Carbon nano-structures show great promise in many applications, including chemical and biological sensors. While carbon nanotubes (CNTs) have been extensively studied in that context for quite some time 12, investigations of graphene as a sensor are only beginning to appear 345. Sensory function of carbon nano-structures is generally implemented in the configuration of a field effect transistor (FET), with a prominent role played by the gate potential that controls the current through the device. Biochemical applications require good understanding of the interaction of carbon nano-structures with aqueous solutions 5, often in the context of the electrochemical top gating 6. While significant progress has been achieved in understanding the interaction of CNT–FETs with the electrolytic environment 789, similar studies involving graphene have appeared only very recently 6, focusing on the screening effect of an ion solution on charge transport through graphene-based FETs 10, as well as on the measurement of the quantum capacitance of graphene as an ultimately thin electrode in an aqueous solution 11.

The top gating of a graphene-based FET with a solid or liquid electrolyte presents several advantages compared to the conventional back gating with a metallic electrode. Upon application of gate voltage, free ions in the electrolyte re-distribute themselves, forming an electrostatic double layer (EDL) at the interface between graphene and the electrolytic solution 12. Depending on the ion concentration, the EDL can be only a few nanometers thick, while still providing efficient shielding of graphene. As a consequence, the capacitance of the EDL in an electrolyte can be much higher than the capacitance of the back gate, which is typically separated from graphene with a layer of SiO₂ a few hundred nanometers thick 12. This property of the EDL enables a much better control of the surface potential on the graphene layer, while requiring a much lower operating voltage that needs to be applied to the reference electrode in the electrolyte than voltages currently used with back gates 12. The applied voltage then modifies the chemical potential of graphene, resulting in changes in its observable properties such as conductance. Since properties of the EDL depend on the ion concentration, monitoring the resulting changes in graphene’s conductance can provide a means for sensor application, e.g., in measuring the amount of salt in the solution.

On the other hand, referring to the electrical model of the electrolytic gating as a series connection of capacitors 13, the high gate capacitance in the electrolyte gives a much more prominent role to the quantum capacitance of graphene than does the back gate 11141516. In addition, doping levels of an electrolytically top-gated graphene have been reported recently 6 to be much higher than those obtained with the conventional back gate 17. At the same time, mobile ions in the solution seem to provide a much more effective screening of charged impurities underneath the graphene, thereby significantly increasing the charge carrier mobility in graphene in comparison with some other high-κ dielectric environments 18. All these facts indicate that electrolytic top gating provides a means to develop high-performance FETs.

While the above few experimental observations reveal quite fascinating aspects of the graphene–electrolyte interaction, theoretical modeling of this system seems to be lagging behind the experiment. It is therefore desirable and tempting to discuss doping of a single layer of graphene by a remote gate electrode immersed in a thick layer of electrolyte by using two simple models: one describing graphene’s π electron band structure in the linear energy dispersion approximation 12, and the other describing the distribution of ions in the electrolyte by a one-dimensional (1D) Poisson–Boltzmann (PB) model, which takes advantage of the planar symmetry of the problem 13. However, it should be emphasized that the experiments involving electrolytic top gating of both CNTs 78 and graphene 610 use rather high voltages, on the order of 1–2 V, which can cause significant crowding of counter-ions at the electrolyte–graphene interface. It is therefore necessary to go beyond the standard PB model by taking into account the steric effects, i.e., the effects of finite size of ions in the solution. To that aim, we shall use the modified Poisson–Boltzmann (mPB) model developed by Borukhov et al. 2223, which retains analytical tractability of the original 1D-PB model. In addition, applied voltages beyond 1 V also require taking into account non-linearity of graphene’s band energy dispersion, giving small but noticeable corrections to the linear approximation.

We shall consider here a simple 1:1 electrolyte representing an aqueous solution of NaF because both the Na⁺ and the F⁻ ions are chemically inert allowing us to neglect their specific adsorption on the graphene surface 1013. In particular, we shall analyze the density of doped charge carriers in graphene at room temperature (RT) as a function of both the applied voltage and the salt concentration to elucidate graphene’s sensor ability. In addition, we shall evaluate the contributions of both the graphene and the EDL in the total gate capacitance in terms of the applied voltage to reveal the significance of quantum capacitance, as well as to elucidate the behavior of the EDL under high voltages. We shall cover broad ranges of both the salt concentration, going from μM to a physiologically relevant value, and the applied voltage, going up to about 2 V.

After outlining our theoretical models for graphene and the EDL in the next section, we shall introduce several reduced quantities of relevance for these two vastly different systems and present our results in the following section. Finally, conclusion follows. Note that we shall use gaussian units (4πε₀ = 1) throughout the paper, unless otherwise explicitly stated.

Graphene is a semi-metal, or a zero-gap semiconductor because its conducting and valence π electron bands touch each other only at two isolated points in its two-dimensional (2D) Brillouin zone 12. The conical shape of these bands in the vicinity of these points gives rise to an approximately linear density of states, , where g_d = 4 is the spin and the band valley degeneracy factor, and v_F ≈ c/300 is the Fermi speed of graphene, with c being the speed of light in vacuum 12. In the intrinsic, or undoped graphene, the Fermi energy level sits precisely at the neutrality point, ɛ_F = 0, also called the Dirac point. Therefore, the electrical conductivity of graphene is easily controlled, e.g., by applying a gate voltage V_A that will cause doping of graphene’s π bands with electrons or holes (depending on the sign of V_A), which can attain the number density per unit area, n, with a typical range of n ∼ 10¹¹–10¹³ cm⁻²12. In a doped graphene, Fermi level moves to , where sgn(n) = ± 1 for electron (hole) doping. At a finite temperature T, one can express the charge carrier density in a doped graphene in terms of its chemical potential μ as 19

where with k_B being the Boltzmann constant. We shall use in our calculations a full, non-linear expression for the π electron band density, ρ(ɛ), given in Eq. 14 of Ref. 12. However, for the sake of transparency, the theoretical model for graphene will be outlined below within the linear density approximation, ρ(ɛ) ≈ ρ_L (ɛ). We note that this approximation is accurate enough for low to moderate doping levels, such that, e.g., , and it only incurs a relative error of up to a few percent when .

At this point, it is convenient to define the potential V_Q = −μ/e, where e > 0 is the proton charge, which is associated with the quantum-mechanical effects of graphene’s band structure 20, and relate it to the induced charge density per unit area on doped graphene, σ = − en, via the Eq. 1,

where dilog is the standard dilogarithm function 21. One can finally use the definition of differential capacitance per unit area, C_Q = dσ/dV_Q, to obtain from Eq. 2 the quantum capacitance of a single layer of graphene as 14

where we have defined the characteristic length scale for graphene,

with the value of λ_Q ≈ 18 nm at RT. Note from Eq. 3 that graphene’s quantum capacitance grows practically linearly with V_Q when this potential exceeds the thermal potential, V_th = 1/(eβ), having the value of ≈26 mV at RT.

We further assume that an upper surface of graphene is exposed to a thick layer of a symmetric z: z electrolyte containing the bulk number density per unit volume, N, of dissolved salt ions. Taking advantage of planar symmetry, we place an x axis perpendicular to graphene and pointing into the electrolyte. The theory developed by Borukhov et al. 2223 to model finite ion size uses the mPB equation for the electrostatic potential V (x) in the electrolyte at a distance x from graphene, given by

where z(=1) is the valency of ions, ε is relative dielectric constant of water (≈80, assumed to be constant throughout the electrolyte), and γ = 2a³N is the packing parameter of the solvated ions, which are assumed to have same effective size, equal to a2223. We note that the standard PB model is recovered from Eq. 5 in the limit γ → 0 13. By assuming the boundary condition V(x) = 0 (and hence dV/dx = 0) at x → ∞, deep into the electrolyte bulk, Eq. 5 can be integrated once giving a relation between the electric field and the potential at a distance x from graphene. Assuming that graphene is placed at x = 0, one can use the boundary condition at the distance d of closest approach for ions in the electrolyte to graphene,

to establish a connection between the induced charge density on graphene, σ, and the potential drop, V_D = V(d), across the EDL as

The total potential,V_A, applied between the reference electrode in the electrolyte and graphene can be written as

where V_pzc = (W_gr − W_ref)/e is the potential of zero charge 13 that stems from difference between the work functions of graphene and the reference electrode, W_gr and W_ref, respectively, and V_cl = 4πhσ/ε′ is the potential drop across a charge-free region between the compact layer of the electrolyte ions condensed on the graphene surface, having the thickness h on the order of the distance of closest approach d2324, and with ε′ < ε taking into account a reduction of the dielectric constant of water close to a charged wall 25. In our calculations, we shall neglect these two contributions to the applied potential in Eq. 8 because V_pzc merely shifts the zero of that potential, while a proper modeling of V_cl involves large uncertainty 2324. However, usually the effects of V_cl can be considered either small 23 or incorporated in the mPB model via saturation of the ion density at the electrolyte–graphene interface for high potential values 22. Consequently, V_D and V_Q represent the two main contributions in Eq. 8, with V_D being the surface potential of graphene that shifts its Dirac point, and V_Q being responsible for controlling the doping of graphene by changing its chemical potential. Finally, we note that all results of our calculations will be symmetrical relative to the change in sign of the applied potential because of our assumption that the effective sizes of the positive and negative ions are equal 2223, but this constraint can be lifted by a relatively simple amendment to the mPB model 24.

Using the relation V_A = V_Q + V_D, we obtain the total differential capacitance of the electrolytically top-gated graphene as

where C_Q(V_Q) is given in Eq. 3, and C_D(V_D) = dσ/dV_D is the differential capacitance per unit area of the EDL, which can be obtained from Eq. 7as 2324,

with being the Debye length of the EDL 13. Note that, in the limit of a very low potential V_D, and hence for low density of ions at the graphene–EDL interface, one can set γ → 0 in Eq. 10 to recover an expression for the EDL capacitance in the standard PB model 13,

We further note that, while Eq. 11 implies an unbounded growth of the EDL capacitance with V_D in the PB model, Eq. 10 suggests a non-monotonous behavior that will eventually give rise to a saturation of the total gate capacitance at high applied voltages.

Given the vast ranges of various parameters of interest in our model, it is of interest to define reduced quantities. With the thermal potential V_th = 1/(eβ), all potentials can be written as. While typical regimes of graphene doping require only, we shall extend this range in our calculations up to aboutto represent doping levels in recent experiments in electrolytic environment 6. Next, referring to Eq. 2, we define the characteristic number density of doped charge carriers in graphene by , which has the value of n₀ ≈ 10¹¹ cm⁻² at RT. Therefore, defining the reduced density by , and hence , we note that may reach up to around 10³612. It is worthwhile mentioning that graphene’s characteristic parameters λ_Q and n₀ are related via ελ_Bλ_Qn₀ = 1, where λ_B = βe²/ε is the Bjerrum length of the aqueous environment, taking the value of λ_B ≈ 0.7 nm at RT 13. Furthermore, it follows from Eq. 3 that the natural unit of capacitance for this system is C₀ = en₀/V_th = λ_Q⁻¹, taking the value of C₀ ≈ 0.6 μF/cm² at RT. Turning now to Eq. 7, one can define the characteristic number density of ions per unit volume in the solution by

which takes the value of N₀ ≈ 1.08 × 10⁻⁶ nm⁻³ ≈ 1.8 μM at RT. Defining the reduced concentration of ions in the bulk of the electrolyte by , it would be of interest to explore a broad range of its values, e.g., . Finally, in order to estimate the packing parameter, we write and take a = λ_B to obtain ν = 2λ_B³N₀ ≈ 7 × 10⁻⁷. With these definitions, Eqs. 2 and 7 now read, respectively,

We now use Eqs. 13and14in conjunction with the relationto eliminate the potential componentsand, and to evaluate the reduced density of doped charge carriers in graphene,, as a function of the reduced applied voltageand the reduced salt concentration. The results are shown in Figs. 1 and 2, covering the following ranges:(corresponding to(corresponding to), and(corresponding to). In Fig. 1, one notices a strong dependence ofon the applied potential forgreater than about 30, which gives approximately equal rates of change for each salt concentration at the highest values of the applied potential. On the other hand, at the lower applied potential values, there exists a much stronger dependence on the salt concentration, which is revealed in Fig. 2, showingversusfor several applied voltages. Indeed, one notices a very strong sensitivity of the doped charge carrier density in graphene to the salt concentration for applied voltagesin the range of salt concentrations. Even though this sensitivity seems to be the strongest at the lowest applied voltages, one should bear in mind that the electrical conductivity in graphene becomes rather uncertain around its minimum value, which extends up to doping densities about n ≈ 10¹¹ cm⁻²2627. Therefore, it seems that would be an optimal range of applied voltages for sensor applications of the electrolytically top-gated graphene in probing salt concentrations in the sub-millimole range.

Next, moving to the capacitance of electrolytically top-gated graphene, C, we note that the reduced capacitances,and, are obtained from Eqs. 3and10as

showing thatandare comparable in magnitude for vanishing potentials when the salt concentration is. Moreover, referring to Eq. 9as an electrical model where graphene and the EDL act as a series connection of capacitors, it follows that graphene’s quantum capacitance C_Q will be promoted as the dominant contribution to the total gate capacitance as the salt concentration increases.

We now use the equality of the right-hand-sides in Eqs. 13and14along with the relationto eliminateandand to evaluate the reduced quantum capacitance of graphene from Eq. 15, as well as the reduced capacitance of the EDL from Eq. 16as functions of the reduced applied voltage. Results are shown in Fig. 3 along with the total reduced capacitance of the system based on Eq. 9, for two reduced salt concentrations,= 1 and 10⁵(corresponding to N ≈ 1.8 μM and 0.18 M, respectively). We show our results for the reduced applied voltages up toin order to elucidate the effect of saturation in the total capacitance that occurs at(corresponding to V_A ≈ 2.21 V) forand at(corresponding to V_A ≈ 1.95 V) for. As can be seen from dotted curves in Fig. 3, showing a non-monotonous dependence of the EDL capacitance on the applied voltage, the saturation effect in the total capacitance of the electrolytically top-gated graphene is a consequence of the steric effect of the electrolyte ions that are crowded at the graphene surface at high applied voltages 23. Even though the voltages where the saturation occurs are relatively high, they may still be accessible in experiments on graphene. Furthermore, we see in Fig. 3 that at intermediate applied voltages, the rate of change of the total capacitance follows closely that of the quantum capacitance, with the value ≈23 μF/(V cm²) that is commensurate with recent measurement 11. At the lowest applied voltages, one notices in Fig. 3 a “rounding” of the total capacitance as a function of voltage for low salt concentrations, which comes from the EDL capacitance. Such rounding is observed in the recent experiment 11.

As a consequence of the vast differences between the capacitances shown in Fig. 3, one expects that there exists a broad variation in the way how the total applied voltage V_A splits between the potential drop V_D across the EDL and the voltage V_Q pertaining to the change in graphene’s chemical potential. We therefore display in Fig. 4 the variation of the fraction r = V_D/V_A as a function of the reduced salt concentration in the electrolytefor several values of the applied voltage V_A. One can see that at low salt concentrations, the potentials V_Q and V_D are roughly comparable in magnitude, although the ratio r increases in favor of the potential drop across the electrolyte as the applied voltage increases. However, this trend is reversed at high salt concentrations and, more importantly, Fig. 4 shows that the most of the applied voltage is used to increase graphene’s chemical potential for a full range of applied voltages when salt concentration N exceeds, say, mM. Besides its importance for applications, the fact that the potential drop across the electrolyte remains very small at high applied voltages also alleviates concern that a high electric field in the electrolyte may cause the onset of voltage-dependent electrochemical reactions on graphene.

We have analyzed the doping of single-layer graphene due to application of the gate potential through an aqueous solution of salt using a modified Poisson–Boltzman model for electrolyte and found great sensitivity of the induced charge density in graphene to the broad ranges of both salt concentration and applied voltage. We have further analyzed differential capacitance of the electrolytically top-gated graphene and found that its quantum capacitance is promoted as the dominant component owing to a reduction in the Debye length of the electric double layer when the salt concentration increases. In this case, very little potential drop appears across the electrolyte, and graphene takes most of the voltage drop to shift its chemical potential. These findings have several important consequences.

First, since graphene’s conductivity is dependent upon its chemical potential, its sensitivity to the salt concentration implies good prospects for applications in biochemical sensors, especially for in vivo electrochemical measurements in biological systems owing to graphene’s natural bio-compatibility. Next, since most of the applied voltage can be used to increase the chemical potential of graphene, as opposed to a potential drop across the electrolyte, one can envision ways to use a very thin top gate (in the form of a liquid or solid electrolyte) that requires relatively low gate voltage to change the chemical potential (and hence conductivity) of graphene in future small scale field effect devices with tunable conductivity. Among other aspects of the increased role of graphene’s quantum capacitance is reduction of the electrical field across the electrolyte. This can help reduce the rates of voltage-induced electrochemical reactions on graphene’s surface, as well as improve the mobility of charge carriers in graphene by reducing their scattering rates on various impurities. Moreover, since quantum capacitance is basically the capacitance associated with change in carrier densities, it can be seen as analogous to the junction capacitance, and the smaller quantum capacitance could in turn lead to faster switching time for graphene-based devices.

Many of these advantages of top gating through an electrolyte are related to a high bulk dielectric constant of the electrolyte, especially in aqueous solutions. So, even though the oxide thickness can be reduced down to around 2 nm in the present generation conventional MOS structures, the much higher dielectric constant of water in comparison with SiO₂ should provide for a higher gate capacitance, translating into much better field effect performance, as discussed above. However, dielectric constant of an electrolyte can be significantly reduced close to a charged surface 2425, and this issue has yet to be discussed in the context of electrolytic top gating of carbon nano-structures.