Rapid industrial scale synthesis of robust carbon nanotube network electrodes for electroanalysis

surprisingly


Introduction
Since Iijima reported his findings about carbon nanotubes (CNT) in 1991 [1], this material has been extensively investigated for various applications [2] due to its exceptional material properties [3][4][5].Particularly in the field of electroanalytical applications, CNTs are widely used in detection of different biomolecules [6][7][8][9].However, despite extensive amounts of research, often these electroanalytical studies lack comprehensive characterization of CNTs physicochemical properties [10][11][12].CNTs properties heavily depend on the synthesis parameters as well as possible purification steps in order to remove residual metal catalyst particles and carbonaceous impurities [5,13].As elec-trochemical performance of a carbon electrode is affected by several factors such as microstructure, surface cleanliness and surface chemistry [14], it is evident that CNTs with different fabrication conditions can exhibit divergent electrochemical properties.This is especially true in detection of inner-sphere (ISR) analytes, which are sensitive to the reactive sites at the electrode surface such as surface oxides [15][16][17] and defects [18].
From a commercial aspect, it is very important to maximize the productivity i.e. yield of a single CNT reactor, so that large quantities of for example electrochemical sensor strips can be manufactured with minimal costs.However, this has to be done without compromising the analytical performance and repeatability of the sensors.A recent studies by Wester et al. [19] and Verrinder et al. [20] used an industrial-scale aerosol chemical vapor deposition (CVD) dry transfer method to demonstrate a simple and inexpensive process for the production of disposable electrochemical test strips for detection of drug molecules.This process, involving the dry transfer of single-walled carbon nanotube (SWCNT) patterns directly onto polymer substrates comprising two SWCNT electrodes and a screen printed silver reference electrode, realizes a high throughput, roll-to-roll compatible, industrially mature fabrication method for the commercialization of disposable CNT-based test strips for quantitative point-of-care (POC) testing from finger-prick whole blood samples.
To explore if it is possible to produce high quality CNT network electrodes rapidly, these same aerosol CVD reactors were used in this work to collect two SWCNT samples at different collection rates, namely Low-Rate and High-Rate, by varying the ferrocene cartridge temperature.To our knowledge, the effects of different SWCNT collection rates or such process conditions in general have not been studied towards the electrochemical performance of SWCNTs in detail.
In order to correlate electroanalytical behavior of carbon nanotubes with their physicochemical properties and synthesis parameters, a thorough physicochemical characterization is required.First, physical characteristics of the SWCNT networks were investigated using scanning electron microscopy (SEM), four point probe sheet resistance measurements and conductive AFM by PeakForce Tunneling (PF-TUNA) method, as well as by ultraviolet-visible (UV-vis) and Raman spectroscopy.SWCNT network structure and morphology were further visualized in higher detail by high-resolution scanning transmission electron microscopy (HRSTEM) and TEM tomography.By utilizing the latter technique, 3 dimensional reconstructions of the CNT networks were realized that provided direct visual access to the network structure.Then, the chemical composition was investigated by X-ray absorption and X-ray photoelectron spectroscopies.Finally, these physicochemical properties were correlated with the electrochemical performance obtained from cyclic voltammetry (CV) and differential pulse voltammetry (DPV) experiments by utilizing both outer-sphere (OSR) redox couples, including Ru(NH 3 ) 6 2+/3+ , IrCl 2−/3− and Fe 2+/3+ , as well as ISR probes.Here morphine (MO), oxycodone (OXC) and paracetamol (PA) were used as a benchmark ISR systems as we have previously studied in-depth SWCNTs suitability in detection of these analgesics [19,[21][22][23].

SWCNT network fabrication
Single-walled carbon nanotube networks were fabricated in a laminar flow reactor using a high-temperature floating catalyst chemical vapor deposition.Briefly, in this process a ferrocene precursor is injected to the reactor chamber that is kept in carbon monoxide atmosphere.Due to the high temperature, ferrocene thermally decomposes to iron particles.These Fe particles catalyze the growth of SWCNTs in the gas phase by decomposing carbon monoxide in the chamber.In the end of the process SWCNTs are collected using a membrane filter.More detailed information about the fabrication process can be found from [24,25].
Here two SWCNT samples were fabricated at different collection rates, 100% and 500%, by modifying the amount of ferrocene available in the gas phase.The amount of iron can be controlled by variating the ferrocene cartridge temperature as it will change the vapor pressure of ferrocene.Collection time was adjusted respectively by in-situ optical rate absorbance monitoring resulting networks with the same thickness.SWCNT network synthesized with slower collection rate is referred as Low-Rate SWCNT, whereas the sample with collection rate of 500% is referred as High-Rate SWCNT.
Region of interest (ROI) i.e. an uniform planar layer covering a suitable hole in the MultiA-holey-carbon grid was selected, using JEM-2800 electron microscope (JEOL).Testing 200 kV TEM, STEM and 80 kV STEM were not suitable due to escalating fogginess in collecting tilt-series procedure.However, tilt-series were automatically successful collected at 80 kV TEM mode with RECORDER application (TEMography) with following settings: Goniometer tilts spanning −72 to + 72 degree, increment steps 2 degree, CCD (GATAN), ORIUS SC200, 2048x2048 camera, pixel X,Y scale 0.262017 nm, at magnification of 400 k.
For tomography, maximum entropy method (MEM) [26,27] was used for 3D reconstruction of aligned.alistacks.Before running tomography, aligned.alistacks were 2 × binned and ROI cropped with scripts initiated in the Aalto Nanomicroscopy Center laboratory.MEM-3D-reconstruction volumes produced were visualized and scrutinized, in stereo mode, using UCSF CHIMERA application.Programs were all run in Mac Pro, Cylinder model, Twelve Core (64 GB).

Scanning transmission electron microscopy (STEM)
Both SWCNT networks were characterized with high-resolution STEM.The samples were prepared by press-transferring a piece of SWCNT network onto a gold-copper TEM grid (Agar Scientific).High-resolution STEM was performed with JEOL JEM-2200FS double aberration corrected microscope with X-ray EDS detector operated at 200 kV.The STEM images were recorded with Gatan 4 k × 4 k UltraScan 4000 CCD camera.

Field emission scanning electron microscope (FESEM)
FESEM imaging of self-standing non-densified samples was performed on a Hitachi S-4700 SEM at 15 kV accelerating voltage.Open-to-close -ratios were computed from images by grey-scale thresholding to binary color in ImageJ.

X-ray absorption spectroscopy (XAS)
The soft X-ray XAS experiments were conducted at the Stanford Synchrotron Radiation Lightsource (SSRL) beamline 8-2, where a bending magnet was used with a 55 degrees incidence angle (magic angle) of X-rays.A resolution of ~200 meV was achieved with a spherical grating monochromator which was operated using 40 × 40 μm slits.X-ray beam spot size was approximately 1 × 1 mm 2 with total flux in order of 10 10 photons/sec.The X-ray energy for iron 2p edge was measured from 695 eV to 735 eV, whereas for carbon and oxygen 1s edge were from 260 eV to 340 eV and 520 eV to 560 eV, respectively.A Keithley picoammeter was used for amplifying the drain current to collect all the data in total electron yield (TEY) mode, where the incoming flux was measured using a nickel grid coated with Au sputtered film.Here, a reference sample was used for energy calibration of the data prior to the data analysis.The C 1s spectra were confirmed to match their energy calibration by observing the core-exciton signature at 291.65 eV [28][29][30] after the reference sample energy alignment.The presented C 1s and Fe 2p spectra are normalized, where the Fe 2p spectra were energy aligned to match metallic iron signature at 707.36 eV [31].The O 1s shows absolute intensity of the measurement where energy alignment is based on a reference sample, calibrating carboxyl at 288.75 eV and Ni L3 in 2nd order at 426.35 eV prior to measurement.The presented C 1s, Fe 2p and O 1s spectra in Fig. 3 shows the average value from three different locations.Furthermore, all the data were background subtracted and energy corrected using IGOR Pro v. 8.02 software.

X-ray photoelectron spectroscopy (XPS)
XPS was performed using an Axis Ultra electron spectrometer (Kratos Analytical) with monochromatic Al Kα irradiation at 100 W under neutralization.Before analysis, pre-evacuation was carried out overnight.High-resolution spectra of C 1s, O 1s, Fe 2p as well as survey spectra of 3-4 locations were recorded for each sample.The data were fitted with CasaXPS software, assuming Gaussian line shapes.100% filter paper (Whatman) was used as an in-situ reference for charge correction [32].

Ultraviolet-visible spectroscopy (UV-vis)
Ultraviolet-visible spectroscopy was carried out on self-standing film samples in the range of 180-1400 nm with 0.5 nm/s scan rate on a Shimadzu UV-2600 spectrophotometer.

Raman spectroscopy
Raman spectroscopy was performed by a Horiba Jobin-Yvon Labram HR confocal Raman system with an 488 nm argon laser with 10 mW power on sample.Spot size of 1 μm was used with an Olympus 100x objective.Spectra were acquired in the range of 50 to 3000 cm −1 with a 600 lines/inch diffraction grating, exposure time of 15 s, and accumulation averaging count of two.Spectroscopic calibration was performed on intrinsic Si wafer (Ultrasil).Spectra were fitted by one Lorentzian peak for D-band, and two Lorentzian peaks for the G-band (G + and G − ), to obtain the I D /I G peak intensity ratios, as explained in literature [33].

Conductive AFM
PeakForce Tunneling AFM (PF-TUNA) was carried out to map both topography and contact current of the SWCNT networks using Bruker Dimension Icon.The instrument was operated in tapping mode with a conductive silicon nitride probe coated with platinum and iridium (PFTUNA probe, Bruker).A probe voltage of 100 mV was used for each measurement.Average contact current was calculated from three different 3 × 3 um areas on each sample.

Sheet resistance
Sheet resistance was measured on glass substrates using a Jandel RM3000 multi height probe.

Electrochemical measurements
Cyclic voltammetry experiments were conducted with Gamry Reference 600 + potentiostat, whereas CH Instruments (CHI630E) potentiostat was used for differential pulse voltammetry.All measurements were carried out in a three-electrode cell, where Ag/AgCl (+0.199V vs. SHE, Radiometer Analytical) was used as a reference electrode and Pt wire (Goodfellow) as a counter electrode.Briefly, the SWCNT electrodes for electrochemical experiments were prepared by press-transferring a smaller piece of the network onto a glass substrate (Thermo Scientific, ISO 8037/I) and electrical contact was prepared with conductive silver paint (Electrolube).The sample was covered with inert PTFE-tape (Saint-Gobain Performance Plastics CHR 2255-2) with a 3 mm hole (radius = 1.5 mm), which was placed on top of the SWCNT network to define the working area of the electrode.Further details from the SWCNT electrode preparation can be found from [23].
The CV measurements were carried out in 1 mM Ru(NH 3 ) 6 2+/3+ , 1 mM IrCl 3−/2− and 1 mM Fe 2+/3+ , where hexaamineruthenium (III) chloride (Sigma-Aldrich) and potassium hexachloroiridate(IV) (Aldrich) were dissolved in 1 M KCl (Merck Suprapur, pH 6.8), whereas ammonium iron(II) sulfate hexahydrate in 0.2 M HClO 4 (VWR Chemicals, pH 1).In addition, 0.2 M HClO 4 was used as an inert supporting electrolyte for defining the potential windows of both Lowand High-SWCNT networks.Morphine hydrochloride (MO), oxycodone hydrochloride (OXC) and paracetamol (PA) were measured with DPV.PA was purchased from Sigma Aldrich, whereas MO and OXC from the University of Pharmacy, Helsinki.MO, OXC and PA were dissolved in 10 mM phosphate buffer saline (PBS, pH 7.4).All DPV experiments were performed in a potential window from −0.4 V to 1 V vs. Ag/AgCl with following parameters: step size 4 mV, sample period 0.2 s, pulse time 0.5 s and pulse size 50 mV.In order to stabilize the background current of the SWCNT electrodes, six blanks were measured for each electrode in PBS before conducting DPVs in analgesics.
All measurements were carried out inside a Faraday's cage at room temperature.New electrodes were used for every electrochemical experiment.In addition, all solutions were purged with N 2 for 30 min prior to the experiment.

Structural and chemical analysis
Two SWCNT networks of the same optical transmittance were collected at widely different rates by varying the ferrocene cartridge temperature.By considering the drastic differences in the collection rate between the two investigated samples, Low-and High-Rate SWCNT, it is surprising to find their physical characteristics practically identical within the experimental error.To begin with, the spectral shapes of each sample in UV-vis spectroscopy are practically indistinguishable (see Supporting Information, Figure S1).The UV-vis spectra are used here as a measure of the amount of CNTs on sample: transmittance at 550 nm is similar in both samples (Table 1), since the collection times were controlled by in-situ optical absorbance.
For Raman spectra, the I D /I G ratios as fitted from Fig. 1A, and tabulated in Table 1, show a small increase in the amount of defects from Low-Rate to High-Rate, but these values are still notably low, and indicate only small numbers of defects in either sample [33].The two samples exhibit similar sheet resistances (Table 1) and comparable PF-TUNA conductivities (Figure S2), where the averages of measurements fall within one standard deviation of each other.It should be noted that while the former measures the resistance between opposite sides of a square of a material (rather than the bulk resistance), the latter mainly probes the z-axis conductivity through the network.When compared to the respective UV-vis %-T at 550 nm, these sheet resistances are noticeably consistent with similar work by Reynaud et al. [34] Lower magnification SEM images are provided in Fig. 1B and 1C for assessing the overall network structure.To estimate network density, open-to-close -ratios of the CNT networks are calculated by grayscale to binary conversion from SEM images and tabulated in Table 1.
The three-dimensional reconstructions from TEM tomography for the both SWCNT networks are presented in Fig. 1D and 1E.Videos for the constructions can be found from the extra material provided with this publication.Previously there have been only a few electron tomography studies from carbon nanotubes, but these only include 3D reconstructions from individual multi-walled carbon nanotubes [35] or composites [36][37][38][39][40].To our knowledge electron tomography reconstructions of single-walled carbon nanotube networks haven't been published before.
From Figs. 1 and 2, it can be observed that in both networks the SWCNTs exist mainly in bundles.The following main observations can be made: (i) High-Rate SWCNT network appears overall less dense and (ii) the average size of the bundles appears to be markedly larger in High-Rate sample when compared to Low-Rate SWCNT.From a process aspect point of view, higher iron precursor rates can increase the amount of carbon nanotube nucleation sites in the aerosol reactor flow.This higher density of CNTs would promote in-flight agglomeration i.e. the bundle size would increase before the CNTs are collected on filter membrane.This can be correlated with the slightly lower PF-TUNA conductivity value since there would be less bundle-to-bundle connections for high network interconnectivity -especially into zdirection (through the network).This can be rationalized as follows: (i) if we assume that the network conductivity is mainly determined by bundle to bundle contact resistance (as the typical lengths of the bundles here are in the range of 20-30 µm with rather wide distribution) and (ii) that the contact area is independent of the bundle diameter (bundles are rigid and do not deform) leads to the assumption that (iii) the network resistance is proportional to the number of contacts between the bundles, which can be expected to be somewhat less in the case of High-Rate SWCNTs with larger bundle size.This chain of arguments is consistent with the results given in [41].However, it must be noted that there exist also opposite views arguing that the contact resistance would actually decrease as a function of increasing bundle diameter [42].In addition, as sheet resistance values appear to support this opposite trend of decreasing network resistance with increasing bundle diameter (even though sheet resistance set up measures different variable than PF-TUNA), further studies are required to settle this issue.
As morphology of the SWCNT networks can vary within the sample, further structural investigation was carried out with high-resolution scanning transmission electron microscopy (STEM).The darkand bright-field micrographs (Fig. 2) shows unambiguously that the density of the Low-Rate SWCNT sample is higher which supports the observation made from the 3D reconstructions.Interestingly, the average size of the Fe catalyst particles appears to be smaller and more evenly distributed in Low-Rate SWCNT i.e. the effective surface area of Fe particles is larger when compared to High-Rate SWCNT.This  can be seen from both 3D reconstruction as well as the STEM micrographs.Furthermore, some of the Fe particles in Low-Rate (Fig. 2F) are encapsulated with graphitic layers which seems to be the case with High-Rate SWCNT (Fig. 2C) as well.Chemical analysis of the samples was carried out by measuring C 1s, Fe 2p and O 1s spectrums with XAS and XPS.The results are shown in Fig. 3 and Table 2, where the elemental compositions are calculated from XPS high-resolution spectra (see Figure S3).Interestingly, with these methods as well, no significant differences are observed between the investigated samples.
Quantitative analyses from the XPS shows that the content of Fe is identical in the network (Table 2) with both collection rates.However, it should be noted that the total surface area of Fe is most likely higher in Low-Rate as the average size of the Fe particles is smaller (Figs. 1  and 2).Furthermore, as XPS is a surface sensitive method, we cannot unambiguously define the exact amount of iron in the networks.Thermogravimetric analysis would provide more accurate information of Fe content in the samples.However, here in this research we did not have the opportunity to use this method.
From the XAS Fe 2p spectra (Fig. 3B) a small but clear difference was observed in the composition of iron.With High-Rate SWCNT network only one peak is observed at the L 3 -edge.This feature arises from metallic Fe [31,48], although a relatively similar shape is seen for FeO in the work of Regan et al. [31].However, based on the Fe-O phase diagram, the thermodynamically most feasible structure below 570 °C with low oxygen loading is metallic Fe.In case of Low-Rate SWCNTs, a smalls houlder-like feature is observed at the higher energy in the L 3 -edge due to carbide and/or oxidized phase of iron [31,48,49].For oxidized phases of iron, Fe 3 O 4 and Fe 2 O 4 , intensity of the oxide peak at L 3 -edge is considerably higher than the metallic peak [49][50][51].As this is not the case here, it is very likely that the signal arising from the iron oxides are masked by other compositions of iron.Thus, authors suggest that the most dominant Fe phases in Low-Rate SWCNT sample are metallic iron and iron carbide.
The higher O 1s oxygen content in High-Rate SWCNT observed with XPS (Table 2 and Figure S3) is likely due to a larger SiO 2 area exposed on the silicon wafer substrate, consistent with higher Si 2p at-%.This is aligned with SEM observations in Fig. 1.Thus, based on these results we cannot unambiguously determine if the oxygen loading varies between the samples.Similar challenge is observed in the XAS O 1s spectra, where the dominating feature at ~535 eV mainly arises from the Si substrate (Figure S4).Although no clear signature features are seen in XAS O 1s, it can be still concluded that some difference is observed between the samples as a small peak is seen before 530 eV in Low-Rate SWCNT.In this case, oxygen could be bound either to carbon and/or iron.However, it should be noted that the XAS C 1s spectra is nearly identical for both samples i.e.C 1s doesn't support that the oxygen containing surface groups of Low-Rate would differ from High-Rate sample.
The authors find these results interesting and surprising.The used aerosol CVD reactor with two widely different collection rates produces SWCNT networks with nearly similar structural and chemical properties.The only distinct differences observed between the samples was the bundle size and the density of the network, which may affect the surface area available in for example electrochemistry.As these features were clearly observed due to the 3D reconstructions, here we highlight the importance of TEM tomography for structural analysis.

Electrochemical characterization
The potential windows (Fig. 4A) in HClO 4 with the threshold current limits of ±20 µA are 2.49 ± 0.02 V and 2.54 ± 0.05 V for Low-and High-Rate SWCNT, respectively.Only minor differences between samples are seen at the cathodic limit.Interestingly, with both SWCNT networks a surface reaction is observed during the 1st cycle at the cathodic scan ~0.2 V after oxidizing the sample at 1.3 V vs. Ag/AgCl.It should be noticed that this phenomena is only observed clearly for one cycle after oxidizing the surface (see supporting information Figure S5).The observed reaction is most probably arising from the Fe particles in the SWCNT network as a similar phenomena is seen with pure Fe wire (Figure S6).Based on Pourbaix diagram of Fe [52] it is likely that it is related to the iron dissolution process and subsequent passivation of the Fe surface.
The electron transfer kinetics of both SWCNT electrodes were characterized utilizing outer-sphere redox couples Ru(NH 3 ) 6 2+/3+ and IrCl 3 . These probes are insensitive to the surface chemistry of the electrode and therefore excellent for investigating the electronic structure of the material.Results for both OSR probes are presented in Fig. 4B, 4D and Table 3 where it can be observed that both SWCNT networks exhibited quasi-reversible behavior within the standard deviation.In Ru(NH 3 ) 6 2+/3+ the potential peak difference ΔE p of High-Rate SWCNT is slightly higher when compared to Low-Rate SWCNT network, whereas the opposite case is observed in IrCl 2−/3− .With increasing scan rate ΔE p increases as well in both OSR probes with both samples.However, in IrCl 2−/3− increase of scan rate does not have a significant effect and less variance is seen between the SWCNT networks when compared to Ru(NH 3 ) 6 2+/3+ .This could be due to the electrostatic interactions between SWCNTs surface and the OSR probes with different charges.Point of zero charge (pzc) of SWCNTs in pH 1 is 0.5 V vs. Ag/AgCl (Figure S7).In KCl the value is less, most probably close to 0.1 V, as pzc depends on the pH of the electrolyte.Thus, it is obvious that the surface charge of SWCNTs is very different for Ru(NH 3 ) 6 2+/3+ at cathodic potentials than for IrCl 2−/3− at higher anodic potentials.Interestingly, the electrochemically active surface areas of Highand Low-Rate SWCNT in Ru(NH 3 ) 6 2+/3+ as well as in IrCl 2−/3− are approximately the same as the current values are nearly identical although it could be expected that Low-Rate SWCNT would have more surface area as density of the network is higher with smaller bundle size.However, evaluating the electrochemical activity of the SWCNTs is not that straightforward as several factors can alter such as defects in the structure, edge and plain sites [18], bundle-to-bundle connections [53] as well as the amount and distribution of Fe catalyst particles [54].Generally, it is expected that with CNTs the electrochemically active area is higher than the geometric area.However, here in case of OSR probes the effective electrochemical area is actually smaller than the geometric area (Table S1).The effective electrochemical area A was obtained from Randles-Ševčík equation: where i p is the peak current, n the number of electrons, F the Faraday constant, C the concentration of OSR probe, v the scan rate, D the diffusion coefficient, R the gas constant and T the temperature.With Fe 2+/3+ a clear difference is seen in electron transfer kinetics between Low-and High-Rate SWCNT networks.It is suggested that Fe 2+/3+ couple kinetics is accelerated by the increase of oxygen containing functional groups, especially carbonyl groups, on the electrodes surface [14].Previously it has been shown with highly ordered pyrolytic graphite (HOPG) [15] and glassy carbon (GC) [17] that electron transfer of Fe 2+/3+ enhances after electrochemical oxidation treatment.However, here both Low-and High-Rate SWCNT are as-fabricated pristine networks.For both samples Fe 2+/3+ exhibits quasi-reversible behavior, but for High-Rate SWCNT electron transfer kinetics is clearly more sluggish.As no remarkable changes are seen on XAS C 1s and O 1s spectra, we suggest that with SWCNTs Fe 2+/3+ kinetics could also depend on other factors such as (i) morphology, (ii) Fe catalyst particles composition [23] and (iii) the distribution and thus the effective surface area of Fe particles as seen from (S) TEM analysis.With Low-Rate SWCNT Fe 2+/3+ kinetics may be accelerated by one or perhaps all of the factors.
The effects of SWCNT network morphology on detection of analgesics was studied with morphine, oxycodone and paracetamol (Fig. 5) with DPV where i p for MO is determined from the first oxidation peak at ~0.4 V vs. Ag/AgCl [21,55].Interestingly, similar trends are observed in all three analytes: (i) the oxidation potential E p,a shifts to anodic direction and (ii) the peak current i p value decreases when the collection rate of SWCNTs during synthesis is increased to 500% i.e. density of the network is lower with larger bundle size.Interestingly, the most clear changes are seen with OXC, where i p reduces approximately to half and E p,a shifts ~100 mV with High-Rate SWCNT.Thus, these results indicate that SWCNTs sensitivity towards oxycodone can be altered by modifying the network morphology.As discussed in Fe 2+/3+ results, changes in the Fe composition as well as the increased effective area of Fe particles may catalyze the detection of OXC.In case of MO and PA structural changes in SWCNT network show less significant effect on the sensitivity of these analytes., IrCl 2−/3− and Fe 2+/3+ with scan rates 100 mV/s and 1000 mV/s.For all measurements N = 3.However, it is clear that these observed changes alter on the interactions with morphine in some level as the second oxidation reaction around 0.8 V is diminished with High-Rate SWCNT.Similar effect is seen when Fe containing SWCNT network is electrochemically oxidized [23].Furthermore, the electrochemical experiments in analgesics show that the selectivity towards MO, OXC and PA can be tailored by changing the amount of iron precursor in the fabrication process.Observed shift of E p,a to anodic direction could be due to stronger absorption of MO/OXC/PA (the reactant) on the surface of High-Rate SWCNT and subsequent surface reaction of the adsorbed species.
Although we cannot unambiguously explain the exact interactions of the analgesics with SWCNTs, these results demonstrate that in electrochemical detection of inner-sphere analytes, physicochemical properties of single-walled carbon nanotubes should be optimized for the analyte of interest.On the other hand, from the industrial point of view we have shown that with faster growth rate (High-Rate SWCNT) it is possible to fabricate good quality single-walled carbon nanotubes in large-scale without any drastic changes in the physicochemical properties when compared to Low-Rate SWCNT.In fact, without TEM tomography these minor changes in the network structure would not have been observed.

Conclusions
In this study, we demonstrated the effect of synthesis rate on the physicochemical properties of SWCNTs.Interestingly, the industrialscale process with two drastically different synthesis rates produced SWCNT networks with very similar physical and chemical properties.The only clear difference was observed in the network density as the higher synthesis rate produced a network with larger bundle size and lower density.Additional minor change was seen in the composi-tion of iron catalyst particles.With High-Rate SWCNTs Fe particles mainly exist in metallic form, whereas Low-Rate SWCNTs also contain carbide and iron oxides.
Throughout extensive electrochemical characterization we showed that synthesis rate did not have significant effects on the network electrical properties as both samples exhibited similar and fast kinetics with both outer-sphere probes Ru(NH 3 ) 6 2+/3+ and IrCl 2−/3− .However, the observed changes in the SWCNT networks did affect the electrochemical detection of surface sensitive inner-sphere analytes.Sensitivity towards all analgesics, especially oxycodone, was enhanced when the density of the network increased.In addition, with slower synthesis rate the oxidation potential of morphine and oxycodone shifted to cathodic direction.Thus, these results provide a proof of principle that selectivity towards these analytes can be modified by altering the morphology of the SWCNT network.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Table 1 Fig. 1 .
Fig. 1.A) Raman spectra normalized to G peak intensity and offset for clarity.SEM micrographs from non-densified self-standing films and the TEM 3D reconstructions illustrating the structure of B,D) Low-Rate and C,E) High-Rate SWCNT networks.

Table 3
The average values of ΔE p and peak current ratios in Ru(NH 3 ) 6