Effect of Reduced Graphene Oxide Film Thickness on a Chemiresistor’s Response to Volatile Organic Compounds and Warfare Agents

Abstract

We explore the performance of a chemiresistor sensor array based on thin layers of reduced graphene oxide (rGO). The rGO is deposited with a spray coating technique to fabricate three samples of different layer thicknesses, which are characterized by atomic force microscopy (AFM) and Raman spectroscopy. We expose the chemiresistors to water vapor, three volatile organic compounds (VOC), ethanol, acetone, and formaldehyde, and two simulants of chemical warfare agents (CWA), dimethyl–methyl phosphonate (DMMP) and dipropylene glycol monomethyl ether (DPGME). The rGO-based sensors show noticeable changes in resistance upon parts per million variations of the analyte concentrations. The largest detection sensitivity 0.02%/ppm is observed with DPGME. Furthermore, we investigate a thickness-dependent signal that depends on the nature of the analyte. We show that comparing the signal measured with only a few rGO layers of different thicknesses can be used to distinguish formaldehyde from other VOC and DMMP from DPGME. Our findings represent a step toward the development of practical sensor arrays based on low cost, scalable graphene-based materials, enabling both sensitive and selective detection of analytes.

The future of health safety increasingly depends on our capacity to detect and recognize the presence of volatile compounds in housing units, research laboratories, workplaces, and warzones. For this reason, gas sensors are a primordial technology to monitor the presence of dangerous gases. Thin films are appealing materials to gas sensing technologies because of their fast response and recovery behaviors due to their electronic properties, which are easily modified by the presence of analytes attached to their surface. Thin films can also be incorporated into compact-size sensors to minimize their footprint and energy consumption. Metal oxide semiconductors (MOS) have been extensively studied as active sensing layers due to their high sensitivity to chemicals, notably to CO, H₂S, NO₂, and HCHO. − However, MOS-based gas sensors perform relatively poorly for selective detection and require a high-temperature operation to reach optimal sensitivity. To address this issue, graphene and its derivatives represent a viable option for sensitive detection of volatile chemicals. These materials are particularly advantageous because of their electrical and mechanical stability, high surface area to volume ratio, ease of surface modification, , tunability by optical illumination, and sensitive response to volatile chemicals at room temperature. , Especially, graphene oxide (GO) and reduced graphene oxide (rGO) are promising derivatives of graphene for deployable sensing technologies because they can be fabricated with low-cost wet chemical methods and then deposited on various substrates with scalable spray coating or spin coating techniques. −

Compact chemiresistors based on GO, fabricated using scalable and cost-effective techniques like spray coating, show promise as the next generation of portable, drone-compatible environmental and safety monitoring sensors. Previous work demonstrated that using a reliable sensing reference can help mitigate the effects of environmental condition fluctuations, enhancing the sensor’s applicability in real-world settings. These sensors are ideal alternatives to bulky, high-cost methods like THz spectroscopy and mass spectrometry, which offer high sensitivity and selectivity at the expense of cost and footprint. Therefore, our results are relevant for evaluating the performance of real-world application devices.

Previous works demonstrated the use of GO and rGO as active materials of a chemiresistor architecture, notably to detect volatile organic compounds (VOC) such as methanol, isopropanol, ethanol, acetone, and formaldehyde. − For example, a drop-casted rGO layer showed a change in resistance between 1 and 2% when exposed to 1000 ppm concentration of ethanol, acetone, isopropanol, or methanol. , Improving the performance of GO- and rGO-based chemiresistors has since become an active field of research. Sensitivity enhancement to some chemicals has been achieved by chemical functionalization, ,, varying the degree of oxidation, or forming a heterojunction with TiO₂, CuO, and SnO₂. ,, For instance, mixing rGO with TiO₂ can result in a 0.4% increase in resistance change when a sensor is exposed to a subppm concentration of formaldehyde. Hybrid films composed of rGO and iron-doped WO₃ have also been explored, where varying the rGO thickness enabled responses of up to 5% when exposed to 1 ppm of NO₂. The literature is scarce on the use of GO and rGO to detect chemical warfare agents (CWA), but those results are nonetheless promising. For example, the sensitive detection of dimethyl–methyl phosphonate (DMMP), a simulant of the chemical warfare agent sarin, was reported using a drop-casted rGO active layer. Under optimal conditions, such chemiresistors produced a response reaching up to a 76% change in resistance when exposed to only a 40 ppm concentration. Improved sensitivity performances to DDMP detection was also demonstrated when a p-phenylenediamine reduction agent is used instead of hydrazine. These results highlight the potential of GO- and rGO-based chemiresistors to detect and potentially identify chemical warfare agents (CWA). Traditional sensors proposed for CWA detection often rely on metal oxide-based sensors, and surface acoustic wave devices. , However, these technologies typically require high operating temperatures, involve complex fabrication processes, or are bulky, which can limit their applications in certain settings. In contrast, our work with the simulants DMMP and dipropylene glycol monomethyl ether (DPGME) demonstrates that sensor arrays based on rGO offer a simpler and cost-effective alternative that can operate at room temperature to achieve significant detection sensitivity. Additionally, the tunable thickness of the rGO layers allows for selective detection without the need for additional functionalization or complex multilayer structures. This unique property provides a competitive advantage by enabling tailored sensor responses to specific target analytes through simple adjustments in the rGO layer thickness, which is less feasible in traditional sensor platforms.

In this article, we examine the sensitivity of chemiresistors relying on a thin film of rGO to detect volatile chemicals relevant to health and security applications. To investigate the impact of rGO layer thickness on the gas sensing performance of a chemiresistor, we fabricated three rGO thin films on Si/SiO₂ substrates, with film thicknesses of 7, 13, and 30 nm. In the first part of this article, we present the rGO thin film fabrication technique and then show the structural and chemical characterization results by atomic force microscopy (AFM) and Raman spectroscopy. We then show resistance measurements obtained when exposing the devices to three volatile organic compounds (VOC): ethanol, acetone, and formaldehyde, as well as two simulants of CWA: DMMP and DPGME. Water vapor is also considered as a reference. To the best of our knowledge, this is the first research on the layer-thickness dependency of an rGO-based chemiresistor. We also show the first results for the detection of DPGME with a chemiresistor. More importantly, we show that an array composed of different rGO layers with varying thicknesses can be used to differentiate the analytes. This key feature for gas identification is demonstrated by comparing distinguishable array signals obtained with formaldehyde and other VOC, and with DMMP and DPGME. This study is designed as a controlled proof-of-concept to investigate how variations in rGO deposition conditions influence sensing selectivity rather than to replicate field-deployment conditions or achieve the lowest possible detection limits.

Experimental Section

Synthesis of Reduced Graphene Oxide Thin Films

The sensor array fabrication is based on chemically reduced GO using hydrazine hydrate in a liquid route, as described in previous work. This reducing agent is extensively used and capable of efficiently reducing functional groups attached to the basal plan as well as the edges of GO. We mixed a nanocolloidal liquid of GO dispersion (Sigma-Aldrich) at a 0.5 mg/mL concentration to a liquid phase of pure hydrazine hydrate (N₂H₄) (Sigma-Aldrich) with a ratio of GO/N₂H₄ equal to 1:0.04 (4% of GO volume). The precursor solution is allowed to rest for 24 h during which the color changes from dark brown to black, revealing the formation of suspended rGO. The dispersion is then sonicated for 30 min at 40 kHz to facilitate the mechanical exfoliation of rGO layers in the solvent. We rely on an airbrush (VH174, VIVOHOME) and a spray coating technique to produce thin film layers on the Si/SiO₂ substrates. Three samples with different thicknesses are obtained using 5, 10, and 15 mL dispersion volumes, which are sprayed on distinct substrates heated at 150 °C.

Sensor Array Fabrication

Chemiresistors are obtained after depositing electrodes on the surface of the rGO layers, allowing our experimental setup to monitor resistance changes as a function of analyte concentration. We use e-beam physical vapor deposition (Angstrom Nexdep Series) and a shadow mask to create two 1 cm long rectangular electrodes separated by 1 cm. We used three metals: Ti, Pd, and Au, with thicknesses of 5, 20, and 120 nm, respectively, ensured good adhesion and electrical conductivity to the rGO thin film. The sensor array is obtained by physically and electrically connecting the three sensors to a circuit board for convenient monitoring of the rGO resistance to allow simultaneous monitoring of the rGO electrical resistance properties, while all three films are exposed to the same experimental conditions.

Material Characterization Methods

We investigate the morphologic properties of the sample with atomic force microscopy (AFM) using a Park NX 10 instrument. Chemical and structural properties are then explored with Raman spectroscopy (XploRa Plus from Horiba Scientific) using a 3.6 mW light source at 532 nm. Gas sensing measurements utilize the setup and technique detailed in previous work. Briefly, gas detection experiments are performed by placing the sensor array in a 250 mL chamber linked to three gas lines with mass flow controllers (MFCs). The first line purges the chamber with pure nitrogen, which is used both as the carrier gas and to purge the sensor chamber before starting gas sensing experiments. The two others are connected to two bubblers containing liquid chemicals and can be employed to create diluted vapor of a target gas, while the two bubbling systems are maintained at a constant temperature of (22.5 ± 0.5) °C. The concentrations of the analytes are fixed by adjusting the flow rates of the MFCs and the gas flow can be adjusted as needed to achieve target analyte concentrations. The sensors are exposed to a fixed concentration of analyte for 10 min followed by a recovery time of 30 min before proceeding to a different concentration or a different analyte. We recorded the voltage/current ratio of each chemiresistor using a Keithley 2400 as the voltage is kept constant at 1 V. We extracted the slope to obtain the resistance. Table lists the details of the tested target gases with their concentration range. Although we purge the sample box for >30 min before the start of each experiment, a small but consistent reduction of the relative humidity inside the sample chamber induces a gradual drift of the chemiresistor’s resistance. We subtract this baseline utilizing an exponential decay fit on the data collected during purging time.

1. Chemical Formula, Vapor Pressure, and Varied Concentrations of the Different Tested Chemicals.

analyte	chemical formula	vapor pressure (Pa) at room temperature	concentration (ppm)
humidity	H2O	2455	200–600
ethanol	C2H6O	6132	1000–2000
acetone	C3H6O	24,598	1000–2000
formaldehyde	CH2O	870	50–150
DMMP	C3H9O3P	78.54	10–30
DPGME	C7H16O3	53.33	10–30

Results and Discussion

To investigate the effects of the rGO layer thickness on the gas sensing performance of a chemiresistor, we fabricate three rGO thin films by spray deposition using different volumes of the precursor solution. AFM measurements shown in Figure are performed to determine the film thickness and investigate the surface topology. We observe that the average thicknesses of the films are 4, 13, and 30 nm for deposited volumes of 5, 10, and 15 mL, respectively. We subsequently refer to these films as rGO₄, rGO₁₃, and rGO₃₀, where the subscripts represent the average thickness in nm. The micrographs also reveal that the deposition volume is linked to clusters and flakes of different sizes. Sample rGO₄ (Figure a) features relatively small flakes of varying sizes with a length between 0.4 and 2 μm. The formation of these flakes can be attributed to nanolayers of GO sticking together during the reduction process. The surface of rGO₁₃ (Figure b) shows the formation of a few bubble-like structures of various sizes, with maximum widths and heights of 5 μm and 160 nm, respectively. Sample rGO₃₀, corresponding to the thicker layer (Figure c), has an inhomogeneous surface also featuring bubble-like structures of even larger sizes with widths and heights reaching up to 10 μm and 800 nm, respectively. The formation of such microbubbles has previously been observed in previous work on rGO, implying local hydration during the reduction process. This phenomenon has also been observed in the fabrication of drop-casted rGO layers and the level of spatial inhomogeneity agrees with results obtained from previously reported GO-like samples. ,

1.

AFM analysis and their corresponding height profile of rGO thin film obtained after the deposition of (a) 5 mL (rGO₄), (b) 10 mL (rGO₁₃), and (c) 15 mL (rGO₃₀), dispersion.

We perform Raman spectroscopy analysis to investigate the chemical properties of the rGO films. Figure a–f presents the first- and second-order spectra for rGO₄, rGO₁₃, and rGO₃₀. All data are normalized to the D peak and deconvoluted utilizing the Lorentzian function. The precise positions of the peaks for the three samples investigated in this work are listed in the Supporting Information. Their positions are found at their expected spectral range position for carbon-like materials. , The Raman spectra of rGO₄, rGO₁₃, and rGO₃₀ are shown in Figure a–c, respectively, and are decomposed into 5 peaks, namely, I, D, D″, G, and D′. In general, we observe that the three samples exhibit comparable spectra, indicating similarities in their structures and composition. However, by delving into their differences, we can uncover some variations seemingly tied to the deposition process of the rGO thin layer. Most notably, the D peak of rGO₄ has the narrowest line width of the three devices, by 11 cm^–1 at full-width at half-maximum (FWHM), pointing at less disorder in that layer. Consistently, the G peak for rGO₁₃ and rGO₃₀ is also red-shifted by ∼15 cm^–1 from the spectral position of 1595 cm^–1 observed with rGO₄ indicate more defects in those two samples. The I, D″, and D′ can also be associated to disorder and their intensity increases with the thickness of the rGO layer. The second-order Raman spectra (Figure d–f), deconvoluted into 4 peaks, also support these conclusions. The 2D peak is indicative of a graphene structure and, although it is found in all samples, it is slightly more pronounced in rGO₄. Finally, a weak D+D” peak visible in rGO₁₃ and rGO₃₀ indicates the presence of defects. A more detailed discussion on the Raman spectrum and the origin of the peaks can be found in the Supporting Information. In summary, all Raman spectra show similarity, indicating that the chemical compositions of the three rGO samples are comparable in our experiments. However, rGO₄ appears to have slightly smaller crystallites and less disorder than the other two samples. These results agree with AFM measurements, showing a rougher morphology associated with larger crystal sizes for rGO₁₃ and rGO₃₀. In summary, although all thin films are fabricated from a spray coating technique using the same solution, the variation of the deposited sample thickness induces distinguishable morphologies and small chemical changes revealed by Raman spectroscopy.

2.

Deconvolution of Raman spectra of the first (a–c) and second order (d–f) on the first and second row, respectively. Results obtained with samples rGO₄ (a, d), rGO₁₃ (b, e), and rGO₃₀ (c, f) are shown from left to right in different columns.

Gas sensing experiments are performed with three chemiresistors using the rGO thin films described above as the active sensing layer. We investigate the dependence of rGO thickness on the gas detection sensitivity as a function of analytes and their concentrations. We consider the following analytes: water vapor, common VOC (ethanol, acetone, and formaldehyde), and CWA simulants (DMMP, and DPGME). The response is defined as the relative change in resistance ΔR/R ₀ = (R – R ₀)/R ₀ with R ₀ representing the baseline resistance value. Figure shows results obtained with devices rGO₄, rGO₁₃, and rGO₃₀ (black, red, and blue lines, respectively) as we successively exposed them to increasing concentrations of analytes. These exposures are separated by a recovery period, during which the sample chamber is purged with pure nitrogen, effectively resetting the sample to its initial conditions. In general, we observe that thinner rGO layers lead to larger maximum values for |ΔR/R|. This result is consistent with the fact that analytes preferentially affect the surface properties of the rGO layer. Since the measured R depends on the bulk properties, a thicker sample of rGO is therefore expected to be less sensitive to the presence of analytes because the majority of the material, which is located away from the surface, remains unaffected. As we increase the analyte concentration, the maximum values |ΔR/R| also increase accordingly. Among VOC, formaldehyde shows the largest thickness dependence, and among CWAs, DMMP also demonstrates a similar trend. Our results present a selectivity mechanism that utilizes the thickness of the rGO layer to identify different chemicals. We present a detailed account of the experimental results in the next paragraph followed by an analysis and discussion.

3.

Signal measured with chemiresistors using rGO₄ (black), rGO₁₃ (red), and rGO₃₀ (blue) as an active layer when exposed to different concentrations of (a) water vapor at 200, 400, and 600 ppm, (b) ethanol at 1000, 1500, and 2000 ppm, (c) acetone at 1000, 1500, and 2000 ppm, (d) formaldehyde at 50, 100, and 150 ppm, (e) DMMP at 10, 20, and 30 ppm, and (f) DPGME at 10, 20, and 30 ppm.

Figure a presents the response of the three chemiresistors to water vapor, with resistance increasing in the presence of humidity. Previous work showed that rGO samples exhibit n-type conductivity when they are reduced with hydrazine hydrate. , As such, water vapor acts as an oxidizing gas that accepts electrons from the surface, thereby reducing the conductivity of the sensitive layer. Additionally, the formation of hydrogen bonds between water molecules and epoxy and hydroxyl further facilitates electron acceptance from the surface. The few percent changes at a concentration ranging from 200 to 600 ppm are also consistent with those previously reported with graphene field-effect transistor and other rGO-based integrated sensors. Figure b–d presents the signal recorded when the rGO-based sensor array is exposed to different concentrations of VOC. We explore the range of 1000–2000 ppm for ethanol (Figure b) and acetone (Figure c), as well as the range 50–150 ppm for formaldehyde (Figure d). At the highest VOC concentrations, the peak signals recorded after exposure to ethanol, acetone, and formaldehyde correspond to a 0.58, 0.6, and 1.23% modulation, respectively. These measurements are consistent with previous work. ,,, For a given concentration of VOC, rGO-based chemiresistors display signals comparable to those obtained with sensors based on graphene , and phthalocyanine-functionalized graphene. Figure e,f shows the rGO-based sensor’s response to CWA simulants (DMMP and DPGME) at concentrations ranging from 10 to 30 ppm. The increase in resistance upon exposure to the analytes indicates the oxidizing nature of these gases. Due to the higher detection sensitivity of the chemiresistor to these two analytes, in comparison to VOCs, a much lower concentration is used to reach maximum signals of 0.21% for DMMP (at 30 ppm) and 0.6% for DPGME (at 30 ppm). In our experiment, rGO-based sensors display the highest sensitivity to DPGME in terms of change in resistance per ppm, with a value approaching 0.02%/ppm as measured with rGO₄. For DMMP, the detection sensitivity of 0.007%/ppm is significantly lower than values previously reported using graphene-based chemiresistors, , porous reduced graphene oxide multilayer frameworks, phenylenediamine reduced graphene oxide, and lead-free perovskite nanocrystals decorating graphene.

Using the data set shown in Figure , we investigate the effect of the rGO layer thickness on the chemiresistor’s signal. We first extract the signal amplitude by extracting the increase in resistance during the 10 min of exposure to an analyte. We then calculate the two parameters α and β, which correspond to the ratio of the maximum signal amplitude measured with rGO₁₃ and rGO₄, and rGO₃₀ and rGO₄, respectively. In Table , we list these ratios after averaging them over the three concentrations considered in the experiment. The error corresponds to the standard deviation. This analysis allows us to test whether the detection of an analyte, which is based on a nontrivial electrical interaction at the interface of the rGO, can be altered by changing the thickness and morphology of the sensing layer. For a given analyte concentration, results obtained with different analytes systematically show a lower resistance peak (see Figure ) for thicker rGO layer thicknesses. Notably, with formaldehyde, we observe the largest ΔR/R ₀ variations among the three rGO devices. One may expect that water is less sensitive to the thickness since it has been shown to have a larger permittivity through a submicrometer-thick GO layer in comparison to acetone and alcohols. Our data do not support this assumption as we observe negligible differences between the values α and β measured with water vapor, ethanol, and acetone. In contrast, the values of α and β obtained with formaldehyde are noticeably smaller than those obtained with the other VOCs. This suggests that a sensor with a thick rGO layer exhibits significantly lower sensitivity to formaldehyde compared to a similar sensor with a thinner layer, whereas both configurations show comparable sensitivity to ethanol and acetone. This feature can be used to distinguish formaldehyde from other analytes. Differences in the parameters α and β between the two CWA simulants are even more dramatic. DMMP shows the strongest dependence on the layer thickness, while the signal produced by DPGME corresponds to the largest value of α across all analytes investigated in this experiment. Our study shows that a chemiresistor array composed of rGO layers with varying thicknesses and morphologies can produce analyte-specific and concentration-dependent responses. The thickness-dependent behavior may be partially explained by differences in the analyte permeability and diffusion time through the rGO layer. This interpretation is supported by the observation that thinner rGO layers exhibit a more rapid and pronounced response immediately after analyte exposure. Morphological and chemical variations, as revealed by AFM and Raman spectroscopy, may also contribute to the observed selectivity. For instance, increased porosity in thicker rGO layers could shorten diffusion paths, potentially mitigating some of the expected thickness-related effects. Additionally, subtle differences in defect density and structural disorder may influence electron transport and adsorption dynamics, further modulating the sensor sensitivity.

2. Parameters Calculated by Dividing the Maximum Resistance Change Observed with the Samples rGO₁₃ and rGO₄ (α), and rGO₃₀ and rGO₄ (β).

analyte	α	β
water	0.85 ± 0.02	0.47 ± 0.01
ethanol	0.86 ± 0.01	0.49 ± 0.02
acetone	0.90 ± 0.05	0.46 ± 0.05
formaldehyde	0.76 ± 0.02	0.42 ± 0.07
DMMP	0.78 ± 0.01	0.33 ± 0.08
DPGME	0.90 ± 0.02	0.42 ± 0.03

In the last experiment, we evaluate the reproducibility of the sensor responses. This is achieved by exposing each rGO layer to an analyte, followed by water vapor. We repeated this process two times while checking for consistency. As shown in Figure , the sensitive layer’s conductivity exhibits a reversible characteristic after exposure to all gases, demonstrating the potential of rGO-based chemiresistors to perform continuous detection of different analytes over extended periods. Moreover, between experiments, we collected measurements under the same conditions to ensure that the three samples considered in this study consistently display the same steady state resistivity and feature the same detection sensitivity to gas (see Additional gas sensing analysis of the Supporting Information). These results validate our experimental methodology, which involves using the same chemiresistors to investigate the response to various analytes and concentrations. While response time was not a primary focus of this study, it could be improved in future work by optimizing enclosure and electrode geometry, reducing rGO film thickness, introducing a prediction algorithm, , and mixing functional material nanomaterials to rGO, such as copper phthalocyanine nanoflowers.

4.

Reversibility test of the sensor array to different chemicals utilizing alternate measurements of vapor of water at 600 ppm with (a) ethanol at 2000 ppm, (b) acetone at 2000 ppm, (c) formaldehyde at 150 ppm, (d) DMMP at 30 ppm, and (e) DPGME at 30 ppm.

Conclusions

We demonstrate a chemiresistor array based on thin layers of reduced graphene oxide (rGO) to detect volatile organic compounds (VOC) and chemical warfare agent (CWA) simulants. The active sensing layers are fabricated with a cost-effective and scalable fabrication technique relying on the chemical reduction of GO and deposition by spray coating to produce thin films with varying thicknesses and morphology. Atomic Force Microscopy (AFM), used to quantify the layer thickness and structural properties, shows the formation of microbubbles in the thicker samples. Raman measurements confirm that the chemical compositions of all samples considered in this work are similar, but thicker samples generally show a higher degree of disorder, oxidation, and defects. Interestingly, we demonstrate that the combination of several rGO layers with different thicknesses within a chemiresistor array produces a resulting signal indicative of the nature of the analyte and not only of its concentration. Our concept relies on a scalable spray coating fabrication technique, offering low cost, high throughput, and compatibility with large-area substrates. These attributes, combined with a simple device architecture, make this approach particularly well-suited for developing cost-effective and disposable sensors for real-world applications. Although this work does not examine device-to-device reproducibility in achieving consistent rGO thickness, any related challenges could be addressed in practical settings through sample characterization and postfabrication screening. To ensure practical viability, extensive testing under practical conditions will be required to validate the sensor’s performance, robustness, long-term stability, and reliability in diverse environments, including humid conditions. Studies, such as those by Park et al., have shown that employing sensing references can reduce the impact of environmental fluctuations, boosting the sensor’s effectiveness in practical applications. Our results suggest a promising approach for the design and fabrication of high-sensitivity gas detectors able to identify a gas mixture for health and security applications. Future studies involving a broader analyte set will be instrumental in uncovering additional pathways for chemical identification and improving selectivity. Such advancements will ensure that the technology can be deployed in a wide range of applications, including environmental and safety monitoring, with the potential for integration into portable drone-mounted systems. The devices used in our experiment show consistent response but structural changes of GO over time have been observed. , Therefore, further studies on the long-term stability of rGO-based sensors are required to ensure that these devices can be stored or deployed over extended periods of time.

The data are available from the corresponding author upon reasonable request.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsomega.5c07277.

• Detailed Raman spectroscopy analysis; gas sensing measurements demonstrating reproducibility over an 18-day period; raw experimental data of the gas sensing measurements displayed in Figure ; electrical characterization of the rGO thin films (PDF)

All authors conceived the experiment. I.B. collected experimental measurements and wrote the first draft. I.B. and J.-M.M. performed the analysis. N.B. and J.-M.M. reviewed and edited the manuscript. All authors have read and agreed to the published version of the manuscript.

The authors declare no competing financial interest.