Influence of High-κ Dielectrics Integration on ALD-Based MoS₂ Field-Effect Transistor Performance

Abstract

The integration of high-κ dielectrics on MoS₂ field-effect transistors (FETs) is essential for the realization of MoS₂ in ultrascaled nanoelectronic devices and circuits. Most studies covering this topic are based on exfoliated MoS₂ flakes or chemical vapor deposition (CVD) grown MoS₂ films, whereas other techniques, such as atomic layer deposition (ALD), are also gaining attention for the growth of MoS₂ in recent years. In this work, we grow large-area MoS₂ by means of plasma-enhanced (PE-)ALD and evaluate the influence of high-κ dielectrics on the properties of ALD-based MoS₂ FETs through electrical characterization combined with surface-chemical and high-resolution scanning transmission electron microscopy (HR-STEM) analyses. We grow HfO_x, AlO_x, or both by means of PE-ALD or thermal ALD on our fabricated devices and show that, in addition to the dielectric constant, three other major parameters related to the processing of the dielectrics can simultaneously affect the MoS₂ FET electrical characteristics and govern its doping. These parameters are the stoichiometry of the dielectric, its carbon impurity content, and the degree to which the MoS₂ surface oxidizes upon the dielectric growth. When grown at 100 °C, our HfO_x films are oxygen-vacant whereas our AlO_x films are oxygen-rich. In addition, carbon impurities are incorporated into the dielectrics at low deposition temperatures, being one of the likely causes of the MoS₂ FET overall n-type performance in all of the studied cases. Our investigations also reveal that PE-ALD of HfO_x or AlO_x oxidizes the MoS₂ surface, whereas thermal ALD AlO_x leaves MoS₂ almost intact. In this respect, if thermal ALD AlO_x of proper thickness is grown between MoS₂ and HfO_x, it can reduce the degree to which the MoS₂ surface oxidizes by HfO_x and meanwhile improve the total dielectric constant, altogether leading to the most optimal electrical performance in ALD-based MoS₂ FETs.

1. Introduction

Ever since layered two-dimensional (2D) transition metal dichalcogenides (TMDCs) were highlighted as potential material candidates for sub-10 nm technology nodes,¹⁻³ there has been a growing interest in validating the feasibility of their implementation into such prospective platforms.³⁻⁶ Among the 2D TMDCs with semiconducting properties, MoS₂ is the most extensively studied material, especially in the context of field-effect transistors (FETs). MoS₂ FET device metrics are continually progressing, with the highest records being so far held for a current density of 700–1135 μA/μm,⁷⁻⁹ a subthreshold swing (SS) close to the thermionic limit (∼60 mV/dec),^10,11 a field-effect mobility (μ_FE) of 200 cm²/(V·s),¹² and a low contact resistance (R_c) range of 123–480 Ω·μm,^7,9,13 although none of these figures of merits have been demonstrated all together in the same device. Besides these fascinating device properties, MoS₂ FETs have demonstrated relatively low variability, good reproducibility, and decent reliability,^10,14,15 as other vital conducts for their realization into more complex circuits.^10,16 However, implementation of emerging materials on technology-relevant scales has its own challenges. As with semiconducting MoS₂, there are yet several major hurdles that first need to be surmounted. Among them, the integration of nanometer-thin high dielectric constant (high-κ) materials with MoS₂ FETs, as well as the large-area synthesis of high quality MoS₂ films, is of crucial importance.

High-κ materials, such as HfO₂, ZrO₂, and Al₂O₃, are commonly used as gate dielectrics in FETs, with the aim of boosting the gate electrostatic control over the channel material.¹⁷ In conventional Si-based FETs, ultrathin layers of high-κ (<2 nm),¹⁸ which are necessary for downscaling the size of devices, can be successfully grown by means of atomic layer deposition (ALD). ALD is a cyclic thin film deposition technique that is well settled in industry and results in thickness-controlled, highly uniform, conformal and continuous films, free of pinholes.¹⁹ During the ALD process, layer growth is mainly governed by surface chemical reactions between the substrate of choice and the ALD reactants.

As far as the ALD growth of dielectrics on 2D MoS₂ is considered, due to the all-surface and chemically inert basal planes of pristine MoS₂, this process is mostly hampered by the lack of surface groups to which the ALD reactants could chemisorb.^20,21 Therefore, nonuniform and defective dielectric films with incomplete surface coverages are often grown on pristine MoS₂, especially for sub-10 nm high-κ thicknesses.²⁰⁻²² Several attempts have been made to address this issue. UV/O₃²³ or O₂ plasma²⁴ surface functionalization and addition of a buffer/seed layer,^10,25,26 prior to the high-κ deposition on MoS₂, as well as employment of plasma-enhanced (PE)-ALD^27,28 were overall beneficial in improving the growth quality of ultrathin films of these dielectrics on MoS₂.

In addition to the gate dielectric role in FETs, high-κ materials can also act as protective or passivating layers for ultrathin nanomaterials and lead to further improvements in the electronic properties of the resulting devices.^12,29 For example, capping MoS₂ FETs with high-κ dielectrics, especially for back-gate devices, has been shown to highly influence the device metrics such as μ_FE, the maximum drive current (I_ON), and hysteresis.^{12,14,30−32} Theoretical studies have associated these improvements mainly with the efficient screening of MoS₂ charged impurities (CIs) by the dipoles of high-κ materials.³³⁻³⁵ CIs are considered as the dominant Coulombic scattering centers in/on 2D materials, and they lead to carrier transport degradation.³³⁻³⁵ MoS₂ CIs can have various sources, but they typically originate from intrinsic structural defects (e.g., Mo/S vacancies),³⁶ environmental adsorbates (e.g., H₂O),^14,37,38 and organic residues formed during the device fabrication.^11,21

Parallel to the suppression of CI scattering, high-κ materials have the capability of doping the underlying MoS₂ as well. This is mainly attributed to the existence of bulk/interfacial fixed charges,^39,40 which are formed when the dielectric film is sub(over)-stoichiometric and contains oxygen-vacant (oxygen-rich) states. The dielectric fixed charges can electrostatically induce carriers of opposite sign into the channel^31,41 or dope the channel via a process known as charge transfer doping.⁴² These charges, furthermore, are shown to reduce the contact resistance (R_c) and improve MoS₂ conductivity by lowering Schottky barrier height (SBH) at the contact-to-MoS₂ interface.⁴³ Both experimental and theoretical studies have shown that the charge transfer doping process via oxygen-vacant (oxygen-rich) high-κ materials (e.g., HfO_x, AlO_x, and TiO_x) can n-type dope (p-type dope) MoS₂^7,26,44,45 mainly because the defect energy bands in such films are distributed in the vicinity of MoS₂ conduction band edge (E_c) (valence band edge (E_v)).^7,44 Therefore, tuning the dielectric film stoichiometry (as recently shown^43,46) can be a way forward for controllably doping the MoS₂ films and improving their electronic properties. In addition to the oxygen-vacant (oxygen-rich) states, the carbon content in high-κ dielectrics is also another doping contributor,⁴⁷ wherewith the carrier density and other MoS₂ FET properties can be adjusted. All of these investigations highlight the determining role of the dielectrics in tailoring the electrical characteristics of MoS₂ FETs.

As mentioned earlier, synthesis of high quality MoS₂ films over large areas is the other challenge for the realization of MoS₂ FETs in ultrascaled circuits. Although chemical vapor deposition (CVD) is one of the primary techniques to fulfill this demand,⁴⁸ in pursuit of a low thermal budget (required for industrial processes) as well as a precise thickness control, (PE-)ALD has gained attention in recent years for the growth of MoS₂ and other 2D TMDCs.⁴⁹⁻⁵¹ However, films obtained by this technique are polycrystalline, with point defects and grain boundaries (GBs) inherent to their synthesis route.⁵²

As far as the growth of high-κ dielectrics on such synthetic and polycrystalline MoS₂ films is concerned, it has been shown that the dielectric layer mostly nucleates along the MoS₂ GBs and defect sites. This leads to the growth of dielectric nanoribbons⁵³ during the initial stages of the dielectric deposition and subsequently results in an incomplete MoS₂ surface coverage for sub-10 nm thicknesses.^53,54 Therefore, it can be inferred that the growth of ultrathin, high-quality, and highly uniform dielectrics on polycrystalline MoS₂ has its own challenges, which can be different from the pristine MoS₂ films obtained by exfoliation. Up until now, the integration of high-κ materials has been mainly studied on exfoliated MoS₂ flakes^12,21,27 or large-grain CVD grown MoS₂ films,^7,28,32 with no studies directly evaluating the growth of dielectrics on ALD-based MoS₂ films.

In this study, we employ an ALD-based approach for the large-area synthesis of MoS₂.^49,55 Because of the principal differences between the morphology of ALD-based MoS₂ films and those obtained by other techniques (e.g., smaller grain size, more GBs, and possibly deviating stoichiometries), the influence of the commonly used dielectrics on the electrical properties of ALD-based MoS₂ films needs to be investigated independently. In this contribution, we grow AlO_x, HfO_x, or combinations thereof on ALD-based MoS₂ FETs. Through current–voltage (I–V) characterization, we evaluate the electrical performance of the fabricated devices. We link our electrical findings to the chemistry involved at the high-κ/MoS₂ interface by means of X-ray photoelectron spectroscopy (XPS) analysis. We also assess the dielectric film coverage on MoS₂ in the early cycles of high-κ deposition using top view scanning transmission electron microscopy (STEM) imaging.

2. Experimental Section

2.1. MoS₂ Film Synthesis

For this work, 7–8 monolayers of MoS₂ were synthesized by a two-step approach, where large-area and thickness-controlled MoO_x films were first grown at 50 °C using PE-ALD⁵⁶ in an Oxford Instruments Plasma Technology ALD reactor (FlexAL). The as-deposited MoO_x films, with a thickness of ∼4.9 nm, were then annealed in a tube furnace, where a H₂S/Ar gas mixture of 10%/90% was introduced at 900 °C for 45 min, to eventually obtain MoS₂ multilayers. Further details of the synthesis method and film specifications are published in a previous work.⁵⁵ In all cases, the film growth took place on highly doped (p⁺⁺) Si substrates, thermally coated with ∼285 nm SiO₂, serving as a global back-gate for the fabricated MoS₂ FETs.

2.2. Back-Gate Device Fabrication

The back-gated MoS₂ FETs were fabricated using a standard two-step electron beam lithography (EBL) technique. Details of the device fabrication are reported in a previous study.⁵⁷ In this work, a combination of 5/95 nm Ti/Au was electron beam evaporated as the contacts to MoS₂. Then, to define the channel areas, MoS₂ was dry etched using a SF₆/O₂ plasma gas mixture. The fabricated devices were finally capped with a 30 nm thick high-κ dielectric layer. A top-view optical microscopy image of the final MoS₂ FETs and cross-sectional schematics are provided in Figures S1a and S1b, respectively.

2.3. MoS₂ FET Capping with PE-ALD Grown HfO_x

Investigations regarding the influence of dielectrics on the back-gate MoS₂ FETs were initiated with the growth of 30 nm HfO_x on the fabricated devices using the PE-ALD method at three different table temperatures of 100, 200, and 300 °C. HfCp(NMe₂)₃ (HyALD) and O₂ plasma were the precursor and the oxygen agent used during this process, respectively.⁵⁸ As HyALD precursor does not react with H₂O, films of HfO_x could only be grown using the PE-ALD method, and thermal ALD of HfO_x could not be examined in this study. Further details and specifications of the growth process are provided in Section S.2.1 of the Supporting Information.

2.4. MoS₂ FET Capping with ALD and PE-ALD Grown AlO_x

Next, 30 nm thick AlO_x dielectrics were grown on the fabricated MoS₂ FETs using both ALD and PE-ALD processes at 100 °C.⁵⁹ We note that throughout this study ALD refers to thermal ALD. For the ALD of AlO_x, trimethylaluminum (TMA, Al(CH₃)₃) vapor and H₂O were employed as the precursor and the oxygen agent, respectively, and for the PE-ALD of AlO_x, TMA and O₂ plasma were utilized. See the Supporting Information for further details of both methods and their processing conditions (Sections S.2.2 and S.2.3).

2.5. MoS₂ FET Capping with ALD/PE-ALD Grown Bilayers of AlO_x/HfO_x

Bilayer stacks of AlO_x/HfO_x were also evaluated as capping layers for the MoS₂ FETs. The processing conditions were similar as to when the AlO_x and HfO_x films were grown individually on the fabricated devices at 100 °C. The thickness of the entire dielectric stack was maintained at 30 nm, while examining 2.5/27.5 nm, 5/25 nm, and 10/20 nm of AlO_x/HfO_x stack combinations. The AlO_x interlayer was grown by both ALD and PE-ALD methods, except in the case of the 10/20 nm combination where only ALD was used for the growth AlO_x.

2.6. Dielectric Thickness Measurements

The data for the high-κ film thicknesses were collected using an in situ spectroscopic ellipsometry (SE) instrument (J.A. Woollam Co., Inc. M-2000F, 1.24–5 eV range). The SE measurements were conducted on a control sample of 450 nm SiO₂/Si. We assume that the growth of dielectrics on MoS₂ is approximately the same as that on SiO₂/Si substrates. The film thicknesses were obtained using a Cauchy-based fitting model in Complete EASE software.

2.7. Electrical Characterization

The room-temperature electrical measurements were performed on 500 nm long and 1 μm wide MoS₂ channels using a Keithley 4200-SCS parameter analyzer, which was connected to a cryogenic probe station (Janis ST-500) operating at a base pressure of ∼1.9 × 10^–4 mbar.

2.8. Surface Characterization

The chemistry involved at the high-κ dielectric/MoS₂ interface was investigated using ex-situ XPS. For this purpose, the dielectric films were grown on blanket films of MoS₂ with equal deposition conditions as to when they were grown on the fabricated MoS₂ FETs, so that consistency could be maintained. The setup used for the XPS analyses was a Thermo Scientific K-alpha KA1066 spectrometer (Thermo Fisher Scientific, Waltham, MA) with a monochromatic Al Kα X-ray radiation source (hv = 1486.6 eV). The measurements were carried out with an X-ray beam spot size of 400 μm, at a takeoff 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/differential charging. The acquired spectra were later chemically quantified and deconvoluted with Avantage software. All the binding energies (BEs) were calibrated with respect to the C 1s adventitious carbon peak, set to 284.8 eV.

2.9. Structural Characterization

The high-κ coverage on MoS₂ was also studied using a top-view STEM imaging with a JEOL atomic resolution microscope (ARM) 200-F, operated at 200 kV and equipped with a 100 mm² Centurio silicon drift detector (SDD) as well as an energy dispersive X-ray (EDX) detector. The analyses took place during the early cycles of (PE-)ALD, and electron transparent SiN_x TEM windows, coated with ∼5 nm ALD SiO₂, were utilized as the imaging substrates. The MoS₂ films were grown on these substrates under equal conditions as when they were being synthesized for devices (explained earlier), followed by relevant dielectric depositions.

3. Results and Discussion

3.1. HfO_x Capping Layer

As already mentioned, high-κ materials can efficiently screen the MoS₂ CIs³³⁻³⁵ and electrostatically dope this 2D layer.^7,26,44,45 HfO_x is one of the most studied dielectrics in the context of FETs. Throughout this section, we evaluate the influence of HfO_x on the electrical performance of our ALD-based MoS₂ FETs.

3.1.1. Electrical Characterization of HfO_x-Capped MoS₂ FETs

Initially, 30 nm HfO_x was grown on the back-gate MoS₂ FETs using PE-ALD⁵⁸ at 100 °C. Figure 1a (left axis) shows the normalized double sweep transfer characteristics (I_DS–V_GS) of a device with and without HfO_x. As can be seen, upon capping of the MoS₂ FETs with HfO_x, the device metrics improve significantly. The maximum drain current (I_ON) increases at least 50 times and reaches 0.5 μA/μm, the current rise in the p-branch is suppressed, and the hysteresis reduces. However, the minimum drain current (I_OFF) slightly increases, and the threshold voltage (V_T) shifts to more negative values.

Figure 1.

(a) Left axis shows the transfer curves of a device with and without HfO_x grown at 100 °C (on a semilog scale). The red arrows determine the direction of forward and backward current sweeps. The right axis shows I_GS in linear scale, (b) transfer curves of the MoS₂ FETs capped with HfO_x films grown at deposition temperatures of 100, 200, and 300 °C.

These observations imply that the MoS₂ channel becomes n-type doped upon the PE-ALD of HfO_x at 100 °C. In addition, the hysteresis reduction could be an indication of environmental adsorbates (e.g., water molecules) and defective states in the MoS₂. By capping MoS₂ with HfO_x, environmental adsorbates are prevented, and the density of defective states on MoS₂ surface could be reduced.^14,30,32 It is also important to evaluate the gate leakage current (I_GS) of the fabricated FETs. As shown in the right axis of Figure 1a, before and after the device capping, I_GS is maintained well below 0.1 nA, inferring that no additional conductive path other than the MoS₂ channel is present in our fabricated devices. All throughout this work, we consider the MoS₂ FETs capped with PE-ALD grown HfO_x at 100 °C as our reference device. Hence, the results obtained from other fabricated devices are compared with this case.

Next, the influence of the HfO_x growth temperature on the electrical performance of the devices was investigated. For this purpose, 30 nm HfO_x films were grown on the as-fabricated MoS₂ FETs at 200 and 300 °C, while keeping all other PE-ALD conditions equal to the reference. As provided in Figure 1b, increasing the HfO_x deposition temperature changes the MoS₂ FET transfer characteristics. Unlike the reference, complete p-type device behavior is observed upon HfO_x growth at 200 °C, with I_ON reaching 0.5 μA/μm. When HfO_x is deposited at 300 °C, a highly degraded performance is noticed, with a minor rise in the p-branch.

Based on a former report by Sharma et al.,⁵⁸ our PE-ALD grown HfO_x films are substoichiometric and contain oxygen vacancies if grown below 400 °C. The oxygen vacancy states in a substoichiometric dielectric (e.g., HfO_x, AlO_x, and TiO_x) are known to be energetically distributed in close proximity to the MoS₂E_c,^7,26,44,45 and introducing only 0.9% of oxygen vacancies into these dielectrics has been shown to significantly n-dope MoS₂.³² Therefore, MoS₂ FETs can become readily n-type doped, and R_c can be reduced⁴³ by the transfer of charges from the dielectric oxygen vacancy states to the MoS₂ + E_c.⁶⁰ Sharma et al. have also demonstrated that the HfO_x film stoichiometry depends on the deposition temperature, and increasing the temperature from 100 to 300 °C results in the HfO_x films that are closer to full stoichiometry.⁵⁸

These earlier findings can clearly explain the observed n-type device characteristics in our reference case, as the HfO_x capping layer grown at 100 °C contains the most oxygen vacancies among the three studied cases. However, the electrical behavior observed for the devices capped with HfO_x films grown at 200 and 300 °C cannot be attributed to the presence of oxygen vacancies. With increasing the growth temperature, the HfO_x films get closer to full stoichiometry⁵⁸ and, hence, lead to less n-type doping. This should manifest itself in a relative I_OFF reduction and a positive shift of V_T. However, a complete p-type device behavior and a decaying performance are noticed for the MoS₂ FETs, with sequentially increasing the HfO_x growth temperature. To explain these electrical observations, the chemistry involved at the HfO_x/MoS₂ interface needs to be investigated in the next step.

3.1.2. XPS Analysis of HfO_x-Capped MoS₂

For this purpose, ex-situ XPS measurements were conducted before and after the growth of ∼2.5 nm HfO_x on bare (as-synthesized) MoS₂ films at 100, 200, and 300 °C. Because of the limited depth resolution of XPS tools (∼7 nm), only 2.5 nm of HfO_x was deposited in this phase, so that the HfO_x/MoS₂ interface could be well evaluated. Depth profiling of the stack was also not an option due to the preferential sputtering of S.

Figure 2 shows the acquired elemental data for the core level spectra of the Mo 3d and S 2p, fitted accordingly, and charge corrected against the C 1s adventitious peak (284.8 eV) (for the C 1 s spectrum, see Figure S2). The bottom panel in each plot corresponds to the bare MoS₂ film with no dielectric capping, and the rest of the panels are offset vertically. Considering the Mo 3d core level spectrum of the bare MoS₂ case, two major peaks (as representatives of the Mo⁴⁺ 3d_5/2 and Mo⁴⁺ 3d_3/2 doublet peaks) are detected at 229.8 and 233.0 eV, respectively.

Figure 2.

Mo 3d and S 2p core level spectra before and after the deposition of 2.5 nm HfO_x at 100, 200, and 300 °C. Panels in each plot are offset vertically with respect to the bare MoS₂.

To improve the fitting, another doublet was also required, with peaks of slightly lower BEs than the Mo⁴⁺ doublet, proposing that our synthetic MoS₂ film is slightly substoichiometric and sulfur-deficient.⁶¹ At the left side of this spectrum (higher BEs), the Mo⁶⁺ oxidation state (MoO_x, x ∼ 3) is also detected, which is mainly due to the exposure of the MoS₂ surface to the ambient air. Finally, as a result of the S 2s core level overlapping into the Mo 3d spectrum, an extra peak is detected at a lower BE of 226.9 eV. Considering the S 2p spectrum, the S 2p_3/2 and S 2p_1/2 doublets are identified at 162.6 and 163.8 eV, respectively.

Upon the growth of ∼2.5 nm of HfO_x on MoS₂ at 100 °C, the intensity of the Mo 3d and S 2p major peaks drops considerably. In addition, a new peak appears at 223.0 eV, which is attributed to the Hf 4d_3/2 chemical state. The integrated area under the Mo⁶⁺ doublet peaks slightly increases as well, and the Mo⁴⁺ doublet peak BEs shift −1.00 eV (relative to the Mo⁴⁺ BEs of bare MoS₂). A similar magnitude of the BE shift is also detected for the S 2p core levels. These observations altogether imply that upon the growth of HfO_x, the concentration of MoO_x (x ∼ 3) species increases on the MoS₂ surface and the MoS₂ surface oxidizes. This is expected since O₂ plasma is used as the oxygen agent during the growth of HfO_x, and the MoS₂ surface is exposed to it from the beginning of the PE-ALD process.

Elevating the growth temperature of HfO_x to 200 and 300 °C, the Mo⁶⁺ doublet peaks shift to higher BEs, and the integrated area under these peaks further increases. The Mo⁶⁺/Mo⁴⁺ ratio changes from 0.44 (at 100 °C) to 1.35 (at 200 °C) and to 1.28 (at 300 °C). Details of the Mo⁶⁺ and Mo⁴⁺ doublet peak areas for all the studied cases are provided in Table S1. In addition, with increasing the growth temperature to 200 and 300 °C, both the Mo 3d and S 2p major peaks shift −1.3 and −1.1 eV, respectively, as compared with the bare MoS₂ case. These findings are jointly pointing in the direction that when HfO_x is deposited at temperatures higher than 100 °C, the concentration of MoO_x (x ∼ 3) increases considerably, and that the MoS₂ surface further oxidizes, mainly because of higher surface reactivity at elevated temperatures.

It is also important to verify that (part of) the MoS₂ films are still structurally intact upon the growth of HfO_x. For this purpose, a Raman analysis was performed. The data provided in Figure S3 reveal that the characteristic MoS₂ Raman modes (A_1g and E¹_2g peaks) are still present after the growth of HfO_x, irrespective of the HfO_x deposition temperature.

3.1.3. Linking XPS Results to I–V Characterization

To correlate the XPS chemical findings with the I–V measurements, it is essential to first point out that the MoO_x compound detected on the MoS₂ surface is generally known as a hole dopant for MoS₂, given that x ≤ 3.^46,62 The MoO_x bandgap (E_g) reaches up to ∼6.5 eV, and the oxygen vacancy states in MoO_x energetically reside close to the MoS₂E_v.⁶²

Based on the XPS data, when the HfO_x is grown at 200 °C on MoS₂, the concentration of MoO_x (x ∼ 3) increases significantly compared to when HfO_x is grown at 100 °C. As a result, more holes can be transferred/induced into the underlying MoS₂. This could explain the p-type conductivity observed for the MoS₂ FETs capped with the HfO_x film grown at 200 °C.

It is also expected that by raising the growth temperature of HfO_x further to 300 °C, the p-type doping effect in the MoS₂ FETs could become more pronounced due to the additional increase in the MoO_x concentration. Although from XPS and Raman spectroscopy it can be concluded that there is still intact MoS₂ under HfO_x, the damage to the MoS₂ surface (created by the O₂ plasma usage during the HfO₂ PE-ALD process) can be a likely cause for the observed device deterioration. Furthermore, one should note that the PE-ALD process of HfO_x takes place in the end of the device fabrication, when the contacts are delineated and the MoS₂ channel is already defined. The growth of HfO_x at an elevated temperature of 300 °C can impose an additional annealing effect to the Ti/Au contacts. This is undesired, as previous studies have shown that annealing the Ti/Au contacts degrades the MoS₂ FET electrical performance and increases R_c.^12,37,57 Therefore, it is speculated that the observed drop in the device metrics could also be attributed to the Ti/Au interface quality that deteriorates upon the growth of HfO_x at 300 °C. In addition to the MoS₂ surface damage and contact annealing effect, the HfO_x films grown at elevated temperatures (>200 °C) usually show higher degree of crystallinity and subsequently higher surface roughness.^27,58 The increase in the HfO_x surface roughness can increase the carrier scattering in MoS₂.²⁷ This can also contribute to the observed degradation in the MoS₂ device performance. The effects mentioned above can altogether explain why the growth of HfO_x at 300 °C does not lead to further p-type doping of the MoS₂ FETs.

In summary, the dielectric growth temperature strongly influences the ALD-based MoS₂ FET characteristics. PE-ALD of HfO_x at 100 °C is oxygen-vacant and leads to an optimal n-type performance in the fabricated devices. When HfO_x is deposited at 200 °C, the MoS₂ surface oxidizes and results in the observed p-type conductivity. Then, by growing HfO_x films at 300 °C, the MoS₂ surface damage, the Ti/Au contact annealing effect, and the HfO_x surface roughness increase can simultaneously contribute to the fully deteriorated device performance.

3.2. AlO_x Capping Layer

In the next step, 30 nm AlO_x capping layers were grown on the fabricated MoS₂ FETs, using both ALD and PE-ALD processes. A deposition temperature of 100 °C was chosen to allow for direct comparison with the reference (HfO_x case) and to prevent device degradation when the dielectric layer is grown at higher temperatures. PE-ALD can have several advantages over ALD, e.g., improved film properties and the possibility to grow the dielectrics at lower deposition temperatures. However, as seen in the previous section and observed in other studies, PE-ALD can induce unwanted damages to the underlying layer.⁶³ As far as MoS₂ is concerned, by carefully choosing the PE-ALD processing conditions, a uniform coverage of ultrathin dielectric films (AlO_x or HfO_x) can be achieved on exfoliated MoS₂ and without severely damaging this 2D layer.²⁷ However, the growth of AlO_x with both ALD and PE-ALD as well as its nucleation behavior on the ALD-based polycrystalline MoS₂ has not yet been thoroughly reported in the literature. Therefore, both methods are investigated for the growth of AlO_x on our fabricated devices.

3.2.1. I–V Characterization of AlO_x-Capped MoS₂ FETs

Figure 3a shows the transfer characteristics of AlO_x-capped MoS₂ FETs and compares the data with the reference device (capped with HfO_x grown by PE-ALD at 100 °C). As can be seen, irrespective of the deposition method, AlO_x capping leads to an overall n-type performance in MoS₂ FETs. However, compared with the reference, in both AlO_x-capped devices, V_T shifts positively and I_OFF reduces, implying a reduced n-type doping concentration in MoS₂. As far as the AlO_x growth methods are concerned, a higher I_ON is achieved when AlO_x is grown by PE-ALD on MoS₂ FETs. This could be attributed to the generally higher dielectric constant of PE-ALD AlO_x films compared to thermal ALD AlO_x films.⁶⁴ A higher dielectric constant is known to screen the MoS₂ CIs and reduce the free carrier scattering more effectively,³⁴ thereby improving the MoS₂ FET ON-current density. The average data shown in Figure S4a and obtained after measuring 3–4 devices at different locations on the individual samples also confirm an overall higher I_ON for the PE-ALD AlO_x case than the ALD counterpart. Among these three studied cases, PE-ALD AlO_x-capped devices exhibit the most optimal performance due to their lower I_OFF and higher I_ON/I_OFF ratio. The average data for I_ON/I_OFF are provided in Figure 3b, and other device metrics (mean values for the maximum μ_FE and I_OFF) are shown in Figure S4b,c, respectively.

Figure 3.

(a) Semilog double sweep transfer curves of the MoS₂ FETs capped with AlO_x films, grown by ALD, PE-ALD at 100 °C, as well as the uncapped case. The arrow determines the relative change in V_T upon the growth of AlO_x on the fabricated devices. (b) Average I_ON/I_OFF ratio after the growth of these dielectrics.

One may assign the as-observed MoS₂ FET overall n-type behavior to the AlO_x oxygen vacancy states.^7,44 However, earlier studies^59,64 have shown that our AlO_x films grown below 200 °C (with both ALD and PE-ALD methods) are oxygen-rich ([O]/[Al] > 1.5). The high oxygen content at low deposition temperatures is mainly associated with the incorporation of a considerable concentration of hydroxyl (−OH) groups into the films. The AlO_x used for capping our MoS₂ FETs is grown at 100 °C, and our films are oxygen-rich. Therefore, the observed n-type device performance cannot be the result of oxygen vacancies.

A recent study has addressed this controversy by highlighting that the AlO_x carbon content can play a significant role in n-type doping the MoS₂ FETs.⁴⁷ Generally, ALD of dielectrics by metal–organic precursors at low deposition temperatures leads to carbon impurities into the films due to inefficient ligand abstraction.⁶⁵ The carbon-related states are energetically distributed inside the AlO_xE_g and close to the oxide midgap,⁴⁷ such that they can donate electrons to the MoS₂E_c via charge transfer doping process. Since we grow the AlO_x films at 100 °C, carbon impurities are anticipated. Therefore, one potential interpretation for the overall n-type behavior of our AlO_x-capped MoS₂ FETs can be the incorporation of carbon impurities into the AlO_x films. As far as the AlO_x oxygen-rich states are concerned, their role is to reduce the MoS₂n-type doping density, since they transfer holes to the MoS₂E_v.^7,44 This has been manifested in the positive shift of V_T and the reduction of I_OFF (relative to the reference case).

3.2.2. XPS Analysis of AlO_x-Capped MoS₂ Films

To investigate the carbon content and other chemical states at the dielectric/MoS₂ interface, XPS analysis was performed before and after the growth of ∼5 nm of AlO_x on bare MoS₂ films. Since the HfO_x capping layer is a reference for all of our analyses in this study, a comparable thickness of HfO_x on MoS₂ is also chemically analyzed with XPS. Figure 4 shows the acquired elemental data for the Mo 3d, S 2p, and C 1s core level spectra, with the bottom panel in each plot as a representative of the bare MoS₂.

Figure 4.

Mo 3d, S 2p, and C 1S core level spectra before and after the growth of ∼5 nm ALD and PE-ALD AlO_x as well as PE-ALD HfO_x, all at the deposition temperature of 100 °C.

Compared with the bare MoS₂ case, after the growth of ALD AlO_x on MoS₂, the peaks associated with the Mo⁴⁺ doublet remain unchanged in the Mo 3d spectrum, and no significant oxidized species are detected (the Mo⁶⁺/Mo⁴⁺ is only 0.07 (see Table S2)). In addition, neither the Mo 3d nor the S 2p major peak intensities drop dramatically. These observations imply that the ALD of AlO_x does not oxidize the MoS₂ surface and that only a very thin layer of AlO_x is deposited on MoS₂ (as also verified by the Al 2p spectrum). As far as the C 1s spectrum is concerned, in addition to adventitious carbon, C–O chemical species are incorporated into the film composite.

In the case of PE-ALD AlO_x, unlike the ALD counterpart, the underlying MoS₂ film is slightly oxidized. However, the oxidation is to a smaller extent as compared with the PE-ALD HfO_x case. The Mo⁶⁺/Mo⁴⁺ ratio for the PE-ALD AlO_x case is 0.11, whereas it is 0.52 for the PE-ALD HfO_x case (see Table S2 for a quantitative analysis). In addition, both the Mo 3d and S 2p major peaks shift −1.0 eV, being another indication of MoS₂ surface oxidation. Considering the C 1s spectrum, C–O and carboxyl chemical states are detected.

Upon the growth of ∼5 nm of HfO_x on MoS₂, similar to what was observed earlier in Figure 2 for the growth of ∼2.6 nm of HfO_x at 100 °C, there is again evidence of MoS₂ surface oxidation. In this case, the Mo⁴⁺ and the S 2p doublets shift −0.9 eV. In addition, the Mo⁶⁺/Mo⁴⁺ ratio increases to 0.52 (see Table S2 for the details). As the Mo⁶⁺ and Mo⁴⁺ doublet peaks are not clearly distinguishable in the graph above, this spectrum is provided separately in Figure S5. In the C 1s spectrum, apart from the C–C adventitious peak, two other chemical states (C–O and C(=O)–OH (carboxyl)) are detected at 287.1 and 289.4 eV, respectively. Compared with the bare MoS₂, the integrated area under the carboxyl peak increases upon the growth of HfO_x, suggesting that a moderate amount of carbon-related content is incorporated into PE-ALD grown HfO_x.

These interface observations draw our attention to several notable points:

First of all, unlike H₂O, the employment of O₂ plasma reactant during the PE-ALD process of AlO_x and HfO_x oxidizes the MoS₂ surface, which is expected to facilitate the growth of these dielectrics on MoS₂.²⁴ The O₂ plasma duration for the HfO_x process is 8 s in each deposition cycle, whereas for the AlO_x it is 2 s (see the Supporting Information for the details of each process, Sections S.2.2 and S.2.3). The shorter plasma exposure time during the PE-ALD of AlO_x can partly explain the lower Mo⁶⁺ content (Table S2) compared with what is obtained from the PE-ALD of HfO_x on MoS₂. The reader should note that for the PE-ALD of AlO_x and HfO_x films, standard recipes (which had been developed earlier) were simply used. This clarifies why processing conditions (e.g., O₂ plasma duration) are different in these two processes.

The second point to highlight is that because H₂O does not react strongly with MoS₂, there is most likely a nucleation delay during the early ALD cycles of AlO_x. Here, it is noted that all the dielectric layer thicknesses, as referred to in this study, were determined on control samples of SiO₂/Si. The nucleation delay of ALD AlO_x on MoS₂ could lead to a thickness deviation between what is deposited on MoS₂ and that on SiO₂/Si during the early cycles of the deposition.

Finally, for both ALD and PE-ALD processes, carbon impurities are incorporated into the dielectric film because of the low growth temperature of 100 °C and subsequently the incomplete ALD reactions.⁵⁹

3.2.3. Linking XPS Results to I–V Characterization

After these extensive XPS analyses, it becomes clear that the AlO_x films, processed by both the ALD and PE-ALD methods, contain a measurable amount of C–O and C(=O)–OH (carboxyl) species. The presence of these carbon-related compounds can be a plausible explanation for the observed n-type device performance during the I–V measurements.⁴⁷ One may also wonder about the role of MoO_x, which is known as a hole dopant for MoS₂. It is worth pointing out that the detected concentration of MoO_x after both the ALD and PE-ALD AlO_x processes at 100 °C is not so significant. The Mo⁶⁺/Mo⁴⁺ ratios (see Table S2) are only 0.07 and 0.11 for the ALD and PE-ALD AlO_x cases, respectively. This is while the Mo⁶⁺/Mo⁴⁺ ratio is 0.52 for the reference case (HfO_x-capped MoS₂ FETs). Therefore, a substantial hole transfer into the MoS₂E_v is less likely.

To summarize, although our ALD/PE-ALD AlO_x films are oxygen-rich, the MoS₂ FETs capped with these dielectrics still show n-type performance, which can be the result of carbon impurities incorporated into the AlO_x films when deposited at low temperatures. Furthermore, the devices capped with PE-ALD AlO_x exhibit the most optimal characteristics, as they show higher I_ON/I_OFF and lower I_OFF among the three studied cases. They also perform better than the ones capped with the ALD AlO_x in the ON-state regime, mostly due to the generally higher dielectric constant of PE-ALD AlO_x than the ALD counterpart.

3.3. AlO_x/HfO_x Bilayer Capping

In conventional FETs, the growth of high-κ bilayers (AlO_x with HfO_x or ZrO_x) is one strategy to increase the total gate dielectric capacitance and to further scale down the dielectric thickness without diminishing the gate electrostatic control over the channel.⁶⁶⁻⁶⁸ The complementary features of AlO_x and HfO_x (or ZrO_x) (e.g., their dielectric constant, E_g, etc.)⁶⁹ make their bilayer stacks suitable options for gate dielectrics. As far as the MoS₂ FETs are considered, in addition to the increased gate electrostatic control,²⁶ the bilayer approach can better screen the CI scattering in MoS₂.³⁴ Therefore, it is expected that the employment of dielectric bilayers further improves the MoS₂ device metrics. To verify this, stacks of AlO_x and HfO_x were examined on our fabricated devices. The AlO_x interlayer was grown by both ALD and PE-ALD methods and the HfO_x by PE-ALD, all at 100 °C. The deposition conditions were kept similar to those when these films were grown individually on the MoS₂ FETs. The entire stack thickness was also retained at 30 nm while different thickness combinations were investigated.

3.3.1. 2.5/27.5 nm of AlO_x/HfO_x

I–V Characterization of 2.5/27.5 nm AlO_x/HfO_x Bilayer Stack

Figure 5 shows the transfer characteristics of the MoS₂ FETs in semilog scale, capped with 2.5/27.5 nm of AlO_x/HfO_x, where the AlO_x interlayer is grown by both ALD and PE-ALD. For the transfer curves in linear scale, see Figure S6. Compared with the reference, when the bilayer approach is followed, V_T shifts more positively (from ∼40 to ∼70 V, in forward sweep), and I_OFF reduces. These observations indicate that upon the growth of bilayer stacks with the oxygen-rich AlO_x interlayer, MoS₂ is n-type doped to a lesser extent as compared to the reference. However, I_ON slightly improves, which can primarily be attributed to the more effective screening of MoS₂ CIs (as scattering centers) when bilayer stacks are employed.³⁴ In fact, the total dielectric constant of a bilayer stack can be higher than that of a single layer, as verified in previous studies where AlO_x/HfO_x stack resulted in higher dielectric constant than HfO_x.^69,70

Figure 5.

Transfer curves (in semilog scale) of the MoS₂ FETs capped with bilayers of AlO_x and HfO_x (where AlO_x is grown by both ALD and PE-ALD processes).

The average statistical data for both the ON- and OFF-states of these devices also show similar trends (Figure S6).

XPS Analysis of 2.5/2.5 nm AlO_x/HfO_x Bilayer Stack

To evaluate the chemical states at the bilayer high-κ/MoS₂ interface, thin bilayers must be used. The average XPS depth of analysis (using Al Kα radiation source (hv = 1486.6 eV)) is limited to the upper ∼10 nm of a stack. Sputter depth profiling to access a deeper lying interface is also not a preferred option because of the potential reduction of oxidic states upon the sputtering process. Therefore, the XPS analysis is performed on only 2.5/2.5 nm thick ALD AlO_x/PE-ALD HfO_x stacks grown on the MoS₂ films.

Figure 6 illustrates the Mo 3d core level spectra of the studied cases. The S 2p and C 1s spectra are also provided in Figure S7. The data are compared to those of the ∼5 nm HfO_x deposited on MoS₂. As can be seen from the plot, after the growth of both bilayer stacks, the Mo 3d major peak intensities do not drop as dramatically as upon the pure HfO_x growth, inferring that less dielectric (than expected) is grown on MoS₂ with the bilayer approach. This is also evidenced from the higher intensity of the Si 2p spectrum for the bilayer cases (especially the ALD AlO_x) than for the HfO_x counterpart (see Figure S7). As mentioned earlier, a nucleation delay likely occurs upon the growth of AlO_x films on MoS₂ and during the initial deposition cycles, as H₂O and O₂ plasma used in the ALD and PE-ALD processes of AlO_x do not react so strongly with MoS₂ (compared with the O₂ plasma in the HfO_x process). This is the major cause of the higher Mo⁴⁺ peak intensity for the AlO_x/HfO_x stack cases than for the single layer HfO_x counterpart.

Figure 6.

Mo 3d core level spectra after the growth of 2.5/2.5 nm AlO_x/HfO_x films on MoS₂ as compared with 5 nm pure HfO_x on MoS₂.

The AlO_x and HfO_x process specifications are provided in Section S.2 of the Supporting Information for further clarifications.

Furthermore, the Mo⁶⁺/Mo⁴⁺ ratio for the HfO_x and the bilayer cases with ALD and PE-ALD AlO_x interlayers are 0.52, 0.32, and 0.31, respectively (see Table S3). In other words, a portion of the MoS₂ surface is oxidized upon the 2.5/2.5 nm bilayer growth and irrespective of the AlO_x processing method. This is unlike the growth of 5 nm single layer ALD and PE-ALD AlO_x on MoS₂, where the Mo⁶⁺/Mo⁴⁺ is only 0.07, 0.11 (pointed out previously in Table S2). Therefore, one can speculate that the 2.5 nm AlO_x is most likely not a closed layer, such that the HfO_x process can still significantly oxidize the MoS₂ underneath (but evidently less than when only pure HfO_x is employed).

STEM Analysis of the Dielectric Growth on MoS₂

To further elucidate the dielectric coverage on MoS₂, the growth and microstructure of the high-κ/MoS₂ stacks are evaluated by using top-view STEM imaging as well as EDX mapping analyses. Initially, the deposition of merely AlO_x films on MoS₂ is discussed. Then, a study of the bilayers is provided.

Figure 7a shows the high angle annular dark field (HAADF) mode STEM image of 2.5 nm ALD AlO_x on MoS₂ as well as the EDX elemental mapping of Al and Mo and the color-coded overlays, respectively. As can be seen from the HAADF-STEM image, the MoS₂ films are polycrystalline with a grain size of on average ∼70 nm.⁵⁵ The GBs on MoS₂ are known to be more reactive than the basal planes, as they are composed of defects that are arranged linearly. Therefore, AlO_x tends to nucleate predominantly at the GBs rather than on the basal planes, as evidenced from the mapping. This leads to a discontinuous layer of ALD AlO_x on MoS₂ after ∼2.5 nm of growth, which has also been observed previously.^53,54Figure 7b displays the EDX overlay mapping of ∼2.5 nm PE-ALD AlO_x on MoS₂. Overall, the nucleation density of PE-ALD AlO_x on MoS₂ is more than when it is grown by ALD on MoS₂. This is mainly attributed to the use of O₂ plasma during the PE-ALD process and its reactive components that promote the precursor chemical adsorption on both the MoS₂ GBs and its basal planes. However, a completely closed layer of AlO_x is not yet achieved, even with the use of PE-ALD.

Figure 7.

(a) HAADF-STEM image of 2.5 nm ALD AlO_x on MoS₂ as well as the EDX elemental mappings of the Al, Mo, and their overlay image. (b) HAADF-STEM image of 2.5 nm PE-ALD AlO_x on MoS₂ as well as the EDX elemental mappings of the Al, Mo, and their overlay image. (c) HAADF-STEM of 2.5/2.5 nm ALD AlO_x/PE-ALD HfO_x on MoS₂ as well as the EDX elemental mappings of Hf, Al, and Mo.

After realizing that 2.5 nm of AlO_x with neither of the deposition methods results in a closed layer of AlO_x on MoS₂, we evaluated the bilayer high-κ growth on MoS₂. Figure 7c provides the STEM-HAADF image of a 2.5/2.5 nm thick AlO_x/HfO_x stack on MoS₂ as well as the elemental mappings of Hf, Al, and Mo. Here, the AlO_x interlayer has only been grown by the ALD method (and HfO_x by the PE-ALD). As can be seen, although the AlO_x interlayer mostly decorates the MoS₂ GBs, HfO_x is homogeneously deposited almost everywhere. Higher resolution HAADF-STEM images are also provided in Figure S8 for further visualization. The closed nature of the HfO_x layer on MoS₂ strongly implies that the XPS results obtained from the 2.5/2.5 nm AlO_x/HfO_x stack are also representatives for stacks with thicker HfO_x overlayers (e.g., 27.5 nm), which were used for capping the MoS₂ FETs.

Linking All the Characterizations

When correlating the STEM images and the EDX mapping analysis jointly with the XPS measurements, one can comprehend that a 2.5 nm AlO_x interlayer (grown by ALD or PE-ALD) is not sufficient for attaining a closed AlO_x dielectric film on MoS₂. Therefore, when the bilayer approach is followed, HfO_x can still reach the underlying MoS₂ and react with the MoS₂ surface. This explains the increase in the Mo⁶⁺ content upon the growth of both bilayer cases on MoS₂ (compared with when only single layers of AlO_x are grown).

However, the intensity of the Mo⁶⁺ chemical state for the bilayer cases is not higher than what is detected for the pure HfO_x case. The Mo⁶⁺/Mo⁴⁺ ratio for the bilayer cases is ∼0.3 and 0.5 for the HfO_x case (Table S3). In other words, for bilayer cases, the concentration of MoO_x on the MoS₂ surface is less than that for the pure HfO_x case. In such a situation, one would expect a relative increase in the n-type doping of the devices capped with AlO_x/HfO_x stacks, since MoO_x is known as a hole dopant for MoS₂, and its reduced concentration denotes less p-type doping of the devices. However, the reverse behavior is observed. V_T shifts positively, and I_OFF reduces, as compared with the reference device (Figure 5a,b). This implies a relative n-type doping reduction (relative p-type doping increase) upon capping the MoS₂ FETs with dielectric bilayers. As previously pointed out, our AlO_x films are oxygen-rich (grown with both ALD and PE-ALD processes). Furthermore, it has been shown that the excess oxygen states of the AlO_x are energetically located in the vicinity of MoS₂E_v and can dope the MoS₂ to p-type.^7,44 This could be a plausible explanation for relatively lower n-type doping (relatively higher p-type doping) of our devices, despite the lower concentration of MoO_x on MoS₂ upon the growth of bilayers.

3.3.2. 5/25 nm AlO_x/HfO_x

The next logic step is to increase the AlO_x interlayer thickness to 5 nm while maintaining the entire stack thickness at 30 nm.

I–V Characterization of the 5/25 nm AlO_x/HfO_x Bilayer Stack

Figures 8a and 8b compare the transfer characteristics of the MoS₂ FETs capped with 5/25 nm of bilayers with the reference, on linear and semilog scales. As can be seen, the growth of the 5 nm AlO_x interlayer shifts V_T to more positive voltages and reduces I_OFF, relative to the reference, again implying that the underlying MoS₂ has become less n-type doped. The approximate V_T values are marked by straight lines in Figure 8a. Upon the ALD growth of a 5 nm AlO_x interlayer, I_ON improves more than 2-fold and reaches 1.2 μA/μm. In addition, the maximum μ_FE increases to 0.1 cm²/(V·s) on average. I_OFF halves and the I_ON/I_OFF ratio becomes the highest (∼250) for the ALD AlO_x interlayer case, as provided in Figure 8c–e and statistically. The ON-state enhancement is most likely the result of a higher dielectric constant for this specific bilayer combination,^69,70 such that the MoS₂ CIs can be more effectively screened. This leads to the best achieved electrical performance to date for our ALD-based MoS₂ FETs. Considering the PE-ALD AlO_x/HfO_x case, although I_OFF is the lowest, I_ON improves only slightly (see Figure S9), causing the maximum μ_FE only as high as 0.04 cm²/(V·s) and the I_ON/I_OFF ratio of ∼150 (Figures 8c and 8e, respectively). For elucidating these observations, the chemistry involved at the high-κ/MoS₂ interface is again investigated by using XPS.

Figure 8.

Transfer curves of the fabricated MoS₂ FETs, capped with bilayers of 5/25 nm AlO_x/HfO_x in (a) linear and (b) semilog scales. The straight lines in (a) and their interception with the horizontal axis correspond to the approximate values of V_T (c–e) are the average statistical data of the maximum μ_FE, I_OFF, and I_ON/I_OFF, respectively.

XPS Analysis of the 5/2.5 nm AlO_x/HfO_x Bilayer Stack

The XPS analysis is performed after the growth of ∼5/2.5 nm AlO_x/HfO_x combinations on bare MoS₂ films. As can be seen in Figure 9, upon the growth of these bilayer stacks, the Mo 3d major peaks do not drop as notably when the pure PE-ALD HfO_x is grown. As discussed before, unlike the PE-ALD of HfO_x, our AlO_x films experience a nucleation delay on MoS₂ during the initial deposition cycles of both ALD and PE-ALD processes. This results in some deviations between the expected and actual AlO_x thicknesses on MoS₂, subsequently leading to a lower drop in the intensity of Mo⁴⁺ doublet peaks. Evaluating the substrate Si 2p core level spectrum (provided in Figure S10) also confirms that less AlO_x is grown on MoS₂, as compared with the case of HfO_x. In fact, when a thicker dielectric is grown on MoS₂, the Si 2p peak intensity fades out due to the limitations in the XPS average depth of analysis. The total counts per second (CPS) area under the Mo⁴⁺ and Mo⁶⁺ doublet peaks are evaluated as well (see Table S4 for further details). For the pure HfO_x case, the Mo⁶⁺/Mo⁴⁺ ratio is 0.75, whereas for both bilayer cases, it does not exceed 0.22, suggesting that the addition of a 5 nm AlO_x interlayer better protects the MoS₂ surface from being oxidized upon the HfO_x deposition. Here, it is remembered that the Mo⁶⁺/Mo⁴⁺ ratio is ∼0.3 when a 2.5/2.5 nm AlO_x/HfO_x stack is grown on MoS₂ (Table S3).

Figure 9.

Mo 3d spectra after the growth of 7.5 nm HfO_x (PE-ALD) and 5/2.5 nm AlO_x/HfO_x where the interlayer is grown by both the ALD and PE-ALD methods.

Linking XPS Data to I–V

Linking the XPS analyses to the I–V measurements, similar trends can be observed as to when 2.5/2.5 nm bilayer stacks are grown on MoS₂. Again, the Mo⁶⁺/Mo⁴⁺ ratios for the bilayer cases are less than that of HfO_x (evidenced in the XPS), implying that the MoS₂ surface is less oxidized to MoO_x. Therefore, the observed V_T positive shift and the I_OFF reduction in the devices capped with 5/25 nm bilayer dielectrics (relative to the reference) cannot be attributed to the hole injection from the MoO_x, and they are mainly due to the oxygen-rich states of AlO_x that reside in close proximity of the MoS₂E_v.

3.3.3. 10/20 nm AlO_x/HfO_x

It is also worth evaluating the 10/20 nm dielectric stack on the fabricated MoS₂ FETs. Figure 10a shows the transfer curves of the devices capped with the ALD AlO_x/PE-ALD HfO_x of 2.5/27.5, 5/25, and 10/20 nm combinations and the reference. Here, we only tried the ALD of AlO_x (and not the PE-ALD of AlO_x) as we deduced that the ALD of AlO_x best protects the MoS₂ surface from being oxidized to MoO_x. As can be seen from Figure 10a, by raising the AlO_x interlayer thickness to 10 nm, I_ON improves nearly 1 order of magnitude and reaches 4.7 μA/μm, as compared with the reference. A similar trend can also be observed for the mean maximum μ_FE, which is shown in Figure 10b. The statistical data for other device metrics are also provided in Figure S11. Despite the improvements in the device ON-state performance, V_T shifts negatively, and I_OFF degrades dramatically for the 10/20 nm bilayer case, implying that our devices become degenerately n-type doped upon the growth of 10/20 nm dielectric stack. As a result of this excess n-type doping, the back-gate bias fails to properly deplete the channel in the OFF-state regime. This suggests that when a 10/20 nm dielectric stack is grown on the MoS₂ FETs, a surface channel is formed.³¹ In fact, only the topmost layers of the MoS₂ are electrically active, which leads to a weak back-gate electrostatic control over the channel.

Figure 10.

(a) Transfer curves of the fabricated MoS₂ FETs capped with bilayer dielectrics of different AlO_x interlayer thicknesses. The AlO_x interlayer is grown by ALD. (b) Mean maximum μ_FE and its dependence on the AlO_x interlayer thickness.

From the material point of view, growing 10 nm of the ALD AlO_x interlayer uniformly covers the underlying MoS₂. This is shown in the STEM top view images provided in Figure S12.

Stability and Repeatability of the Optimal Devices

Based on the above discussions, among all the studied cases, the optimal device performance is obtained when 5/25 nm of ALD AlO_x/PE-ALD HfO_x is grown on the fabricated MoS₂ FETs. It is also worthwhile evaluating the stability and repeatability of the data obtained from such devices. Figure 11a compares the I–V performance of the same device after eight months of being initially measured. As evidenced, the device metrics are persistent over a long period of time, indicating a high degree of device stability when capped with such a dielectric stack. The I–V performance of a device fabricated few months after, on another wafer and with similar processing conditions is also shown in Figure 11b. The data verify that the obtained results for devices capped with bilayers of such a thickness combination (5/25 nm) are repeatable.

Figure 11.

(a) Initial transfer characteristics of a fabricated MoS₂ FET, capped with 5/25 nm of ALD AlO_x/PE-ALD HfO_x dielectrics, and after eight months of being remeasured. (b) Transfer characteristics of two devices made with same processing conditions on two different wafers.

4. Conclusions

To conclude, this work extensively investigates the integration of high-κ dielectrics with ALD-based MoS₂ FETs by means of electrical, surface-chemical, and material characterization techniques.

The electrical characterization data show that the employment of high-κ dielectrics improves the overall device metrics in ALD-based MoS₂ FETs. While other studies have associated such improvements mainly with the effective screening of MoS₂ charged impurities (CIs) by a high-κ dielectric as well as the dielectric doping effect, we show that three additional parameters related to the processing of the dielectrics can simultaneously govern the MoS₂ FET electrical performance and control its doping. These are the stoichiometry of the dielectrics, the carbon impurity content in the dielectrics, and the oxidation degree of the MoS₂ top surface upon the dielectric growth.

Despite the fact that our AlO_x films are oxygen-rich, which typically leads to p-type doping in MoS₂ FETs, an overall n-type performance is observed in the devices capped with these dielectrics. Surface chemical analysis reveals that carbon-related species are incorporated into the AlO_x films when grown at 100 °C, being one plausible explanation of the overall observed n-type characteristics in MoS₂ FETs capped with oxygen-rich AlO_x. Among the single layer dielectrics studied, the PE-ALD AlO_x films exhibit the most optimal device electrical performance, as the resulting devices exhibit a high ON-state current while maintaining low OFF-state characteristics.

Surface-chemical analyses further show that the PE-ALD of HfO_x and AlO_x dielectrics oxidize the MoS₂ surface to MoO_x, whereas the (thermal) ALD AlO_x leaves the underlying MoS₂ surface almost intact. Therefore, if a layer of (thermal) ALD AlO_x is employed between MoS₂ and PE-ALD HfO_x, device metrics improve further. Among the bilayer cases, the 5/25 nm (thermal) ALD AlO_x/PE-ALD HfO_x dielectric stack leads to the most optimal device electrical characteristics. We relate this to the overall higher dielectric constant of the bilayer stack as well as a fair protection of the MoS₂ surface against oxidation upon the growth of HfO_x. The MoS₂ FETs capped with this bilayer combination also show a high degree of stability over time with repeatable results.

Our findings and the connections we make between the surface-chemical, material, and electrical analyses provide guidelines to further optimize and improve the properties of synthetic-based MoS₂ FETs of polycrystalline MoS₂ nature.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsanm.4c02214.

• MoS₂ FET optical microscopy top-view and cross-sectional schematics; process specifications for ALD and PE-ALD of AlO_x and HfO_x; HfO_x capping layer on MoS₂ (XPS and Raman data analyses); AlO_x capping layer on MoS₂ (MoS₂ FET device metrics and XPS data analysis); bilayer AlO_x/HfO_x capping layer on MoS₂ (MoS₂ FET device metrics, XPS analyses, HR-STEM images); references (PDF)

The authors declare no competing financial interest.