Highly stretchable strain sensors based on Marangoni self-assemblies of graphene and its hybrids with other 2D materials

Graphene and other two-dimensional materials (2DMs) have been shown to be promising candidates for the development of flexible and highly-sensitive strain sensors. However, the successful implementation of 2DMs in practical applications is slowed down by complex processing and still low sensitivity. Here, we report on a novel development of strain sensors based on Marangoni self-assemblies of graphene and of its hybrids with other 2DMs that can both withstand very large deformation and exhibit highly sensitive piezoresistive behaviour. By exploiting the Marangoni effect, reference films of self-assembled reduced graphene oxide (RGO) are first optimized, and the electromechanical behaviour has been assessed after deposition onto different elastomers demonstrating the potential of producing strain sensors suitable for different fields of application. Hybrid networks have been then prepared by adding hexagonal boron nitride (hBN) and fluorinated graphene (FGr) to the RGO dispersion. The hybrid integration of 2D materials is demonstrated to become a potential solution to increase substantially the sensitivity of the produced resistive strain sensors without compromising the mechanical integrity of the film. In fact, for large quasi-static deformations, a range of gauge factor values up to 2000 were demonstrated, while retaining a stable performance under cyclic deformations.


Introduction
Nowadays highly flexible and highly-sensitive strain sensors are being required for the development of soft electronic devices that cover a range of applications, such as human body measurements and motion detection, sports performance monitoring or even human-machine interfaces and soft robotics [1][2][3][4][5][6][7][8][9]. These sensors can be employed to monitor object movements by detecting the shifts in the electrical signal (often resistance or capacitance) that are generated upon the application of an external deformation [10]. Their performance relies on several parameters, such as the relative change of the electrical signal known as gauge factor (GF), linearity, response and recovery time. Their advantages compared to the traditional sensors (i.e. metal and semiconductor strain gauges) are the high efficiency and resolution even at the nanoscale [11][12][13], the ability to be embedded in structural materials [14], high stretchability and durability [15], low power consumption [16] and, eventually, biocompatibility [17].
The poor stretchability and sensitivity of conventional metals or inorganic semiconductor-based strain sensors have restricted their application in flexible electronic devices to some extent, and hence many efforts have been devoted to find suitable candidates to overcome these limitations. The introduction of nanomaterials such as carbon nanotubes (CNTs) [18] and, more recently, graphene and related materials (GRMs) [5,9,19,20] and other 2D materials [21][22][23][24], has given rise to a new era of innovative strain sensors [16,[25][26][27][28], owing to their outstanding electrical, mechanical, optical and chemical properties, and to the possibility to be integrated into polymers as flexible support material [13]. For instance, Fu et al [29] demonstrated that monolayer graphene synthesized via chemical vapor deposition (CVD) and deposited on a flexible substrate can act as a strain sensor with high sensitivity, reaching a GF of 151 for deformations up to 4.5%. More recently, Liu and his co-workers [2] presented a graphene woven fabric/polydimethylsiloxane (PDMS) composite capable of detecting feeble human motions with an extremely high piezoresistive GF of 223 at a strain of 3% and excellent durability. Such a sensor was applied to convert human motions into sounds and music of different instruments. By using the same polymer substrate, Wang et al [19] proved that the morphology and periodicity of the as-formed graphene ripples can withstand large deformation, thus making it ideal for flexible electronic applications. Similar results were also found by Anagnostopoulos et al [14], where the electrical resistance of CVD graphene/epoxy resin system was found to decrease with the imposition of mechanical deformation, possibly due to the opening-up of the structure and the associated increase of electron mobility. Further advances are represented by the possibility of bestowing sensing multifunctionality; for instance, Xu et al [30] fabricated a multifunctional wearable device based on two different reduced graphene oxide (RGO) films that can simultaneously detect physiological signals and volatile organic compound biomarkers without inducing signal interference.
Important parameters in the production of a strain sensor are the scalability and its low cost, which can make it attractive to a plethora of applications. Recently, a novel selfassembly technique based on the Marangoni effect [31,32] has been presented to fabricate large-area ultrathin graphene films with 'fish scale' like microstructures that have been demonstrated to be suitable for strain sensing when deposited onto flexible substrates, with GF up to 1037 at 2% strain [33].
In particular, it was shown that -driven by Marangoni flowgraphene flakes dispersed in ethanol can form films in distilled water due to potential difference in surface tension of the two liquids [34] (i.e. the flakes move from low to high surface tension regions). As a result, the graphene flakes can self-organize and stack very densely with each other through π-π interaction [33,35].
Herein, we report on the ecofriendly and cost-effective development of strain sensors based on Marangoni selfassembly of graphene and of its hybrids with other 2D materials that can withstand very large deformation (up to 30%, i.e. one order of magnitude higher than those explored by Li et al [33]) and exhibit highly sensitive piezoresistive behaviour. In particular, reference films of self-assembled RGO were prepared and optimized by investigating the effect of concentration on structural and electrical properties. Then, the electromechanical behaviour of the optimal RGO film was assessed after deposition onto several elastomeric substrates -i.e. thermoplastic polyurethane (TPU), PDMS and fluoroelastomer -demonstrating the potential for producing strain sensors suitable for different fields of application. Finally, by using the same methodology, two types of hybrid coatings were prepared by adding hexagonal boron nitride (hBN) and fluorinated graphene (FGr) to the RGO dispersion and, very interestingly, this approach has been found to increase substantially the sensitivity of the produced resistive strain sensors. For quasi-static deformations up to 40%, a range of record GF values were obtained reaching up to 2000, while for cyclic deformations a stable performance is observed despite the viscoelastic nature of the substrates.

Experimental section
Dispersions of reduced graphene oxide and hybrids with other 2DMs The GO has been synthesized from natural graphite flakes (NGS Naturgraphit GmbH, Germany) by a two-step oxidation process, as it has been previously reported [36][37][38].

Elastomeric substrates
Three different elastomers were employed as substrates, namely thermoplastic polyurethane (TPU), polydimethylsiloxane (PDMS) and fluoro-elastomer (Viton). The dimensions of all the specimens were 100 × 5 × 3 mm 3 . In particular, films from a polyesterbased TPU (Elastollan 890AN, gently supplied by BASF) were produced by compression molding under 20 bars at 150°C. PDMS films (Sylgard 184, DOW) were synthesized by casting the monomer/initiator mixture in ratio 10/1 and then curing at 150°C for 10 min. Viton substrate (V732, gently supplied by Trelleborg) was used as received and cut in the desired dimensions.

Strain sensors fabrication
The selected dispersion of RGO or hybrid RGO/FGr or RGO/hBN in 60:40 H 2 O/EtOH solution was dispensed on the surface of DI water contained in a large beaker, and ethanol quickly spread due to Marangoni effect enabling particle self-assembling. Once the graphene-based film is formed on the water surface, it can be deposited on a suitable substrate by employing the 'scooping' technique [39][40][41]. Accordingly, the chosen substrate is firstly immersed in the fluid over which the film has been formed and then lifted very slowly, scooping the self-assembled layer from the water. After completion of the deposition step, specimens were heated to 50°C for 30 min in oven in order to remove any remaining solvent.

Electrical characterization
Electrical properties of produced self-assembled films were measured through van-der-Pauw method [42]; electrical contacts were formed by pasting copper foil strips to the end of each specimen through a conductive silicone adhesive. A four-point probe positioned on the sample provided the average resistivity under the examined area. Measurements were conducted by using the digital source meter Keithley 2420.

Electromechanical characterization
Quasi-static and continuous cyclic deformation were imposed onto the specimens and, simultaneously, electrical resistance was measured in order to obtain the sensitivity and to evaluate the performance of the produced graphene-based strain sensors. To do so, a MTS mini-Bionix 848 servo-hydraulic testing frame was used. The testing apparatus was capable of applying a fully strain-controlled external load via the embedded LVDT system, while the MTS TestStar 40 controller provided a high-rate real-time recording of the experiment parameters (time, displacement, and force). All measurements took place by using a 25kN load cell with a strain rate of 50 min −1 for the static tensile tests, while for the dynamic tests two deformations ranges (3%-6% and 8%-16% strains) at 0.1 Hz were applied, respectively. Especially for the PDMS the deformation ranges applied were 0%-3% and 0%-6% at 0.1 Hz and 0.5 Hz in order to simulate also deformations appear in the human body [11]. The gauge length of all specimens tested was 60 mm. During the mechanical measurements, the electrical response was simultaneously recorded by using Keithley 2420 multimeter.

Raman spectroscopy
Raman spectra were acquired with a Raman microscope (InVia Reflex, Renishaw, UK) equipped with at 514 nm (2.41 eV) laser. The laser power on the sample was kept below 1 mW to avoid local heating, while an Olympus MPLN100x objective (NA = 0.90) was used to focus the beam on the samples. Raman mapping was performed on an area of 80 × 50 μm 2 at a step of 1 μm and 1 s of acquisition time. All Raman peaks were fitted using Lorentzian functions.

Scanning electron microscopy (SEM)
The morphological features of the samples were investigated by using field-emission scanning electron microscopy (FESEM, Zeiss SUPRA 35 VP).

Atomic force microscopy (AFM)
AFM images were collected in a peak-force tapping mode, with a Dimension Icon instrument (Bruker Corporation, USA). Scanasyst-air tips with nominal tip radius of 8 nm, typical spring constant of ∼0.4 N m −1 and a frequency of ∼70 kHz were used for the characterization. Images were recorded at 20 × 20 μm 2 with 512 × 512 lines per image, and a scan rate of 0.9 Hz. The thickness of the produced films deposited on silicon was evaluated using the cross-section analysis of the Nanoscope Analysis software. Several steps were measured for each deposited layer at the edges to allow statistical analysis of data.

UV-Vis spectroscopy
UV-Vis transmission spectra were recorded using a Hitachi U-3000 reverse optics spectrophotometer, while the IR spectra were measured using the Attenuated Reflectance Technique (ATR) of Bruker Equinox FTIR spectrometer.

Results and discussion
The stoichiometry, the quality, and the morphology of the produced reduced graphene oxide (RGO) have been investigated in detail via x-ray photoelectron spectroscopy, x-ray diffraction (XRD), Raman spectroscopy and scanning electron microscopy (SEM). The C 1s spectrum (figure S1) was deconvoluted into six components, corresponding to C-C sp 2 and C-C sp 3 carbon bonds, C-O(H) epoxides and hydroxides, carbonyls C=O, carboxyls O-C=O(H) and the π-π * transition loss. The % components derived from the C 1s peak deconvolution prove the high reduction rate achieved by heating the GO to 1050°C in an inert atmosphere. This is also confirmed by XRD (figure S2); in fact, by comparing the spectra for the produced GO and its reduced form, the former shows a graphitic peak at 2θ = 10.3°attributed to its oxygen functionalities, while for the latter this peak is absent and a broad peak at 2θ = 23.5°is emerged due to the decrease of intercalated oxygen atoms [43]. Furthermore, (figure S3) characteristic Raman spectra for GO and RGO reveal certain differences; in particular, as the reduction takes place, the ratio of the intensities of D (∼1350 cm −1 ) to G (∼1580 cm −1 ) peaks are found to increase [44,45]. SEM images of GO and RGO flakes (figure S4) reveal large particles, with size of the order of few hundreds of microns, with a crumpled morphology. On the surface, nanoscale wrinkles and corrugations can be noticed. As for the RGO case, it seems that the thinnest regions and borders are made of very thin crystals, only few layers thick, crumpled and folded several times.
Self-assembled RGO ultrathin films have been then prepared using the single-step Marangoni method presented by Liu et al [33]. Actually, the ethanol with the RGO flakes moved from low surface tension areas (ethanol-rich) to high surface tension areas (water-rich). Eventually, the π-π interactions bind RGO flakes each other and form the ultrathin film. It is worth mentioning that the formation of the film on the surface of the deionized water is observed after only a few seconds. In order to assess the quality of the produced RGO self-assembly, the floating films have been subsequently deposited on target substrates. The morphology and the homogeneity of the films produced from four different concentrations of RGO dispersions (namely, A1 = 0.16 mg ml −1 , A2 = 0.08 mg ml −1 , A3 = 0.04 mg ml −1 , A4 = 0.01 mg ml −1 ) has been examined by using optical microscopy and atomic force microscopy (AFM). The optical images shown in figure A1 clearly reveal that the RGO assemblies fully cover the target substrate. In particular, the higher homogeneity is achieved in the self-assembly obtained from the less concentrated dispersion, while large aggregates can be observed in the films obtained from dispersions with higher RGO concentration. AFM images highlight the characteristic structure of stacked and closely packed RGO flakes, and a significant roughness associated with the crumpled morphology of the nanoparticles ( figure 1(B)). The estimated thickness that has been measured from the corresponding height profiles (figure S5) was found to be 10 nm for the lower concentration and 25 nm ca. for the higher concentration of RGO dispersions. Further proof of the homogeneity of the RGO self-assembly obtained from A4 dispersion is provided by Raman spectroscopy. In fact, the homogeneous spatial distribution of intensity of the G peak presented in the Raman mapping (figure 2(A)) confirms that RGO has been successfully transferred with high coverage of the target substrate. The produced films are transparent in the visible spectrum, as revealed by UV/Vis spectroscopy ( figure 2(B)). More specifically, the transparency of the self-assembled film increases from 59.2% for the high concentration to 77.5%, for the low concentration of RGO dispersions.
Taking into consideration that graphene absorbs only 2.3 % of visible light, fabricated thin films are expected to have a thickness of few tens of nm. Finally, the dependence of sheet resistance R s of the produced films as a function of the concentration was also measured and presented in figure 2(C). For the higher RGO concentration, the sheet resistance is of the order of 2.5 kΩ per square. As the concentration is decreased a gradual increase of the sheet resistance of the films is observed. Such a difference is due to the amount of charge carriers (RGO flakes) on the glass substrate. In light of the morphological and electrical characterization, it is clear that the optimum film is obtained by using the RGO dispersion with the lowest concentration.
The characteristic stacking structure of the RGO selfassemblies and their electrical properties made them promising candidate for the development of flexible and stretchable strain sensors. In order to investigate this phenomenon, the optimum Marangoni-driven self-assembly has been deposited onto a stretchable, elastomeric substrate (PDMS) and the electromechanical behaviour has been examined through quasi-static and cyclic experiments. In figure 3(A) the relative resistance change (ΔR/R 0 ) is presented as a function of the applied quasi-static deformation. It is evident that ΔR/R 0 increases up to 30% of the applied strain and this has been ascribed to the sliding and thus the relative movement between the adjacent RGO flakes that affects the electron pathways and the tunneling effect [19,46,47]. In our case, the experimental data of relative resistance changes (ΔR/R 0 )   can be perfectly fitted via exponential growth equations (y = A + B·e bx ) in the whole strain region for all the systems as can be also seen in figure 3(A). This behaviour has been previously predicted by first principles calculations combined with nonequilibrium Green's function method by Li et al [33]. In order to quantify the strain sensing performance of the RGO/PDMS system, the gauge factors (GF) have been evaluated from the slope of the curve in four strain regions (namely, 0-5, 5-10, 10-20 and 20%-30% of applied strain) and are presented in figure 3(B). It is interesting noting that the RGO/PDMS system can approach GF values up to 1500. The recovery and the stability over time of the RGO/PDMS system has been investigated also through cyclic measurements ( figure 3(C)). The dependence of the resistance upon the application of a dynamic loading is shown with a frequency of 0.1 and 0.5 Hz for two ranges of strains from 0% to 3% and from 0% to 6%. It is evident that the resistance follows the external applied deformation even after several cycles. It is interesting noting that, as with most nanocomposites, some conditioning is observed, here over the first 15-20 cycles. It is generally associated with the viscoelastic behaviour of the substrate, which enables the relaxation of the nanoparticle network to a slightly better connected state after every deformation [11]. However, after 50 cycles the response is extremely stable. In both cases, the same performance is monitored. For the 0.1 Hz, resistance changes 0.18 for 0%-3% strains and from 0.4 for the 0%-6% strains, respectively. In this way the sensor can detect even weak changes, showing capabilities for monitoring an increase of blood pressure or the occurrence of tachycardia. As for the 0.5 Hz the resistance changes found to be 0.14 and 0.27 0%-3% and 0%-6% strains respectively, proving the excellent sensitivity of the examined system.
The produced Marangoni RGO self-assembly can be easily deposited also onto other elastomeric materials (e.g. TPU and Viton) and the strain sensing behaviour is still retained. However, it is evident that the electromechanical response depends on the target elastomeric substrate. Actually, in the case of the film deposited onto PDMS, the resistance changes are much more intense compared to the other two systems and this could be ascribed to possible higher adhesion between the RGO film and the PDMS [33]. It is important to note here that the deformation range under investigation is well outside the range of expected fracture of the employed elastomeric substrates (figure S6). Very interestingly, the possibility to adopt elastomeric substrates with different chemical structure (e.g. thermosets and thermoplastics) is a key factor in the design of highly sensitive strain sensors for a wide range of applications, from body motion (e.g. integration into shoe soles, in the case of TPU) to sensors for harsh environments (e.g. smart sealings, in the case of Viton).
We have therefore demonstrated that, by using an environmentally friendly and cost-effective method, we can produce graphene-based highly sensitive strain sensors with several elastomeric substrates that can withstand large deformations (up 30%) and that can be applied in different applications, from flexible electronics, body motions sensing and high-end applications under harsh environments. In principle, one can think that the sensitivity of the Marangoni RGO film can be improved by further decreasing the concentration of the initial dispersion. However, using less concentrated RGO dispersions is detrimental to the uniformity and the integrity of the film. A practical way to circumvent this, is to introduce non-conductive elements in the Marangoni self-assembly of conductive RGO particles. The presence of the non-conductive elements in the network has a two-fold function; first of all, is expected to reduce the available paths for electrons to travel through the network and thus making it much more sensitive to external stimuli. Secondly, the non-conductive particles still act as synergetic building blocks in the hybrid network, thus retaining the integrity of the film. To this aim, we fabricated Marangonibased ultra-thin films of simultaneously conductive (RGO) and non-conductive 2D materials (e.g. fluorinated graphene and hBN). More detailed, ultra-thin films from hybrid dispersions of fluorinated graphene (FGr) and RGO in concentrations 30/70, 60/40, 70/30 and 75/25 w/w have been prepared.
Raman mappings collected for hybrid systems with different concentrations of FGr have been derived on the bases of the spectral background of the acquired spectra, which is very strong in the case of FGr and absent in the case of RGO ( figure S7). The contours depicted in figure 4(A) reveal the presence of FGr particles in the hybrid self-assembly deposited on the target substrate (increasing with the FGr concentration in the initial dispersion), and suggest the subsequent reduction of the available conductive paths of the surface network. This is indeed confirmed from the sheet resistance measurements that are presented in figure 4(B), revealing an increase from 10 KΩ per square for the neat RGO network to almost 1 MΩ per square for the FGr/RGO 75/25 system. This is expected to make the network much more sensitive to the relative movement between the adjacent RGO flakes that affects the electron pathways and the tunneling effect. It is important to note at this point that the produced films retain high uniformity and integrity.
As already done for RGO films, the electromechanical behaviour of the produced hybrid films deposited on PDMS has been investigated. As shown in figure 5(A), the strain sensitivity increased significantly compared to the neat RGO film at the same strain level. Moreover, an increase of the GF of the fabricated sensor is observed with the gradual increment of FGr content in the hybrid film ( figure 5(B)). Finally, resistance dependence of the FGr/RGO 70:30 has been also studied upon the application of a dynamic loading with a frequency of 0.5 Hz for three ranges of strains from 0% to 2.5% and from 0% to 5% and 0%-10% where it is evident that the resistance is following the externally applied load. For concentrations of FGr higher than 75:25 the film was not conductive and thus cannot be used as strain sensor.
As anticipated, in the hybrid system, the observed improvement of the strain sensitivity with FGr content can be ascribed to the reduction of the conductive paths arising from the increased amount of the non-conductive flakes in the Marangoni assembly. As further proof of that, hBN has been explored as non-conductive phase in the hybrid ultra-thin films, as well. As shown in figures 5(A) and (B), also the hBN/RGO hybrid system exhibits significant improvement of the strain sensitivity compared to the neat RGO network. Nevertheless, since hBN particles possess larger lateral size compared to FGr flakes (figure S8), their disruptive effect on the conductive network becomes significant even at lower weight fractions. This finding suggests that the lateral size of the non-conductive phase can be a critical parameter for the realization and the optimization of the strain sensors based on hybrid 2D systems.
In biomechanics, researchers strive to measure accurately the motion of segments of the human body. Recognizing body posture and motion is an important physiological function that can help keep the body in balance. Motion sensors are often used to diagnose balance disorders and track energy expenditure. For this reason, small sensor units can be installed at various locations on the human body to provide measurements over time. Ultra-thin graphene-based strain  sensors are sensitive enough to measure small electrical signals, which can be used to detect motion of the human body. Hybrid FGr/RGO self-assemblies have been used in order to fabricate high sensitivity body motion sensors. More detailed, FGr/RGO 70:30 was deposited on PDMS and then it was attached on human wrist and finger for body motion detection. As it is presented in figure 6 the fabricated sensor accurately captures the motion of both finger and wrist motion with high accuracy as it is also demonstrated from the videos in SI.

Conclusions
The development of highly sensitive strain sensors based on self-assemblies of graphene and its hybrids with other 2D materials has been shown. The adopted process exploits the development of Marangoni forces that drive the self-assembly of continuous ultrathin films of 2D particles and is environmentally friendly and cost-effective. More specifically, ultra-thin films of RGO and hybrid systems of RGO either with fluorinated graphene or with hexagonal boron nitride have been produced and deposited on elastomeric substrates. RGO films were optimized, and the strain sensing ability has been demonstrated for elastomeric substrates with different chemical structures. The RGO films exhibited high sensing capabilities, especially in the case of PDMS with a GF of 1500 at the strain range 20%-30%, and remarkable stability upon cyclic strains and frequencies. In the case of hybrid systems, a strongly connected network of ultrathin 2DM domains has been formed while the addition of non-conductive elements makes the whole network more sensitive to external deformations since the conductive paths are significantly reduced due to the presence of the non-conductive units. Indeed, the GF was found to be higher compared to the neat RGO film in all cases. Particularly in the case of hBN/ RGO the GF was found to be higher than 2000 in the strain range 9%-13%. Finally, highly sensitive and stable body motion sensors have been fabricated from hybrid FGr/RGO, and have been demonstrated effective for real applications, such as in situ tracking electronic skin, wearable sensors, and health monitoring platforms.

Acknowledgments
This work has received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement n. Graphene Core 3 881603.

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).

Conflicts of interest
There are no conflicts to declare.