Filiberto
Ricciardella
*a,
Sten
Vollebregt
a,
Tiziana
Polichetti
b,
Mario
Miscuglio
cd,
Brigida
Alfano
be,
Maria L.
Miglietta
b,
Ettore
Massera
b,
Girolamo
Di Francia
b and
Pasqualina M.
Sarro
a
aDelft University of Technology, Faculty of Electrical Engineering, Mathematics and Computer Science, Department of Microelectronics, Delft, Feldmannweg 17, 2628 CT Delft, Netherlands. E-mail: filiberto.ricciardella@gmail.com
bENEA – Materials and Devices Basic Research Laboratory, Piazzale Enrico Fermi, 1, I – 80055 Portici (Napoli), Italy
cItalian Institute of Technology, Nanochemistry Department, Via Morego, 30, I-16163 Genova, Italy
dUniversity of Genova, Department of Chemistry and Industrial Chemistry, Via Dodecaneso, 33, I-16146 Genova, Italy
eUniversity of Napoli “Federico II”, Department of Physical Sciences, Via Cinthia, I-80126 Napoli, Italy
First published on 5th April 2017
The crystal structure of graphene flakes is expected to significantly affect their sensing properties. Here we report an experimental investigation on the crystalline structure of graphene aimed at exploring the effects on the gas sensing properties. The morphology of graphene, prepared via Chemical Vapor Deposition (CVD), Liquid Phase Exfoliation (LPE) and Mechanical Exfoliation (ME), is inspected through Raman spectroscopy, Atomic Force Microscopy (AFM) and Scanning Electron Microscopy (SEM). CVD and LPE-graphene structures are found to be more defective with respect to ME-graphene. The defects are due to the jagged morphology of the films rather than originating from intrinsic disorder. The flatness of ME-graphene flakes, instead, explains the absence of defects. Chemiresistors based on the three different graphene preparation methods are subsequently exposed to NO2 in the concentration range 0.1–1.5 ppm (parts per million). The device performance is demonstrated to be strongly and unambiguously affected by the material structure: the less defective the material is, the higher the response rate is. In terms of signal variation, at 1.5 ppm, for instance, ME-graphene shows the highest value (5%) among the three materials. This study, comparing simultaneously graphene and sensors prepared via different routes, provides the first experimental evidence of the role played by the graphene level of defectiveness in the interaction with analytes. Moreover, these findings can pave the path for tailoring the sensor behavior as a function of graphene morphology.
In this work, we present the first comparative study on Gr synthetized via CVD, LPE and ME aiming at investigating the role played by the Gr crystalline structure on the interaction mechanism with analytes, in particular towards NO2. The outcomes of the Raman investigation, in addition to the Atomic Force Microscopy (AFM) and Scanning Electron Microscopy (SEM) analyses, originally disclose that CVD-Gr and LPE-Gr possess more defect structures compared to ME-Gr. The defects originate from the jagged morphology of the films rather than from intrinsic disorder. From the sensing point of view, a more defected morphology points out the presence of high-energy binding sites whilst a smoother structure consists mostly of low-energy binding sites, prevalently localized on the basal plane. Therefore, through the investigation of the kinetics upon the analyte interaction, we demonstrate that the low level of ME-Gr defectiveness induces a faster interaction with analytes. This conclusion is not only proven on ME-Gr and LPE-Gr displaying the steepest and the least steep slope, respectively, but also supported by the intermediate behavior of CVD-Gr that results to have a structure in between the other two kinds of materials. Consequently, the interaction kinetics is slower with the increasing level of the material defectiveness. Due to the absence of the signal saturation during the gas flow (Fig. 5), we infer that a more appropriate parameter to perform this study can be the signal slope rather than the current variation. The analysis we hereby present has been implemented adopting NO2 as the target gas since it is widely adopted in the gas sensing research field as a standard for oxidant analytes.6 Also, ME-Gr is well known to be sensitive to this analyte24 and in several studies we have uncovered that both LPE-Gr and CVD-Gr are much more sensitive to NO2 compared to other species.25–28
LPE-Gr suspension was synthesized by dispersing graphite flakes (product 332461, Sigma-Aldrich) at 1 mg ml−1 in a water/IPA mixture (7:
1 v/v) and sonicating in an ultrasonic bath kept at 30 W for 48 h.40 Next, a purification step to sift thinner flakes from un-exfoliated graphite crystallites was performed by applying a relative centrifugal g-force (RCF) roughly equal to 200g for 45 min. The estimated concentration of the suspension resulted to be 0.2 mg ml−1.
ME-Gr was produced by micromechanical exfoliation of natural graphite blocks and then transferred to Si wafer covered with 90 nm thick thermally grown SiO2.
To further examine the CVD-Gr, LPE-Gr and ME-Gr morphology, a NTEGRA AURA atomic force microscope (AFM) was used, operating in tapping mode with an n-doped Si NSG tip, rate 0.60 Hz and 512 lines. Based on the Gr type, different areas were scanned, ranging from 10 μm × 10 μm for CVD-Gr, to 50 μm × 50 μm in the case of ME-Gr and LPE-Gr. The surface topography was also investigated through a Scanning Electron Microscope (SEM) Philips XL50, using a beam acceleration voltage of 15 kV. For the sake of clarity, since SEM analysis could induce damage in the film, the images were acquired on a twin triad of samples.41,42
I–V measurements on such prepared graphene-based resistors were performed in the range [−1, 1] V through a semi-automatic probe-station equipped with an Agilent 4156C semiconductor parameter analyzer.
As a subsequent step, resistors based on CVD-Gr and ME-Gr were bonded on a chip by means of Al wires having a diameter equal to 30 μm in order to perform on those samples sensing measurements, as addressed in the following. The resistor based on LPE-Gr does not need to be bonded since the pad areas of a few mm2 allow directly to lay down the probes for the sensing measurements.
As far as LPE-Gr is concerned, the 2D band of all spectra is not composed by a single Lorentzian but is generally fitted by the two sub-components. In addition, as shown in Fig. 1b, on average the normalized intensities obey the relation I(2D1) > I(2D2), confirming that also in this case the film can be assumed as composed of FLG.6,26,33,34 Therefore, we can finally infer that, independently from the preparation technique, the three investigated sample surfaces are mostly composed of FLG.
A similar analysis was performed on the D peak, taking into account the ratio I(D)/I(G), especially for CVD-Gr and LPE-Gr, since ME-Gr spectra do not show this feature. In Fig. 2, the map of the I(D)/I(G) ratio, the respective histogram of the ratio distribution and the scatter plot of I(D)/I(G) as a function of FWHM(G) are reported: the panels (a–c) refer to CVD-Gr, and the panels (d–f) correspond to LPE-Gr. As featured by the average spectra (Fig. 1a), in both cases the presence of the D peak is further confirmed over the total mapped surface (panels (a) and (e)). Fig. 2b shows that, in CVD-Gr, the D intensity distribution has a wide dispersion on the mapped area and the most frequently observed value is around 0.2. On the other hand, LPE-Gr is characterized by a sharper distribution of I(D) alongside the sample surface, with the central value at around 0.35. These two almost double values reveal that, although both structures present defects, the sample morphology can also differ between them and surely is quite different from the ME-Gr one, having no D peak. To have a deeper insight, through the scatter plot I(D)/I(G) versus FWHM(G) (Fig. 2c and f) accomplished on the two sets of the collected Raman spectra, the kind of defect characterizing both samples can be evaluated. As stated by Torrisi et al.,32 the lack of correlation in both datasets suggests that the major contribution to the D peak does not originate from intrinsic disorder, but it is more related to the flake structure. As a result, the substantially different structures among the three inspected samples can be finally claimed and further evidence on this respect is delivered by AFM and SEM images (Fig. 3).
Both SEM and AFM analyses (Fig. 3) attest that ME-Gr presents a flat surface having continuous flakes with a mean lateral size in the range of a few tens of microns, as essentially shown by the SEM image (Fig. 3g) and AFM phase (Fig. 3i). As a counterweight, CVD-Gr (Fig. 3a–c) and LPE-Gr (Fig. 3d–f) are mostly composed of flakes with a mean lateral size around one order of magnitude lower, as shown by the scale bars of the SEM images. These findings match the conclusions achieved by means of Raman analysis, especially concerning the D peak. In fact, based on the laser spot size (∼3 μm), the jagged structure of the films acts as the origin of the defects which are indeed due to the flake edges and are totally absent onto the wider surface of ME-Gr. In both cases, i.e. CVD-Gr and LPE-Gr, the rise of these defects is intrinsically associated with the synthesizing routes, since CVD-Gr replicates the catalyst structure35–38 and being the LPE-Gr jagged structure due to the disrupting role played by the ultrasonic waves.39
Once the differences in morphology of the three realized materials were effectively proven, as a next step the sensing properties of such prepared materials were addressed. Devices were realized (see the Materials and methods section) using the Gr sample described in the previous section. The linear behavior of the I–V characteristics (Fig. 4) confirms that ohmic contacts were successfully established between Gr prepared according to the three different approaches and the metal contacts.
The devices were rigorously subjected to the same test protocol described in the Experimental section and Fig. 5 shows the dynamic current behavior towards the exposure to NO2. All graphs are normalized to the current value at the gas inlet of the first pulse.
In Fig. 5, a diverse general trend between the three curves is observed. While the graph associated with LPE-Gr (green line) has the tendency to continuously grow and scarcely recover after each single exposure, the opposite can be noticed for ME-Gr (black line). In the last case, the recovery phase is clearly distinguishable after each pulse and the final value of the current is even lower with respect to the beginning of the cycle. CVD-Gr (red line) displays an intermediate behavior between the other two, highlighting a slight recovery phase after the exposure window, even if the overall feature is not comparable to the behavior of the ME-Gr. The argument is further validated by taking into consideration the values of the normalized current at the end of each restoration process. The intermediate behavior of CVD-Gr appears more evident especially at concentrations lower than 0.5 ppm where a slight inversion of this baseline current is highlighted compared to the continuous rise-up related to LPE-Gr and the drastic decrease observed for ME-Gr that is already relevant at around 1 ppm.
These outstanding results denote the first significant proof of concept that the Gr structure can affect the sensing properties, especially bearing in mind that all three films consist of FLG, as previously demonstrated by the Raman analysis.
In an attempt to gain a deeper understanding, the signals during the single beats reported in Fig. 5 were compared, in particular considering the first two steps, where the three curves are mainly overlapping. Fig. 6 shows the magnification of the responses during the first and second gas pulses, at 1.5 and 1.32 ppm, respectively, as indicated in the right y-axis of the panels. During the pulse, a different rising rate for the three devices can be noticed, as further highlighted in the insets of Fig. 6, where the slopes of the current response are enlarged. Also, in Table 1, the fitting values of the slope are compared for both exposure steps and all three devices. It is noteworthy that a clear trend can be remarked: in both cases, ME-Gr shows the fastest rise compared to CVD-Gr and LPE-Gr that, in turn, represents the slowest one. The same was observed for the other gas pulses whose cycles are formed, exploiting the differential method introduced in our previous work (see Fig. S4 in the ESI†).25
![]() | ||
Fig. 6 Magnification of the first (a) and second (b) step of the current dynamic response shown in Fig. 5. During the exposure windows towards NO2 (blue rectangles), a different rate of signal growth can be noticed. The curves in panel (b) were shifted to the same starting point in order to facilitate the reader understanding. Insets. Fitting slopes of the rising signal for the three kinds of Gr. A decreasing level of steepness is shown encompassing ME-Gr (black line), CVD-Gr (red line) and LPE-Gr (green line). |
Slope (10−4) | ||
---|---|---|
Step 1 (1.5 ppm) | Step 2 (1.32 ppm) | |
ME-Gr | 2.17 ± 0.04 | 1.82 ± 0.03 |
CVD-Gr | 1.42 ± 0.01 | 1.07 ± 0.01 |
LPE-Gr | 0.42 ± 0.01 | 0.33 ± 0.01 |
The key concept relating the Gr structure and the diverse behaviors towards the analyte is tracked down within the context of adsorption sites having high or low binding energy, as reported by Lu et al.18 In that work, similar behaviors to what hereby presented were also addressed as associated with differently structured nano-materials. The authors claimed that diverse regimes and, in turn, different slopes in the response curves during the exposure towards NO2 are due to different interaction mechanisms between the sensing layer and the gas molecules. In particular, fast responses, corresponding to steeper lines, are mainly attributed to sites with low binding energy, such as the sp2-carbon localized on the plane. On the other hand, binding sites having high-energy, such as defects, are responsible for slow responses, i.e. flatter lines.
In this respect, the correspondence between the conclusions achieved in that paper and the experimental data hereby discussed is quite consistent. In fact, ME-Gr, being composed of smooth flakes with a mean lateral size in the range of a few tens of microns, is more prone to provide only low-energy binding sites localized on the surface. In turn, faster responses can be observed, as shown by the response curves in Fig. 6 (black lines). On the opposite, LPE-Gr and CVD-Gr mainly consist of rough flakes having a mean lateral size of a few microns, i.e. an order of magnitude lower compared to ME-Gr. Then, for instance, at a fixed area of the sensing layers prepared according to the three approaches, the flat surface related to ME-Gr is much more predominant than in the other two cases that, instead, are characterized by a higher density of defected structures (Fig. 3a, d and g). As the defects are considered high-energy binding sites, the interaction between molecules and that kind of sites bears out the slower rate of the slopes, as reported in Fig. 6 and Table 1.
Also, despite both LPE-Gr and CVD-Gr exhibiting defected structures (see Raman spectra in Fig. 1), looking at Table 1 and insets of Fig. 6, the slim disparity discerned between them can be explained by considering what has been stated with reference to Fig. 2b and e. The slightly wider diversity of CVD-Gr morphology with respect to the LPE-Gr one can also be interpreted as a simultaneous presence of defected points, as already demonstrated, but also encompassing some zones within the basal plane free of defects.
In other words, on the same sample, not only high-energy binding sites are present, but also sites localized on the plane, indeed low-energy binding sites, are distributed, justifying also the intermediate behavior between ME-Gr and LPE-Gr previously identified (Fig. 5).
The behavior of chemi-resistors based on differently prepared materials towards NO2 was inspected. The experimental data attested a clear correlation between the flake structure and the behavior towards the analyte. ME-Gr showed a faster response rate during the exposure time towards the gas. On the contrary, CVD-Gr and LPE-Gr resulted to have a slower response. The CVD-Gr intermediate behavior between ME-Gr and LPE-Gr is explained by the fact that CVD-Gr consists of a diversified structure. Low-energy binding sites localized on the plane and responsible for the fast regime are present, similarly to ME-Gr. At the same time, as occurs for LPE-Gr, high-energy binding sites, such as defected points, exist, determining the slow rate.
The remarkable findings hereby addressed and, in general, the correlation between the sensor behavior and the purity level of the material justified the best performances reached by ME-Gr based devices compared to the other two. Furthermore, we demonstrated that CVD-Gr represents a promising route to attain comparable results, especially taking into account the large scale production achievable by CVD compared to the manual fabrication of ME-Gr. Therefore, the outcomes can pave routes of possible applications and research developments in the sensing field, mainly related to the capability of tailoring the device performance based on the flake defectiveness level.
Footnote |
† Electronic supplementary information (ESI) available: Details of Raman characterization of samples addressed in the main text. The E-beam lithography procedure adopted for the ME-Gr based sensor realization. A differential method applied to the dynamic sensor behavior. See DOI: 10.1039/c7nr01120b |
This journal is © The Royal Society of Chemistry 2017 |