Enhanced Charge Carrier Mobility in Two‐Dimensional High Dielectric Molybdenum Oxide

Abstract

We demonstrate that the energy bandgap of layered, high-dielectric α-MoO3 can be reduced to values viable for the fabrication of 2D electronic devices. This is achieved through embedding Coulomb charges within the high dielectric media, advantageously limiting charge scattering. As a result, devices with α-MoO3 of ∼11 nm thickness and carrier mobilities larger than 1100 cm2 V−1 s−1 are obtained. In atomically thin two-dimensional (2D) materials, free charges have quantized energy levels in one spatial dimension, while they are mobile in the other two.1 The interest in such 2D materials increased in the late 1970s and early 1980s driven by a large amount of experimental outcomes from the development of molecular beam epitaxy (MBE) equipment for the deposition of high quality thin films of III-V semiconductors.2 This resulted in the realization of high electron mobility transistors, with thin films of different bandgaps that can operate at very high frequencies. In such devices, the conduction band energy of at least one of the films is forced under the Fermi level at the junction, creating a narrow quantum well, and the quantization of free charges.3 Such structures can offer electron mobilities larger than 106 cm2 V−1 s−1 (at near zero Kelvin temperatures), which is significantly larger than that of silicon.4 However, the high cost of the rare-earth materials, lack of compatibility, and technological difficulties for creating III-V structures hinders their widespread adaptation. The emergence of graphene has revived the interest in 2D materials, due to its many favorable properties including enhanced electron mobilities that exceed 105 cm2 V−1 s−1.5 However, graphene lacks the semiconducting characteristics of prevalent materials, such as silicon with their natural energy bandgap.5, 6 Even though there have been advancements in introducing a bandgap to graphene,7 the required complex synthesis processes always resulted in significant loss in the much desired carrier mobilities.8 It is suggested that alternative layered materials such as semiconducting transition metal dichalcogenides (e.g., molybdenum disulfide–MoS2 – one of the most researched in this group) would solve the issue of introducing a bandgap, but still results in relatively low carrier mobilities.9, 10 Recent reports on atomically thin MoS2 indicate possible approaches to increase its mobility by introducing a high dielectric top-gate material (HfO2) that reduces Coulomb scattering to attain values smaller than 220 cm2 V−1 s−1, which is still less than values sought after.10 MoO3 crystal structure and intercalation for sub-stoichiometry. (a) 1.4 nm octahedral structures held together by van der Waals forces. (b) Optical microscopy images showing evolution of a thin MoO3 flake with duration of H+ intercalation. (c) Micro-Raman spectra corresponding to optical images in (b). (d) Photon energy vs. (αhv)2 curves derived from the absorbance spectrum of a MoO3 flake with progression of catalyzed H+ intercalation. Intersection of the linear fit with the x–axis indicates the bandgap energy, used to depict the change in estimated optical bandgap with progression of the intercalation process (inset). The measurement for (d) was performed by intercalating thicker flakes in order to observe changes in the absorbance spectra. MoO3, in its intrinsic form, has a wide bandgap (>2.7 eV) resulting in a low carrier concentration, which is not suitable for electronic devices.17 It is well known that α-MoO3 can be reduced to form MoO(3–x) and increasing x can lower the bandgap.18, 19 We have previously demonstrated that intercalating ions such as H+ can interact with oxygen atoms of α-MoO3, gradually leaving the material as H2O vapor, and result in oxygen deficiencies.20 In low-κ materials, this process increases the number of embedded Coulomb charges and has a possible detrimental effect on electron scattering. On the contrary, according to Equation 1, as α-MoO3 is a layered high-κ material, the Coulomb scattering effect is minimal and acoustic phonon scattering has the most impact at room temperature.2 We extensively investigated the reduction process of α-MoO3 flakes to gain perfect control of the Coulomb charge incorporation process. Flakes of stratified α-MoO3 was synthesized through thermal evaporation of MoO3 powder, followed by mechanical exfoliation to realize perfect α-MoO3 planar crystals (Figures S2–S4).11 We studied the intercalation in these layers via their exposure to H+ ions.20 Similar intercalation behavior was also observed upon exposure to H2 gas, as H2 readily breaks onto the MoO3 surface to form H+ ions (Figures S5 and S6). However, the H2 gas exposure procedure has an advantage as HyMoO3 readily loses H2O to ambient, producing the desired sub-stoichiometry MoO(3–x).20 The electronic behavior of HyMoO3 and MoO(3–x) are similar as they both have the same number of Coulomb charges. Similarly, highly energetic beams and annealing in a high vacuum environment can also reduce the material.21 So any of the aforementioned methods can be equally used for obtaining MoO(3–x). In brief, by introducing the oxygen vacancies, Mo6+ neighboring the oxygen vacancies in the MoO3 lattice would be reduced to Mo5+, injecting an electron to the conduction band.18, 21 The additional electron from such Mo5+ is delocalized within the layers, since it is not tightly bound to any Mo6+ core cation. This will give rise to a gap state in between the valence and conduction bands of MoO3 hence narrowing its band gap. Optical images of an α-MoO3 flake (Figure 1b) show the progress of the liquid phase intercalation with time, generating a change of appearance from transparent to Prussian blue. Raman spectra of the same flake reveal the evolution of the MoO(3–x) from the α-MoO3 (Figure 1c), based on the fact that the sharp peak of MoO3 at 820 cm−1 gradually disappeared, and a broad peak at ∼780 cm−1 emerged instead. The Raman peak shift at 780 cm−1 is a strong indication for the presence and strength of sub-stoichiometric crystal formation.20 The bandgap reduced from 3.25 to 2.65eV after exposure to H2 gas (Figure 1d) and to smaller values less than 2 eV afterwards (which can be extrapolated). Collectively, these results confirm the presence of sub-stoichiometric MoO(3–x), and the ability to reduce bandgap as a function of x. In this work, the aforementioned process is adopted to reduce stoichiometry in thin layers of molybdenum oxide, in order to produce a 2D layer with reduced bandgap and large electron concentration, hence making it suitable for FET applications. The flakes were mechanically exfoliated onto 300 nm SiO2 on highly p-doped Si substrates as well as on Si substrates with an optically smooth gold layer. Transmission electron microscopy (TEM) image of a flake before exfoliation (Figure 2a) and scanning tunneling microscopy (STM) image after the exfoliation (Figure 2b) show parallel lattice fringes with spacing of 0.39 nm, which is also in agreement with the STMs for both before and after losing H2O.(see supporting information) These indicate perfect crystallinity of the layers and also perfectly matched crystal lattices of MoO3 and MoO(3–x). Characterization of MoO3 and MoO(3–x) flakes and the configuration of FETs. (a) TEM image of an exfoliated MoO3 crystal with inset showing magnified lattice arrangement with parameters matching α-MoO3 (ICDD No: 35-0609). (b) STM image of a sub-stoichiometric MoO(3–x) flake with similar lattice spacing to that of MoO3 and inset of (a). This also indicates conductivity of the flake. (c) AFM scan of a MoO3 FET fabricated by electron beam direct writing with the corresponding thickness profile of the flake. The measured thickness of ∼11 nm indicates the presence of eight fundamental layers in MoO3. (d) Illustration of I–V characterization setup. (e,f) Cross-sectional schematic of formation of sub-stoichiometric states in MoO3 as a result of electron beam writing and intercalation. (g) Micro-Raman spectra of the MoO3 flake in (c) with broad peak at 780 cm−1 (absent in background measurement of substrate) indicating presence of MoO(3–x). In order to fabricate field effect transistors (FETs) with thin layered oxide and assess the charge carrier mobility, the thickness of the exfoliated layers were identified using scanning electron microscopy (SEM) and atomic force microscopy (AFM). Platinum/gold electrodes were formed using a combination of electron beam lithography and photolithography. See Materials and Methods in the supporting information for details of the fabrication process. Figure 2c depicts the AFM image of a FET with a ∼11 nm flake comprised of 8 fundamental atomic layers of MoO3, each with the thickness of approximately 1.4 nm.11 This was the thinnest flake, which could be incorporated into a device, as our electron beam had a detrimental effect on the formation of contacts to thinner samples. Many devices were fabricated with varying MoO3 thicknesses from several microns to the minimum of 11 nm, with and without the application of electron beam lithography, to benchmark the performance of the devices and contacts (Figures S10 and S12). The material in the structure of the FETs were reduced (Figures 2d–2f), generating oxygen vacancies resulting in the formation of sub-stoichiometric MoO(3–x). Raman spectra of the 2D layer, used in the FETs (Figure 2g), also contains the expected Raman peak shift feature at 780 cm−1, corresponding to that observed in MoO(3–x) (Figure 1c). The experimental I–V characteristics of the MoO3 FET. (a) IDS vs. VDS characteristics of the FET with varying back-gate voltages (VGS) in steps of 0.4 V from -2 to +2 V. (b) Corresponding IDS–VGS curves of the FET acquired at VDS values of 20, 50, and 100 mV. The enhanced performance of our FETs can be explained by the physics of high-κ layer dominating the Coulomb scattering effect, as extracted from Equation 1. However the theoretical mobility is slightly larger than that of the measured. This can be due to the surface roughness of SiO2 (Figure S14), which increases the roughness scattering effect. Something which is of importance is the formation of near Ohmic contacts under the electron beam established Pt electrodes. Our Raman measurements show that the highly energetic beam produces a highly reduced MoO(3–x) layer under and near the Pt electrodes, which is almost metallic (Figure S15). During the FET operation, electrons can be easily transferred from Pt to this near metallic layer, which can then be further injected down into the underneath layers through van der Waals gaps. This electronic charge transfer from the conducting non-stoichiometric layer to neighboring more stoichiometric layers has been confirmed by density functional theory (DFT) simulations. As shown in Figure 4, the conduction states in neighbor layers L1, L2, L1′ and L2′ are pushed down below the Fermi level resulting in reduced energy gaps between these layers' valence states and conduction states (Figures S16 and S17). Thus, while the lattice matching remains perfect, the free charge carriers can be exchanged between the layers and eventually guided within the planes. Density functional theory (DFT) simulations of charge injection in layered α-MoO3. (a) Schematic illustrating ball and stick atomic model of bulk α-MoO3 with large, medium and small spheres denoting Mo, O, and H atoms, respectively. Only five MoO3 layers are shown for clarity: Layer L0, where hydrogen atoms are adsorbed near the oxygen atom located at the corner of MoO3 octahedra, nearest layers L1 and L1′, and next-nearest layers L2 and L2′. (b) Partial density of states projected on each atomic site and summed for each layer. For nearest and next-nearest layers, ΔG denotes reduction in the band gap from their bulk, defect-free state. We showed that using the Born approximation equation, one can predict and design the electronic characteristics of a 2D semiconducting metal oxide. We theoretically demonstrated that the 2D α-MoO3 has high carrier mobility. However, the natural bandgap of α-MoO3 is so wide, that it does not thermionically generate enough charge carriers, required for the operation of FETs. The Coulomb charges were used in order to reduce the bandgap of the 2D MoO3. These charges can be introduced by either intercalated H+, or the reduction of MoO3 to MoO(3–x). This process produces gap states, resulting in the reduction of the materials bandgap. In low-κ materials, the incorporation of Coulomb charges decreases the carrier mobilities due to the Coulomb scattering effect. However, in high-κ MoO(3-x) the Coulomb scattering effect is minimized and the carrier mobility still remains high. Additionally, the Coulomb charges also generate near Ohmic contacts at the drain and source electrodes, that allow the injection of electrons into the FET channel. Here, we exfoliated α-MoO3 and reduced to MoO(3–x). By implementing a ∼11 nm thick 2D high-κ semiconductor, we were able to reduce the Coulomb scattering potential effect significantly, which resulted in the carrier mobility values >1100 cm2 V−1 s−1 that readily exceeded those of doped and low dimensional silicon24 and that of single layer MoS210 (see supporting information for a comparison of the mobilities for different materials). Well known procedures for the large scale synthesis and manipulation of 2D MoO3 and many other transition metal oxides,13 as well as their relative abundance in nature and their great optical and catalytic properties, renders them ideal building blocks of future electronic and optical devices, sensors, and integrated circuits. It is also suggested that the work to be expanded to other low dimensional MoO3 structures, such as nanorods and nanobelts in the future.25 Supporting Information is available from Wiley Online Library or from the author. SS and MB acknowledge the Australian Post-Doctoral Fellowships from the Australian Research Council through Discovery Projects DP110100262 and DP1092717, respectively. SS and KKZ acknowledge the Australian Research Council for equipment funding through the Linkage, Infrastructure, Equipment, and Facilities Grant LE100100215. The work was partially supported by the CSIRO Sensors and Sensor Networks Transformational Capability Platform and CSIRO Materials Science and Engineering Division. JD and NM thank Philip Chan, the Monash HPC cluster, and the MASSIVE and NCI facilities funded by the Australian Government for their support for computational resources. The authors also would like to acknowledge the facilities, and scientific and technical assistance, of the Australian Microscopy & Microanalysis Research Facility at the RMIT Microscopy & Microanalysis Facility, at RMIT University. As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials are peer reviewed and may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.