Contact morphology and revisited photocurrent dynamics in monolayer MoS2

Two-dimensional (2D) layered transition metal dichalcogenides (TMDs) have emerged as promising materials for electronic, optoelectronic, and valleytronic applications. Recent work suggests drastic changes of the band gap and exciton binding energies of photo-excited TMDs with ultrafast non-radiative relaxation processes effectively heating the crystal lattice. Such phenomena have not been considered in the context of optoelectronic devices yet. We resolve corresponding ultrafast photo-conductance dynamics within monolayer MoS2 and demonstrate that a bolometric contribution dominates the overall photoconductance. We further reveal that a focused laser illumination, as is used in many standard optoelectronic measurements of MoS2, modifies and anneals the morphology of metal contacts. We show that a junction evolves with lateral built-in electric fields, although Raman- and photoluminescence spectra indicate no significant changes such as a crystal phase transition. We highlight how such optimized devices can drive ultrafast electromagnetic signals in on-chip high-frequency and THz circuits.


Introduction
Future applications of transition metal dichalcogenides (TMDs) rely on the fabrication of good and well-defined contacts.  Although a lot of progress has been made in the fabrication of metal contacts with reduced contact resistance, 3,6,7,10,25,26 the corresponding impact on optoelectronic phenomena in TMDs is not yet fully understood. 4,6 In fact, low resistant contacts remain still the key bottleneck for the realization of a high device performance, and they are especially interesting for optoelectronic devices to avoid depletion regions at the contacts. In general, a combination of top and edge contact is used to describe the morphology of contacts to TMDs. 6 The top contact is formed either by a van der Waals gap between metal and the TMD-crystal or by the creation of covalent bonds to the two-dimensional semiconductor, leading to Fermi level pinning 7,25 or the metallization 6,25 of the TMD in the contact region. As for pure edge contacts, the fabrication turns out to be rather difficult for obvious reasons and so far, has only been reported for graphene. 27 Thus, understanding the interfacial properties of the metal-TMD interface, dominating for most contact geometries, is a key aspect. Considerable efforts have been made by investigating different contact metals by means of density functional theory (DFT) calculations 25,26 as well as the realization of field effect transitors (FETs) with different metals. 3,4,7,10 These studies show, that for monolayer MoS2, Ti/Au contacts exhibit the least contact resistance and a Schottky barrier height of ~200 meV. 3,7,25 However, the influence of the metal contact on the underlying MoS2 remains unclear. This holds in particular, if one considers the presence of defects such as sulfur vacancies at the basal plane of the 2D material. Defects may not only cause Fermi level shifts of up to 1 eV over tens of nanometers, 4 they also open the possibility for the formation of covalent bonds between the contact metal and the TMDs. 6,25 Whereas covalent bonds may reduce the interface resistance, a significant impact on the Fermi level in the 2D material is expected. Recent reports 3 show by evaluating the transfer curves of field effect transistors, that the work function of the contact metal has very weak influence on the Fermi level pinning in MoS2. 7 Noteworthy, the best contacts so far have been achieved by inducing a metallic phase transition within the 2D semiconductors, e.g. via a phase transition from 2H to 1T in MoS2 FETs. 5 The Coulomb screening in monolayer TMDs can be significantly altered by the presence of a large density of photogenerated charge carriers. 15,[28][29][30][31][32] In turn, it has been demonstrated that both the quasi-particle band gap and the excitonic binding energies are renormalized after a pulsed photoexcitation of the TMDs. The renormalization effects are based upon an interplay of excitationinduced dephasing, phase-space filling and the mentioned screening of Coulomb interaction in combination with ultrafast non-radiative relaxation and recombination processes. 15,[28][29][30][31][32] Particularly, the latter give rise to a transfer of the photo-induced excess energy to the phonon bath. 32,33 In turn, the lattice heats up to several tens of Kelvin, and the corresponding cooling process can last for hundreds of picoseconds depending on the interfacial thermal conductance to the substrate and to the metal contacts of the TMDs. 32,34 Generally, there have been several reports on photocurrent and photoconductance phenomena in TMDs. 2,8,10,12,13,16,17,21 They mainly discuss the observed phenomena in terms of photo-thermoelectric and photovoltaic effects. However, the above-mentioned, drastic photo-induced ultrafast dynamics of the energy scales and local temperatures have not been considered in the discussion of photoconductance phenomena in TMDs yet. 4

Results and Discussion
Our study is based on several tens of different monolayer MoS2-flakes, which are micromechanically exfoliated from bulk crystals and then transferred by an all-dry viscoelastic stamping technique (cf. methods). 35 We utilize either Si/SiO2 or sapphire-substrates for the different spectroscopy methods (cf. methods). Fig. 1a depicts an optical image of a MoS2-flake with varying layer thickness (monolayer -1L, trilayer -3L, up to bulk) which is placed on top of a 10-nm-thick Ti contact. To demonstrate the laser annealing effect (cf. Fig. 1b and methods), we focus a laser with varying dwell times onto the flake (cf. dot pattern and black lettering in Fig. 1c).
Then, we perform atomic force microscopy (AFM) and Kelvin probe force microscopy (KPFM) to characterize the morphology and the work function change within the same flakes (cf. Figs. 1d, 1e, and methods). The impact of the laser-annealing on the work function of monolayer MoS2 is found to be as large as E ~ -0.16 eV (lower panel of Fig. 1b, and black lettering in Fig. 1e). A comparative study on thermally annealed monolayers suggests values even up to -0.4 eV (Figs. 2a to 2d, and methods). By utilizing Raman and photoluminescence (PL) measurements during the laser-annealing process, we can exclude metallization or the creation of a phase transition from the 2H to the 1T or 1T' phase 5 for monolayer MoS2 in contact with Ti or Au as possible explanations (cf. Supplemental Figs. S1 and methods). However, we can observe an energy difference of the A1g Raman mode of contacted and non-contacted monolayer MoS2, indicating a change of the charge carrier density 36 in the order of Δn ≈ -2.8•10 12 cm -2 (cf. Supplemental Fig. S2), which is consistent with a shift of the Fermi energy of ΔE ≈ -0.16 eV. Noteworthy, we find that the work function adjustment by the laser-annealing is a persistent effect. It can be observed by performing the KPFM measurements in ambient conditions one week after the actual laser-annealing. This clearly distinguishes the effect from volatile doping effects, such as the previously observed 5 photogating 36 or work function modulation 37 of monolayer MoS2, which are reversible processes related to adsorbed species and ambient gases. Moreover, we do not observe the work function shift for MoS2-flakes positioned directly on Si/SiO2 or sapphire as substrates, i.e. without metals involved (data not shown). In our understanding, the laser-annealing reduces the van der Waals gap between the contacting metal and the MoS2 (cf. Fig. 1b), resulting in an apparent work function adjustment. We note, however, that this gap alteration must be smaller than about ~1 Å, because it cannot be resolved in our AFM measurements (cf. Fig. 1d). it is determined relative to the value of Ti in air. 38,39 By the impact of a focused laser, we find that the work function of MoS2 can have a local gradient of ~0.16 eV on ~0.5 µm (cf. lower panel in Fig. 1b). This suggests that after an annealing step, horizontal Ti/MoS2 -MoS2 junctions do not necessarily consist of a Schottky barrier. 7,9,17,18,25,40 Instead, they rather comprise an effective junction with built-in electric fields due to the different doping levels (Figs. 3b, 3c, and 3d). In this respect, 1L MoS2 covered by the contact metal yields a shifted and pinned Fermi-level position, while MoS2 away from the contact is intrinsically n-doped, 1,4,8 forming an effective junction with built-in electric fields close to the contact's edge.
We point out that all of our investigated optoelectronic metal/MoS2-monolayer/metal devices (about 20 optoelectronic devices in the course of three years) finally turned into having a contact morphology as sketched in Fig. 3. We explain this observation by the long exposure times to a laser as is necessary for optoelectronic experiments. In other words, scanning photocurrent experiments with a focused laser spot anneal the contacts of metal/MoS2-monolayer/metal devices, and this annealing gives rise to a permanent renormalization of the contact energy landscape. 6 To resolve the impact of the annealed contacts on the optoelectronic dynamics, we present timeintegrated data as well as time-resolved photocurrent data on the same 1L MoS2 (cf. Figs. 4, 5, and S3). We utilize photon energies Ephoton exceeding the ones of the so-called A-and B-exciton in the MoS2 with absorbance energies of EA-exciton = 1.89 eV and EB-exciton = 2.03 eV in our devices. 41 With an absorbance of ~0.03 at Ephoton = 2.25 eV (for the time-resolved data in Fig. 5) and a maximum laser fluence of 210 µJcm -2 , we can estimate the initial maximum photogenerated electron-hole pair density to be in the order of ~1.75 ˑ 10 13 cm -2 . This charge carrier density is below the so-called Mott-threshold of an exciton-dissociation, 15 but in a regime where a renormalization of both the quasi-particle band gap and the exciton binding energy occurs. The renormalizations happen on timescales as fast as femtoseconds and prevail up to several hundreds of picoseconds after the photo-excitation. 15,[28][29][30][31][32] Generally, we observe that the time-resolved photocurrent signal Isampling as in Fig. 5 can be separated into four components: I1, I2, I3, and the offset with different characteristic timescales. The initial photocurrent response I1 (blue in Fig. 5) has a temporal Gaussian shape with a FWHM of ~3 ps, which exceeds the experimental resolution of ~1 ps of the utilized THz-time-domain circuit. 42,43 One possible explanation of this first contribution is the so-called photovoltaic effect, as it is termed in standard time-averaged photocurrent experiments. 2,8,10,12,13,16,17,21 On ultrafast timescales as in Fig. 5, this photovoltaic effect translates to both, the so-called transient displacement current jD and the lifetime-limited current jlifetime. The former is typically written as 44 with εr (ε0) the relative (vacuum) permittivity and / the ultrafast change of the local electric field E, when the laser pulse impinges on the sample. In principle, this change includes the 7 photogeneration of charge carriers as well as the renormalization of the quasi-particle band gap and the exciton binding energy. Importantly, jD allows to map-out the polarity of electric fields present in the investigated samples. For a positive (negative) bias voltage Usd, we observe that the sign of I1 is negative (positive) in the center of the sample (cf. top panel of Fig. 5e). Since I1 has an opposite sign at the two contacts, this ultrafast contribution corroborates the existence of builtin electric fields at the annealed metal contacts. The 'temporal duration' of jD is given by the dielectric response εr of the overall circuit, which is fundamentally limited by a phonon-excitation in the underlying sapphire substrate at ~12 THz; 43 i.e., significantly faster than the observed constant FWHM of ~3 ps across the whole device. This consideration brings the lifetime-limited current jlifetime into play 44 with eff the effective conductivity of the locally photo-excited electron bath and 1 a corresponding lifetime. From equation (2), we find the same sign for the built-in electric fields at the laser-annealed contacts as in the case of jD, as it is consistent with the sketch in Fig. 3d. The notation of eff implies that the locally excited electron bath is thermalized, as it can be assumed to occur at ~100s of femtoseconds. 32 At the utilized laser fluences of ~100 µJcm -2 and beyond, it has been demonstrated that the non-radiative relaxation and recombination lifetimes are only a few picoseconds. 32 In turn, there is not a clear timescale separation between jD and jlifetime, and we interpret the observed timescale of ~3ps with the non-radiative relaxation and recombination times 32,45 and fast capture mechanisms by traps via Auger processes 46 or surface defects. 47 Consistent with this interpretation, we observe that the FWHM of I1 increases with a lower laser fluence (data not shown). 32 Furthermore, we can neglect a photo-thermoelectric current for this 8 first contribution I1, as we do not detect it for a laser excitation below the band gap of MoS2 (cf. supplementary Fig. S4). In other words, within the given experimental noise, I1 seems to comprise all ultrafast non-radiative processes involving the non-equilibrium dynamics and energetics of charge carriers, which are photogenerated above band gap.
The second contribution I2 shows a fast rise followed by an exponential decay ∝ − / 2 with 2 ~ 100 ps (cf. orange contribution in Fig. 5a), pointing towards a lifetime-limited current as in equation (2). Since I2 is most pronounced at the contact regions (second panel in Fig. 5e), it also points towards the presence of electric fields as discussed for I1. In this respect, we associate I2 to photoexcited carriers having an extended lifetime 2. This interpretation agrees well with the timescale of an energy transfer between the electron bath and the phonon bath, 32 but also with carriers captured by defects via Auger processes and slow trap states 47 or photo-injected charge carriers in indirect-gap side valleys. 48 We note, however, that I2 can be detected for a laser excitation below band gap as well (cf. supplementary Fig. S4). Then, the signal is ~10-fold reduced but with an equivalent timescale. In this respect, I2 comprises a thermo-electric contribution, which is consistent with the fact that the heated electron bath can drive corresponding thermoelectric currents at the contacts' interfaces. 13 The third component I3 (green in Fig. 5), differs from the previous two contributions by the clearly Iphoto (cf. Fig. 4), we can estimate a maximum temperature increase of the overall device to be is the prevailing bolometric photoconductance with a cooling time in the order of the laser repetition time (~13.2 ns = 76 -1 MHz -1 ). We note that a heated phonon bath contributes two-fold to the bolometric photoconductance in TMDs; i.e by a combined change of the mobility and charge carrier density as is known for semiconductors (cf. Supplementary Fig. S5), and by a temperatureinduced change of the quasi-particle band gap. 33 Supplementary Fig. S3). The remaining silicon strip serves as an on-chip photodetector (Auston-switch) as we use for ultrafast photocurrent measurements. 42,43,52 In a second optical lithography step, we evaporate 10 nm/300 nm Ti/Au to define the contacts in form of two coplanar striplines. The striplines contact the monolayer MoS2 and the read-out of the Auston-switch. Each individual stripline has a width of 5 µm and a distance of 10 µm in-between two coplanar ones. The MoS2-flakes are placed at a distance of about 400 µm from the Auston-switch. The striplines have a total length exceeding 58 mm. The geometry allows for both time-integrated and time-resolved photocurrent measurements on individual MoS2-flakes.
Optical microscopy. All samples are characterized by optical microscopy to investigate the layerthickness and morphology of the individual MoS2-flakes. Fig. 1a depicts such an optical image of a MoS2-flake with varying layer thickness placed on top of a 10-nm-thick Ti contact (on a 525µm-thick p-type Si substrate with 285-nm SiO2 layer). The monolayer, trilayer, and bulk areas have different contrast and they are marked as 1L, 3L and bulk, respectively.

Laser-annealing.
We perform laser-annealing by continuously illuminating individual spots on the samples to locally introduce heat to the MoS2-flakes on Ti (Fig. 1b). For the presented results, we use a Kr/Ar-laser with a wavelength of 488 nm, which is focused with a 100x objective onto the sample stored at ~10 -3 mbar. However, we note that we achieve similar results with a Ti:Sapphire laser at 780 nm. In both cases, the laser spot is scanned across the samples by the help of a x-y-z-piezo scanner (PI) with a specific dose per laser spot position. Fig. 1c introduces a corresponding dose pattern as is applied to the MoS2-flake shown in Fig. 1a. At each of the depicted positions (gray dots in Fig. 1c), the laser is focused at a dose of 0.4 kJ/cm 2 forming a uniform grid with 1 µm spacing. Moreover, a dose gradient is superimposed to the pattern in the shape of a 'T-U-M' logo (shaded dark area in Fig. 1c). There, the dose ranges from ~10 4 to ~5×10 4 J/cm 2 . We adjust the laser dose by varying the dwell time of each spot illuminated with a power of ~1.5 mW. Fig. 2d illustrates the measured work function shift induced by laser-annealing as a function of the overall applied laser doses. The doses are again adjusted by varying the dwell time with a laser wavelength of 488 nm, a laser power of 1.5 mW, and a spot diameter of ~1 µm. The corresponding KPFM traces are extracted across a laser-annealed spot and fitted by a single Gaussian (cf. Fig. 1b). In Fig. 2d, the amplitude of the Gaussian is plotted as a measure for the maximum shift. We find that the work function shift seems to saturate at ~ -0.2 eV (red line in Fig.  2d).
Atomic force microscopy. All atomic force microscopy (AFM) measurements are performed in ambient conditions using a Bruker MultiMode 8 atomic force microscope by means of PeakForce frequency-modulation with a SiN cantilever and a Si tip with a triangular shape. 53 Fig. 1d depicts the AFM image of the MoS2-flake on top of the Ti contact after laser-annealing. In the topography profile of the sample, there are no indications from the laser-annealing step. Only small protruding clusters are visible on the MoS2-flake, which can be attributed to residual contaminations introduced by the viscoelastic stamping technique. 35 Kelvin probe force microscopy. Kelvin probe force microscopy (KPFM) measurements are carried out in ambient conditions using the same system as for AFM measurements in dual pass mode (lift height 50 nm). 53 Fig. 1e shows a KPFM map carried out simultaneously to the AFM results in Fig. 1d. Clearly, the 'T-U-M' pattern is visible in the KPFM signal of the MoS2-flake placed on the thin Ti film. We detect that the work function is shifted by up to ΔE ≈ -(0.1 -0.2) eV by the laser-annealing for the highest laser dose (Fig. 1e and lower panel in Fig. 1b). Noteworthy, we observe that the monolayer regions, which are not laser-annealed, show no difference in work function. This observation indicates that prior to laser-annealing, the van der Waals gap between the MoS2-flake and the Ti contact is rather large, causing no significant Fermi level pinning. 6,7 Given the fact that Ti is often used to promote the adhesion of Au contacts, the question arises if Ti or Au induces an additional work function shift of 1L MoS2. Therefore, we directly compare the Fermi level pinning of monolayer MoS2 on Ti and Au before and after a thermal annealing step, as described in the next section.  38 We are aware of possible changes due to oxidation of the Ti while exposed to ambient conditions, however, this can cause both negative and positive changes. 39 As we are focusing on the relative work function shifts, we neglect any offset of the absolute values caused by oxide formation. For MoS2 on Ti, we find then roughly -4.45 eV and on Au -4.25 eV, suggesting a change of the work function difference due to the different metals by ~0.2 eV. In average, these values agree well with the previously reported work function of monolayer MoS2 on top of p-type Si by KPFM measurements also in ambient conditions 37 . In a third step, the sample is thermally annealed for 45 min at 200 °C and in a vacuum of ~1 mbar. Figs. 2b and 2c show the corresponding KPFM traces after the thermal annealing step with the individual shift of the work functions indicated by red arrows. We find a significant difference in work function of the MoS2-flake before and after the annealing (red arrows Fig. 2b). Within the overlap region, the 10-nm-thick Ti on top of the Au contact seems to define the work function of MoS2. Fig. 2c shows the KPFM traces of the bare metals (along the dotted line (2) in Fig. 2a). For Ti, the work function increases slightly, while for Au the work function slightly decreases (indicated by arrows). We interpret the small changes to stem from the removal of lithography residues during the thermal annealing.
Raman and photoluminescence spectroscopy. Raman and photoluminescence spectroscopy are powerful tools to probe a variety of fundamental properties of MoS2 such as the number of 13 layers, 41,54 charge carrier density, 36 charge carrier lifetime, 45,47,48 temperature, 51 etc. We utilize Raman and photoluminescence spectroscopy to clearly identify the number of layers for the region of interest for all presented samples, and most importantly, we can exclude thermal degradation or an annealing induced phase transition (cf. Supplemental Fig. S1). 5 Temperature-dependent transport spectroscopy. We take current-voltage (IV)-characteristics on annealed two-terminal metal-MoS2-metal devices over a temperature range from 305 to 325 K (cf. Supplementary Fig. S5). In the devices, the Ti/Au-contacts (5/20 nm) cover the monolayer MoS2-flake from the top on a sapphire-substrate (cf. Supplemental Figs. S2 and S5). All IVmeasurements are performed using an Oxford Instruments Optistat setup to control the temperature at a pressure of ~10 -5 mbar. The source-drain current is recorded utilizing a current voltage amplifier at a gain of 10 7 (cf. Supplemental Fig. S5).
Scanning photocurrent spectroscopy. Scanning photocurrent measurements are performed utilizing a Ti-Sapphire laser together with a non-linear optical fiber to photoexcite the MoS2flakes. 2,8,10,12,13,16,17,21 The photon energy is set to be Ephoton = 2.25 eV at a pump fluence F = 210 µJ/cm 2 , a repetition rate 76 MHz, and a pulse duration ~1 ps. The samples are kept at room temperature and vacuum conditions at a pressure of 10 -5 mbar throughout the optoelectronic measurements. Before the presented photocurrent measurements, all samples are laser-annealed. We record the photocurrent signal by chopping the laser beam at a frequency of 1.7 kHz together with a lock-in detection. A current voltage amplifier is utilized to amplify the signal before lockin detection. Fig. 4a depicts an optical microscopy image of a monolayer MoS2-flake contacted by two metal striplines. For better visibility, the MoS2-flake is highlighted by solid white lines. For detecting photocurrent maps, the laser spot with diameter ~2.5 µm is scanned across the sample while the photocurrent response as well as the reflectivity are simultaneously recorded for each spot. Fig. 4b shows such a reflectivity map of the MoS2-flake, while Figs. 4c and 4d show the corresponding photocurrent maps for Usd = -5 V and +5 V. In the figures, the solid blue lines indicate the position of the MoS2-flake, while the dashed gray lines highlight the positions of the contacting metal striplines. In the electronic circuitry, the right stripline represents the source contact, while the left stripline is the drain contact connected to ground. For negative (positive) bias, we observe a pronounced negative (positive) photocurrent signal close to the right (left) metal stripline (cf. Figs. 4c and 4d). For better visualization, Fig. 4e shows a cross-section of the photocurrent signal Iphoto for both Usd = -5 V and +5 V (along the dashed-dotted line illustrated in Figs. 4b-d). The cross-section of the reflectivity is also shown with a maximum reflectivity at the striplines (gray lines). We clearly observe that the time-integrated photocurrent is mainly generated at the contact regions but there are also finite smaller contributions in the center region. Noteworthy, for zero bias Usd = 0 V, we observe a photocurrent response at both contacts with equal amplitude but opposite sign (data not shown). All time-averaged data are consistent with recent reports on the photoconductance and photocurrent phenomena of TMDs. 2,8,10,12,13,16,17,21 Ultrafast photocurrent spectroscopy. For measuring the ultrafast optoelectronic dynamics of the monolayer MoS2-flakes, we use an on-chip THz-time domain photocurrent detection scheme based on coplanar striplines 34,42,43,52,55,56 (cf. Supplemental Fig. S3). A short laser pulse with Ephoton = 2.25 eV at a pump fluence F = 210 µJ/cm 2 , a repetition rate 76 MHz, and a pulse duration of < 1ps is used to excite charge carriers in the MoS2. The corresponding photocurrent response in the MoS2-flake gives rise to electromagnetic transients in the striplines, which propagate along the striplines. We utilize a time-delayed laser pulse in combination with an Auston-switch 44 as an on- 14 chip read-out of the electromagnetic transients. The probe laser pulse has a photon energy of 1.59 eV and a temporal width of 100 fs. The current Isampling across the Auston-switch samples the electromagnetic transients on the striplines as a function of the time-delay Δt between the two laser pulses. It is directly proportional to the ultrafast photocurrents in MoS2. Hereby, the photocurrents in MoS2 can be measured with a (sub-) picosecond time-resolution 43 . Fig. 5 depicts the timeresolved photocurrent Isampling(Δt) of the monolayer MoS2 excited at the drain contact (Figs. a and  b) and at the center of the channel (Figs. c and d) on short (0-250 ps) and long (0-1600 ps) timescales. We can consistently fit the time-resolved data Isampling with four components: an initial ultrafast response I1 with a Gaussian lineshape and a FWHM ~ 3 ps (blue in Fig. 5), a second response I2 as a Gaussian convoluted decay-function with a decay time 2 (orange in Fig. 5), a third comparably long lasting response I3 with a rise time 3 rise and a decay time 3 decay (green in Fig. 5) and an offset. The offset is particularly observable for Δt < 0; shortly before the pump laser hits the MoS2-flake in the repetitive scheme with a repetition time of 76 MHz -1 ~13.2 ns.

Contact morphology and revisited photocurrent dynamics in monolayer MoS2
Eric with a linear bolometric temperature coefficient and Δ = − 0 the temperature difference with respect to T0 = 305 K. The analysis yields annealed = (+0.011 ± 0.001) K −1 after the laserannealing and prior = (−0.018 ± 0.002) K −1 prior to the annealing. We note that after annealing (Supplemental Fig. S5a), the rather symmetric IV characteristics can be understood in terms of a metal-semiconductor-metal device, in which the semiconducting temperature dependence of the monolayer MoS2 dominates with a positive bolometric coefficient annealed. In contrast, the as-fabricated device (i.e. without annealing) shows rather asymmetric IV traces together with a factor ~10 higher overall device resistance (Supplemental Fig. S5b). We attribute both findings to a poor contact interface between the monolayer MoS2 and at least one of the metal leads, resulting in irregular charge injection properties. Microscopically, this can be related to residuals at the interface, resulting in a larger van der Waals gap as discussed before 6 . Moreover, we find a negative bolometric coefficient prior prior to annealing (Supplemental Fig. S5d), which we can explain, if we consider that the overall device characteristic is governed by the metal leads and the poor contacts, rather than by the monolayer MoS2.   Fig. 5e). The fits to the tree distinct contributions include I1, I2 (orange) and I3 (green) are highlighted in color, although we note that I1 is not resolvable within the given noise; i.e. we cannot detect I1 for an excitation below the band gap of MoS2.