied by localized bonding. Therefore, in order to get an explanation of the experimental results we have to allow for the possibility of dipole formation arising from the adsorption of neutral molecular oxygen species (Fig. 1.7). These species are neglected in all mechanistic description of gas sensing on semiconducting metal oxides.

Fig. 1.6

Changes in work function (black), band bending (dark grey) and electron affinity (light grey) due to 300 ppm O₂ (-—-) at 200 °C (left) and at 400 °C (right). Copyright Elsevier, reproduced with permission from Ref. [67]

Fig. 1.7

Adsorption of O₂ on a reduced SnO₂ (110) surface. There is a stable state for a twisted conformation on a fourfold Sn²⁺ site (a) It is expected to accept negative charge under building of ionosorbed species and thus without influence on electron affinity (b) A stable conformation tilted from the normal, where a Lewis acid/base interaction leads to a local dipole with a negative partial charge on the tin and thus to a decrease in χ, is also reported (c) Copyright Elsevier, reproduced with permission from Ref. [67]

A critical look at the available experimental data shows that the concept of oxygen ionosorption is based exclusively on phenomenological measurements. Despite trying for a long time, there has not been any convincing spectroscopic evidence for “ionosorption”. Neither superoxide ion O₂ ⁻, nor charged atomic oxygen O⁻, nor peroxide ions O₂ ²⁻, nor CO⁺ have been observed under real working conditions of sensors (see a recent review [109]).

With regard to the two main forms of charged oxygen species on the surface (superoxide ion O₂ ⁻ and charged atomic oxygen O⁻) and widely used in the mechanistic description of gas sensing properties and modelling of oxide conduction mechanism, it appears that:

• 1. The superoxide ion (O₂ ⁻) has been observed only after low-temperature adsorption <150 °C on reduced SnO₂;

• 2. There has not been any spectroscopic evidence for the formation of charged atomic oxygen (O⁻) on SnO₂.

Moreover, several findings, such as (i) the formation of superoxide ion (O₂ ⁻)only at low adsorption temperatures(<150 °C) on reduced SnO₂, (ii) absence of a high-temperature oxygen desorption (peak at 400–550 °C, attributed to adsorbed oxygen) if the superoxide-ion is present at the surface, (iii) decrease in oxygen intensity with increasing evacuation temperature; herewith the amount of O₂ desorbed is equal to the number of superoxide ion centres, (iv) correlation between TPD, EPR, IR and electrophysical studies on reduced SnO₂, allows us to conclude that the superoxide ion does not undergo transformations into charged atomic oxygen at the surface and represents a dead-end form of low-temperature oxygen adsorption on reduced metal oxide.

As known, the superoxide ion can undergo the following chemical changes on the surface: (i) lose an electron (to the CB) and leave as gaseous O₂ and (ii) gain an additional electron (becoming a peroxide ion O₂ ²⁻), followed by cleaving to form atomic oxygen and the lattice oxygen anion (O²⁻). According to the spectroscopic data (there is no evidence either for peroxide ion or for charged atomic oxygen) the transformation from superoxide ion to atomic oxygen does not happen on SnO₂. This indicates two competing channels for oxygen adsorption—molecular and dissociative. A similar mechanism has been recently postulated for TiO₂ [110] and Ag [111]. On TiO₂, only η²-coordinated dioxygen decomposes to oxygen adatom and a filled oxygen vacancy (in contrast, the η¹-coordinated dioxygen desorbs at 410 K [110]). On Ag, upon heating, physisorbed oxygen transforms into molecular chemisorbed I α-O₂ (“end-form”) which does not dissociate into the atomic form due to the high conversion barrier; only molecular chemisorbed II β-O₂ (“transformable form”) accessible only through a direct interaction from the gas phase (and not accessible from physisorbed form) dissociates into atomic oxygen [111].

The long sought efforts to quench “high-temperature” oxygen species (claimed to represent charged atomic oxygen—O⁻) has not yielded any measureable success. No such paramagnetic species have been observed on high-temperature oxygen treated oxides (TiO₂, SnO₂, ZnO) [95, 112, 113]. Moreover, the EPR evidence of the surface O⁻ species formed due to oxygen adsorption is very contradictory. Furthermore, from a review of the literature there isn’t convincing evidence of their formation on n-type semiconducting oxides due to their direct interaction with dioxygen. Likewise, it is not possible to connect the high-temperature peak in the TPD spectra and the change in the conductivity with the formation of surface O⁻ species. Consequently, the conclusions on O⁻ formation on SnO₂ are not supported by any spectroscopic data. Accordingly, the picture of oxygen adsorption on SnO₂ has to be modified in the following way (Fig. 1.8).

Fig. 1.8

Scheme of oxygen interaction with metal oxides showing the superoxide ion as a dead-end form. Modified from Ref. [145]

5 Oxygen-Vacancy Model (Reduction-Reoxidation Mechanism)

This model focuses on oxygen vacancies at the surface, which are considered to be “the determining factor in the chemiresistive behavior” [114]. Tin dioxide, the most extensively investigated sensing material, is oxygen-deficient and, therefore, an n-type semiconductor, whose oxygen vacancies act as electron donors. Alternate reduction and reoxidation of the surface by gaseous oxygen (Mars—van Krevelen mechanism)control the surface conductivity and therefore the overall sensing behaviour. In this model, the mechanism of CO detection is represented as follows: (i) CO removes oxygen from the surface of the lattice to give CO₂, thereby producing an oxygen vacancy; (ii) the vacancy becomes ionized, thereby introducing electrons into the conduction band and increasing the conductivity; (iii) if oxygen is present, it fills the vacancy; in this process one or more electrons are taken from the conduction band, which results in a decrease in conductivity.

Numerous experimental and theoretical works have evaluated the reduction-reoxidation mechanism (see, for example, references [52, 114–119]); this mechanism still dominates in almost all spectroscopic studies (see, for example, references [120–125], Table 1.2). For example, it was found that oxygen promotes water vapour dissociation on SnO₂ at 330–400 °C [126]: an increase in the concentration of hydroxyl groups (peaks at 3640 cm⁻¹) was observed for low oxygen (2000 ppm) and water vapour (3 ppm) concentrations and evolved towards saturation. This effect was explained by the reaction:

Table 1.2 Some examples of main finding from in situ and operando case studies of gas sensing on semiconducting metal oxides

$$ {\text{H}}_{2} {\text{O}}\;({\text{gas}}) + {\text{O}}^{ - } ({\text{ads}}) + 2{{\text{Sn}}_{\rm Sn}}^{ \times } \leftrightarrow 2({{\text{Sn}}_{\rm Sn}}^{ \times } - {\text{OH}}) + {\text{e}}^{ - } ({\text{CB}}) $$

At first sight this seems to provide evidence for the ionosorption model. However, this effect (i.e. the increase in the concentration of hydroxyl groups during oxygen exposure) can be explained just as well, by completely different processes within the framework of the oxygen vacancy model. For example, an EPR signal of single ionised oxygen vacancies (V ^•_O ) at 1.89 was observed after wet air treatment of SnO₂ at 200 °C [127]. Accordingly, the observed influence of water and oxygen can be described by the two reactions:

$$ \begin{aligned}&{\text{H}}_{2} {\text{O}}\;({\text{gas}}) + 2{{\text{Sn}}_{\rm Sn}}^{ \times } + {{\text{O}}_{\rm O}}^{ \times } \leftrightarrow 2({{\text{Sn}}_{\rm Sn}}^{ \times } - {\text{OH}}) + {{\text{V}}_{\rm O}}^{ \bullet } + {\text{e}}^{ - } ({\text{CB}})\\&2{{\text{V}}_{\rm O}}^{ \bullet } + {\text{O}}_{2} \;({\text{gas}}) + 2{\text{e}}^{ - } ({\text{CB}}) \leftrightarrow 2{{\text{O}}_{\rm O}}^{ \times }\end{aligned} $$

However, the problems associated with oxygen adsorption and detection of reducing gases in an oxygen-free atmosphere are questioning the validity of the reduction-reoxidation model (see detailed discussion below).

As we mentioned above, ionosorbed oxygen has never been observed in operando and in situ studies on metal oxide sensors under working conditions [123, 124]. By contrast, operando and in situ spectroscopy provides very strong evidence of the reaction and ionization of oxygen vacancies under operating conditions of sensors [120–124].

The in situ FT-IR studies [123, 128] of SnO₂ under working conditions (at 375–450 °C) showed an increase of the intensity in the broad band in the region of 2300–800 cm⁻¹ (so-called X-band) with increasing oxygen content. The proximity of the absorption edge with respect to the ionization energy of the second level of oxygen vacancies (1400–1500 cm⁻¹ ~ 170–180 meV) is indicative of the electronic transition from this level to the conduction band (i.e. photoionisation of V ^•_O to V ^••_O )[129]. Accordingly, this band can serve as an indicator of the electron concentration in the neighborhood of oxygen vacancies in the SnO₂. Similar effects were observed on Ga₂O₃, AlVO₄, WO₃ [124]. However, this interpretation was considered to be in contrast to the early electrophysical measurements on SnO₂ [130] which showed that the donor levels in SnO₂ are located at around 30–150 meV below the conduction band, and will be completely ionized at the sensor operating temperatures [131, 132]. The next problem is related to (i) the ionization of oxygen vacancies and consequently, and (ii) to the diffusion processes in the oxide lattice. For SnO₂, for example, it is assumed that the surface defects do not act as electron donors; they have to migrate a small distance into the bulk to become ionized [52]. The diffusion coefficients for this process are low, and, accordingly, the defects are immobilized at the operating temperatures [116]. Nevertheless, diffusion at grain boundaries and at the surface can be much faster than bulk diffusion [12].

A different situation seems to appear in quasi-one-dimensional SnO₂ structures (nanowires) [133, 134], where the oxygen adsorption involves two steps (i) healing of oxygen vacancies by adsorbed oxygen (fast) and (ii) diffusion of as-formed oxygen ions in the bulk healing (annealing) oxygen vacancies (slow) (Fig. 1.9).

Fig. 1.9

One-dimensional model of oxygen diffusion in nanowires. According to the ionosorption model, adsorbed oxygen creates a depletion region close to the surface (dashed line) and then a fast change of R_NW is observed. The new equilibrium between oxygen in the environment, C_G, and the concentration of oxygen at both the nanowire surface, C_S, and its external shell, C₁, creates a gradient with the inside C₂ favoring ion migration into the bulk. This diffusion is associated with long-term drifts of R_NW. Copyright Wiley–VCH, reproduced with permission from Ref. [134]

Surprisingly, even at room temperature an exponent of 1/6 was found in the power law for the oxygen partial pressure dependence of the sensor signal (i.e. conductivity) indicating an intrinsic case in the defect chemistry of SnO₂ (Fig. 1.10) [116]:

Fig. 1.10

Response of one SnO₂ nanowire with r = 20 nm to increasing oxygen partial pressure at room temperature (T = 298 K). Resistance is normalized to the experimental value in synthetic air environment. (Inset) log−log plot of resistance as function of oxygen partial pressure. A linear behavior of slope n = 1/6 is observed. Copyright Wiley–VCH, reproduced with permission from Ref. [134]

$$\begin{aligned} & \frac{1}{2}{{\text{O}}_{2}}+{{\text{V}}_{\rm O}}^{\bullet \bullet }+2{{\text{e}}^{/}}\leftrightarrow {{\text{O}}_{\text{O}}} \\ & \left[ {{\text{e}}^{/}} \right]={{2}^{1/3}}{{\text{K}}_{\text{O}}}^{-1/3}{{\text{p}}_{\text{O}}}^{-1/6},\; {\text{considering}}\; {\text{the\; electroneutrality \;condition}} \\ & \left[ {{\text{e}}^{/}} \right]=2\left[ {{V}_{\text{O}}}^{\bullet \bullet } \right] \\\end{aligned}$$

These findings contrast with bulk film studies which have shown that (i) the surface exchange reaction, i.e. the incorporation of adsorbed oxygen together with the fast electron transfer, is the rate-determining step [118] and (ii) significant oxygen exchange is observed on SnO₂ only at temperatures above 400 C [135].

6 Reduction as a Secondary Process: an Open Issue of Detection of Reduction Gases in Oxygen-Free Conditions

The mechanism of detection of reducing gases in oxygen-free atmospheres requires consideration of the following four experimentally confirmed observations:

• 1. Recovery of sensor resistance to its initial value in an inert gas (N₂, Ar, He) after removing a reducing gas (CO, CH₄, H₂) from the test atmosphere.

• 2. Missing correlation between the degree of oxide reduction and the magnitude of gas sensing response.

• 3. Missing correlation between the gas combustion (oxidation) and the magnitude of gas sensing response.

• 4. Decrease of sensor signals (relative resistance change) with increasing oxygen concentration.

We have to note that a lack of experiments does not allow addressing properly these issues; some important points are discussed below.

Let us take as an example of CO detection in the oxygen-free conditions (alternating CO/N₂ and N₂ flows): What happens when CO is removed from the surrounding atmosphere? From electrical measurements one knows that the sensor resistance (or conductance) recovers its initial value (Fig. 1.11). However, within the framework of the reduction-reoxidation mechanism, gaseous oxygen is required for the reverse process (“vacancy refilling”). Unfortunately, the consideration of this problem has been avoided in spectroscopic studies by alternating CO/N₂ (or Ar) and O₂/N₂ (or Ar) flows, whereas “realistic” conditions require alternating CO/N₂ (or Ar) and N₂ (or Ar) flows.

Fig. 1.11

Transient resistance change of undoped, Pd and Pt doped SnO₂ sensors in oxygen free atmosphere (N₂ balance) [146]. Note, that the sensor resistance recovers its initial value after removing the target gas from the surrounding atmosphere

The oxide reduction as well as reduction/reoxidation of catalytic additives like Pd and Pt seems to be a secondary process that is not connected with the overall sensor response. Figure 1.12 a and b shows the electrical response combined with a change of Sn(II) concentration revealed by in situ Mössbauer spectra for nanocrystalline SnO₂ in the presence of CO and dry air at 380 °C. It was shown that the conductance changes simultaneously with the change of the tin oxidation state (which in turn indicates the formation of oxygen vacancies). A rapid and pronounced increase in Sn(II) spectral contribution was observed just after CO admission into the reactor. The Sn(II) component disappeared 1 min after air admission. It was also noted that (i) a very low Sn(II) content (1 mol %) was sufficient for the conductance to change by 1000 times and (ii) a further increase of Sn(II) concentration up to 14 mol % under exposure to CO did not significantly change the conductance. Similar findings have been also reported for H₂ detection with Pd-promoted SnO₂ sensors. Figure 1.13 [136] show the correlation between the electrical conductance and the oxidation states of Pd and Sn during the cycling of Pd–SnO₂ film in H₂ and O₂ gas mixtures. At 373 K, the conductance changes without any variation of the Pd and Sn oxidation states. At higher temperatures, the oxidation state of Pd varies considerably depending on the atmospheric composition. However, there is no direct correlation between the conductance and the oxidation states of Pd and Sn, i.e. even at 573 K the conductance changes by several orders of magnitude without any measurable variation of the oxidation states of both metals. These results indicate that oxidation and reduction of Pd nanoparticles and SnO₂ matrix are the secondary processes, which are not responsible for the sensitivity to H₂.

Fig. 1.12

a Mössbauer spectra of SnO₂ in a 1 % CO/N₂ atmosphere and then in dry air at 380 °C; b Electrical response of SnO₂ and change of Sn(II) concentration in 1 % CO/N₂ and dry air at 380 °C under 4 l/h gas flow rate. Copyright the Royal Chemical Society, reproduced with permission from Ref. [121]

Fig. 1.13

(On the left) Variation of the electrical conductance a normalized absorption at Pd K-edge (b, left scale) and Pd²⁺ fraction in the Pd²⁺/Pd⁰ mixture (b, right scale) for Pd–SnO₂ film at 573 K (solid lines) and 373 K (dotted lines) during the alternative exposure to 20 % O₂ in He and 1000 ppm H₂ in He. The broken line corresponds to pure SnO₂ at 573 K. (On the right) Operando XAS and conductance studies. The correlation between the conductance of Pd–SnO₂ film and the oxidation states of tin a and palladium b at 573 K (white squares, solid lines) and 373 K (black triangles, dotted lines). Fraction of Pd²⁺ is the concentration of Pd²⁺ in the Pd²⁺/Pd⁰ mixtures; the fraction of Sn²⁺ is the concentration of Sn²⁺ in the Sn²⁺/Sn⁴⁺mixture. The arrows indicate the direction in which the system changes during exposure to H₂ and O₂ Copyright the Royal Chemical Society, reproduced with permission from Ref. [136]

No direct correlation exists between sensor activity (i.e. sensor signal) and consumption of target gases obtained from the on-line gas analysis. As demonstrated in Fig. 1.14 herein at 200 °C the sensor shows relatively high activity in C₃H₈ detection (sensor signal is about 6–400 ppm C₃H₈) the combustion, however, is almost negligible [137]. The same also holds for higher temperatures and other analytes. Several recent works have also demonstrated that there has been no direct correlation between sensor (SnO₂, TiO₂…) response to CO and the CO₂ production (“catalytic activity”) [138–140].

Fig. 1.14

(Left) Sensor signal of SnO₂ sensors exposed to different analytes in dry air dependent on operating temperature of sensors and (right) overall gas combustion measured by on-line PAS Copyright Elsevier, reproduced with permission from Ref. [137]

The strong argument is coming from the observation that in the “oxygen-free” atmosphere the sensor response (i.e. relative change in the conductance) is even higher than in “oxygen” containing atmosphere (Fig. 1.15). To explain these findings, an assumption about the formation of ionosorbed donor-like CO⁺ species is made (see, for example [141] and Fig. 1.16). However, like in the case of ionosorbed oxygen species we face here a problem of common chemical understanding and missing spectroscopic evidence.

Fig. 1.15

(left) Relative change in the conductance (sensor signal, defined as (G_CH4-G₀)/G₀, where G_CH4 and G₀ are the conductance values measured at 450 C under CH₄ supporting gas, respectively.) for undoped SnO₂ as a function of CH₄ concentrations for different O₂ concentrations (N₂ balance). Copyright Elsevier, reproduced with permission from Ref. [85]. (Right) Sensor signal of a 0.2 wt % Pt doped SnO2 thick film sensor at 300 C as a function of CO concentration for different O₂ concentrations (0, 10, 25, 50 ppm, 1 and 5 %). Copyright Elsevier, reproduced with permission from Ref. [141]

Fig. 1.16

Schematic representation CO interaction with SnO₂ as function of the O₂ concentration: in the absence of O₂, CO adsorbs as electron donor and leads to a decrease of the sensor resistance. In the presence of O₂, CO reacts mainly with adsorbed O₂ species. The contribution of each mechanism depends on the O₂ concentration. Copyright Elsevier, reproduced with permission from Ref. [141]

7 Summary and Outlook

Since the development of the first models of gas detection on metal-oxide-based sensors much effort has been made to describe the mechanism responsible for gas sensing. Despite progress in recent years, a number of key issues remain the subject of controversy; for example, the disagreement between electrophysical and spectroscopic investigations, as well as the lack of proven mechanistic description of surface reactions involved in gas sensing. The state-of-the-art description and understanding of gas sensing on metal oxides cannot explain all effects observed on operating metal oxide sensors.

Obviously, our ability for understanding fundamental physicochemical phenomena is limited by dominant physicochemical paradigms. Accordingly, the interpretation of spectroscopic data depends on the model a priori chosen for the mechanistic description.

Besides general applicability for describing the mechanism of gas sensing on semiconducting metal oxides, the ionosorpion model works very well in explaining also quite unobvious results under a priori made unproven assumptions; typical examples are:

• higher response to reducing gases such as CO in absence of oxygen compared with that in the presence of oxygen; here the assumption is made about the formation of ionosorbed donor-like CO⁺ species (see, for example [141]);

• large response for smaller crystals; obviously this interpretation does not involve and does not require the explicit formation of any ions at the surface. In case of the interaction with gaseous oxygen, the key is only the decrease of electron density (depletion) in the surface regions of a metal oxide; this effect is more pronounced in materials with higher surface/volume ratio (see, for example, Ref. [103]);

• reversible p- to n- and to p-type transition on semiconducting metal oxides induced by gas adsorption [142, 143].

The ionosorption theory fails to explain several important issues, among them:

• 1. the missing spectroscopic evidence as well as theoretical confirmation of ionosorbed oxygen species and

• 2. electron affinity changes due to the oxygen adsorption.

Excluding the electron transfer to adsorbed species by assuming the localization of electrons in solid material in close vicinity to adsorbed/gaseous species eliminates immediately the issue of missing spectroscopic evidence of oxygen ions. The polarizability effects can also be involved for explaining the competition between e.g. CO and oxygen: in the presence of CO, adsorbed O₂ molecules will also attract the electrons from the highly polarizable CO molecules thus reducing the O₂ strength to attract the conduction electrons [144]. This assumption can also explain large response for smaller crystals (see above). However, it fails to explain the higher response—meaning the increase in the conductance—to reducing gases such as CO in absence of oxygen compared with that in the presence of oxygen.

Even if numerous experimental and theoretical works have evaluated the reduction-reoxidation mechanism and this mechanism still dominates in almost all spectroscopic studies, the reduction-reoxidation model fails to explain

• 1. the missing correlation between the conductance of sensing materials and the degree of reduction under operating conditions (from operando studies);

• 2. kinetics of oxygen exchange;

• 3. recovery of the sensor resistance to its initial value after exposure to reducing gases in oxygen-free conditions where—according to the reduction-reoxidation mechanism, gaseous oxygen is required for the reverse process (“vacancy refilling”);

• 4. lack of gaseous products of the oxidation of reducing gases (e.g. CO₂) from simultaneous on-line gas analysis.