Non-invasive on-skin sensors for brain machine interfaces with epitaxial graphene

Objective. Brain–machine interfaces are key components for the development of hands-free, brain-controlled devices. Electroencephalogram (EEG) electrodes are particularly attractive for harvesting the neural signals in a non-invasive fashion. Approach. Here, we explore the use of epitaxial graphene (EG) grown on silicon carbide on silicon for detecting the EEG signals with high sensitivity. Main results and significance. This dry and non-invasive approach exhibits a markedly improved skin contact impedance when benchmarked to commercial dry electrodes, as well as superior robustness, allowing prolonged and repeated use also in a highly saline environment. In addition, we report the newly observed phenomenon of surface conditioning of the EG electrodes. The prolonged contact of the EG with the skin electrolytes functionalize the grain boundaries of the graphene, leading to the formation of a thin surface film of water through physisorption and consequently reducing its contact impedance more than three-fold. This effect is primed in highly saline environments, and could be also further tailored as pre-conditioning to enhance the performance and reliability of the EG sensors.


Introduction
Brain-machine interfaces aim to be a communicative bridge between the human brain and external electronic systems, integrating seamlessly biological electric pulses (biopotentials) with electronic devices. Non-invasive interfaces can be employed to detect the cortical electrical activity (brain waves) through electroencephalogram (EEG) analysis to interpret human intentions and translate them, for instance, into commands directed to electronic robots [1,2]. Brainmachine interfaces are gaining substantial impetus from recent advances in brain science [3][4][5]. EEG sensors are non-invasive in nature, and they can be hence easily employed for any wanted duration without undergoing delicate surgery procedures. However, their signal collection is significantly challenged by the presence of the skull, acting as a physical barrier with additional complexity caused by the presence of hair and variable scalp conditions [3]. EEG sensors can be classified into two types: wet and dry sensors [3,4]. Semi-dry sensors, typically based on hydrogels, have recently demonstrated encouraging potential as EEG sensors, however, their development is still at an early stage [5][6][7][8]. Although wet electrodes typically offer a lower contact impedance with the scalp, they are not desirable for commercial applications due to their cumbersome deployment outside of a clinical environment [4]. On the other hand, dry on-skin electrodes are more versatile and easy to use [8][9][10][11]. Dry EEG sensors can be made wearable, reusable, and they can be easily integrated as part of a cap or helmet, opening up a variety of applications outside of clinical settings [9,10]. The successful implementation of dry electrodes depends on achieving a lower sensor/scalp impedance, prolonged reliability of the electrodes, while causing minimal skin reaction or discomfort. More conventional dry EEG sensors have a rather uncomfortable bulky form-factor, while still yielding rather high impedance with poor biocompatibility and limited robustness. Recent advances in soft nanomaterials including nanocarbons have been shown as promising alternative towards improved signal collection at the skull [12][13][14][15][16][17][18][19][20][21][22][23][24][25][26].
Few-layers epitaxial graphene (EG) grown on silicon carbide on silicon substrates could become a material of choice for EEG electrodes that can operate with high reliability in external environments. Graphene's excellent conductivity and biocompatibility, highly adherent to a robust mechanical electrode structure would also offer the capability of integration with silicon electronics for further added functionalities [28][29][30][31][32][33][34]. Silicon carbide has already gained attention as material for neural interfaces due to its physical robustness, chemical inertness, hemocompatibility and compatibility with neural cells [35]. The EG on SiC can enhance those beneficial properties by combining the advantages of both materials. Using SiC on silicon substrates brings the additional advantage that silicon can be made flexible through thinning it down below 50 µm thickness, which would be impossible using SiC substrates [36].
Here, we report of a novel dry EEG electrode based EG from cubic SiC on silicon for brainmachine interfaces. The graphene electrodes show enhanced electrochemical properties as compared to commercial dry EEG electrodes, particularly an outstanding reliability. In addition, we report a new key phenomenon we refer to as 'surface conditioning' . We observe a dramatic reduction of the contact impedance of the graphene electrodes with the skin over repeated testing, thanks to the formation of surface functional groups promoting graphene wetting and better ion intercalation, all leading to a strongly improved sensor performance.

Experimental methods
3C-SiC (cubic silicon carbide) films epitaxially grown on highly doped Si (100) substrates have been purchased from NOVASiC; a chemical and mechanical polishing procedure was performed (StepSiC ® by NOVASiC (France)) [37]. The SiC layer thickness is ∼500 nm. A catalytic alloy approach was employed for the growth of EG on SiC on silicon [28,29]. A cryopump deposition chamber operating with CD Ar + ion and 200 mA current was used for sputtering the metallic layers. A sequential deposition of a Ni layer (∼10 nm) followed by a Cu layer (∼20 nm) was used to fabricate EG. Annealing the metal-coated samples at ∼1100 • C for an hour under vacuum condition (∼10 −5 mbar) leads to breaking the Si-C bonds thanks to the presence of Ni, and release of the carbon atoms which form graphene on the surface. Ni silicides are the by-product of this process, which are later removed together with any metal residues on the surface using chemical wet etching for 9 h (Freckle solution) [28,29]. EG/SiC/Si coupons of 1 × 1 cm 2 were used for this study. A Zeiss Supra 55VP SEM was employed for surface imaging. Raman spectra were collected with a WiTec Raman microscope using a green laser (532 nm); the Raman spectra reported here is an average of 2500 single spectra collected over a 100 µm 2 area map. X-ray photoelectron spectroscopy (XPS) was carried out with a Specs PHOIBOS 100 Analyzer operated with a Mg Kα xray source and 3 mm spot size. The contact angle measurements were performed using a tensiometer (Sigma 701, Biolin Scientific). An electrochemical workstation (CH Instruments, 660 E Model) was used to assess the electrochemical performance of the cells operating in a three-probe configuration (figure S1 (available online at stacks.iop.org/JNE/18/066035/ mmedia)). The primary electrochemical properties of the 3CSiC films and the EG grown on silicon carbide films were characterized in a three-electrode cell using a solution of 0.1 M NaCl in water as electrolyte, with respect to Ag/AgCl electrodes. Cyclic voltammograms (CVs) of the electrodes were taken at a potential window of 0.8 V and electrochemical impedance spectroscopy (EIS) measurements were conducted in the 0.01 Hz-1000 kHz frequency range with a signal amplitude of 5 mV.

On-skin measurements and helmet tests
The on-skin EIS tests were carried out by mounting the EG on silicon coupon of an area of 1 cm × 1 cm. Double sided carbon tape was attached to the back of the highly doped silicon substrate using a pin button for electrical connection. Note that cubic heteroepitaxials SiC on silicon grows unintentionally n-type doped, and is hence conductive [34]. In addition, when a doped silicon wafer of p-type polarity is used as the substrate for SiC growth, the SiC/silicon interface becomes highly leaky due to a carrier inversion phenomenon [53,54], causing altogether the graphene to become effectively in electrical contact with its substrate, hence enabling this simple electrical contact approach from the silicon backside. The resistivity of the highly doped p-type silicon substrates were found in the range of 10-20 Ω cm.
In all on-skin measurements, a three-electrode setup was configured using the EG sensor as the working electrode (WE), Ag/AgCl as the reference electrode (RE) and gold as the counter electrode (CE) with a 4 cm distance between the electrodes. All the EIS measurements were taken across a frequency range of 0.1-10 000 Hz at 5 mV. Variations of the electrode/skin contact impedance is common due to factors such as the occurrence of skin flakes, sweat, and changing environmental conditions. Five EG sensors were fabricated, and the sensors have been tested over ten times to investigate reproducibility.
Each sensor was first tested in a three-electrode cell setup. Subsequently, they underwent on-skin EIS measurements on forearms and forehead. Once the skin contact was established, three to four on-skin EIS sequential measurements were taken at 5 min intervals without breaking the skin contact. Each complete measurement took 3-4 min. The average impedance values and their variation were evaluated in the stable value region, reached after three to five repeated measurements (surface conditioning). In addition, in order to explore the robustness of the sensors, we have also performed measurements after 10 min interruption of the skin contact, as well measurements in a saline solution soaked cellulose cloth surface (0.01 M phosphate-buffered saline (PBS)).
We have also used liquid exfoliated graphene (LEG) on SiC as a way to compare the performance of the EG versus that of large-grain (several microns), drop-cast graphene. The large flakes of fewlayer graphene were synthesized by liquid exfoliation via sheer mixture following the Coleman et al methods [39,40]. The synthesized graphene showed large few layer graphene flakes (figure S5(a)) and typical low-defect Raman spectrum ( figure S5(b)). The low concentrated dispersion of graphene in isopropanol was drop-casted on the highly doped 3C SiC/Si substrate and dried before on-skin measurement.
The helmet setup [38] used for this work contains eight channels connected to eight commercial foam sensors (EEE Holter Technology Co., Ltd, Hsinchu, Taiwan). The two frontal channels, which are in contact with the forehead skin, were replaced by the EG sensors for comparison. The wearable sensors were exclusively tested on the researchers who have co-authored this work, who have all agreed to carry out and consented to publish this research.

Results and discussion
A schematic of the EG electrode used for EEG is shown in figure 1(a). A highly doped silicon wafer was employed to provide a suitable electrical connection through the back of the electrodes. The surface of the EG-based EEG electrodes was characterized by SEM-energy dispersive spectroscopy (EDS) mapping for elemental analysis (figures 1(b) and (c)).
The EDS data show the typical elements expected for the EG synthesized on 3C-SiC/Si substrates [28][29][30]. Details about the graphitic nature and carbon lattice defects of the EG were evaluated by extensive Raman spectroscopy mapping at a wavelength of 532 nm. The Raman spectra show four dominant peaks: the longitudinal optical peak of SiC at ∼970 cm −1 and graphene-related peaks of D peak at ∼1340 cm −1 , G peak at ∼1580 cm −1 , and 2D peak at ∼2680 cm −1 (figure 1(d)) [36,41]. The D peak arises from defects within graphene lattice. The intensity ratio between D and G peaks has been widely used as a measure to compare defect density for graphenebased materials, whereas the intensity ratio between the 2D and the G peaks is an indicator of the graphene thickness [36,41]. Here, the I D /I G and the I 2D /I G averaged intensity ratios indicate the presence of lowdefect, and one to two layers graphene, respectively. In particular, the I D /I G ratio of 0.2 is an indication of an average graphene grain size of about 90 nm [32,52].
Na + and Cl − in water solution are known to be major components of human sweat and thus main skin electrolytes [42]. CV was hence used to investigate the electrochemical behavior of EG and SiC on silicon films in 0.1 M aqueous NaCl electrolyte at potential limits of 0.8-0.0 V vs Ag/AgCl electrode in a three-electrode system, at a scan rate of 100 mV s −1 (figure 2(a)). From the CV curves it is observed that the EG possesses a substantially enhanced (about 1.8 times) capacitive behavior compared to the SiC film on highly doped silicon (figure 2(a)) [43,44]. The CV curve of SiC shows redox peaks (oxidation peak at 0.28 V and reduction peak at 0.18 V), possibly due to the presence of metallic impurities. We also observe that the initial CV curve of EG shows a slightly resistive behavior, which is reduced upon repeated cycling (the 100th cycle is shown in figure 2). To quantify and compare the transfer impedance between the electrodes and the electrolyte, we employed the EIS technique.
The EIS curves in the aqueous solution (figures 2(b) and S1) indicate an improved transfer impedance for EG after 100 cycles, as compared both with the initial EG cycle, as well as the reference SiC. The measured charge transfer impedance (R ct ) for SiC/Si was ∼160 Ω, whereas the initial R ct of the EG was around 60 Ω, reducing to about 20 Ω after 100 cycles.

On-skin measurements and surface conditioning
To further evaluate the electrochemical properties of the EG electrode for on-skin applications, the electrode was mounted on a pin button connector using double-sided carbon tape ( figure 3(a)). The mounted EG electrode (figure 3(b)) was then placed in contact with the skin surface in a three-electrode configuration as a sensor to measure the impedance spectroscopy. Figure 3(c) represents the circuit diagram of the on-skin EIS measurements and figure 3(d) shows the measurement setup. The WE (green) connected to the EG-based sensors, the CE (red) connected to a gold electrode, and the RE (white) connected to the Ag/AgCl wet electrodes. Gold was chosen as CE type thanks to its highly inert nature. Figure 4(a) shows the Nyquist impedance plot of the mounted on-skin EG sensor on skin with repeated testing. The skin-contact resistances with impedance reduced drastically after repeated tests and become stable from 3rd test. Figure 4(b) demonstrates the benchmarking of EG sensor to two types of commercial dry EEG sensors measured in exactly the same configuration (conductive foam-based, and springloaded gold pin sensors, provided by EEE Holter Technology Co., Ltd, Hsinchu, Taiwan). Note that both the foam and EG-based sensors have both the same extent of electrode area exposed to the skin (1 × 1 cm 2 ), so they can be directly compared, whereas direct comparisons with the spring-loaded sensor are more difficult because of its pin-contacts and bulky nature (see inset in figure 4(b)).
Results for the first and third repeated (sequential) tests of the EG sensors are also shown, for comparison. While the EG-based sensors do not show an impressive performance at the first test, the impedance resistance reduces drastically after repeated testing. By the third sequential test, the EG sensor reaches an impedance of about 130 kΩ at 50 Hz, making it superior as compared to both the commercial spring-loaded pin sensor (impedance around 830 kΩ at 50 Hz) and commercial conducting foam sensor (impedance around 665 kΩ at 50 Hz). Figure 4(c) shows the sequential evolution of the electrode-skin contact impedance values for the EG sensors at 50 and 100 Hz (average of five measured samples for each point). We can observe that stable impedance values are obtained approximately after two to three sequential tests ( figure 4(c)). The measured values of contact impedance once reached stability are found to be 550 ± 10 kΩ at 5 Hz, 435 ± 10 kΩ at 10 Hz, 130 ± 10 kΩ at 50 Hz, and 80 ± 5 kΩ at 100 Hz. The fluctuation of the values of skin contact resistance with dry electrodes are due to the varying skin moisture condition, potential skin abrasions, and the occurrence of flakes from the scalp [27]. Such values compare favorably with the state-of-the-art EEG sensors [45,46] as well as to previously reported graphene-based EEG sensors, as shown in table S4.
The electrochemical characterization indicates a consistent and marked reduction of contact impedance between the graphene sensors and the skin over repeated sequential testing, until a stable plateau is reached at impedance values that are less than one fourth of the initial ones ( figure 4(c)). The repeated tests after keeping the sensor on the air exposure for 10 min showed the similar behavior of reaching the stable values on 2nd test ( figure 4(c)).
In addition, a very similar behavior is also observed when the sensors are placed in contact with the aqueous 0.1 M PBS-soaked cellulose cloth in figure 4(d), with a marked reduction in impedance over time. Note that the contact impedance values with the cloth are overall substantially lower than those with the skin, as with average values around ∼6.5 ± 0.5 kΩ at 50 Hz, and ∼4 ± 0.5 kΩ at 100 Hz after reaching stability. This all indicates that the presence of a saline solution at the interface between the graphene electrodes and their contacts (skin or cloth) is very beneficial towards lowering the overall contact impedance.
Hereon, we refer to this phenomenon as 'surface conditioning' of the EG.

X-ray photoelectron analysis
There is very limited information available to-date regarding a possible on-skin surface conditioning effect of graphene EEG sensors. To this purpose, the EG electrodes have been further characterized to identify any physical and chemical changes after the repeated contact impedance tests in contact with the skin. Optical microscopy and SEM analysis do Through the deconvolution of the C1s peak of the pristine EG surface, we find the presence of an estimated 65.1, 25.4, 5.9 and 3.7 at.% of C-C (graphenic), Si-C, C-OH (chemisorbed water) and COOH (carboxyl) bonds, respectively. After repeated testing, a minor decrease of C-C bonds, and high relative decrement of Si-C bonds is observed, while the amount of COOH remains roughly the same, but a significant increment of C-OH occurs. This delivers a strong indication of the formation of surface or edge functionalization with hydroxyl groups on the EG after skin contact. The deconvolution of O1s XPS peaks was also carried out in figures 5(e) and (f). Three types of oxygen bond configurations are inferred: COOH at 531.8 eV, C-OH at 533 eV and chemisorbed water at 534.2 eV. On the pristine graphene, the ratios of the three types of bonds are about 29.8, 15.8 and 54.4 at.% for COOH, C-OH and chemisorbed water, respectively. After repeated testing, the COOH amount remains roughly unvaried, while a strong increment of C-OH (64.5 at.%) and decrement of chemisorbed water (8.3 at.%) were observed.

Surface conditioning mechanism
Typically, the formation of oxygen functional groups on the graphene surface leads to formation of defects on the basal planes [32]. Therefore, Raman spectroscopy was employed to explore a possible increase in the number of defects created after repeated testing, to reflect the larger amounts of functional groups. However, no significant change is observed in the D peak intensity nor the ratio of I D /I G ( figure S2(c)).   (b) Schematic of the chemisorption and physisorption of water on the surface of epitaxial graphene, proceeding through an increased functionalization of the EG grain boundaries with hydroxyl groups after contact with the skin. Over time, a thin layer of water molecules (boundary layer) develops to cover the EG grains, leading to a reduced contact impedance with the skin.
This suggests that the formation of the additional C-OH bond occurs mainly at the edges or grain boundaries of the graphene, rather than the basal plane, due to the higher surface energy of the defective grain boundaries promoting the formation of functional groups [47]. Interestingly, a similar surface functionalization phenomenon was observed in our recent study, where water splitting was purposely induced in-situ in a supercapacitor cell with graphene electrodes and acidic electrolyte (sulfuric acid and poly vinyl alcohol) [43]. This functionalization, promoted by the availability of free protons and hydroxyl groups in the electrolyte, led to an increased amount of C-OH and COOH functionalization on the graphene, which in turn, among others, greatly improved the wettability of the electrodes to the electrolyte and led to a marked enhancement of the double-layer capacitance.
Similarly, in this case, the increased hydrophilicity of the graphene leads to a better electrolyte wetting and consequently also an enhanced ion intercalation of the EG, hence a reduced contact impedance with the skin. We propose here a mechanism for the reduction of the contact impedance between the skin and the graphene electrode after prolonged contact.
To confirm the change in hydrophilicity originated by the functionalization, contact angle measurements were carried out on the EG sensor surface before any contact with the skin (pristine) and also after ten sequential tests (conditioning). It was observed that the contact angle went down from 73 ± 2 • for the pristine surface to 34 ± 2 • after surface conditioning (insets of figure 6(a)).
We suggest that during the graphene-skin measurements and contact, the skin electrolyte (sweat) leads to a chemical functionalization of the grain boundaries of graphene of oxygen functional groups (hydroxyl and carboxyl) [42]. Sweat is a body fluid emitted by the skin containing mainly ions such as sodium, potassium, calcium, magnesium, chloride and lactate, with a slightly acidic pH [48].
Defect-less graphene is highly hydrophobic and would not be easily wetted by sweat. However, this EG does possesses a relatively large amount of grain boundaries (∼90 nm average grain size), which are chemically active sites. Upon EG-skin contact, hydroxyl groups already present in the sweat solution can form C-OH bonds at the grain boundaries of EG ( figure 6). Subsequently, water molecules of sweat can be adsorbed on the surface (physisorption) starting from hydroxyls at the grain boundaries, and extending over time to cover-partly or fully-the grains forming a boundary layer held by weak van der Waals forces on the graphene surface. The decrease of the intensity of C-C and Si-C after repeated testing, is a further indication that a layer is being formed on the top of the graphene upon contact with the skin. This semisolid layer leads overall to an improved wetting of EG and facilitates ion intercalation in the graphene, leading to a substantial reduction of the contact impedance of the sensors with the skin.
To understand the stability of the surface water thin layer formed on the EG, we performed repeated cycles of (a) sequential measurements on-skin, followed by (b) 10 min of contact interruption with the skin /exposure to the environment, and subsequently (c) again sequential measurements on-skin ( figure 4(c)). This test indicates the contact impedance increases again after the contact with the skin is interrupted, likely due the evaporation of the water boundary layer. However, the low contact impedance with the skin can be restored again once the contact is re-established. There is also an indication of faster surface re-conditioning, indicating that once the additional C-OH functionalization is in place on the grain boundaries, the formation of the water boundary layer can immediately take place. Recent literature on water adsorption on EG is in line with the proposed mechanism of the formation of water boundary layer [49].
To emphasize the crucial effect of grain boundaries in this mechanism, the extent of surface conditioning upon sequential on-skin testing of the EG sensors was compared to that of sensors using defect-free LEG (figure S5), drop-casted on the silicon carbide on silicon. LEG has micron-sized grains, as compared to the ∼100 nm grain sizes of the EG. The EIS analysis (figure 6(a)) shows a markedly different behavior. While the EG electrodes show a dramatic ∼threefold reduction of contact impedance upon sequential testing, the reduction of contact impedance for LEG electrodes is limited to ∼1.5 (table S7). This analysis proves the importance of the defects/grain boundaries of the graphene in the surface conditioning mechanism. Hence, in order to obtain a maximum benefit in terms of contact impedance with the skin, the use of a graphene with relatively low grain size like the EG here used, is preferred to the use of large grainsize graphene.
In addition, delamination of the graphene from the sensor was observed using the LEG electrode ( figure S6). The additional advantage of the EG is its high adhesion to the substrate which prevents delamination [34]. This is further corroborated by the stability of the EG sensors upon long-term onskin contact as well as the resilience to corrosive saline environments. The long-term contact stability was observed by performing on-skin measurements on the forehead at an interval of 15 min without interrupting the contact with the skin over 2 h. The results showed a very good stability of the impedance values measured at 10 Hz ( figure S8(a)). The stability of the EG sensors in harsh corrosive conditions was also confirmed by measuring the sensors after immersion in a solution of concentrated sodium chloride solution (3 M) for up to 120 days ( figure S8(b)). The excellent stability is among others attributed to the high adhesion of the EG to its substrate.
Finally, open circuit potential (OCP) and possible potential shifts were also verified. The open circuit potential had been recorded for the graphene sensors (on-skin of the forearm) during 300 s at 0.5 samples s −1 , against an Ag/AgCl RE. The corresponding measurements indicate an OCP with the on-skin electrolyte of about 38 ± 6 mV, with negligible potential drift ( figure S9). This compares well with current literature using an equal input voltage of 5 mV, for example the OCP (∼58 mV) of Ag/AgCl coated polyurethane based dry electrodes [50].
Within the frequency range investigated in this work (0.1-1000 Hz), we did not observe any obvious DC potential shifts. The DC potential shifts are typically more pronounced, for slow EEG activity at very low frequency (<0.1 Hz) [51].

EEG helmet-testing
The typical EEG sensors rely on the detection of frequency generated from the brain by placing the electrodes over the head. To observe the performance of the EG sensors for EEG signal collection, initially we have placed the EG sensor on the forehead skin as a WE, together with silver and gold as RE and CE, respectively, all held tightly to the forehead using an elastic head band (inset figure 7(a)). The skin-contact impedance measured on the forehead without the elastic head band showed lower impedance value (90 ± 5 kΩ at 50 Hz, 128 ± 5 at 10 Hz and 145 ± 5 at 5 Hz) compared to the skin-contact impedance measured on the forearm (130 ± 10 kΩ at 50 Hz). With the aid of the elastic band (inset figure 7(a)) we observed further reduction of the skincontact impedance (44 ± 4 kΩ at 50 Hz, 68 ± 4 kΩ at 10 Hz and 76 ± 4 kΩ at 5 Hz) from the Nyquist Figure 7. (a) Nyquist plot of EIS on the forehead skin using EG sensor as working, Ag/AgCl as reference and gold as counter electrodes by attaching on the skin surface with an elastic head band (inset), (b) a skull map showing the location of the sensors used on the helmet, (c) comparison of signals collected with two channels on forehead using epitaxial graphene (EG), commercial pin and commercial foam sensors, (d) comparison of on-skin impedance values of the sensors mounted the helmet system. Note that a prolonged exposure of the polymer-graphene based electrodes to the skin would typically tend to degrade their performance, for example, because of the graphene's delamination [16]. In this case, exposure to sweat is beneficial as we observed the enhanced conductivity for saline soaked cloth, and we experience no delamination, as the EG displays high adhesion to the underlying SiC layer [34]. plot in figure 7(a). We attribute this improvement to the tighter sensor-skin contact ensured by the elastic band, the harder contact surface offered by the forehead and the sweat generated underneath the band. Finally, the EG-based EEG sensors have been mounted with the help of a simple pin button socket on a brain-interface helmet system ( figure 7(b)) [38]. The headset used for collecting the EEG signal has eight channels and the two channels at the forehead, which are in contact with the skin, were replaced by the EG based EEG sensors ( figure 7(b)). The graphene-based sensors also show the blinking signal in a similar way to the foam-based sensor ( figure 7(c)). This indicates the promising potential of the graphene-based EEG sensors for future brain interface applications. Figures 7(c) and (d) report a comparison of the signal and impedances from the two forehead channels of the helmet system when EG, commercial foam and commercial pin sensors are used. To evaluate the subject-to-subject variability, the tests of the EG sensors in the helmet system were conducted on a group of eight subjects. From the analysis of the collected data, bearing in mind the variability of skin and hair condition of the different subjects, a reasonable consistency was observed (table S9).

Conclusions
In summary, we demonstrated that EG fabricated on silicon carbide on silicon substrates can form robust and reliable electrodes with low contact impedance for brain-machine interface applications. Skin-contact impedance has a crucial effect on the performance of EEG sensors. A repeated sequential testing of the EG sensors leads to a dramatic reduction of over 75% of the contact impedance with the skin. We attribute this improvement to an in-situ formation of oxygen functional groups on the graphene grain boundaries upon exposure to the skin, as well as the formation of a thin boundary layer of physisorbed water covering the grains. We refer to this phenomenon as surface conditioning of the EG. This conditioning is consistent and repeatable, and is primed in saline environments. We anticipate that this knowledge may be used to develop a preconditioning procedure to further optimize the EG EEG sensors' performance. The EG sensors appear hence very competitive with current literature on dry sensors for EEG, EMG and ECG, particularly in terms of resilience, which is crucial to develop reallife reliable on-skin EEG sensor for brain-computer interfaces.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors.