Introduction

Optical and electronic properties of graphene can be modulated by continuously tuning the charge carrier density using electrostatic gating1,2,3,4,5,6, electrochemical doping7 or charge transfer by adsorption of molecular species8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23. Besides electronic transport, the doping of graphene has a marked effect on the fundamental electron-phonon interactions, like the breakdown of the adiabatic Born-Oppenheimer approximation1,2,3,4, the interplay between adiabatic and non-adiabatic effects23 or the interference between all the quantum pathways involved in inelastic light scattering5. These effects have been conveniently investigated by Raman spectroscopy, which is also sensitive to the number of layers24, their stacking ordering25,26,27,28, the nature and density of defects29,30,31 and the in-plane strain variations32,33. As a consequence, Raman spectroscopy of active modes in graphene, like the G and 2D bands, is being considered as a high-throughput technique to characterize graphene and to probe the inelastic light scattering phenomena. Moreover, several groups24,34,35 have reported that no significant spectral changes were observed by collecting Raman spectra of graphene exfoliated on SiO2/Si substrate, with a laser power Plaser ranging from 0.04 to 4 mW. Therefore, Raman spectroscopy is considered as non-invasive when performed with Plaser in the mW range.

Yet, the non-invasive character of Raman spectroscopy has been questioned by the reports of possible photo-induced effects and some authors recommended to use more cautious experimental conditions (laser power limited to 70 μW36 or Ar annealing34). Laser irradiation has even been shown to induce irreversible damages of graphene37,38. Here, we report that the charge carrier density of exfoliated graphene lying on a hydrophilic SiO2/Si substrate can be finely and reversibly tuned optically with visible photons. The influence of the laser power and the surface chemistry of the SiO2/Si substrate on the optical doping of exfoliated graphene in air is examined. Our results also imply that Raman spectroscopy of graphene performed with usual experimental conditions (1 mW under a 100× objective) can be invasive as it directly alters the charge carrier density.

Results

In a first series of experiments, Raman spectra of exfoliated graphene excited with a 532 nm laser are collected as a function of the incident laser power, Plaser, at a fixed location. The results obtained for three different types of samples are compared. F1 was exfoliated on a hydrophilic SiO2/Si substrate, F2 on a less hydrophilic substrate and F3 was suspended over a trench etched into the substrate (see methods). The quality of the samples is preserved during the whole experiments, as evidenced by the absence of the D band (see Supplementary Fig. S1c). All Raman spectra are fitted by Lorentzian functions to extract the position of the G band (ωG), the full width at half maximum (FWHM) of the G band (ΓG), the 2D band position (ω2D) and the ratio between the integrated intensities of the 2D and G bands (A2D/AG).

For the sample F1, Plaser is increased from 0.05 mW up to 1.5 mW and decreased back to 0.05 mW, as shown in Figure 1. The G band at 1580 cm−1 for Plaser = 0.05 mW downshifts to reach a minimum value 1576.5 cm−1 for Plaser = 0.5–0.6 mW. With the further increase of Plaser, it upshifts up to 1579 cm−1 (Fig. 1a). Correspondingly, ΓG rises from 7 cm−1 up to 12 cm−1 for Plaser = 0.5–0.6 mW and then decreases to 10 cm−1 (Fig. 1b). In the same conditions, the 2D band continuously downshifts from 2666 cm−1 down to 2663 cm−1 (Fig. 1c). The A2D/AG ratio displays a maximum for Plaser around 0.5–0.6 mW (Fig. 1d). The comparison of the concomitant evolution of ωG, ω2D, ΓG and A2D/AG with electrostatic gating results1,2,3,4 indicates that the graphene doping is modified upon light irradiation. In particular, the coincidence of the maxima of ΓG and of A2D/AG with the minimum of ωG is a signature of neutral graphene. The doping type is determined from the evolution of ω2D3,4: a larger upshift for p-doping (below 0.5–0.6 mW) than for low-level n-doping (above 0.5–0.6 mW). Interestingly, the changes are observed to be mostly reversible when Plaser is decreased back to its initial value (Fig. 1). The doping of graphene deposited on a hydrophilic SiO2/Si substrate can be continuously and reversibly tuned from the initial p-doping to quasi-neutral and eventually n-doping with moderate laser power. From the shifts and widths of the G band3,4, the inital doping of F1 was estimated to be p ≈ 4 × 1012 cm−2 at 0.05 mW. For Plaser = 1.44 mW, the carrier density was estimated to be n ≈ 3 × 1012 cm−2. Similar observations were made with different laser wavelengths: 488 nm, 514 nm and 660 nm (see Supplementary Figs. S3, S4, S5 and S6).

Figure 1
figure 1

Reversible evolution of Raman spectra as a function of the incident laser power Plaser for the graphene flake F1.

(a) the G band position ωG, (b) the G band FWHM ΓG, (c) the 2D band position ω2D and (d) the integrated intensities ratio A2D/AG. The graphene flake F1 is exfoliated on a hydrophilic substrate. Plaser is increased from 0.05 mW up to 1.5 mW and decreased back to 0.05 mW as shown by the arrows in (a). The color code of each point corresponds to the chronological order in which the measurements have been carried out as depicted on the right hand side color bar (for the corresponding time see Supplementary Fig. S2).

Full size image

We now examine the influence of the underlying substrate by comparing these results to the ones obtained on sample F2 deposited on a less hydrophilic substrate and to sample F3 suspended over a trench etched into the substrate. The relative variations of the 2D and G band positions as a function of Plaser are collected in Figure 2. This particular representation disentangles doping and strain effects33. We clearly observe three different behaviors in the plots presented on Figure 2. When the doping of graphene is continuously tuned from p-type to n-type, the 2D versus G positions presents the particular v-shape shown in Figure 2a. The G band consistently upshifts with an increasing charge carrier density (for both electrons and holes). The 2D band downshifts monotonically as the doping varies from p to n. Figure 2b displays the 2D vs G behavior for a flake lying on a less hydrophilic substrate. The G and 2D bands continuously upshift as Plaser increases. The small upshift of the 2D band is expected for low level n-type doping3,4,23. Here, ΓG is maximum for the lowest incident power and continuously decreases with the increasing laser power and so does the A2D/AG ratio (see Supplementary Fig. S7 and S9). Considered together, these variations indicate that this supported graphene is close to neutral for the lowest Plaser and becomes n-doped as Plaser is increased (for Plaser = 2.6 mW, n ≈ 4–5 × 1012 cm−2 from refs. 3 and 4). Once again, these doping variations are reversible when Plaser is decreased back. For the suspended graphene flake, F3, the high values of ΓG (14 cm−1) and A2D/AG ratio (9.5) remain constant through the entire power sweep (Supplementary Fig. S8). This indicates that F3 remains neutral as the laser power increases in agreement with S. Berciaud et al.35 Thus, the small shifts of ωG and ω2D (Fig. 2c) cannot be ascribed to any doping level variation.

Figure 2
figure 2

Comparison of the relative evolutions of the 2D band position (ω2D) versus the G band position (ωG) as a function of Plaser for three graphene flakes (F1, F2 and F3).

F2 was exfoliated on a less hydrophilic SiO2/Si substrate than F1. F3 was suspended over a trench etched into the substrate. The color code of each point corresponds to the incident laser power Plaser as displayed on the right hand side color bar. (a) F1 is p-doped at low Plaser, it becomes quasi-neutral around 0.5 mW and n-doped for higher Plaser. (b) F2 is initially quasi-neutral and becomes n-doped with the increasing Plaser. (c) The suspended graphene flake, F3, is neutral and stays neutral with the increasing Plaser. The measured shifts for F3 are only due to laser heating effects. The different G band and 2D band positions (ωG, ω2D) of the different graphene samples (F1, F2, F3) in their quasi-neutral state are attributed to strain fluctuations from one sample to the other.

Full size image

We can now wonder whether these laser-induced doping variations are uniform across the graphene surface. To this end, Raman mapping was performed at different laser powers on graphene exfoliated onto a hydrophilic substrate and partially suspended over a pre-patterned trench (sample F4, see optical microscopy image in supplementary Fig. S1b where the mapped area is shown). Maps of the ωG, ΓG, ω2D and A2D/AG are displayed on Figure 3 for seven Plaser ranging from 0.1 mW to 3 mW. The constant position of the G and 2D bands across the whole supported area of the sample indicates that the strain and doping are homogeneous, therefore allowing to single out the influence of optical doping by the varying laser power. When the laser power is increased from 0.1 mW to 1.5 mW, the G band, initially at 1584 cm−1 (ΓG = 7.5 cm−1) shifts down to 1582 cm−1 (ΓG = 12 cm−1). Further increase of the laser power inverts the effect and the G band upshifts back to 1583.5 cm−1 (ΓG decreases down to 9.5 cm−1). For the same power sweep, the 2D band continuously downshifts from 2676 to 2672 cm−1. Similarly, the A2D/AG ratio increases from 4.7 for Plaser = 0.1 mW, to 5.1 for Plaser between 0.5 and 1.5 mW and then decreases back to 4.7 for higher laser powers. These observations confirm that graphene supported on hydrophilic SiO2 evolves continuously from p-type to n-type doped graphene when the laser power is increased.

Figure 3
figure 3

Raman maps measured for 7 different laser powers on sample F4.

The rows correspond from top to bottom to the G band position ωG, the G band FWHM ΓG, the 2D band position ω2D and the integrated intensities ratio A2D/AG. The graphene flake is covering an etched pool of the substrate that has a capsule shape and is visible on the right hand side of each map. The black upper left corners correspond to bare SiO2/Si surfaces. A greyscale version of this figure is available as Supplementary Fig. S10.

Full size image

Interestingly, the irradiation of the suspended graphene yields a very different behavior. Indeed, Figure 3 shows that both G and 2D bands downshift from 1581.5 to 1578.5 cm−1 and from 2673 to 2663.5 cm−1 respectively as the laser power is increased. However, ΓG and the A2D/AG ratio remain constant (13 cm−1 and 5.9 respectively). Therefore the doping level remains unchanged through the entire power sweep and the downshifts of the 2D and G bands are ascribed to classical laser-induced heating effects that are also observed on sample F3 (see Supplementary Fig. S8, S11 and S12).

A careful examination of the signals recorded on the graphene edges and on the trench contour reveals that the optical doping is less effective in these specific locations. In particular, the doping never reaches n-type even for the highest Plaser. We attribute this local effect to the supplementary p-doping near graphene edges which has already been observed by Raman spectroscopy39,40.

Discussion

The absence of doping upon irradiation of suspended graphene clearly shows that the substrate contributes to the mechanism involved in the optical doping of graphene. We thus investigated the influence of substrate cleaning procedures on the efficiency of the optical doping. For highly hydroxylated (hydrophilic) substrates, freshly cleaned by O2 plasma and/or piranha treatments prior to graphene deposition, we observe the ambipolar behavior with a doping evolving from p-type to n-type upon increasing laser power. Our results confirm that graphene exfoliated on a hydrophilic substrate is p-doped41,42 in contrast to the quasi-intrinsic state of suspended graphene35. Across the different samples studied here, we found that the Plaser for which graphene is neutralized falls in the range 0.5–2 mW. These values depend on the laser power density (laser spot size, see Supplementary Fig. S14) and graphene absorption (laser wavelength and SiO2 thickness, see Supplementary Figs. S3, S4, S5 and S6). The density of hydroxyl groups can be reduced by performing a thermal annealing of the samples under argon before graphene deposition. In this case, graphene was observed to be initially quasi-neutral and n-type doping was obtained upon illuminating the sample with increasing Plaser. Finally, when graphene is transferred onto substrates used as-received without any cleaning, the doping of some flakes remain constant for all the entire Plaser range (not shown). Therefore, the absence of any significant spectral changes previously observed24 might be accounted for by the deposition of graphene on as-received substrates. Finally, it should be pointed out that the optical doping is not specific to graphene micromechanically exfoliated onto SiO2/Si substrates but similarly occurs for graphene deposited on standard glass (not shown).

It thus appears that the starting value of the graphene doping level is directly linked to the substrate preparation. Cleaning procedures that render the substrate hydrophilic tend to result in more p-doped supported graphene, in agreement with previous studies43,44,45,46,47, which also demonstrated the involvement of adsorbed water in the p-doping of graphene by atmospheric oxygen36,42,45,47,48. Irreversible or slow modulation of charge carrier density was demonstrated by changing the atmosphere36,42 or by illuminating graphene with UV or visible light36,37,49. In the latter cases, the invoked mechanisms relied on dopant removal and more specifically oxygen derivatives. The characteristic times of such phenomena were found to be in the range of minutes36,49 or even hours37. Other authors have probed the dynamics of charge transfer between the H2O/O2 redox couple and graphene by electrical measurements alone or combined with Raman spectroscopy. It was shown that the equilibrium is established after minutes48 or hours42.

Our study shows that the charge carrier density can be conveniently tuned by adjusting the incident laser power even without a gating electrode. This effect does not involve the chemical modification of graphene since no D band emerges upon extended irradiation. By contrast with previous reports36,37,49, this laser-induced doping is reversible with a characteristic time that was preliminarily evaluated to be less than 1 s (see Supplementary Fig. S15), i.e. orders of magnitude faster. This suggests that this phenomenon, although related to similar environmental effects, involves a different mechanism. O2 and H2O are playing a key role as illustrated by the weak dependency of ωG on Plaser found after adsorbates removal by Ar annealing at 150°C of graphene lying on O2 plasma treated SiO2/Si substrates34. The transfer of “hot” electrons from optically excited graphene to the H2O/O2 redox couple or superoxide anion proposed in reference 36 could only increase the hole doping in graphene upon visible light exposure, which is not supported by our observations. To get more insight on the underlying mechanism of the observed effect, we have measured the evolution of the G-band Stokes/AntiStokes integrated intensity ratio as a function of Plaser (see Supplementary Fig. S11). We were thus able to establish that, for graphene lying on a substrate, its local temperature increases by about 100°C/mW. This behavior was found to be similar for all the substrates tested (hydrophilic or less hydrophilic, 285 nm or 90 nm of SiO2). Considering the occurence of a possible thermoelectric effect, the thermoelectric voltage that can build up in our experiments is expected to be of the order of few mV/mW50, i.e. negligible when compared to the Fermi level shifts measured around 200 meV/mW. Furthermore, since the Seebeck coefficient is 0 for intrinsic graphene and since its sign changes with graphene doping type, the thermoelectric effect will tend to make the laser-exposed (hot) region evolve toward lower doping concentration. This is incompatible with our observations that the laser induces a doping type reversal for F1 (switching from p-type to n-type) and a n-type doping for F2, that was initially intrinsic. We thus tentatively attribute the observed effect to a local and reversible perturbation of the chemical equilibrium between graphene, the substrate and the atmosphere by laser-induced heating. The fast dynamics and the reversibility of the effect cannot be related to dopant removal but can be compatible with a temporary destabilization of species. Noteworthy, graphene lying on SiO2/Si exposed to H2O alone, with only traces of O2, has been shown to be n-doped42. Similarly, water significantly reduces hole doping of graphene deposited on mica51. The elucidation of the laser-assisted charge carrier density tuning mechanism deserves further investigations which are beyond the scope of this paper.

In conclusion, we have shown that a low power visible laser light can be used to reversibly tune the charge carrier density of graphene lying on a substrate, with subsecond characteristic time. This effect is highly sensitive to the substrate hydrophilicity and completely suppressed in suspended graphene. The continuous tuning of the doping in graphene from p-type to n-type has been achieved on O2 plasma treated SiO2/Si substrates. The observed phenomenon is attributed to a local and reversible perturbation of the chemical equilibrium established between graphene, the substrate and the atmosphere by laser-induced heating.

One technical implication of our study for the entire scientific community using Raman spectroscopy of graphene as a routine characterization technique is that it should be considered as potentially invasive as far as electronic properties are concerned. In particular, the laser induced-modification of graphene doping could account for recent discrepancies between Raman and electrical transport measurements45. It would be interesting to extend the present work to graphene on metals47 and on silicon carbide52 to assess how carefully Raman experiments on graphene must be performed.

On another hand, the ability to tune the charge carrier density with visible photons opens a wide set of opportunities to develop optically gated graphene electronic devices and a new approach to graphene optoelectronics. Finally, this effect should allow to study the interplay between graphene properties and the environment and to trigger laser-assisted functionalization of graphene leading to more advanced devices53,54.

Methods

Sample fabrication

Four samples (F1 through F4) were prepared. They consisted of a 500 μm thick highly p-doped monocrystalline Si (100) substrate, with a thermal oxide layer of 290 ± 5 nm (in F1 and F3, oxide grown in O2; in F2 and F4, oxide grown in H2O) and a square matrix of metallic marks every 200 μm to ease the identification of graphene flakes. Samples F3 and F4 were further processed by etching matrices of 5 × 1 μm trenches with depths of 160 nm (F3) and 480 nm (F4) in between the metallic marks with a CF4 ICP-RIE process.

Cleaning of F1 and F3 was performed by sonication in organic solvents (N-methyl-2-pyrrolidone, acetone and isopropanol; VLSI quality), immersion in piranha bath (H2SO4/H2O2 3:1) during 3 hours and Ar/O2 (3:1) plasma cleaning in a Fischione 1020 for 20 min. On the other hand, F2 and F4 were cleaned by sonication in organic solvents and O2 plasma cleaning in a PVA TePLA 300 at 800 W for 15 min. In addition, sample F2 was annealed during 1 hour at 400°C in a 300 sccm Ar flow at atmospheric pressure.

The contact angles of deionised water on three Si (100) substrates with a 290 ± 5 nm layer of thermal oxide (grown in H2O atmosphere) have been measured by the sessile drop method. Sample CA1 was only rinsed in water and dried in a N2 flux to eliminate silicon particles resulting from cleaving, sample CA2 was cleaned using the same procedure as for F1 and F3 (i.e. piranha plus light O2 plasma) and sample CA3 was cleaned as F2 and F4 (i.e. strong O2 plasma). The resulting advancing quasi-static contact angles were 60 ± 1° for CA1, unmeasurable or below 10° for CA2 and 18 ± 1° for CA3.

Finally, graphene was deposited on all samples by exfoliation from kish graphite with the scotch tape method55.

Micro-Raman spectroscopy

Raman spectra were recorded using an Acton spectrometer fitted with a Pylon CCD detector and a 600 grooves/mm grating (~ 2.5 cm−1 between each CCD pixel). The samples were excited with a 532 nm (2.33 eV) CW frequency doubled Nd:Yag laser through a x100 objective (N.A. 0.9). The FWHM of the focused laser spot is 400 ± 20 nm. At 1 mW laser power, the power density impinging on the sample is thus about 800 kW cm−2 (8 mW μm−2). Optimized focus conditions have been checked for each measurement. The samples are mounted on a three-axis piezoelectric stage to ensure the precise positioning and focusing of the laser spot. The maps were recorded with a 0.2 μm step in X (0.4 μm in Y) to probe the suspended graphene independently with a minimum of 2 to 3 points in the middle of the pool. The laser power was tuned with a variable neutral density filter controlled by a servomotor. The laser power was continuously measured by a calibrated photodiode put behind the beamsplitter. The whole experimental setup (spectrometer, piezoelectric stage, photodiodes, servomotor) were controlled by a dedicated and home-made Labview application. During laser power sweep or Raman mapping experiments, the acquisition times are varying from few seconds (2–5 s) at the highest power up to 30–60 s at the lowest power. The laser power sweep experiments typically last 20–60 min.

The results were reproduced on 8 hydrophilic samples (including F1 and F4, 3 others on 285 nm SiO2/Si and 3 on 90 nm SiO2/Si), 5 less hydrophilic substrates (including F2 and 4 on as-received SiO2/Si substrates) and 3 suspended graphene flakes (including F3 and the suspended part of F4). All the flakes have been measured on different locations and the measurements have been repeated several times. On most of the samples Raman maps at different laser powers have also been measured.