Giant persistent photoconductivity in monolayer MoS2 field-effect transistors

Monolayer transition metal dichalcogenides (TMD) have numerous potential applications in ultrathin electronics and photonics. The exposure of TMD based devices to light generates photo-carriers resulting in an enhanced conductivity, which can be effectively used, e.g., in photodetectors. If the photo-enhanced conductivity persists after removal of the irradiation, the effect is known as persistent photoconductivity (PPC). Here we show that ultraviolet light (wavelength = 365 nm) exposure induces an extremely long-living giant PPC (GPPC) in monolayer MoS2 (ML-MoS2) field-effect transistors (FET) with a time constant of ~30 days. Furthermore, this effect leads to a large enhancement of the conductivity up to a factor of 107. In contrast to previous studies in which the origin of the PPC was attributed to extrinsic reasons such as trapped charges in the substrate or adsorbates, we unambiguously show that the GPPC arises mainly from the intrinsic properties of ML-MoS2 such as lattice defects that induce a large amount of localized states in the forbidden gap. This finding is supported by a detailed experimental and theoretical study of the electric transport in TMD based FETs as well as by characterization of ML-MoS2 with scanning tunneling spectroscopy, high-resolution transmission electron microscopy, and photoluminescence measurements. The obtained results provide a basis towards the defect-based engineering of the electronic and optical properties of TMDs for device applications.

GPPC arises mainly from the intrinsic properties of ML-MoS2 such as lattice defects that induce a large amount of localized states in the forbidden gap. This finding is supported by a detailed experimental and theoretical study of the electric transport in TMD based FETs as well as by characterization of ML-MoS2 with scanning tunneling spectroscopy, high-resolution transmission electron microscopy and photoluminescence measurements. The obtained results provide a basis towards the defect-based engineering of the electronic and optical properties of TMDs for device applications.

TOC TEXT
We report on a giant persistent photoconductivity in monolayers of MoS2 with a time constant of more than 30 days after irradiation with UV light. Our detailed microscopy study demonstrates that the main reason of this phenomenon are the intrinsic structural defects resulting in strong spatial variation of the monolayer band structure.
Persistent photoconductivity (PPC) has long been studied in amorphous as well as highly-compensated wide-bandgap bulk semiconductors and was attributed to the presence of large spatial fluctuations of the potential energy of charge carriers (electrons and holes). 1,2,3 In case of transition metal dichalcogenides (TMD), the PPC effect with the respective time constant, , of 10 2 -10 4 s was reported for monolayer MoS2 (ML-MoS2) after their irradiation with visible light at room temperature (RT). 4,5,6,7 An even higher  10 6 s was observed for few layers of MoS2 after UV irradiation ( = 254 nm). 8 In these studies, the PPC effect was related to the charge traps caused by inhomogeneities either in the substrate 4,6,9 or in the adsorbates 8,9 on the TMD surface. Here we show that the long living photo-generated charge carriers may also originate from intrinsic, material specific lattice defects resulting in a prolonged recombination time of photo-generated carriers.
We demonstrate the extremely long living giant persistent photoconductivity (GPPC) in field effect transistors (FETs) fabricated from single crystalline ML-MoS2 grown by chemical vapor deposition 10,11 (CVD) after their exposure to UV light ( = 365 nm), Fig.   1. At RT the photo-generated charge carriers lead to an increase of the conductivity by a factor of up to ~10 7 , which depends on both the applied gate voltage (V g ) and the irradiation intensity (Fig. 1b). The high conductivity state persists for a long time with a time constant of ~30 days (3×10 6 s) at V g = 0 V (Fig. 1c). We explain these experimental findings with a model considering the presence of large spatial fluctuations of the potential energy of carriers (electrons and holes) in the ML-MoS2.
These fluctuations lead to a spatial separation of photo-generated carriers, as electrons (holes) concentrate in the minima (maxima) of random potential energy landscape (see schematic in Fig. 1d) resulting in a giant increase of their recombination time. 12,13,14 Transport of these photo-generated carriers displays two regimes: (i) the thermal activation regime at room temperature (RT) and the variable-range hopping regime at low temperatures (LT). Carrying out a quantitative analysis of the transfer characteristics in both regimes, we extract such parameters of the random potential energy landscape as the characteristic amplitude and the correlation radius as well as the variation of the concentration of photo-generated carriers with time (inset in Fig. 1c) and UV irradiation intensity (Fig. 2e, f). Large fluctuations of the potential energy result in a substantial concentration of strongly localized states in the forbidden energy gap. By performing scanning tunneling spectroscopy (STS) we experimentally confirm the presence of such localized states in ML-MoS2. Atomically-resolved transmission electron microscopy (TEM) enables us to correlate these findings with the density of the point defects in the samples.

Transport measurements of MoS2-FETs and observation of GPPC
In Fig. 1a a schematic representation of the experimental setup for the transport measurements is shown. As grown ML-MoS2 were characterized by optical microscopy, Raman spectroscopy, and atomic force microscopy (Figs. S1-2).
Afterwards, the FETs were fabricated using e-beam lithography. An optical microscopy image of a typical device is shown in the inset of Fig. 1b. We measured the transfer characteristics of these devices, i.e. the drain-source current, I ds , versus V g before and after UV irradiation (λ = 365 nm, intensity of ~30 mW/cm 2 for 5 minutes). All measurements were performed in a high vacuum (~10 -6 mbar) and under dark conditions; these data are shown in Fig. 1b. Note that we studied a possible effect of the UV irradiation induced damage of the ML-MoS2 by conducting Raman spectroscopy before and after the irradiation and did not find any noticeable changes in the spectra (see Fig. S3).

Figure 2.
Transfer characteristics of a MoS 2 FET at the pristine condition and after UV irradiation with various intensities. The irradiation time for each intensity was 5 min, and all transfer curves were recorded in dark immediately after irradiation. Experimental transfer curves at RT (a) and at 6 K (b). Theoretically calculated transfer curves RT (c) and at 6 K (d). Calculated photoelectron density as a function of the irradiation intensity at RT (e) and at 6 K (f). The dashed lines in (e) and (f) are presented as a guide for the eye. The GPPC at a function of time for this device is presented in Fig. S4.
From the transfer characteristics of the pristine device (black line in the Fig. 1b) we estimate a field effect mobility of 1.5 cm 2 /Vs, which is a typical value for CVD grown ML-MoS2. 11,15 Directly after UV irradiation, we observe a very strong enhancement in the I ds (see red line in Fig. 1b) of up to 10 7 at V g = -40 V, which is close to the device threshold voltage. We found that the GPPC persists even for days at RT, albeit decaying in its strength over time (transfer characteristics recorded between 1 and 30 days are shown in Fig. 1b; Fig. S4 shows the full data set). During these measurements the devices were always kept in high vacuum and under dark conditions. The obtained decay of I ds over time is shown in Fig. 1c at V g = 0 V. These data can be described using a two-stage exponential decay function. In the initial stage, the I ds decays with a time constant of  1  1 day, whereas in the following stage, the GPPC relaxation slows down yielding a time constant of  2  34 days. Similar values were obtained for more than 10 devices made of ML-MoS2 synthesized in different CVD experiments (see, e.g., Fig. S5). Note that after a few months, the transfer characteristics of the devices completely recover to their pristine state before irradiation (Fig. S6a). We also found that vacuum annealing (~10 -2 mbar) at 170 °C results in a significantly faster decrease of the persistent photocurrent at RT (see SI p. S6 and Fig. S6b), which agrees well with the thermally enhanced recombination of the photo-generated carriers. Similar as after their recombination at RT, a subsequent UV irradiation of the device induces the initially observed GPPC effect.
To investigate the GPPC in detail, we measured the transfer characteristics of the MoS2-FETs at RT and low temperature (LT, 6 K) after UV irradiation of varying intensities between 2 and 30 mW/cm 2 for 5 min as shown in Fig. 2a and 2b, respectively. In both cases, we observed an enhancement of the I ds with increasing UV intensity. At LT, the absolute values of I ds were lower than at RT for all V g , which we attribute to the variable range hopping type of transport caused by the strong localization of the charge carriers. 16,17,18 Model of the photo-induced charge transport in the presence of spatial fluctuations of the band structure To rationalize the experimental observations of the GPPC in MoS2-FETs, we apply the following model. We assume that the crystal lattice defects and lattice strain lead to large spatial fluctuations of the potential energy of carriers in the ML-MoS2, predominantly because of the deep lying traps in the proximity of the conduction band (CB) / valence band (VB) edges as schematically depicted in Fig. 1d. We describe these fluctuations effectively by a coordinate dependent potential with zero mean value, / , where is the typical amplitude of the random potential and in our case ≫ , T is the temperature, k B is the Boltzmann constant, and is the correlation radius. 12 By using such a model, we carry out a quantitative analysis of the transfer characteristics in two regimes. At RT, the conductivity of the ML-MoS2 is defined by thermally induced activation of localized electrons surpassing the percolation level (the mobility edge), . The is the characteristic energy above which electrons are delocalized, i.e. propagate along a conduction channel experiencing only weak scattering 12,13 . In this regime, the concentration of delocalized electrons depends strongly on the temperature as well as V g and thus allows one to tune the conductivity (see dotted line in Fig. 2a). The conductivity in the thermal activation regime is obtained as where µ e is the quasi-Fermi energy level considering both equilibrium electrons (µ e,D ), i.e. the electrons which are present in the absence of irradiation, and non-equilibrium, photo-generated electrons (µ e,l ) (see Fig. 1d), and σ 0 is a coefficient which depends only weakly on both V g and the intensity of the UV radiation. In strongly disordered materials with many localized states, µ e is determined by the concentration of electrons according to 12,19 µ , where is the concentration of equilibrium electrons given by , where ε r = 3.9 and d = 300 nm are the relative dielectric constant and the thickness of the SiO2 gate insulator, respectively, is the vacuum permittivity, is the electron charge, is the concentration of non-equilibrium electrons, is the intensity of UV irradiation, n 0 is the maximum possible concentration of electrons in the CB and is the density of states. The density of localized states in the bandgap is approximated by 13, 20 where is the energy of localized states, and 0 CB E is the bottom of the CB in the absence of spatial fluctuations. Note that for disordered two-dimensional (2D) materials the percolation level is given as 13 The calculated transfer characteristic for the pristine, i.e. non-irradiated sample (see dotted line in Fig. 2b), are in a good agreement with the experimental data (dotted line in Fig. 2a with . µ . Using (4) to fit the experimental transfer characteristics for the pristine sample (the dotted lines in Fig. 2b,d), we extract the correlation radius of the random potential U(r), r corr = 5 nm. Now, by fitting our experimental transfer data at different UV irradiation doses, and , to the theoretical model (see Fig. 2c and 2d) we calculate the variation of the concentration photo-generated electrons, n non-eq (J, t), with time at RT (inset Fig. 1c) and with UV intensity at RT and LT (see Fig. 2e, f). In agreement with our experiment, the decay of the long living photo-generated electrons with time shows a bi-exponential decay process (see Fig. 1c). We attribute the two different exponents to the variation of the spatial separation between photo-generated electrons and holes, i.e. a part of photogenerated electrons and holes are localized in close proximity and thus recombine faster, whereas another part is separated at longer distances further away from each other and therefore recombine slower. 14 Furthermore, the previously reported photogating 4, 6, 7 effect may also contribute to the faster exponent. Summarizing this part, we conclude that our experimental transport data and the theoretical analysis suggest large spatial fluctuations of the band structures in MoS2-FETs, with the minima of the CB and maxima of the VB serving as trap sites for the photo-generated carriers.
As we show next, our spectroscopy and microscopy study enables us to identify atomic vacancies 21, 22, 23 and strain 24 in the MoS2 monolayers as main reasons for the spatial variation of the band structure responsible for the observed GPPC effect. On the other hand, such extrinsic sources as adsorbates on the FET channel 6 or trapped ). Therewith we conclude that the apparent monolayer roughness obtained by STM is caused by its interaction with the substrate playing a minor role in the observed fluctuations on the CB and VB. We expect the intrinsic structure to be responsible for that.

Correlation between spatial inhomogeneity of the band structure with lattice defects and strain
In order to study the origin of the observed band structure fluctuations in the ML-MoS2, we performed the structural study using the aberration corrected high-resolution transmission electron microscopy (HRTEM). 27 A representative unprocessed image of the ML-MoS2 is shown in Fig. 4a, which clearly demonstrates a presence of the sulfur vacancies. 28,29 In Fig As the differentiation between single (S1) and double (S2) sulfur vacancies is not possible by counting the black dots in Fig As a result, the concentration of the S2 vacancies was found to be 0.067(2) vac/nm 2 , which corresponds to 8.5 % of the total concentration. Summarizing these results, we ascribe the point detects in ML-MoS2 (S1 and S2 vacancies), their concentration and spatial distribution to the observed fluctuation of the band structure presented in the previous section.
As ML-MoS2 grown by the CVD method is known to build up biaxial strain during the cooling step due to mismatch of the thermal expansion coefficients with the underlying SiO2 substrate, 24 this lattice strain can also cause the band structure fluctuations, 30 which contribute to the observed photoconductivity in the MoS2-FETs. From the

Comparison with monolayer WS2-FETs
To further support the defect induced origin of the observed GPPC in MoS2-FETs, we carried out a comparative study of WS2-FETs made of CVD grown monolayers.
Evaluation of the HRTEM data presented in Fig. 4c-d shows that in this case the total intrinsic concentration of sulfur vacancies in ML-WS2 is 0.49 (9) vac/nm 2 , which is a factor of 1.6 lower than in ML-MoS2. Additionally, the concentration of S2 vacancies is about 0.022(5) vac/nm 2 , which means that the relative concentration of S2 vacancies in ML-WS2 in comparison to ML-MoS2 is a factor of 3 lower. In agreement with this evaluation, we found that in WS2-FETs the PPC is significantly weaker in comparison to MoS2-FETs (see Fig. S16). At similar conditions the PCC reveals a time constant () of only ~6 hours. In contrast to MoS2-FETs our control experiments with WS2-FETs show that rather the external factors, i.e. the adsorbates, the monolayer/substrate interaction and photogating 4,6,7 than the internal structural defects contribute to the observed PPC of in the latter devices (Fig. S17, SI Part 2 for details).

Modification of the optical emission by UV irradiation
Finally, in addition to the modification of the transport properties by UV irradiation, we also expect that the optical properties of the ML-MoS2 are modified. After the irradiation, a significant amount of VB electrons is excited and localized in the trap states below the CB, which has to result in a significantly quenching of the photoluminescence (PL). We performed PL emission mapping of as grown ML-MoS2 crystals on SiO2/Si substrate before and after UV irradiation, see Fig. 5a and Fig.5b, respectively. As can be seen, after the irradiation with UV irradiation the PL emission is significantly diminished, which is in agreement with our expectation. We further tested this effect by preparing suspended ML-MoS2 on TEM grids and observed similar behavior (see SI Fig. S18 and SI Part 6 for details). Thus, the GPPC induced by UV irradiation also allows one to effectively modify the related physical properties of ML-MoS2. optoelectronics. Furthermore, efficient routes towards defect engineering of ML-TMDs may enhance their applicability. We anticipate that the GPPC effect can be effectively exploited further for applications in electronic, optoelectronic and biotechnological devices (see, e.g., Refs. 32,33,34,35,36 ).

Methods
The ML-TMDs used in this study were synthesized by CVD method. 10,11 The basic properties of the monolayers after the synthesis were characterized by Raman and Xray photoelectron spectroscopy as well as by optical and atomic force microscopy.
After this characterization, the FET devices were microfabricated by electron beam lithography. The electric transport measurements before and after the UV irradiation ( = 365 nm) were conducted in a vacuum probe station (10 -6 mbar). The PL measurements were conducted at RT using the excitation wavelength of 532 nm. The electronic band structure of the monolayers down to the atomic scale was characterized by low temperature (1.1 K) scanning tunneling spectroscopy measurements. The chromatic (Cc) and spherical (Cs) aberration-corrected highresolution low electron energy TEM was conducted at RT using the SALVE microscope. 27 All experimental procedures and the data evaluation are presented in SI in details.

Data availability
Data presented in this study are available on request from the authors.

Acknowledgements
We

Additional information
Correspondence and request for materials should be addressed to A.T. (andrey.turchanin@uni-jena.de).

CVD growth of monolayer MoS2 (ML-MoS2) and monolayer WS2 (ML-WS2)
ML-MoS 2 and ML-WS 2 crystals were grown by the CVD process. 1, 2 Silicon substrates with a thermally grown SiO2 layer of 300 nm were used as substrates (Siltronix, roughness 0.3 nm RMS). The growth was carried out in a two-zone tube furnace with a tube diameter of 55 mm. The substrates were cleaned initially by ultrasonication in acetone for 5 min

S4
For the growth of the 2D h-BN layer we followed the steps described by Orlando et al. 3 They showed that the adsorption of borazine on Ir(111) at room temperature followed by an annealing step at 800°C reduces the number of different adsorption configurations and results in an h-BN layer exhibiting a low defect density. We thus exposed the substrate to borazine (B 3 N 3 H 6 , Katchem spol. s.r.o.) vapor with a pressure of 5x10 -8 mbar for 10 min with the substrate being held at room temperature. Immediately after that we increased the substrate temperature to 800°C for 10 min followed by an annealing step at 800°C for another 10 min without borazine exposure in order to guarantee the complete dehydrogenation of the precursor molecules on the substrate. This growth progress is known to be monolayer-terminated. 3,4 The quality of the h-BN layer was verified with XPS. Fig S11b shows the corresponding B 1s and N 1s spectra. We used a polar angle of 70° to increase the surface sensitivity of the measurement. Both spectra show only one component, which was modelled by an asymmetric Mahan line shape. 5 The asymmetry of the peak can be explained by the slight corrugation of the h-BN layer as well as by scattering of the photoelectrons at electronic states near the fermi edge. 5 The amount of contaminations like oxygen and carbon was less than 1%. Point defects in the h-BN layer should cause a second component in the B 1s and N 1s spectra at lower binding energy which is not visible in the observed XP spectra 6 . In conclusion, our results show a high-quality h-BN layer which is a suitable substrate for STS investigations of a MoS 2 film.

Basic characterization by optical microscopy, atomic force microscopy and Raman spectroscopy of the grown monolayers
After the CVD growth the transition metal dichalcogenides (TMDs) were characterized by optical microscopy (OM), atomic force microscopy (AFM) and Raman spectroscopy (see The Raman spectra were acquired using a Bruker Senterra spectrometer operated in backscattering mode. Measurements at 532 nm were obtained with a frequency-doubled S5 Nd:YAG Laser, a 50× objective and a thermoelectrically cooled CCD detector. The spectral resolution of the system is 2-3 cm −1 . For all spectra the Si peak at 520.7 cm −1 was used for peak shift calibration of the instrument. The Raman spectrum shown in Fig.   S2b reveals the characteristic peaks of ML-MoS2 at 384cm -1 and 404 cm -1 , which are originated from the in-plane (E 1 2g band) vibrations of the Mo-S bonds and out-of-plane (A1g band) vibrations of S atoms in the MoS2 lattice, 7 respectively. The difference between the peak positions is 20 cm -1 confirming that the crystal is a monolayer. 7

Preparation of ML-MoS 2 and ML-WS 2 FET devices
After the growth, the ML-MoS 2 and ML-WS 2 crystals were transferred onto the device nm) electrodes were deposited by e-beam evaporation process followed by the dissolution of the PMMA resist in acetone for 2h. The heavily p-doped silicon base functioned as the gate electrode and 300 nm thermal oxide functioned as the gate dielectric.

Electrical transport measurements.
The electrical characterization was carried out with two Keithley 2634B source measure units (SMU). One SMU was used to change the voltage of the gate (V g ) with respect to the source/drain in the range between -60 and 60 V for the back-gated devices in vacuum (~10 -6 mbar). The other SMU was used to apply the source-drain voltage (V ds ) and measure the source-drain current (Ids). A Lakeshore cryogenic vacuum needle probe station TTPX was used to measure the devices in vacuum at a residual pressure about equation µ , where L is the channel length, W is the channel width, C ox is the capacitance of the 300 nm gate oxide and Vds is the source-drain voltage 2 . For UV irradiation a light emitting diode (LED) with a wavelength of 365 nm (Thorlabs, M365L2), with a typical power output of 360 mW, was used. Also LEDs with wavelengths 455 nm (Thorlabs, M455L3) and 617 nm (Thorlabs, M617L4), respectively, were used to irradiate the devices. The LEDs were controlled using a Thorlabs four-channel LED driver (DC4100).

Scanning tunneling microscopy and spectroscopy (STM/STS)
CVD grown ML-MoS2 crystals were transferred onto a hBN layer grown on a single crystalline Pt (111)  at about 120 °C for 2h. For STS we directly measured the derivative of the tunneling current dI/dV using the lock-in technique. We recorded STS spectra in a grid with dimensions of 50 nm × 50 nm and 1 nm spacing in each direction (2500 spectra in total).
No hysteresis was observed between forward (1.5 V  -2.75 V) and backward (reverse) bias sweeps, which proves that we do not permanently influence the band structure by the measurement process itself. In order to visualize the spatial distribution of trap states we plotted their onsets as color maps as shown in Fig 3b and 3c.

High resolution transmission electron microscopy (HRTEM)
The HRTEM images were acquired with the Cc/Cs-corrected Sub-Angstrom Low-Voltage Electron microscope (SALVE) 8 . A voltage of 60 kV was used with typical dose rates of about 10 5 e -/nm 2 s. The values for the chromatic aberration Cc and the spherical aberration Cs were between -10 µm to -20 µm. All Cc/Cs-corrected HRTEM images were acquired with bright atom contrast and recorded on a 4k x 4k camera with exposure times of 1 s.

Photoluminescence (PL) measurements
PL from ML-MoS2 crystals was characterized with a MicroTime 200 laser-scanning confocal fluorescence microscope from PicoQuant GmbH. A pulsed laser of wavelength 532 nm and repetition rate of 80 Hz was used to excite the ML-MoS2 crystals and measure their PL emission with a single-photon avalanche diode (SPAD) detector. A microscope objective of 40x magnification and a numerical aperture 0.65 was used to focus the laser onto the crystals, forming a spot of diameter ~1 µm. The PL emission was collected using the same objective. PL maps were acquired by raster scanning of the microscope objective and collecting the PL emission in the spectral range of 650 -720 nm using a band pass filter, essentially to collect the A-exciton and trion emissions. Care was taken in all the measurements to block the excitation light reaching the detector using a dichroic mirror and a notch filter for the excitation wavelength of 532 nm, in addition to band pass and long pass filters.

Influence of substrates and adsorbates on the GPPC
To further investigate the reason of the GPPC and to answer the question whether it is of extrinsic or intrinsic origin, we performed electrical measurements as proposed by Kis et al. 9 It is suggested that the surrounding material of the MoS 2 channel, especially the substrate, plays an important role in photocurrent dynamics and the PPC by providing traps for charge carriers, which, however, can be discharged by applying a short gate pulse. 9 In Fig. S7, we show the rise of Ids by irradiation with UV light. After the removal of the UV irradiation, the I ds remains nearly constant. We then applied several short gate pulses from -60 V to 0 V to test whether the transfer curves can be restored to the initial state. The effect of gate pulses is minimal as can be seen in Fig. S7. The GPPC behaviour was also tested after the application of the gate pulses and it can be seen that the effect is still present and the device shows similar GPPC decay behaviour even after the application of the gate pulses (Fig. S8).
Furthermore, we have studied the hysteresis of the MoS 2 -FETs. The respective transfer curves between forward and backward Vg sweeps of UV irradiated MoS2-FET devices are shown in Fig. S9, S10. If a significant density of charge traps was existing at the interface between the MoS2 channel and the SiO2 surface, one would expect significant hysteresis in the transfer characteristics of the FETs. 10,11,12 Absence of large hysteresis indicates S8 that additional charge traps either do not exist in significant concentration in our MoS 2 -FET samples, or do not contribute much to the here observed GPPC. Also, it is observed that the hysteresis of the devices remained the same after the application of several gate pulses (Fig. S10).
We conducted the gate pulse experiments on the WS 2 -FET devices as well where only a PPC effect is present instead of GPPC (Fig S16). In Fig. S17, the rise and fall of Ids by irradiation with UV light is shown. In this case, the I ds values before UV irradiation can be restored by applying a single short gate pulse as shown in Fig. S17. Thus, we cannot exclude a substrate effect to contribute to the PPC in our samples. Since the ML-WS 2 samples have fewer defects (as revealed in our HRTEM study) in comparison with ML-MoS2, the substrate effect is more prominent. However, for the case of MoS2-FETs, the substrate seems to play a minor role, and the defect related spatial fluctuations in the band structure is the prominent mechanism of GPPC.

Evaluation of the defect density by HRTEM
For the error evaluation, the evolution of the defect concentration was analysed for MoS 2 and WS2 (cf. fig. 4). Furthermore, a linear behaviour for low defect concentrations  Fig. S14. In total, three data sets for WS2 and also for MoS2 were evaluated. For the confidence intervals, we took ∆ √ for the vacancies and ∆ √ for the evaluated area. For the total accumulated dose, a huge confidence interval is assumed because of the uncertainties of the previous electron beam irradiation before the first image was recorded. Here, we assumed an error for the irradiation time of t = 10s, so that the confidence interval for the accumulated dose, depending on the dose rate , becomes ∆ 10 • (Fig. S14).

S9
In Fourier filtered Cc/Cs-corrected HRTEM images the location of the vacancies can be clearly identified but not the defect type. Therefore, the differentiation between double and single sulphur vacancies was performed in the unfiltered image by checking the contrast of each vacancy. Fig. S15 shows a Cc/Cs-corrected HRTEM image of MoS2 with single (S 1 ) and double (S 2 ) sulfur vacancies. For differentiation, the intensity profile along the line-scans are given (blue for double S2 vacancy and red for the single S1 vacancy) showing that the contrast for single S 1 vacancies is noticeable reduced compared to a pristine S column. Due to the absence of the sulfur atoms in the S 2 vacancy, no S-peak occurs in the profile (blue curve).

Estimation of the lattice strain in CVD grown ML-MoS2
Typically, ML-MoS2 grown by CVD at high temperature conditions are subjected to significant tensile strain due to the mismatch of thermal expansion coefficients of monolayers and the SiO2/Si substrate. 13 Furthermore, these monolayers are also subjected to electron doping from the SiO 2 substrate. The strain inside a CVD grown monolayer is estimated using a well-known analysis based on spectral positions of firstorder Raman modes as described previously. 14,15 Employing the Grüneisen parameters ( ), electron doping factors ( ) from the literature, 16,17 the strain is estimated relative to the exfoliated ML-MoS 2 , which are typically considered to be free of strain. The strain in ML-MoS2 grown with the CVD method is estimated to be 0.34±0.08 %. The parameters considered in the estimation of strain inside the ML-MoS 2 are shown in the table below.

Effect of visible light irradiation on the transport in MoS2-FETs
In addition to UV irradiation (365 nm), we also tested irradiation of MoS2-FETs with 455 nm and 617 nm light (Figs. S19, S20), and the devices did not show any GPPC. The devices returned nearly to their initial conditions after several hours. This indicates that irradiation with higher energy photons are required for causing the GPPC, while the exact reason for this is presently not well understood.

Photoluminescence (PL) measurements of suspended ML-MoS2
By preparing suspended ML-MoS 2 samples one could ultimately exclude substrate effects. We have prepared such samples by transferring ML-MoS 2 on to Quantifoil type TEM grids. Such a grid has a thin carbon film with an array of holes with a diameter of 2 µm separated by 2 µm. Thus, the ML-MoS 2 regions spread over the holes can be considered as (partially) suspended areas. The emission maps of a MoS2 crystal on such a TEM grid before and after UV irradiations (Fig. S18a, b) show a similar behavior as observed in as-grown samples. The areas that appear as bright dots are the suspended regions of the ML-MoS 2 , and it can be observed that the PL intensity is diminished after the UV irradiation, however, quantitatively not by the same magnitude as observed on SiO 2 /Si substrates. In order to interpret the UV induced quenching quantitatively, histograms of the PL intensity were extracted from these maps as shown in Fig. S18c, d.
The fraction of frequencies of bins with counts in the range of 0 -40 (the range where PL is absent) increased significantly from 81% to 92%, and bins with counts higher than 40 (the range where PL is present) decreased after the UV exposure. This observation infers that as a result of the UV exposure, the PL signal across the suspended UV irradiated MoS 2 crystal is quenched and thus causing a reshaping of the histogram. UV irradiation triggers photoexcitation and charging of traps, which is the prerequisite for the PPC, and thereby suggests that this effect is present even on suspended samples.    The decay shows similar behavior with or without the application of short gate pulses.  2.6x10 -6 I(V G ,t) = C 1 *e (-t/ 1 ) + C 2 *e (-t/ 2 ) + I 0 (V G )  Spatial maps of PL on pristine and UV exposed regions on the Quantifoil grid reveal a clearly visible quenching effect, however being quantitatively not of the same order of the quenching observed with UV exposed crystal on the SiO 2 /Si substrate (shown in Fig