Drift suppression of solution-gated graphene field-effect transistors through electrolyte submersion

In solution-gated graphene FETs (SG-GFETs), cations in electrolyte solutions can intercalate between graphene and SiO2. Such permeation affects substrate-induced hole doping effects, resulting in drifts in the charge neutrality point (CNP) of SG-GFETs. In this study, we investigated the effect of submerging GFETs in electrolyte solutions on CNP values. The results revealed that the CNP decreased considerably from approximately 180 mV to nearly zero with the increase in the immersion period. The CNP drifts during electrical measurements were also suppressed by the prolonged submersion. These insights can be used for developing improved SG-GFETs.

Device fabrication was performed according to a previously described procedure. 20)The GFETs were fabricated by transferring polycrystalline CVD-graphene onto a p-doped Si substrate with 290 nm SiO 2 by etching the underlying copper foil.Source/drain electrode lines were formed with 10 nm Ti and 90 nm Au through electron-beam evaporation.The electrodes were coated with an insulating polymer, but the Au-graphene contact areas were exposed.The graphene films were patterned into an array of six rows and nine columns using a 300 × 300 μm 2 mask through oxygen plasma etching, resulting in 54 GFETs integrated into the chips.Figure 1(a) displays a typical optical image of a GFET.Each GFET had a channel width and length of 100 μm.An optical microscope (DSX510, Olympus) and a deep learning method were used to ensure that graphene films were not torn or contaminated. 21)igure 1(b) displays a schematic of the measurement setup.A custom-built portable measurement system was used to characterize the devices. 22)A buffer solution of 0.01 ×PBS was prepared by diluting Dulbecco's phosphate-buffered saline [D-PBS(-), Nacalai Tesque] with deionized water by a factor of 100 for electrical measurements.A drain-source bias (V DS ) of 10 mV was applied, and a gate-source bias (V GS ) was imposed through the solution by using a doublejunction Ag/AgCl reference electrode.The drain-source current (I DS ) was recorded while sweeping V GS , and V GS at the minimum I DS (I DS,min ), denoted as CNP, was calculated for each GFET using polynomial fitting.
Figure 1(c) displays the mean I DS -V GS curve for an array of SG-GFETs.The plot displays the hole-doped graphene.The primary source for the doping is the negatively charged dangling oxygen bonds in SiO 2 substrates. 19)][25] As displayed in Fig. 1(d), the mean I DS -V GS was not identical but drifted during continuous electrical measurements.The CNP value shifted by more than 75 mV toward the negative direction after 1 h of measurement.The drift was initially substantial but gradually decreased.The trend of the drifts was consistent with those of previous reports. 15,16)The drift may be attributed to the cations intercalated between graphene and SiO 2 .The cations gradually diffused between the graphene and SiO 2 interface and screened the negatively charged SiO 2 surface, counteracting the p-type doping effect from the substrate.
We investigated the effects of cation permeation by submerging the device in 0.01 ×PBS prior to electrical measurements.Two devices were prepared under the following conditions: (i) without submersion and (ii) with submersion for a month (t immerse = 720 h).Electrical measurements were performed on these devices in 0.01 ×PBS [Fig. 2 To investigate permeation kinetics, t immerse dependence was evaluated.Multiple devices were prepared at various t immerse , followed by subsequent electrical measurements.Figure 3(a) displays the initial CNP values corresponding to the CNP obtained from the first V GS sweep as a function of t immerse .This trend was visualized by applying Gaussian process regression, which is a nonlinear and nonparametric regression tool that infers a continuous function from a set of individual data points. 26)Despite variations among devices, the initial CNP values exhibited a decreasing trend with increasing t immerse .In addition, the CNP drift decreased with the increase in t immerse .We defined ΔCNP as the change during the measurements over 60 min.As displayed in Fig. 3(b), ΔCNP approached zero with the increase in t immerse .These results indicate the effective prevention of permeation during the measurements by lengthening t immerse , resulting in CNP drift suppression.
To obtain spatial information and permeation dynamics, a Raman microscope (RAMANtouch, Nanophoton) with an    excitation wavelength of 532 nm and a 20× objective lens was used for a single GFET submerged in 0.01 ×PBS.Timelapse Raman images (25 × 408 μm 2 ) were captured from the middle of the graphene channel since the submersion initiation.G-band and 2D-band peak positions (ω G , ω 2D ) were obtained for every pixel from Raman images.Figure 4(a) displays the average (ω G , ω 2D ) points for the Raman images, revealing downshifts in both peaks.Here, (ω G , ω 2D ) are sensitive to changes in carrier density and mechanical strain. 27)The peaks were upshifted along the unit vectors of e H = (1, 0.7) under an increase in n H and e T = (1, 2.2) under tensile strain ε.Setting (ω G 0 , ω 2D 0 ) = (1585, 2676) as the origin (O) of the ω G -ω 2D space, 28) which corresponds to charge-neutral and unstrained graphene, any given P(ω G ,ω 2D ) can be vector-decomposed as OP = αe H + βe T , where α and β are constants, respectively.Based on the obtained α, Fermi energy E F is derived through E F = (−18α − 83) × 10 3 . 28)In addition, E F is a function of n H as  ( ) where ħ is the reduced Planck's constant and v F = 1.1 × 10 6 m s −1 is the Fermi velocity of graphene on the SiO 2 substrate. 28,29)Therefore, n H can be derived from (ω G , ω 2D ) using these equations.Because (ω G , ω 2D ) were obtained from every pixel in Raman images, the corresponding n H images were generated.Figure 4(b) displays that n H is relatively homogeneous across the area.Time-lapse n H images in Figs.4(b)-4(f) reveal a gradual decrease in n H with the increase in t immerse .This trend was consistent with that of the CNP drift, as displayed in Fig. 1(d).Furthermore, Figs.4(b)-4(f) reveals a uniform decrease in n H across the study area.These results indicate that the primary pathway for cation permeation was through nanopores in the basal plane.Although the intercalation of water underneath graphene from the edges to the center was visualized in real time using Raman spectroscopy, 30) this process was not observed in Figs.4(b)-4(f).Because the graphene films used in this study were polycrystalline and wet-transferred onto SiO 2 substrates, some nanopores, atomic defects, cracks, and wrinkles may be present in the basal plane. 31)These defects likely served as additional entrances for ion diffusion, resulting in a uniform decrease in n H .
In conclusion, we investigated the kinetics of ion permeation and its effect on the electronic properties in GFET configurations.The intercalated cations screened the negative charge on the SiO 2 surface, counteracting the p-type doping effects originating from the substrate.Consequently, the diffusion process induced drifts in the CNP in SG-GFETs.By submerging GFETs in 0.01 ×PBS prior to electrical measurements, the initial CNP values and CNP drift decreased with the increase in t immerse .These results indicate that the cations sufficiently penetrated at the onset of the measurement, and further permeation was suppressed after prolonged submersion.Raman microscopic analysis revealed a uniform decrease in n H across the graphene film, indicating that the primary pathway for permeation was through the nanopores in the basal plane rather than from the edges.The findings of the study provide valuable insights for the development of reliable sensing platforms based on SG-GFETs.
(a)]. Figure 2(b) displays mean I DS -V GS obtained from the initial V GS sweep.The plots were considerably affected by submersion.To understand this phenomenon, the feature values of CNP, I DS,min , and the minimum transconductance g m,h were compared [Figs.2(c)-2(e)].As displayed in Fig. 2(c), the CNP values decreased considerably with prolonged submersion.This decrease indicates that the cations penetrated at the onset of the measurement, thus counteracting the initial p-type doping effect from the substrate.Despite shifts in CNP, I DS,min and g m,h remained nearly unchanged.These results indicate that submersion did not induce charge-scattering centers.Furthermore, Fig. 2(f) confirms the remarkable suppression of the CNP drift with submersion.The CNP value shifted by −77 mV during 1 h of measurements for t immerse = 0 h, whereas the CNP value shifted by less than 10 mV for t immerse = 720 h.The results indicate that in the pre-immersed GFETs, further permeation was prevented during measurements.

Fig. 1 .
Fig. 1.(a) Optical image of a typical graphene field-effect transistor (GFET).(b) Schematic of the measurement setup.(c) Mean I DS -V GS of an array of solution-gated-GFETs (SG-GFETs) at 0 min (red) and 60 min (blue).The shadows represent their standard deviations.(d) Time course of the mean I DS -V GS .The white dashed line represents the values of charge neutrality point (CNP).

Fig. 2 .
Fig. 2. (a) Schematic of the experimental procedure.One device was submerged in 0.01 ×PBS for 720 h prior to electrical measurements.(b) Mean I DS -V GS obtained from the first V GS sweep.(c)-(e) Violin plots of CNP (c), minimum I DS (I DS,min ) (d), and minimum transconductance (g m,h ) (e) of the arrays of SG-GFETs.(f) Time course of CNP.The colors represent the pre-immersion periods t immerse , red: 0 h, blue: 720 h.The shadows represent their standard deviations.(a)(b)

Fig. 3 .
Fig. 3. (a)(b) Immersion-time (t immerse ) dependence of initial CNP values (a) and ΔCNP (b).Here, ΔCNP is defined as the change during the measurements over 60 min.Each dot represents different chips.The dashed line represents the median prediction of Gaussian process regression, and the shadows represent 70% of Bayesian credible intervals.

Fig. 4 .
Fig. 4. (a) The average G-and 2D peak (ω G , ω 2D ) points for Raman images captured at the middle of graphene channels.Unit vectors of e T and e H are parallel to the dashed lines, respectively.The colors correspond to the time since submersion.(b)-(f) Time-lapse images of hole carrier density n H at 0 h (b), 1 h (c), 2 h (d), 3 h (e), and 4 h (f).The scale bar is 50 μm.