Introduction

Monolayer molybdenum disulfide (MoS2) is a two-dimensional transition metal dichalcogenide (2D TMD) that features a direct band gap of ~1.8 eV1, excellent charge mobility2, and an ultrathin structure measuring 0.615 nm in thickness3. These characteristics have positioned monolayer MoS2 at the forefront of logic device scaling efforts in research and industry, particularly for use as transistor channels4,5,6. However, the low formation energy of chalcogen (i.e. sulfur) vacancies (Vs) in monolayer MoS2 leads to a relatively high Vs density on the order of 1012–1013 cm−2 7,8,9. For comparison, the formation energy for a single Vs in MoS2 is estimated to be 2.35 eV9, compared to the formation energy of 7.5 eV for a single carbon vacancy (Vc) in graphene10 which has a Vc density on the order of 1010–1011 cm−2 11. The high defect density of monolayer MoS2 impairs the reliability of both its mechanical and electronic characteristics6,12,13, which hinders its applicability. Therefore, the reliability of monolayer MoS2 is a key concern that requires further understanding and focused efforts for improvement.

In our previous work, we employed atomic force microscopy (AFM)-based monotonic loading and long-term static fatigue loading to assess the mechanical reliability of monolayer MoS213. The Weibull modulus reflects the level of scatter in the data and serves as an indicator of mechanical reliability. A higher Weibull modulus demonstrates more consistent mechanical properties and from an engineering standpoint, a high Weibull modulus can be just as critical as high strength. In monotonic loading, samples were loaded monotonically to fracture, and the fracture strength was measured. It was revealed that monolayer MoS2 has a fracture strength Weibull modulus (m) comparable to that of engineered ceramics, m ~ 10–20, which is much lower than the m of metals but notably higher than those of traditional ceramics14,15. Although the use of Weibull statistics for micro- and nanostructures is relatively recent, it has been shown that the Weibull distribution can accurately model the probability of failure, with less than 5% error, for structures containing as few as six representative volume elements, provided m > 216. While understanding the reliability of fracture strength is important, material failure in real-world applications is often driven by fatigue from prolonged stress exposure, and this applies to 2D materials as well17,18,19,20. In static fatigue loading, samples are subjected to a fraction of the average fracture strength, and the load is sustained until failure occurs due to fatigue. Static fatigue testing revealed that monolayer MoS2 has a static fatigue Weibull modulus (M) of ~0.5, similar to that of, again, engineered ceramics such as silicate glass21,22,23. An M less than unity indicates a large spread in lifetime15. For example, structural materials such as steel and glass fiber reinforced polyester composites have M values of 1–10 24,25. During static fatigue testing of monolayer MoS2 it was also observed that ~20% of samples failed prematurely while being loaded to a fraction of the average fracture strength. This low reliability was attributed to the combined effect of defects and thermal fluctuations. This raises a major concern for the industrial adoption of monolayer MoS2. Other fatigue studies of monolayer MoS2 have also shown that defects and lattice reconfigurations are the initial points for brittle, catastrophic failure19 and that the fatigue life decreases sharply when the excitation frequency approaches the monolayer’s natural frequency20.

Tribological characteristics such as scratch and wear resistance are also indicators of the mechanical reliability of layered materials. Another study of ours combined AFM with molecular dynamics simulations to statistically analyze the cyclic wear reliability of graphene, monolayer MoS₂, and monolayer tungsten diselenide (WSe₂)26. It was revealed that graphene exhibited exceptional durability, surviving over 3,000 cycles at 85% of the critical normal load, the threshold load at which wear is initiated, before failure. In contrast, monolayers of MoS₂ and WSe₂ lasted only 500 cycles or fewer. We identified the type and density of defects as key factors contributing to the relatively low cyclic wear reliability of TMDs, consistent with other work that has investigated the mechanical reliability of TMDs18,19. Studies on nanoscale wear have mostly focused on single scratch tests to establish fundamental governing mechanisms of wear27, although there is an increasing number of studies on the cyclic wear reliability of 2D monolayers28. For example, it was shown that as the thickness of fluorinated graphene is decreased its wear reliability increases due to enhanced charge transfer between the substrate and the 2D material promoting adhesion29 and that the that the wear reliability of graphene is substantially lower near its step edges due to peel induced rupturing30. However, there is a lack of systematic studies on the role of defect density and wear resistance in 2D semiconductors such as monolayer MoS2.

Although advancements in growth and synthesis are required to improve material quality, the ultrahigh surface area-to-volume ratio of 2D materials make them ideal candidates for property optimization through precise defect engineering31,32,33,34. Notably, treatment by a nonoxidizing organic superacid bis(trifluoromethane) sulfonimide (TFSI) has been demonstrated to increase the photoluminescence quantum yield (PLQY) of sulfur-based 2D TMDs from less than 1% to near unity (>95%)35. The drastic increase in PLQY is attributed to the Lewis acid nature of TFSI which makes the material more intrinsic by electron counterdoping, in which electrons are withdrawn (e.g., hole doping) from monolayer MoS₂ via surface charge transfer 36. This chemical treatment is a straightforward process with a low thermal budget and has been demonstrated to enhance the performance of sulfur containing, 2D TMD-based optoelectronic37 and electronic38 devices. While this treatment can boost the PLQY of sulfur-based 2D TMDs by means of electron counterdoping, it also reduces the density of sulfur vacancies39 and passivates vacancy sites by hydrogenation40,41 (Fig. 1a). Reducing the density of defects and passivation of defect sites in an atomically thin material can have a substantial impact on its mechanical and electronic characteristics. However, the effects of TFSI treatment on the mechanical properties of MoS2 remain unexplored. Here, we show that this simple defect repair and passivation scheme by treating monolayer MoS2 with TFSI can be leveraged to boost its mechanical reliability and achieve uniform current distribution.

Fig. 1: Young’s modulus, fracture strength, and static fatigue.
Fig. 1: Young’s modulus, fracture strength, and static fatigue.
Full size image

a Schematics of defect healing and passivation of monolayer MoS2 containing a sulfur vacancy. b Optical image of large, exfoliated monolayer MoS2 flakes on pre-patterned Au/Ti coated SiO2/Si substrate containing hundreds of holes. Scale bar, 20 µm. c Atomic force microscope (AFM) topographic image of monolayer MoS2 covering multiple holes. Scale bar, 2 µm. d AFM topographic images of as-exfoliated (left) and treated (right) monolayer MoS2 samples, each covering a hole with a diameter of 2 µm. Scale bar, 500 nm. e Schematic of the loading profiles for AFM-based monotonic and static fatigue loading. Inset shows a scanning electron microscope image of a focused ion beam (FIB)-modified single crystal diamond tip with a well-defined radius which was used to indent suspended monolayer membranes. Inset scale bar, 200 nm. f, g 2D Young’s modulus (E2D) and 2D fracture strength (\({\sigma }_{\max }^{2D}\)), respectively, of as-exfoliated and treated MoS2 monolayers. The central line in each box plot represents the median. The top and bottom edges of the box show the upper and lower quartiles, while the whiskers extend to the maximum and minimum non-outlier values. 30 samples each for the two monolayer MoS2 conditions were tested. h Weibull plots of failure probability of as-exfoliated and treated MoS2 monolayers under static fatigue loading at 85% of their respective \({\sigma }_{\max }^{2D}\). t is time in seconds, lines are linear fit with 95% confidence interval, and 50 samples each for the two monolayer MoS2 conditions were tested. The static fatigue Weibull modulus (M) for each monolayer MoS2 condition is shown. The gray shaded areas represent the regime of premature failure, defined as failure occurring within ~3 s of loading to a fraction of \({\sigma }_{\max }^{2D}\) (left shaded area), and the runout regime, where no failure occurred for at least 5000 s (1.39 h) (right shaded area).

Results

Static fatigue reliability

Monolayer MoS2 samples were prepared by mechanical exfoliation following the deposition of a 2.5 nm Ti adhesion layer and a 7.5 nm Au layer on SiO2/Si substrates containing hundreds of patterned holes (Fig. 1b-d)42. The single layer structure of the monolayers was verified using Raman spectroscopy and AFM-based thickness characterization (see Supplementary Note 1). AFM-based nanomechanical characterization was performed on suspended membranes of monolayer MoS2. The availability of hundreds of suspended membranes is key to obtaining statistically significant data. Individual monolayer MoS2 membranes were monotonically loaded to failure by using single crystal diamond tips which were modified into a spherical form (Fig. 1e). Further details on tip modification are given in Methods and additional images of the modified tips are shown in Supplementary Note 2. We determine the Young’s modulus and fracture strength from the load-deflection data (see Supplementary Note 3)43,44. Nanomechanical characterization was performed for both as-exfoliated monolayer MoS2 and TFSI-treated monolayer MoS2, with the latter referred to as treated monolayer MoS2 herein. The effectiveness of the superacid treatment was confirmed using PL spectroscopy, which, as expected, shows an increase of greater than two orders of magnitude (~120x) in PL intensity (see Supplementary Note 4). In addition, XPS measurements confirmed an increase in the S to Mo stoichiometric ratio in the treated samples (see Supplementary Note 5), consistent with previous reports35,38.

Monotonic loading tests revealed that the pristine monolayers exhibit a 2D Young’s modulus (E2D) of 200 ± 15 N m⁻¹ and a 2D fracture strength (\({\sigma }_{\max }^{2D}\)) of 18.4 ± 1.5 N m−1 (Fig. 1f, g), corresponding to the higher end of previously reported values13,45,46. The treated monolayers demonstrate enhanced mechanical properties, with an E2D of 228 ± 16 N m−1 and a \({\sigma }_{\max }^{2D}\) of 23.8 ± 1.5 N m−1 (Fig. 1f, g). This increase in E2D and \({\sigma }_{\max }^{2D}\) is attributed to a reduction in defect density resulting from the repair of sulfur vacancies. In terms of reliability of the fracture strength, the fracture strength Weibull modulus, m, increased from 14.6 ± 1.2 to 18.6 ± 0.8 while remaining in the range reported for engineered ceramics (see Supplementary Note 6). While improvements in E2D, \({\sigma }_{\max }^{2D}\), and m are promising, they are not particularly impactful from a technological standpoint and do not address reliability concerns. Importantly, fatigue testing reveals a significant improvement in static fatigue, a key long-term time-dependent failure behavior parameter, following the treatment. Static fatigue tests were conducted by AFM-based loading of monolayer MoS2 to 85% of its \({\sigma }_{\max }^{2D}\) and maintaining this load until fracture occurs (Fig. 1e). The \({\sigma }_{\max }^{2D}\) used in static fatigue testing of individual suspended monolayer membranes represents the average fracture strength of three neighbouring membranes within the same monolayer flake. 85% of \({\sigma }_{\max }^{2D}\) is chosen because failure probability at this load is relatively high and cases of premature failure and survival are still observed for monolayer MoS213. Weibull statistical analysis of static fatigue data reveals that the repair and passivation of sulfur defects drastically improve the mechanical reliability of monolayer MoS2. Failure probability under static fatigue is presented in (Fig. 1h) in the form a Weibull plot for time-dependant loading described by:

$$P\left(t\right)=1-\exp \left[-{\left(\frac{t}{{t}_{0}}\right)}^{M}\right]$$
(1)

Where t is the failure time and the fitting parameters t0 and M are the characteristic lifetime and Weibull modulus, respectively, for static fatigue. Notably, the M value for treated MoS2 is 0.81 ± 0.04, which is near unity and twice that of the as-exfoliated MoS2 at 0.39 ± 0.01. In addition, treated samples do not exhibit premature failure which is defined as failure occurring within ~3 seconds of being loaded to a fraction of \({\sigma }_{\max }^{2D}\) or failure during the ramp to the fraction of \({\sigma }_{\max }^{2D}\) (see Supplementary Note 7). Furthermore, test runouts in which failure did not occur for at least 5000 s (1.39 h) were observed at a higher frequency for treated monolayers. Together, the observed increases in m and M underscore the critical role of defect repair passivation in enabling the long-term mechanical reliability required for practical deployment of 2D semiconductor-based devices.

Wear reliability

Tribological experiments demonstrated that treated monolayer MoS2 exhibits enhanced scratch resistance and cyclic wear reliability. Contact-mode AFM with a diamond tip was employed to identify as-exfoliated and treated monolayer MoS₂ samples on Au/Ti/SiO₂/Si by characterizing their thickness. Scratch tests were conducted using diamond tips that were preconditioned by rubbing them against a bare SiO₂ substrate until the friction-load curves stabilized, indicating that the tips had become rounder and blunter (see Methods). Then, the blunt diamond tip was slid across monolayer MoS₂ on a flat SiO2 surface for a distance of 2 µm, while the normal force (FN) was gradually increased from 0 to a maximum of 4 µN (see Supplementary Note 8). 19 scratch tests were performed on both as-exfoliated and treated monolayer MoS₂ samples. In Fig. 2a the relationship between the lateral force and the FN applied by the AFM probe during a scratch test is shown. A sharp change in the slope indicates a rise in friction due to the probe contacting the underlying SiO₂, marking the onset of wear failure (see Supplementary Note 8). Figure 2b presents the determined critical normal force (FCN) needed to induce wear failure in a single pass for both as-exfoliated and treated monolayer MoS₂. On average, the FCN increased by 60%, from 1.4 ± 0.5 µN to 2.2 ± 0.6 µN, following TFSI treatment. Figure 2a shows a representative plot of lateral versus normal force corresponding to the average FCN of as-exfoliated and treated monolayer MoS₂. Lateral versus normal force plots from 19 individual scratch tests for both as-exfoliated and treated monolayer MoS₂, whose FCN values are shown in Fig. 2b, are provided in Supplementary Fig. 8. Figure 2c displays post-mortem friction maps of four scratch tests each from as-exfoliated and treated monolayer MoS₂ samples where the AFM probe was slid from top to bottom as the normal load was increased. It should also be noted that the scratch test data for pristine and treated samples do exhibit some overlap, which could be attributed to the non-uniform effects of the treatment or slight discrepancies in the contact area. It is important to note that FCN is highly dependent on contact pressure and the orientation of the flake relative to the AFM probe. These factors introduce variability, making the precise quantification of FCN particularly challenging26. Although this enhancement in scratch resistance appears noteworthy, treated MoS₂ still falls short compared to graphene, which can resist a ~ 2x higher FCN 26.

Fig. 2: Scratch and cyclic wear testing.
Fig. 2: Scratch and cyclic wear testing.
Full size image

a Normal versus lateral force plot of scratch tests of as-exfoliated and treated monolayer MoS2. b Box plot of the critical normal force (FCN) for scratch testing of as-exfoliated and treated monolayer MoS2. The central line in each box plot represents the median. The top and bottom edges of the box show the upper and lower quartiles, while the whiskers extend to the maximum and minimum non-outlier values. 19 samples each for the two monolayer MoS2 conditions were tested. c Atomic force microscopy (AFM) friction maps of as-exfoliated (left) and treated (right) monolayer MoS2 showing post-mortem scratch tests. Scale bars, 500 nm. d Averaged lateral force on the middle of a wear track versus the cycle number during a cyclic wear test on treated monolayer MoS2. In Stage 1 the lateral force is gradually increased by gradually increasing the normal load until it reaches 90% of FCN. In Stage 2, the lateral force is kept stable until, finally, a sudden and drastic jump in lateral force indicates wear failure. Inset shows a schematic of cyclic wear testing. e Normalized normal force (FN) versus number of cycles to failure for as-exfoliated and treated monolayer MoS2. 16 scratch tests and 20 cyclic wear tests are plotted for each monolayer MoS2 condition.

While the treatment slightly improved the scratch resistance of monolayer MoS₂, cyclic wear tests demonstrated a substantial improvement in wear reliability. The cyclic wear reliability of monolayer MoS₂ was assessed at 80% and 90% of the respective FCN for both as-exfoliated and treated samples. Figure 2d presents the lateral force at the center of the wear track as a function of cycle number from a single cyclic wear test. In stage 1 (Fig. 2d), the normal load is gradually increased to a fraction of the respective FCN, followed by a steady friction state in stage 2. The test is concluded by the moment of failure which occurs when the AFM probe contacts the substrate, resulting in a sudden spike in the friction signal. The results of 16 scratch tests and 20 cyclic wear tests each for as-exfoliated and treated monolayer MoS2 are shown in Fig. 2e, where the normalized force is plotted against the number of cycles to failure for both as-exfoliated and treated monolayer MoS₂. The results of the scratch tests are averaged and presented as a single column at cycle 1. At 90% FCN, the as-exfoliated MoS₂ exhibited occasional premature failures, where failure happened within the first 20 cycles, and typically failed within 500 cycles. However, after treatment with TFSI, its cyclic wear life improved significantly, lasting up to 5000 cycles. A similar trend was observed at 80% FCN, where the treated monolayer MoS₂ endured up to 21,000 cycles before failure, reaching wear reliability of graphene26. Importantly, no premature failures were observed in the treated MoS₂ under both normal force conditions. AFM images of as-exfoliated and treated MoS2 after cyclic wear tests are shown in Supplementary Note 9.

Simulations of failure mechanisms

Though a lower density of defects could result in improved mechanics, treated monolayer MoS2 still contains sulfur vacancies as it has been shown that TFSI treatment can reduce the defect density of monolayer MoS2 from ~1013 cm−2 to ~1012 cm−2 39. Thus, it is particularly surprising that fatigue and wear premature failure cases which are observed in pristine monolayers are not observed in treated ones. This points to a possible change in the fracture mechanisms. It has been experimentally shown that as result of TFSI treatment S-H bonds form at some sulfur vacancy sites of monolayer MoS241. DFT simulations show that the fracture strength increases slightly with a reduction in the density of defects while hydrogenation of a vacancy site does not change the fracture strength (see Supplementary Table 4). To gain mechanistic insight into how passivation of vacancy sites via hydrogenation affects the static fatigue behavior of defective MoS2, we performed ab initio molecular dynamics (AIMD) simulations of an MoS2 system containing a single sulfur vacancy and one in which the sulfur vacancy is hydrogenated, both at a constant biaxial strain of 13%, a fraction (~85%) of the failure biaxial strain of monolayer MoS2 which is estimated to be 15–16%47. This approach resembles the static fatigue loading at 85% of fracture strength which was performed experimentally. In the AIMD simulations where fracture occurred, a change in the fracture mechanism was observed following passivation (Fig. 3a–c). In the single vacancy system, fracture always begins with the breakage of the third bond counting from the defect, with crack propagation happening ~40 fs after that (Fig. 3d). Before any bond breakage, the central S atom (i) maintains a largely planar geometry due to biaxial strain. After the initial bond breakage, an adjacent S atom (ii) also adopts a planar geometry. This state resembles the strained configuration of two vacancies located side by side, which is expected to have low stability. In the hydrogenated system the presence of a second fracture mechanism is observed: breakage of the first bond counting from the defect, with initiation of crack propagation taking at least one order of magnitude longer (Fig. 3c, d).

Fig. 3: Fracture mechanism simulations.
Fig. 3: Fracture mechanism simulations.
Full size image

a–c Schematics of the observed fracture nucleation mechanisms. a shows top-down and side views of model of monolayer MoS2 containing a single sulfur vacancy which is circled in red. The top-down view of the model after first failure event occurring at bond 3 (b) and after failure event occurring at bond 1 (c). Breakage at bond 3 happens between atom iii and the Mo atom below it in the figure, leaving atom ii in a mostly planar geometry, similar to that of atom i in the single vacancy system. d Fracture process of monolayer MoS2 with a sulfur vacancy (Vs) and monolayer MoS2 with a hydrogenated sulfur vacancy (Vs, hydrogenated) in independent ab initio molecular dynamics (AIMD) runs. e Nudged elastic band (NEB) calculations of energy barrier (Ea) and energy variation (ΔE) of monolayer MoS2 systems under biaxial strain-induced fracture nucleation. The dotted lines are polynomial fits. Inset in energy barrier shows an example NEB calculation.

To demonstrate that this shift in fracture mechanism is rooted in physical-chemistry rather than the stochasticity of the AIMD simulations, we also performed DFT-based nudged elastic band (NEB) calculations for both bond fracture scenarios (Fig. 3e). In the single vacancy system, breakage of bond 3 has a lower energy barrier (Ea) than breakage of bond 1 across a wide range of applied biaxial strain, while the opposite is true for the hydrogenated system. Furthermore, the energy variation (ΔE) is more negative after the breakage of bond 3 for the single vacancy system, and of bond 1 for the hydrogenated system. In addition, fracture at bond 1 for the hydrogenated system is the most stabilizing process across all four scenarios, explaining why the hydrogenated system takes longer to fail after the initial breakage of bond 1. Further, the physical-chemistry behind this shift in mechanism can also be understood by analyzing the strength of bond 1 in each system which can be quantified by its Integrated Crystal Orbital Hamilton Population (ICOHP) value (see Supplementary Table 5). The presence of a vacancy lowers the ICOHP of bond 1, making it stronger when compared to the pristine material, as the subcoordinated Mo atoms near the vacancy make stronger bonds with the remaining S atoms. The same trend is maintained at 13% biaxial strain. However, when the vacancy is hydrogenated, the ICOHP increases under applied strain, as the Mo atoms are no longer subcoordinated. This leads to the weakening of bond 1, leading to its preferential breakage in the hydrogenated system. These results suggest that hydrogenation induces a shift in failure mechanism.

DFT simulations also provide insight into the failure mechanisms associated with scratch and cyclic wear testing. In the case of monolayer MoS2 with two sulfur vacancies, the vacancy sites are assumed to be symmetrical with respect to the basal plane, as this configuration represents the most stable arrangement for two closely spaced sulfur vacancies in MoS₂ (see Supplementary Fig. 12). However, upon relaxation of the system, the single and double vacancy sites induce area contractions of −0.6% and −1.1%, respectively, indicating tension near the vacancies. These tensile sites can act as weak points that facilitate the initiation of wear. In contrast, the hydrogenated system experiences a slight area expansion of +0.3%, suggesting mild compression near vacancy sites that can impede the onset of wear. It has also been shown that during AFM-based wear testing of 2D materials there is a stress concentration on the side of the AFM probe sliding on the 2D layer26. This results in asymmetric loading conditions, where sulfur atoms on the top layer undergo tension while bottom-layer atoms are compressed. Healing a sulfur vacancy significantly reduces the stress concentrations, which in turn can enhance wear resistance.

Atomic-resolution current

The reliability of MoS2 based-devices is also closely linked to their electronic characteristics12, which are directly altered by TFSI treatment. For example, TFSI-enabled sulfur vacancy healing has been used to modulate the work function of monolayer MoS2 for improved band alignment at metal-semiconductor interfaces38 and to reduce electron doping induced during fabrication of 2D materials-based electronic devices48. To gain understanding of the impact of passivation and defect healing on electronic characteristics and of MoS2, we performed atomic-resolution conductive atomic force microscopy (C-AFM) experiments.

Recently, it has been demonstrated that C-AFM can be employed to obtain true atomic-resolution imaging of surfaces by recording out-of-plane current under ambient conditions49. This is enabled by an atomically sharp conductive pathway and high-speed imaging. When the tip scans the sample surface with a light load and under the application of a bias voltage, a likely scenario involves portions of the insulating absorbate layer at the tip-sample interface breaking down to leave an angstrom-scale electrically conductive pathway that can detect individual atoms, even though a larger area of the tip is still in contact with the sample50. Meanwhile, imaging at high speeds (corresponding to ∼100 nm/s and higher) enables maintaining the atomic sharpness of the electrical contact by containing the time spent by the AFM tip on a particular pixel below 1 ms, the relevant regime for atomic-scale aging processes at nanoscale contacts51. 5 × 5 nm2, out-of-plane, current maps of as-exfoliated and treated MoS2 at a bias voltage (Vbias) of 1 V are shown in Fig. 4a, b along with their corresponding Fourier transforms. The histograms of the current at Vbias = 1 V of as-exfoliated and treated MoS2 is shown in Fig. 4c. The current distribution in treated samples is drastically more uniform as indicated by a tighter distribution of current in the histograms which demonstrates more homogeneous atomic-scale current distribution under an applied bias. Figure 4d presents the standard deviation of current measurements for both as-exfoliated and treated MoS₂ which are calculated to be 4.1 ± 0.9 nA and 0.6 ± 0.4 nA, respectively. These values were calculated from 5 × 5 nm² current maps, with four images acquired at each applied bias voltage (Vbias = −0.5 V, 0.7 V, and 1 V) for both monolayer MoS₂ conditions. The corresponding current maps and their histograms are provided in Supplementary Note 11. This phenomenon can be further understood through DFT calculations in which we modeled four MoS₂ systems: pristine monolayer MoS2, monolayer MoS2 with a single sulfur vacancy, monolayer MoS2 with two sulfur vacancies, and monolayer MoS2 in which a sulfur vacancy is hydrogenated (Fig. 4e–h). Further details of simulations are provided in Supplementary Note 10. DFT calculations estimate a band gap of ~1.6 eV for pristine MoS2 (Fig. 4i). However, the presence of single or two sulfur vacancies introduces unoccupied mid-gap states (Fig. 4j, k). Importantly, these midgap states do not appear in the hydrogenated system (Fig. 4l). The midgap states are attributed to the shifted electronic density (Fig. 4m–p) of the subcoordinated Mo atoms. The electronic density map of the hydrogenated system (Fig. 4p) closely resembles that of pristine MoS2 (Fig. 4m), reinforcing the role of hydrogen passivation in restoring electronic uniformity. Non-uniform current distribution is a key factor driving long-term performance degradation in devices such as batteries and fuel cells52,53. Therefore, achieving uniform current distribution could significantly enhance the reliability and commercial viability of monolayer MoS2-based electronic devices. This underscores the need for more systematic research on this topic.

Fig. 4: Atomic-resolution imaging via conductive atomic force microscopy (C-AFM) and density functional theory (DFT) calculations of electronic structure.
Fig. 4: Atomic-resolution imaging via conductive atomic force microscopy (C-AFM) and density functional theory (DFT) calculations of electronic structure.
Full size image

a–c 5 × 5 nm2 current images of as-exfoliated (a) and treated (b) MoS₂ samples, and their corresponding Fourier transforms (blue circles highlight bright spots which confirm the hexagonal lattice structure) and (c) their histograms of current, all acquired at a bias voltage (Vbias) of 1 V. Scale bars, Fourier transforms in (a, b), 2.5 nm−1. d Standard deviation of current of as-exfoliated and treated MoS2. The standard deviation values were calculated from 5 × 5 nm² current maps (See Supplementary Note 11 for the current maps in which the standard deviation values were extracted). Twelve images were analyzed for each of the two monolayer MoS₂ conditions, with four images acquired at each bias voltage: Vbias = –0.5 V, 0.7 V, and 1 V. The central line in each box plot represents the median. The top and bottom edges of the box show the upper and lower quartiles, while the whiskers extend to the maximum and minimum non-outlier values. e–p DFT models (e–h), density of states (DOS) as predicted by DFT (i–l), and electronic density maps as predicted by DFT (m–p), corresponding to pristine monolayer MoS2 (e, i, and m), monolayer MoS2 containing a single S vacancy (f, j, and n), monolayer MoS2 containing a double S vacancy (g, k, and o), and monolayer MoS2 with a hydrogenated vacancy (h, l, and p). Electronic density maps are shown on the Mo plane, with red representing higher density and blue representing lower density.

Fourier transforms of atomic-resolution current images were used to determine the lattice periodicity of MoS₂. In each Fourier transform image, the distances between bright spots in reciprocal space, highlighted by blue circles in the Fourier transform images (Fig. 4a, b), were measured to extract this periodicity. The average periodicity was found to be 3.06 ± 0.10 Å for as-exfoliated MoS₂ and 3.10 ± 0.07 Å for the treated MoS₂ (see Supplementary Note 12). These values are very close to the expected lattice constant (a) of ~3.15 Å3,54,55. This periodicity corresponds to the average spacing between the bright features in the current images, and confirms the hexagonal lattice structure. Because the Mo–Mo and S–S distances in MoS₂ are nearly identical, it is not possible to definitively assign these bright features to either Mo or S atoms. However, they are likely due to sulfur atoms, which reside on the surface and are more directly probed by the tip.

C-AFM enables the detection of both point defects, such as individual missing atoms (e.g., sulfur vacancies), and extended defects, which appear as broader regions of altered conductivity spanning several nanometers (see Supplementary Fig. 21). Extended defects, potentially caused by charge accumulation from donor atoms56, are commonly observed in as-exfoliated MoS₂ but are rarely seen in treated samples. These defects can reach lateral dimensions of approximately 5 nm and are likely contributors to premature failure in static fatigue and wear testing of as-exfoliated samples. Point defects were observed in both as-exfoliated and treated samples, though they were more prevalent in the former. Since these defects consistently appeared in the same locations across multiple scans, surface contamination effects can be ruled out. Additional confirmation that these features represent intrinsic crystal defects comes from differential conductance spectroscopy (dI/dV vs. V), which revealed gap states in the band gap, typically associated with sulfur vacancies on MoS₂ surfaces (see Supplementary Fig. 22). In addition, given that the dominant defect type in MoS2 are S vacancies and the frequent appearance of missing bright spots in atomic-resolution C-AFM images further supports the assignment of the bright features to sulfur atoms.

Discussion

We have reported that the passivation and repair of sulfur vacancies in monolayer MoS2 via TFSI treatment can be utilized to boost its mechanical reliability and obtain uniform atomic-scale current distribution. The technique improves the static fatigue reliability and cyclic wear reliability of monolayer MoS2 by ~2x and ~10x, respectively. It also prevents cases of premature failure which were observed during the static fatigue and cyclic wear testing of as-exfoliated monolayer MoS2 samples. Simulations suggest that passivation and defect healing alter the failure mechanism, delaying failure propagation and reducing stress at vacancy sites. In addition, the elimination midgap states upon passivation and defect healing explain the uniform out-of-plane current distribution in treated MoS2. The results of this study will benefit the development of 2D semiconductor-based devices by demonstrating a quick and process-compatible strategy to enhance their reliability. We also emphasize the need for a deeper understanding of the reliability of 2D semiconductors and 2D semiconductor-based devices, and call for additional comprehensive studies on the subject.

Methods

Monolayer MoS2 sample preparation

Holes of 2 µm in diameter and ~1.5 μm in depth were fabricated on Si wafer with a 280 nm SiO2 layer using electron beam lithography (Raith EBPG 5000+, resist: AR-P6200-13) patterning and reactive ion etching (Oxford PlasmaPro 100 Cobra ICP-RIE). Monolayer MoS2 samples were prepared by mechanical exfoliation following the electron beam evaporation (Angstrom Engineering Nexdep Electron Beam Evaporator) of a 2.5 nm Ti adhesion layer and a 7.5 nm Au layer on SiO2/Si substrates containing hundreds of patterned holes. Deposition of the Ti and Au layers were performed at a rate of 0.2 Å s−1. Mechanical exfoliation was performed within 5 min after deposition by using a polyester base adhesive tape (Nitto Denko) and bulk MoS2 crystals (synthetic crystal flux zone growth MoS2, 2D semiconductors). Monolayers were also mechanically exfoliated on SiO2/Si. Oxygen plasma treatment at 100 W for 5 min was performed on SiO2 before exfoliation to increase the yield of monolayers. Monolayer flakes were identified using an optical microscope (Zeiss Scope A1) and characterized via Raman spectroscopy and AFM topographical imaging.

Raman and PL spectroscopy

Raman and PL spectroscopy measurements (Renishaw inVia Confocal Raman Micro-spectrophotometer) of MoS2 on SiO2/Si were carried out at a laser wavelength of 532 nm, 1800 I mm−1 grating, 20x objective, and spot size ~ 1 µm. The laser power was kept below 1 mW to avoid local heating induced by the laser.

XPS

Measurements were performed on a ThermoFisher Scientific EscaLab 250Xi instrument. The X-ray source was monochromatized Al K-Alpha with a 650 µm spot size. Photoelectron spectra were collected at an electron emission angle of 90° relative to the surface plane. A low-energy electron/ion (Ar) flood source was used for charge neutralization. Survey spectra were acquired at a pass energy of 100 eV. The corresponding point density on the energy axis was 1 eV/step. These scans were performed in order to identify all of the species detectable on the surface. Regional scans were performed at high energy resolution (lower pass energy (5 eV), and with correspondingly higher point density on the energy axis, 0.1 eV between points. The dwell time for the acquisition of these spectra was 50 ms. Measurements were optimized to minimize analysis time and x-ray exposure. Spectra were collected across as-exfoliated and treated MoS2 flakes on SiO2 for binding energies at the peak of interest. For XPS peak analysis and deconvolution, the software Advantage was employed. The S/Mo ratios were determined from the integrated areas of the S 2p and Mo 3 d peaks factored by their corresponding relative sensitivity factors.

AFM tip modification

AFM tips used in the monotonic loading and static fatigue testing were modified. The original tip apex (K-TEK nanotechnology, D300 tips), with a sharpness of 5–10 nm radius as specified by the manufacturer, introduces uncertainties in measuring the tip shape and effective radius by using SEM. These uncertainties can lead to considerable errors in nanomechanical measurements. To address this, we used a FIB system (Hitachi NB5000) equipped with Ga+ ions to modify the tips, reshaping them into a spherical form with a larger radius. The milling process was carried out sequentially from the lower portion of the tip to the apex, resulting in a conical profile and a final apex radius of ~100 nm. Initially, an accelerating voltage of 40 kV and a probe current of 20 pA were applied, followed by a lower accelerating voltage of 10 kV and a probe current of 10 pA for final adjustments.

SEM

An SU7000 microscope (Hitachi Ltd.) with an accelerating voltage of 1.5–5 kV was used to image AFM tips.

AFM

Topographical imaging, monotonic quasi-static loading, static fatigue testing, friction mapping, scratch testing, and cyclic wear testing were all conducted using an Asylum Cypher atomic force microscope (Oxford Instruments) with single crystal diamond tips to mitigate tip wear during nanomechanical experiments. K-TEK nanotechnology, D300 tips were used for monotonic loading, static fatigue, scratch testing, and cyclic wear experiments. K-TEK Nanotechnology, USA D10 tips were used for friction mapping. All AFM measurements were conducted at room temperature of 20–25 °C and the relative humidity of ~30%.

Monotonic quasi-static loading in which Young’s modulus and fracture strength values were extracted from was performed by initially applying a small force to the samples (e.g., 50 nN or 100 nN), followed by displacement-controlled deflection at a loading rate of 20 nm s−1 for 400 nm. Unlike the force-controlled loading method, this approach ensures that only the fracture force required for each sample is applied, preventing the AFM tip from making contact with the substrate after the sample fractures, thereby better preserving the tip. The spring constant of each cantilever was calibrated by averaging spring constant values obtained the Sader and thermal methods. FIB-modified tips were used in monotonic loading experiments.

Static fatigue testing was carried out by applying a user-defined force, lower than the fracture force, to the center of the hole, with the force maintained through feedback loop control. Failure, if it occurred, was signaled by large (>1 μm) and abrupt changes in the tip height and deflection signals. The spring constant of each cantilever was calibrated by averaging spring constant values obtained the Sader and thermal methods. FIB-modified tips were used in static fatigue experiments.

Friction force microscopy (FFM) mapping was performed using the normal and lateral signals of equipped with a sharp diamond-tipped cantilever. The cantilever spring constants were calibrated using the normal57 and lateral58 Sader methods, as well as lateral sensitivity calibration based on the test probe method59. The normal force was carefully selected to ensure that no detectable wear occurred on either the cantilever tip or the 2D materials. Topographic images were acquired simultaneously with friction force images during FFM operation.

Scratch tests were conducted using single crystal diamond tips with an initial tip manufacturer-defined tip radius of 5–10 nm according to manufacturer specifications. The cantilever spring constants were calibrated using the normal57 and lateral58 Sader methods, as well as lateral sensitivity calibration based on the test probe method59. To minimize uncertainties that may be introduced due to variations in the shape of the tip, the tips were preconditioned by rubbing them against a bare SiO₂ substrate until the friction-load curves stabilized, indicating that the tips had become rounder and blunter. Intermittent adhesion tests on the SiO₂ substrate were conducted to verify consistent contact quality and to ensure the tip bluntness was maintained after each test. To convert the lateral voltage signal to lateral force, the trace and retrace voltage signals were multiplied by lateral sensitivity, torsional stiffness, and 1 h−2, where h is the height of the AFM tip plus half of the thickness of the cantilever. To determine the critical normal force, the probe is brought to contact and scans a 2 µm single line in lithography mode at a constant speed of 500 nm s−1 and 90° scan angle while the normal force is continuously increased from 0 to 4 µN.

Cyclic wear experiments were conducted using single crystal diamond tips with an initial tip manufacturer defined tip radius of 5–10 nm according to manufacturer specifications. The cantilever spring constants were calibrated using the normal57 and lateral58 Sader methods, as well as lateral sensitivity calibration based on the test probe method59. To minimize uncertainties that may be introduced due to variations in the shape of the tip, the tips were preconditioned by rubbing them against a bare SiO₂ substrate until the friction-load curves stabilized, indicating that the tips had become rounder and blunter. Intermittent adhesion tests on the SiO₂ substrate were conducted to verify consistent contact quality and to ensure the tip bluntness was maintained after each test. A tip was brought into contact and a single 500 nm line was scanned in reciprocating motion with a speed of 5 µm s−1. Initially, normal load is set to 0 N and then gradually increased until a desired load is obtained within the first 5–10 cycles. Counting cycles to failure begins when the desired normal load is reached, and continues until the friction signal spikes which indicates the failure of the monolayer flake as it reaches the friction level of the substrate.

An Asylum Research, Cypher VRS was used for the conductive atomic force microscopy (C-AFM) imaging. C-AFM was carried out in ambient conditions (temperature near the sample between 30 and 35 °C, relative humidity in the room between 30% and 45%). The detectable current range of our setup is ±20 nA. Doped-diamond-coated (Nanosensors, CDT CONTR and CDT-NCHR) and Ti/Ir-coated (Asylum Research, ASYELEC.01-R2) conductive AFM probes were utilized for imaging. Atomic-resolution imaging was typically performed on areas of 30 × 30 nm2 and below, at a high scanning rate of 15.62 Hz with 256 scan lines. The bias voltage was kept between −0.5 V and 1 V. The interaction force between the AFM probe and the sample was kept constant at the snap-in level (corresponding to several nN), and no additional normal force was applied during the measurements.

Weibull statistical analysis

The Weibull plot for static fatigue was constructed by the following steps after obtaining the failure time for static fatigue of an X number of samples13,60,61:

  1. (1)

    Rank the failure times in ascending order and assign a running index to each.

  2. (2)

    Calculate the cumulative failure probability using the formula \(P=\frac{(i-0.5)}{X}\), where i ranges from 1 to X.

  3. (3)

    Plot the data in the form of ln[-ln(1-P)] versus ln(σ) or (vs lnt) and fit the resulting points with a straight line. The slope of this line represents the Weibull modulus m or M.

Computational methodology

Density Functional Theory (DFT) calculations were performed on VASP62 with a plane-wave basis set. The GGA/PBE exchange-correlation functional and standard PAW pseudopotentials were employed. An energy cutoff of 550 eV was used. The thresholds for energy and forces were 10−5 eV and 10−2 eV Å−1, respectively, during ionic and cell relaxation. A vacuum of 20 Å was included perpendicularly to the 2D material in order to avoid spurious interactions across the periodic boundary conditions. A rectangular supercell was used to model the point defects, as shown in the Supplementary Material, with a k-point mesh of 2 × 2 × 1. Stress-strain curves were generated by changing the dimensions of the supercell and allowing the atomic positions to relax, as typically done in the literature63. When van der Waals interactions were included, they were modelled according to Grimme’s D3 correction as implemented on VASP. Formation energies were calculated with respect to the most stable elemental forms of the elements present in each system. Hydrogenation energies were calculated similarly, with respect to H2 gas. Crystal Orbital Hamilton Population (COHP) calculations were performed with LOBSTER64, and Bader charge analysis was done with VASPKIT65. Ab initio molecular dynamics (AIMD) simulations were performed within an NVT ensemble by using the Nosé-Hoover thermostat with a time step of 2 fs. Energy barriers were determined from Nudged Elastic Band (NEB) calculations. NEB calculations were performed with a climbing image (CI) and a total of six images unless shown otherwise.