Introduction

Two-dimensional (2D) transition metal dichalcogenides (TMDCs)1, particularly molybdenum disulfide (MoS2), have emerged as key materials in the development of advanced nanoelectronic and optoelectronic devices2. Their atomic-scale thickness and tunable, layer-dependent electronic properties3 make them highly attractive for next-generation applications, including low-power, flexible, and transparent devices4,5,6,7,8,9. Monolayer MoS2, with its direct bandgap of 1.8 eV10, strong photoluminescence (PL)11, and remarkable stability1, exhibits an on/off current ratio of 108 in n-type field-effect transistors (FETs)12, solidifying its prominence in 2D semiconductor research. However, the realization of stable p-type MoS2 is equally critical for fully unlocking its potential in complementary metal-oxide-semiconductor (CMOS) technology based on 2D materials13,14,15. The ability to precisely control the transition between n-type and p-type behavior16,17 in MoS2 would enable the formation of complementary logic circuits18 and p–n junctions19,20, both fundamental building blocks for integrated circuits, transistors, and optoelectronic devices21,22,23. Despite its importance, achieving stable and controllable p-type doping in MoS2 remains a significant challenge, limiting the development of fully integrated TMDC-based circuits.

Various methods have been explored to achieve p-type doping, including plasma treatment24,25,26, charge transfer27,28, and contact engineering28,29. However, these approaches are often complex, difficult to scale, and result in limited control over doping levels. More importantly, they frequently produce nonuniform doping, which is unsuitable for practical, large-scale device fabrication. Substitutional doping has been shown to induce p-type conductivity in MoS230,31. However, these studies were limited to microscale or local doping, whereas wafer-scale uniform substitutional incorporation is still challenging. Recently, Ghoshal et al. reported a light-activated dedoping approach in wafer-scale MoS232, which demonstrates excellent spatial fidelity and scalability, but relies on interface trap states and offers limited direct control over dopant species and concentration profiles. In contrast, substitutional doping via chemical vapor deposition (CVD) has shown promise as a more stable and intrinsic method for achieving p-type MoS233,34,35,36,37,38. Among these dopants, Nb atoms, when incorporated into the MoS2 lattice, induce hole carriers by modifying the electronic structure through charge transfer from Nb d-orbitals39, offering greater control over doping concentration and stability. However, the carrier mobility in Nb-doped MoS2 is considerably lower than that of pristine MoS2, often due to increased scattering and structural distortions caused by the Nb atoms36. Enhancing device mobility remains a critical challenge. Furthermore, most studies on Nb-doped MoS2 have focused on small-scale single crystals33,34,36, typically limited to a few micrometers. Achieving uniform, wafer-scale doping with precise control of electronic properties is still a major hurdle, but is crucial for the practical integration of MoS2 in large-scale electronic applications.

Here, we demonstrated the metal-organic chemical vapor deposition (MOCVD) growth of wafer-scale Nb-doped MoS2 films on c-plane sapphire substrates, achieving precise control over doping levels to induce unconventional p-type behavior. By first growing monolayer MoS2 via MOCVD and then introducing Nb dopants, Nb atoms were substitutionally incorporated into the MoS2 lattice, achieving wafer-scale uniformity and precise doping controllability, as confirmed by Raman, photoluminescence (PL), scanning transmission electron microscopy (STEM), energy-dispersive X-ray spectroscopy (EDX), electron energy loss spectroscopy (EELS) and X-ray photoelectron spectroscopy (XPS) measurements. A theoretical model based on surface reaction kinetics and reaction thermodynamics is employed to simulate Nb substitution on MoS2, revealing a clear doping correlation and highlighting how the interplay between temperature and precursor pressure governs substitution efficiency. Moreover, the Nb-doped MoS2 exhibited a transition from n-type to p-type behavior with increasing doping levels, achieving an on/off current ratio of 104 and an average field-effect mobility of 2.36 cm2 V−1 s−1, with a device yield of 90%, placing it among the well-performing p-type MoS2 reported to date. Our findings represent a major advancement in the development of p-type MoS2 and establish a scalable method for fabricating wafer-scale 2D materials with precisely controllable doping levels. This capability is crucial for enabling complementary logic circuits and the realization of CMOS technology using 2D materials, paving the way for the widespread integration of TMDC-based electronic devices in next-generation low-power, flexible, and transparent electronics.

Results and discussion

Substitutional niobium doping in monolayer MoS2 via MOCVD

Nb-doped MoS2 was synthesized using a home-built MOCVD system, schematically illustrated in Supplementary Fig. 1. A 2-inch c-plane sapphire wafer served as the growth substrate, while molybdenum hexacarbonyl (Mo(CO)6), diethyl sulfide ((C2H5)2S), and niobium ethoxide (Nb2(OC2H5)10) were employed as gaseous precursors to ensure stable, controllable, and uniform growth. The growth process involved several stages by continuously tuning the growth temperature and precursor supply: monolayer MoS2 growth, an intermediate interval, and subsequent Nb doping, as illustrated in Fig. 1a. Further details on the growth process are provided in Supplementary Fig. 2 and the “Methods” section.

Fig. 1: Substitutional niobium doping in monolayer MoS2 via MOCVD.
Fig. 1: Substitutional niobium doping in monolayer MoS2 via MOCVD.
Full size image

a Schematic illustration of the MOCVD process for Nb-doped MoS2 growth. b Optical micrograph of the as-grown Nb-doped MoS2 film on a sapphire substrate. c, f Experimental and simulated STEM images of Nb-doped MoS2, respectively. QSTEM simulations were performed using the same parameters as those used in the experimental setup. d, e Intensity profiles along the lines shown in (c, f), respectively. g, h HAADF-STEM image and corresponding EELS elemental mapping of Nb-doped MoS2; Mo, S, and Nb signals are represented in green, yellow, and purple, respectively. Scale bars: 0.5 nm. i–k Magnified and fitted EELS spectra of the Nb, Mo, and S edges, respectively, showing the systematic evolution of spectral intensity with increasing Nb doping concentration. The original data is provided in Supplementary Fig. 4. l AFM image with a height profile of the as-grown Nb-doped MoS2 on sapphire, corresponding to the region shown in (b). m, n Raman and PL spectra of pristine MoS2 and Nb-doped MoS2 at various doping levels.

Figure 1b displays a typical optical micrograph of the Nb-doped monolayer MoS2 film on a c-plane sapphire substrate. A clear contrast between the continuous Nb-doped MoS2 film and the bare sapphire substrate is observed by intentionally creating a scratch. To identify Nb atoms in the MoS2 lattice and accurately determine the doping levels, STEM measurements, EDX, EELS and XPS analysis were performed on Nb-doped MoS2 samples. A typical STEM image in Fig. 1c reveals the atomic lattice structure of monolayer MoS2 in the 2H phase. The line profiles in Fig. 1d show a lower intensity along Nb atoms compared to Mo atoms, further supported by quantitative STEM simulations in Fig. 1e, f. However, since Nb and Mo differ by only one atomic number in the periodic table, distinguishing them based on atomic contrast in STEM images is challenging. Therefore, EDX spectroscopy was employed to identify Nb dopants, as shown in Supplementary Fig. 3. Although the K and L edge energies for Nb and Mo are very close, non-negative matrix factorization was employed for peak deconvolution to resolve the overlapping signals and accurately differentiate Nb and Mo contributions. Elemental mapping and spectra from the corresponding regions illustrate the spatial distribution of S, Mo, and Nb atoms, confirming a uniform distribution of all three elements within the analyzed area and enabling quantitative determination of the Nb doping concentration.

Furthermore, as shown in Fig. 1g, h, the EELS elemental mapping clearly reveals the spatial distribution of Nb together with Mo and S signals, indicating that Nb atoms are substitutionally incorporated into the MoS2 lattice. In the EELS spectra (Supplementary Fig. 4), distinct energy-loss edges corresponding to S-L2,3 (orange), Nb-M4,5 (purple), and Mo-M4,5 (green) are clearly observed. Compared with pristine MoS2, the Nb-doped samples display an emerging Nb-M4,5 edge at approximately 205–208 eV, confirming the incorporation of Nb within the lattice. The magnified and fitted spectra (Fig. 1i–k) show a systematic increase in the Nb-related peak intensity with higher Nb concentration, while the Mo-M4,5 and S-L2,3 edges exhibit only subtle variations, further supporting the substitutional replacement of Mo by Nb atoms. Quantitative evaluation of the Nb concentration was performed by integrating the intensity of the Nb-M4,5 edge and normalizing it to the Mo-M4,5 edge using their respective ionization cross-sections, according to: \({C}_{{{\rm{Nb}}}}=\frac{{I}_{{{\rm{Nb}}}}/{\sigma }_{{{\rm{Nb}}}}}{({I}_{{{\rm{Nb}}}}/{\sigma }_{{{\rm{Nb}}}})+({I}_{{{\rm{Mo}}}}/{\sigma }_{{{\rm{Mo}}}})}\), where \({I}_{{{\rm{Nb}}}}\) and \({I}_{{{\rm{Mo}}}}\) are the integrated edge intensities and \({\sigma }_{{{\rm{Nb}}}}\), \({\sigma }_{{{\rm{Mo}}}}\) are the corresponding partial ionization cross-sections. In addition, XPS measurements were conducted across multiple regions of the wafer (see Supplementary Fig. 5) to quantify the large-scale doping concentration. The Nb concentrations derived from EDX, EELS, and XPS analyses are in good agreement, confirming the controllability of Nb incorporation achieved by MOCVD growth.

Figure 1l demonstrates the atomic force microscopy (AFM) image captured from the scratched region in Fig. 1b, revealing a smooth surface with a thickness of 0.7 nm, confirming the successful monolayer growth. Figure 1m, npresent Raman and PL spectra, respectively, collected from pristine MoS2 and Nb-doped MoS2 with varying doping levels. All samples exhibit a characteristic Raman peak separation (E12g and A1g) of less than 20 cm−1, along with a strong PL A exciton emission and no detectable lower-energy indirect transition, confirming their monolayer nature and direct band-gap character. Compared to pristine MoS2, Nb-doped monolayer MoS2 samples exhibit blue-shifted Raman signals and PL peaks, with the magnitude of the shifts increasing with higher doping levels. The presence of sapphire signals at the same location confirms that the shift is not due to noise. The observed shifts in Raman and PL spectra can be attributed to strain and charge doping effects induced by Nb doping, consistent with previous reports33,36. Meanwhile, the FWHM of the Raman modes remains nearly unchanged across doping levels, indicating that Nb substitution does not introduce detectable long-range lattice disorder, as Nb and Mo have nearly identical atomic radii (rNb = 1.46 Å, rMo = 1.45 Å).

Thermodynamic-kinetic modeling of Nb doping

To gain insight into the doping behavior of Nb atoms during MOCVD growth, we developed a reaction model that considers the adsorption, substitution, and decomposition processes occurring on the MoS2 surface. As shown schematically in Fig. 2a, gaseous Nb species are delivered to the surface and adsorbed. It indicates that Nb atoms preferentially substitute Mo sites rather than occupy interstitial positions, as substitution exhibits a significantly lower formation energy (≈ −3.4 eV, see Supplementary Information). The displaced Mo atoms are subsequently desorbed from the surface after being replaced by Nb. This process follows a thermally activated reaction pathway, where the substitution must overcome an energy barrier ΔEa, and competes with both desorption and decomposition reactions. While the diagram illustrates a simplified substitution mechanism, assuming a constant effective sticking coefficient and active-site density under steady-state MOCVD growth conditions, the actual process involves multiple concurrent surface reactions, which are captured by our kinetic model (detailed in Supplementary Information).

Fig. 2: Thermodynamic-kinetic modeling of Nb substitution on MoS2.
Fig. 2: Thermodynamic-kinetic modeling of Nb substitution on MoS2.
Full size image

a Schematic illustration of the thermally activated substitution mechanism. Nb precursors are adsorbed onto the MoS₂ surface, followed by thermal activation that enables substitution at Mo lattice sites. The energy barrier ΔEa separates the adsorption and substitution states. Precursor flux, governed by Nb partial pressure, controls the frequency of surface collisions and affects substitution efficiency. b Peff-sub as a function of Nb partial pressure, illustrating that increased flux enhances substitution probability. Inset: pressure-dependent trends at different temperatures. c Effective substitution probability Peff-sub as a function of temperature, showing a non-monotonic behavior due to the competition between substitution and decomposition. Inset: temperature-dependent trends under different precursor partial pressures. d Contour map of Peff-sub as a function of temperature and Nb partial pressure. Red dashed lines and gray gridlines indicate cross-sections corresponding to data in the insets of (b, c), respectively.

To evaluate substitution efficiency, we define the effective substitution probability Peff-sub, which refers to the effective probability per unit time for Nb atoms adsorbed on the surface to replace a Mo atom through surface exchange reactions (both the desorption of displaced Mo atoms and the bond formation of the substituted Nb with neighboring S atoms). Figure 2b shows that increasing the partial pressure of Nb precursors (PNb, by adjusting the carrier gas flow rate of the Nb precursor) leads to a steady rise in Peff-sub, primarily due to the enhanced flux of incident species and increased surface coverage (see Supplementary Equation 1). However, when PNb exceeds a certain threshold, the increase in doping concentration gradually slows down, which may be attributed to precursor decomposition saturation or other kinetic limitations. In contrast, the effect of temperature is non-monotonic, as seen in Fig. 2c. At low temperatures, surface reactions are kinetically limited, and substitution is inefficient. As the temperature increases, the reaction rate rises, leading to a higher substitution probability. However, further temperature increase accelerates decomposition processes, ultimately reducing Peff-sub. This trend highlights the existence of an optimal thermal window, where substitution is favored while decomposition remains suppressed (see Supplementary Equation 6).

The combined influence of temperature and precursor pressure is visualized in Fig. 2d, where a continuous surface map of Peff-sub reveals a distinct reaction space. Along the temperature axis, substitution efficiency increases and then decreases, while it rises monotonically with increasing pressure. This indicates that high precursor flux can be broadly beneficial, whereas temperature must be finely tuned to balance activation and stability. Such mapping provides a valuable reference for experimentally identifying growth windows that enable precise and uniform doping across large areas.

To further examine parameter sensitivity, the insets of Fig. 2b, c present cross-sections of the surface map in Fig. 2d, showing temperature-dependent substitution probabilities at different Nb partial pressures and, conversely, pressure-dependent trends under varying temperatures. At higher pressures, the temperature window for effective substitution broadens, and the peak probability increases, indicating enhanced doping efficiency with abundant precursor supply. Conversely, at elevated temperatures, Peff-sub becomes more sensitive to precursor pressure, highlighting the need for coordinated control of both parameters. These results reinforce the coupled nature of thermodynamic and kinetic effects in determining substitution efficiency during MOCVD growth. Supplementary Table 1 summarizes the experimental growth parameters and measured Nb doping concentrations, which clearly illustrate the influence of different growth parameters on the doping outcome and its extent, providing experimental validation of the proposed model.

Despite capturing the essential thermodynamic-kinetic balance governing Nb incorporation, the present model remains a simplified and primarily qualitative framework. Consequently, while the model successfully reproduces the general doping trends observed experimentally, a fully quantitative description remains challenging. Future work will therefore focus on integrating first-principles calculations with machine-learning-assisted modeling to establish a more quantitative and predictive correlation between growth parameters and doping behavior. Such an approach is expected to provide deeper mechanistic understanding and more practical guidance for optimizing large-scale, controllable p-type doping in 2D semiconductors.

Wafer-scale uniformity of Nb-doped MoS2 film

We next investigated the wafer-scale uniformity of Nb-doped MoS2 films. Figure 3a presents a typical photograph of a wafer-scale Nb-doped MoS2 film on a 2-inch c-plane sapphire substrate. Optical micrographs and AFM images, taken from various regions across the 2-inch film, demonstrate excellent surface quality and uniformity, with a roughness of less than 200 pm, as shown in Fig. 3b, Supplementary Figs. 6 and 7. The AFM images display very few grain boundaries, indicating large grain size and high growth quality across the entire wafer. By selective oxidation at elevated temperature40, the grain size of the as-grown Nb-doped MoS2 films was estimated to be over 100 µm from optical microscopy observations (Supplementary Fig. 8). Due to the presence of unsaturated bonds and higher chemical reactivity along grain boundaries, Nb atoms are more likely to incorporate through these boundary regions. However, direct experimental confirmation of this hypothesis remains challenging, as the limited spatial resolution of EELS and STEM prevents precise localized compositional analysis at individual grain boundaries. Figure 3c, e present Raman and PL line scans across the 2-inch Nb-doped MoS2 film with a doping concentration of 2% (11 spectra with 5 mm steps), with no significant variations in peak position or linewidth observed. Slight variations in Raman peak intensity and position can be observed, which originate from local strain fluctuations induced during wafer-scale MOCVD growth. To further assess uniformity, high-resolution Raman and PL mappings were performed on several areas of the same wafer, as shown in Supplementary Figs. 9 and 10. Statistical analysis of 10,000 Raman and PL spectra revealed an average separation between the E12g and A1g modes of 19.48 cm−1 with a standard deviation of 0.17 cm−1 (Fig. 3d), and an average PL peak position of 1.834 eV with a standard deviation of 3.6 meV (Fig. 3f ). Additionally, EDX measurements, as shown in Supplementary Fig. 11, were collected from various areas of the wafer to evaluate the uniformity of Nb doping. The elemental maps for Nb, Mo, and S reveal a consistent and even distribution of dopants across the entire film, indicating that Nb doping is uniformly incorporated throughout the Nb-doped MoS₂ film. The absence of significant fluctuations in doping levels or material composition highlights the precision of the MOCVD method in controlling both doping concentration and material quality over large areas. These results confirm the exceptional uniformity of our Nb-doped MoS2 films, spanning from the nanometer to centimeter scale. This uniformity in elemental composition further supports the high-quality growth of the film, ensuring consistent structural and electronic properties across different regions of the wafer, which is crucial for reproducibility in large-scale device applications.

Fig. 3: Wafer-scale uniformity of Nb-doped MoS2 film.
Fig. 3: Wafer-scale uniformity of Nb-doped MoS2 film.
Full size image

a A photograph of the as-grown Nb-doped MoS2 film on a 2-inch sapphire substrate. b AFM image of the as-grown Nb-doped MoS2 film, demonstrating a smooth surface on the sapphire substrate. c, e Raman and PL line scans, respectively, taken at room temperature across the 2-inch Nb-doped MoS2 film on a sapphire substrate, with 11 spectra collected at 5 mm intervals. d, f Statistical distributions of the Raman peak separation between E12g and A1g (Δ), and the PL peak position, derived from the 10,000 spectra collected from the mapping zones shown in Supplementary Figs. 6 and 7, respectively.

Electrical transport in wafer-scale Nb-doped MoS2

To evaluate the electrical transport properties of the Nb-doped MoS2 film, an array of two-terminal FET devices was fabricated by transferring the samples onto a Si/SiO2 substrate, as depicted in Fig. 4a. The device structure, illustrated in the inset of Fig. 4a, consists of Ti/Au contacts serving as the source and drain electrodes, with the Si substrate acting as the back gate and a 270 nm SiO2 layer functioning as the gate dielectric. The detailed fabrication process is provided in the “Methods” section.

Fig. 4: Electrical transport in wafer-scale Nb-doped MoS2.
Fig. 4: Electrical transport in wafer-scale Nb-doped MoS2.
Full size image

a Optical microscope image showing an array of two-terminal devices fabricated after transferring the Nb-doped MoS2 film onto a Si/SiO2 substrate. Inset: Schematic side view of a two-terminal device. Scale bar: 100 μm. b Gating characteristics of Nb-doped MoS2 with various doping levels, compared to pristine MoS2. c Gating characteristics of 80 FET devices with 2% Nb-doped MoS2 from the array shown in (a). Inset: linear scale representation of the data. d Distribution of the on-state current for 2% Nb-doped MoS2 across 80 FETs at Vds = 100 mV and Vg = −60 V. e Statistical distribution of field-effect mobility and subthreshold swing (SS) for the devices shown in (c). f Performance comparison of p-type MoS2 across various studies via different dopants and methods. “ML” represents monolayer and “FL” represents few layers.

Figure 4b presents the gating characteristics of pristine MoS2 and Nb-doped MoS2 with varying doping levels. The doping concentrations were precisely controlled during the MOCVD process and verified through EDX analysis, as detailed in previous sections. The transfer curve for pristine MoS2 (red curve) exhibits typical n-type behavior, with an on/off current ratio of 105 and a field-effect mobility of 9.63 cm2 V−1 s−1 (defined as \(\mu =\frac{d{I}_{{ds}}}{{{dV}}_{g}}\times \frac{L}{W{C}_{i}V}\), where L and W are the channel length and width between two voltage probes, respectively, Ci is the capacitance between MoS2 channel and the back gate per unit area, and V is the voltage measured between the two electrodes)12. In the case of 0.5% Nb-doped MoS2 (orange curve), n-type characteristics are still observed, though the on/off ratio and mobility are significantly reduced compared to pristine MoS2. This decline in performance is likely attributed to the charged-impurity scattering and carrier trapping induced by substitutional Nb dopants, which create local potential fluctuations and slightly perturb the lattice. Interestingly, 1% Nb-doped MoS2 (green curve) begins to exhibit ambipolar behavior, where both electron and hole conduction are present. At negative gate voltages, a small p-type conduction emerges, while n-type behavior is retained at positive gate voltages. This ambipolarity is attributed to the higher concentration of Nb dopants, which shifts the Fermi level and modifies the band structure, enabling hole conduction at negative gate voltages. As the Nb doping level increases further, p-type behavior becomes more pronounced, starting from 1.5% Nb-doped MoS2. The transfer curve for the 1.5% Nb-doped sample shows a clear shift toward p-type conduction, characterized by an on/off current ratio of approximately 1 × 103 and a field-effect mobility of 2.06 cm2 V −1 s−1. This transition to p-type behavior suggests that Nb dopants effectively alter the band structure, lowering the Fermi level closer to the valence band and thereby promoting hole transport at negative gate voltages. In the 2% Nb-doped MoS2 sample, the p-type behavior becomes even more dominant, with a higher on/off current ratio of approximately 1 × 104 and a field-effect mobility of 2.43 cm2 V −1s −1, indicating enhanced hole transport as the Nb concentration increases. This further incorporation of Nb enhances p-type conductivity by increasing the density of available holes, reducing electron concentration, and shifting the Fermi level deeper into the valence band. The improved on/off ratio and mobility at this doping level suggest that 2% Nb doping strikes a better balance between maintaining structural integrity and facilitating efficient p-type charge carrier mobility.

The gating characteristics and on-state current distribution for 80 FET devices, each with a 5 μm channel length, fabricated over a 5 × 4 mm2 area, are presented in Fig. 4c, d. Due to the exceptional uniformity of the MOCVD-grown Nb-doped MoS2 film, a device yield of 90.0% was achieved, with a 15.2% variation in on-state current (σIon/Ion, where σIon and Ion represent the standard deviation and mean value, respectively) across the centimeter-scale region. Statistical analysis of field-effect mobility and subthreshold swing (SS) for the 80 FETs displayed Gaussian distributions in both parameters, as depicted in Fig. 4e. The average field-effect mobility was measured at 2.36 cm2 V−1 s−1, with the highest value reaching 3.23 cm2 V−1 s−1. Variations in mobility and SS were 19.2% and 10.3%, respectively, similar to those reported in devices fabricated from CVD-grown pristine MoS2 films. This consistency suggests that any performance variations likely stem from the transfer and fabrication processes rather than the inherent material quality. Overall, although the absolute mobility values in our Nb-doped MoS2 are relatively low compared with pristine n-type MoS2 or conventional semiconductors, the device performance ranks among the highest reported for p-type MoS2 transistors, including both monolayer and few-layer devices with various doping methods (Fig. 4f and Supplementary Table 3), highlighting the unique advantage of our approach in achieving stable, efficient, and high-performing doping. This underscores the effectiveness of our MOCVD strategy in combining excellent electronic performance with wafer-scale material uniformity.

Conclusions

In conclusion, we have demonstrated the successful MOCVD growth of wafer-scale Nb-doped MoS2 films, achieving precise control over doping levels and enabling a transition from n-type to p-type behavior. Substitutional incorporation of Nb atoms into the MoS2 lattice led to uniform and controllable doping across the wafer, confirmed by multiple characterization techniques. Additionally, our simulation reveals a clear doping correlation governed by temperature and precursor pressure, providing a theoretical basis for optimizing growth conditions and maximizing substitution efficiency. The films exhibited remarkable electrical properties, positioning Nb-doped MoS2 as one of the most promising p-type variants reported so far. This work addresses a critical challenge in achieving stable, controllable p-type doping in 2D materials, which is essential for the development of complementary logic circuits and the advancement of CMOS technology, paving the way for their integration into next-generation low-power, flexible, and transparent electronics.

Methods

Material synthesis

Our Nb-doped MoS2 films were synthesized using a custom-built MOCVD system, as shown in Supplementary Fig. 1. The growth process, which involved multiple stages with continuous adjustments to temperature and precursor supply, is outlined in Supplementary Fig. 2. Prior to loading into the chamber, the c-plane sapphire substrate was annealed in air for 6 h and spin-coated with a 0.3 mol/L NaCl solution in deionized water. NaCl facilitated growth by lowering the energy barrier41 and reducing nucleation density42. For the MoS2 growth stage, molybdenum hexacarbonyl (Mo(CO)6) and diethyl sulfide ((C2H5)2S) were used as precursors, introduced into the quartz tube by argon at flow rates of 10 and 3 sccm, respectively. The precursors were kept in a bubbler maintained at 25 °C. Small amounts of hydrogen (H2) and oxygen (O2) were also introduced into the growth chamber to eliminate carbon contaminants and ensure the formation of a monolayer through a mild etching effect43. MoS2 growth occurred at 870 °C under atmospheric pressure for 30 min. After MoS2 growth, the precursor and O2 supplies were stopped, and argon was introduced at a high flow rate to purge any remaining Mo and S precursors, thereby preventing excessive MoS2 deposition. Once the temperature was increased to 980 °C, Nb2 (OC2H5)10 was introduced into the chamber with argon as the carrier gas at a flow rate of 3 sccm to initiate Nb doping. The doping level was finely controlled by adjusting the exposure time. During this doping phase, 8 sccm of H2 was continuously supplied to minimize carbon contamination. Following the doping process, the chamber was allowed to cool naturally.

Raman and photoluminescence (PL) spectroscopy

Raman and PL spectra, along with mapping, were obtained using a Renishaw inVia confocal Raman microscope at room temperature. A 532 nm laser was employed with a 3000 g/mm grating for Raman spectral collection and with a 300 g/mm grating for PL spectral collection. The laser spot size was approximately 1 μm, and the laser power was maintained at 1 mW to prevent any potential damage to the samples during the measurements.

Atomic force microscopy (AFM)

AFM measurements were performed using an Asylum Research system to evaluate the surface morphology and roughness of the films. The root-mean-square roughness was computed on representative areas of the wafer from a 5 × 5 µm flattened window using the Asylum Research analysis software.

Scanning transmission electron microscopy (STEM), energy-dispersive X-ray spectroscopy (EDX) and electron energy loss spectroscopy (EELS) measurements

The as-grown samples were transferred onto a silicon nitride TEM grid for STEM analysis. The transfer process involved spin-coating the samples with poly(methyl methacrylate) (PMMA) A2 at 2000 rpm for 60 s, followed by drying on a hot plate at 180 °C for 5 min. After immersion in water for 5 min, the PMMA layer with the samples was detached through water tension and carefully transferred onto the silicon nitride TEM grid. The samples were baked at 65 °C for 10 min to enhance adhesion, and the PMMA was then removed by immersing the samples in acetone overnight. To further clean the surface, the samples were annealed in high vacuum at 250 °C for 6 h.

High-resolution high-angle annular dark field (HAADF) STEM imaging was performed using a double aberration-corrected FEI Titan Themis microscope. The microscope was operated at an accelerating voltage of 80 kV to avoid electron-beam-induced damage. The beam convergence half-angle was set to 20 mrad, corresponding to a collection angle of 49.5–198 mrad. A probe current of ~20 pA was used, and to minimize sample drift, fast scanning was performed at 512 × 512 pixels with a dwell time of 8 µs per pixel.

EDX and EELS analyses were conducted on the same regions as the HAADF-STEM imaging using a double aberration-corrected FEI Titan Themis microscope equipped with a Super-X EDX detector and a Gatan Quantum 965 EELS spectrometer. The measurements were performed at an accelerating voltage of 80 kV with a probe convergence semi-angle of 20 mrad. For EELS acquisition, a collection angle of 36 mrad and an energy dispersion of 0.25 eV per channel were used. The energy resolution, determined from the zero-loss peak, was approximately 0.9 eV. Spectra were acquired in scanning mode with a dwell time of 8 µs per pixel and subsequently processed using Gatan Digital Micrograph software for background subtraction, deconvolution, and elemental quantification.

X-ray photoelectron spectroscopy (XPS) measurements

XPS measurements were performed using a Thermo Fisher Scientific ESCALAB 250Xi system equipped with a monochromatic Al Kα (hν = 1486.6 eV) X-ray source. The base pressure during measurement was maintained below 1 × 10−9 mbar. The pass energy was set to 20 eV for high-resolution scans and 100 eV for survey spectra. All binding energies were calibrated using the C 1 s peak at 284.8 eV as a reference.

Wafer-scale sample transfer

To transfer the as-grown samples from sapphire substrates, samples were spin-coated with PMMA A2 at 2000  rpm for 1 min, then baked at 180 °C for 10 min. The edges of the PMMA layer were gently scratched with tweezers, and the sample was then covered with thermal-release tape and immersed in water for 15 min. This process detached the film from the sapphire substrate, which adhered to the thermal-release tape. After drying the sample with a nitrogen gun, it was transferred onto a Si/SiO2 target substrate, and bubbles between the film and substrate were carefully removed. To enhance adhesion, the sample was baked at 55 °C for 1 h, and the thermal-release tape was removed by heating to 135 °C. Finally, PMMA was removed by immersing the sample in acetone overnight, followed by annealing at 250 °C in high vacuum for 6 h to remove any remaining residues.

Device fabrication

The samples were first transferred onto a Si/SiO2 substrate before proceeding with device fabrication. The desired geometry was defined using laser writer photolithography, with LOR 5A and AZ 1512 HS used as photoresist layers. The samples were etched using a XeF2 chemical etching process. The photoresist layers were cleaned by immersing the samples in dimethyl sulfoxide (DMSO) overnight. Electrode patterns were defined using laser writer photolithography, and Ti/Au (2/80 nm) was deposited using e-beam evaporation to form the electrodes. A final lift-off process in DMSO was used to remove the remaining photoresist. Throughout the fabrication steps, the samples were maintained in an inert Ar atmosphere inside a glovebox to prevent sample degradation from exposure to air.

Transport measurements

Electrical transport measurements were carried out in a closed-cycle cryogen-free probe station. Prior to the measurements, the sample was annealed in high vacuum at 120 °C for 6 h to remove any adsorbates or residues. Two-probe DC electrical measurements were performed using a Keithley 2636B electrometer, with both source-drain voltage and back-gate voltage applied to evaluate the transport properties of the sample.

Thermodynamic-kinetic modeling

To model the Nb substitution process on the MoS2 surface, a thermodynamic-kinetic reaction framework was constructed based on the Langmuir–Hinshelwood mechanism. The model incorporates adsorption, substitution, desorption, and decomposition reactions, with rate constants following the Arrhenius equation. Surface coverage and substitution probability were derived under steady-state assumptions. The gas-phase flux of Nb precursors was calculated using the kinetic theory of gases. Simulations were performed using custom Python scripts. Key parameters, including activation energies and pre-exponential factors, were estimated from the literature or treated as tunable inputs to qualitatively explore trends. The effective substitution probability was computed as a function of growth temperature and Nb partial pressure to identify optimal doping conditions. Detailed equations and parameter definitions are provided in the Supplementary Information.