Introduction

The ability to spatially modulate properties of semiconductors has been a key factor in the technological success of this class of materials. The prime examples include complementary metal oxide semiconductor (CMOS) technology that relies on local donor and acceptor doping and semiconductor active optoelectronics, which relies on changing material composition along the electrical current flow. To realize the technological potential of two-dimensional semiconductors, a reliable, simple and scalable method to modulate their electronic properties and form homo- or heterojunctions is desired1.

Monolayer (ML) transition metal dichalcogenide (TMD) p–n junctions can be induced by electrostatic gating2,3, however very strong electric fields are necessary in order to invert the conduction type of the material. A double layer forming at the interface of the thin films of TMDs with ionic gels or liquids has been shown to generate sufficiently strong fields, yet spatial modulation of the carrier concentration has not been possible at short distances4,5. In addition, multiple gates or patterned substrates needed to realize such devices restrict the range of possible device designs. Doping with molecules adsorbed on MLs from gases or solvents has also been demonstrated6,7,8, but weak forces between adsorbants and the ML TMDs render them unstable and difficult to control especially if spatially selective doping is to be achieved9. Substitutional doping of TMDs during crystal growth has already been demonstrated10,11,12, but only some methods of post-growth substitutional doping that have been reported offer spatial modulations. For example doping by plasma or laser assisted chemical doping has been used to replace chalcogen atoms in TMDs13,14.

Since the band gap of ML TMDs depends on their electrostatic environment15,16, it has been proposed that substrate engineering may be potentially used to fabricate homojunction devices. Heterojunctions of ML TMDs have been fabricated by changing of reagents during the CVD growth17,18,19,20. These as-grown heterojunctions have geometries given by the crystal growth, which may limit their functionalities. While selective seeding may overcome this issue, size control still remains an issue21,22.

Ion implantation is one of the commonly used doping techniques in standard semiconductor processing23. Compared with chemical doping13,14 it has the advantage of being versatile; almost any element can be introduced in the lattice of the target material. The method does not introduce contamination; it relies on physical processes and is performed in vacuum. Local doping can be introduced using lithographically defined masks or focused ion beams. Ion implantation into bulk materials requires ion energies of at least several keV, which increase with the desired depth of the implantation. Several successful attempts have been made to create defects in TMD MLs using ions with energies typical for bulk implantation24,25,26,27,28,29,30,31, but it is intuitively clear that lower energy of ions should be more suitable for doping atomically thin targets. The doping of graphene with boron and nitrogen by implantation with ions at energies below 20 eV has been confirmed by a detailed analysis of electron energy loss spectra (EELS) acquired with atomic resolution in the scanning transmission electron microscope (STEM)32 while dark-field STEM analysis of MoS2 implanted with selenium (Se) at 10 eV indicated incorporation of Se into sulfur (S) sites33. However, the mechanism of implantation is not fully understood, especially with account for the differences in the response of low-dimensional materials to ion bombardment as compared to those in bulk systems34,35.

Here, we study ion implantation of 80Se+ into ML MoS2 at an energy of 20 eV. Since Se atoms are isoelectronic with S, substitutional Se atoms are chemically inactive and thus stable, making this material system an ideal model to study the implantation process. The choice of implantation conditions has been guided by analytical potential (AP) and density-functional theory (DFT) molecular dynamics (MD) simulations. The implanted MLs have been characterized using Raman and photoluminescence (PL) spectroscopies with the results further compared with the predictions of band structure obtained from DFT.

Results

Simulations of implantation process

Prior to carrying out the experiments on ion implantation onto ML MoS2, we theoretically studied the behavior of the system under ion irradiation in vacuum. Specifically, we used atomistic computer simulations at AP and DFT levels of sophistication to investigate the energetics of defects, which can be formed in the irradiated samples at ion energies used in the experiments. We also assessed the probability of defects to appear upon ion impacts. The details of our calculations are given in the Methods section.

It is intuitively clear that a low-energy ion colliding with a ML can be reflected back from the ML, incorporated in the substitutional or interstitial position in the lattice of the ML, or adsorbed on the surface as an adatom. At higher ion energies the ion will go through the ML with or without defect creation in the ML. The likelihood of each of these scenarios depends on the ion type, its energy, and the formation energy of a particular defect.

Our implantation hypothesis is further explored by the DFT calculations, results of which are summarized in Fig. 1. The calculations show that the energetics of Se substitution versus adsorption defines the defect structures during implantation with ions at low energies. As revealed by the DFT calculations, the energy needed to create a S vacancy in MoS2 (Fig. 1a) is lower than that required to create a molybdenum (Mo) vacancy (Fig. 1b). We stress that our interest is in the formation of defects upon irradiation, not in their equilibrium concentration. We define the defect formation energy here as the difference between the energies of the pristine system and that with this defect, as previously done for the system of vacancy plus atom taken to infinity in the simulations of effects of electron irradiation36. At the same time, defects already present in the system will influence the interaction of the projectiles with the system. For example, it is energetically more favorable for a Se atom to be incorporated into the existing vacancy (Fig. 1c) than to form an adatom (Fig. 1d). The energy corresponding to healing of a vacancy defect by Se atom in MoS2 is about 5.82 eV, which is significantly higher than the Se adsorption energy of 2.39 eV, suggesting that defect healing in MoS2 ML by Se is thermodynamically preferred. There is an energy gain of 0.6 eV in swapping Se–S atoms in the implanted MoS2 ML, so that S is incorporated in the lattice and Se becomes an adatom (Fig. 1e). This can happen if the displaced S adatoms can be chemically adsorbed at the sites, where Se healed S vacancies. Furthermore, the adsorption of a Se dimer into a S vacancy can be expected as a relatively stable configuration, especially under the Se-rich environments (Fig. 1f).

Fig. 1: Energetics of various defect configurations in MoS2 ML, which can form upon Se ion impacts.
figure 1

The energy required to form a S vacancy (a) and Mo vacancy (b) is determine by taking the corresponding atom to infinity. Filling of a S vacancy by a Se atom (c) is energetically preferable over forming an adatom (d). Panel e illustrates the energetics of an atom exchange and f an interaction of the S vacancy with Se dimer.

Full size image

The electronic structure calculations showed that the existence of a S vacancy introduces a highly localized acceptor state in the band gap of MoS2 ML, i.e., 1.1 eV above the top of the valence band maximum, see Fig. 2b.

Fig. 2: Calculated projected densities of states.
figure 2

for a pristine material, b a single S vacancy, c a substitutional Se impurity, and d a Se adatom. The Fermi energy is set to 0 eV, and also chosen to coincide with the valence band maximum of the pristine system.

Full size image

At low implantation levels, the filling of the S vacancy with a Se atom in the substitutional configuration recovers the band structure of a MoS2 ML (Fig. 2c). Closer analysis, which takes into account the fact that implanted Se atoms can be displaced in the out-of-plane direction, shows that the direct band gap of the compound increases with the fraction of Se incorporated in the top S sub-lattice of the ML (Supplementary Fig. 7). Simultaneously, the conduction band edge decreases in energy at the Q point (midway between the K and Γ points), and the band gap thus changes from direct to indirect above a certain Se concentration. Our calculations predict that this transition occurs at a Se concentration of about 20% (in the top chalcogen layer). This behavior is different from the calculations of relaxed random and asymmetric MoSSe compounds, which feature a direct band gap37,38. Since in our calculations we fixed the ML lattice constant to that of pristine MoS2 in order to match the experiment (in which MoS2 is supported on SiO2), the difference is due to strain in our film. Finally, Se adatoms, which are predicted to remain on the MoS2 surface, add an occupied donor state in the band gap at about 0.2 eV above the valence band edge (Fig. 2d). This state is mostly composed of Se p- and Mo d-states indicating strong hybridization between the two states.

Considering the large variation in the formation energies of defects39 that can be created during ion implantation, we used MD simulations to find out the likely atomic configurations of MoS2 ML after an impact of Se+ ion. In the simulations we launched an ion at normal incidence and with kinetic energy in the range between 10 and 25 eV towards the target, which was a free-standing ML MoS2. Both AP and DFT force models were used to describe the interactions between the atoms: AP simulations were repeated 128 times for randomly chosen impact points in the minimum irreducible area40 of MoS2 ML. In both approaches the ion was assumed to be a neutral atom, as for such low ion energies, it is likely to neutralize before it reaches the surface of the ML. Owing to the high computational costs, only 10 impact points were used in the DFT MD. The results of the simulations are summarized in Table 1. At the lowest considered ion energy of 10 eV, Se atoms are most likely to be adsorbed on the MoS2 ML as adatoms. It is less likely that the Se ion is reflected back, and there is a small probability that Se will substitute S-atom in the lattice. When ion energy increases, the probability for Se substitution into the S-site increases and according to AP simulations reaches almost 40% at ion energy of 25 eV. Simultaneously the fraction of ions, which incorporate as adatoms drops from 76% at 10 eV to 22% at 25 eV. A minimum of 20 eV ion energy is needed before S vacancies can be generated upon Se ion impacts.

Table 1 Probabilities by MD simulations of the most likely outcomes of a Se ion colliding with MoS2 ML.
Full size table

Under normal incidence, Se was found to replace the S atoms in the top layer of pristine MoS2. No Mo vacancies were found under this implantation condition, even when a vacancy already existed in the system (see Supplementary movie 1 that shows lattice dynamics upon head-on impact of a Se ion into Mo atom neighboring a S vacancy). Moreover, here we used non-spin-polarized DFT. As recently shown41, the initial kinetic energy of the projectile should correspond in this case to a slightly higher (by the Se atom spin polarizaiton energy) kinetic energy of the ion. At such low energies and relatively heavy ions with normal incidence, the presence of the substrate should also decrease the number of vacancies in the system42,43, especially in the bottom layer44. We also note that the analysis of the simulated final atomic configurations did not provide any evidence for the substitution of atoms in the bottom layer.

The DFT and AP results are in overall good agreement. In most cases the probabilities of various outcomes of an ion impact agree within the discretization error bar (10% for full DFT MD, 0.8% for AP MD). A discrepancy can be seen in the probability of generating a vacancy with ions implanted at 20 and 25 eV. For 20 eV, the DFT MD predicts much larger concentration of vacancies of 20 ± 10% as compared to 2.3 ± 0.8% when using AP MD. Besides, the threshold for producing substitutional impurities appears to be higher in DFT MD. Nevertheless, both methods give thresholds for efficient Se substitution in the S sub-lattice at the energy between 15 and 20 eV. At these energies the probability of having an adatom is still higher than the probability of forming a vacancy.

Sample preparation and ion implantation

Bulk MoS2 crystals were used for mechanical exfoliation into MLs. The exfoliated MLs were transferred on SiO2/Si chips with pre-defined electrical contacts, as described in the Methods section. A typical sample is shown in Fig. 3. Care was taken to avoid surface contamination and each ML to be implanted was tested using atomic force microscopy (AFM) and Raman spectroscopy to assure their quality.

Fig. 3: A micrograph of a typical device prepared for implantation.
figure 3

The exfoliated MoS2 flake containing a ML is placed on a Si/SiO2 substrate with metal contact pads. The thicker part of the MoS2 flake makes contact with the Ti/Au contact pad. A filter was applied on the optics for better contrast on the ML.

Full size image

80Se+ ions were selected for the implantation and their energy was set to 20 eV. Since the deceleration voltage is set relative to the potential of the ion source anode, the given ion energy is the upper limit of the energy distribution of the ions in the beam. The low-energy tail of the distribution is caused by the energy spread of the ions extracted from the source. The selection of this ion energy was based on the simulations discussed in the previous section as one at which we expect relatively high probability of Se atoms incorporating in the S sub-lattice and low probability of forming S vacancies. The implantation fluence, determined from the measurement of the time integrated current of impinging ions, was varied between 0.8 and 5 · 1014 cm−2. The reliability of this measurement method was verified as described in the Methods section. With the atom density of 1.11 · 1015 cm−2 in the top S sub-lattice of the ML MoS2, the range of used ion fluences corresponds to 0.07–0.45 Se ions per S-atom. Assuming Poisson distribution of the collisions of ions with the S atoms, 7% of S atoms will experience a collision with a Se ion at the lowest fluence, while the fraction of S atoms experiencing collisions with Se atoms is 36% (including 6% of atoms experiencing more than one collision) at the highest fluence. In this case, we expect the compounds to be asymmetric, with Se occupying predominantly the top S-sub-lattice, while the implantation into the bottom S-sublayer can be possible upon non-normal propagation of ions or after the generation of vacancies in the top layer.

Optical properties of Se-implanted MoS2 MLs

Raman spectroscopy is commonly used to determine the composition of TMDs, including that of MoS2−2xSe2x45,46,47. Figure 4 shows room temperature Raman spectra (under 532 nm excitation) of a MoS2 ML and a MoS2 ML implanted with 80Se+ ions at the ion energy of 20 eV and a fluence of 0.8 · 1014 cm−2. The spectra are normalized to the first order optical phonon mode of Si substrate at 520 cm−1 48 (not shown for clarity). Both spectra contain a signal from the out-of-plane A' and the in-plane E′ vibrational modes of MoS2, at 405 cm−1 and 385 cm−1, respectively. All the remaining features, except from the one at 271 cm−1, are also present in the pristine ML and attributed to modes in MoS2 ML (LA(M) and 2LA(M) at around 200 cm−1 and 450 cm−1)26 or of the Si substrate (2TA at around 300 cm−1)48. The mode at 271 cm–1 has been reported in MoS2−2xSe2x alloys at low Se concentration (x = 0.04–0.16)45,49. The same feature was observed by Lu et al.50 and was attributed to the out-of-plane vibrations between Mo and the neighboring S and Se layers, \(A_{{\mathrm{MoSSe}}}^\prime\). The mode at 360 cm−1, while present as a shoulder in the pristine ML, has become stronger after implantation.

Fig. 4: Room temperature Raman spectra of mechanically exfoliated MoS2 ML before and after implantation.
figure 4

The implantation conditions were with 80Se+ at 20 eV at the fluence of 0.8 · 1014 cm−2. The spectra were acquired with a 532 nm laser excitation. Lines in the spectra are labeled with the vibrational modes that they originate from. E′ and A′ are vibrational modes of MoS2.

Full size image

In order to assign the observed features to the isolated defects introduced by implantation, we calculated Raman spectra for the three defects that can be generated in the ML according to our MD simulations: Se substitution, Se adatom, and S vacancy. Scanning transmission electron microscopy images (see Supplementary note 7) of a ML implanted with Se provided an independent verification of the incorporation of ions into the chalcogen sub-lattice and the presence of adatoms and vacancies.

The results of the DFT simulations of Raman modes in Supplementary Fig. 1 show that the Se substitution model reproduces well the peak at 271 cm−1 found in the experiments, whereas the spectrum from Se adatom or vacancy models are inconsistent with it. More specifically, the peak at 271 cm−1, which is found already at low Se concentrations, arises from S–Mo–Se bonds45,49. It is relatively sharp (i.e., reflecting small dispersion) since it falls into the phonon band gap between the acoustic and optical modes of MoS2. The precise wavenumber of the mode depends on the bond length and angle used in the calculations; in the measured samples the bond length depends critically on the strain of the ML. The mode at 360 cm−1 appears for all the three defects as it is caused by breaking the lattice symmetry. The model predicts additional features at around 220 and 300 cm−1, especially strong for the case of S vacancies. These are not clearly observed in the samples with low implantation levels, where the model is directly applicable. We also do not observe the broadening of the E′ line of MoS2 predicted by the model, so vacancies are the least likely origin of the mode at 360 cm−1. Similarly, we do not observe the 225 cm−1 line predicted for Se adatoms. These simulated features are described in more detail in Supplementary note 1. With an established implantation process (see Supplementary note 3), a set of MoS2 samples was implanted with Se ions, varying the fluence between 0.8 and 5.0 · 1014 cm−2 while keeping the ion energy at 20 eV. During the implantation, the ML target was electrically connected to the ground of the system, while the substrate was heated to 150 °C.

As discussed already and also shown in Fig. 5a, implantation at the fluence of 0.8 · 1014 cm−2 brought only small changes to the Raman spectrum of the ML, a slight broadening of MoS2 A′ and E′ modes accompanied by the appearance of the weak \(A_{{\mathrm{MoSSe}}}^\prime\) signal at 271 cm−1. Increasing the fluence to 2.0 · 1014 cm−2 led to several changes. The \(A_{{\mathrm{MoSSe}}}^\prime\) evolved into two lines, one at 276 cm−1 and another at 280.5 cm−1. In addition to the signal from S–Mo–Se vibrational modes, several MoS2 ML Raman modes are modified in these implanted samples. In particular, the MoS2 E′ vibrational mode shifts to the lower wavenumbers and it develops a strong shoulder at 375 cm−1, which indicates broken periodicity of the lattice26, which in turn can be here a consequence of the presence of Se atoms in the S sub-lattice as well as the presence of other defects. This evolution of the E′ line is the same as seen in the MoS2−2xSe2x MLs from other reports45,46,47. At the highest fluence of 5.0 · 1014 cm−2, the MoS2 A′ and E′ lines become even weaker and shift further to lower wavenumbers, while the shoulder of E′ line at 375 cm−1 is replaced by a well-defined peak at 350 cm−1. This mode has been assigned to in-plane vibration of S–Mo–Se bonds50, the stronger of the two E′ modes of the random and asymmetric MoS2−2xSe2x compound45,46,49,50. The other E′ mode of that compound is expected at 210 cm−1 51. We observe some enhancement of an asymmetric line at 220 cm−1 and further studies would be needed to differentiate it from the longitudinal acoustic, LA(M) phonon mode of MoS226. At the highest fluence, we observe a strong A′ mode of S–Mo–Se (\(A_{{\mathrm{MoSSe}}}^\prime\)) at 281 cm−1 accompanied by a shoulder at 276 cm−1may be the low wavenumber branch of the A′ line, which is predicted and observed to split in MoSxSe2−x at higher x values49.

Fig. 5: Raman spectroscopy studies of MoS2 MLs implanted with Se ions at 20 eV with different fluences.
figure 5

a Room temperature Raman spectra of MoS2 MLs implanted at four different fluences. The spectra are normalized to the Si signal at 520 cm−1 (not shown) and offset vertically for clarity. b Change of intensities of A′ Raman lines with ion beam fluence. Peak areas are obtained by fitting with Voigt profiles with Origin (OriginLab). Error bars are associated standard error from the fitting routine. The blue curve with the axis on the right shows the ratio between the out of-plane-vibration of S–Mo–Se bonds, \(A_{{\mathrm{MoSSe}}}^\prime\), to that of the pristine S–Mo–S bonds, A′.

Full size image

The measured upshift of the \(A_{{\mathrm{MoSSe}}}^\prime\) wavenumber is consistent with an incomplete replacement of S atoms with Se atoms and the formation of the S–Mo–S1−2xSe2x compound, as we do not expect a full replacement of the S sub-lattice at the used fluences. It is not possible to convert this shift into the composition of the compound. Direct comparison of Raman spectra to experimental data or model calculations (e.g., modified random element isodisplacement) in surveyed literature4,28,49,51,52 may not give a good estimate of Se content due to two reasons. Firstly, our implanted ML is likely strained with the lattice constant remaining at the value for pristine MoS2 ML. Compressive strain is known to cause shifting of Raman modes in TMD MLs53,54. Secondly, the Se atoms are randomly distributed in these works, while for our samples Se atoms are mostly located in the top chalcogen layer. The reported peak positions of pure Se–Mo–S Janus compounds, which may be more appropriate reference points for the asymmetric alloys studied here, are inconsistent with each other28,51,52 making it impossible to use them as a measure of Se concentration. Even though the intensity of A′ lines from S–Mo–Se increases with the Se ion flux fluence and the intensity of A′ line of MoS2 bonds decreases (see Fig. 5b), the ratio of the two cannot be used either because of the impact of all other defects on A′ of MoS2 (Supplementary Fig. 1).

The room temperature PL spectra of implanted MLs, an example of which is shown in Fig. 6a, are dominated by the sub-band gap emission, consisting of two broad bands Xs and X, positions of which evolve with the fluence of Se (Supplementary Fig. 6). Considering the simulations from Table 1, adatoms are the most likely defects and since the binding energy of Se adatoms is low compared with other defects (see Fig. 1), we used post-implantation ex-situ 3-h long annealing at a moderate temperature of 200 °C and a vacuum of 2–5 · 10−3 mbar to remove them. The PL spectrum of the post-annealed sample shows only one PL line, with Xs at 1.65 eV effectively removed. The Raman spectra of the same sample before and after annealing are compared in Fig. 6b. The increased intensity of the Raman \(A_{{\mathrm{MoSSe}}}^\prime\) mode following the annealing is observed indicating potentially further incorporation of Se in the substitutional sites. No other changes are noticeable, which is consistent with the predicted weak effect of Se adatoms on Raman signal (Supplementary Fig. 1). The PL peak of the post-implantation annealed sample is at a lower energy than the trion of pristine MoS2 and similar to the PL energy from the MoS2−2xSe2x compound (x ≈ 0.25)45,47. The PL is quenched at 40 K (Supplementary Fig. 8) indicating that it is a thermally activated process, as could be expected from our DFT calculations, which predict an indirect bandgap for sufficiently high level of implantation (Supplementary Fig. 7). The redshift of the measured PL line emission compared with MoS2 exciton (or trion) is not reflected in the calculated band gap at the K-point. This discrepancy may be due to strain introduced by the implantation process. The alternative, PL from S vacancies, can be excluded since it was measured to be located around 1.85 eV at room temperature55 and remained intense at lower temperatures25,56. While the full picture of the PL is not completely clear, its presence at room temperature indicates further that implantation preserved the quality of the material, a condition for the radiative recombination to remain a competitive exciton delay channel.

Fig. 6: The effect of ex-situ post-implantation annealing on Se-implanted MoS2 ML.
figure 6

a Room temperature PL spectrum from as-implanted ML (red line) shows a broadband emission comprising of two peaks Xs and X. The one at lower energy Xs is removed by post-implantation ex-situ annealing at 200 °C (blue line). The remaining higher energy peak X is red-shifted compared to the XT emission line from the measurement before implantation (black line). b Room temperature Raman spectra, which are normalized to A′ peak (of MoS2) to compare the intensities of \(A_{MoSSe}^\prime\).

Full size image

Discussion

To summarize, low-energy ion implantation into ML TMDs was studied on an example of 80Se+ implantation into ML MoS2. The experiments were guided by the simulations of ion interactions with MoS2 ML using AP and DFT potentials MD methods. With an ion energy of 20 eV, successful implantation required good electrical contact between the sample and the ground potential of the implantation system and elevated sample temperature. Se substitutional incorporation in S sites was verified with Raman spectroscopy, which showed very clear signal from S–Mo–Se bonds at ion fluences as low as 0.07 Se ion per S-atom in the top sub-lattice, confirming high Se incorporation rate as predicted by the MD methods. The ion fluence of 0.18 Se ion per S-atom in the top sub-lattice was sufficient to observe a Raman signal close to that reported for MoS2−2xSe2x compounds. Post-implantation annealing at 200 °C was effective in removing the defects responsible for low-energy photoluminescence from the implanted samples. Further atomic characterization techniques would be needed to quantify the implantation results.

Considering that the typical implantation level required to introduce doping is below what can be observed in Raman spectroscopy, and much below the level used in this work, we believe that low-energy implantation is highly promising as a MoS2 doping method. Since at such low ion energies, a selective area implantation can be performed by mechanical or even electrostatic masking of the ML target, the method has potentially strong practical value especially since it is also compatible with the standard semiconductor technology.

Methods

Molecular dynamics simulations

Atomistic calculations: All MD simulations have been carried out using the large-scale atomic/molecular massive parallel simulator (LAMMPS) code57. Periodic boundary conditions are applied to the in-plane directions. The interactions through the MoS2 ML were described by the Stillinger-Weber (SW) potential58, and the interaction between the ion (Se) and the ML atoms was described by the Ziegler-Biersack-Littmark (ZBL) universal repulsive potential59 smoothly joined the SW potential at small separations. This supercell corresponds to the 11 × 19 unit cells of the ideal lattice with dimensions of about 60 × 60 Å2 in a rectangular representation. Impact points were randomly selected via 128 sites within the minimum irreducible area of the lattice. We have used an adaptive time step for impact simulations, which is defined based on the velocity of the fastest moving atoms in the system, varying from almost one attosecond to one femtosecond. The calculations of the binding energies involved in Se sputtering on MoS2 was performed by using density-functional theory (DFT) with the PBE exchange-correlation functional, as implemented in the VASP code60. An energy cutoff of 500 eV has been used for the 4 × 4 supercell calculations. The Brillouin zones of the supercells were sampled using a using 4 × 4 × 1 Monkhorst-Pack grid points. The maximum force on each atom is set to be less than 0.01 eV/Å for optimized structures. In the simulations, different possibilities of Se adsorption positions have been considered and the most stable structures have been used for binding energy calculations. A lower energy cutoff of 350 eV and 2 × 2 × 1 k-point grid were used in DFT MD simulations.

Band structure calculations

In order to extract the data for Supplementary Fig. 7, band structure calculations of Se-implanted MoS2 were carried out with the DFT-based code FLEUR61 using the PBE exchange-correlation functional. A supercell of 3 × 3 primitive unit cells is used for the different concentrations of Se doping in the upper chalcogen layer. The properties of the resulting ML are crucially depending on the relaxation of the material, which was performed up to a maximum residual force for any atom of 5 · 10−2 eV/Å. The overall lattice constant is kept the same throughout the study of different concentrations. For very high concentrations, especially full replacement of the top layer, this implies that the material is strained influencing the band structure. Starting from a lattice constant of a = 3.15 Å, c = 12.3 Å, and an internal structure parameter of z = 0.124 for the hexagonal lattice structure, the layers are separated by four times the bulk distance. The self-consistent calculation was converged on a grid of 50 k points up to \(10^{ - 8}\;e/a_0^3\). The band structure was then backfolded to the primitive unit cell, and the relevant minima and maxima were extracted.

Sample preparation

Si/SiO2 chips with Ti/Au contacts (pre-patterned by electron beam lithography) were used as substrates for the optical measurements. The chips were cleaned in acetone and isopronanol (IPA) under bath sonication, blown dry with a N2 gun before treatment with oxygen plasma (300 W, 200 sccm for 10 minutes) prior to ML transfer. MoS2 MLs were mechanically exfoliated from a bulk crystal using polydimethylsiloxane (PDMS) stamps (Gel-pak DGL X4 films) and transferred onto the substrate using a dry viscoelastic transfer process62. This was done in a N2 filled glovebox. An electrical contact between the ML part of MoS2 and the metal contacts was provided by a few layer thick part of MoS2. The ML quality was later checked by Raman spectroscopy and AFM.

Se ion implantation

Prior to implantation, the samples were fixed on a special holder by bronze tips. The bronze tips make contact with the Au pads and, thus, the ML. The samples were then evacuated and kept at a pressure of 10−9 mbar for several hours to clean the surface from adsorbants. The ions are extracted from the source using a high voltage of 30 kV. Owing to the single positive charge, the ions possess an energy of 30 keV. To decelerate the ions to 20 eV, the potential of the sample holder and, thus, of the sample is set to 29.98 kV. Tests of implantation fluences were performed by implanting various ions into tetrahedral amorphous carbon (ta-C) thin films on Si substrates samples and subsequent analysis of the results using Rutherford backscattering spectrometry (RBS), more details are available in Supplementary note 2. For ions with energies between 20 and 100 eV and species Se, P, Fe we find that the implanted fluence determined by RBS is within 65–90% of the fluence determined by the ion charge measurement. Taking into account the possible reflection of ions and uncertainty in the implanted area, the ion charge measurement provides reliable values of the implanted fluence.

Raman and PL spectroscopy

Spectra were acquired at room temperature and in ambient pressure with a confocal Raman microscope (Renishaw inVia) using a 532 nm excitation laser (Coherent Compass 315M 150SL). The laser power was set below 250 μW for sufficient signal intensity without heating the sample under long exposure. The Raman signal was collected via an objective lens (50×, NA = 0.75, Leica N-plan EPI). The signal was dispersed using a 2400 l/mm grating and recorded using a CCD camera, giving a spectral resolution of 1 cm−1. For the PL signal, a grating of 600 l/mm was used, giving 0.147 nm resolution.