Carbon nanotube-based chemical sensing

(1)

CARBON NANOTUBE-BASED CHEMICAL

SENSING

by

Vera Schroeder

M.S. Chemistry, 2013

Rheinisch-Westfalische Technische Hochschule Aachen Submitted to the Department of Chemistry

in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY IN CHEMISTRY

at the

Massachusetts Institute of Technology

Signature of Author:

June 2019

Signature redacted

( Certified by: Accepted by: MASSACHUSETTS INSTITUTE MASSACHUSETTS INSTITUTE OF TECHNOLOGY

MAY 2

3 2019

LIBRARIES

Department of Chemistry ,Vay 14, 2019

3ignature redacted

Tir not"Lf. Swager John D. MacArthur Professor of Chemistry Thesis Supervisor

Signature redacted

...Robert W. Field Haslam and Dewey Professor of Chemistry Chairman, Departmental Committee on Graduate Students

(2)

This doctoral thesis has been examined by a Committee of the Department of Chemistry as follows:

Signature redacted

Yogesh Surendranath

Paul M Cook Career Development Assistant Professor Thesis Committee Chairperson

Signature redacted

Tirt4 y M. SwagerJ

John D. MacArthur Professor of Chemistry Thesis Supervisor

Signature redacted

Christopher C. Cummins

Henry Dreyfus Professor of Chemistry Thesis Committee Member

(3)

(4)

CARBON NANOTUBE-BASED CHEMICAL

SENSING

by

Vera Schroeder

Submitted to the Department of Chemistry

on May 10, 2019 in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy in Chemistry

In this thesis, we introduce approaches to carbon nanotube-based sensing for applications in environmental monitoring, disease diagnostics, and food analysis:

In Chapter 1, we introduce carbon nanotube-based sensing. We describe parameters that give rise to the sensing capabilities of CNT-based sensors and discuss important performance parameters of carbon nanotube sensors.

In Chapter 2, we demonstrate voltage-activated sensing of carbon monoxide using a sensor comprising iron porphyrin and functionalized single walled carbon nanotubes (F-SWCNTs). Modulation of the gate voltage offers a predicted extra dimension for sensing. Specifically, the sensors show significant increase in sensitivity toward CO when negative gate voltage is applied. In Chapter 3, we describe the design of a sensor for the highly selective detection of acrylates using conditions for the aerobic oxidative Heck reaction. The sensors mirror the catalytic processes and selectively respond to electron deficient alkenes by adapting a catalytic reaction system to modulate the doping levels in carbon nanotubes.

In Chapter 4, we introduce sensor arrays consisting of imidazolium-based ILs with different substituents and counterions to provide selective responses for known biomarkers of infectious diseases of the lungs.

In Chapter 5, we discuss a sensor array comprised of platform 20 functionalized SWCNT sensing channels for the classification of cheese, liquor, and edible oil samples based on their odor. We classify unknown food samples using a k-nearest neighbors model and a random forest model trained on extracted features. This protocol allows us to accurately differentiate between five cheese and five liquor samples (91% and 78% respectively) and only slightly lower (73%) accuracy for five edible oils.

Thesis Supervisor: Timothy M. Swager

(5)

LIST OF FIGURES

Figure 1-1: Schematic of sensing mechanisms in CNT-based sensors: (a) at the sidewall or the length of the CNT (intra-CNT), (b) at the CNT-CNT interface (inter-CNT), and (c) at the interface between the metallic electrode and the CNT (Schottky barrier)... 22 Figure 1-2: Intra-CNT (semiconducting) sensing mechanism through changes in charge carrier concentration or mobility. Hypothetical transfer (l-Vg) curves and band diagrams before (black) and after (red) the exposure to the analyte for three different sensing mechanisms. The dotted line in the band diagram corresponds to the metal workfunction of the electrode and diagrams are given for both p- and n-type semiconductors interfaces with a metal. (a) n-Doping of the CNT induces a shift of the I-V curve to more negative voltages. The band diagram shows a hole doped CNT. (b) Schottky barrier modulation corresponds to a change of the barrier height between the work function of the metal electrode and CNT and asymmetric change in conductance for electron and hole transport. The band diagram shows change in barrier height for hole transport. (c) Change in mobility can be induced by the addition of resistive elements or carrier scattering which reduces the conductivity in both p- and n-type materials ... 24 Figure 1-3: G-bands in the of Raman spectra of SWCNTs when interacting with electron donating and accepting molecules: (1) tetrathiafulvalene, (2) aniline, (3) pristine SWCNT, (4) nitrobenzene, (5) tetracyanoquinodimethane, and (6) tetracyanoethylene... 25 Figure 1-4: CNT-based chemical sensor designed based on inter-CNT mechanism. (a) Illustration of polymer swelling upon exposure of a CNT/polymer composite to solvent vapors. Reproduced with permission from Ref. ". Copyright 2007, Elsevier. (b) Schematic illustration of a chemiresistive sensor comprising SWCNTs and metallosupramolecular polymer (MSP) showing the polymer degradation upon exposure to chemical warfare agent mimic diethyl chlorophosphate (D E C P ). ... 2 7 Figure 1-5: Three sensor architectures to probe Schottky vs intra-tube sensing mechanisms. Schematic for (a) device with bare CNTs, (b) device with passivated CNT-electrode contacts, and (c) device with passivated length of CNTs that are not in contact with the metal electrode. ... 28 Figure 1-6. Hypothetical sensing response curve (a) and calibration curve (b) of a sensor exposed to increasing concentration of the analyte. ... 31

Figure 2-1: Carbon monoxide detection by a bio-inspired sensors. Schematic of a field-effect transistor (FET) substrate with Au source-drain electrodes, and Ti adhesion layer deposited on SiO2 dielectric layer and Si gate electrode. Chemical structures of pyridyl-functionalized single wall carbon nanotubes (F-SWCNTs) and iron porphyrin (Fe(tpp)CI04), depicting the coordination chemistry of the pyridyl group to the iron center of the porphyrin. ... 38 Figure 2-2: Sensing responses at no gate voltages. (a) Average changes in the conductance and standard deviations (N 6 sensors) in response to 2 min exposures to 200 ppm of CO for F-SWCNTs without Fe(tpp)C104 (black), pristine SWCNTs with Fe(tpp)C104 (green), and three

(8)

densities of functionalization (red, blue, violet). (b) Conductance changes of F-SWCNT-1 with Fe(tpp)CIO4 in response to various concentrations of CO gas diluted in N2... ... ... .. . . 40

Figure 2-3. Robustness and selectivity of the CO sensors. (a) Conductance curves of F-SWCNT-1 with Fe(tpp)CIO4 sensors in response to 2 min of 200 ppm of CO gas in air (42 % relative humidity) and dry N2. (b) Comparison between the response to CO in both N2 and air to the

respo nses to C O 2 a nd 0 2... 42

Figure 2-4: UV-Vis investigation of reactivity of Fe(tpp)C104 to CO in solution of THF. (a) Fe(tpp)CIO4 before and at various times after addition of Na metal. (b) Photograph of the color change with the addition of sodium metal and subsequent addition of carbon monoxide. (c) Non-reduced Fe(tpp)CI04 before and after addition of CO. (d) Blue shift in the spectra of fully reduced porphyrin upon addition of C O . ... 43 Figure 2-5: Enhancement in sensitivity via application of the gate voltage. (a) Conductance curves of F-SWCNT-1 with Fe(tpp)CIO4 sensors in response to 2 min of 200 ppm of CO at +3, 0, and

-3 V gate voltage. (b) Change in conductance towards an exposure of 2 min at 200 ppm of CO as a function of the gate voltage. Dashed line to guide the eye. (c) Schematic of band diagram of

SWCNT and Fe"'py(tpp)CIO4 (py = pyridine) and interactions between the two upon application

o f g a te v o lta g e . ... 4 5 Figure 2-6: Computed change in the Fermi energy (AEF) upon addition of Fel porphyrin and subsequent addition of CO or 02 relative to the Fermi energy of the pristine SWCNT with inserts of the ground-state geometries. For these molecules the Fermi level is defined as the level of the H O M O . ... 4 6 Figure 3-1. Sensors using reactivity schemes to detect ethyl acrylate. (a) Schematic of sensing device containing gold electrodes on a glass substrate, a network of SWCNTs, and a liquid selector mixture (green). (b) Average response and standard deviations (N 2 6 sensors) of SWCNT-sensor using a mixture of Pd(OAc)2, NaOAc, trihexyl(tetradecyl)phosphonium chloride, and iodo-, bromo- or chlorobenzene as the co-selector towards 50.Oppm of ethyl acrylate in N2.

... 7 0 Figure 3-2. 1H-NMR spectrum of reaction product of Heck reaction between iodo-benzene and ethyl acrylate in CDC13 isolated from sensing device after 14h exposure to 100ppm ethyl acrylate.

Literature reference: (400 MHz, CDC 3): 6 1.34 (3H, t, J = 7.1 Hz, CH3), 4.27 (2H, q, J = 7.1Hz, OCH2), 6.44 (1H, d, J = 16.0, EtOCCH=), 7.37-7.39 (3H, m, Ar-H), 7.51-7.52 (2H, m, Ar-H), 7.69 (1 H, d, J = 16.0,ArCH=).3 . . . ._. ₇₁

Figure 3-3. (a) Average response and standard deviations (N 2 6 sensors) of device containing different solvents using [Pd(CH3CN)4](BF4)2, 4,5-diazafluorenone, and phenylboronic (ratio

0.05:0.1:1) towards 50.0 ppm ethyl acrylate diluted in air (relative humidity = 46%). (b) Average changes in the conductance and standard deviations (N 2 6 sensors) in response to 1 min exposures (shaded area) for sensors missing the boronic acid coupling partner (black curve), sensor missing the Pd source (red curve), sensors missing the ligand (blue curve), and the complete catalytic mixture 1 (green curve) sensors and insert with components of catalytic mixture

(9)

Figure 3-4. (a-b) Response of 1-device towards different concentrations of ethyl acrylate in air compared to the short-term exposure limit as defined by OSHA.127_{(c) Average response and} standard deviations (N 2 6 sensors) of device towards 50 ppm ethyl acrylate (shaded areas) freshly prepared and after storage on benchtop for two weeks. ... 74 Figure 3-5. Average response and standard deviations (N 6 sensors) of 1-device towards different classes of molecules: (a) volatile organic compounds, (b) simple and activated alkenes, (c) strained cyclic alkenes and small gaseous analytes, (d) and alkynes... 75 Figure 3-6. Sensor response of devices containing different ligands and palladium sources compared to published126

,138 reaction yields: (1) 4,5 diazaflourenone/[Pd(CH₃CN)₄](BF₄)2, (2)

4,4'-dimethyl-2,2'-dipyridyl/[Pd(CH3CN)4(BF4)2 (3) no ligand/[Pd(CH3CN)4](BF4)2 (4)

2,9-dimethyl-1,1 0-phenanthroline/[Pd(CH3CN)4](BF4)2 (5) phenanthroline/[Pd(CH3CN)4](BF4)2 (6) 5,5'-dimethyl-2,2'-dipyridyl/[Pd(CH3CN)4](BF4)2 (7) 6,6'-dimethyl-2,2'-dipyridyl/[Pd(CH3CN)4](BF4)2.,

(8) 2,2'-bipyridine/[Pd(CH3CN)4(BF4)2, (9) phenanthroline/Pd(OAc)2,(10) 2,2'-b ip y rid in e /P d (O A c )2... 7 6

Figure 4-1. Schematic representation of device fabrication and synthesis. (a) Fabrication of the chemiresistive sensor array using various IL-SWCNT pastes. (b) Synthesis of methyl imidazole-based IlLs with Cl- and anion exchanges from Cl- to FeC4- and PF- for each IL. ... 88 Figure 4-2. Sensing performance of chemiresistive array. (a) Exemplary sensing traces of exposures to ethanol, toluene, and heptanal (diluted in dry N2) for Cl and PF6-containing IL/CNT

composites. (b) Sensing response patterns of all nine different IL-SWCNT-based chemiresistive sensors to nine representative volatile organic compounds related to human breath. ... 92 Figure 4-3.Stability and selectivity measurements (a) Sensing response of device to 1000 ppm toluene diluted in dry N2, after 1-3 days, after 12-15 days and after 23-26 days of storage under

ambient conditions without additional precautions. (b) Relative responses of [l] containing sensors to toluene diluted in dry N2 as a function of storage time. (c) Principal component analysis

(PCA) of sensing response to VOC biomarkers related to TB. The arrows indicate the direction of the response migration in the PCA plot after 1-2 (=X), 12-15 (=Y) and 23-26 (=Z) days. (d) Scree p lo t o f P C A ... 9 4 Figure 4-4. Gas sensing under ambient and humid conditions. (a) Principal component analysis of sensing response to five TB related VOCs and two VOC mixtures diluted in air (200 ppm

heptanal, 200ppm 3-heptanone, 200ppm 2-methylpentane, 1000 ppm toluene, 100ppm benzene, and 1000ppm benzene and toluene mixtures). (b) PCA sensing response to toluene, healthy breath and simulated diseased breath (healthy breath + 0.85 % saturated toluene vapor) using a ir a s th e c a rrie r g a s . ... 9 6 Figure 6-1. Schematic of sensing device with carbon-based electrodes de-posited on a Kapton substrate. The active layer of SWCNTs and selectors is deposited between the electrodes. All 20 se le cto rs a re liste d in T a b le S 6 -1...1 14 Figure 6-2. Example of sensing response for one selector (S4) towards cheddar, Mah6n, and pecorino. The response is represented as a change in conductance normalized to the conductance at the start of the exposure (AG/Go). The exposure starts at t = 60 s and ends at t =

(10)

180 s (marked by dashed vertical lines), the recovery period starts at t = 180 s and ends at 360 s. The response is an average of 12 separate sensing experiments, the blue/green shaded areas represent the standard deviation of the response. The purple shaded area represents the data u s e d fo r c la s s ific a tio n ... 1 1 7 Figure 6-3. a,b) Selector accuracies for both the (a) f-RF and (b) KNN models using single selectors for differentiating between cheddar, Mahon and pecorino cheese. The shaded, grey area corresponds to random guessing (33 % accuracy between cheddar, Mahdn, and pecorino), selectors with accuracies around this threshold do not assist in this particular application. c,d) Example results from the combinatorial selector scan for both the (a) f-RF and (b) KNN models on the cheese dataset (cheddar, Mah6n and pecorino) using combinations of four selectors. Plotted are the top three (1-3), three medium (2423-2425), and the bottom three (4843-4845) combinations. Each selector combination was trained 36 times, plotted are the average accuracies and standard deviations...119 Figure 6-4. Selector scan results showing only the highest accuracy selector combinations for e a c h u s e c a s e ... 1 2 1

Figure 6-5. Sensing response for a) S4, b) S5, c) S6, and d) S20 towards five cheeses. The response is represented as a change in conductance normalized to the conductance at the start of the exposure (AG/Go). The exposure starts at t = 60 s and ends at t = 180 s (marked by dashed vertical lines). Each response is an average of 40 separate sensing experiments, the shaded area represents the standard deviation of the response...123 Figure 6-6. PCA analysis of extracted features from the five-cheese dataset showing the first two p rin c ip le c o m p o n e n ts . ... 12 5 Figure 6-7. Top 16 overall most important features in the five-cheese f-RF model. The importance is averaged over 50 f-RF models trained and tested on shuffled data splits. The top 16 features for cheese, liquor and edible oil are listed in Table S2-4. ... 126

LIST OF SUPPLEMENTARY FIGURES

Figure S2-1. (a) Photograph of the experimental gas sensing setup at KAUST showing the mass flow controllers ans gas enclosure. (b) Photograph of sensor connected to the test clip outside of the PTFE gas enclosure. (c) Photograph of the sensing setup at KAUST setup showing a more detailed view of the Alicat Scientific mass flow controllers. (d) Photograph of the PTFE gas enclosure show ing the gas inlet and outlet... 51 Figure S2-2. Reaction of pyridyl-iodonium salt with pristine SWCNTs... 53 Figure S2-3. Characterizations of functionalized SWCNTs: (a) Raman spectroscopy, (b) TGA, (c) X P S , a nd (d ) U V -V is-N IR . ... 54

(11)

Figure S2-4. 1_{H NMR spectrum of the reaction filtrate of a functionalization of SWCNTs using 0.05} equivalents of sodium naphtalide and 0.05 equivalents of the pyridyl iodononium salt measured in in C D C 13 .... 5 5

Figure S2-5. SEM images of the devices comprising F-SWCNTs and iron porphyrin. (a) Devices fabricated by a one-step method of drop-casting a mixed solution of F-SWCNTs and Fe(tpp)ClO 4 and a magnified image (b). (c) Surface of pure F-SWCNTs before infusion of Fe(tpp)C104 (d). 56 Figure S2-6. Differential pulse voltammetry under N2 and schematic explanation of processes

during DPV experiment. (A) DPV of pristine SWCNTs decorated with Fe(tpp)CI04 before and after addition of pyridine. (B) DPV of functionalized SWCNTs decorated with Fe(tpp)C104 before and after addition of pyridine. (C) DPV of functionalized SWCNTs decorated with Fe(tpp)CIO4 before and at certain points during exposure to CO. (D) DPV of functionalized SWCNTs decorated with Fe(tpp)C104 under CO before and at certain points during purging with N2 ...... ... . . . 57

Figure S2-7. Trace of the change in conductance of the CO sensors with prolonged exposure to 200 ppm of CO. A deviation from the linear response occurred near 10 min and the saturated regim e w as observed after 15 m in ... 58 Figure S2-8. Conductance changes of F-SWCNT-1 with Fe(tpp)C10 4 in response to various concentrations of C O gas diluted in N2... .. ... ... ... .... ... ... ... . . . . 59

Figure S2-9. UV-Vis investigation of reactivity of Fe(tpp)C104 in THF solution (1 pM). (a) UV-Vis spectra of Na-reduced species before and at various times after purging with air to demonstrate aerobic reoxidation. (b) UV-Vis spectra of fully reduced porphyrin upon addition of CO before and after vortexing in air for 15min to show irreversibility of the binding of CO. (c) UV-Vis spectra of Fe(tpp)C104 as-is before and after various times of vortexing in air. Figure S9 (d-e) show the same samples as figure 4a and S9(b-c) in the range of 450-600 nm. (d) UV-Vis spectra (450 nm to 600 nm) of Fe(tpp)CI04 before and at various times after addition of Na metal. (e) UV-Vis spectra (450 nm to 600 nm) of fully reduced porphyrin upon addition of CO before and after vortexing in air for 15min. (f) UV-Vis spectra (450 nm to 600 nm) of Fe(tpp)C10 4 as-is before and after various times o f v o rte x in g in a ir...6 0 Figure S2-10. Ground-state geometries of (5,5) SWCNT fragment (a) pristine, (b) with a Fe"porphyrin, and with small gaseous ligands: CO (c), and 02 (d). Schematic illustrating the high and low level of the ONIOM calculation of (5,5) SWCNT fragment (e) pristine, (f) with a Fe"porphyrin, and with small gaseous ligands: CO (g), and 02 (h)... 62 Figure S2-1 1. Frontier orbitals (contour value = 0.02 e-a.u.3_{) and DOS for pristine (5, 5) SWCNT}

(a), (5, 5) SWCNT with Fe"porphyrin (b), (5, 5) SWCNT with Fe'porphyrin and CO (c),and (5, 5)

SW C N T w ith Fe"porphyrin and 0 2. ... 63

Figure S2-12. Computed change in the Fermi energy (AEF) of the ONIOM calculation upon addition of Fe"porphyrin and subsequent addition of CO or 02 relative to the Fermi energy of the pristine SW CNT with inserts of the ground-state geometries... 64 Figure S2-13. Transfer characteristic of P-SWCNT, F-SWCNT-1, and F-SWCNT-1 + Fe(tpp)C104 with the constant drain-source voltage of 0.1 V ... 65

(12)

Figure S3-1. 1 H-NMR spectrum of reaction product of Heck reaction in CDC13 isolated from

sensing device after 14h exposure to 100ppm ethyl acrylate. A) Between iodo-benzene and ethyl acrylate using Pd(OAc)2, NaOAc, trihexyl(tetradecyl)phosphonium chloride, and iodobenzene in a ratio of 0.04:1.5:10:1.0. b) Between phenyl boronic acid and ethyl acrylate using catalytic m ix tu re 1 ... 8 0 Figure S3-2. Additional controls for proof-of-principle system. Average changes in the conductance and standard deviations (N 6 sensors) in response to 1 min exposures of 50 ppm ethyl acrylate in N2 (shaded area) for sensors with CNT-only (black curve), sensors with the IL,

trihexyl(tetradecyl)phosphonium chloride (red curve), sensors missing the Pd-source, Pd(OAc)2 (green curve), and sensors with the complete catalytic mixture containing Pd(OAc)2, NaOAc, trihexyl(tetradecyl)phosphonium chloride, and iodobenzene (green curve). ... 81 Figure S3-3. Proof-of-principle sensor for ethyl acrylate. Average changes in the conductance and standard deviations (N ! 6 sensors) of SWCNT-sensor using a mixture of Pd(OAc)2, NaOAc, trihexyl(tetradecyl)phosphonium chloride, and iodo-, bromo- or chlorobenzene as the co-selector towards different concentration of ethyl acrylate in N2. The dotted lines are used to guide

v is u a liz a tio n .... 8 2 Figure S3-4. Response towards repeated exposures of ethyl acrylate. Average changes in the conductance and standard deviations (N 6 sensors) in response to 1 min exposures of 100 ppm ethyl acrylate in air (shaded area) for sensors containing [Pd(CH3CN)4](BF4)2, 4,5-diazafluorenone and phenylboronic acid ... 83 Figure S3-5. Response towards repeated exposures of ethyl acrylate. Average changes in the conductance and standard deviations (N 6 sensors) in response to 1 min exposures of 50 ppm ethyl acrylate in air (shaded area) for sensors containing selector mixture 1...83 Figure S3-6. Additional controls for system 1. Average changes in the conductance and standard deviations (N 6 sensors) in response to 1 min exposures of 50 ppm ethyl acrylate in air (shaded area) for sensors with CNT-only (black curve), sensors with phenylboronic acid (red curve),

sensors with phenylboronic acid and [EMIM][PO2(OEt)2] (blue curve), and sensors with the

com plete catalytic m ixture 1 (green curve) ... 84 Figure S4-1. Sensing response to 1000 ppm ethanol in air with [IL1 ][Cl]-SWCNT...102 Figure S4-2. Schematic of EBC setup. Exhaled breath is collected and condensed, to simulate diseased breath a biom arker is added in the gas phase...103 Figure S4-3. FT-IR spectra of (a) Cl- containing ILs; (b) PF6 containing ILs; (c) pristine [IL1][CI], [IL1][PF6], and their SW C N T com posites. ... 104 Figure S4-4. (a) Raman spectra of FeC4- containing ILs on Si wafer (excitation at 532 nm). (b) Raman spectra of pristine SWCNT, FeC4- containing ILs and their composites. ... 105 Figure S4-5. SEM images of (a) dropcast pristine SWCNT; pasted (b) [IL1][Cl]-SWCNT, (c)

[IL2][CI]-SWCNT and (d) [IL3][CI]-SWCNT on the silicon substrates. (White arrow: sweeping d ire ctio n , sc a le b a r: 1 p m )...10 5

(13)

Figure S4-6. Response towards 2 min exposures of benzene (1000 ppm), toluene (1000 ppm), heptanal (200 ppm) and ethanol (1000 ppm) at different mass ratios (SWCNT: IL = 1:40, 1:10,

1 :4 a n d 1 :1 )...1 0 6 Figure S4-7. Average conductance changes of three IL-SWCNT chemiresistive sensors in

response to 2 min exposures to various concentrations of ethanol diluted in N2 (linear fit)...107

Figure S4-8. Fingerprint-like sensing response of a cross-reactive array with IL selectors (SWCNT: IL = 1:10). Each bar represents the average response of 3-4 sensors exposed to the analyte. Vertical error bars show standard deviation from the mean based on three exposures of sensors to e a c h o f th e a n a lyte s ... 10 8 Figure S4-9. Raman spectra of [IL1][Cl]-SWCNT depending on the exposure time of saturated (a) ethanol and (b) acetone vapors. (c) Raman spectra of pristine SWCNT depending on the exposure time of saturated ethanol vapor (excitation at 785 nm). ... 108

Figure S4-10. Raman spectra of [ILl][CI], pristine SWCNT, [IL1][Cl]-SWCNT and [ILl][Cl]-SWCNT after 10 consecutive measurements with 1000 ppm ethanol normalized to [0, 1] (e xc ita tio n a t 5 3 2 n m )...10 9 Figure S4-11. The contact angle of nine different ILs on Si wafer at room temperature (-25 % R H ). ... 1 1 0 Figure S4-12. (a) Schematic illustration of the exposure of an external magnetic fields to the sensor device. (b) Sensing response of [ILl][FeCI4]-SWCNT, [L2][FeCl4]-SWCNT and [IL3][FeCI4]-SWCNT to 1000 ppm ethanol diluted in dry N2 (the magnetic field was exposed for 1 h ). ... 1 1 1 Figure S6-1. 1H NMR (600 MHz) of selector S9 collected in CDC13... .... .... ... ....132

Figure S6-2. 1H NMR (600 MHz) of selector S10 collected in CDC13.... ... ... ... ...132

Figure S6-3. 1H NMR (600 MHz) of selector Sl1 collected in CDC13.... ... ... ... ....133

Figure S6-4. 1H NMR (400 MHz) of selector S12 collected in CDC13... ... .... ... .. ....135

Figure S6-5. Gas sensing equipment. a) C2Sense Kapton sensing substrate with 16 sensing electrodes. b) C2Sense Sensing Node with gas in-and outlet. ... 136

Figure S6-6. Selector accuracies for both the (a) f-RF and (b) KNN models using single selectors for differentiating between water and acetone. The shaded, grey area corresponds to random guessing (33 % accuracy between cheddar, Mahon, and pecorino), selectors with accuracies around this threshold do not assist in this particular application. Each training was performed 36 times, plotted are the average accuracies and standard deviations...142

Figure S6-7. PCA analysis of the the 20 selector datasets for acetone and water...143 Figure S6-8. Sensing response for a) S1, b) S2, c) S3, and d) S4 towards acetone and water.

(14)

Figure S6-9. Sensing response for a) S5, b) S6, c) S7, and d) S8 towards acetone and water.

... 1 4 4 Figure S6-10. Sensing response for a) S9, b) S10, c) S11, and d) S12 towards acetone and water.

... 1 4 5 Figure S6-11. Sensing response for a) S13, b) S14, c) S15, and d) S16 towards acetone and w a te r...1 4 5 Figure S6-12. Sensing response for a) S17, b) S18, c) S19, and d) S20 towards acetone and w a te r...1 4 6 Figure S6-13. Selector accuracies for both the f-RF and KNN models using single selectors for differentiating between (a,b) cheese (cheddar, Mah6n and pecorino), (c,d) liquor (rum, vodka, whiskey), and (e,f) edible oil (canola, olive, walnut). The shaded, grey area corresponds to random guessing (33 % accuracy between cheddar, Mahon, and pecorino), selectors with accuracies around this threshold do not assist in this particular application. Each training was performed 36 times, plotted are the average accuracies and standard deviations...147 Figure S6-14. Best performing combination from the combinatorial selector scan for both the f-RF and KNN models using combinations of 1-5 selectors on the (a,b) cheese dataset (cheddar, Mah6n and pecorino), (c,d) liquor (rum, vodka, whiskey), and (e,f) edible oil (canola, olive, walnut). Each selector combination was trained 36 times, plotted are the average accuracies and sta n d a rd d e v ia tio n s ... 14 8 Figure S6-15. Example results from the combinatorial selector scan for both the f-RF and KNN models using combinations of four selectors on the (ab) cheese dataset (cheddar, Mah6n and pecorino), (cd) liquor (rum, vodka, whiskey), and (ef) edible oil (canola, olive, walnut). Each selector combination was trained 36 times, plotted are the average accuracies and standard d e v ia tio n s ... 1 4 9 Figure S6-16. PCA analysis of the three-class, 20 selector datasets for all categories (cheese, liquor and oil). a) PC1-2 for cheese b) PC2-3 for cheese c) PC1-2 for liquor d) PC2-3 for liquor e)

P C 1-2 fo r o il f) P C 2 -3 fo r o il ... 15 0 Figure S6-17. Sensing response for a) S7, b) S8, c) S12, and d) S15 towards five liquor samples. The response is represented as a change in conductance normalized to the conductance at the start of the exposure (AG/GO). The exposure starts at t = 60 s and ends at t = 180 s (marked by dashed vertical lines). The response is an average of 40 separate sensing experiments, the shaded area represents the standard deviation of the response. ... 151 Figure S6-18. Sensing response for a) S5, b) S7, c) S8, and d) S13 towards five edible oils. The response is represented as a change in conductance normalized to the conductance at the start of the exposure (AG/GO). The exposure starts at t = 60 s and ends at t = 180 s (marked by dashed vertical lines). The response is an average of 40 separate sensing experiments, the shaded area represents the standard deviation of the response...152

(15)

Figure S6-19. PCA analysis of extracted features from the five-cheese dataset showing the first tw o principle com ponents. a) PC1-2 b) PC2-3 ... 153 Figure S6-20. PCA analysis of extracted features from the five-liquor dataset showing the first two principle com ponents. a) PC 1-2 b) PC 2-3...153 Figure S6-21. PCA analysis of extracted features from the five-oil dataset showing the first two principle com ponents. a) PC 1-2 b) PC 2-3...154 Figure S6-22. Top 16 overall most important features in the five-cheese f-RF model. The importance is averaged over 50 f-RF models trained and tested on shuffled data splits...155 Figure S6-23. Top 16 overall most important features in the five-liquor f-RF model. The importance is averaged over 50 f-RF models trained and tested on shuffled data splits...156 Figure S6-24. Top 16 overall most important features in the five-oil f-RF model. The importance is averaged over 50 f-RF models trained and tested on shuffled data splits...157 Figure S6-25. Sensing response for a) S1, b) S2, c) S3, and d) S4 towards three cheeses....158 Figure S6-26. Sensing response for a) S5, b) S6, c) S7, and d) S8 towards three cheeses. ... 159 Figure S6-27. Sensing response for a) S9, b) S10, c) S11, and d) S12 towards three cheeses.

... 1 5 9 Figure S6-28. Sensing response for a) S13, b) S14, c) S15, and d) S16 towards three cheeses.

... 1 6 0 Figure S6-29. Sensing response for a) S17, b) S18, c) S19, and d) S20 towards three cheeses.

... 1 6 0 Figure S6-30. Sensing response for a) S1, b) S2, c) S3, and d) S4 towards three liquor samples.

... 1 6 1 Figure S6-31. Sensing response for a) S5, b) S6, c) S7, and d) S8 towards 3 liquor samples.161 Figure S6-32. Sensing response for a) S9, b) S10, c) S11, and d) S12 towards three liquor s a m p le s . ... 1 6 2 Figure S6-33. Sensing response for a) S13, b) S14, c) S15, and d) S16 towards three liquor

s a m p le s . ... ... 1 6 2 Figure S6-34. Sensing response for a) S17, b) S18, c) S19, and d) S20 towards three liquor s a m p le s .... 1 6 3 Figure S6-35. Sensing response for a) S1, b) S2, c) S3, and d) S4 towards three edible oils..163 Figure S6-36. Sensing response for a) S5, b) S6, c) S7, and d) S8 towards three edible oils..164

(16)

Figure S6-37. Sensing response for a) S9, b) S10, c) S11, and d) S12 towards three edible oils.

... 1 6 4 Figure S6-38. Sensing response for a) S13, b) S14, c) S15, and d) S16 towards three edible oils.

... 1 6 5 Figure S6-39. Sensing response for a) S17, b) S18, c) S19, and d) S20 towards three edible oils.

... 1 6 5

LIST OF SCHEMES

Scheme 3-1. Aerobic oxidative Heck arylation of olefins. ... 68 Scheme 4-1. Synthesis of 1-(nonyl)-3-methylimidazolium chloride ([IL1][Cl])... 99 Scheme 4-2. Synthesis of 1-(3-phenylpropyl)-3-methylimidazolium chloride ([1L2][Cl])... 99 Scheme 4-3. Synthesis of 1-(2-(2-(2-hydroxyethoxy)ethoxy)ethyl)-3-methylimidazolium chloride ([lL 3 ][C l]). ... 1 0 0 S c h e m e 6 -1 . S ynthe sis of S 12 . ... 134

LIST OF TABLES

Table 6-1. Optimal 4-selector accuracya analysis on the five-class classification problems for both the f-R F a nd K N N m o d e ls ... 12 2

LIST OF SUPPLEMENTARY TABLES

Table S2-1. Reaction conditions used to functionalize SWCNTs... 53 Table S2-2. Summary of X-ray photoelectron spectroscopy (XPS) analysis of pristine SWCNTs and functionalized SWCNTs used in this paper...54 Table S2-3. UL2034 Required Alarm Points for different levels of CO... 58 Table S2-4. Fe-SWCNT bond lengths (dFe-CNT, A), Fe-gas bond lengths (dFe-L, A), and Fermi level

(17)

Table S2-5. Atomic dipole moment corrected Hirshfeld population, and Becke1 _{charges on (5,5)}

SWCNT, (5,5) SWCNT and Fe"porphyrin, (5,5) SWCNT and Fe"porphyrin and CO, and (5,5)

SW C N T and Fe"porphyrin and 0 2... 64

Table S6-1. Structure of selectors, origin of selectors, and composition of selector solution. ..129 Table S6-2. Full length feature names, cheese data. ... 155 Table S6-3. Full length feature names, liquor data. ... 156 Table S6-4. Full length feature names, oil data. ... 157

(18)

1 INTRODUCTION

SENSING

OF CARBON NANOTUBE-BASED

CHEMICAL

A - -i.-~

Anamytes

Signai

A*

Selectors

s

Adapted and reprinted in part with permission from: Vera Schroeder, Suchol Savagatrup, Maggie He, Sibo Lin, and Timothy M. Swager "Carbon Nanotube Chemical Sensors"

Chem. Rev., 2019, 119 (1), pp 599-663. doi.org/10.1021/acs.chemrev.8b00340

(19)

-18-1.1 Introduction

Carbon nanotubes (CNTs) have now been a subject of research for more than 20 years. Mirroring this academic endeavor is the worldwide commercial interest, leading to the production capacity of several thousand tons of CNTs per year.1 _{These developments have paved ways to}

the wide array of emerging applications2 3 _{in microelectronics,}4

,5 computing,6 medicinal therapy,7

electrochemical biosensors,8 and chemical sensors.910 _{However, the field is far from mature and}

our understanding of the chemical and physical properties of these materials continues to grow. At the outset, it is fair to state that the chemistry of CNTs remains dubious and often imprecise." Although advances in the production have allowed preferential synthesis of single-walled carbon

nanotube (SWCNTs) with metallic or semiconducting properties with selectivity of 90 to 95%,1213 production of pure semiconducting tubes remains cost-prohibitive. Commercial supplies of SWCNTs, despite improvements in consistency, are still polydisperse in length, diameter, and chirality. Separations methods by density-gradient centrifugation with selective surfactants,1 4 conjugated polymer wrappings,15

,16 or by gel chromatography17-18 are not readily scaled.

Bottom-up syntheses have seen heroic efforts,19 _{but remain far from full realization. Similarly, multi-walled}

carbon nanotubes (MWCNTs) can be produced in high volume through large-scale chemical vapor deposition (CVD), however they suffer from structural deviations, contaminations, and impurities that often require costly treatment for removal.

One may ask why CNTs continue to garner such attention, given the complexity and what some chemists may even refer to as impurities. Clearly, scientific curiosity is one answer. The other motivation driving CNT research is their unusual optical, electrical, mechanical, and chemical properties. CNTs are unique organic electronic wires with shape persistence. These rr-electron wires, possessing quantized rr-electronic states with coherence lengths that are longer than what is possible for conducting polymers, making them ideal building blocks for nanoelectronic devices. Furthermore, CNTs can be organized in nanowire networks and with the addition of

(20)

19-recognition elements are ideal for sensing applications. Indeed, it was understood from the early 1990s that molecular and nanowire architectures could produce sensors with superior sensitivity, benefitting from the restricted transport along percolative paths and the large surface area-to-volume ratio.20 _{Hence, these principles were translated quickly to create the first example of}

SWCNT chemical sensors by Dai and coworkers.21 _{The authors strived to realize the concept of}

nanowire in its purest form by connecting electrodes with an individual SWCNT to observe the change in its conductivity when exposed to oxidative p-doping (NO2) and reductive un-doping

(NH3) gases. This study is certainly historic in the field of SWCNTs. However, similar to the onset of every area, much more progress was required to usher CNT platforms into versatile and useful

sensors.

It is indisputable that selectivity underpins the utility of any chemical sensor. Of course, sensitivity and stability must also be given the appropriate weight. The advancements in system integrations and electrical interfaces have lowered the stringent requirements of these latter parameters. For examples, electrical signals can be isolated and amplified, and trace analytes can be captured and released using pre-concentrators. However, without selectivity, the sensors are often rendered ineffective as a result of confounding effects in real-world environments such as interfering species, varying humidity, and fluctuation in temperature. Specificity, or perfect selectivity, is often not needed; and, robust sensors can be created from arrays of sensing elements, with each sensor having limited discriminating ability.2 2 _{Array-based sensors, such as}

a CNT-based chemical nose/tongue, are applicable to most types of chemical sensing and continue to progress toward the idea of a "universal sensor." Nevertheless, it is seldom a disadvantage to incorporate sensors that are inherently selective to the target analytes. SWCNTs are natural sensing materials as their transport properties are extremely responsive to their environment. They have suffered from limitation in selectivity at the inception of this field. Indeed, this limitation contributed to the relatively few commercial CNT-based sensors in spite of a

massive world-wide research effort.

(21)

-Approaches to increase selectivity predominantly include functionalization with selectors (e.g., polymer-wrapping and sidewall attachments). Quite often, the sensing performance of CNTs depends not only on the molecular recognition, but also on the response of the collective system, which can be affected by non-specific chemical, thermal, and mechanical interactions. The mechanism of chemical sensing may likely be intra-CNT and inter-CNT in nature; however, the other interfaces (CNT-electrodes and CNT-dielectric) must also be considered. In functionalization of SWCNTs, it was proposed initially that non-covalent attachments were preferred, as a result of the simplicity of the technique and the small perturbation on the base transport properties of the SWCNTs. Early applications of these methods to immobilized proteins appeared to give excellent performing biosensors2 3 However, detailed follow-up studies by the same researchers later revealed that the interfaces between the metal electrodes and the CNTs were non-innocent, and the interactions at these locations constitute the major responses for these sensors. Hence, if the primary response occurs at locations other than sites comprising

receptors/selectors, the sensor will lack predictably selective responses.

1.2 Chemical Sensing Mechanisms

The discussion on the exact mechanisms that cause the response of carbon nanotube-based sensors is very much alive. In contrast to conducting polymers, whose behavior can be described via molecular mechanisms, the properties of CNTs need to be described beyond local molecular structures, as is done in solid state physics. As a result of their extended rr-system, the frontier orbitals of CNTs are best described through band structures rather than discrete molecular orbitals. Accordingly, chemical intuition is often not sufficient when trying to predict or describe

CNT-based sensing mechanisms. This section covers several mechanisms that give rise to signals in CNT-based chemical sensors of different architectures and functionalization techniques. Responses of CNT-based sensors are attributed to effects arising within the tubes

(22)

-(intra-CNT), effects arising at contact points between tubes (inter-CNT), or effects due to the contact between the tubes and the electrodes (Schottky barrier modulations) (Figure 1-1). The strength of these different mechanisms can depend strongly on the analyte, the defect concentrations in the CNTs, and the device architecture.

Analyte

(a) Intra-CNT (c) Schottky Barrier

/Z

(b) Inter-CINT

Figure 1-1: Schematic of sensing mechanisms in CNT-based sensors: (a) at the sidewall or the length of the CNT (intra-CNT), (b) at the CNT-CNT interface (inter-CNT), and (c) at the interface between the metallic electrode and the CNT (Schottky barrier).

1.2.1 Intra-CNT

Intra-CNT sensing mechanisms are modes of interaction between analyte and individual nanotubes or nanotube bundles. They include changes in the number or mobility of charge carriers and generation of defects on the walls of the tubes.

Charge transfer induced directly or indirectly by analyte interactions will modulate the conductance of the CNT by changing (decreasing or increasing) the concentration of the majority charge carriers. Under ambient conditions, CNTs are p-doped as a result of physisorption of oxygen molecules on their surface. Thus, exposure to further p-dopants will increase the hole conduction and cause a decrease in the resistance, while n-type dopants will induce the reverse effect.21 22 _{Direct charge transfer between the analyte and CNTs has been identified as a major}

(23)

-sensing mechanism for polar analytes. 1 8_-30 _{In some cases this mechanism has a more localized}

nature. For example, interactions of a Lewis basic localized pair of electrons can create a local pinning force for cationic carriers as opposed to the fractional transfer of electron density to delocalized CNT states. For individual SWCNTs,2 1_' 263 1 _{charge transfer between analyte and tube} can be observed experimentally through the current-voltage (I-V) characteristics, photoemission spectroscopy (PES), and Raman spectroscopy.

Investigation of I-V characteristics through field-effect transistor (FET) experiments is a powerful tool for probing the sensing mechanism of CNT-based devices. When plotting the current through the CNT material as a function of the applied gate voltage (transfer curve), different sensing mechanisms induce characteristic changes. Adsorption of electron donating species (charge transfer to the tube from the analyte) induces negative charge in the CNT, thus n-doping the CNT and shift the threshold voltage towards a more negative gate voltages and vice versa (Figure 1-2a).21 _{Modulation of the metal/CNT junction induces asymmetric conductance change,}

as electron- and hole-conductions are affected differently (Figure 1-2b). Lastly, a reduction of the charge carrier mobility through charge carrier trapping or scattering sites induces a reduction in conductance (Figure 1-2c).3 2

,33 Any perturbation of the ideal SWCNT structure introduces charge scattering sites which reduce the mobility of the charge carriers and thus the conductance. Using I-V curves, changes in charge carrier mobility have been observed for scattering through adsorption of charged or polar species34-3 8or via deformation of the tube.39

(24)

-Electron Conduction C wU

/ c:zz

C w

b) Schottky Barrier Modulation IV-Curve 0.0 V, (V) c) Change in Mobility Hole Conduction C w Hole Conduction Electron Conduction I-Electron Conduction 0.0 IV-Curve 0.0

V,

(V)

0) I-a) C w 0)

Figure 1-2: Intra-CNT (semiconducting) sensing mechanism through changes in charge carrier concentration or mobility. Hypothetical transfer (l-Vg) curves and band diagrams before (black) and after (red) the exposure to the analyte for three different sensing mechanisms. The dotted line in the band diagram corresponds to the metal workfunction of the electrode and diagrams are given for both p- and n-type semiconductors interfaces with a metal. (a) n-Doping of the CNT induces a shift of the I-V curve to more negative voltages. The band diagram shows a hole doped CNT. (b) Schottky barrier modulation corresponds to a change of the barrier height between the work function of the metal electrode and CNT and asymmetric change in conductance for electron and hole transport. The band diagram shows change in barrier height for hole transport. (c) Change in mobility can be induced by the addition of resistive elements or carrier scattering which reduces the conductivity in both p- and n-type materials.

- 24 -0 C IV-Curve 0.0

V (V)

0.0 Hole Conduction a) Doping

(25)

The effect of charge transfer on the doping levels of CNTs can also be estimated from shifts in the Raman spectrum.- Wherein a shift of the G-band-stretching of the sp2 _{C-C bond}

in graphitic materials-towards higher wavenumbers is indicative of an electron-accepting analyte and a shift towards lower frequencies is indicative of electron-donating analyte (Figure 1-3). This shift can have a magnitude of 30 cm-1 for strong dopants and has been observed for inorganic40

and organic dopants.4546

e- donating e accepting

1450 1500 1550 60 1650

Wavenumber

cm-Figure 1-3: G-bands in the of Raman spectra of SWCNTs when interacting with electron donating and accepting molecules: (1) tetrathiafulvalene, (2) aniline, (3) pristine SWCNT, (4) nitrobenzene, (5) tetracyanoquinodimethane, and (6) tetracyanoethylene.

In addition to charge transfer effects, analytes can also promote the degradation of the CNT sidewalls. Particularly, the chemisorption of NO2 via formation of nitro- and nitrite-groups has been identified as a plausible sensing mechanisms.47-49 Soylemez et al.50 reported a

chemiresisitive glucose sensor based on poly(4-vinylpyridine) (P4VP) wrapped SWCNTs functionalized with glucose oxidase. Upon exposure to glucose, hydrogen peroxide is formed which oxidizes the SWCNT sidewall. The degradation of the conjugated sp2 network of pristine CNTs increases the number of defect sites of the SWCNT, also observable as an increased D/G peak intensity ratio of the Raman spectrum. Strong localized interactions associated with carrier pinning can manifest increases in the D/G peak ratios.

(26)

-1.2.2 Inter-CNT

For devices consisting of a network of CNTs, mechanisms at the interface between tubes can have significant influence on the electronic properties of the overall network. Small changes in distance between two CNTs dramatically influence the contact resistance as the probability of charge tunneling decreases exponentially with distance.5152 _{The inter-tube conduction pathways}

can be modulated either by partitioning of analytes into interstitial spaces between tubes or by swelling of the supporting matrix/wrapper. Alternatively, an analyte can trigger the disassembly of a molecular/polymer wrapping of the CNTs, Figure 1-4.

Swelling of a matrix material causes a decrease in the bulk conductance of CNT networks by increasing the width of tunneling gaps (Figure 1-4a). For example, Ponnamma et a.53 reported

the influence of swelling on the electronic properties of a MWCNT-natural rubber composite. The swelling index was determined by quantifying the equilibrium uptake of a given solvent for all tested composites. They reported that the swelling index correlates with the magnitude of the decrease in conductance for all tested samples. Similar sensing behavior has been observed for

porphyrins towards different VOCs,54 _covalently55

-56 and non-covalently57

.58 attached polymers

towards VOCs.

Alternatively, it is also possible to create sensing systems in which new conducting pathways are formed. Ishihara et al. reported the design of a sensor based on the de-wrapping of SWCNTs, which provided an increase of conductance by five orders of magnitude.59 _{In this} case the SWCNTs were wrapped with a metallosupramolecular polymer designed to depolymerize upon contact with an electrophilic analyte (chemical warfare agent mimic, diethyl chlorophosphate), the depolymerization caused the formerly isolated SWCNTs to come into electronic contact (Figure 1-4b). Similarly, Lobez et al.60 _{and Zeininger et al.}61 _{used CNTs} wrapped by poly(olefin sulfone) (POS) polymers to detect ionizing radiation. Upon exposure to

(27)

--i

radiation, the meta-stable POS spontaneously depolymerizes with fragmentation resulting in an increase in the interconnections between CNTs and overall CNT network conductivity.

a

in solvent vapor in dry air SWCNT ti

b

0 MSP

DECP %

Insulated network Conductive network

Figure 1-4: CNT-based chemical sensor designed based on inter-CNT mechanism. (a) Illustration of polymer swelling upon exposure of a CNT/polymer composite to solvent vapors. Reproduced with permission from Ref. 56. Copyright 2007, Elsevier. (b) Schematic illustration of a chemiresistive sensor comprising SWCNTs and metallosupramolecular polymer (MSP) showing the polymer degradation upon exposure to chemical warfare agent mimic diethyl chlorophosphate

(DECP).

1.2.3 Schottky Barrier (SB) Modulation

In certain cases, the device performance is not only influenced by intra- and inter-CNT effects, but also by modulation of the junction of metal electrode and CNT (Schottky barrier). To differentiate between the previous mechanisms and effects at the electrode/CNT interface, several groups have observed the sensing behavior with and without passivation of the

CNT/electrode contacts. In these experiments, passivating layers are deposited selectively over the whole device, over the areas where CNTs are in contact with the electrode, or over the length of CNTs that is not in contact with the electrodes (Figure 1-5).

(28)

-aCNT So,

b

C

SKO2

Si02

Figure 1-5: Three sensor architectures to probe Schottky vs intra-tube sensing mechanisms. Schematic for (a) device with bare CNTs, (b) device with passivated CNT-electrode contacts, and

(c) device with passivated length of CNTs that are not in contact with the metal electrode.

Bradley et al. contact-passivated different areas of a pristine CNT-device with SiO2 and tested the response of the resulting sensors towards NH3.62 They found that complete coverage

with SiO2 drastically attenuated the response to NH3 exposure, proving SiO2 a suitable passivating material. Coverage of the electrode/CNT contact areas resulted in a sensor with comparable responsiveness and faster reversibility than the non-passivated sensor. From this result, they concluded that the NH3 sensing mechanism is the result of processes occurring over the length

of the CNT, and not at the CNT/electrode interfaces. Liu et al. used poly(methyl methacrylate) (PMMA) to passivate the channel of the electrode contact areas of a CNT-FET and detect NO2

and N H3.63 In contrast to Bradley et al., they observed changes in the transfer characteristics for both channel- and electrode-passivated devices suggesting that effect on both the metal/CNT junction and the length of the CNT have significant contributions to the sensor signal. Zhang et

al. also employed PMMA to passivate the contact of a CNT device used to detect NO2, however

they found that the sensing response is mainly due to the interface between electrode and CNT.64 Similarly, Peng et al. used devices partially passivated by Si3N4 and found that the sensing

(29)

-response towards NH3 mainly results from the metal-CNT junction.65 Considering the

inconsistencies between these reports, it is understandable that the debate on the sensing mechanism persists.

To clarify the contradictory findings just discussed, Salehi-Khojin et al. investigated the sensing behaviors of CNTs with different defect levels.66 They reported that the dominating sensing mechanism is strongly dependent on the bottlenecks in the conduction pathways. For highly conductive CNTs, the sensing behavior is dominated by mechanisms influencing the electrode-tube junctions while the response of devices with less conductive, defect-rich CNTs is dominated by intra-tube effects. Generalization remains difficult as a result of the fact that there can be different types of defects in CNTs and other components in the sensor material can influence intra-tube and metal-tube electron transport.

Apart from the nature of the CNTs, the choice of metal for the electrode also influences the behavior of the CNT/electrode junction. Kim et al. reported increased sensing responses from CNT devices containing Pd instead of Au electrodes. 67 This result was attributed to the stronger

interactions between the Pd surface and CNTs and the resulting barrier-free electronic transport across the junction, which agreed well with several theoretical studies.8_'6 9 _{Zhang et al.} demonstrated that targeted disruption of this Pd/CNT junction is a useful approach to fabricate H2

sensors using CNTs without any further functionalization.70

1.3 Performance Parameters

The promise of practical chemical sensors has motivated the constantly expanding research on CNTs. All analytical methods must possess the capability to provide quantitative, or in some cases, qualitative measurements. Chemical sensors, by definition, are devices with the

(30)

-capability to recognize and transduce the chemical information of the samples. In the simplest cases, the chemical information is the analyte concentration, and the read-out is a change in a readily measured signal. Ideally, a sensor is sensitive, selective, and stable. Achieving these figures of merit continues to be a challenge for the development of all sensors. Furthermore, chemical sensors are an alternative to large equipment in analytical lab and it is implicit that they are inexpensive, physically small (ideally portable), and robust under field conditions. CNT-based sensors are viewed as being well suited for these requirements. Additionally, qualities such as reproducibility, rapid response times, and low drift are demanded for practical sensors. This section will introduce and provide brief descriptions of the relevant figures of merit and how they are measured. This section will conclude with a discussion of challenges arising from the properties of CNTs.

1.3.1 Parameters of Performance and Figures of Merit: What they are and how to Measure them?

The sensing response curve describes how the devices response to the exposure of the analyte as a function of time. Figure 1-6a illustrates a hypothetical curve when the device is exposed to increasing concentration of the analyte. The choice of the architecture of the sensing device will govern the type of signal measured by the sensor. The sensing traces are often reported as the relative changes in measured resistance (R), current (/), conductance (G = / / V;

the symbol is not to be confused with Gibb's free energy), capacitance (C), power gain (RFID), and resonant frequency (fo) of the device vs. time. Different conventions have been adopted for plotting these measurements-most popularly normalized differences AX/Xo, X/Xo, or simply AX (where X = R, I, G, C, or gain). For example, the change in conductivity (-AG/Go) is calculated by observing the normalized difference in the current before (lo) and after (/) the exposure to the analyte using equation 1-1:

-Z (%) =

['

x 100 (eq. 1-1)

(31)

-0 0.0 b Time U) b

0

Calculated L D Sensitivity Dynamic range Concentration

Figure 1-6. Hypothetical sensing response curve (a) and calibration curve (b) of a sensor exposed to increasing concentration of the analyte.

The calibration curve (Figure 1-6b) shows the relationship between the response of the

sensor and the concentration of the standard solution.71 _{The standard solutions for the analyte}

should be carefully chosen to cover the relevant concentration range. Sensitivity is defined as the capability to discriminate small differences in concentration or mass of the analyte. In other words, sensitivity of the sensor is the slope of the calibration curve. The range of the concentration of the standard solution measured for the calibration curve constitutes the dynamic range; the range of which the signal is linearly proportional to the concentration is the linear range.72

Limit of detection (LOD) is the lowest amount of an analyte in a sample which can be

detected with reasonable certainty. The theoretical or calculated LOD, as established in literature,7_3-5 _{is determined as the concentration that corresponds to the signal at three standard} deviations of noise above the baseline, using the calibration curve. Briefly, the root-mean-square noise (rmsnoise) of the sensors is first calculated as the deviation of the sensing response curve from the appropriate polynomial fit. The LOD is then calculated using equation 1-2:

LOD = 3 * rmSnoise (eq. 1-2)

slope

(32)

-The slope in Equation 2 is the linear regression fit of the sensor response vs. concentration curve (slope of the calibration curve). The drive for achieving lower LOD is often governed by application-dependent requirements. These values for environmental safety are published by regulatory agencies. The ability to detect low concentration of the analyte is often coupled dynamic range of the sensors which describes the range of concentration the sensor can be calibrated. Sensors with low LOD often deviate from linearity at high concentrations as a result of saturation.

Selectivity of a single sensor is defined as the ability to discriminate the analyte of interest

from other species within the sample.76 _{Specificity is also used to express selectivity: in a literal} interpretation a specific sensor recognizes only the target analyte and no other compounds. Such ideal sensors are rare as a result of similarity between analytes and lack of perfect molecular recognition. Cross signaling, also known as cross reactivity, between sensors and analytes leads to a compounding of the signals and loss of selectivity. Selectivity is measured from the signals arising from the analyte and the possible interfering compounds separately with the assumption that there are no synergistic cooperative effects of multiple analytes and the sensor. The calibration curves (signal vs. concentration) of each is then compared. The ratio of the signals of the analyte to those of the interfering compounds defines the selectivity of the sensor. This method is operationally simple and generally adequately quantifies the desirable signal relative to possible cross signaling. However, it may overlook the competing effects between different compounds and the analyte. The second method comprises replicating the matrix (simple or complex) of the real-world samples in which the analyte is mixed with expected interfering compounds. In such case, the selectivity of the sensor can then be determined by the differences in signal of the analyte with and without the interfering agents.

Stability is defined as the capability of the sensors to produce repeatable outputs for an

identical environment over time. To measure the stability of the sensors, repeated measurements should be taken over time or over many cycles of exposures to the analyte. In a more methodical