Introduction

Measuring the ionic content or pH of an analyte is critical for a wide variety of applications, ranging from medical diagnostics, early disease detection, and genome sequencing to industrial process control, homeland security, and environmental monitoring1,2. Solid-state sensors, such as the ion-sensitive field-effect transistors (ISFETs), offer considerable benefits over state-of-the-art electrochemiluminescence, electrochemical, and optical sensing platforms for point-of-care applications. ISFETs enable rapid, real-time, low-power, low-cost, and label-free electronic detection in a compact setup that is highly compatible with the standard complementary metal–oxide–semiconductor (CMOS) process3,4. In addition, ISFETs allow on-chip integration of sensor and measurement circuitry on the same substrate using monolithic integration5. Unlike metal–oxide–semiconductor (MOS) field-effect transistors (FETs), the gate metal in ISFETs is removed and replaced by a fluid gate which consists of a reference electrode capacitively coupled to the oxide layer through an electrolyte solution6. The pH sensing is based on the modulation of surface potential and channel conductance, when hydrogen ions in the electrolyte combine with hydroxyl groups on the gate dielectric, causing a shift in the threshold voltage7. This shift in threshold voltage for a unit change in pH is defined as the voltage sensitivity of the pH sensor8. For classical ISFET devices, the maximum achievable voltage sensitivity for pH sensing is 59 mV pH−1 at 25 °C, known as the Nernst limit9. Enhancement of sensitivity, beyond the Nernst limit, is necessary to improve the resolution of the pH sensor, especially for detecting small pH changes in biological samples with ultralow concentrations. Several efforts have been made to achieve super-Nernstian sensitivity (>59 mV pH−1) in ISFETs. Recent studies have exploited the steep-subthreshold behavior in negative capacitance FETs10 and tunnel FETs11 to demonstrate ISFETs with sensitivities beyond the Nernst limit. A more common approach to enhance sensitivity uses capacitive amplification through asymmetric dual gate configurations9,12,13,14,15,16. Compared with classical single gate pH sensors, the combination of a fluid gate and back gate offers significant signal amplification, overcoming the charge screening effect caused by the counter ions in the electrolyte17. The threshold voltage shift is modulated by the top- (Ctop) and bottom- (Cbottom) gate capacitances, allowing the sensitivity to be increased by simply adjusting the ratio, Ctop/Cbottom18. A simple approach to achieve super-Nernst sensitivity is by increasing the thickness of bottom dielectric or changing the dielectric constant (making Cbottom < Ctop). However, thicker dielectric limits miniaturization and restricts the use of these sensor devices in point-of-care applications. Considering the ultimate aim of scaling down the device to nanoscale dimensions, an alternate method of enhancing sensitivity needs to be explored, which is independent of the thickness of the gate dielectric. This can be achieved by exploiting the interface traps and defects in order to modulate the device capacitance, without (necessarily) increasing its dimensions.

Several studies have reported ultrascaled FET-based biosensors using transition-metal dichalcogenide (TMDC)-based two-dimensional (2D) materials, like molybdenum disulfide (MoS2) or tungsten diselenide (WSe2), as the channel material19,20,21,22,23. Compared with silicon, the atomically thin 2D materials offer desired scalability, good electrostatics, and reduced flicker noise, leading to highly sensitive nano biosensors with excellent signal-to-noise ratio (SNR)24. Recently, FETs with Van der Waals WSe2(top)/MoS2(bottom) vertical heterojunction have been intensively investigated for their unique electrical properties such as band-to-band tunneling, negative differential resistance, and ambipolar transfer characteristics, arising from the atomically sharp interface and strong interlayer coupling between the MoS2 and WSe2 films25,26,27,28,29. However, the use of TMDC 2D heterostructures has been limited to novel electronics, tunneling FETs, memory devices, photodetector, etc. In most of these applications, the nonidealities such as interface traps, defects, etc., deteriorate device performance. Alternatively, these nonideal effects can also be exploited and even engineered for sensing applications, as reported in recent literatures for super-Nernstian graphene-ISFETs30,31,32. In these devices, sensitivity above the theoretical Nernst limit is achieved through the process of charge-transfer doping, which directly modulates the carrier density of graphene with changes in pH level. However, this requires the fabrication of defect-engineered nanocrystalline graphene grain boundaries, which may add to processing complexities. Here, we propose a WSe2(top)/MoS2(bottom) pH sensor, which uses the interface effects of Van der Waals heterostructure to surpass the Nernst sensitivity limit of 59 mV pH-1 at room temperature. Unlike the defect-engineered graphene sensors, these still utilize an oxide to sense the pH by protonation or deprotonation. The sensing and signal transduction are decoupled using a double-gate FET architecture (fluid gate for sensing and back gate for gate voltage), while the charge screening due to the heterostructure interface is used to provide a gain over the Nernst limit.

In this work, we present experimental demonstration of a WSe2(top)/MoS2(bottom) vertical heterostructure ISFET for pH sensing beyond the Nernst sensitivity limit. The double-gated FET structure uses a thin high-k hafnium oxide (HfO2) as the top and bottom dielectric. The heterostructure enables the modulation of back gate transconductance by introducing a screening layer of interface and channel charges. This enables super-Nernst sensitivity by lowering effective back gate capacitance without increasing the device dimension and compromising the overall device performance. The atomically thin 2D heterostructure channel along with use of thin high-k dielectric allows extreme device scalability that is crucial for next generation nano-biosensors, targeted for point-of-care applications. In addition, we have also used technology computer-aided design (TCAD) simulations to elucidate the working principle of our device and also to investigate pH sensors with the flipped MoS2(top)/WSe2(bottom) heterostructure as the channel material.

Results and discussions

Device characterization

Figure 1a shows the schematic of the ISFET device with the WSe2(top)/MoS2(bottom) vertical heterostructure. Few-layer flakes of exfoliated MoS2 (bottom) and WSe2 (top) were stacked together and sandwiched between two ~30 nm thick, HfO2 layers, serving as the top and the bottom gate dielectrics. The bottom HfO2 is deposited on a highly doped p-type silicon (Si) substrate, which also acts as a global back gate. Previous studies have reported pH sensors with HfO2 as the sensing surface due to the high pH sensitivity, low drift voltage, low hysteresis, and low body effect of HfO233,34,35. Moreover, the medium permittivity and high bandgap of HfO2 provide excellent chemical stability when in contact with Si and silicon dioxide35. The MoS2 layer in the proposed device was contacted by nickel (Ni) electrodes and showed n-channel operation36. Palladium (Pd) electrodes were used to make the contacts to the top WSe2 layer and shows p-channel (hole) transport37. The individual contacts on the flakes were also connected to form a single pair of source/drain contact. Figure 1b shows the optical image of the WSe2(top)/MoS2(bottom) heterostructure, with the source and drain metal contacts. High-resolution transmission electron microscope (HR-TEM) imaging was used to analyze the cross section of the fabricated device. The HR-TEM images shown in Fig. 1c show excellent interface quality with ~1.2 nm Van der Waals gap between HfO2–MoS2 and MoS2–WSe2 interfaces. This gap is expected due to the dry transfer of the exfoliated flakes. There is minimal gap in WSe2–HfO2 interface, as the top HfO2 is deposited directly on WSe2. Both the top and bottom HfO2 dielectric are also of excellent quality without any visible cracks or nonuniformities. The same was also confirmed by atomic force microscopy (AFM) on the fabricated device (on the channel after top HfO2), which also showed good uniformity of deposition without any island formation (see Supplementary Fig. 1). The measured physical characteristics of the top and bottom HfO2 films are provided in Supplementary Table 1. The top HfO2 was slightly thicker and showed a slight increase in roughness due to process variations. A more detailed physical and surface analysis of the same HfO2 films has been reported in the recent study by Ganapathi et al.38

Fig. 1: Device structure and dry electrical characteristics.
figure 1

a Schematic of the WSe2(top)/MoS2(bottom) ISFET device, b optical image, showing the WSe2 on MoS2 along with the metal contacts (scale bar 10 μm), c HR-TEM of the channel region of HfO2/WSe2(top)/MoS2(bottom)/HfO2 stack (scale bar 5 nm), and d transfer and e output characteristics of the ISFET measured in the dry environment.

Full size image

The device was initially characterized in a dry environment by measuring the transfer and output characteristics as a function of back gate (VBG) and drain voltage (VDS), respectively. The transfer curve, in Fig. 1d, shows ambipolar characteristics, with p (VBG ↓ ID ↑) and n regions (VBG ↑ ID ↑) labeled. The transfer characteristics of the reverse sweep are considered to focus on the hole transport region. For VBG < 0 V, the p-type WSe2 turns ON and starts conducting current through holes (p-channel operation), whereas for VBG > 0 V, n-type MoS2 begins to conduct current by electrons (n-channel operation). The ON regions of these two materials are mutually exclusive, i.e., when one is ON and conducting, the other is fully depleted or even inverted. At large VBG, minority inversion carriers can form in one channel while the other is in accumulation. However, their contribution toward the overall device current is very small. The space charge regions at the p–n junctions of the hetero-interface, as well as the corresponding band offsets, pose a significant potential barrier for the minority carriers to overcome. Furthermore, measurements on individual back-gated MoS2 and WSe2 FETs show minimal currents in inversion mode due to the large Schottky barrier at their respective contacts (see Supplementary Fig. 2). Also, the lack of any negative differential resistance (NDR) eliminates the possibility of band-to-band tunneling in our devices28. Our findings are consistent with the results reported by Lee et al. for a similar WSe2(top)/MoS2(bottom) heterostructure39. The curves in Fig. 1e display the output characteristics of the proposed device at different VBG. The drain current increases with increasing VBG during n-channel operation (from VBG = 0 to 2 V), whereas it increases with decreasing VBG for p-channel operation (from VBG = −0.5 to −2 V). The linearity of the curves indicates good (~ ohmic) contacts with a low Schottky barrier.

The current in the n-channel region is ~2 orders higher than the p-channel due to better back gate coupling for MoS2. The back-gated transfer characteristics of the standalone WSe2 FET (Supplementary Fig. 2) show much higher currents, highlighting weaker back gate control in the WSe2(top)/MoS2(bottom) heterostructure. The characteristics of the n-channel region are similar to that of a back-gated MoS2 FET, showing minimal effects of heterostructure on the bottom layer. The reverse is valid for top-gate coupling, which allows the WSe2 layer to be better sensitive to pH changes during wet measurements. The bidirectional sweep of the VBG reveals hysteresis in the device. The hysteresis direction confirms the lack of any ferroelectric polarization in the HfO2 films. The clockwise direction of the hysteresis arises from interface traps in the HfO2–MoS2 and MoS2–WSe2 interfaces40. These interface traps play a key role in reducing the back gate transconductance without the need for a thicker bottom gate dielectric. The density of these interface traps (Dit) is crucial for understanding the operation of the device. MOS capacitors are a popular way to extract interface characteristics. However, the small area of the exfoliated flakes (compared with the pads) prevents such measurements. The subthreshold slope (SS) can provide an approximation of Dit (higher Dit causes larger SS). But the complexity of the device structure and the transfer characteristics make such approximations difficult. Hence, an analysis based on TCAD simulations is provided, which numerically estimates a Dit of more than ~1012 cm−2 at the MoS2–WSe2 interface41.

pH sensing with WSe2(top)/MoS2(bottom) heterostructure ISFET

For the wet characterization and pH sensing, phosphate buffered saline (PBS) solutions of different pH values were prepared. These solutions were drop-casted onto the top HfO2 layer of the ISFET. A platinum wire, biased at 0 V, was used as the reference electrode (VFG). This reference electrode, in contact with the electrolyte solution, also serves as the fluid gate of the ISFET. The back gate transfer characteristics were obtained at multiple pH values, ranging from pH = 5.65 to 8.25, in both forward- and reverse-sweep directions, as shown in Figs. 2a and 2b, respectively.

Fig. 2: Wet electrical characteristics and pH sensing.
figure 2

Transfer characteristics of the WSe2(top)/MoS2(bottom) ISFET at different pH values during a forward and b reverse sweep of the back gate voltage (VBG), c schematic of the pH sensing mechanism showing the formation of surface charge on oxide and electric double layer (EDL) in the electrolyte, and d simplified equivalent model showing signal transduction in a double-gated device.

Full size image

The effective surface potential of the top dielectric is determined by the combination of fluid gate voltage, applied through the reference electrode, and the surface charge, modulated by the pH value of the electrolyte. A lower pH value results in protonation (OH + H+ = OH2+), giving rise to positive surface charges, while at higher pH values, the top HfO2 surface is deprotonated (OH–H+ = O−), yielding negative charges. These surface charges are partially screened by the counter-ions in the electrolyte, which move toward the charged surface and forms an electric double layer. The resultant surface charge or the change in potential at the oxide surface (\({\Psi}_{0_{{{{\mathrm{pH}}}}}}\)) causes a corresponding shift in the device threshold voltage, which, in turn, determines the sensitivity of the pH sensor. A schematic of this sensing mechanism is shown in Fig. 2c.

Figure 2d represents a simplified model of the signal transduction once a surface charge and potential have developed at the oxide surface. A dual gated configuration can be represented as two transistors in parallel with a common source and drain. The top gate with a transconductance \(g_{{{{\mathrm{m}}}}_{{{\mathrm{t}}}}}\) acts as the sensor for the oxide surface potential \({\Psi}_{0_{{{{\mathrm{pH}}}}}}\) (represented as an ac source) due to the pH. The fluid gate bias VFG is kept at a reference voltage of 0 V. The device current is controlled by the back gate voltage (VBG) and has a transconductance of \(g_{{{{\mathrm{m}}}}_{{{\mathrm{b}}}}}\). When operated at constant current mode, the voltage sensitivity is given by Eq. 1. Hence, super-Nernstian sensitivity can be obtained if \(g_{{{{\mathrm{m}}}}_{{{\mathrm{b}}}}} <\, g_{{{{\mathrm{m}}}}_{{{\mathrm{t}}}}}\) providing a gain in the sensitivity12,42,43.

$$S_{{{\mathrm{V}}}} = \frac{{{\Delta}V_{{{{\mathrm{BG}}}}}}}{{{\Delta}{{{\mathrm{pH}}}}}} = \left( {\frac{{g_{{{{\mathrm{m}}}}_{{{\mathrm{t}}}}}}}{{g_{{{{\mathrm{m}}}}_{{{\mathrm{b}}}}}}}} \right)\left( {\frac{{{\Delta}{\Psi}_{0_{{{{\mathrm{pH}}}}}}}}{{{\Delta}{{{\mathrm{pH}}}}}}} \right)$$
(1)

To measure the sensitivity of our device, we plot VBG against the corresponding pH of the electrolyte solution at a fixed drain current (IDS = 55 nA), as shown in Fig. 3a. The voltage sensitivity (Sv) is derived directly from the slope and amounts to 362 mV pH−1, clearly exceeding the Nernst limit by more than six times. The response, with a nonlinearity of 9%, indicates excellent sensing performance. The Sv can be tuned by adjusting the VBG, as shown in Fig. 3b. Here, Sv is determined at a constant drain current and plotted with respect to the VBG of the transfer curve at pH = 7 for both forward and reverse sweeps. The peak Sv in the subthreshold, linear, and saturation region of the WSe2(top)/MoS2(bottom) ISFET is displayed in Figs. 3c and 3d for the forward and reverse sweeps, respectively. In the forward sweep, Sv is high only in the extreme linear regions of both p and n sides. On the other hand, the p-channel region in the reverse sweep shows high Sv throughout subthreshold, linear, and saturation regions. In fact, the highest Sv (362 mV pH−1) is obtained in the reverse sweep when the back gate is biased in the linear region of hole transport. Although super-Nernstian sensitivity is visible in both the forward and the reverse sweeps, it is more prominent in the reverse sweep direction. The enhanced sensitivity, beyond the theoretical Nernst limit of 59 mV pH−1, originates from the transconductance (gm) modulation between the fluid-gate and the back gate, since \(g_{{{{\mathrm{m}}}}_{{{\mathrm{b}}}}} <\, g_{{{{\mathrm{m}}}}_{{{\mathrm{t}}}}}\).

Fig. 3: Sensitivity.
figure 3

a Back-gate voltage (VBG) against pH at constant drain current, b Voltage sensitivity (SV) with respect to the VBG of the transfer curve at pH = 7. Peak SV at linear, saturation, and subthreshold region of operation during c forward sweep and d reverse sweep for the WSe2(top)/MoS2(bottom) ISFET.

Full size image

In the present work, the gm mismatch arises from the hetero-interface properties, whereas it originates from capacitance mismatch in devices with a thicker back gate dielectric44. The donor traps in the MoS2-WSe2 interface capture holes from WSe2 during the reverse sweep and reduces \(g_{{{{\mathrm{m}}}}_{{{\mathrm{b}}}}}\), leading to enhanced Sv. In addition, the combination of screening effect45 and mobility degradation46 increases Sv in the linear region of operation. In the former case, the inversion charges in the bottom MoS2 channel lower the \(g_{{{{\mathrm{m}}}}_{{{\mathrm{b}}}}}\) when the device is biased in strong linear region of the p-channel side. In addition, high vertical electric field degrades the mobility of carriers through the channel through surface scattering, reducing \(g_{{{{\mathrm{m}}}}_{{{\mathrm{b}}}}}\) and improving Sv. A similar effect is observed in the strong linear region of the n-channel region during the forward sweep. In all cases, the lowest Sv is observed in the subthreshold region, primarily due to the high gate leakage.

Charge-screening effect for high sensitivity

To further understand the role of screening and interface traps in enhancing Sv, we investigate the electrical characteristics of WSe2(top)/MoS2(bottom) heterostructure ISFET using TCAD. All simulations were carried out with Silvaco ATLAS device simulation tool, which self-consistently solves Poisson’s equations along with current continuity equations to obtain the drain current as a function of gate voltage. Silvaco ATLAS TCAD device simulator tool cannot directly model devices based on 2D materials. As a workaround, we have modeled MoS2 and WSe2 using the material parameters available in the literature and considering the 3-D equivalent of their 2-D density of states. Using a few layers (~6 nm) of exfoliated flakes means that the quantum effects dominant in monolayer or bilayer materials can be avoided. This approach is consistent with previous studies which used TCAD tools to simulate 2-D material based FETs, with reasonable accuracy41,47,48,49. The mobile ions in the electrolyte solution are modeled as charge carriers in a semiconductor material such that pH values correspond to the charge density at the electrolyte/dielectric interface50. The use of HfO2 as the gate dielectric and sensing layer has been well explored in the literature and usually shows near-ideal sensitivity even in single gate configurations15,33,35. Hence, the modeling of surface protonation or deprotonation on the HfO2 surface is not included in the simulations. The TCAD model is calibrated against the drain current vs back gate voltage curve of pH = 7 in the reverse-sweep direction. Figure 4a shows a good agreement between experiment and simulation. A slight deviation is observed at high VBG due to gate-leakage in the practical device, which is not captured in the TCAD simulations. The inset of Fig. 4a shows the simulation results for a wide range of pH values.

Fig. 4: TCAD simulations.
figure 4

a TCAD simulations calibrated against experimental transfer curve at pH = 7 in the reverse-sweep direction for the WSe2(top)/MoS2(bottom) ISFET. The inset shows simulation results for pH values ranging between 5 and 9. b Drain current vs fluid-gate voltage characteristics at different pH values, extracted using TCAD simulations. c Simulated magnitude of the electric field across the WSe2 channel v/s gate voltage (for VFG and VBG when the other is set to 0 V). d Simulated electric field contours with applied VFG and VBG (arrows indicate the field direction). e Simulated density of channel (WSe2) charge v/s gate voltage (for VFG and VBG when the other is set to 0 V).

Full size image

Figure 4b shows the variation in the drain current with respect to the change in the fluid-gate voltage (VFG) at different pH values, when VBG = 0 V, as obtained from the TCAD simulations. The sensitivity with change in VFG is found to be constant across all operating regions and is ~59.5 mV pH−1, very close to the Nernst limit. This is expected as there is no gain mechanism with a single gated device. However, the device can still provide near-ideal sensitivity even when operated as a single gated ISFET.

The TCAD simulations confirm the presence of interface traps (~1012 cm−2) at the MoS2–WSe2 interface as well as inversion carriers in the bottom MoS2 channel (Supplementary Fig. 3). The higher density of traps in WSe2–MoS2 interface and their distribution of energy levels have a profound impact on the device operation. The much lower currents and larger subthreshold slope of the p-channel region are due to these donor-interface traps that capture holes from the WSe2 channel as back gate bias is applied. On the contrary, the traps in the HfO2–MoS2 interface are shallower and of lower density, with minimal effects on the n-channel region. The donor traps in WSe2–MoS2 interface effectively weaken or screen the controlling of the WSe2 channel charges by the applied back gate voltage. In addition, the inversion charges that form in the bottom MoS2 layer do not contribute significantly to device current (due to band offset barriers) but can again weaken the back gate control by screening.

To illustrate the effects of screening, parameters such as electric field and charge are plotted against the back gate and fluid gate voltages when the other is set to 0 V (Fig. 4(c–e)). Figure 4c shows the magnitude of the electric field across the WSe2 channel, which is an order of magnitude lower with the back gate bias than with the applied fluid-gate voltage. The distribution of the electric field is illustrated in Fig. 4d for VBG = −3 V (VFG = 0 V) and VFG = −3 V (VBG = 0 V). With the back-gate voltage, the field lines are mainly confined within MoS2, whereas with the applied fluid-gate voltage, there is much higher field penetration. This is evident by considering the electric field direction in the respective channels. Even at VBG = −3 V (VFG = 0 V), the field in the WSe2 channel is directed away from the back gate, i.e., there is a certain amount of band pinning at the WSe2–MoS2 interface due to the interface traps. However, this does not occur when VFG = −3 V (VBG = 0 V). These effects are also visible in the charge induced in the WSe2 channel by the respective gates, which is shown in Fig. 4e. The channel charge in WSe2 induced by the back-gate bias is much lower than the channel charge formed by the fluid-gate for a considerable range of voltages. These voltages correspond to the range in VBG where the super-Nernstian response is visible. These factors emphasize the role of screening in reducing the back-gate coupling and the necessity of interface and inversion charges to cause it. This reduced back-gate coupling and hence reduced \(g_{m_b}\) directly translates to the high sensitivity in our heterostructure-based ISFET.

MoS2(top)/WSe2(bottom) pH sensor

Next, we simulate the flipped MoS2(top)/WSe2(bottom) heterostructure device, using the calibrated TCAD model, with MoS2 (n-type) stacked on top of WSe2 (p-type). To be consistent with the simulation setup for WSe2(top)/MoS2(bottom) ISFET, the same donor trap density has been used, while the acceptor traps are being ignored. Although, in a real device, the presence of acceptor traps will reduce the current in the n-region, having an identical simulation setting provides a fair comparison between the MoS2(top)/WSe2(bottom) and the WSe2(top)/MoS2(bottom) ISFETs. Figure 5a shows the simulated drain current of the flipped MoS2(top)/WSe2(bottom) ISFET, plotted against VBG for different pH values at VFG = 0 V. Compared with the WSe2(top)/MoS2(bottom) ISFET, the drain current obtained is higher in the MoS2(top)/WSe2(bottom) ISFET. This is because of the lower Schottky barrier height in MoS2 and the absence of acceptor traps in the interface in the simulation setup. The Sv of the simulated MoS2(top)/WSe2(bottom) device, shown in Fig. 5b, displays super-Nernstian sensitivity when the n-layer is ON. On the contrary, the sensitivity remains within the Nernst limit in the p-region. The density of inversion electrons (Ninv) in the WSe2 layer, with change in VBG, is also displayed in Fig. 5b. A higher Ninv corresponds to an enhanced SV, confirming the contribution of the screening effect in improving sensor performance. Mobility degradation due to higher vertical field at higher VBG may also play a role in enhancing SV.

Fig. 5: Simulated MoS2(top)/WSe2(bottom) ISFET.
figure 5

a Drain current vs back-gate voltage characteristics at different pH values, when VFG = 0 V. b Voltage sensitivity and density of inversion carriers (Ninv) in the bottom channel against back-gate voltage.

Full size image

In summary, we have demonstrated a double gated super-Nernstian ISFET for pH sensing, using a 2D heterostructure of WSe2(top)/MoS2(bottom) as the channel material. The proposed device shows a maximum sensitivity of 362 mV pH−1 for pH sensing due to the reduced back-gate transconductance. The novelty of this work lies in the underlying sensing mechanism that is attributed to the traps and defects formed at the heterointerface. The bottom semiconducting channel acts both as a screening layer and a site for interface traps, contributing to the enhanced sensitivity. The proposed device and its flipped variant (MoS2(top)/WSe2(bottom)) both offer high sensitivity with great tunability, depending on the region of operation (subthreshold, linear, and saturation). Although the Nernst limit can be surpassed in other dual-gate ISFETs by tailoring the thickness and the material of the top and bottom gate dielectrics, the thick back gate oxide limits device scaling. The proposed device, on the other hand, with thin high-k gate oxides and atomically thin 2-D channel materials, offers extreme scalability, enabling ultrascaled, next-generation biosensing platforms for point-of-care applications.

Methods

Device fabrication and material characterization

In all, 30-nm-thick HfO2 was deposited (e-beam evaporation) on p++ Si substrate, which is used as a global back gate. The HfO2 thin film deposition parameters were optimized based on the reports by Ganapathi et al.38. MoS2 flakes (source: Graphene supermarket) were mechanically exfoliated onto this substrate using scotch-tape and polydimethylsiloxane (PDMS) stamps. The samples were cleaned with acetone, isopropyl alcohol (IPA), and deionized water after each process. Flakes, ranging between 4 and 7 nm in thickness, were identified optically and later confirmed using AFM (Bruker) (Supplementary Fig. 4). Source/drain contacts with a channel length of 1 µm were patterned using e-beam lithography (Raith Pioneer), and nickel/palladium (nickel in contact with MoS2) of thickness 15 nm each were deposited using e-beam evaporation, followed by liftoff. WSe2 flakes (source: 2D Semiconductors) of 4–7 nm thickness were transferred on top of the MoS2 flake using PDMS stamps after aligning them using the stage of an optical lithography setup (MJB4 mask aligner). Source/drain contacts of palladium/gold, each with a thickness of 15 nm, were connected to the WSe2 layer, aligned with the contacts on MoS2. The structural and chemical integrity of the flakes were verified using Raman spectroscopy (Supplementary Fig. 4). HfO2 (~30 nm) was deposited on top, forming the top dielectric, which acts as the pH sensing layer. This layer also isolates the electrodes from direct contact with the electrolyte. Oxide on the contact pads was etched using buffered hydrofluoric acid (BHF) solution. For the cross-section HR-TEM of the device, the sample was prepared using focused ion beam (FIB) Helios G4-UX and imaged using Titan Themis 300 kV from ThermoFisher.

Device characterization

PBS was used as the electrolyte. PBS solution with different pH values, ranging from pH = 5 to 8, was prepared and the pH was accurately determined using a standard laboratory pH meter (Eutech Instruments). The electrolyte solution is then drop-casted (~10 µl) onto the exposed region (top HfO2 layer) of the ISFET for sensing. A fluid gate, made up of a platinum electrode, connects to the top HfO2 layer through the electrolyte solution, and is used to maintain a constant reference potential during measurement. Finally, the electrical measurements were carried out using Keysight B1500 Semiconductor device analyzer at room temperature.

TCAD simulations

The current–voltage characteristics of WSe2(top)/MoS2(bottom) and MoS2(top)/WSe2(bottom) ISFET heterostructure devices were simulated using Silvaco ATLAS simulator51. Material parameters of MoS2 and WSe2, such as carrier mobility, were defined based on the reports by Mirabelle et al.47,48 and Arora et al.52. Shockley–Read–Hall (SRH) and Auger models53,54 for carrier generation–recombination were also included. Quantum effects have been ignored in the TCAD to simplify the simulation procedure. The electrolyte was modeled using an undoped semiconductor. The details of this modeling are given in Supplementary Notes.

The simulated ISFET was then calibrated against the experiment–transfer curve, corresponding to pH = 7. The device parameters such as mobility, doping, interface traps, and Schottky barrier heights were adjusted in the TCAD model to match the experimental results accurately. These final values after calibration of these parameters are provided in Supplementary Table 2, along with the material parameters used in the simulation. To better fit the subthreshold region of the transfer curve and its shape, interface traps and fixed–oxide charges were added at various interfaces. These are summarized in Table 3 of the Supplementary Information.