On the Contact Optimization of ALD-Based MoS2 FETs: Correlation of Processing Conditions and Interface Chemistry with Device Electrical Performance.

Despite the extensive ongoing research on MoS2 field effect transistors (FETs), the key role of device processing conditions in the chemistry involved at the metal-to-MoS2 interface and their influence on the electrical performance are often overlooked. In addition, the majority of reports on MoS2 contacts are based on exfoliated MoS2, whereas synthetic films are even more susceptible to the changes made in device processing conditions. In this paper, working FETs with atomic layer deposition (ALD)-based MoS2 films and Ti/Au contacts are demonstrated, using current-voltage (I-V) characterization. In pursuit of optimizing the contacts, high-vacuum thermal annealing as well as O2/Ar plasma cleaning treatments are introduced, and their influence on the electrical performance is studied. The electrical findings are linked to the interface chemistry through X-ray photoelectron spectroscopy (XPS) and scanning transmission electron microscopy (STEM) analyses. XPS evaluation reveals that the concentration of organic residues on the MoS2 surface, as a result of resist usage during the device processing, is significant. Removal of these contaminations with O2/Ar plasma changes the MoS2 chemical state and enhances the MoS2 electrical properties. Based on the STEM analysis, the observed progress in the device electrical characteristics could also be associated with the formation of a continuous TiS x layer at the Ti-to-MoS2 interface. Scaling down the Ti interlayer thickness and replacing it with Cr is found to be beneficial as well, leading to further device performance advancements. Our findings are of value for attaining optimal contacts to synthetic MoS2 films.

Introduction

Atomically thin, all-surface, two-dimensional (2D) transition metal dichalcogenides (TMDCs) with electrical properties different from bulk materials have attracted significant interest in recent years for next generation nanoelectronic device schemes, beyond 10 nm complementary metal oxide semiconductor (CMOS) technology nodes.[1−3] TMDCs constitute a wide library of compounds, exhibiting diverse properties, ranging from semiconductors to (semi)metals and superconductors.[4,5]Among semiconducting TMDCs, MoS2 is the most extensively studied material for electronic circuits and switches, specifically in the context of field effect transistors (FETs), owing to its outstanding electrical properties including a low subthreshold swing (∼60 mV/dec),[6]Ion/Ioff ratios exceeding 10,7[7−9] and reasonably high mobility values.[7,10] High degree of mechanical stability,[11,12] low variability,[13] and high reliability[8] as well as compatibility with conventional silicon CMOS fabrication process flows are other notable features of this material.[14−16] However, paving the way toward the implementation of 2D TMDCs as emerging materials into the mainstream CMOS platforms is not as straightforward as it appears to be. There are still several challenges that need to be addressed.

One of the critical issues is that 2D TMDC FETs, in general, are Schottky barrier transistors.[17] In other words, a potential barrier against the efficient carrier injection from the metal contacts toward the TMDC channel (due to their work function misalignments) is invariably formed, resulting in current rectification and high contact resistance (Rc) values.[18,19] In traditional bulk semiconductors, this issue is successfully resolved by using ion implantation, to heavily dope the regions below or around the contacts.[20] As a result, the metal work function is aligned with the semiconductor conduction band minimum (Ec), such that the carrier transport toward the channel is facilitated with contacts becoming transparent (Ohmic). However, despite the ongoing research,[21,22] no such reliable and controllable doping technology for the 2D TMDC-based FETs, without damaging or perturbing the 2D layer properties, has been fully established yet.

In addition to the work function misalignment between the 2D TMDC and the contacts, the so-called Fermi levelpinning (FLP) phenomenon is the other contributor to the Schottky barrier and the high Rc.[23] Regardless of the metal choice and its work function value, it has often been observed that the MoS2 FETs exhibit an n-type behavior, disobeying the Schottky–Mott model, where the barrier height for the electron injection is determined by the difference between the metal work function and the semiconductor electron affinity level.[24,25] The n-type observations are ascribed to the FLP at the metal-to-MoS2 interface due to the combination of factors, including metal work function modulations as well as defect-/metal-induced gap state formations, which are most often energetically distributed in close proximity to MoS2Ec.[26] The origin of the gap states is strongly processing-related. These states could be intrinsic and formed at the time of layer growth (in the case of synthesized films)[27,28] or they could be induced during the device fabrication (e.g., when depositing metal contacts and when using lithographic processes for patterning).[29]

The majority of metals reacts with MoS2, when brought into contact.[30] According to theoretical calculations, upon this interaction, the MoS2 Fermi level (EF) position shifts away from its intrinsic level, denoting that MoS2 becomes doped at the contact regions.[31] Inert metals, such as Au, exhibit weak van der Waals (vdW) interactions, whereas more chemically reactive metals (such as Ti and Cr) form covalent bonds with the underlying MoS2 and disrupt the MoS2 electronic band structure.[31,32] In case the formation energy of metal-to-sulfur (Ti–S, Cr–S) is less than the formation energy of molybdenum-to-sulfur (Mo–S),[33] a sulfur-deficient MoS2 forms at the interface upon metal deposition.[26] Apart from the metal choice, it has been experimentally observed that the chemical compound formed at the metal-to-MoS2 interface strongly depends on the metal deposition conditions.[34,35] For example, McDonnell et al.[35] have demonstrated that when a Ti contact is deposited on MoS2 at high vacuum (HV) (pressure >10–6 mbar), Ti is likely to react with oxygen and form TiO at the interface, owing to the background oxygen present at such pressures. On the other hand, for ultrahigh vacuum (UHV) environments (pressure <10–9 mbar), Ti shows a stronger reaction with the underlying MoS2, resulting in the formation of Ti–S bonds at the interface[35] and disruption of the semiconducting layer beneath as well as its electronic band structure,[31,36] leaving behind gap states mainly of Mo character.[26,31,36] Several efforts have been made in recent years to control the gap state formation. Insertion of ultra thin interfacial layers of oxides such as Ta2O5,[37] Al2O3,[38] and TiO2[38,39] at the interface was overall beneficial in mitigating the impact of the deposited metal on the MoS2 layer beneath. This has in most of the cases led to Fermi level de-pinning, Schottky barrier height lowering and Rc reduction for the MoS2 FETs. Similarly, when growing the Ti metal in a HV environment, the inevitable formation of interfacial TiO, as a barrier layer against the direct metal interaction with MoS2, has positively influenced the transport properties.[29] These studies altogether indicate that connecting the interface chemical reactions with the processing conditions is vital for appropriately understanding the electrical behavior of the MoS2 FETs, a perspective that should not be overlooked or underestimated.

It is also commonly known that the process of device fabrication can itself induce additional impurities and defects in the contact areas. In a standard device fabrication flow, where the contacts are defined by means of lithographic and patterning processes, the opened areas intended for the contacts may not always be fully cleaned from the resist residues. The presence of these contaminants can subsequently lead to low-quality contacts.[29] In conventional CMOS process flows, a plasma cleaning step is usually followed after the development step to ensure a complete resist removal.[20] This cleaning treatment is reckoned to be a harsh process for MoS2 films, often creating additional defects or forming other compounds on their surface.[40] Nonetheless, it has been reported that a mild and short O2 plasma cleaning step can effectively eliminate the resist residues without damaging the MoS2 lattice, leading to noninterrupted contacts, and thus improved electrical performance.[29] H2 plasma and a subsequent usage of carbonyl disulfide have also been shown to clean and restore the 2D film properties.[41]

High-temperature post deposition annealing is another standard technique in industry for mitigating the impurities and defects at the metal-to-semiconductor interfaces.[20] However, for MoS2 FETs, this treatment resulted in contradictory outcomes. Some groups observed performance enhancements,[42,43] whereas others reported deteriorations in electrical features.[7,44] These inconsistencies originate from the chemical interactions at the metal-to-MoS2 interface that can create/annihilate defects, re-emphasizing that the MoS2 FETs are defect-mediated systems. Therefore, the relation between the processing conditions, the interface chemistry, and the device electrical performance has to be independently investigated for every device fabrication process flow.

It is also noteworthy that the majority of the abovementioned studies have been conducted on exfoliated MoS2, whereas the role of intrinsic and process-induced defects in transport properties at the contact sites can be even more pronounced for synthetic MoS2 films.

In this work, atomic layer deposition (ALD) is employed for the growth of MoS2 films. Although chemical vapor deposition (CVD) is still the preferred synthesizing technique to achieve the best quality 2D TMDCs in large areas,[10,45,46] in the quest of Ångstrøm level thickness control and highly conformal layers (suitable for 3D structures) while maintaining a low thermal budget, required for industry processes, ALD has also been gaining attention in recent years.[47−50] We comprehensively study the impact of device processing conditions on the chemistry involved at the metal-to-MoS2 interface and connect the two to the current–voltage (I–V) results, obtained from electrically characterizing the fabricated FETs, with ALD-based MoS2 films and Ti/Au contacts. In pursuit of optimizing the contacts to our MoS2, we investigate the influence of thermal annealing and plasma cleaning the contacts (with O2 and Ar) on the overall device electrical performance while scaling down the Ti interlayer thickness. We relate the observed I–V characteristics to the interface chemistry, using X-ray photoelectron spectroscopy (XPS) and cross-sectional scanning transmission electron microscopy (STEM) analyses. Furthermore, the impact of changing the contact metal type will be discussed. Our findings shed light on the path toward a better understanding of the interface chemistry as well as the carrier transport in ALD-based MoS2 FETs and their strong reliance on the device processing conditions.

Experimental Section

MoS2 Film Synthesis

For this study, multilayer MoS2 films with a thickness of about 3.6 nm (7–8 monolayers) were synthesized using a two-step approach, where plasma-enhanced ALD was employed for the large-area and thickness-controlled growth of MoO on a commercial ALD reactor from Oxford instruments (FlexAL). Then, the as-deposited MoO films were annealed in a H2S environment, with 10% H2S/90% Ar, at 900 °C, for 45 min, to eventually obtain multilayer MoS2. Further details of the film synthesis are published in our previous work by Sharma et al.[51] In all cases, the growth took place on degenerately doped (p++) silicon substrates with ∼285 nm thermal SiO2, acting as global back-gates for the fabricated MoS2 FETs.

Fabrication of Back-Gate MoS2 FETs

The back-gate MoS2 FETs were fabricated by the common electron beam lithography (EBL) patterning technique. For this purpose, a polymer resist (950K PMMA A4) was spin-coated on the as-synthesized MoS2 films, with a spin speed of 4000 rpm, for 60 s. After baking poly(methyl methacrylate) (PMMA) at 180 °C, for 5 min, a contact layout design was transferred on it by an EBL exposure, using a commercial RAITH micrograph system (EBPG5150). Upon developing the patterned PMMA with an organic developer [methyl isobutyl ketone (MIBK)/isopropyl alcohol (IPA) mixture with a 1:3 ratio], the contact windows (source and drain regions) were opened. The development step was then followed by the deposition of 20/80 nm of Ti/Au on MoS2, using an e-beam evaporator, operated at room temperature, with a base pressure of ∼4 × 10–7 mbar and a deposition rate of 1 Å/s. Subsequently, contacts were delineated by a lift-off process in an acetone solvent, where the PMMA together with the deposited metal were removed from the nonexposed regions. For defining the MoS2 channel regions and isolating the individual MoS2 devices, a second lithography step (to transfer a relevant overlay pattern onto the samples) was employed. After this exposure and the PMMA development, MoS2 was dry-etched from the opened areas, in a reactive ion etching (RIE) reactor (GP-RIE, Oxford Instruments), using SF6/O2 plasma at room temperature, with a flow rate of 16/4 sccm, for 20 s, at a pressure of 22.5 mTorr and a forward power of 25 W. Then, the PMMA layer (which was protecting the channel regions during the plasma etching process) was removed in acetone. Finally, 30 nm of HfO (high-κ) was grown on the fabricated MoS2 FETs, using PE-ALD[52] at 100 °C (FlexAl ALD reactor, Oxford Instruments). The HfO deposition is mainly for capping the fabricated devices and screening the charge impurity scatterings as one of the major mobility-limited mechanisms for MoS2 channels.[53] The optical microscopic top view image of the as-fabricated MoS2 FETs and the final device schematic cross section are provided in our Supporting Information (see section S.1).

Annealing the MoS2 Contacts

Investigating the impact of annealing the contacts on the electrical performance of the ALD-based MoS2 FETs, the fabricated devices were vacuum annealed at 300 °C, for 2 h, at a base pressure of ∼7 × 10–6 Torr (HV), in the ALD reactor and prior to the HfO deposition step.

Plasma Cleaning the Contact Openings on MoS2

To efficiently remove the possibly existing organic residues from the surface of our MoS2 films, a mild plasma cleaning step right before the contact deposition step with two different gases (O2 and Ar) was examined. The plasma cleaning conditions were kept similar for both gases, being introduced into the chamber with a flow rate of 20 sccm, for 5 s, at a pressure of 200 mTorr and using a plasma power of 50 W.

Scaling Down the Ti Interlayer Thickness

Evaluating the influence of the Ti interlayer on the overall device electrical performance, various Ti thicknesses in the range of 20 nm down to 2.5 nm were examined while retaining the whole Ti/Au contact stack thickness at 100 nm. A complete Ti removal was also inspected. For all the cases, the evaporation conditions were constant (a base pressure of ∼4 × 10–7 mbar and a deposition rate of 1 Å/s).

Replacement of Ti/Au with Cr/Au

For further optimizing the contacts to our MoS2 films, the Ti interlayer was replaced with Cr. The Cr evaporation took place in the same chamber and under the similar conditions to those of the Ti case (a base pressure of ∼4 × 10–7 mbar and a deposition rate of 1 Å/s).

Electrical Characterization

The electrical performance of the fabricated MoS2 FETs was evaluated by characterizing their room temperature I–V response. In all the cases, the measurements were conducted on 500 nm long and 1 μm wide MoS2 channels. The setup used for this purpose was a cryogenic probe station (Janis ST-500) with a base pressure of ∼1.9 × 10–4 mbar, connected to a Keithley 4200-SCS parameter analyzer.

Surface Characterization

Correlating the electrical characterizations with the surface chemistry, ex situ XPS was performed on the MoS2 surface after its synthesis, contact pattern development, and plasma cleaning treatments, with both Ar and O2 gases. For this purpose, a Thermo Scientific K-alpha KA1066 spectrometer (Thermo Fisher Scientific, Waltham, MA) with a monochromatic Al Kα X-ray radiation source (hν = 1486.6 eV) was utilized. The measurements were carried out with an X-ray beam spot size of 200 μm, at a take-off angle of 60° and a pass energy of 50 eV, combined with an electron flood gun, to efficiently neutralize the existing charges on the samples as well as to correct for the nonuniform and differential charging. The acquired spectra were later chemically quantified and deconvoluted with Avantage software. All the peaks were also further binding energy calibrated with respect to the C 1s adventitious carbon peak (284.8 eV).

Structural Characterization

The microstructure of the metal-to-MoS2 interface was investigated using cross-sectional STEM analysis, conducted using a JEOL ARM 200F operated at 200 kV. The sample that received O2 plasma cleaning treatment, prior to the contact deposition, was selected for this evaluation. The common lift-out procedure with focused ion beam (FIB) was followed to prepare an imaging specimen on a Mo support grid. A Mo grid was chosen rather than a Cu grid, as previous studies have shown that redeposited Cu by the FIB process can react with the sulfur in MoS2, creating undesirable CuS precipitates on the transmission electron microscopy (TEM) specimen surface. Prior to the so-called FIB milling step, the contact stack was covered with a protective layer of SiO2 and Pt, using electron beam-induced deposition (EBID) and ion beam-induced deposition (IBID), respectively.

Results and Discussion

Throughout the Results and Discussion section, the reference sample is represented by the ALD-based MoS2 FET with a 3.6 nm thick, 1 μm wide, and 500 nm long MoS2 channel and 20/80 nm of Ti/Au contacts (no plasma cleaning of the contact areas or post annealing treatments). The device shows typical transfer curves (IDS–VGS) with dual sweep (Figure a) and output curves (IDS–VDS) (Figure b). As can be seen from Figure a, the fabricated device exhibits n-type characteristics with the maximum drain current (Ion) reaching 180 nA/μm, at VGS = 150 V and the highest applied lateral field of VDS = 5 V. The minimum current (Ioff) (at VGS = −150 V) is few nA/μm for all the VDS values (see the semilog plot shown as the inset) and does not decrease monotonically with more negative VGS. The latter is due to the dielectric encapsulation effect and suppression of hole contribution, which is observed in other related studies as well.[54,55] In addition, the hysteresis and, hence, the threshold voltage (VT) of both backward and forward sweeps show strong dependence on the applied VDS (apart from the their general dependence on VGS). All these observations indicate that more than one phenomenon is contributing to the MoS2 charge transport. Some of the causes can be listed as follows: defect states at the contact-to-MoS2 interface, MoS2 structural defects (grain boundaries and dislocations) and defect states/charged trapping sites at the dielectric-to-MoS2 interface as well as high Rc.

Figure 1

(a) Double sweep transfer curves of the ALD-based MoS2 FET, for different lateral fields, on a linear scale (the red arrows indicate the direction of forward and backward sweeps). The inset shows a semilog plot, for better visualizing the current change with VDS in the off-state regime. (b) Forward sweep output IDS–VDS characteristics for different VGS values.

Hysteresis usually increases with VGS and is weakly affected by VDS. However, in our study, the hysteresis increases with VDS, indicating a profound influence of defects and trapping of carriers inside the MoS2 lattice. The nonlinearity in the output characteristics (shown in Figure b) is also due to Schottky barrier formation between MoS2 and the contacts. Recent reports from several researchers have shown that all the abovementioned effects are responsible for VT dependence on VDS.[8,29,56−58] Therefore, further studies are followed to tailor the contacts and to investigate the influence of different processing conditions, with the aim of improving the device electrical performance.

Annealing the ALD–MoS2 Contacts

As mentioned earlier, annealing the contacts is a common practice in conventional CMOS process flows to improve the electrical performance, through reducing the density of defects at the metal-to-semiconductor interfaces and eliminating the surface charges from the semiconducting channel regions.[20] However, inclusion of this step and its impact on the overall device electrical characteristics have not yet been studied for ALD-based MoS2 FETs with Ti/Au contacts. To investigate that, the fabricated devices were thermally HV annealed at 300 °C, for 2 h and prior to the HfO (high-κ dielectric) deposition. Figure a compares the transfer curves of a nonannealed (reference) sample versus the annealed one at VDS = 5 V, on a semilog scale. For both cases, the Ti/Au contacts were 20/80 nm thick. As can be seen from the plot, upon the HV annealing treatment, Ion does not improve, whereas Ioff slightly drops. In addition, VT shifts from negative to positive values for both back and forward sweeps. The latter is more evident on a linear scale, provided as the inset of Figure a. These observations certainly imply that when the HV annealing treatment is employed, the chemical state at the Ti-to-MoS2 interface changes, but not in favor of improving the device on-state performance. The reduction in Ion could be explained by the interdiffusion of the elemental compounds (Ti, Mo and S), which is facilitated at elevated temperatures. As a result, a disordered and an intermixed region would be formed at the Ti-to-MoS2 interface, as has previously been verified for the annealed Ti contacts to the exfoliated MoS2 films.[59] Furthermore, the annealing process in this study is performed in a HV chamber. The presence of background water and/or oxygen and its possible reaction with the MoS2 surface at the channel regions, as well as the Ti-to-MoS2 interfaces (from the Ti edge sites), may lead to the formation of oxide-containing compounds, all of which could impede the enhancements in the device on-state.

Figure 2

(a) Transfer curves of the annealed and the nonannealed cases, on a semilog scale (the inset shows the plot with a linear scale, and the green arrow indicates the positive shift of VT upon the annealing treatment). (b) Forward sweep VT and its dependence on VDS for both the annealed and the nonannealed devices.

Further elaborating the discussion, VT values of the reference and the annealed samples are extracted and compared in Figure b. The data are derived using the Y-function method[60] (See the Supporting Information, Section S.2 for details of the extraction method). Considering the reference sample, modulation of the drain current (by changing VDS) results in shifting VT toward more negative values (from −8 to −37 V). As mentioned before, this indicates that the defect states (at the contact-to-MoS2 and/or the oxide-to-MoS2 interface) dope MoS2 to the n-type and contribute to the current. However, the annealed device shows a completely reversed behavior. As VDS increases, VT shifts to more positive values (from +25 V to +35 V). The positive shift of VT with increasing VDS together with the observed reduction in Ioff suggest that the defect density close to Ec of MoS2 decreases and that MoS2 becomes relatively less n-type doped upon the annealing treatment. Moreover, the inset of Figure a shows that the hysteresis reduces after this process, inferring the defect density mitigation at the oxide-to-MoS2 interface. All these observations are pointing toward a direction that a portion of the conduction in the MoS2 FETs is carried out by the defect states (as formerly also suggested by McDonnell et al.[61]), such that by annealing the contacts, the influence of these states become quite less pronounced.

Figure b also reveals that for all the VDS values, the overdrive voltage (Vov = VGS – VT) does not drop below 100 V (at VDS = 5 V, Vov = 150 V – 35 V (annealed case) or Vov = 150V + 37 V (reference case)). Therefore, our devices work in the linear operating regime.

O2 and Ar Plasma Cleaning of the Contact Openings

Further optimizing the contacts for our MoS2 FETs, the effect of plasma cleaning, prior to the metal contact deposition, is assessed in this Section. As discussed earlier, upon using PMMA (or any other polymer resist) for patterning the contacts with EBL, traces of the resist may contaminate the opened areas, subsequently causing formation of discontinuous contacts to MoS2 and affecting the overall device electrical performance. Marinov et al.[41] have recently reported that H2 plasma treatment at 300 °C is highly effective in clearing their synthetic WS2 surface from organic residues. They exclusively highlighted that the cleaning process has to be performed at sufficiently high temperatures to impose the least damage to the 2D lattice. In a former effort, Bolshakov et al. also suggested inclusion of a room temperature plasma step, to clean up the surface of their exfoliated MoS2, prior to the contact deposition.[29] Therefore, to efficiently remove the possibly existing residual components from the surface of our ALD-based MoS2 films, we examined a mild and short plasma cleaning step at room temperature, right before the contact deposition, with O2 and Ar plasma sources, which are regularly employed in microfabrication. The plasma parameters (time, flow rate, power, and pressure) were initially optimized on a blanket MoS2 film for the O2 gas source. Then, to allow for direct comparison, these conditions were kept similar for both O2 and Ar. Here, we also note that performing the cleaning process at 300 °C (as suggested by Marinov et al.[41]) is not applicable in our study because PMMA evaporates at such high temperatures, while it needs to remain for the contact delineation step. After the plasma exposures, Ti/Au contacts were deposited with a similar thickness and evaporation conditions to those of the reference sample.

I–V Analysis

Figure a demonstrates the transfer characteristics of the samples that have been treated with the O2 and Ar plasmas and compares their behavior with respect to the uncleaned (reference) sample, at VDS = 5 V. As can be seen, employing both the O2 and Ar plasmas, prior to the contact deposition, has been effective in improving the on-state performance, with increasing Ion up to one order of magnitude and reaching nearly 1.5–2 μA/μm. Hence, the average maximum field effect (FE) mobility, extracted from measuring various devices (typically three to four devices), has improved six to eight times (Figure b). See the Supporting Information, Section S.3 for the mobility extraction technique. Despite the significant enhancement in Ion as well as the mobility, Ioff increases, and VT shifts more negatively (from −37 V for the reference case to −42 V for the O2 plasma case and to −70 V for the Ar plasma case), both indicating that the MoS2 channel is degenerately n-type doped after the plasma cleaning processes. As a consequence of this excess carrier concentration, the back-gate oxide is unable to fully deplete the channel and fails to efficiently modulate the drain current. However, the increase in Ioff for the O2 plasma case is less severe than when the Ar plasma-treated sample is considered. This suggests different chemical interactions at the metal-to-MoS2 interface, based on the plasma being utilized.

Figure 3

(a) Impact of the O2 and Ar plasma cleaning treatments (prior to the contact deposition) on the device electrical performance. (b) Average extracted maximum FE mobility values for all the three cases of study.

XPS Analysis

Bridging the I–V characterization to the chemistry involved at the metal-to-MoS2 interface, ex-situ XPS analysis was performed on the MoS2 surface, after its synthesis, contact pattern development, and cleaning treatments, with both the O2 and Ar plasmas. Figure shows the acquired elemental data from the XPS measurements for the core level spectra of the Mo 3d, S 2p, O 1s and C 1s, binding energy calibrated with respect to the C 1s sp3 component of the adventitious peak, set to 284.8 eV. The very bottom panel in each plot corresponds to the as-synthesized MoS2 film, prior to the EBL patterning and any processing treatments. The rest of the panels are offset vertically. After fitting and deconvoluting the peaks, the Mo 3d core level spectrum of the as-synthesized MoS2 shows two major peaks at 229.2 and 232.4 eV, representing the Mo 3d5/2 and Mo 3d3/2 binding energies for the Mo4+ oxidation state (MoS2), respectively. Two minor peaks with a slight shift to relatively lower binding energies at 228.9 and 232.1 eV are also discernible. These peaks are attributed to lower oxidation states, Mo (x = 2+, 3+), suggesting that our MoS2 film is substoichiometric/sulfur-deficient, intrinsic to its synthesis route.[62] Furthermore, as a result of the S 2s core level overlapping with the Mo 3d spectrum, an additional peak at 226.3 eV is discernible. Minor Mo components with oxidation states other than 4+ are also detected at higher binding energies, inferring that the surface of our synthetic MoS2 film contains a slight amount of oxidized species (MoO), upon its exposure to the ambient air. Evaluating the S 2p spectrum, signatures of the S 2p3/2 and S 2p1/2 core levels are detected at 162.0 and 163.2 eV, respectively. In addition, an extra peak is found at higher binding energies (171.8 eV) related to the S–O compound formation, reflecting MoS2 surface oxidation. As far as the O 1s and C 1s core level spectra are concerned, no specific chemical state, other than the adventitious oxygen and carbon, in the forms of C–O, O–H and C–C, can be distinguished. For better viewing these peaks, they are separately provided in the Supporting Information, Section S.4.

Figure 4

Mo 3d, S 2p, O 1s, and C 1s core level spectra of the as-synthesized MoS2, after the contact pattern development and after the different plasma cleaning treatments (offset vertically).

After defining the contact areas with EBL and PMMA development in MIBK/IPA solution, the binding energies for both the Mo 3d and S 2p peaks shift +0.1 eV, indicating that the MoS2 chemical state might have been changed and that its EF position displaced +0.1 eV closer to Ec, upon the development process.[29] In addition, the O 1s and C 1s spectra clearly demonstrate the presence of new organic compounds of C–O and O–H nature, with binding energies located at 531.8 eV in the O 1s spectrum and at 286.4 and 288.5 eV in the C 1s spectrum, respectively. As can be seen, the presence of these residual species on the MoS2 surface leads to a significant peak intensity drop in the Mo 3d and S 2p core level spectra, relative to the as-synthesized case. For better visualization, the as-mentioned spectra are separately provided in the Supporting Information, see Section S.5. The observed shift in the MoS2 binding energies together with the detection of organic compounds on the MoS2 surface comprehensively pinpoint that after the development step, residues of the resist material (PMMA) are still present on the MoS2 surface. Therefore, utilization of the developer solution (MIBK/IPA) alone may not be sufficient to completely remove the PMMA from the opened areas on the MoS2 surface. This non-negligible concentration of organic impurities could be considered as one of the main sources of changes in the MoS2 chemical state and its EF position.[29]

Applying a mild and short O2 plasma, the peak intensities in both the C 1s and O 1s spectra drop, and their full width at half-maximum (FWHM) increase, compared with the former case, where the PMMA residual components were still considerable on the MoS2 surface. Hence, the cleaning process is effective in removing at least a part of the residues from the MoS2 surface, leading to peak intensity enhancements in both the Mo 3d and S 2p core level spectra. The binding energies for both the Mo 3d and S 2p major peaks are also shifted +0.5 eV after the O2 plasma treatment, implying that the MoS2EF position is shifted +0.5 eV closer to Ec as well (relative to the as-developed case). In addition to the EF displacement, an extra doublet appears in the Mo 3d core level spectrum with peak positions of 232.4 and 235.7 eV, attributed to the Mo6+ oxidation state (MoO3). Meanwhile, in the S 2p spectrum, fingerprints of other oxidized species (e.g. MoSO) become more pronounced,[35] with peak positions located at 162.4 and 163.5 eV. All these findings indicate that the surface of MoS2 is slightly oxidized upon the O2 plasma cleaning process.

Finally, Ar plasma is examined for removing the organic residues. Upon this treatment, the peak intensities pertaining to the residual organics in the O 1s and C 1s spectra further reduce, and their FWHM increase accordingly. This comparison is with respect to the previous case of study, implying that the application of Ar plasma is more efficient than that of O2 plasma in cleaning the MoS2 surface from the resist contaminations. As far as the Mo 3d and S 2p core levels are concerned, the peak positions in both spectra shift +0.2 eV, relative to the as-developed case, as an indication of the EF shift toward Ec with the same magnitude. Unlike the O2 plasma treatment, in the Mo 3d spectrum, two minor peaks corresponding to lower oxidation states [Mo (x = 0, 1+)] appear upon the surface exposure to the Ar plasma. The presence of these oxidation states can be attributed to the preferential sputtering of sulfur when Ar plasma is employed, leaving behind Mo states of mostly metallic nature. Sulfur sputtering is also evident when considering the S 2p core level spectrum and the decreased binding energies for the minor peaks, as a sign of sulfur-deficient MoS species.[35] On the other side of the Mo 3d spectrum (higher binding energies), minor components of the Mo5+ state occur as well. This indicates that the sulfur-deficient MoS2 surface is slightly oxidized, likely due to the background oxygen present inside the chamber during the Ar plasma cleaning or the sample exposure to the ambient air after the process.

STEM Analysis

To further investigate the chemical species formed at the metal-to-MoS2 interface and its influence on the device electrical performance, a cross-sectional STEM analysis was performed for the MoS2 FETs that received O2 plasma cleaning treatment, prior to the metal deposition. Figure a shows a high-angle annular dark field (HAADF)–STEM image of the Au/Ti/MoS2 contact stack. At the interface between Ti and MoS2, an extra interlayer with a different contrast than Ti and MoS2 can be clearly distinguished. A higher-resolution STEM image, obtained in annular bright field (ABF) mode, provided in Figure b, reveals that this interlayer is amorphous and evenly distributed along the interface with a thickness of about 2 nm. Because the MoS2 surface is initially exposed to the O2 plasma, it is expected that organic residues do not play a significant role in the formation of such an interlayer.

Figure 5

(a) Cross-sectional HAADF STEM image of the O2 plasma-treated MoS2 contact stack, (b) higher-resolution ABF–STEM image of the contact stack, to better visualize the interlayer between the MoS2 and Ti (the yellow dashed lines determine the interlayer region), (c) HAADF image displaying the location of the EDX–STEM line scan, highlighted by the purple line and (d) compositional profile along the distance of the EDX line scan.

For exploring the compositional distribution at the MoS2-to-metal interface, an energy-dispersive X-ray (EDX)–STEM analysis has been carried out along the stack, as indicated by the purple line in Figure c. The acquired data, plotted in Figure d, elucidate that the sulfur concentration is still considerable at the interface between MoS2 and Ti, unlike the oxygen intensity that drops significantly beyond the SiO2 substrate and remains only as the background. The latter is further clarified by providing the raw peak intensities in the Supporting Information (see Section S.6 and the associated discussion). The Mo profile, on the other hand, does not fall to zero, beyond the MoS2-to-Ti interface. This is mainly because the sample is mounted on a Mo support grid. Therefore, the Mo background signal is unavoidably counted over the full range of the line scan. Altogether, the acquired elemental profiles suggest that TiS is the most likely formed chemical compound on the O2 plasma-cleaned MoS2 surface, rather than TiO. Now, considering the vacuum pressure during the Ti/Au stack deposition in the e-beam evaporator, which is ∼4 × 10–7 mbar, and the key role of the background oxygen in the chemical interactions at the metal-to-MoS2 interface (discussed earlier), one can deduce that under near-UHV conditions, formation of TiO is less probable. As a result, Ti tends to mostly react with S atoms of MoS2,[35,36] forming an amorphous interlayer of about 2 nm in thickness. It is also important to point out that the formation of MoO3 species during the O2 plasma cleaning treatment and before the contact deposition (based on the XPS analysis) appears to be negligibly influencing the interface interactions because only a limited amount of oxygen (less than a monolayer) is detected by the XPS.

Linking XPS and STEM to I–V

Re-evaluating the measured I–V data and the extracted FE mobility (shown in Figure a,b), one could comprehensibly correlate the observed electrical behaviors with the MoS2 surface chemical modifications, imposed by the plasma cleaning treatments. As concluded from the XPS analysis, both the Ar- and O2 plasma cleaning treatments remove the organic residues from the MoS2 surface up to some extent and alter the MoS2 chemical state. The equally detected displacements in peak binding energies of both the Mo 3d and S 2p core level spectra reveal that MoS2EF is shifted further toward Ec upon both plasma exposures. This could explain the observed improvements in the device on-state characteristics (Ion and mobility), compared with when no cleaning treatment is employed on the MoS2 surface (and the resist residual concentration remains significant), suggesting formation of less-impurity-interrupted contacts to MoS2 upon the Ti/Au deposition. The cross-sectional STEM images in Figure a,b and the detection of an evenly distributed TiS layer all along the Ti-to-MoS2 interface (without a discontinuity) also verify the progress made in the contact quality upon the plasma cleaning treatments. Based on these investigations, conducted for the O2 plasma-cleaned case, one could also infer that for the uncleaned MoS2 surface, the organic resist residues may prevent the direct Ti reaction with the S atoms in MoS2. The presence of these impurities may lead to the formation of lower-quality and discontinuous contacts to the underlying MoS2, as also reported by Bolshakov et al.,[29] explaining the generally lower electrical performance, for the reference case, as compared with the plasma-cleaned counterparts.

Apart from the on-state performance enhancements, the observed VT negative shifts together with the dramatic increase in Ioff (in Figure a) could then be readily linked to the MoS2 chemical changes and the shifts in the EF position, upon the cleaning treatments. Based on the XPS analysis, it was found that after the O2 and Ar plasma exposures, MoS2EF shifts +0.5 eV and +0.2 eV closer to Ec, compared with when no plasma cleaning step is yet employed. Because the shift of MoS2EF (toward Ec) is larger for the O2 plasma-treated sample, one would expect that the magnitude of VT increases and Ioff becomes higher for this case, compared with its Ar plasma-treated peer. However, a reverse behavior is observed. In other words, for the O2 plasma-treated sample, VT shifts only −5 V with respect to the reference, and Ioff is 0.06 μA/μm, whereas for the Ar plasma-treated case, ΔVT is −33 V and Ioff is 0.16 μA/μm. Elaborating the observations, it is important to note that the XPS analysis has been merely carried out before the metal contact deposition on the MoS2 surface. After any metal deposition, MoS2EF is expected to move even closer to Ec,[29,34] with its magnitude being highly dependent on the degree of chemical interactions at the metal-to-MoS2 interface. When the Ar plasma is applied on the MoS2 surface, the concentration of metallic Mo and MoS increases (as confirmed by the XPS). Furthermore, the Ar plasma has proven to be more effective in removing the organic residues. As a result, once the Ti/Au contacts are deposited on the Ar plasma-cleaned MoS2, the underlying MoS2 is more degenerately doped, and its EF position is shifted closer to Ec, leading to more carrier injection and transport toward the MoS2 channel, and thus higher Ioff. Considering the O2 plasma-treated sample and the difference in the nature of these two gases, the MoS2 surface does not represent metallic Mo or MoS species. Therefore, after the contact deposition, it is speculated that the MoS2 layer beneath Ti/Au is less excessively doped, leading to lower Ioff than in the case of Ar plasma treatment.

Plasma Cleaning and Annealing of the MoS2 Contacts

With the purpose of further enhancing the device electrical characteristics, plasma cleaning together with the HV annealing treatments were also examined. Employing both treatments has their own benefits and drawbacks. We report that Ioff reduces, and VT shifts positively. Despite this progress in the off-state regime, Ion and the FE mobility degrade, although the as-mentioned parameters are still better than when no cleaning treatment is applied. For further details, see Section S.7 in the Supporting Information and the associated discussion.

Scaling Down the Ti Interlayer Thickness

Up to this point, the Ti thickness has been maintained at 20 nm. Striving toward further optimizing the contacts for our ALD-based MoS2 FETs, it is necessary to benchmark the influence of scaling down the Ti thickness on the overall device electrical performance. Ti is generally used as an adhesive interlayer between the MoS2 and Au contacts. It is expected that reducing the Ti thickness facilitates the Au wavefunction penetration into the MoS2, leading to further electrical improvements.[31] To verify this, various Ti thicknesses, ranging from 20 nm down to 2.5 nm, were examined, meanwhile maintaining the whole Ti/Au stack thickness at 100 nm. Complete removal of this interlayer was also inspected. Initially, no plasma cleaning treatment was employed to clean up the MoS2 surface. Figure a shows the transfer curves of the fabricated MoS2 FETs with Ti thickness variations. At first sight, it is explicit that upon the Ti thickness reduction, Ion gradually increases and improves up to one order of magnitude, when Ti is completely removed from the contact stack (100 nm Au). On the other hand, Ioff deteriorates upon this reduction and VT shifts to more negative values, altogether implying that the MoS2 gets further n-type doped. The increase in the MoS2 doping concentration and the improvement in the device on-state performance are directly associated with further facileness of the Au wavefunction penetration into the MoS2, as pointed earlier. This progress is also evident from the extracted FE mobilities, as shown in Figure b. On average, the maximum FE mobility increases about five times upon decreasing the Ti thickness and reaches nearly 0.06 cm2/(V·s). The average mean Ion/Ioff ratio is also shown in this Figure, at its right axis. As can be seen, this ratio is maximum (about 40) when the Ti/Au stack is 5/95 nm, making this combination optimum as the contacts to MoS2.

Figure 6

(a) Transfer curves of different Ti interlayer thicknesses, (b) average FE mobility at the left axis and the average Ion/Ioff ratio at the right axis as a function of reducing Ti thickness.

It is also of interest to gain insights into the impact of plasma cleaning treatment before the deposition of various Ti thicknesses. Figure a–d provides the average statistical data (on device figures of merit) for the Ar plasma-cleaned contact openings, as a function of Ti thickness and relative to their uncleaned peers. Excluding the pure Au contact analysis at first, the Ar plasma cleaning treatment significantly increases Ion, shown in Figure a. As earlier discussed, this is due to the improved quality of contacts to the MoS2 and the continuous TiS compound formation by adding the plasma cleaning step. The on-state performance is further continued to enhance with reducing the Ti thickness down to 2.5 nm. A similar trend is also observable in the extracted mean maximum FE mobilities, as provided in Figure b. Meanwhile, Ioff increases upon the plasma exposure, resulting in the Ion/Ioff ratio to fall below 20 for all the Ti thicknesses. The reason for such a dramatic increase in Ioff is attributed to the increase in the doping density, either because of extended Au wave function penetration with reducing the Ti thickness or formation of metallic Mo and MoS on the MoS2 contact openings upon the Ar plasma cleaning treatment.

Figure 7

Average (a) Ion, (b) FE mobility, (c) Ioff, and (d) Ion/Ioff ratio for the uncleaned and Ar plasma-cleaned contacts as a function of Ti interlayer thickness, respectively.

Considering the pure Au case, a deviating behavior is observed. Upon the Ar plasma cleaning treatment, Ion as well as the FE mobility values slightly drop on average, whereas Ioff increases (Figure a–c, respectively), all compared to when no treatment is applied. It is already known from theoretical calculations that Au inclines to only weakly react with the surface of MoS2 (unlike Ti that tends to form covalent bonds to this 2D layer), leaving behind an extra tunneling barrier against the efficient carrier injection at the interface. The tunneling barrier is in addition to the Schottky barrier and leads to higher Rc values for MoS2 FETs.[31] In our fabricated devices, however, as has been previously confirmed by the XPS data, the organic residual concentration on the MoS2 surface is considerable when no plasma cleaning step is included. Therefore, it is speculated that the presence of surface impurities forms a bridge between Au and MoS2 and lowers the interface tunneling barrier, leading to the observed improvements in the average Ion and mobility (Figure a,b) of the devices (compared with the uncleaned Ti cases). On the other hand, application of the Ar plasma, prior to the pure Au deposition, results in a lower concentration of organic residues on the MoS2 surface and likely an increase in the tunneling barrier. Thus, the on-state performance (Ion and mobility) slightly drops on average and does not follow the trend observed for Ar plasma-cleaned Ti cases. In addition, based on the XPS data, the metallic Mo and MoS concentrations upon the Ar plasma cleaning process increase, causing the MoS2 layer and its electronic band structure to disrupt, leading to an increase in Ioff. Here, we note that the increase in Ioff, after the Ar plasma treatment, does not follow the trend shown in Figure c, where Ioff increases up to one order of magnitude by decreasing the Ti thickness. This could still be associated with weaker interactions between the Au- and the plasma-cleaned MoS2 surface, compared to when no adhesive interlayer is added. The transfer curves of the Au contacts with and without the Ar plasma cleaning treatment are provided in Figure S8 of the Supporting Information.

Analyzing the device electrical performance, for varied Ti thicknesses, reveals that among all the studied cases, the Ti/Au stack of 5/95 nm, without any Ar plasma cleaning treatment, is the optimal contact to our synthetic MoS2 films. This is mainly because the Ion/Ioff ratio, as one of the other important device figures of merit, is the highest when 5 nm of the Ti interlayer is deposited.

Substitution of the Ti Interlayer with Cr

Continuing the optimization path of the contacts for the ALD-based MoS2 FETs, the 5 nm Ti interlayer was replaced with 5 nm of Cr, meanwhile the metal deposition conditions (e.g., pressure and deposition rate) were kept constant. Initially, no plasma cleaning treatment was employed. The reason for choosing Cr is that Cr is known to have a lower affinity for oxidation, compared with Ti.[34,35] Therefore, it is expected that its interaction with the organic contaminations as well as the underlying MoS2 surface is relatively weaker. The schematics of the energy band diagrams before and after the contact deposition on MoS2, for both cases, are provided in the Supporting Information (Section S.9). The transfer curves of the MoS2 FETs with Cr and Ti interlayers are compared in Figure a. As can be seen from the plot, utilization of Cr instead of Ti leads to better controlling Ioff and improving Ion, suggesting that the degree of MoS2 doping and its electronic band structure disruption is mitigated with the Cr substitution. This is first because the Cr–S formation energy is higher than that of Ti–S, and both are lower than that of Mo–S, based on the Gibbs free energy of formation (ΔGof).[34,35] ΔGof for Cr2S3 is −148.19 kJ/mol, for TiS it is in the range of −573 to −390 kJ/mol, depending on x, and for MoS2 it is −117.99 kJ/m.[34,35] These values clearly indicate that Ti–S formation is more favorable than Cr–S. Second, the electronegativity difference between Cr–S is lower than that of Ti–S, implying that Cr–S bonds are weaker than Ti–S bonds, and the degree of MoS2 disruption is lower in the case of Cr contacts. In addition, Cr is not expected to react with the residual components as strongly as in the case of Ti (Cr–C and Cr–O electronegativity differences are less than those of Ti–C and Ti–O). As a result, on average, Ioff for the Cr/Au stack drops to half of the value obtained from Ti/Au, and Ion increases 1.5 fold, as evident in Figure b on the right and left axes, respectively.

Figure 8

(a) Transfer curves of Cr/Au and Ti/Au 5/95 nm contacts, (b) Average Ion (left axis) and average Ioff (right axis) for both contact types and (c) impact of plasma cleaning treatments before the Cr/Au deposition.

Inclusion of the plasma cleaning step for the Cr contacts and its impact on the electrical performance are also evaluated. Figure c compares the measured transfer data for the Ar and O2 plasma-treated MoS2 FETs, with respect to the uncleaned reference. Analogous to the Ti/Au cases, addition of the plasma cleaning step improves the device on-state features (Ion and the mobility), while the off-state performance deteriorates because MoS2 becomes degenerately doped and the back gate fails to efficiently deplete the channel, leading to a significant increase in Ioff and a significant negative shift in VT.

The average overall electrical characteristics of the fabricated MoS2 devices with both types of metal contacts are also evaluated. As can be seen from Figure a,b, substitution of Ti with Cr (without any initial plasma cleaning) doubles the mean Ion and FE mobility values. Although this improvement might be minor, the mean Ioff for the Cr/Au stack decreases as well (see Figure c). As a result, the Ion/Ioff ratio, provided in Figure d, improves on average and reaches ∼110. These assessments on the overall electrical characteristics of the fabricated MoS2 devices, with both types of contacts, reveal that the plasma cleaning treatments are beneficial in improving the device on-state performance. However, due to the lack of gate electrostatic control over the off-state current, the noncleaned devices are performing more optimally than their plasma-cleaned counterparts. One way to control Ioff is to thin down the MoS2 thickness, which will be discussed in a forthcoming paper. Among the two studied interlayers, Cr exhibits a better performance than Ti, as a result of its higher Ion/Ioff ratio.

Figure 9

Benchmarking the 5/95 nm Cr/Au contacts with different plasma cleaning pretreatments comparative to Ti/Au. (a) Mean Ion, (b) mean FE mobility, (c) mean Ioff and (d) mean Ion/Ioff, respectively.

All in all, the provided results and the associated discussions lead to better recognizing the chemistry involved at the metal-to-synthetic MoS2 interfaces and their direct influence on the overall device electrical performance.

Conclusions

In this study, working ALD-based MoS2 FETs have been successfully demonstrated, and the role of device processing conditions in the chemistry involved at the metal-to-MoS2 interface as well as their correlation with the electrical performance have been investigated in detail. This work highlights that understanding and tailoring the device-processing conditions as well as the contacts to MoS2 are vital for achieving optimal electrical performance, especially for the synthetic MoS2 films, where the role of intrinsic and process-induced defects in transport properties is even more pronounced than for their exfoliated counterparts.

In pursuit of optimizing the contacts to our MoS2 FETs, several efforts have been made. These are HV annealing the contacts after their deposition, plasma cleaning the contact opening areas and scaling down the Ti interlayer thickness as well as its substitution with Cr.

We have shown that annealing the Ti/Au contacts to MoS2, in a HV environment, is not effective in improving the on-state device performance. However, VT shifts positively and Ioff lowers, indicating that upon the Ti/Au HV annealing, MoS2 becomes relatively less n-type doped.

Furthermore, inclusion of the O2 or Ar plasma cleaning step (prior to the contact deposition) enhances the device on-state characteristics, up to one order of magnitude. This is while Ioff deteriorates, due to MoS2 being degenerately doped. For linking the role of interface chemistry to the electrical observations, XPS and STEM analyses have been conducted. The XPS study reveals that the impurity concentration on the MoS2 contact opening areas is significant. This is the result of resist usage during the device fabrication that could hinder the utmost performance, being achieved from our fabricated devices. The analysis also shows that addition of a mild plasma cleaning step removes a considerable portion of the resist remainders and shifts the MoS2EF position closer to Ec, all of which are directly associated with the observed improvements in the device on-state performance. Apart from the surface impurity reduction, the electrical progress in our fabricated MoS2 FETs is also attributed to the formation of an evenly distributed and continuous TiS layer, all along the MoS2 surface and upon the Ti deposition, as confirmed by the STEM analysis.

Scaling down the Ti interlayer thickness is proved to be necessary as well, for further unleashing the capability of our MoS2 films and for being possibly implemented in transistor platforms.

In the last attempt, the replacement of the Ti interlayer with Cr, in the contact stack, has been evaluated. The results show that upon the Cr usage, the gate electrostatic control over the off-state current improves, meanwhile the Ion range retains.