Uncovering the doping mechanism of nitric oxide in high-performance P-type WSe₂ transistors

Abstract

Atomically thin two-dimensional (2D) semiconductors are promising candidates for beyond-silicon electronic devices. However, an excessive contact resistance due to ineffective or non-existent doping techniques hinders their technological readiness. Here, we unveil the doping mechanism of pure nitric oxide and demonstrate its effectiveness on wafer-scale grown monolayer and bilayer tungsten diselenide (1L- and 2L-WSe₂) transistors, where doping bands induced by nitric oxide can realign the Schottky barrier and approach p-type unipolar transport. This doping approach, combined with a scaled high-κ dielectric, yields WSe₂ transistors with high performance metrics. For monolayer WSe₂, we achieved an on-state current of 300 μA/μm (at a drain-to-source voltage of –1 V and overdrive voltage of –0.8 V), contact resistance of 875 Ω·μm, peak transconductance of 400 μS/μm, and a subthreshold swing of 70 mV/dec, while preserving on/off ratios >10⁹, minimal variability, and good stability over 24 days under moderate thermal conditions. For bilayer WSe₂, the devices exhibit an on-state current of 448 μA/μm and contact resistance of 390 Ω·μm, further showcasing the scalability and effectiveness of the NO doping method. Our findings establish NO doping as a promising technique for realizing high-performance p-type 2D transistors and advancing next-generation ultra-scaled electronic devices.

Introduction

The continued downscaling of silicon (Si) metal-oxide-semiconductor field-effect transistors (MOSFETs) faces fundamental limits due to short-channel effects (SCEs). As gate lengths approach the ~10 nanometer range, SCEs such as source-to-drain tunneling, drain-induced barrier lowering (DIBL), and reduced electrostatic gate control significantly degrade the performance of Si MOSFETs¹. Semiconducting two-dimensional transition metal dichalcogenides (2D TMDs), on the other hand, offer a promising alternative for ultra-scaled FETs^2,3,4,5,6,7. Their atomic body thickness and ideally self-passivated surfaces enable superior electrostatic control over the channel potential, particularly in the monolayer form. These remarkable properties position 2D monolayer TMDs as compelling candidates for next-generation semiconductor materials^8,9. However, their nature of being “interface-only” transistors renders these atomically thin materials highly susceptible to perturbations from metal contacts, defects, and gate dielectrics^{10,11,12,13,14,15,16,17}. In particular, the presence of metal-induced gap states (MIGS)^12,18,19 and defect-induced gap states (DIGS)^{11,20,21,22,23} at the metal-semiconductor interface can induce strong Fermi-level pinning. In addition to MIGS and DIGS, the gate dielectric can introduce additional trap charges at the contact edges, further pinning the Fermi-level at the metal contact interface and resulting in unfavorable threshold voltages V_TH. This Fermi-level pinning and the lack of robust doping techniques lead to the formation of substantial Schottky barriers (SBs)^{13,19,24,25,26,27}. Accordingly, a highly resistive contact forms at the source/drain-to-channel interfaces, hindering exploiting the full potential of 2D TMD-based devices^5,7.

To address the challenge of high SBs in 2D TMD devices, recent advancements in contact engineering for monolayer TMDs have enabled the development of high-performance n-type FETs (n-FETs). Those devices exhibit on-current density (I_ON) values reaching hundreds of µA/µm at a drain-to-source voltage (V_DS) of 1 V^{12,15,16,18,19,28,29}. However, the superior on-state performance is often accompanied by poor off-state characteristics, such as unfavorably high subthreshold swing (approaching 1 V/dec) and susceptibility to SCEs, thereby undermining the inherent advantages of using 2D materials as channel materials. This issue arises mainly because the semiconductor-dielectric interface is frequently suboptimal in terms of equivalent oxide thickness (EOT) and interface trap density. As a result, a large overdrive voltage V_OV (tens of volts) is frequently required to turn on the transistors. Scaling EOT and improving the dielectric interface is essential for 2D n-FETs operating at much-scaled V_OV < 5 V^{14,15,16,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43}. To satisfy the requirements of future technology nodes, where overdrive voltages should ideally be lower than 1V (V_OV < 1 V), simultaneously optimizing both on- and off-state characteristics becomes paramount for realizing the full potential of 2D transistors^19,44.

Notably, the development of p-type monolayer TMD transistors faces even greater challenges. The on-current density (I_ON) and subthreshold swing (SS) of the best-performing monolayer TMD p-FETs lag significantly behind their n-FET counterparts^{13,20,21,22,23,27}. Recent efforts to implement van der Waals (vdW) p-contact schemes, achieved through refined e-beam evaporation techniques²² or metal-transfer processes⁴⁵, have yet to deliver significant improvement or establish a path towards manufacturable processes that meet the performance metrics for future transistor technologies. Furthermore, simultaneously achieving optimal on- and off-state performances of 2D p-FETs is still lacking.

Another approach to improving the contact resistance of p-type transistors relies on effective doping techniques. While various p-doping strategies have yielded some improvements in on-currents, these frequently come at the expense of low on-off current ratios, which often remain below one order of magnitude^26,27,46,47. In our previous work, we reported a much-improved on-current density (approximately 300 µA/µm) using pure nitric oxide doping on monolayer WSe₂ single crystals¹³. However, this high current density was observed in a small number of devices and was at the expense of degradation in SS (~ 215 mV/dec) and on/off current ratio (~ 10⁶)¹³.

In this study, we report a significant advancement in the performance of 2D p-type transistors fabricated on wafer-scale (2-inch) grown chemical vapor deposition (CVD) 1L-WSe₂ and 2L-WSe₂ films, demonstrating improvements in both on-state and off-state characteristics across hundreds of devices treated with nitric oxide (NO) doping. To gain deeper insights into the impact of NO doping on the SB realignment and associated electron/hole carrier injection, we developed an analytical SB transistor model that accounts for both transport and electrostatics. This model exhibits good agreement with experimental results and doping band calculations using density functional theory (DFT), revealing a significant reduction in SB height from approximately 250 meV to 50 meV after NO doping. Through this doping technique, combined with EOT scaling (high-κ HfO₂ with EOT ~ 1.5 nm), NO-doped 1L-WSe₂ FETs achieved several notable device metrics: I_ON = 300 µA/µm at V_DS = –1 V and V_OV = –0.8 V, low contact resistance, R_C = 875 Ω µm, high transconductance g_m = 400 µS/µm, and low SS ~ 70 mV/dec. Importantly, this approach preserves large on/off current ratios ( > 10⁹) and enables unipolarity in 1L-WSe₂. Furthermore, arrays of NO-doped 1L-WSe₂ transistors fabricated on wafer-scale CVD-grown 1L-WSe₂ films demonstrate relatively small device variability (R_C, SS, I_ON, g_m, and threshold voltage, V_TH) and good stability over 24 days⁴⁸. By extending this doping technique to bilayer WSe₂, devices can achieve an on-state current of 448 µA/µm at V_DS = –1 V and V_OV = –1 V, along with a low contact resistance of 390 Ω·µm, a value that is getting comparable to state-of-art n-type monolayer transistors^12,18,49. Our doping approach enabled by nitric oxide is in principle applicable to other low-dimensional materials. The combination of the above performance specifications (on-state, off-state, contact resistance, variability, and stability) and the proposed doping mechanisms advances our understanding and technological applicability of 2D p-type transistor^{13,21,22,23,45,50,51}.

Results

Mechanism of nitric oxide doping in 1L-WSe₂

To study the doping band formation in NO-doped 1L-WSe₂, Density Functional Theory (DFT) simulations were conducted to elucidate the influence of selenium vacancies (V_Se) and NO-dopants on the electronic band structure of 1L-WSe₂. Figure 1a, b displays the band structures, atomic configurations, and local density of states (LDOS) for undoped and NO-doped 1L-WSe₂, respectively. Notably, in comparison to undoped 1L-WSe₂, V_Se generates a localized state near the valence band edge (as illustrated in Supplementary Fig. 1a). When NO molecules are bound to V_Se sites, these molecules induce doping bands close to mid-gap, as illustrated in Fig. 1b. As a result, the Fermi level shifts downward toward the valence band, similar to conventional doping in three-dimensional semiconductors, which is further illustrated in the schematic band diagram (Fig. 1c). On the other hand, when only oxygen is bound to V_Se sites, there is no additional doping band in the midgap (Supplementary Fig. 1b). To assess the presence and density of selenium vacancies (V_Se) in 1L-WSe₂, we performed scanning tunneling microscopy (STM) measurements on CVD-grown 1L-WSe₂ films (details in Methods). The estimated V_Se density, as shown in Supplementary Fig. 2, is approximately $6\times {10}^{12}$ cm^-2, consistent with typical values reported in the literature⁵².

Fig. 1: Mechanism of nitric oxide doping in monolayer WSe₂.

a Local density of states (LDOS) as a function of the electron energy (E) relative to the Fermi level (E_F) of pristine monolayer WSe₂. Inset: atomic structure of monolayer WSe₂ (gray, tungsten atoms; green, selenium atoms). b LDOS of monolayer WSe₂ with a single nitric oxide (NO) molecule. The presence of nitric oxide dopants introduces an acceptor doping band near the mid-gap region. Insets show the corresponding atomic structures obtained from density functional theory (DFT) calculations. c Schematic density of states (DOS) plots of the doping impact as extracted from a and b. E_C and E_V denote the conduction band minimum and valence band maximum, respectively. The blue shaded regions represent the filled valence bands, while the red shaded region represents filled conduction bands for pristine WSe₂ and the doping-induced acceptor band introduced by NO in WSe₂. X-ray photoelectron spectroscopy (XPS) spectra of d W (4 f) peak; e Se (3 d) peak; and f Raman spectra of monolayer WSe₂ before and after NO treatment at various temperatures. The black lines represent the spectra of the undoped monolayer WSe₂, while the red, blue, and green lines indicate the spectra after NO treatment at temperatures of 100 °C, 150 °C, and 200 °C for 2 h, respectively.

To investigate the surface chemistry of MOCVD-grown 1L-WSe₂ films under NO exposure at various temperatures, we utilized X-ray photoelectron spectroscopy (XPS) to qualitatively analyze the doping effect and support our doping band theory, as illustrated in Fig. 1d,e and Supplementary Fig. 3. Details regarding MOCVD-grown 1L-WSe₂ growth can be found in previous studies^24,38,53. The evolution of W (4 f) spectra clearly demonstrates the change of W chemical states and thus reveals the doping mechanism under NO treatment (Fig. 1d). At the beginning, pristine WSe₂ shows a set of W (4 f) peaks at 32.2 eV and 34.4 eV. After annealing at a NO environment at 100 °C, two new sets of W peaks appear: the set at lower binding energies of 31.8 eV and 34.0 eV, corresponding to W-N species⁵⁴ and the set at higher binding energies of 35.5 eV and 37.7 eV, corresponding to WO₃⁵⁵. As the annealing temperature is increased to 150 °C, the WO₃ peaks become more prominent, while the WSe₂ and W-N peaks become weaker. Annealing at 200 ^oC results in WO₃ peaks dominating and weak W-N peaks in the XPS spectra, indicating most WSe₂ and W-N have been oxidized after high-temperature treatment. Meanwhile, the evolution of Se (3 d) XPS spectra shows that, in addition to WSe₂, SeO₂ components emerged after annealing at high temperatures (150 °C and 200 °C, Fig. 1e). A distinct redshift in the binding energy observed in the Se (3 d) peak and O (1 s) after NO treatment at different temperatures indicates p-type doping in WSe₂, as shown in Fig. 1e and Supplementary Fig. 3.

To further evaluate the effect of the doping temperature on the WSe₂ crystallinity, we have analyzed Raman spectra under the same conditions as XPS (Fig. 1f). The spectra of 1L-WSe₂ films show good crystallinity for treatment temperatures up to 150 °C, as evident from the absence of a significant peak broadening in the A_1g and E_2g vibrational modes. However, after NO treatment at 200 °C for 2 h, a notable drop in intensity suggests partial oxidation and loss of crystallinity. Based on Raman and photoluminescence characterizations (Supplementary Fig. 4), we conclude that NO doping does not compromise the macroscopic crystal structure of WSe_2, despite the considerable oxidation indicated in the XPS analysis.

We also examined the electrical behavior of 1L-WSe₂ films treated under 150 °C for 2 h, as shown in Supplementary Fig. 5. Most devices exhibited no current conduction, despite only part of the 1L-WSe₂ being oxidized, as confirmed by XPS. This insulating behavior resembles that observed in devices subjected to a pure oxygen treatment⁵⁶. These findings underscore the significance of optimizing the doping temperature for incorporating nitrogen into monolayer WSe₂ to achieve the desired electrical characteristics while minimizing oxidation.

To reduce the oxidation impact, we chose an intermediate NO treatment temperature between 100 °C and 150 °C. Specifically, we employed a gentler doping process (120 °C for 30 minutes) for our statistical study on monolayers, while bilayer WSe₂ samples underwent higher doping temperatures and extended durations (130 °C for 2 h). It is important to note that in bilayer WSe₂, the bottom layer can still function as a channel even if the top layer becomes oxidized. The yield under these conditions is higher than 90%, indicating an optimized doping temperature and time.

Unipolar transport and Schottky barrier realignment

Understanding the physics and chemistry involved in the formation of a doping band around mid-gap, along with a shift in the Fermi level toward the valence band, enables us to explain the positive threshold voltage shift of the hole branch, the realignment of the SB, and the p-type unipolarity following NO doping, as illustrated in Fig. 2a. Note that the SB at the metal-semiconductor interface is significantly influenced by Fermi-level pinning, which results from either metal-induced gap states (MIGS) or defect-induced gap states (DIGS) pinning. MIGS are a result of the metal electron wave functions penetrating the semiconductor at the interface, creating a continuum of states and pinning the Fermi level within the bandgap. DIGS pinning occurs when the Fermi level is pinned at the energy level with the highest density of defect states. These two pinning mechanisms compete at the metal/semiconductor interface, with the Fermi level typically fixed at the overall charge neutrality level (Fig. 2a). It is important to note that NO molecules can diffuse into the channel region beneath the source/drain contacts near the edge. If the density of doping states surpasses the influence of MIGS and DIGS, the resulting Fermi-level is primarily determined by the doping density, as illustrated in the band diagram at the metal contact interface (Fig. 2a left).

Fig. 2: Doping-band induced unipolarity and Schottky barrier realignment.

a Schematic of a 2D field-effect transistor (FET) with a monolayer semiconductor (WSe₂) channel, platinum (Pt) metal contacts, and a local bottom gate (LBG) after nitric oxide (NO) doping. The energy band diagram at the semiconductor-contact interface shows how NO doping induces Schottky barrier (SB) realignment via doping-induced pinning (left), while the diagram at the semiconductor-dielectric interface illustrates how the doping band pins the Fermi level (E_F) in the off-state (right). Defect bands (blue for acceptor-like traps and red for donor-like traps) are displayed in the gate dielectric and Fermi window (gray) within the semiconductor channel. V_BG denotes the back-gate voltage, Φ_SB,n and Φ_SB,p refer to the Schottky barrier height (SBH) for electrons and holes, respectively, and ΔΦ_SB represents their change upon NO doping. b I_D-V_G characteristics at V_D = –0.1 V for one representative long-channel device (L_CH = 780 nm) prior and post NO doping at 120 °C for 30 min. I_D is the drain current, V_G is the gate voltage, and V_D is the drain-to-source voltage. The blue shaded square marks the region of reduced SBH for holes (Φ_SB,p), while the red shaded square indicates increased SBH for electrons (Φ_SB,n)and doping-band-induced Fermi level pinning in the bandgap.

Figure 2b compares the characteristics of an exemplary 1L-WSe₂ device before and after NO doping. Notably, the NO doping shifts the threshold voltage of the hole current branch by approximately 0.8 V, enhancing the on-state current at a maximum gate voltage of –4V  by a factor of 10. Furthermore, it is intriguing to observe a widened off-state, with the threshold voltage of the electron branch shifted more positively relative to the hole branch. To fully understand the observed suppression of the electron branch in terms of both magnitude and threshold voltage, we propose a two-fold mechanism (Fig. 2a). Firstly, NO doping reduces the Schottky Barrier height (SBH) for holes at the source/drain, facilitating hole injection from the source into the channel, while simultaneously increasing the electron SBH, which suppresses the electron flow from the drain (left, Fig. 2a). Secondly, the introduced doping band pins the Fermi level within the bandgap, located between the valence band and the doping band itself (shaded grey band in the WSe₂ band profile, Fig. 2a right). This slows down the band movement towards the conduction band edge, leading to a positively shifted electron threshold voltage and an increased gate voltage needed to reach the electron branch. These two combined effects of NO doping on both the contact and channel regions are evident in the significantly enhanced on-state hole current (I_ON) and a substantial stretch of the electron current branch observed in the transfer characteristics of NO-doped 1L-WSe₂ p-FETs (Fig. 2b). We conclude, that the on-current improvement primarily stems from two key changes induced by NO doping: a favorable shift of the threshold voltage (V_TH) toward more positive values and a reduction in the SBH for holes. The observed larger gate voltage window between the electron and hole branches, as well as the much larger SS for the electron branch, are both attributed to inefficient gating due to doping-band-induced Fermi level pinning in the bandgap.

Schottky-barrier MOSFET model with doping band

To quantify the SBH reduction in NO-doped 1L-WSe₂ transistors, we employed an analytic Schottky barrier MOSFET model based on the Landauer formalism (details in Methods and Supplementary Fig. 6 and Note), building upon our previous works^57,58,59,60, which has been adopted to extract various SBH for various 2D semiconductors^59,61. Note that the common approach for extracting SBH based on temperature-dependent measurements, originally introduced by Appenzeller et al. ^62,63, is not applicable to devices with SBHs smaller than 100 meV. As shown in Fig. 2b, the combination of ultra-thin 2D channel bodies, scaled gate dielectrics, and small SB heights in our nitric oxide treated devices results in an indistinct SB slope regime, making it difficult to extract the SBH using the temperature-dependent method.

To investigate the impact of NO doping on the extracted SBH, we applied the SB-MOSFET model (see the Methods section) to undoped long-channel devices and compared the results with those of devices treated with nitric oxide. For undoped long-channel devices (L_CH = 780 nm), the hole SBH (Φ_SB,p) was extracted approximately 250 meV (Fig. 3a). Note that in the Landauer model simulation, we only consider the device characteristics in the off-state ( | V_G | <|V_TH | ), where the scattering from the channel is negligible^59,60. After NO doping (Fig. 3b), the device exhibits a unipolar-like behavior, characterized by a high hole current branch and a delayed, stretched electron branch, along with a substantial reduction of Φ_SB,p from approximately 250 meV to 50 meV. This significant SBH reduction for holes implies a corresponding increase of the SB for electron injection (Φ_SB,n from 1.35 eV to 1.55 eV) and is thus expected to decrease the electron branch current according to the comparison of the red solid lines in Fig. 3a and red dashed line in Fig. 3b. Note that due to the use of an ultra-thin ~6 nm HfO₂ gate dielectric in our experiment (with an extracted screening length of only Λ ~3 nm), a measurable electron tunneling current (red dashed line in Fig. 3b) is expected according to the conventional SB model^57,58,59,60, even with the large Φ_SB,n of 1.55 eV. It is worth noting that the determination of the SBH is subject to some uncertainty due to the ultra-scaled screening length.

Fig. 3: The concept of doping-band induced unipolarity and Schottky barrier realignment.

a, b I_D-V_G at V_D = −0.1 V for one representative long-channel device (L_CH = 780 nm) prior and post NO doping with Landauer model fitting. The solid points are experimental results, while the solid lines are fitting results with the red line describing the electron branch current and the blue line capturing the hole branch current. The extracted Schottky barrier height (SBH) for holes (Φ_SB,p) was found to be approximately 0.25 eV prior doping. After NO doping, the device exhibits unipolar behavior with a substantially reduced Φ_SB,p of 0.05 eV. c x-axis shifted I_D-V_G characteristics of devices prior and post NO doping to align the minimum current (V_G=V_min). A clear on-current enhancement and significant suppression of the electron branch current can be observed. d Illustration of the movement of the Fermi level and the realignment of the Schottky barrier (SB) induced by nitric oxide (NO) doping.

The simulation clearly fails to capture the stretched electron branch observed in the experimental data, primarily due to the conventional SB-MOSFET model’s omission of key electrostatic effects, such as quantum capacitance limits, interface traps, and doping⁶⁴. To gain deeper insights into the impact of NO doping on the electrical characteristics and accurately describe the observed behavior in the electron branch, we refined the SB-MOSFET model by incorporating a “doping band” in WSe₂ (Method and Supplementary Figs. 6, 7). Specifically, we incorporated an NO-induced doping band into the electrostatic equation, following a methodology similar to previous work that accounted for interface traps^64,65,66. The energetic position of the doping band is extracted from DFT calculations (Fig. 1a,b), and the doping density is estimated as ${C}_{G}\times {\Delta V}_{{TH}}/q$, where C_G is extracted using the measured dielectric constant of our 6 nm HfO₂ layer (ĸ~15) and incorporating the capacitance from the 0.3 nm van der Waals gap in series⁶⁷, ${\Delta V}_{{TH}}$ is the experimentally observed V_TH shifts in the hole branch. Good agreement between our simulations and the experiment – including the increased separation of the electron and hole branch – is observed for a doping band located approximately 0.9 eV above the valence band with an energetic spread of 0.15 eV and an integrated doping density of $6\times {10}^{12}$ cm^-2 (red solid line, Fig. 3b). Interestingly, the doping density matches our selenium vacancy density extracted from STM quite well, implying possible NO bonding on the vacancy sites. The simulated Gaussian doping band profile is shown in Supplementary Fig. 7.

By carefully comparing the transfer characteristics before and after doping and aligning them at the same minimum current (V_G=V_min), we can evaluate the impact of SBH reduction on the on-state characteristics (Fig. 3c). It is apparent that the enhancement in hole current in the on-state is primarily a result of the reduced SBH, since the subthreshold characteristics of the hole branches are almost identical, while most of the impact of the NO doping band manifests itself in a stretch of the gate voltage axis. To further illustrate these effects, Fig. 3d depicts Fermi-level movement and SB realignment induced by NO doping. Note that if the doping band is too close to the valence band edge, it could potentially degrade the SS of the hole branch by introducing states similar to interface traps. To validate the low SBH after doping, we conducted cryogenic electrical measurements at 40 K (Supplementary Fig. 8). The output characteristics exhibit nearly linear behavior at a channel length of 180 nm, indicating a small SBH in our devices after NO doping.

Electrical characteristics of NO-doped 1L-WSe₂ p-FETs

To understand the effects of the NO doping on the electrical characteristics and current injection from the contacts, we fabricated transfer length method (TLM) devices using the fabrication processes outlined in the Methods section. The extraction of the dielectric constant of HfO₂ using area-dependent capacitance measurements on a metal-insulator-metal (MIM) structure is shown in Supplementary Fig. 9.

Figure 4a compares transfer characteristics (drain-to-source current, I_DS, versus gate-to-source voltage, V_GS) of representative 1L-WSe₂ devices with channel lengths (L_CH) of 55 nm and 780 nm, both before and after NO doping. NO-doped 1L-WSe₂ FETs exhibit a positive V_TH shift and unipolar p-type transport within the gate voltage span, maintaining strong electrostatic control with an on/off ratio exceeding 10⁹. The on-current at V_DS = -1V and V_OV = –0.8 V reaches 300 µA/µm, and SS of long-channel devices also improves from ~110 mV/dec to ~90 mV/dec after NO doping. Note that the overdrive voltage is defined as V_OV = V_G - V_{TH, LIN}, with V_{TH, LIN} being the linear extrapolation of the threshold voltage at peak-transconductance. The improvement of SS in the deep-off state may be attributed to the annealing removing residuals or interface traps, which needs further study. More transfer characteristics of representative short-channel devices (L_CH of 55 nm) are provided in Supplementary Fig. 10. A clear on-current enhancement can be observed after shifting the I_D-V_G curve to the same V_TH, as shown in Fig. 3c.

Fig. 4: Electrical characteristics and contact resistance of scaled NO-doped 1L-WSe₂ p-FETs.

a I_D-V_G at V_D = -0.1 V and -1V of representative devices with L_CH = 55 nm and L_CH = 780 nm, both before (black) and after NO doping at 120 ^oC for 30 minutes (blue). The subthreshold swing (SS) is determined via the extraction of the minimum slope from the I_D-V_G curves. b I_D-V_D characteristics of representative devices with L_CH = 55 nm and L_CH = 780 nm for gate voltages from –4 V to -2.5 V in steps of 0.25 V. After NO doping, the on-current of short-channel device (L_CH = 55 nm) at V_DS = -1V reaches 300 µA/µm. c Total resistance as a function of channel length (L_CH) at different overdrive voltage (V_OV). Contact resistance (R_C) and sheet resistance (R_sh) extraction using the transfer-length method (TLM) for NO-doped 1L-WSe₂ p-FETs on 6-nm-thick HfO₂ dielectric (EOT ~ 1.5 nm). Total resistance (R_TOT) is extracted at V_DS = -0.1 V. d R_C (left axis) and R_sh (right axis) versus V_OV. e Extracted conductive mobility (µ_con) from TLM structure and field-effect mobility (µ_FE) versus V_OV. Cumulative distribution function (CDF) plots of f R_C and g mobility: Conductive mobility (µ_con) and field-effect mobility (µ_FE). µ_con is extracted from TLM structures and µ_FE is extracted from 780 nm-long-channel NO-doped 1L-WSe₂ p-FETs. h Distribution of I_D-V_D of the short channel (L_CH = 55 nm) NO-doped 1L- WSe₂ p-FETs. The black line represents the total resistance (R_TOT) of these representative short-channel devices.

Figure 4b shows output characteristics (I_DS-V_DS) of NO-doped 1L-WSe₂ FETs (L_CH = 55 nm and 780 nm) for gate voltages from –4 V to –2.5 V in steps of 0.25 V. Channel length scaling significantly boosts the drain current, with I_ON increasing from ~80 µA/µm to ~300 µA/µm at V_DS = –1.5 V. Further scaling and V_TH optimization are essential to achieve even higher performance specs in NO-doped 1L-WSe₂ p-FETs with I_ON-values reaching mA/µm current densities and SS approaching the ideal 60 mV/dec limit. Remarkably, NO-doped 1L-WSe₂ p-FETs can achieve a low contact resistance (R_C) of 875 Ω·µm at a scaled overdrive voltage (V_OV) of –0.9 V. This is comparable to the channel resistance (R_CH = R_TOT-2R_C) of 1.755 kΩ·µm (Fig. 4c, d). We note that in order to extract an accurate value for R_C, R_CH should be of the same order of magnitude or much smaller than R_C. If R_CH dominates such that R_C is over an order of magnitude smaller than R_TOT for the smallest channel in the TLM structure, the validity of the extracted R_C becomes questionable. A more detailed discussion on this subject can be found in the referenced review⁶⁸. Figure 4e shows that the extracted conductivity mobility (µ_con) from this particular TLM is around 40 cm²V⁻¹ s⁻¹, which aligns with the field-effect-mobility (µ_FE) extracted from long-channel length (L_CH = 780 nm) devices.

Figure 4f,g shows cumulative distribution function (CDF) plots of R_C (V_OV = –0.7 V) and mobility extracted from 21 TLM sets. For the mobility extraction, µ_con is extracted from TLM structures at V_OV = –0.7 V, and µ_FE is extracted from the maximum transconductance of long-channel (780 nm) devices. Each data point of µ_con is derived from the analysis of one TLM structure. The median R_C at V_OV = –0.7 V is around 1.94 kΩ·µm, median conductivity mobility is around 50 cm²V⁻¹ s⁻¹, and field-effect mobility is around 63 cm²V⁻¹ s⁻¹. The highest mobility is around 80 cm²V⁻¹ s⁻¹, which is close to the theoretical mobility values predicted on high-κ dielectrics⁶⁹. Figure 4h illustrates the high reproducibility of our short-channel (L_CH = 55 nm) device performance after NO-doping, even on a linear I_D-V_D current scale. The total resistance of these representative short-channel devices is approximately 1.5 to 3 k Ω·µm, which aligns closely with the extracted contact resistance, indicating a reliable extraction of our contact resistance⁶⁸. The statistical distribution of contact resistance values and the mobility extracted from TLM structures is provided in Supplementary Fig. 11. Note that since most of the pristine devices are not reaching the device on-state, it is not meaningful to extract R_C and µ_con from TLM structures in this situation.

To demonstrate the broader applicability and effectiveness of nitric oxide doping on other low-dimensional materials, we also fabricated MoTe₂ back-gated transistors and performed nitric oxide doping. Similar to the device characteristics observed from 1L-WSe₂ channels, nitric oxide doping induces a transition in MoTe₂ FETs from n-type dominant ambipolar behavior to nearly unipolar p-type behavior, as illustrated in Supplementary Fig. 12. More detailed characteristics can be found in our previous work, where the combination of work function engineering and NO p-doping are adopted to achieve high-performance inverters based on few-layer MoTe₂⁷⁰. These results show that NO doping is a relatively universal p-type doping method for low-dimensional materials.

Variability and stability of NO-doped 1L-WSe₂ FETs

To evaluate the variability of the NO doping in 1L-WSe₂ devices upon scaling, we examined the statistics of key device metrics (Fig. 5). Figure 5a–d present I_D-V_G characteristics of a total of 228 FETs with channel lengths (L_CH) varying from 780 nm to 55 nm. An optical image of the fabricated transistor arrays is displayed in Fig. 5e. Figure 5i displays a cross-sectional scanning transmission electron microscopy (STEM) image of our short channel 1L-WSe₂ transistors, which features a 6 nm HfO₂ dielectric, 0.5/30/30 nm Ti/Pt/Au metal contacts, and a channel length of 55 nm. Such a scaled oxide thickness is crucial for achieving both optimal on-state and off-state performance simultaneously. The variability of the threshold voltage (V_{TH, LIN}), subthreshold swing (SS), and on-current (I_ON at an overdrive voltage V_OV = –0.5 V) as a function of L_CH are summarized in Fig. 5f–h. The SS is determined via the extraction of the minimum slope from the I_D-V_G curves.

Fig. 5: Channel length scaling and variability of NO-doped 1L-We₂ FETs.

a–d Distribution of transfer characteristics I_D-V_G of pristine (light-color) and NO-doped (dark-color) 1L-WSe₂ p-FETs with channel lengths of 55 nm (a), 180 nm (b), 380 nm (c) and 780 nm (d) at V_DS = –1 V. e Optical microscopy (OM) of large transistor arrays. Scale bar: 100 μm. f–h Boxplots with Gaussian fitting of key device metrics, including V_TH, _LIN, SS, and I_ON at V_OV = -0.5 V. V_{TH, LIN} is extracted by linear extrapolation at the point of peak-transconductance. The color of the symbols corresponds to the channel length in a–d. For each group of data, a boxplot was created and Gaussian fitting was performed. The error bars represent the standard deviation of each distribution, as obtained from Gaussian fitting. The graphs show the mean (square symbols), standard deviation, lower quartile (25%), median (50%), upper quartile (75%), interquartile range (25–75%), and maximum and minimum (cross symbols). i A cross-sectional scanning transmission electron microscopy (STEM) image of one representative short channel 1L-WSe₂ transistors, which feature a 6 nm HfO₂ dielectric, 0.5/30/30 nm Ti/Pt/Au metal contacts, and a channel length of 55 nm. j Cumulative distribution function (CDF) plots of device metrics. From the left to right, threshold voltage (V_{TH, CC}) extracted at a constant current of 10 pA/µm at V_D = -0.1 V, the minimum subthreshold swing (SS), and maximum transconductance (g_max). k I_D-V_G of one representative 55 nm-long-channel device at V_D = -1V. l Statistical bar graph of maximum current (I_max) from 150 devices. Gray colors denote pristine 1L-WSe₂ p-FETs. Blue colors denote NO-doped 1L-WSe₂ p-FETs, while green for the same devices after 24 days. The dashed horizontal line denotes the averaged I_max of p-FETs.

We did observe a slight change in the variability of device metrics as a function of L_CH. The SS exhibited a slight degradation at the shorter channel length of 55 and 180 nm, likely attributable to a non-ideal back-gate structure with high metal height⁷¹, a non-passivated top surface of 1L-WSe₂¹⁵, or an un-optimized dielectric interface between hafnium dioxide (HfO₂) and WSe₂¹⁵, as also evident from negative V_{TH, LIN}-values ranging from –2.5 V to –3.5 V among all the channel lengths. Further studies are needed to uncover the impact of scaling (length and EOT) and gate stack process, which is not within the scope of this study. To fairly compare on-current densities among devices after NO doping, the same overdrive voltage (V_OV = V_G - V_{TH, LIN} = –0.5 V) is employed across all devices. Upon scaling, the on-state current (I_ON) increased by a factor of approximately four, as shown in Fig. 5h.

To assess the long-term stability and evaluate the reproducibility of our NO doping approach, we employed cumulative distribution function (CDF) plots and extracted mean values to analyze V_TH and SS for all devices across all channel lengths, as well as maximum transconductance (g_max) for short-cannel devices (55 nm). To compare devices before and after NO doping, we define threshold voltage, V_{TH, CC} at a constant current of 10 pA/µm with V_D = –0.1 V, as most of the pristine devices do not reach the on-state to extract V_{TH, LIN}. This analysis involved comparing pristine devices, freshly NO-doped devices, and NO-doped devices after 24 days (Fig. 5j). The distribution of transfer characteristics of corresponding devices is shown in Fig. 5a–d and Supplementary Fig. 13. Specifically, NO-doped 1L-WSe₂ p-FETs exhibited a positive shift in V_{TH, CC} (ΔV_{TH, CC} ~ 0.7 V) compared to pristine devices (blue vs. black), as shown in Fig. 5j. The positive shift is attributed to a doping-induced Fermi-level shift. This improvement in SS (ΔSS ~ –20 mV/dec) suggests that NO-doping can also be considered a means to optimize the channel and gate dielectric interface, similar to dipole engineering in silicon transistors. Notably, the smallest SS reaches ~70 mV/dec, representing a relatively low value among high-performance TMD p-FETs^{13,21,22,23,45,50,51}. From this comprehensive analysis, it is apparent that NO doping can effectively tune V_TH and improve off-state performance. In terms of on-state performance, NO-doped 1L-WSe₂ p-FETs achieve a mean transconductance of g_max ~ 300 µS/µm (Fig. 5j). Furthermore, the minimal change in the statistical distribution of the maximum current (I_max) from 150 FETs (L_CH = 55 nm) over a 24-day period confirms the good stability of NO doping (Fig. 5k,l). Unlike previous reports on NO₂ or NO_x doping, pure nitric oxide doping is crucial to maintain the doping effect due to its molecular bonding on the vacancy site, as predicted by previous studies²⁶. Additionally, pure nitric oxide doping did not lead to increased gate leakage current, as illustrated in Supplementary Fig. 14.

To assess the thermal stability of NO doping in WSe₂, we conducted annealing experiments at temperatures up to 200 °C for 30 min in a nitrogen environment, followed by several hours of cool-down to 50 °C, as shown in Supplementary Fig. 15. The devices demonstrate improved electrical performance after annealing at 100 °C, with slight degradation observed at 200 °C, indicating that the NO-induced doping remains stable under these moderate thermal conditions.

High-performance NO-doped 2L-WSe₂ FETs

To further illustrate the robustness of nitric oxide doping, we fabricated bilayer WSe₂ (2L-WSe₂) transistors with TLM arrays using a similar fabrication process but on a 3.5 nm thin high-κ HfO₂ dielectric (EOT ~ 0.85 nm), as depicted in the device schematics (Fig. 6a) and the transmission electron microscopy (TEM) image in Fig. 6b. The Raman spectra exhibit the multilayer B_1g vibration mode at ~309 cm⁻¹ before and after NO doping, suggesting a well-preserved crystallinity after doping. Figure 6d, e shows the linear and log-scale transfer characteristics, of pristine 2L-WSe₂ devices, after NO doping at 130 °C for 1 h (red), followed by another NO doping at 130 °C for 2 h (blue), and final NO doping at 140 °C for 2 h (green). As shown in the output characteristics (Fig. 6f), after the second doping step, the on-current can reach as high as 448 uA/um at V_DS = –1 V and V_OV = –1.25 V, which is 60 times greater than that of the pristine device at such a scaled voltage. Note that extending doping time further improves on-state performance but at the cost of SS degradation (~ 155 mV/dec). These results suggest that the doping band within the bandgap can overlap with the Fermi window (the gray window in the band diagram in Fig. 6g), thereby reducing the gate efficiency in the device off-state. It is essential to optimize both on-state and off-state performance in NO-doped devices. Heavier (degenerate) p-type doping achieved through increasing doping temperature and time could potentially be employed in the spacer region for future nanosheet transistor structures. The extracted contact resistance from the hero TLM structure can reach below 390 Ω·µm, which is among the lowest reported values among 2D p-type transistors. Note that the extracted contact resistance is close to half the total resistance (Fig. 6f), implying an accurate TLM extraction⁶⁸.

Fig. 6: High-performance NO-Doped 2L-WSe₂ FETs with ultralow contact resistance.

a Schematic of a 2D FET with a bilayer WSe₂ channel after NO doping. b A cross-sectional scanning transmission electron microscopy (STEM) image of a representative short-channel bilayer WSe2 transistor, featuring a 3.5 nm HfO₂ dielectric, 0.5/20/20 nm Ti/Pt/Au metal contacts, and a channel length of 55 nm. c Raman spectra of bilayer WSe₂ before and after NO treatment at 130 °C for 1 h  d linear-scale and e log-scale I_D-V_G at V_D = –1V of representative devices with L_CH = 55 nm, both before (black) and NO doping at different steps. f I_D-V_D of a short channel (L_CH = 55 nm) NO-doped 2L-WSe₂ p-FET. The black line represents the total resistance (R_TOT) of this representative short-channel device. g Schematic of the energy band diagram of the semiconductor-dielectric interface, where the doping band starts approaching the Fermi-window when increasing the doping temperature and time duration. Defect bands (blue for acceptor-like traps and red for donor-like traps) are displayed in the gate dielectric and Fermi window (gray) within the semiconductor channel. h Total resistance as a function of channel length (L_CH) at V_OV = -1V for two TLM sets. Total resistance (R_TOT) is extracted at V_DS = -0.1 V.

Supplementary Fig. 16a, b presents a histogram of the contact resistance (R_C) distribution for NO-doped bilayer WSe₂ p-FETs and the cumulative distribution function (CDF) of R_C for both NO-doped monolayer and bilayer WSe₂ devices, clearly showing that bilayer WSe₂ devices exhibit lower median R_C (692 Ω·µm) compared to monolayer devices (1.94 kΩ·µm). This comparison underscores the advantage of bilayer WSe₂ in achieving reduced contact resistance due to a smaller bandgap and potentially enhanced doping efficiency in the bilayer configuration.

Discussion

In Fig. 7a–g, we have benchmarked our devices against the best-reported 2D TMD p-type transistors utilizing various doping and contact strategies. A comprehensive benchmarking table that includes all the key device metrics is shown in Table 1^{13,21,22,24,25,26,27,38,46,50,51,72,73,74}. Figure 7a, b presents the comparison of log-scale transfer characteristics of 1L-WSe₂ and 2L-WSe₂ p-type FETs in our work at V_DS = –1 V alongside previous literature, including our earlier work^13,21,22,51. In this comparison, the gate voltages V_G are shifted to maintain a constant off-state current (I_OFF) of 100 pA μm⁻¹, a typical value for low-power (LP) logic devices. Note that this benchmarking comparison is common practice across various channels and technologies. While some previous studies demonstrated on-currents in the tens of µA/µm range (Fig. 7a), those devices required significantly higher overdrive voltages (tens of volts) and typically exhibited much lower on/off-current ratios than the ~9 orders of magnitude we achieved^13,21,22,51. Our device delivers one of the highest I_ON within a smaller gate voltage window (i.e., supply voltage V_DD = V_G-V_OFF) of 2.5 V. Note that most prior 1L-TMD p-FETs employ a thicker EOT ( > 10 nm) compared to our devices. This highlights the crucial need of EOT scaling to optimize both on- and off-state performance simultaneously (Fig. 7b). In particular, in our previous work presented at the International Electron Device Meeting (IEDM), we introduced a hybrid doping strategy, which integrates NO molecular and WO_x charge transfer doping⁷². A high on-current I_ON of 545 μA/μm at a V_DS of –1 V was demonstrated, although with degradation of off-state performance (SS ~ 550 mV/dec) in this case.

Fig. 7: Benchmarking state-of-the-art 2D p-FETs.

a Transfer characteristics of NO-doped 1L- and 2L-WSe₂ p-FETs with 55-nm L_CH and the recently reported high-performance p-FETs. b Zoom-in characteristics of p-FETs, highlighting the scaled V_G-V_OFF due to scaled equivalent oxide thickness (EOT) in this work. V_OFF is extracted from constant off-current (0.1 nA/µm). c I_ON vs. SS of state-of-the-art 2D p-FETs and d g_m vs. SS of 2D p-FETs with scaled dielectric. Benchmarking plots of e I_ON vs. L_CH f I_ON vs. I_ON / I_OFF of state-of-the-art 2D p-FETs. The channels include multilayer and monolayer WSe₂ devices. g R_C vs. carrier density of reported p-contacts in the literature. The blue and green stars mark the results of this work. A benchmarking table (Table 1) includes the key device metrics shown in the benchmarking plots.

Table 1 Benchmarking state-of-art WSe₂ p-type transistors

For a fair comparison of on-state and off-state device performance metrics, we benchmark the I_ON (at V_DS of -1 V) vs. SS and g_m vs. SS for state-of-the-art p-FETs in Fig. 7c, d. Our devices reach the desirable corner (higher I_ON, g_m, and lower SS). In particular, the monolayer devices exhibit a transconductance (g_m) almost four times higher and a lower subthreshold swing (SS) superior to most of previous reports (Fig. 7d)^73,75. Figure 7e, f shows the benchmarking plots of I_ON vs. L_CH and I_ON vs. I_ON / I_OFF of state-of-the-art 2D p-FETs, respectively. Our devices exhibit a higher current across various channel lengths with an I_ON / I_OFF ratio approximately up to 10⁹. Additionally, our NO-doped 1L-WSe₂ p-FETs achieve relatively low contact resistance, while NO-doped 2L-WSe₂ p-FETs reach one of the lowest contact resistance and one of the highest transconductance among all TMD p-FETs (Fig. 7g). Note that the contact resistance demonstrated here is even lower than some reports on multilayer materials with essentially no on-off ratio ( < 10), as shown in Table 1. Further refinement of gate dielectric stacks to reduce charge trapping, hybrid doping strategies to enhance carrier densities, and contact interface engineering to minimize the impact of source/drain tunneling barriers is needed to fully realize the potential of monolayer or bilayer WSe₂ as a channel material for ultra-scaled p-FETs. Moreover, further work will be necessary to improve the long-term and thermal stability of the NO doping effect.

Methods

CVD monolayer WSe₂ film

2-inch monolayer WSe₂ (1L-WSe₂) films were purchased from 2D semiconductor cooperation.

CVD bilayer WSe₂ film

Bilayer WSe₂ (2L-WSe₂) films were provided by Nexstrom Pte. Ltd.

DFT Calculation

A first-principle calculation was performed based on density-functional theory (DFT) by using the Vienna ab initio simulation package (VASP). Electron-electron interactions were modeled using the Perdew-Burke-Ernzerhof (PBE) exchange-correlation function. The structures were relaxed until the total energy change was below 10⁻⁵ eV. The energy cutoff and k-points were set to 650 eV and 11 × 11 × 1, respectively, for WSe₂. The band structure of 2 × 2 × 1 WSe₂ supercells with one Se vacancy was simulated.

Scanning tunneling microscopy (STM) characterization

The sample was prepared using the same wet transfer process mentioned in device fabrication, utilizing the same material source. It was then loaded into the Omicron ultrahigh-vacuum (UHV) Scanning Tunneling Microscope (STM) chamber and underwent 24 h of UHV annealing to eliminate potential contaminants and residues. The sample was scanned at room temperature with an iridium-platinum tip, and the topographical images were recorded at a tunneling current of 500 pA and ± 0.5 V. The STM data were analyzed using WSxM software.

X-ray photoelectron spectroscopy (XPS) characterization

XPS was acquired using a non-monochromatic Mg Kα X-ray radiation (hν = 1253.6 eV) at 150 W with high-resolution spectra recorded at a constant pass energy of 20 eV. The electron energy analyzer was Omicron EAC 125, and the analyzer controller was Omicron EAC 2000. XPS data were analyzed using CasaXPS software with Finite Lorentzian (LF) function and Shirley background function used for the line shape and background fittings, respectively.

Optical characterization

Raman spectroscopy was conducted with a Thermo Scientific DXR3xi Raman Imaging Microscope using a 532 nm green laser source with a spot size of around 1 μm. Each Raman mapping was conducted over ~10 μm² on a MOCVD-grown 1L-WSe₂ film.

Device fabrication

A local bottom gate (LBG), composed of a Cr (3 nm)/Au (12 nm) metal stack, was defined using electron-beam lithography. Subsequently, an approximate 6 nm hafnium dioxide (HfO₂) layer was deposited via atomic layer deposition (ALD) at 200 °C to serve as the gate dielectric. Monolayer WSe₂ films (CVD growth), grown on a sapphire substrate, were transferred onto the LBG. Reactive ion etching using Cl₂/O₂ at a power of 40 W was employed to define a channel width of 0.42 μm. Then, e-beam lithography was employed again to define Ti (0.5 nm)/ Pt (30 nm)/Au (30 nm) source/drain (S/D) contacts. The devices are then subjected to a flow of nitric oxide gas.

Electrical characterization

The electrical characterization under vacuum (~ 1 × 10⁻⁵ torr) was performed using a Keysight 4156 C Precision Semiconductor Parameter Analyzer. The cryogenic electrical characterization was performed in a Lakeshore CRX-VF cryogenic probe station using the Keysight B1500A Semiconductor Device Parameter Analyzer.

Scanning transmission electron microscopy (STEM) characterization

High-resolution scanning transmission electron microscopy (STEM) was conducted on a Thermo Fisher Scientific TALOS200X operated at 200 kV. For the TEM samples, fabrication was doneusing a Thermo Fisher Helios 4 G Dual Beam SEM/FIB. A thin Pt protection layer was deposited on the top surface to protect the sample from ion beam damage. The lamella was then carefully thinned to ∼100 nm by using ion milling under low voltage (2 keV).

Analytical Schottky barrier transistor model

We have employed Landauer formalism and WKB tunneling to model the impact of the SB⁵⁹, while taking into account the quantum capacitance. For estimating contact resistance values, we assume that total transmission is governed by WKB tunneling at the source and drain. We note that this assumption becomes less accurate as the device approaches its on-state, requiring more precise channel transmission modeling. However, for devices in their off-state and near on-state conditions (V_G < V_TH), our simulations effectively approximate the device transport behavior, facilitating good SBH extraction. A triangular-shaped Schottky barrier is assumed to model WKB tunneling. A similar approach was employed in our previous work on black phosphorus⁵⁹. This assumption is valid for small screening lengths as present in our devices with thin gate dielectrics.