Multivariable Sensors for Ubiquitous Monitoring of ... - ACS Publications

Sep 7, 2016 - presented in section 12. Figure 1. Examples of diverse application scenarios of gas sensors in the Internet of Things and Industrial Int...
8 downloads 15 Views 27MB Size
Review pubs.acs.org/CR

Multivariable Sensors for Ubiquitous Monitoring of Gases in the Era of Internet of Things and Industrial Internet Radislav A. Potyrailo* GE Global Research, Niskayuna, New York 12309, United States ABSTRACT: Modern gas monitoring scenarios for medical diagnostics, environmental surveillance, industrial safety, and other applications demand new sensing capabilities. This Review provides analysis of development of new generation of gas sensors based on the multivariable response principles. Design criteria of these individual sensors involve a sensing material with multiresponse mechanisms to different gases and a multivariable transducer with independent outputs to recognize these different gas responses. These new sensors quantify individual components in mixtures, reject interferences, and offer more stable response over sensor arrays. Such performance is attractive when selectivity advantages of classic gas chromatography, ion mobility, and mass spectrometry instruments are canceled by requirements for no consumables, low power, low cost, and unobtrusive form factors for Internet of Things, Industrial Internet, and other applications. This Review is concluded with a perspective for future needs in fundamental and applied aspects of gas sensing and with the 2025 roadmap for ubiquitous gas monitoring.

CONTENTS 1. Introduction 2. Diversity of Applications and Requirements for Modern Gas Sensors 3. State of the Art of Existing Sensing Concepts 4. General Principles of Multivariable Sensors 5. Multivariable Nonresonant and Resonant Impedance Sensors Dielectric Polymers Conjugated Polymers Macrocycles Metal Oxides Carbon Allotropes Ligand-Capped Metal Nanoparticles 6. Electromechanical Multivariable Resonant Sensors 7. Multivariable Field-Effect Transistor Sensors 8. Multivariable Photonic Resonant Sensors Material-Based Multivariable Photonic Sensors Structure-Based Multivariable Photonic Sensors 9. Other Multivariable Sensor Technologies 10. Design Criteria for Multivariable Sensors Design Criteria for Multivariable Nonresonant Impedance Sensors Design Criteria for Multivariable Resonant Impedance Sensors Design Criteria for Multivariable Electromechanical Resonant Sensors Design Criteria for Multivariable Field-Effect Transistor Sensors

Design Criteria for Multivariable Photonic Sensors Based on Functionalized Plasmonic Nanoparticles Design Criteria for Multivariable Photonic Composite Colloidal Crystal Film Sensors Design Criteria for Multivariable Photonic Interference-Stack Sensors 11. Benefits of Multivariable Sensors 12. Summary and Development Trends of SingleOutput and Multivariable Sensors Looking Back Looking at the Present Looking into the Future Author Information Corresponding Author Notes Biography Acknowledgments References

A C D F I J K M M N N Q R U U W Z AA

AA AB AB AB AE AE AE AG AI AI AI AI AI AI

1. INTRODUCTION Monitoring of gas-phase chemicals such as gases and vapors using portable instruments has been traditionally important in numerous applications including industrial and home safety, environmental surveillance, process monitoring, homeland security, and others.1−3 To meet diverse detection needs, developed complementary technologies include direct spectros-

AA AA AA AA

Received: March 19, 2016

© XXXX American Chemical Society

A

DOI: 10.1021/acs.chemrev.6b00187 Chem. Rev. XXXX, XXX, XXX−XXX

Chemical Reviews

Review

Figure 1. Examples of diverse application scenarios of gas sensors in the Internet of Things and Industrial Internet applications. Image prepared by GE Global Research.

These new performance characteristics will be attractive in established and emerging sensing scenarios. This Review has four broad goals in order to stimulate research in this rapidly expanding multidisciplinary area. The f irst goal is to deliver critical analysis of recent developments of sensors based on the multivariable response principles as provided by the number of independent outputs generated by the sensor (also known as sensor dispersion). The second goal is to summarize the design criteria for a new generation of individual sensors based on diverse transducers (e.g., electrical, photonic, electromechanical, and other transducers) developed to operate in a multivariable mode when coupled with inorganic, organic, polymeric, and composite sensing materials. The third goal is to compare performance of individual multivariable sensors and conventional sensor arrays and to illustrate that modern multivariable sensors are a disruptive sensor technology. The fourth goal is to present a 2025 roadmap for multivariable gas sensors with the analysis of the significant driving forces in sensor developments. The Review is structured in 11 sections. Diverse applications and requirements for modern gas sensors are summarized in section 2. Existing sensing concepts are analyzed in section 3, followed by the discussion of principles of multivariable sensors in section 4. A critical analysis of diverse types of multivariable sensors and associated sensing materials is provided in sections 5−9. This analysis is accomplished by the type of multivariable transducer that governs design choices for particular sensing materials. Discussed multivariable sensors include nonresonant and resonant impedance sensors (section 5), electromechanical resonant sensors (section 6), field-effect transistors (section 7), photonic resonant sensors (section 8), and other multivariable sensor technologies (section 9). The design criteria for diverse types of multivariable sensors are summarized in section 10. Benefits of multivariable sensors are further summarized in comparison with conventional sensor arrays in section 11. The outlook and the roadmap for multivariable gas sensors are presented in section 12.

copy, gas chromatography, mass spectrometry, ion mobility spectrometry, and chemical sensors. Sensors for monitoring of “gases” (such as any gas-phase chemicals) where a sensing material is applied onto a suitable physical transducer4,5 have a suite of attractive operational advantages over other portable instruments. These advantages include tunable sensitivity, continuous real-time determination of the concentrations of specific sample constituents, small power consumption, operation without consumables, and unobtrusive form factors. Unfortunately, existing sensors also have several performance limitations such as high crosssensitivity and poor selectivity to various gases,6 inability to preserve detection accuracy in the presence of unknown interferences, and sensor drift, especially noticeable in outdoor applications and in detection of low analyte levels. As a result, these limitations often can revoke advantages of sensors in their intended practical applications. Thus, field uses of gas sensors are often most successful when their poor selectivity is not important, concentrations of measured gases are high enough to make the sensor drift unnoticed, or frequent recalibration is acceptable.7,8 Given the tremendous interest of the chemical research community in improving selectivity of sensor systems (50 000+ publications on “selectivity of gas or vapor sensors”), this Review provides a critical analysis of the recent progress in the development of a new generation of gas sensors based on the multivariable response principles to overcome the insufficient gas-selectivity limitation of existing sensors. The design criteria of these individual multivariable sensors involve a sensing material with multiresponse mechanisms to different gases and a multivariable transducer with several partially or fully independent outputs to recognize these different gas responses. Performance capabilities of inorganic, organic, polymeric, biological, composite, and formulated sensing materials are discussed that have been explored for multivariable gas sensing. These sensing materials, when coupled with electrical, optical, and electromechanical transducers designed for operation in a multivariable mode, provide performance capabilities previously unavailable from conventional sensor systems. These new individual sensors quantify individual components in gas mixtures, reject interferences, and have a self-correction ability against environmental instabilities. B

DOI: 10.1021/acs.chemrev.6b00187 Chem. Rev. XXXX, XXX, XXX−XXX

Chemical Reviews

Review

Figure 2. Anatomy of conventional gas sensors. (A) Required proper combination of a sensing material and a physical transducer to achieve a desired sensor response. (B−G) Examples of sensing materials: (B) semiconducting metal oxide, (C) metal−organic framework,73 (D) single-walled carbon nanotubes, (E) graphene, (F) gold nanoparticles functionalized with organic ligand,74 and (G) atomic-layered molybdenum disulfide.75 (H−S) Examples of physical transducers: electrical nonresonant transducers (H) resistor with interdigital electrodes, (I) capacitor, and (J) field effect transistor;100 electrical resonant transducers (K) resistor−inductor−capacitor (RLC) resonant transducer with an integrated circuit memory chip as a passive radio frequency identification tag and sensor operating at ∼13 MHz, (L) coil-based transducer operating at ∼100 MHz, (M) dual-split-ringbased transducer operating at ∼5 GHz; electromechanical resonant transducers (N) thickness shear mode device, (O) acoustic wave device,101 (P) microcantilever;102 optical transducers (Q) reflected light opto-pair,103 (R) distributed fiber-optic transducer, (S) localized plasmon resonance transducer.102 (C) Reprinted with permission from ref 73. Copyright 2014 Nature Publishing Group. (F) Reprinted with permission from ref 74. Copyright 2013 Wiley-VCH Verlag GmbH & Co KGaA. (G) Reprinted with permission from ref 75. Copyright 2015 American Chemical Society. (J) Reprinted with permission from ref 100. Copyright 2012 American Chemical Society. (L) Logo pictured courtesy of General Electric.

devices.12,13 The Industrial Internet is the integration of complex machinery with networked sensors.14 Exemplary existing and emerging applications of sensors are summarized in Figure 1 and include environmental monitoring and protection, industrial safety and manufacturing process control, monitoring of agricultural emissions, public safety, medical systems, wearable health and fitness, automation of residential homes and industrial buildings, transportation, and retail.15−23 Examples of classes and types of measured gases and volatiles of interest for these applications include environmental background (e.g., O2, CO2, and H2O), transportation/ industrial/agricultural atmospheric pollutants (e.g., CO2, CO, O3, H2S, NH3, NOx, SO2, CH4, industrial fumes, and waste odors), breath biomarkers (e.g., NO, H2S, NH4, acetone, ethane, pentane, isoprene, and hydrogen peroxide), and public/ homeland safety hazardous volatiles (e.g., toxic industrial chemicals, chemical warfare agents, and explosives).18,24−28 Diverse types of volatiles need to be monitored over their broad range of concentrations ranging from part-per-trillion to percent, often mixed with chemical interferences such as ubiquitous variable background (indoor and outdoor urban

2. DIVERSITY OF APPLICATIONS AND REQUIREMENTS FOR MODERN GAS SENSORS At present, “classic” analytical instruments based on gas chromatography (GC), mass spectrometry (MS), ion mobility spectrometry (IMS), and direct spectroscopy are preferred for high-selectivity detection, despite their relatively large power demands, cost, and size.9,10 These instruments could be inconvenient, even in portable configurations with the reduced carrier gas, vacuum, and power demands,11 but are an unavoidable alternative over existing sensors. Meanwhile, there are numerous scenarios when highselectivity advantages of “classic” analytical instruments even in their microfabricated implementations would be canceled by specific application requirements (e.g., unobtrusive form factor, no external power for operation, no vacuum or carrier gases, no radioactive sources). The most prominent of these scenarios are Internet of Things and Industrial Internet applications. The Internet of Things is the network of everyday objects with embedded sensors and connectivity to increase the value of these objects by exchanging data with the users and other C

DOI: 10.1021/acs.chemrev.6b00187 Chem. Rev. XXXX, XXX, XXX−XXX

Chemical Reviews

Review

Table 1. Gas-Response Mechanisms of Representative Sensing Materials sensing material dielectric polymers conjugated polymers metalloporphyrins, phthalocyanines, related macrocycles cavitands zeolites metal−organic frameworks metal oxides monolayer-protected metal nanoparticles carbon nanotubes graphene molybdenum disulfide plasmonic nanoparticles with soft organic layers plasmonic nanoparticle/metal oxide nanocomposite films colloidal crystals from core/shell nanospheres iridescent scales of tropical Morpho butterflies bioinspired photonic interferencestack nanostructures

gas-response mechanisms

ref

dispersion, polarizability, dipolarity, basicity, acidity, and hydrogen-bonding interactions changes in density and charge carrier mobility, swelling, and conformation transitions of chains hydrogen bonding, polarization, polarity, metal center coordination interactions, π-stacking, and molecular arrangements

76 77, 78 79, 80

intracavity host−guest complexation with hydrogen bonding, CH-π, and dipole−dipole as the main specific interactions molecular discrimination by size, shape, molecular kinetic diameter van der Waals interactions of the framework surface, coordination to the central metal ion, hydrogen bonding of the framework surface, size exclusion physisorption, chemisorption, surface defects, and bulk defects depending on operation temperatures (ambient to ca. 1000 °C) and utilizing different metal oxides and dopants electron tunneling between metal cores, charge hopping along the atoms of ligand shell

81

charge transfer from analytes and polarization of surface adsorbates, gas-induced Schottky barrier modulation charge transfer induced by adsorption/desorption of gaseous molecules acting as electron donors or acceptors, leading to changes in conductance charge-transfer mechanism involving transient doping of sensing layer changes in interparticle spacing, refractive index of the organic layer, and reflectivity of the metal nanoparticle network film charge exchange with the nanoparticles, change in the dielectric constant surrounding the nanoparticles, dependent on the type of a metal oxide and its morphology for operation at 300−800 °C vapor-induced changes of optical lattice constant of colloidal crystal with cores and shells of nanospheres responding to diverse vapors lamella/ridge nanostructures with gradient of surface chemistry induce spatial control of sorption and adsorption of analytes and probed with light interference and diffraction chemically functionalized nanostructures with weak optical loss induce spatial control of sorption and adsorption of analytes probed with light interference and diffraction

90, 91 92

82 83, 84 85−87 88, 89

93, 94 74 95 96 97, 98 99

available catalytic and MOS sensors operate with 0.1−1 W power requirements.42−44 that allow their long-term applications in stationary and short-term applications in batterypowered systems. Recent reduction of needed power was achieved by reducing the duty cycle of operation45−47 and applying self-heating principles of sensor operation.48−51 Reduced power, size, and cost provided opportunities for MOS sensors to be integrated into smartphones for monitoring of air pollution.52 Reducing the operating temperature of MOS sensors was also demonstrated under controlled lab conditions.53−57 The reduced-temperature operation of such MOS sensors changes the gas-detection mechanisms58 and requires further significant theoretical and practical validation. Cost reduction is targeted in many recent reports that include waferlevel fabrication,28,50,59 printing,60−62 roll-to-roll,63,64 selfassembly,65,66 and other techniques. This Review is focused on the development of multivariable sensors to achieve an improved reliability of their performance by their enhanced selectivity. Several topics are outside the scope of this Review. In particular, design-for-manufacturability and cost reduction of sensors are the topics of a recent review67 addressing the demands for Internet of Things and Industrial Internet applications in volumes of billions and even trillions of sensors.30,32 Aspects for low-power operation and energyharvesting approaches for sensors have been also recently reviewed.68−71 Diverse architectures of wireless sensor networks have been recently reviewed as well.72

air, industrial air, human odors and breath, exhaust of transportation engines, etc.) and at expected operation temperatures (ambient indoor and outdoor temperatures, body temperature, and exhaust of transportation engines). At present, physical sensors dominate in Internet of Things applicationsmicrophones, accelerometers, gyroscopes, and compasses are being shipped at ∼1 billion units each annually.29 However, the market for physical, chemical, and biological sensors is expected to grow to a cumulative trillion units by 2022.25,30−32 The top five requirements for modern sensors for Internet of Things and Industrial Internet applications include (1) reliability, to provide accurate readings in diverse environmental conditions; (2) low power, to extend battery life or to eliminate its need thus simplifying detection logistics; (3) low cost, to accommodate the need for their large deployed numbers; (4) appropriate real-time communication capability; and (5) data security.33,34 Applying a combination of these requirements to new sensors significantly enhances their value. In particular, having new sensors at low cost but with unreliable performance significantly limits their value. In contrast, a new generation of sensors is desired to be not only low cost but also as reliable as their more expensive traditional analytical devices and at a fraction of the power needed to operate.35−39 Aligned with these requirements,33,34 developments of new sensors have been already focusing on reliability, low power, and low cost. Diverse sensing materials have been improved by understanding the key degradation mechanisms and reducing their effects. Recent reports include improvement of stability and poison resistance of metal oxide semiconducting (MOS) sensors40 and improvement stability of ligand-capped metal nanoparticles by new ligand-attachment chemistry.41 Significant power reduction has been demonstrated for sensors that operate at elevated temperatures of hundreds of degrees. Established

3. STATE OF THE ART OF EXISTING SENSING CONCEPTS In gas sensing, a proper combination of a sensing material and a physical transducer is needed to achieve a desired sensor response (Figure 2A). Some of the most widely studied types of sensing materials are illustrated in Figure 2B−G73−75 as D

DOI: 10.1021/acs.chemrev.6b00187 Chem. Rev. XXXX, XXX, XXX−XXX

Chemical Reviews

Review

Figure 3. Typical gas cross-sensitivity patterns of new types of reversible sensing materials. (A) Chemiresistor with a p-type semiconductor NiO preferential response to formaldehyde over other volatiles.116 (B) Polymeric sensing film formulated with two types of fluorescent phosphonate cavitandsfluorescence response to vapors of different alcohols.117 (C) Thickness shear mode resonators with two types of immobilized DNA response to model analytes.118 (A) Reprinted with permission from ref 116. Copyright 2015 Elsevier. (B) Reprinted with permission from ref 117. Copyright 2011 Wiley-VCH Verlag GmbH & Co KGaA. (C) Reprinted with permission from ref 118. Copyright 2015 Elsevier.

Figure 4. Typical gas cross-sensitivity patterns of established types of reversible sensing materials. (A) Electrochemical sensor calibrated for ethylene oxide.119 (B) Metal oxide semiconductor sensor calibrated for methane.40 (C) Catalytic combustion sensor calibrated for methane.120

Figure 5. Typical gas cross-sensitivity patterns of gas dosimeter materials based on irreversible or slow-recovery chemical reactions. (A) Response of a reduced graphene oxide-decorated cotton yarn to NO2 (analyte) and other volatiles.121 (B) Response of an atomic-layered MoS2 to NO2 (analyte) and other volatiles.75 (C) Response of colorimetric formulated composition to formaldehyde (analyte) and other volatiles.122 Insets in (A−C) are output signals of dosimeters (1) before, (2) during, and (3) after exposures to analytes. (A) Reprinted with permission from ref 121. Copyright 2015 Nature Publishing Group. (B) Reprinted with permission from ref 75. Copyright 2015 American Chemical Society. (C) Reprinted with permission from ref 122. Copyright 2015 Institute of Electrical and Electronics Engineers.

electrical nonresonant and resonant transducers, electromechanical resonators, and optical transducers. Recent important advances in gas sensors include outstanding sensitivity in vacuum and clean carrier gas90,92,104−108 and rapid response times.105,109−114 These improvements in sensitivity and response times of sensing materials were demonstrated with the reduction of size of sensing features down to zerodimensional nanoparticles, one-dimensional nanowires, two-

examples for detection of reducing or oxidizing gases, volatile organic compounds, and combustible and toxic gases. A summary of diverse vapor-response mechanisms of several classes of sensing materials such as inorganic, organic, polymeric, biological, and composites is presented in Table 1. To probe a sensing material response upon gas exposures, numerous types of transducers were developed. Figure 2H− S100−103 illustrates examples of such transducers that include E

DOI: 10.1021/acs.chemrev.6b00187 Chem. Rev. XXXX, XXX, XXX−XXX

Chemical Reviews

Review

Figure 6. Evolution of gas-sensing concepts. (A) “Classic” single-output gas sensor with known insufficient selectivity. (B) Assembly of individual single-output sensors into an array and chemometric processing of the array output. (C) Individual gas sensor based on multivariable response principles.

conventional sensors as shown in excellent studies with sensor arrays containing up to 65 536 elements.131,135−140 The field of sensor arrays (also known as electronic noses) has matured to an understanding of their applicability and limitations outside controlled laboratory conditions (e.g., an uncorrelated drift of each sensor in an array, inability to provide accurate quantitation of multiple vapors in their mixtures, and inability to operate in the presence of high levels of known and unknown interferences). The state of the art in sensor arrays and their prospects has been critically analyzed in “classic” and recent reviews.141−166 Sensor arrays can be also complimented with hyphenated methodologies where sensing materials are interrogated with different transducers to probe diverse properties of the material. Several “classic” and recent examples include sensor arrays based on several transduction principles,167−170 probing organic semiconducting materials with thickness-shear mode and field-effect transducers,171 probing carbon nanotubes with acoustic and optical readouts,172 probing nanopore-sorbed volatiles with opto-calorimetric readout,173 and probing adsorbed volatiles with piezotransistive and photoacoustic readouts.174 On the basis of the developments of single-output sensors, their arrays, and hyphenated readouts, a new generation of gas sensors is emerging that utilizes multivariable response principles (Figure 6C). Multivariable sensors and microanalytical systems producing multivariable response were reviewed in “classic”175,176 and recent reviews.177−179 Critical analysis of the recent developments in multivariable sensors based on diverse transduction principles and their critical comparison is the focus of the next sections of this Review.

dimensional sheets, and three-dimensional nanostructures. Such size reduction allowed not only the higher surface area for the analyte to interact with the material but also implementation of new physical phenomena on the nanometer scale.48,109,115 The major differences in gas sensitivity between diverse materials are their mechanisms of interaction with different classes of gases such as reducing or oxidizing gases, volatile organic compounds, and combustible and toxic gases at different temperatures ranging from ambient to ∼1000 °C as presented in Table 1. Unfortunately, existing sensors have poor gas selectivity and insufficient stability. These features affect sensor reliability, which is one of the critical aspects for the broad acceptance of sensors.33,34 Often, new sensing materials respond not only to an intended analyte vapor but also to other vapors, for example, as shown in Figure 3A−C116−118 exhibiting significant vapor cross-sensitivity. The origin of this limitation is in the conflicting requirements for sensor selectivity versus reversibility.4 The full and fast reversibility of sensor response is achieved via weak interactions between the analyte and the sensing film, whereas the high selectivity of sensor response is achieved via strong interactions between the analyte vapor and the sensing film. Insufficient vapor selectivity is also known for existing sensors. For qualitative comparison with new sensing materials, typical gas cross-sensitivity patterns of the most widely implemented types of commercially available materials-based sensors such as electrochemical, MOS, and catalytic combustion sensors40,119,120 are visualized in Figure 4A−C.40,119,120 This information is typically available in the sensor product specification and is utilized to estimate the expected levels of false alarms in anticipated applications. Such comparison illustrates that neither new nor established single-output sensors have a desired minimal gas cross-sensitivity. This problem of poor selectivity can be reduced by implementing sensing materials that utilize strong irreversible or slow-recovery chemical reactions. Recent examples of such developments are illustrated in Figure 5A−C.75,121,122 This approach allows operation of a single sensing element for several dosimetric measurements followed by element replacement or element resetting using an external UV, thermal, or other type of energy. An evolution of gas-sensing concepts that had led to multivariable sensors is depicted in Figure 6. “Classic” gas sensors based on a single output (i.e., zero-order sensors123) are schematically depicted in Figure 6A. Such sensors have been recently extensively reviewed.124−130 To improve selectivity, individual sensors are assembled into arrays where the output of the array is processed with multivariate analysis tools (Figure 6B). Examples of the most widely used tools for multivariate analysis of sensing data are summarized in Table 2. Since the 1980s, combining sensors into arrays131−134 is a common compromise to mitigate poor selectivity of individual

4. GENERAL PRINCIPLES OF MULTIVARIABLE SENSORS To overcome the insufficient selectivity limitation of existing sensors and sensor arrays and to improve their reliability, a new generation of gas sensors is emerging based on multivariable response principles. Multivariable sensors (also known as intelligent,180 multiparameter,181 high-order,177 or multidimensional signatures182 sensors, virtual multisensor systems,183 or virtual sensor arrays184,185) provide several partially or fully independent responses from a sensor.63,99,179,186 General design criteria for multivariable sensors involve roles of (1) a sensing material with diverse responses to different gases, (2) a multivariable transducer to provide independent outputs and to recognize these different gas responses, and (3) data analytics to provide multianalyte quantitation, rejection of interferences, and drift minimization. The common term “design” reflects a quantitative outcome of creating new materials or devices for particular applications as predicted by existing knowledge. Design of transducers and data analytics tools are two such examples. The complex nature of interactions between the F

DOI: 10.1021/acs.chemrev.6b00187 Chem. Rev. XXXX, XXX, XXX−XXX

composition, preparation, and end-use conditions of sensing materials often makes difficult their “design”.187 Instead, “material tuning” is often performed to meet specific needs.5,187 Thus, at present, “design” of sensing materials should be viewed as an ultimate future goal rather than the currently available fully developed tool. While new computational tools are decoding the intricate interplay of mechanisms ranging from atomic to macroscopic scales with an increasing accuracy, these tools are not replacing yet the detailed experimental tuning of materials.188−191 Selectivity and stability of sensing materials are important examples of remaining computational challenges that require experimental validations. For new multivariable sensors, a subset of sensing materials, already applied to single-output sensors (Table 1), is also of interest because some of these materials have diverse vaporresponse mechanisms that can be probed with a multivariable transducer. Further, tools for multivariate analysis of sensing data from individual multivariable sensors can be adapted from those used for sensor arrays (Table 2). Multivariable sensors provide performance capabilities previously unavailable from not only conventional single-output sensors but also from sensor arrays. These new individual sensors quantify individual components in gas mixtures, reject interferences, and have selfcorrection ability against environmental instabilities.74,99,180,192−194 These capabilities are provided by the number of independent outputs (also known as dispersion, dimensionality, or order) generated by the sensor. A single-output sensor affords a single correlation between a gas concentration and a sensor output and provides a one-dimensional (1-D) response or 1-D dispersion. In such a sensor, gases with known cross-sensitivity produce different response magnitudes but without their discrimination (Figure 7A). Such sensors are valuable to measure known contaminants in the absence of interferences. Sensors with more than one independent output are critical for emerging applications. As the simplest case, Figure 7B depicts the response of a multivariable sensor with two outputs where different gases have their own unique response directions. These independent sensor outputs can be either raw sensor responses or can be weighted contributions of several partially independent outputs. Even this simplest 2-D dispersion allows discrimination between closely related analytes or correction for some environmental interferences. The value of the multivariable sensor increases with its ability to discriminate and quantify gases in the presence of known and unknown interferences and to correct for multiple environmental effects. This increased value is provided by the increased number of independent outputs leading to multidimensional dispersion (Figure 7C). Such value of individual multivariable sensors becomes higher than that of a sensor array not only because of similar180,195,196 or better99 short-term performance over sensor arrays but also because of an improved capability for long-term stability. In this Review, dispersion of the reported multivariable sensors was determined based on results of multivariate analysis of their individual outputs. The criterion for reporting dispersion of a multivariable sensor was its highest reported dispersion at which the sensor demonstrated the consistent diverse vapor-dependent responses. The most widely implemented multivariate analysis technique in multivariable sensors is PCA followed by DA, ANN, and other techniques summarized in Table 2. To illustrate the PCA approach, performance of a multivariable sensor has been simulated and

Determines correlations between the independent variables and the sensor response by finding the direction in the multidimensional space of the sensor response that explains the maximum variance for the independent variables. The key outputs of the developed multivariate models are residual errors of calibration and cross-validation. Regression analysis technique based on PCA by regressing the dependent variables on a set of independent variables based on a standard linear regression model, but uses PCA for estimating the unknown regression coefficients in the model.

Models the difference between the classes of data and maximizes the ratio of between-class variance to the within-class variance. Requires an input of distinction between independent variables and dependent variables.

description

Unsupervised algorithm that reduces a multidimensional data set for its easier interpretation by calculating orthogonal principal components (PCs) oriented in the direction of the maximum variance within the data set. The first PC contains the highest degree of variance, and other PCs follow in the order of decreasing variance. Thus, PCA concentrates the most significant characteristics (variance) of the data into a lower dimensional space.

algorithm

A system of a large number of simple highly interconnected processing elements (“neurons”) that exchange messages between each other to process information by their dynamic state response to external inputs. The connections have numeric weights that can be tuned based on experience, making neural nets adaptive to inputs and capable of learning. Classifies samples using a dendrogram representation. Often, a Ward’s method is applied that shows the Euclidean distance between the samples. The Ward’s method is a minimum variance method, which takes into consideration the minimum amount of variance between the samples and gases (analyte and interferents) to define a cluster. Supervised learning models with associated learning algorithms that analyze data and recognize patterns, used for classification, regression analysis, and outliers detection by finding the decision hyperplane that maximizes the margin between the classes. The vectors (cases) that define the decision hyperplane are the support vectors. Separates a multivariate signal into additive subcomponents by assuming that the subcomponents are statistically mutually independent non-Gaussian signals. A powerful technique for revealing hidden factors that underlie sets of random variables, measurements, or signals.

Review

principal component analysis (PCA) discriminant analysis (DA) artificial neural network (ANN) hierarchical cluster analysis (HCA) support vector machines (SVM) independent component analysis (ICA) partial least-squares (PLS) regression principal component regression (PCR)

Table 2. Examples of Typical Chemometrics Tools Applied for Data Analysis of Sensor Arrays and Multivariable Sensors

Chemical Reviews

G

DOI: 10.1021/acs.chemrev.6b00187 Chem. Rev. XXXX, XXX, XXX−XXX

Chemical Reviews

Review

Figure 7. Importance of sensor response dispersion for reliable performance. (A) 1-D dispersion of a single-output sensor, sensor affords a single correlation between a gas concentration and sensor output. Gases with known cross-sensitivity have different response magnitudes without their discrimination. (B) 2-D dispersion of a multivariable sensor with two independent outputs; different gases have their own unique response directions, affording the possibility for correction for some environmental interferences. (C) 3-D dispersion of a multivariable sensor to monitor multiple gases in the presence of known and unknown interferences and closely related gases of different classes and to correct for multiple environmental effects.

Figure 8. Illustration of typical PCA results from a computer-simulated multivariable sensor. (A) “Spectrum” of a sensor with indicated changes in peak height, peak position, and peak width upon application of simulated environmental effects. (B) Three levels of spectral changes in peak height, peak position, and peak width. (C, D) Visualization of relations between measured spectra by presenting PCA scores plots of PC1 vs PC2 and PC1 vs PC2 vs PC3, respectively. (E) Plot of the signal-to-noise (S/N) of each PC in the model. (F) Loadings plot that depicts contributions of each variable to different PCs.

processed using PCA as shown in Figure 8. This simulated sensor had a “spectrum” across 150 variables (a.k.a. dimensions such as frequencies or wavelengths) with an added random noise (Figure 8A). Three environmental effects (e.g., analyte vapor, humidity, and temperature) were assumed to change the peak height, peak position, and peak width of the spectrum (Figure 8A). Three levels of each of the effects were simulated (shown in Figure 8B), with three replicates per each level (not shown for simplicity). From the raw responses with a total of 30 samples of spectra (initial position, three effects at three levels, n

= 3 for each), a PCA classification model was built. Several outputs from a PCA model are important for understanding and visualization of spectral changes in the sensor. One output is a scores plot that visualizes relations between measured spectra by presenting principal components (PCs) against each other or as a function of experimental time. A PC1 vs PC2 plot in Figure 8C illustrates that the contributions from the first two PCs (52.14% and 42.54%, respectively) did not cover 100% of the variance produced by the sensor. Thus, Figure 8D depicts a 3-D plot of PC1 vs PC2 vs PC3 illustrating that PC3 also correlated well with H

DOI: 10.1021/acs.chemrev.6b00187 Chem. Rev. XXXX, XXX, XXX−XXX

Chemical Reviews

Review

sensing material that can be correlated to a single or multiple analytes of interest and can provide correction of environmental effects.194,205 The real part of the complex permittivity of the sensing material is its “dielectric constant”. The imaginary part of the complex permittivity of the sensing material is its “dielectric loss factor” and is directly proportional to the conductivity of the sensing material. In impedance sensors, measurements can be done at a single206 or at multiple frequencies.207−210 While measurements at a single frequency provide a useful sensor, analysis at multiple frequencies could achieve an enhanced selectivity. The change in R and C of portions of the sensing structure besides the bulk of the sensing material can also contribute to sensor response. These contributions can originate from several independent regions of the film/transducer system such as contact resistance and capacitance of the film/electrode interface, surface resistance and capacitance, and substrate/ film interface resistance and capacitance211 as visualized in Figure 10. As an example, parts A and B of Figure 11 show

the three types of environmental effects. Another output is a signal-to-noise (S/N) plot of each PC in the model that determines which PCs have low S/N. Figure 8E depicts that the built PCA model had three PCs that had S/N ≫ 3. The third output is the loadings plot that reveals contributions of each variable to different PCs. Figure 8F depicts that loadings of the first three PCs had different spectral shapes.

5. MULTIVARIABLE NONRESONANT AND RESONANT IMPEDANCE SENSORS Multivariable impedance sensors typically measure vapormodulated changes of resistance R and capacitance C of a sensing structure in nonresonant and resonant configurations. Measurements of R and C in a nonresonant configuration can be performed by using an electrode RC circuit structure (see Figure 9A) that incorporates a sensing film.197−199 Adding an inductor

Figure 9. Design and operation of multivariable impedance vapor sensors. Simplified equivalent circuits of (A) nonresonant and (B) resonant sensor configurations to probe vapor-modulated changes of resistance Rs and capacitance Cs of a sensing material. (C) Resonance impedance spectrum (real part Zre and imaginary part Zim of resonance impedance) and representative parameters for multivariate analysis: frequency position Fp and magnitude Zp of the real part of the resonance impedance spectrum, the resonant F1 and antiresonant F2 frequencies of the imaginary part of the resonance impedance spectrum, and their impedance magnitudes Z1 and Z2, respectively.

Figure 10. Details of the sensing region where a sensing material is applied onto the sensor electrodes. Effects of the sensing film are pronounced in resistance and capacitance related to the film width between electrodes, film thickness, contact resistance and capacitance, surface resistance and capacitance, and substrate/film interface resistance and capacitance.

L to such an electrical circuit provides a resistor−inductor− capacitor (RLC) resonator (see Figure 9B), where changes in R and C can be measured in a resonant circuit configuration.200 Sensor-excitation conditions also can increase dispersion of sensor response.201,202 The advantage of resonators is sometimes a higher sensitivity over nonresonant structures.203 Resonant response can be the resonance impedance spectrum (Figure 9C). Measurements can be done of the real part Zre and imaginary part Zim of resonance impedance, and representative parameters for multivariate analysis can be determined such as frequency Fp and magnitude Zp of the real part of the resonance impedance spectrum, the resonant F1 and antiresonant F2 frequencies of the imaginary part of the resonance impedance spectrum, and their impedance magnitudes Z1 and Z2, respectively.200,204 To provide useful impedance response that will have diversity for different gases, the sensors should have the ability to change independently measured real and imaginary impedance values. These independent changes can originate from several types of contributions of the sensing structure. One type of contribution is the change in R and C from the bulk of the sensing material itself that is related to changes in the complex permittivity of the

transmission electron microscope images of film/transducer interface, respectively, without and with a contact layer engineered to enhance sensor selectivity via an additional vapor-responsive RC circuit structure. Impedance gas sensing was first demonstrated decades ago,195,212−215 including impedance measurements in gas mixtures,180,216 and has been the foundation for the modern RC and RLC sensor developments. These sensors can be probed using analyzers that can measure multiple output parameters from a single sensor.217 Besides operating in the radio frequency (MHz) frequency range with RLC resonant sensors,200 such sensors also have been demonstrated in the microwave (GHz)218−220 frequency ranges. Numerous portable and single-chip analyzers have been developed to be applicable for these measurements.221−233 To date, multivariable impedance sensors are the most broadly explored multivariable sensors. They have been demonstrated with a wide range of sensing materials that exhibit independent or partially independent changes in their complex permittivity for discrimination between multiple gases, rejection of interferences, and correction for environmental effects. In this section, several diverse types of materials are I

DOI: 10.1021/acs.chemrev.6b00187 Chem. Rev. XXXX, XXX, XXX−XXX

Chemical Reviews

Review

Figure 11. Transmission electron microscope images of film/transducer interface without (A) and with (B) an engineered contact layer for the enhancement of sensor selectivity.

analyzed to illustrate the power of multivariable sensing using electrical nonresonant and resonant impedance sensors. The most studied sensing materials include dielectric198,201,202,205,215,220,234 and conjugated192,195,204,235−237 polymers, macrocycles,215,238 metal oxides,182,206,239−241 carbon allotropes, 2 42, 24 3 and ligand-capped metal nanoparticles.178,192,201,207,244 Less-explored materials are transition metal dichalcogenide monolayers MX2, with M being a transition metal atom (e.g., Mo, W) and X being a chalcogen atom (e.g., S, Se, Te).245 Dielectric Polymers

Dielectric polymers exhibit gas-sensing mechanisms such as dispersion, polarizability, dipolarity, basicity, acidity, and hydrogen-bonding interactions.76 In single-output sensors, measurements are performed with capacitance readout246,247 unless these polymers are formulated with conducting nanoparticles to measure the change in sensor conductivity.248 Such sensors have known insufficient selectivity; thus, they have been assembled into classic sensor arrays.234,249 Several reviews analyze performance of dielectric polymers in gas sensing.76,250 In an important initial study using impedance measurements, dielectric polymers were deposited as films onto interdigital electrodes and exposed to several organic vapors.215 While absorbed vapors had only negligible effect on the sensing film conductivity G, sensor resistance R exhibited significant changes due to the relation to sensor capacitance C as R = G/(G2 + (2πf C)2) where f was the measurement frequency. Following that initial study, polymer-coated nonresonant198,234 and RLC resonant sensors201,202,205,220 have been demonstrated for gas sensing. Examples of dielectric polymers that have been explored with multivariable RC and RLC sensors include ethylcellulose, poly(ethyl acrylate), poly(etherurethane), poly(vinyl acetate), cyanopropylmethylphenylmethyl silicone, dicyanoallyl silicone, and polysiloxane with pendant hexafluoro-2propanol groups.201,202,205,215,220 Poly(etherurethane) has been used most extensively on RLC resonant transducers and has demonstrated important advancements in multivariable sensing using analysis of resonant spectra shown in Figure 9C. Diverse volatiles were discriminated with 2D dispersion of the data shown by the PCA scores plot in Figure 12201 with PCA responses roughly related to vapor dielectric constant. Self-correction against fluctuations of ambient temperature was demonstrated (Figure 13)194,251 by taking advantage of different temperature- and gas-induced effects on the equivalent circuit components of the sensor (dielectric sensor substrate, metal sensor coil, dielectric sensing film, and

Figure 12. 2-D dispersion of a multivariable RLC resonant sensor with poly(etherurethane) sensing film presented as PCA scores plot in discrimination of eight volatiles at their single concentrations.201 Dielectric constants of respective solvents are shown in parentheses.

semiconductor memory chip). Such an approach will be attractive in applications where temperature stabilization of a gas sensor or addition of auxiliary temperature or uncoated reference sensors is prohibitive. When RLC sensors were tested with relatively high concentrations of vapors, 3-D sensor dispersion was achieved even using only three model vapors (water, toluene, and tetrahydrofuran) as illustrated in Figure 14.201 The PC3 contribution of that sensor was 0.01% but was correlated with concentrations of vapors. Another RLC sensor was based on a conventional flexible passive radio frequency identification (RFID) tag with an integrated circuit memory chip and its antenna inlay laminated with poly(etherurethane) film. Using four vapors (toluene, acetone, ethanol, and water), this sensor had 3-D dispersion and demonstrated a strong PC3 contribution of 13% as shown in Figure 15A252 due to the carefully selected sensor-excitation conditions. Using a roll-toroll manufacturing process, ∼5000 flexible RFID sensors were made (Figure 15B).63 Binary and ternary mixtures of model vapors were quantified using RLC sensors with poly(etherurethane) sensing films.201 Resonant RLC sensors also can be evaluated by measurements of response delay, resonant frequency, and other parameters.253−255 In recent multivariable sensors operating in the microwave range, measurements of response delay were combined with resonant frequency and response amplitude.220 J

DOI: 10.1021/acs.chemrev.6b00187 Chem. Rev. XXXX, XXX, XXX−XXX

Chemical Reviews

Review

Figure 13. Response of the developed multivariable resonant RLC sensor to various concentrations of water vapor at variable temperatures. (A) PCA scores plot. (B) Actual vs predicted values of water vapor concentrations at four different temperatures. Reprinted with permission from ref 194. Copyright 2013 Elsevier.

scores plots of PCA models. These results illustrate 2-D dispersion of these sensors where the chosen polymer modulates the selectivity of the sensor and its contributions to PC1 and PC2. Conjugated Polymers

Conjugated polymers (also known as conducting polymers256,257 or intrinsically conducting polymers258) exhibit several mechanisms of gas response including changes in density and charge carrier mobility, polymer swelling, and conformational transitions of polymer chains.77,78 Sensors with singleoutput measurements of resistance259,260 or capacitance261 of conjugated polymers have known insufficient selectivity; thus, they have been assembled into classic sensor arrays.259,260 Several reviews analyze applications of conjugated polymers in single-output sensors and their arrays.256−258,262 In one important early study, impedance measurements at relatively low frequencies (20 Hz−10 kHz) were utilized to explore changes of relative permittivity of a “classic” conjugated polymer such as polypyrrole upon exposure to different volatiles (methanol, acetone, ethyl acetate, and ethanol).235 The patterns of R and C responses to these volatiles were different and comparable to those produced using conventional multisensor arrays. Impedance measurements were also explored at high frequencies in resonant sensor configurations using polypyr-

Figure 14. 3-D dispersion of a multivariable RLC resonant sensor with poly(etherurethane) sensing film presented as PCA scores plots in discrimination of three volatiles at their various concentrations.201 Dielectric constants of respective solvents are shown in parentheses.

Response signatures of sensors coated with cyanopropylmethylphenylmethyl silicone and polysiloxane with pendant hexafluoro-2-propanol groups are illustrated in Figure 16 as

Figure 15. 3-D dispersion of a multivariable RLC resonant sensor with poly(etherurethane) sensing film presented as PCA scores plot and an example of manufactured sensors. (A) 3-D sensor dispersion in discrimination of four volatiles at their various concentrations. Dielectric constants of respective solvents are shown in parentheses. (B) Example of roll-to-roll manufactured sensors on a flexible substrate. Reprinted with permission from ref 63. Copyright 2012 The Royal Society of Chemistry. K

DOI: 10.1021/acs.chemrev.6b00187 Chem. Rev. XXXX, XXX, XXX−XXX

Chemical Reviews

Review

Figure 16. PCA scores plots illustrating 2-D dispersion of multivariable RLC resonant sensors based on measurements of response delay, resonant frequency, and response amplitude. Sensing films: (A) cyanopropyl methyl phenylmethyl silicone and (B) polysiloxane with pendant hexafluoro-2propanol groups. Data adapted and reproduced with permission from ref 220. Copyright 2015 Elsevier.

Figure 17. Dissipation factor measurements using multivariable RLC resonant sensors with conjugated polymers. (A) Response to individual vapors of methanol, acetone, and ethyl acetate. (B) Response to individual vapors of acetone and methanol and their binary mixtures. (A) Reprinted with permission from ref 195. Copyright 1995 Institute of Physics. (B) Reprinted with permission from ref 180. Copyright 1997 Elsevier.

Figure 18. Applications of metallophthalocyanines for multivariable vapor sensing. (A) Diversity of resistance and capacitance responses obtained from nonresonant impedance measurements of tetrakis-t-butyl phthalocyaninatonickel film.215 (B) Diversity of dissipation spectra obtained from resonant measurements of cobalt phthalocyanine film.238 (A) Reprinted with permission from ref 215. Copyright 1996 Elsevier. (B) Reprinted with permission from ref 238. Copyright 2006 American Institute of Physics.

role195 and polyaniline.196 In these sensors, the dissipation factor was an indicator of the change in dipole−dipole interactions upon sorption of vapors into polymers. Figure 17A shows the dissipation factor vs frequency response of polypyrrole to methanol, ethyl acetate, and acetone vapors.195 A

specific frequency of the sensor dissipation factor was related to vapor type, while the response magnitude was proportional to vapor concentration. It was suggested that these specific frequency responses may be correlated to the permittivity of the polymer film. This technique was further applied to analyze L

DOI: 10.1021/acs.chemrev.6b00187 Chem. Rev. XXXX, XXX, XXX−XXX

Chemical Reviews

Review

resistance or capacitance of metal oxides have known insufficient selectivity; as a result, they have been assembled into classic sensor arrays as recently reviewed.148 Different techniques to improve the selectivities of individual MOS sensors have been explored. One such technique utilized a carefully designed temperature modulation of a metal oxide sensing film to take advantage of different mechanisms of gas response of metal oxides and their gas selectivity at different temperatures.239,272−277 For example, this approach provided an impressive ability to resolve numerous odors with a single sensor using a conventional resistance readout but by rapidly (85% accuracy.371 In summary, reported multivariable field-effect transistors with inorganic and organic sensing films typically demonstrate 2-D365,366 and 3-D193,370 response dispersion.

8. MULTIVARIABLE PHOTONIC RESONANT SENSORS Photonic multivariable sensors can be categorized as materialand structure-based. Material-based sensors utilize materials with multifunctional properties that allow several partially or fully independent responses. Such materials comprise units that are much smaller than the wavelength of interrogation light. Structure-based sensors utilize physical structures that are responsible for the multivariable performance of these sensors. Such structures comprise units that are comparable with the wavelength of interrogation light. Material-Based Multivariable Photonic Sensors

Material-based multivariable photonic sensors go back to ratiometric probes that provided light-intensity-372 or temperature-independent373 measurements. Follow-up developments included sensing materials formulated with several reporter moieties.374 Multiwavelength luminescent and colorimetric reporter moieties were demonstrated to respond to different chemicals provided by several reporter units in a moiety. Such reporters were developed to exhibit absorbance, fluorescence, electroluminescence, and other changes in relation to detected chemicals.375,376 Compounds for dual-, triple-, and quadruplechannel sensing of volatile and nonvolatile analytes have been demonstrated.377−383 Developments in nanotechnology provided controlled synthesis of plasmonic nanoparticles for resonant vapor sensing. Classic sensor arrays based on plasmonic nanoparticles functionalized with soft384 and rigid385 layers have been reported. Plasmonic nanoparticles have been functionalized with “soft” organic layers for multivariable gas sensing at room temperature.74,386 Nanocomposites of plasmonic nanoparticles with “rigid” inorganic layers of metal oxides provided the ability for gas sensing at elevated temperatures.95,387 The mechanisms of optical vapor response using plasmonic nanoparticles functionalized with a soft organic layer include (1) variation in the interparticle spacing related to the type and concentration of the vapor, (2) variation in the refractive index of the organic layer related to the partition coefficients of vapors sorbed into this ligand shell, and (3) variation of the reflectivity of the metal nanoparticle network film affected by the variations of the film thickness74 (Figure 35A). These effects allowed U

DOI: 10.1021/acs.chemrev.6b00187 Chem. Rev. XXXX, XXX, XXX−XXX

Chemical Reviews

Review

Figure 36. Au-CeO2 nanocomposite film for multivariable gas sensing at high temperature. Model gases: H2, CO, and NO2 at 500 °C. PCA scores plots showing the data projected onto the PC axes: (A) PC1 vs PC2 and (B) PC2 vs PC3. Nonoverlapping clusters indicate a unique response to each of the three analytes. Data marker size increases with increasing concentration. Reprinted with permission from ref 95. Copyright 2012 American Chemical Society.

Figure 37. Structurally colored colloidal crystal film formed from composite core/shell nanospheres for multivariable gas sensing. (A−D) Differential reflectance spectra for four vapors: water, acetonitrile (ACN), dichloromethane (DCM), and toluene, respectively, at various concentrations. (E) PCA scores plot of response of the colloidal sensor film for four tested vapors demonstrating 3-D dispersion. (F) Reflected light image of the sensing film on a Teflon support. Reprinted with permission from ref 96. Copyright 2008 Institute of Electrical and Electronics Engineers.

The mechanisms of vapor response using plasmonic nanoparticles in nanocomposite films with metal oxides and operating at high temperatures (300−800 °C) involve the charge exchange with the nanoparticles or a change in the dielectric constant surrounding the nanoparticles, dependent on the type of metal oxide and its morphology.95 The material of reported nanoparticles was typically gold or gold alloys; examples of metal oxides in such plasmonic films include CuO, ZnO, TiO2, NiO-SiO2, SiO2, BaO, CeO2, and yttriastabilized zirconia (YSZ, ZrO2, and Y2O3).95,385,387−393 Most of the reported nanocomposite plasmonic films with metal oxides have not been explored yet for their multivariable responses to different gases. Only several metal oxides have been used in multivariable plasmonic sensing. They did show important results such as 3-D dispersion. Au-CeO2 nanocomposite films were used for multivariable sensing of individual

discrimination of individual vapors and their mixtures using gold nanoparticles functionalized with a 1-mercapto-(triethylene glycol) methyl ether ligand, selected for its amphiphilic properties to respond to both polar and nonpolar vapors. Six diverse vapors (water, methyl salicylate, tetrahydrofuran, dimethylformamide, ethyl acetate, and benzene) produced plasmonic responses affecting the peak and the short- and long-wavelength shoulders of the plasmonic band. The PCA scores plot illustrated that this sensing film discriminated most of the tested vapors (Figure 35B) ranked in the order of the refractive index of the respective solvent except for water vapor, likely because water was the only polar protic solvent among tested vapors. Sensor response to these volatiles produced only 2-D dispersion. This sensing film was further utilized to quantify ethyl acetate and benzene in their mixtures using PLS. V

DOI: 10.1021/acs.chemrev.6b00187 Chem. Rev. XXXX, XXX, XXX−XXX

Chemical Reviews

Review

H2, CO, and NO2 gases at 500 °C.95 Spectral multivariate analysis was used to gauge the inherent selectivity of the film between the separate analytes. The PCA model had three PCs shown in Figure 3695 and demonstrated identifiable responses for each gas. Similarly, the use of a Au-TiO2 nanocomposite film at 500 °C also provided three PCs in its PCA model.385 To eliminate the need for a white light source and to take advantage of a thermal emission of the sensing material at high temperature, lithographically patterned Au nanorods were used upon tuning of their absorption peak into the nearinfrared. An Au nanorod−YSZ nanocomposite film was used for multivariable sensing of H2, CO, and NO2 gases at 500 °C.387 The plasmonic absorption spectrum of the sensing film overlapped with the thermal energy emitted by the tube furnace and the calculated spectral irradiance from Planck’s distribution, providing opportunities for the detection of white light and thermal spectra from the sensing film over the 600−1000 nm wavelength range. The multivariable gas response achieved by using a passive thermal emission of the film was compared to results using a white light source for film illumination. Illumination with the white light source provided a slightly better discrimination between three gases versus thermal imaging where the two reducing gases were not discriminated well. Other plasmonic materials attractive but not demonstrated yet for multivariable sensing include nanostructured metal hydrates,394 template-fabricated plasmonic nanoholes on analyte-sensitive substrates,395 and hybrid nanocomposites.396

the design features of these nanostructures that provided geometries that are difficult to reproduce using existing nanofabrication tools but attractive for sensing. In particular, iridescence of tropical Morpho butterflies (Figure 38A)99 is

Structure-Based Multivariable Photonic Sensors

Structure-based multivariable photonic sensors have been recently reported to demonstrate improvements over conventional single-output sensors and sensor arrays. Previous photonic structure-based vapor sensors operated on singleoutput vapor quantitation principles based on detection of wavelength shift of the resonance peak measured with a spectrometer397,398 or signal-intensity change measured at a single wavelength.399 These sensors were based on porous silicon, self-assembled colloidal particles, mesoporous photonic crystals, inverse opals, and high-Q resonators as analyzed in recent reviews.400−411 Traditionally, for monitoring of multiple analytes, such sensors were combined in an array with each sensor having a partial response selectivity to a certain class of analytes.397,412 Recently, multivariable vapor sensing has been accomplished using a composite structurally colored colloidal crystal film selfassembled from polystyrene core nanospheres with a sol−gel shell (Figure 37).96 The detection mechanism was based on the vapor-induced changes of the optical lattice constant of this composite colloidal crystal array where the polystyrene cores of the nanospheres were preferentially responding to nonpolar vapors while the sol−gel shells were affected mostly by polar vapors. The associated vapor-induced variations in the shape of the Bragg diffraction band were resolved using PCA of differential reflectance spectra. For the evaluation of the response of the sensor, four model vapors of different polarities were selected (water, acetonitrile, dichloromethane, and toluene). The best selectivity was obtained between nonpolar vapors, with less resolution of polar vapors. The built PCA model provided a 3-D response dispersion of the sensor. Natural biological nanostructures have been the recent focus of growing attention for bioinspired sensing and other technological applications.97,413−415 This interest is driven by

Figure 38. High vapor-selectivity of natural Morpho scales. (A) Iridescent coloration of Morpho sulkowskyi scales. Images of (B) bare and (C) Al2O3-coated ridge nanostructures with lamellae. (D) Image of conformal epicuticle on the lamellae. An out-of-plane microrib is also visible. (E) Schematic of the tree-like tapered structure of natural butterfly scales with its chemical gradient of surface polarity. (A, E) Reprinted with permission from ref 99. Copyright 2015 Nature Publishing Group. (B, D) Reprinted with permission from ref 98. Copyright 2013 The National Academy of Sciences USA.

attractive because it is produced by microscopic scales that have a nanostructure with an open-air architecture98 that allows vapors to interact with all its regions. Ridges of the scales act as a diffraction grating and lamella of the ridges act as multilayer interferometric nanoreflectors.416 These biological nanostructures were visualized using electron microscopy of bare (Figure 38B)98 and Al2O3-coated (Figure 38C) samples. Initially, iridescent scales of the Morpho butterflies provided an unexpected diverse optical response to different vapors.97 This finding inspired further studies in this direction.414,417−427 The origin of this vapor selectivity was found to be a gradient of W

DOI: 10.1021/acs.chemrev.6b00187 Chem. Rev. XXXX, XXX, XXX−XXX

Chemical Reviews

Review

Figure 39. High vapor selectivity of natural Morpho scales. (A−C) Differential reflectance spectra upon exposure to water, methanol, and ethanol vapors, respectively, at various concentrations. (D) PCA scores plot illustrating discrimination of water, methanol, and ethanol vapors with a 3-D dispersion. Reprinted with permission from ref 97. Copyright 2007 Nature Publishing Group.

Figure 40. Tuning of vapor selectivity of Polyommatus icarus butterfly scales. PCA scores plots of spectral responses to diverse vapors at their various concentrations of (A) bare wing scales and (B) wing scales conformally covered by 5 nm Al2O3. Adapted and reproduced with permission from ref 425. Copyright 2014 Elsevier.

discrimination of all the vapors (Figure 39D) demonstrating strong 3-D dispersion.97 To explain the experimentally observed results, spectral responses were computed upon a uniform and gradient coverage of the Morpho structure with different adsorbed vapors.98 With a uniform coverage, computed spectra did not explain the experimental results. Simulations of gradient coverage produced reflectance spectra that formed individual response directions in the PCA scores plot, similar to experimental results.97 Besides using scales of Morpho butterflies, vapor-sensing experiments were also performed using other types of iridescent butterflies.425,428 Scales of Polyommatus icarus were utilized for measurements of responses to diverse model vapors (Figure 40A).425 When a 5 nm thick conformal Al2O3 coating was applied onto the photonic structure of the Polyommatus icarus butterfly, this thin film intended to isolate the epicuticle from the vapors. Indeed, the vapor-response pattern was significantly

surface polarity of the