+
+
+

+ Introduction +

+
+

+ Graphene field-effect transistors (GFETs) have emerged as exceptionally sensitive platforms for chemical and biological sensing + + + 1 + + , + + 2 + + , + + 3 + + , + + 4 + + , + + 5 + + + . The atomically thin graphene channel offers a fully exposed surface and exceptional carrier mobility, enabling strong field-effect responses to adsorbates + + + 6 + + + . In typical single-gate GFET biosensors, baseline stability and limit-of-detection (LOD) are degraded by several well-known issues + + + 7 + + , + + 8 + + , + + 9 + + + . One major problem is signal drift and hysteresis: the graphene transfer characteristics (e.g., Dirac point) tend to shift irreversibly or slowly over time (in some cases, even exacerbated by applied electrostatic gates) due to charge trapping and adsorbates + + + 10 + + , + + 11 + + , + + 12 + + + . Likewise, substantial hysteresis is seen in graphene-on-oxide devices: as the gate voltage is swept, charge transfer and capacitive effects (from adsorbed molecules or ions) cause positive or negative voltage shifts of the conductance minimum + + + 13 + + , + + 14 + + , + + 15 + + , + + 16 + + + . Unexplained signal drift and interfacial phenomena at the nanoscale remain key obstacles to the stable operation of graphene-based FET biosensors + + + 7 + + + . +

+

+ These performance-limiting instabilities are further compounded by the reliance on single-gate architectures (majority of reported GFET research)—typically employing an electrolyte top gate for liquid-phase analyte detection or a back gate for solid-state or gaseous phase sensing + + + 17 + + + . This static-gate mode of operation often yields relatively small signal amplitudes, limiting sensitivity and signal-to-noise ratio. To enhance the sensing response, many studies employ dynamic gate sweeps to probe features such as the position of the Dirac point, transconductance, and carrier mobility—parameters that are modulated by analyte interactions + + + 17 + + , + + 18 + + + . However, gate voltage sweeps can induce charge trapping/detrapping, hysteresis, and temporal signal drift. To overcome the limitations associated with small signal amplitudes and drift-prone gate sweeping, there is a pressing need for signal amplification strategies that do not compromise measurement stability or noise performance. +

+

+ Dual-gate transistor architectures present a compelling alternative by enabling capacitive signal amplification through asymmetric gate coupling + + + 19 + + , + + 20 + + , + + 21 + + , + + 22 + + + . In dual-gate transistors, the channel is modulated simultaneously by two independent gate electrodes—for example, a solid-state dielectric back gate and an electrolyte top gate. When these gates are designed with significantly different gate capacitances—typically with the electrolyte gate providing orders of magnitude higher capacitance than the back gate—a small perturbation at the top gate (e.g., due to analyte binding) can result in a disproportionately larger signal response when the back gate compensates via capacitive coupling + + + 22 + + + . This asymmetric dual-gating framework enables signal amplification without the need for rapid gate sweeping. This concept has been previously explored in a limited number of non-graphene sensing platforms. For example, a dual-gated silicon nanowire FET, incorporating both a liquid gate and a back gate, demonstrated pH sensitivity beyond the Nernst limit + + + 21 + + + . More recently, another study utilized a dual-gated GFET with an ionic liquid top gate and an actively controlled back gate under high-vacuum conditions to enhance pH sensitivity via feedback mechanisms + + + 22 + + + . In this configuration, the graphene channel was isolated from the analyte medium and served solely as an electronic amplifier within the feedback loop. +

+

+ Despite these promising reports, dual-gated GFET sensors remain underexplored, particularly those that fully leverage graphene’s intrinsic surface sensitivity and chemical tunability—whether through direct functionalization with molecular probes or through electrolyte-gated interactions with charged analytes in solution. A key challenge for dual-gate GFETs is gate leakage in liquid environments. While the use of local high-κ back gates has been shown to substantially reduce effective oxide thickness (EOT) and suppress leakage, these benefits have largely been demonstrated under dry conditions + + + 23 + + , + + 24 + + , + + 25 + + + . However, in a liquid environment, the back-gate electrode area must be carefully minimized to avoid faradaic currents due to defects in the oxide. Microscopic processing defects that remain insulating in air can become active leakage pathways for ions in solution. Furthermore, the dielectric material itself is susceptible to electrochemical degradation and water-assisted etching under bias, leading to faradaic currents and device failure + + + 26 + + , + + 27 + + , + + 28 + + + . We believe this challenge is a key reason why dual-gate GFETs have not been widely adopted in research or industry. +

+

+ In this work, we overcome longstanding limitations of GFET sensors by introducing a field-effect amplified, active dual-gated architecture engineered for enhanced sensitivity and operational stability with low noise. Our design integrates a local high-κ hafnium dioxide (HfO + + 2 + + ) back gate, patterned beneath the graphene channel, paired with a liquid-phase electrolyte top gate. This asymmetric dual-gate configuration acts as a capacitive voltage divider: perturbations at the graphene-electrolyte interface—such as ion adsorption or analyte binding—modulate the surface potential, which is then amplified at the back gate via capacitive coupling. To achieve active feedback, we employ readily available operational amplifiers to dynamically control the back gate voltage in real time. This approach enables cost-effective, on-demand modulation of the gate potential, facilitating sensor calibration and drift compensation without requiring complex electronics. +

+

+ To systematically evaluate sensing performance, we define and compare seven distinct operational modes, encompassing a range of static and dynamic biasing conditions, both in single- and dual-gate configurations. Each mode was assessed across various analyte categories, including pH changes, small molecules (electroactive biogenic neurotransmitter amines), volatile organic compounds (isopropyl alcohol), environmental contaminants (PFAS), and large biomolecules (cytokine IL6). This wide applicability underscores the versatility of the sensing platform. Among all configurations, the Differential Mode Fixed (DMF; described in later sections) mode demonstrates the highest signal fidelity and sensor performance. In terms of performance metrics, DMF achieves up to 20× higher sensitivity, > 15× lower signal drift, and enhanced signal-to-noise ratios (up to 7×)—well beyond what is attainable using conventional single-gate GFETs. +

+

+ In summary, the proposed dual-gated, field-effect amplified GFET architecture establishes a new standard for graphene-based sensors by simultaneously delivering ultrasensitive detection, ultralow drift, and on-demand signal amplification. By integrating advanced materials and device engineering with tailored electronic circuit design and rigorous operational mode analysis, this platform achieves unprecedented sensor performance. Moreover, the approach is broadly transferable to other 2D materials and nanoscale FET systems, providing a versatile blueprint for next-generation biosensors and environmental monitors requiring both exceptional sensitivity and long-term stability. We envision this platform enabling real-time, high-precision sensing in complex environments, with the potential to achieve single-molecule detection limits in portable, low-power devices. +

+
+
+
+
+
+

+ Results +

+
+

+ Device physics and amplification model +

+

+ The device architecture utilized in this study is a generalized dual-gated GFET, incorporating two independent gating mechanisms: (1) a top gate formed by an electrolyte interface and (2) a local solid-state back gate. A schematic representation of the structure is shown in Fig. + + 1a + + (Inset shows a representative fabricated device; as a comparison, similar schematic for global gated devices is shown in Fig. + + 1b + + ). This dual-gate configuration enables versatile modulation of the graphene channel potential and charge carrier density via capacitive coupling from both gates + + + 29 + + + . +

+
+
+
+ + Fig. 1: Dual-gated graphene field-effect transistors (GFET) with one gate floated. + +
+
+
+ + + Fig. 1: Dual-gated graphene field-effect transistors (GFET) with one gate floated. + +
+ + + Full size image + + + +
+
+
+

+ + a + + Schematic of dual-gated GFET featuring an independent top gate and a locally patterned back gate with a solid-state HfO + + 2 + + dielectric, enabling capacitive modulation from both interfaces. Inset: optical micrograph of a fabricated device (scale bar: 30 μm). + + b + + Comparison schematic of a conventional global back-gated GFET with a thick SiO + + 2 + + dielectric. + + c + + Top Gate Fixed (TGF) mode: the top gate is biased while the back gate is floated. + + d + + Top Gate Sweep (TGS) mode: the top gate is swept with the back gate floated to obtain full transfer characteristics. + + e + + Representative transfer curves in TGS mode for aqueous PBS electrolyte media. + + f + + Back Gate Fixed (BGF) mode: the back gate is held at a fixed bias while the top gate is floated. + + g + + Back Gate Sweep (BGS) mode: the back gate is swept with the top gate floated to evaluate solid-state gating performance. + + h + + Representative transfer curves in BGS mode, demonstrating improved gating efficiency and lower voltage operation using the HfO + + 2 + + dielectric compared to conventional thick-SiO + + 2 + + -based devices. + + e + + , + + h + + Insets) Statistical distribution of device performance parameters ( + + N + + = 18 for TGS parameters, + + N + + = 63 for BGS parameters). The central line represents the median, the box edges represent the interquartile range (IQR) (25th to 75th percentiles), and the whiskers extend to 1.5 × IQR. Individual data points beyond the whiskers indicate outliers. +

+
+
+
+
+

+ To model the operation of the dual-gate GFET, we employed a standard electrostatic model involving two capacitances in series: the geometric capacitance of the gate dielectric and the quantum capacitance of graphene + + + 30 + + + . The quantum capacitance arises due to the linear dispersion relation of Dirac fermions in graphene, which leads to a low density of states near the Dirac point. The total gate capacitance for each gate, + + \({C}_{{eff}}^{(i)}\) + + , is thus given by Eq. + + 1 + + : +

+
+
+ + $$\frac{1}{{C}_{{eff}}^{(i)}}=\frac{1}{{C}_{{geo}}^{(i)}}+\frac{1}{{C}_{q}},$$ + +
+
+ (1) +
+
+

+ where + + \(i=\{\mathrm{TG},\mathrm{BG}\}\) + + denotes top and back gate, + + \({C}_{q}\) + + is the quantum capacitance of graphene, and + + \({C}_{{geo}}^{(i)}\) + + is the geometric capacitance for the top gate and back gate. For the top gate, the geometric capacitance is the double layer capacitance, + + \({C}_{{dl}}\) + + , formed by the electrolyte and is heavily influenced by the ionic strength, media type, and dissolved species. For the back gate, the geometric capacitance is the oxide capacitance, + + \({C}_{{ox}}\) + + . Using this electrostatic framework, the drain-source current ( + + \({I}_{{DS}}\) + + ) in the linear regime is modeled using the standard drift-diffusion approximation for GFETs + + + 31 + + , + + 32 + + + : +

+
+
+ + $${I}_{{DS}}=\mu \frac{W}{L}{V}_{{DS}}\left({C}_{{eff}}^{{TG}}({V}_{{TG}}-{V}_{{TG},{Dirac}})+{C}_{{eff}}^{{BG}}({V}_{{BG}}-{V}_{{BG},{Dirac}})+{{en}}^{{{\prime} }}\right),$$ + +
+
+ (2) +
+
+

+ where + + \(\mu\) + + is the carrier mobility, + + \(W\) + + and + + \(L\) + + are width and length of the graphene channel, + + \(e\) + + is the electron charge, + + \({n}^{{\prime} }={n}_{0}^{{\prime} }\) + + is intrinsic doping including trapped charges and charge puddles, + + \({V}_{{DS}}\) + + is the drain source voltage, + + \({V}_{{TG}}\) + + and + + \({V}_{{BG}}\) + + are the top and back gates respectively, + + \({V}_{{TG},{Dirac}}\) + + and + + \({V}_{{BG},{Dirac}}\) + + are the Dirac peak locations for top and back gate respectively. The interaction between the graphene channel and a target analyte during a measurement event modifies the carrier density. This can be modeled as + + \(n^{\prime}\) + + = + + \({n}_{0}^{{\prime} }+\triangle {n}^{{\prime} }+a\) + + . Here, + + \(a\) + + represents the target signal: the charge density induced solely by the analyte-graphene interaction. In contrast, + + \(\Delta {n}^{{\prime} }\) + + represents the error signal: the unwanted change in carrier density due to drift, hysteresis, or trap filling that occurs during the measurement interval. Hence, in a GFET at a fixed gate potential, the change in current is given by: +

+
+
+ + $$\Delta {I}_{{DS}}=\mu \frac{W}{L}{{eV}}_{DS}(\Delta {n}^{{\prime} }+a),$$ + +
+
+ (3) +
+
+

+ where + + \(\Delta {n}^{{\prime} }\) + + is the added drift/hysteresis due to the performed measurement, + + \(\Delta {I}_{{DS}}\) + + is the signal change, and + + \(a\) + + is the contribution from charge introduced due to the analyte-graphene/gate interaction. To evaluate the sensor performance, we define the effective Signal-to-Noise Ratio (SNR). We consider the “Signal” to be the analyte contribution ( + + \(a\) + + ) and the “Noise” to be the sum of the electrical noise floor and the drift-induced error ( + + \(\Delta {n}^{{\prime} }\) + + ), as both limit the resolution of the sensor. If we assume the noise in + + \({I}_{{DS}}\) + + to be + + \({N}_{I,{DS}}\) + + , the SNR—when + + \(\Delta {I}_{{DS}}\) + + is used as the signal—using Eq. + + 3 + + can be calculated as: +

+
+
+ + $${{SNR}}_{I,{DS}}=\frac{a}{\Delta {n}^{{\prime} }+\frac{{N}_{I,{DS}}}{\mu \frac{W}{L}e{V}_{{DS}}}}$$ + +
+
+ (4) +
+
+

+ If the gate potential is fixed for entirety of the measurement, we can assume + + \(a\) + + is much larger compared to + + \(\Delta {n}^{{\prime} }\) + + , as is the case where measurements commonly bias the gate and continuously read + + \({I}_{{DS}}\) + + . Thus, + + \({{SNR}}_{I,{DS}}\) + + of change in these cases is mostly proportional to + + \(a/{N}_{I,{DS}}\) + + . In contrast, if + + \(a\) + + is small/comparable to + + \(\Delta {n}^{{\prime} }\) + + , + + \({{SNR}}_{I,{DS}}\) + + is small, as is the case with measurements where the gate is swept rapidly. In these measurements, other metrics such as the location of the Dirac peak is used instead, ideally, this is equal to: +

+
+
+ + $$\Delta {V}_{{DP},(i)}=\frac{e(\triangle {n}^{{\prime} }+a)}{{C}_{{eff}}^{(i)}},$$ + +
+
+ (5) +
+
+

+ where + + \(\triangle {V}_{{DP},(i)}\) + + is the change in the Dirac peak location due to change in + + \(\triangle {n}^{{\prime} }\) + + . Similar to Eq. + + 4 + + , we can calculate the SNR for Eq. + + 5 + + —assuming + + \({N}_{V,{DP},(i)}\) + + as the noise in + + \({V}_{{DP},(i)}\) + + —as: +

+
+
+ + $${\mathrm{SNR}}_{V,{DP},(i)}=\frac{a}{\Delta {n}^{{\prime} }+{\frac{{C}_{{tot}}^{(i)}}{e}N}_{V,{DP},(i)}}$$ + +
+
+ (6) +
+
+

+ Generally, in order to find the Dirac peak, multiple + + \({I}_{{DS}}\) + + measurements are performed (assume + + \(N\) + + ). This means that we can estimate (from Eq. + + 2 + + ) that: +

+
+
+ + $${{C}_{{eff}}^{(i)}N}_{V,{DP},(i)}\times f(N)=\frac{{N}_{I,{DS}}}{\mu \frac{W}{L}e{V}_{{DS}}},$$ + +
+
+ (7) +
+
+

+ where + + \(f\left(N\right)\propto \sqrt{N}\) + + is a function of + + \(N\) + + guaranteed to be greater than unity since multiple measurements reduce the uncertainty + + + 33 + + + . Hence, from Eqs. + + 4 + + , + + 6 + + and + + 7 + + , + + \({{SNR}}_{V,{DP},(i)}\) + + > + + \({{SNR}}_{I,{DS}}\) + + provided + + \(\Delta {n}^{{\prime} }\) + + does not increase substantially enough to overshadow the decrease in measurement uncertainty due to the sweeps as compared to the fixed gate method, this is generally the case for slow sweeps. +

+

+ For a simultaneous dual gating with feedback that compensates for changes in + + \({I}_{{DS}}\) + + by sweeping the back gate, we require that ideally, + + \(\Delta {I}_{{DS}}=0\) + + . Hence, using Eq. + + 2 + + , we have: +

+
+
+ + $${C}_{{eff}}^{{TG}}\Delta {V}_{{TG}}+\Delta {V}_{{BG}}{C}_{{eff}}^{{BG}}+e\Delta {n}^{{\prime} }+{ea}=0$$ + +
+
+ (8) +
+
+

+ Which follows that the measured signal + + \(\Delta {V}_{{BG}}\) + + is given by: +

+
+
+ + $$\Delta {V}_{{BG}}=-\frac{\left(e\Delta {n}^{{\prime} }+{ea}+{C}_{{eff}}^{{TG}}\Delta {V}_{{TG}}\right)}{{C}_{{eff}}^{{BG}}}=-\frac{{C}_{{eff}}^{{TG}}}{{C}_{{eff}}^{{BG}}}\Delta {V}_{e,{TG}},$$ + +
+
+ (9) +
+
+

+ where + + \(\Delta {V}_{e,{TG}}=e\frac{\Delta {n}^{{\prime} }+a}{{C}_{{eff}}^{{TG}}}+\Delta {V}_{{TG}}\) + + is the equivalent top gate shift due to either molecule interaction with the gate or the graphene channel. Hence, following this analysis, for simultaneous dual gating—where we design a feedback system to fix + + \({I}_{{DS}}\) + + by actively changing + + \({V}_{{BG}}\) + + and fixing + + \({V}_{{TG}}\) + + —we find that Eq. + + 5 + + still holds, with the signal defined as: +

+
+
+ + $$\Delta {V}_{{BG}}=\frac{e(\Delta {n}^{{\prime} }+a)}{{C}_{{eff}}^{{BG}}}$$ + +
+
+ (10) +
+
+

+ Similarly, from Eq. + + 7 + + , we preserve the gains in SNR. It is important to emphasize that the feedback mechanism modulates + + \({V}_{{BG}}\) + + to compensate for the total change in channel carrier density + + \(e(\Delta {n}^{{\prime} }+a)\) + + , without distinguishing between the analyte signal ( + + \(a\) + + ) and the error signal ( + + \(\Delta {n}^{{\prime} }\) + + ). By maintaining the top gate at a fixed potential (unlike sweep-based modes), the large hysteresis-induced component of + + \(\Delta {n}^{{\prime} }\) + + is eliminated. Consequently, the feedback loop predominantly amplifies the analyte-induced surface potential shift, resulting in a high SNR. +

+

+ In conclusion, we find that compared to fixed gate systems, the signal with our dual-gating approach is amplified by + + \({C}_{{eff}}^{{TG}}/{C}_{{eff}}^{{BG}}\) + + (from Eq. + + 9 + + ), while SNR is also improved since lower speed gate sweep rates lead to less + + \(\triangle {n}^{{\prime} }\) + + . This necessitates that we estimate both + + \({C}_{{eff}}^{{TG}}\) + + and + + \({C}_{{eff}}^{{BG}}\) + + to calculate the expected gain in sensitivity. From literature review (and experimental evidence in later sections), we find that + + \({C}_{{geo}}^{{TG}}={C}_{{dl}} > {C}_{q}\) + + and + + \({C}_{{geo}}^{{BG}}={C}_{{ox}}\ll {C}_{q}\) + + . Accordingly, Eq. + + 9 + + can be approximated to: +

+
+
+ + $$\frac{\Delta {V}_{{BG}}}{\Delta {V}_{e,{TG}}}=-\frac{{C}_{{dl}}{C}_{q}}{{C}_{{ox}}({C}_{{dl}}+{C}_{q})}$$ + +
+
+ (11) +
+
+

+ Using approximate values for + + \({C}_{{dl}}\approx 36\,\frac{\mu F}{c{m}^{2}}\) + + , + + \({C}_{q}\approx 7\,\frac{\mu F}{c{m}^{2}}\) + + , and + + \({C}_{{ox}}\approx 0.6\,\frac{\mu F}{c{m}^{2}}\) + + in our system, we estimated a gain of approximately + + \(\frac{\Delta {V}_{{BG}}}{\Delta {V}_{e,{TG}}}\approx 10\) + + (see Section S + + 3 + + ). However, as these capacitances change—especially + + \({C}_{{dl}}\) + + or + + \({C}_{q}\) + + due to changes in the media, molecule interaction, or gate potentials—we expect to see varied amplification factors. Since + + \({C}_{q}\) + + is slightly smaller than + + \({C}_{{dl}}\) + + , it plays a significant role in determining the amplification factor. +

+

+ Electrical characterization and operational modes +

+

+ The device fabrication and experimental validation were conducted using various electrolyte media and analytes, as described in “Methods” section. The local back gate consisting of a thin HfO + + 2 + + dielectric is deposited over patterned electrodes on a silicon dioxide substrate schematically shown in Fig. S + + 1 + + of Section S + + 1 + + . Briefly, photolithography and atomic layer deposition (ALD) were performed on a Si/SiO + + 2 + + wafer to pattern a tri-metal local gate, deposit a 35 nm HfO + + 2 + + dielectric layer, and etch openings for the source and drain contacts. Graphene was then transferred onto the local gate stack via a wet transfer method, followed by etching and passivation to complete device fabrication illustrated in Fig. S + + 2 + + . The finalized device stack was integrated with the measurement readout circuitry shown in Section S + + 2 + + and tested using protocols for various media and analytes. Overall, across tested sensors, we achieved a 94% yield for successful source-drain contacts and graphene presence under dry conditions. We achieved a yield of > 90% for functional back-gate modulation under dry conditions, confirming the integrity of the gate stack in ambient air. However, only 65% of devices exhibited functional back-gate response in liquid. Electrical gate leakage failures accounted for 35% of devices, while resistive failures were observed in 6%. +

+

+ The dual-gated GFET architecture facilitates the exploration of multiple operational modes by independently configuring the top and back gate voltages. These modes include single-gate operation (with one gate floated or fixed), dual-sweep operation (both gates swept), and differential feedback operation. Table + + 1 + + summarizes the seven operational modes evaluated in this study. +

+
+
+
+ + Table 1 Operational modes of the dual-gated GFET and their configurations + +
+
+ + + Full size table + + + +
+
+
+

+ In Top Gate Fixed (TGF) mode, the top gate is held at a constant voltage while the back gate is left floating, as illustrated in Fig. + + 1c + + . In this configuration, the device functions effectively as a single top-gated GFET, where the electrolyte forms an electric double layer (EDL) at the graphene-electrolyte interface. This EDL serves as a high-capacitance gate dielectric, enabling modulation of the graphene channel at ultra-low gate voltages. The elevated EDL capacitance enhances charge carrier accumulation in the channel, facilitating real-time monitoring of the drain current during analyte exposure. In the Top Gate Sweep (TGS) mode, the top gate voltage is swept while the back gate remains floating, as illustrated in Fig. + + 1d + + . This approach enables acquisition of full transfer characteristics, allowing extraction of transconductance and precise tracking of Dirac point shifts caused by surface interactions. These shifts result from electrostatic gating by target analytes or specific molecular interactions at the graphene surface or gate electrode. For example, adsorption of charged biomolecules such as DNA or proteins onto the graphene channel induces local doping, causing concentration-dependent shifts of the Dirac point. Furthermore, reactions or binding events at the gate-electrolyte interface—such as protonation of functional groups—can indirectly modulate the gating field experienced by the graphene channel + + + 9 + + + . +

+

+ The nature and strength of the electrolyte play a significant role in device behavior. Electrolytes with smaller ions or higher ionic strength can form thinner and more compact EDLs, resulting in higher effective gate capacitance + + + 30 + + + . Aqueous electrolytes—such as phosphate-buffered saline and ionic liquids—enable strong gating but can differ markedly in their dielectric behavior, viscosity, and ion mobility. These differences directly affect the position and sharpness of the Dirac point. Specifically, weaker electrolytes or those containing bulky solvated ions—such as organic solvents like acetonitrile—result in broader or shifted transfer characteristics due to the formation of a more diffuse EDL. Table + + S1 + + summarizes the extracted conductance and double layer capacitance for different media, including deionized water (DIW), phosphate-buffered saline (PBS), potassium chloride (KCl), acetonitrile (ACET) with and without 10 mM KCl, and an ionic liquid (IL). +

+

+ As expected, ionic media such as PBS and 10 mM KCl exhibited high capacitance (36–46 µF/cm + + 2 + + ) and low solution resistance (~0.2 kΩ), consistent with efficient ionic screening. In contrast, DIW and pure acetonitrile showed low capacitance in the sub-pF/cm² range with much higher solution resistance, reflecting their low ionic conductivity. The ionic liquid demonstrated intermediate capacitance and resistance, highlighting its unique electrochemical properties. These results validate the strong dependence of interfacial capacitance on electrolyte composition, which is critical for sensor performance in different media. +

+

+ Example TGS data of devices measured in PBS is shown in Fig. + + 1e + + , displaying the characteristic Dirac peak of graphene, which shifts as multiple gate scans are performed. A distribution of the initial resistance at zero gate ( + + R + + + + SD,0 + + + ), ratio of resistance at the Dirac peak to + + R + + + + SD,0 + + + ( + + R + + + + Mod + + + ), and the location of the Dirac peak ( + + V + + + + DP + + + ) for 18 devices also individually plotted in Fig. S + + 4 + + . Having established the baseline response of the liquid gate, we next isolated the solid-state back gate to characterize the device’s intrinsic dielectric performance independent of the electrolyte. +

+

+ In Back Gate Fixed (BGF) mode, the back gate voltage is held constant while the top gate is floated, allowing partial investigation of solid-state gating contributions, as shown in Fig. + + 1f + + . In contrast, Back Gate Sweep (BGS) mode involves actively sweeping the back gate while the top gate remains floated, as shown in Fig. + + 1g + + . These configurations help isolate and characterize the intrinsic behavior of the device structure, particularly the effects of solid-state dielectric modulation in the absence of an electrolyte interface. +

+

+ A key challenge with back gate operation—especially in ambient conditions—is the inherent high doping of graphene due to adsorbed moisture, oxygen, and charged species from the environment + + + 15 + + , + + 16 + + + . This ambient doping often masks or shifts the Dirac point, making precise calibration difficult. Moreover, solid-state gating in thin dielectric systems is prone to leakage currents and potential dielectric breakdown, especially under high electric fields. These effects can compromise long-term device reliability. +

+

+ However, in our devices fabricated using ALD of HfO + + 2 + + as the back gate dielectric, we observe clear and reproducible Dirac peaks at significantly lower back gate voltages, as shown in Fig. + + 1h + + . (Distribution of device parameters is plotted as insets in Fig. + + 1h + + . Individual device data for 63 devices are presented in Fig. S + + 5 + + ). This contrasts sharply with devices using, for example, 285 nm SiO + + 2 + + oxide, which require back gate voltages exceeding 80 V + + + 15 + + , + + 16 + + + . This improvement is attributed to the higher dielectric constant ( + + k + + ~ 25) of HfO + + 2 + + , resulting in approximately a 50× increase in capacitance, which enables efficient capacitive coupling at reduced gate biases. Compared to conventional global back-gated SiO + + 2 + + devices (Fig. + + 1b + + ), our architecture delivers superior performance. The lower geometric capacitance of thick SiO + + 2 + + (~50× smaller than that of our devices) limits its suitability for low-voltage or battery-powered applications. Additionally, careful processing and cleaning minimize trapped charges and charge puddles, further stabilizing device operation. While Fig. + + 1h + + confirms the high dielectric constant and integrity of the HfO + + 2 + + back gate in ambient air, the introduction of an electrolyte significantly alters the electrostatic environment. As detailed in Section S + + 6 + + , the addition of PBS electrolyte results in a positive shift of the Dirac point (p-doping) and increased hysteresis compared to the air-stable baseline + + + 34 + + + . +

+

+ From a system integration perspective, global back-gated configurations—such as those employing a common SiO + + 2 + + back plane—are less suitable for array-based sensing. Due to the fabrication variability inherent in two-dimensional materials, a single leaky device in a globally gated array can affect all sensors sharing the same gate, compromising measurement integrity. In contrast, our locally patterned back gate architecture provides device-level control, enabling selective gating and isolation of individual sensors. This localized gating approach enhances robustness, improves fault tolerance, and offers clear advantages for multichannel biosensing and scalable integration. Beyond independent single-gate characterization, the unique architecture of these devices allows for simultaneous dual-gate sweeping to resolve complex electrostatic coupling. +

+

+ In the Addition Mode (AM), both the top electrolyte gate and the back solid-state gate are sequentially swept during a single measurement, as illustrated in Fig. + + 2a + + . This dual-gate operation enables complex, high-resolution mapping of the GFET transfer characteristics, providing detailed insights into charge neutrality and electrostatic coupling across the graphene channel. Notably, to the best of our knowledge, this is the first demonstration of a dual-gated GFET system in which both a solid oxide back gate and a liquid/aqueous electrolyte top gate yield clearly resolved Dirac peaks under simultaneous sweep conditions. +

+
+
+
+ + Fig. 2: Dual-gated operational modes in GFETs with both gates biased. + +
+
+
+ + + Fig. 2: Dual-gated operational modes in GFETs with both gates biased. + +
+ + + Full size image + + + +
+
+
+

+ + a + + Addition Mode (AM): sequential sweeping of both top liquid-gate and back solid-gate reveals distinct Dirac peaks for each gate. + + b + + 2D gate voltage maps obtained in PBS electrolyte showing the addition effect of dual gates. The resolution and shape of these maps depend on the scan dynamics. In our dual-gate mapping, we utilize a rastering approach where one gate serves as the “fast scan” (continuously swept) and the other as the “slow scan” (stepped incrementally). + + c + + Differential Mode Fixed (DMF): top gate is held constant while an operational amplifier adjusts the back gate in feedback to maintain a fixed channel current, enabling real-time signal amplification. + + d + + Differential Mode Sweep (DMS): top gate is swept while the back gate is dynamically adjusted, allowing investigation of transient and hysteresis effects. + + e + + Amplified back gate response DMS mode ( + + V + + + + BG + + + ) to the top gate sweep ( + + V + + + + TG + + + ) in PBS, with an observed slope ~10×, consistent with predicted feedback gain. ( + + e + + Inset) Statistical distribution of observed device slopes depicted by a box plot ( + + N + + = 5). +

+
+
+
+
+

+ Previous studies have typically relied on asymmetric gate structures in which one gate—often the electrolyte—dominates electrostatic control, rendering the secondary gate largely ineffective in modulating the channel. In contrast, our devices achieve sufficiently balanced electrostatic coupling between the top and back gates. While the top gate capacitance remains approximately an order of magnitude larger than that of the back gate (amplification factor + + ∼ + + 10), the high-κ HfO + + 2 + + dielectric ensures that the back gate is strong enough to independently resolve the Dirac point within a low voltage window, a capability often lacking in standard SiO2-based devices. +

+

+ We must note that repeated back gate sweeps in AM reveal a key limitation: the combined effects of hysteresis, charge trapping, and ionic drift lead to progressive broadening and eventual loss of the Dirac peak. This degradation is particularly pronounced under electrolyte gating conditions. As shown in Fig. S + + 7a + + , the Dirac peak is clearly visible during the initial sweeps but diminishes with continued operation. Although AM operation offers valuable insights into gate coupling and dielectric properties, its practical use in sensing is constrained by hysteresis-induced drift and the loss of peak resolution over time. To mitigate this issue and investigate AM further, we systematically performed dual-gate sweeps using aqueous electrolyte media (PBS, shown in Fig. + + 2b + + ) by sweeping the top gate with fixed back gate potentials. We observed a shift in the Dirac peak with a slope of around + + \(m=13.6\) + + . While the geometric capacitance ratio ( + + \({C}_{{dl}}/{C}_{{ox}}\) + + ) calculated in Section S + + 3 + + suggests a potential gain of ~36, the effective amplification is fundamentally limited by the quantum capacitance of graphene ( + + \({C}_{q}\) + + ). As modeled in Eq. + + 11 + + , the series contribution of + + \({C}_{q}\) + + reduces the theoretical effective gain to ~10. Our measured value of 13.6 lies in close agreement with this + + \({C}_{q}\) + + -limited prediction. The variation between the predicted ( ~ 10) and measured (13.6) values is attributed to the variable nature of + + \({C}_{q}\) + + , which increases significantly away from the Dirac point ( > 20μF/cm + + 2 + + ), thereby increasing the coupling efficiency during the sweep. We measured the quantum capacitance of graphene using AM as described in Section S + + 7 + + and found the value to be ~2 + + \(\frac{\mu F}{c{m}^{2}}\) + + near the Dirac point increasing to > 50 + + \(\frac{\mathrm{\mu F}}{{\mathrm{cm}}^{2}}\) + + farther from it + + + 35 + + + . This demonstrates that we expect to see a varied amplification factor >10—as predicted by theory—in cases where the quantum capacitance of graphene or the double layer capacitance is larger. +

+

+ The differential feedback mode represents a novel sensing architecture that leverages real-time electrostatic feedback to achieve intrinsic signal amplification. In Differential Mode Fixed (DMF) mode, the top gate is held at a fixed bias while the back gate is dynamically modulated via a closed-loop feedback mechanism implemented using off-the-shelf electronic components such as operational amplifiers or digital circuits (Fig. + + 2c + + ). The op-amp senses the current flowing through the graphene channel and adjusts the back gate voltage ( + + \({V}_{{BG}}\) + + ) to restore the channel to its predefined operating point. This enables the system to translate small changes in top gate potential—caused by molecular binding or environmental shifts—into amplified back gate voltage responses ( + + \({\triangle V}_{{BG}}\) + + ). We compared different feedback modes as described in Section S + + 8 + + and found that, while op amp-based feedback can effectively implement dynamic modulation, the high capacitance of the electrolyte can, in some cases, lead to instability in the op amp response. To mitigate this, in the present study we employed a “digital op amp” approach, implemented using a digital-to-analog converter (DAC) and analog-to-digital converters (ADCs), which eliminates the risk of feedback oscillations associated with the analog op amp configuration. We define a unifying parameter “Signal” for our tests as either + + \({S=\triangle V}_{{BG}}\) + + for differential mode measurements, + + \(S={\triangle V}_{\{{BG},{TG}\},{Dirac}}\) + + for sweep methods, and + + \(S={{\triangle I}_{{ds}}R}_{{gain}}\) + + (gain of amplifier as discussed in Section S + + 8 + + ) in case of static modes. +

+

+ We expect an amplification factor of around >10 with PBS samples since the ratio + + \(\frac{{C}_{{eff}}^{{TG}}}{{C}_{{eff}}^{{BG}}} > 10\) + + in our system. This amplification factor may change based on changes in media and added molecules which may change the capacitive coupling. In this study, this factor is desirable since the molecules of interest in this study (e.g., redox-active neurotransmitters) typically generate + + \({|\triangle V}_{{TG}}|\le 0.5V\) + + leading to + + \({|\triangle V}_{{BG}}|\le \sim 15V\) + + nearing the limits for safe wearable/portable battery sensors. Translating the signal into larger electrical shifts via feedback control not only improves signal-to-noise ratios but also simplifies the downstream analog-to-digital conversion and processing. +

+

+ It is important to recognize that the amplification factor is tunable and dependent on the dielectric and electrochemical configuration of both gates. Should different classes of analytes or sensing environments be employed—e.g., those requiring organic solvents or non-polar media—the selection of the electrolyte (including top gate electrode) and back gate oxide must be carefully optimized. This tunability provides flexibility for tailoring the sensor design to the biochemical context, making this approach highly versatile. +

+

+ In addition to signal amplification, we also investigated the effect of DMF on the intrinsic electrical noise characteristics of the GFET. Specifically, we measured the low-frequency 1/f noise—commonly associated with charge carrier fluctuations and interfacial traps—which is a major limiting factor in the detection of small signals. Our measurements revealed that DMF operation significantly suppresses the current noise compared to open-loop configurations such as TGF (As shown in Fig. + + S8 + + ). This noise suppression arises from the active stabilization of the channel current via the feedback loop, which dynamically compensates for fluctuations by modulating + + \({V}_{{BG}}\) + + . As a result, the residual noise is effectively transferred from the current domain to the voltage domain, where it appears as minor variations in the amplified + + \({V}_{{BG}}\) + + signal. +

+

+ In Differential Mode Sweep (DMS) mode, the top gate is swept while the differential feedback loop actively adjusts the back gate, enabling dynamic characterization of the graphene channel response under a range of electrochemical conditions as shown in Fig. + + 2d + + . Figure + + 2e + + plots the measured + + \({V}_{{BG}}\) + + against applied + + \({V}_{{TG}}\) + + in PBS and we find the slope (amplification factor) to be nearly ~10× confirming + + \(\frac{{C}_{{eff}}^{{TG}}}{{C}_{{eff}}^{{BG}}}\approx 10\) + + in our system. To validate the dependence of this amplification factor on the electrolyte properties, we repeated the dual-gate mapping in Deionized Water (DIW). As shown in Fig. + + S7b + + , the reduced double-layer capacitance of DIW resulted in a significantly shallower slope of approximately 1.2. This confirms that the + + \({V}_{{TG}}\) + + vs. + + \({V}_{{BG}}\) + + relationship is tunable based on the media capacitance. Together, DMF and DMS constitute a powerful dual-gate framework with built-in electrostatic amplification, uniquely suited for low-signal environments. +

+

+ In summary, this section establishes the operational role and expected electrical behavior of each gating configuration. The following sections analyze how these operating modes translate into sensing performance across chemical, biological, and environmental targets. +

+

+ Sensor performance and benchmarking +

+

+ To benchmark the performance of various operational modes discussed earlier and highlight the superiority of DMF, we evaluated a range of biological, environmental, and chemical targets. This assessment quantifies improvements in key sensing metrics, including sensitivity, limit of detection (LOD), SNR, hysteresis, and drift. To validate the universality of the DMF amplification mechanism, we selected a panel of analytes representing distinct electrostatic gating regimes. These analytes serve as model systems for specific sensing challenges: (1) pH validates the platform’s response to fundamental ionic gating and Nernstian shifts; (2) Neurotransmitters test the resolution of small surface-doping signals from redox-active species; (3) IL-6 represents macromolecular sensing which validates the platform’s utility for immunoassays involving large capture probes; and (4) VOCs and PFOA evaluate performance in gas-phase and liquid-phase adsorption and environmental contaminant monitoring. This selection confirms that the capacitive amplification described by theory is intrinsic to the device architecture and agnostic to the specific biochemical origin of the surface potential shift. To empirically verify this theoretical universality, we first examined the platform’s response to the simplest gating mechanism: protonation changes in the electrolyte. +

+

+ The detection of pH changes is a canonical application of liquid-gated GFETs due to the sensitivity of the electric double layer and the protonation/deprotonation of surface functional groups. In the single top gate (TGF/TGS) mode, pH sensitivity is manifested as a shift in the Dirac point voltage with increasing H + + + + + or OH + + − + + ion concentration. As shown in Fig. + + 3a + + , the pH sensitivity of TGF was 214 mV/pH and that of TGS exhibited significant drift (50.52%/hr; discussed later). Similarly, we calculated the sensitivity across different measurement modes described previously tabulated in Table + + 2 + + . We observed that dual-gate operation, particularly in differential feedback modes (DMF and DMS), enhances both the sensitivity and stability of pH sensing as shown in Fig. + + 3a + + . Dual-gate operation in DMF mode achieved a remarkable pH sensitivity of 1314 mV/pH, representing over a 6× amplification compared to single-gate modes—with an SNR of 21.64 and reduced drift of 0.66%/hr. This sensitivity represents the amplified back-gate readout voltage ( + + \({\triangle V}_{{BG}}\) + + ), not the intrinsic surface potential shift at the electrolyte interface. While the interfacial potential shift remains governed by the Nernst limit (~59 mV/pH), the DMF feedback loop multiplies this signal by the capacitance ratio, effectively acting as an in-situ voltage amplifier. These improvements are highlighted in Fig. + + 3a + + . The results, plotted in Fig. + + 3a + + and tabulated in Table + + 2 + + , show that DMF operation reduced drift and hysteresis by more than an order of magnitude compared to TGS, confirming the theoretical predictions. Following the confirmation of fundamental ionic gating, we investigated the system’s ability to resolve signals from more complex, redox-active small molecules. +

+
+
+
+ + Fig. 3: Sensing performance of dual-gated GFETs for pH, redox-active small molecules, IL-6 (as a representative protein), and PFAS. + +
+
+
+ + + Fig. 3: Sensing performance of dual-gated GFETs for pH, redox-active small molecules, IL-6 (as a representative protein), and PFAS. + +
+ + + Full size image + + + +
+
+
+

+ The three solid lines in each plot represent data from three independent devices and the error bars indicate the Standard Deviation of 6 repeated measurements on each device. + + a + + pH sensing: comparison of Top Gate Fixed mode (TGF) with Differential Mode Fixed (DMF), showing enhanced stability and sensitivity in DMF. + + b + + Small molecule detection: sensing of redox-active small molecules demonstrates amplified response in DMF compared to TGF. + + c + + Protein detection (IL-6): IL-6 detection in TGF and DMF modes, with DMF providing higher sensitivity and stronger signal due to electrostatic feedback. + + d + + PFOA detection: comparison of TGF and DMF modes for PFOA sensing, highlighting improved response under DMF operation. ( + + a + + – + + d + + Inset) Statistical distribution of observed device parameters depicted by a box plot ( + + N + + = 3). +

+
+
+
+
+
+
+
+ + Table 2 Summary of sensing metrics for various operational modes + +
+
+ + + Full size table + + + +
+
+
+

+ Many small molecules such as dopamine (DA), serotonin (SER), norepinephrine (NEP), epinephrine (EP), uric acid (UA), and ascorbic acid (AA) are redox active, undergoing reduction–oxidation reactions at specific applied potentials + + + 36 + + , + + 37 + + , + + 38 + + + . These reactions modulate the interfacial charge density and, in turn, the electrostatic potential sensed by graphene. Consistent with our pH results, we observed a 21× amplification in sensor sensitivity for detecting these molecules, as shown in Fig. + + 3b + + . Owing to differences in the formal potentials of various redox-active neurotransmitters, the top-gate response is distinct for each analyte and is further amplified and measured using the DMF mode. Having successfully resolved small redox-active species, we proceeded to challenge the sensor with larger, sterically complex protein biomarkers to validate its clinical utility. +

+

+ The detection of large biomolecules such as interleukin-6 (IL-6) is critically important in clinical diagnostics, particularly for inflammatory disorders, autoimmune conditions, and cancer progression monitoring. IL-6 is a cytokine with a molecular weight of ~21 kDa and a hydrodynamic diameter of approximately 4–6 nm. Unlike small molecules or ions, large biomolecules pose unique challenges for GFETs due to limited charge transfer, slower diffusion kinetics, and steric hindrance near the graphene sensing surface. +

+

+ To enable IL-6 detection, we functionalized the graphene surface with IL-6 antibodies, as described in “Methods” section, and subsequently tested the devices with increasing concentrations of IL-6. The results for TGF and DMF modes are shown in Fig. + + 3c + + . We observed that sensitivity is enhanced by 8× in DMF compared to TGF. As shown in Section S + + 9 + + , we achieved an LOD of ~10 ng/mL for TGF and ~1 ng/mL for DMF. This improvement highlights the capability of the feedback mode to resolve minute surface potential shifts caused by antibody-antigen binding that are otherwise buried in the noise floor of passive operating modes. Upon IL-6 binding, a net change in the local charge distribution and interfacial dipole potential shifts the Dirac point voltage, which in DMF mode was further amplified into a back-gate voltage response. These results, while demonstrated here for IL-6 detection, highlight a generalizable sensing paradigm applicable to a wide range of analytes and capture chemistries. The dual-gating strategy can, in principle, be extended to other clinically relevant targets such as cancer biomarkers (e.g., prostate-specific antigen, carcinoembryonic antigen), whole cells (e.g., circulating tumor cells, bacteria), and nucleic acids (DNA, RNA, microRNA). Likewise, the graphene transducer can be functionalized with diverse capture elements—including antibodies, aptamers, nucleic acids, and even intact cells—enabling both molecular and cellular detection in a label-free manner. Similar GFET-based devices have been reported for CRISPR-mediated nucleic acid detection with attomolar sensitivity via Cas-mediated cleavage products + + + 39 + + , + + 40 + + + , aptamer-functionalized GFETs for thrombin and cytokine sensing + + + 41 + + , + + 42 + + + , and antibody-modified GFETs for viral antigens and small-molecule toxins + + + 43 + + + . The generality arises from graphene’s high interfacial sensitivity, the tunable surface chemistry for immobilizing different biorecognition layers, and the capability of the DMF mode to amplify surface-potential changes into measurable electronic signals. Demonstrating the platform’s adaptability beyond biological targets, we next evaluated its performance in detecting critical environmental pollutants. +

+

+ Perfluoroalkyl and polyfluoroalkyl substances (PFAS) represent a critical class of persistent organic pollutants widely used in industrial and consumer products due to their hydrophobic and lipophobic properties. Their environmental persistence and bioaccumulation potential pose serious health risks, necessitating sensitive and field-deployable detection technologies. In this work, we demonstrated the applicability of dual-gated GFETs for detecting representative PFAS compounds— specifically perfluorooctanoic acid (PFOA). The results are shown in Fig. + + 3d + + and tabulated in Table + + 2 + + . TGF showed a sensitivity of 229 mV/dec with a LOD of 88 ppb and an SNR of 3.34, while DMF amplified the sensitivity to 3318 mV/dec, lowered the LOD to 3.45 ppb, and improved the SNR to 22.49. These results confirm the suitability of dual-gated GFETs for sensitive, real-time detection of PFAS contaminants in water. To further expand the scope of environmental monitoring beyond liquid contaminants, we investigated the sensor’s response to gas-phase volatile organic compounds. +

+

+ VOCs such as ethanol, acetone, and isopropyl alcohol are crucial indicators in environmental monitoring, breath analysis, and industrial safety. VOC molecules adsorb onto the graphene surface via physisorption or π-π interactions, modulating the carrier density and inducing a measurable shift in the Dirac point. The extent and direction of the shift depend on the electron-donating or withdrawing nature of the VOC; for example, 2-propanol typically acts as an electron acceptor, shifting the Dirac point toward more negative gate voltages. +

+

+ We performed VOC sensing (gas environment) using BGF, BGS, and DMF modes, as described in “Methods” section and Section S + + 10 + + . VOC detection was performed using BGF and DMF modes (Fig. + + 4a + + ), and the performance metrics are tabulated in Table + + 2 + + . BGF exhibited minimal drift (0.17%/hr; discussed later), but switching to BGS resulted in significant drift (38.91%/hr). Importantly, DMF maintained low drift (2.65%/hr) while still enabling active sensing. This demonstrates the advantage of DMF for stable, low-noise VOC detection. +

+
+
+
+ + Fig. 4: VOC sensing and drift analysis. + +
+
+
+ + + Fig. 4: VOC sensing and drift analysis. + +
+ + + Full size image + + + +
+
+
+

+ + a + + Detection of isopropyl alcohol using dual-gated GFETs operated in Differential Mode Fixed (DMF) and Back-Gate Fixed mode (BGF), showing enhanced signal response in DMF. + + b + + Signal drift comparison in BGF, DMF, and back-gate sweep (BGS). + + c + + Signal drift comparison in top and back-gate sweeps with current change and Dirac shift as the signals. + + d + + Signal drift comparison in top-gate modes: Top-Gate Sweep (TGS), Top-Gate Fixed (TGF), Differential Mode Sweep (DMS), and DMF, highlighting the superior drift suppression achieved with DMF. Bold lines in ( + + a + + , + + b + + , + + d + + ) represent the average response of the device population ( + + N + + = 4, 4, and 5, respectively), while lighter lines indicate individual device traces. Insets are box plots that represent the device parameter statistical distribution. +

+
+
+
+
+

+ Stability analysis and drift comparison +

+

+ Long-term signal stability is a critical parameter in the design and operation of graphene-based chemical and biological sensors. Signal drift—which manifests as a gradual shift in the Dirac point or sensor output over time in the absence of analyte changes—can significantly impair sensitivity, reproducibility, and the reliability of quantitative measurements. Drift is often attributed to charge trapping/detrapping at the dielectric interface, ionic migration in the electrolyte, or hysteresis effects related to surface interactions. To evaluate and compare drift performance across different gating configurations, we systematically measured the time-dependent response of GFET sensors under both air-phase and liquid-phase conditions. The results are summarized in Fig. + + 4b–d + + . +

+

+ Figure + + 4b + + presents the signal drift observed during VOC sensing using three different back-gate configurations: BGF, BGS, and DMF. In the BGS mode, where the back-gate voltage is continuously swept during measurement, we observed significant signal drift—up to 38.91% per hour—which severely compromises signal stability and renders real-time sensing unreliable. In contrast, maintaining the back gate at a constant voltage in BGF mode drastically reduced drift to just 0.17% per hour, indicating that the sweeping process itself induces substantial disturbance, likely due to dynamic charge trapping in the gate dielectric or substrate, however, in this mode SNR is lower. The DMF configuration, which incorporates active feedback from the source-drain current to dynamically adjust the back gate, showed a moderate drift of 2.65% per hour. While this is higher than the static BGF case, it represents a >15× reduction in drift compared to BGS, and crucially, it preserves the amplification gain described earlier. +

+

+ We also compared the drift if instead of + + \({\triangle I}_{{DS}}\) + + as the signal, we use + + \({\triangle V}_{{Dirac}}\) + + for the sweep modes, considering that the shift in Dirac peak location as the signal is very commonly sensor characterization + + + 43 + + + . The results for both top and back gate sweep modes are summarized in Fig. + + 4c + + , showing that the drift in + + \({\triangle I}_{{DS}}\) + + and + + \({\triangle V}_{{Dirac}}\) + + is comparable. For both + + \({\triangle I}_{{DS}}\) + + and + + \({\triangle V}_{{Dirac}}\) + + as the signal, the absolute drift is ~50% and ~40% in top gate and back gate sweep modes, respectively. One discernible difference observed is that while + + \({\triangle V}_{{Dirac}}\) + + drift is linear, the + + \({\triangle I}_{{DS}}\) + + drift is more non-linear. Nevertheless, this drift complicates sensor calibration and, furthermore, is dependent on the molecule/media under test, as well as any changes in the sweep window utilized + + + 44 + + + . +

+

+ Signal drift in electrolyte-gated modes—relevant for biological and chemical sensing in liquid environments—is shown in Fig. + + 4d + + for four configurations: TGS, TGF, DMF, and DMS. The TGS mode exhibited the highest drift at 50.52% per hour, consistent with the known instability of continuously swept electrolyte gates, where ionic redistribution and interfacial charge trapping dominate. TGF mode, where the electrolyte gate is held at a fixed potential, achieved the lowest drift, as expected, by minimizing ionic motion and electrostatic perturbation. Notably, the DMF mode reduced drift to just 0.66% per hour, representing a ~76× improvement over TGS, while still enabling dynamic control of the back gate and enhanced signal amplification. The DMS mode, where both gates are swept, produced a drift of ~7.03% per hour, which, while higher than DMF, still reflects a 7× reduction in drift compared to TGS. +

+

+ The stark differences in drift performance across various operational modes highlight the critical role of gate modulation strategy in dual-gated GFET systems. In both air and liquid environments, gate voltage sweeps exacerbate charge trapping and hysteresis, particularly at the dielectric and graphene—electrolyte interfaces, ultimately degrading long-term reliability. By contrast, the active back-gate control in DMF mode provides real-time compensation of electrostatic fluctuations, dramatically stabilizing the sensor response. This enhanced drift suppression is key for enabling continuous, long-term monitoring in real-world sensing applications. +

+
+
+
+
+
+

+ Discussion +

+
+

+ In this work, we introduce a dual-gated graphene field-effect transistor (GFET) architecture that integrates a high-κ local back gate with an electrolyte top gate, enabling real-time, feedback-stabilized signal amplification for chemical and biological sensing. This asymmetric dual-gate design addresses longstanding limitations of single-gate GFETs—such as signal drift, hysteresis, and limited sensitivity—by eliminating the need for dynamic gate sweeps and reducing charge trapping and dielectric relaxation. Through systematic evaluation of multiple biasing modes across a broad set of analytes (pH, neurotransmitters, volatile compounds, environmental toxins, and proteins), we demonstrated that the Differential Mode Fixed configuration achieves superior performance, delivering >20× higher sensitivity, >7× noise reduction, and <15× drift compared to conventional methods. Maintaining constant transconductance via back-gate feedback enables over tenfold signal amplification while preserving a low-noise baseline, supporting robust, label-free detection in aqueous and physiologically relevant environments. Finally, the use of scalable materials and straightforward feedback electronics makes this platform readily adaptable to other 2D materials and miniaturized sensing technologies. While the Differential Mode Fixed mode offers significant advantages in signal fidelity, we acknowledge certain operational limitations inherent to this active feedback architecture: the stability of the feedback loop is sensitive to the capacitive load of the electrolyte, the high amplification factor requires sufficient voltage headroom at the back gate; and since the signal gain is derived from the capacitance ratio, the amplification factor is dependent on the ionic strength of the media, requiring calibration when transitioning between electrolytes with vastly different double-layer capacitances. Overall, this work defines a new operating paradigm for graphene-based sensors and provides a blueprint for designing high-performance, drift-resistant FET biosensors capable of real-time, multichannel analysis in complex environments. +

+
+
+
+
+
+

+ Methods +

+
+

+ Materials +

+

+ All chemicals and reagents were used as received without further purification. Phosphate-buffered saline (1× PBS, pH 7.4) was obtained from Thermo Fisher Scientific (Dulbecco’s formulation without magnesium and calcium). Acetonitrile, BMIM (1-Butyl-3-methylimidazolium hexafluorophosphate) ionic liquid, serotonin (5-HT), epinephrine, norepinephrine, dopamine, ascorbic acid, uric acid, and perfluorooctanoic acid (PFOA), were purchased from Sigma-Aldrich. Interleukin-6 (IL-6) antibodies (monoclonal), bovine serum albumin (BSA), 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide (EDC), and N-hydroxysuccinimide (NHS) were sourced from Thermo Fisher Scientific and Sigma-Aldrich. Monolayer graphene grown on copper foil was procured from Graphenea Inc. Silicon wafers (500 microns, + + P + + < 100 > , 0.001—0.005 ohm-cm) with thermally grown silicon dioxide (~285 nm) were purchased from Nova wafer Inc. +

+

+ Fabrication of substrates for local back gating +

+

+ Substrates for dual gating (with local back gate) were fabricated using standard photolithography and atomic layer deposition (ALD) techniques. A silicon wafer with a 285 nm thermally grown silicon dioxide layer was sequentially cleaned in acetone and isopropanol (IPA) and dried under nitrogen. A bilayer resist stack (LOR5A and SPR3012) was spin-coated and patterned using a Heidelberg MLA150 maskless aligner, and the exposed regions were developed in CD-26 developer to define the local gate areas. A tri-metal gate stack comprising 5 nm titanium (Ti), 20 nm gold (Au), and 10 nm platinum (Pt) was deposited via electron-beam evaporation. Lift-off was performed in acetone and PRS3000 remover, followed by oxygen plasma cleaning (25 W, 30 min; Harrick plasma cleaner) to remove residual organics and enhance surface hydrophilicity. Hafnium dioxide (HfO + + 2 + + ) was then deposited to a thickness of 35 nm by ALD (Lesker ALD150LE), and the thickness was confirmed using spectroscopic ellipsometry (J.A. Woollam). To expose the source and drain contact regions, a second photolithography step using the same bilayer resist stack was carried out, and the patterned HfO + + 2 + + areas were etched using reactive ion etching (PlasmaTherm Versalock) with an Ar/CF + + 4 + + gas mixture. The remaining resist was stripped in acetone, completing the fabrication of the local back-gate structure with HfO + + 2 + + dielectric. +

+

+ Graphene transfer and device fabrication +

+

+ Graphene transfer was carried out using a poly(methyl methacrylate) (PMMA)-assisted wet transfer method. Commercial monolayer graphene on copper foil was spin-coated with PMMA and baked at 150 °C for 2 min. To remove backside graphene, the reverse side was exposed to 25 W oxygen plasma in a Harrick cleaner for 15 min. Copper was then etched in ammonium persulfate solution (Transene company), and the PMMA/graphene stack was transferred onto the prepared gate stack through multiple rinsing in deionized (DI) water baths, followed by nitrogen drying. +

+

+ To enhance adhesion of graphene to the gate-stack, the fabricated substrate was exposed to a brief 1-min plasma treatment before picking up the floating film. Water was allowed to evaporate at 50 °C for 15 min, followed by baking at 150 °C for 15 min. The PMMA layer was removed by soaking in acetone for 4 h, followed by a final bake at 200 °C for 15 min to improve adhesion. To define the active graphene channel, the SPR3012 resist was spin-coated, and photolithography was used to pattern non-channel regions. Exposed graphene was removed via plasma etching, and the resist was stripped by soaking in acetone for 4 h. Finally, an SU-8 encapsulation layer was spin-coated, exposed, and developed to passivate the device and define the electrolyte-exposed sensing window. A commercial platinum wire electrode (BASi, Inc.) was used as the top gate. +

+

+ Circuit integration and electrical measurements +

+

+ Each fabricated GFET was wire-bonded onto a custom-designed printed circuit board (PCB) that housed the measurement and control electronics. The system was managed by an Atmega2560 microcontroller. Analog signal conditioning was performed using TL074 operational amplifiers, with analog-to-digital conversion (ADC) handled by an MCP3204 module and digital-to-analog conversion (DAC) provided by an MCP4822 module. Relay switching between operational modes and grounding states was implemented using ULN2003A Darlington arrays, driven by GPIO lines expanded via an MCP23S08 I/O expander. The implementation of this relay-based multichannel architecture has a distinct impact on operational scalability and data integrity. The platform hosts multiple independent GFET sensors on a single silicon chip, which are individually addressable via the PCB interface. By utilizing electromechanical relays to provide complete galvanic isolation for inactive devices, the system eliminates electrical crosstalk common in passive matrix arrays. This design ensures that the multichannel readout preserves the intrinsic noise floor and sensitivity of a single isolated sensor while facilitating the robust statistical validation presented in this work. +

+

+ Each device could be independently switched to active or grounded states through relay control to minimize electrostatic discharge and leakage during inactive phases. Different biasing configurations (e.g., single-gate static, dual-gate dynamic) were established by selectively routing voltage and current lines through the relays. Drain current was sampled at 1 kHz over a 500 ms window and averaged to yield a steady-state value, while low-frequency noise characterization was performed by collecting 2-s current traces for offline analysis. For feedback modes (DMF, DMS), devices were initialized to a target current corresponding to a back-gate potential of 4–6 V. This biasing strategy ensures the sensor operates within the optimal dynamic range of the feedback amplifier, preventing signal saturation. For static modes (TGF, BGF), the signal is defined as the change in output voltage of the transimpedance amplifier, calculated as + + \({V}_{{out}}={I}_{{DS}}\times {R}_{{gain}}\) + + , where + + \({R}_{{gain}}\) + + is the feedback resistance. +

+

+ A custom graphical user interface (GUI), developed using the Panel Python library, enabled real-time data acquisition, device switching, and visualization. Backend communication with the circuit was implemented using Python and C++ serial protocols. +

+

+ Measurement procedures and setups +

+

+ For pH sensing, potassium chloride electrolyte in DIW was adjusted using hydrochloric acid (HCl) or sodium hydroxide (NaOH), and pH was verified using a VWR Versatile bench-top pH meter calibrated with standard buffer solutions. Each pH value was tested across six devices. +

+

+ Small-molecule sensing (e.g., dopamine, serotonin, epinephrine, ascorbic acid, and uric acid) was performed by preparing stock solutions in PBS and diluting serially to desired concentrations. Measurements were repeated on 6 devices per analyte to ensure reproducibility. Similarly, PFOA was tested with dilutions in DIW. +

+

+ Volatile organic compound (VOC) detection was carried out in a custom-built sealed exposure chamber. Isopropanol was introduced via clean dry air (CDA) gas flow after aspirating a fixed volume of the liquid-phase VOC. Each exposure cycle was followed by purging and baseline recovery. Tests were conducted across 8 devices. +

+

+ For IL-6 biosensing, graphene channels were functionalized using a carbodiimide crosslinking strategy. Briefly, pyrenebutyric acid (PBA) was adsorbed onto the graphene surface, followed by activation with EDC and NHS in MES buffer. Monoclonal IL-6 antibodies were then immobilized, and non-specific binding was blocked using 1% BSA in PBS. Devices were incubated with varying concentrations of IL-6 and washed before measurement. Each concentration point was tested across 6 devices. +

+

+ For concentration response curves, measurements were repeated 6 times per concentration point for each device to assess repeatability. In the plots, data points represent the mean of these measurements, and error bars represent the standard deviation ( + + N + + = 6 scans). To accurately determine the Limit of Detection (LOD), particularly for IL-6 and PFAS, we utilized a log-log linearization method (detailed in Section S + + 9 + + and Fig. + + S9 + + ). The LOD was extracted as the concentration corresponding to a signal of 3σ above the baseline noise floor. +

+

+ Unless otherwise stated, gate sweeps were performed continuously to simulate real-time monitoring conditions. For Top Gate Sweep (TGS) measurements, the gate voltage was swept at a rate of 10 mV/s (4 consecutive sweeps for representative curves). For Back Gate Sweep (BGS) measurements—conducted in ambient air or gas environments—the gate voltage was swept at a rate of 0.2 V/s (3 consecutive sweeps). No inter-sweep delay was applied to capture the transient hysteresis dynamics relevant to rapid sensing applications. For drift measurements, specific protocols were established: Liquid-phase sweeps (TGS/DMS) were conducted at 10 mV/s (approximately 1 sweep/3 min), while gas-phase sweeps (BGS) were conducted at 25 mV/s (approximately 1 sweep/7 min). +

+
+
+
+