Ion Migration Studies in Exfoliated 2D Molybdenum Oxide via Ionic

Jun 11, 2018 - (20,29,30) This material is well-known for its large work function(31 ..... of Basic Energy Sciences (BES), Materials Sciences and Engi...
0 downloads 0 Views 4MB Size
Research Article Cite This: ACS Appl. Mater. Interfaces 2018, 10, 22623−22631

www.acsami.org

Ion Migration Studies in Exfoliated 2D Molybdenum Oxide via Ionic Liquid Gating for Neuromorphic Device Applications Cheng Zhang,†,§ Pushpa R. Pudasaini,†,§ Akinola D. Oyedele,§,‡ Anton V. Ievlev,§ Liubin Xu,† Amanda V. Haglund,† Joo Hyon Noh,† Anthony T. Wong,† Kai Xiao,§ Thomas Z. Ward,∥ David G. Mandrus,†,∥ Haixuan Xu,† Olga S. Ovchinnikova,§ and Philip D. Rack*,†,§

ACS Appl. Mater. Interfaces 2018.10:22623-22631. Downloaded from pubs.acs.org by UNIV OF NEW ENGLAND on 01/11/19. For personal use only.



Department of Materials Science and Engineering and ‡Bredesen Center for Interdisciplinary Research and Graduate Education, University of Tennessee, Knoxville, Tennessee 37996, United States § Center for Nanophase Materials Sciences and ∥Materials Science and Technology Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831, United States S Supporting Information *

ABSTRACT: The formation of an electric double layer in ionic liquid (IL) can electrostatically induce charge carriers and/or intercalate ions in and out of the lattice which can trigger a large change of the electronic, optical, and magnetic properties of materials and even modify the crystal structure. We present a systematic study of ionic liquid gating of exfoliated 2D molybdenum trioxide (MoO3) devices and correlate the resultant electrical properties to the electrochemical doping via ion migration during the IL biasing process. A nearly 9 orders of magnitude modulation of the MoO3 conductivity is obtained for the two types of ionic liquids that are investigated. In addition, notably rapid on/off switching was realized through a lithiumcontaining ionic liquid whereas much slower modulation was induced via oxygen extraction/intercalation. Time of flight− secondary ion mass spectrometry confirms the Li intercalation. Density functional theory (DFT) calculations have been carried out to examine the underlying metallization mechanism. Results of short-pulse tests show the potential of these MoO3 devices as neuromorphic computing elements due to their synaptic plasticity. KEYWORDS: 2D materials, neuromorphic device, ionic liquid, transition metal oxides, ion migration



materials, as was demonstrated by Schladt et al. in TiO210 and by Li et al. in SrTiO3.11 More studies have been subsequently performed to investigate the ion migration between the IL and the sample, which induces drastic electrical property changes. The relationship of the gate induced metallization and the crystal orientation in VO2 was reported by Jeong et al., and a significant expansion of the lattice was observed which was accompanied by the metallization.12 Interestingly, it has been found in WO3 thin films that an IL electric field induces the migration of only a small amount of oxygen can lead to major changes to the structural and electrical/optical properties.13

INTRODUCTION

Ionic liquids (IL) have been extensively studied to investigate electrostatic effects on fundamental physics problems in strongly correlated systems.1−3 An electric double layer is formed at the IL−channel interface, supporting high electric fields, which have been used in field-effect transistor (FET) devices to electrostatically gate carrier densities.4,5 The gate bias induced changes of several different materials were initially reported to be due to electrostatic charge accumulation. Examples include the transition from paramagnetism to ferromagnetism found by Yamada et al. in Co-doped TiO2 and the formation of the metallic state or even superconducting state found in ZrNCl, MoS2, and SiTiO3.3,6−9 Later studies, on the other hand, have shown that the formation of oxygen vacancies is the cause of the metallization in some © 2018 American Chemical Society

Received: April 9, 2018 Accepted: June 11, 2018 Published: June 11, 2018 22623

DOI: 10.1021/acsami.8b05577 ACS Appl. Mater. Interfaces 2018, 10, 22623−22631

Research Article

ACS Applied Materials & Interfaces

Figure 1. Electrical transfer (a) and output (b) measurement results for pristine MoO3 FET devices via bottom SiO2 gate biasing. The MoO3 behaves as an insulator and experiences no gate modulation in the current between ±60 V. Inset in (b) is a schematic of a MoO3 flake patterned with source/drain and bottom gate electrodes. The dimensions of the channel is 5 × 4 μm2.

Figure 2. Metallization of a typical MoO3 device through IL bias over time. The insulator-to-metal transition is reversible, and the transfer curves measured at each gate voltage/time increment are shown from left to right: +1 V IL bias for up to 60 min (a), −1 V bias for up to 30 min (b), −2 V bias for up to 60 min (c), and +2 V bias for up to 80 min (d). The original off-to-on current ratio of the Id is ∼106, and the reversible switching range was from ∼60 μA to 300 nA. All measurements were conducted with Vsd = 1 V. An optical micrograph of a MoO3 side gate device is shown as an inset in (a).

(trifluoromethanesulfonyl)imide) and LiClO4−PEO. For the BMIM/Tf2N IL, oxygen anions are extracted and intercalated, whereas the lithium IL intercalates and extracts Li cations in and out of the MoO3 lattice. The Li cations are shown to have much faster ion mobility and thus much faster current modulation. Both metallization processes can in principle act as a nonvolatile memory, and the different intercalation rates could facilitate interesting neuromorphic computing platforms.13,34,35 Time of flight−secondary ion mass spectrometry (ToF-SIMS) has also been employed to correlate the ion migration resulting from the gating process.

Recently IL electrostatic gating14,15 and electrostatic oxygen extraction16,17 were used to modulate the ON/OFF current and electrically activate amorphous InGaZnO thin film transistors. Additionally, in situ AFM imaging of the HMIMTF2N IL in TFT devices recently elucidated the IL−cation interface structure to the associated IGZO conductance.18 Here we present our investigation of IL gated molybdenum trioxide, MoO3, which was exfoliated from a bulk single crystal. Transition metal oxides MoO3 and WO3 have been studied as charge injection layers in organic semiconductor devices,19 such as light-emitting diodes4,20 or organic solar cells.21,22 MoO3 in particular has garnered attention for gas sensing,23,24 biosensing,25,26 energy storage27,28 applications, and the potential in optoelectronic and resistive switching devices.20,29,30 This material is well-known for its large work function31,32 and has a monoclinic structure at room temperature with a band gap of ∼3.1 eV.33 It consists of two Mo atoms in the bilayer which is separated by three oxide atoms. The stoichiometric oxide is insulating, and no insulator to metal transition with temperature has been reported. In this report, we show that the MoO3 can be metallized via two types of ILs: BMIM-Tf2N (1-butyl-3-methylimidazolium bis-



RESULTS AND DISCUSSION MoO3 thin flakes were mechanically exfoliated from the bulk single crystals and contacted with Ti/Au electrodes (see details in the Experimental Section). The bottom-gate (through the common silicon substrate) device configuration is shown in Figure 1b, inset. Prior to IL gating, devices were measured to obtain the baseline properties. Figure 1 shows the measurement result for an as-fabricated MoO3 bottom-gated device. The device behaves as an insulator, and no drain current (Id) modulation was observed with the application of a gate voltage 22624

DOI: 10.1021/acsami.8b05577 ACS Appl. Mater. Interfaces 2018, 10, 22623−22631

Research Article

ACS Applied Materials & Interfaces

Figure 3. Id vs time for pulsed bias applied through LIL gate. With a +2 V LIL gate bias pulse Id increased from pA level to over 1 mA within 50 s. When the gate voltage is turned off, Id decays gradually to ∼10 μA. When the second +2 V pulse is applied, Id approaches mA currents with a slightly slower ramping rate. The conductivity reverses with a −2 V gate pulse and slowly decrease its Id to ∼1 nA in ∼45 min. When the negative pulse is removed, the Id returned to μA level. Insets are magnified regions at on/off points on a linear scale.

between −60 and 60 V, as shown in the transfer curves in Figure 1a. The Id of the device is below 10−8 A at an S−D voltage up to 4.1 V. Figure 1b shows the output curves at various gate voltages. Figure 2 shows examples of the tunable metallization process for the exfoliated MoO3 using BMIM-Tf2N IL, as schematically shown in Supporting Information Figure S1a. After 30 min of +1 V IL biasing, the Id increased more than 3 orders of magnitude from 0.1 nA to μA level, as shown in Figure 2a. The conductance continues to increase to ∼60 min, where the current reaches over 80 μA. The total increase of the channel current is close to 6 orders of magnitude during the gateinduced metallization. A negative bias voltage was then applied to investigate the reversibility of the process where Figures 2b,c show the results via different gate voltages and times. When −1 V bias was applied, Id starts to decrease and tended to saturate at ∼20 μA in 30 min. When the bias voltage was increased to −2 V, the process was accelerated, and the Id was suppressed to sub-μA levels, though still much higher than the pristine device current. This partially reversible process allows the conductivity adjustment via IL bias voltage and time. A +2 V bias was then applied to confirm the reversible nature of the insulator/metal transition. Raman spectroscopy was carried out before and after this series of gating experiments, and the result is shown in Figure S1c. The intensities of characteristic peaks decrease, but no obvious shift is observed, which indicates that the major structure of the MoO3 remains the same. The tunable conductivity through IL gating we observed in MoO3 is similar to what was recently observed in WO3 metallization.13 Also, the ion exchange during the IL gating process has been previously studied on VO2 thin films by Jeong et al. using an 18O isotope.2 Both studies demonstrate that the oxygen migration and the subsequent oxygen vacancies play a critical role during the metallization process. Also, a recent report by Yang et al. shows that MoO3 IL devices (using the BMIM-TF2N) can be metalized via IL biasing.35 Notably, they demonstrated a very small current modulation under vacuum,

at albeit shorter gating times. Current modulations as a function of the relative humidity revealed a correlation in the achievable current modulation to the water content, and the positive and negative current modulation was attributed to H+ and OH− intercalation, respectively. The positive IL gate voltage induced an increase in the channel conductance and was attributed to an O2− cation reacting with intercalated H+ to form a hydroxyl (OH−) bond and a concomitant Mo6+ to Mo5+ valence change, thus electron doping the conduction band. A negative gate bias (at an albeit higher voltage) reduced the channel conductance and was attributed to removal of the H+ via OH− accumulation at the channel. As our measurements were performed in a vacuum, we propose the conductivity change in MoO3 is a result of oxygen migration into and out of the MoO3 via IL gating. The data suggests that the metallization (generating oxygen vacancies) process is faster than the reversible (oxygen intercalation) process. In fact, based on this mechanism, the metallization can also be realized by repeatedly cycling the gate voltage. The current change as a function of the number of IL gate cycles is shown in Figure S2. During the asymmetric continuous scanning, the voltage sweep rate was set at 1 V/min from −1 to 2 V. In this process the device was metallized similar to the applying a constant bias over time, with a comparatively lower Id when saturated. The inset of Figure S2 shows the Id at zero gate bias versus the cycle number. In this case, larger positive charging voltage and faster oxygen extraction rate results in the generation of oxygen vacancies which induce the metallization. As oxygen anions can migrate in and out of the MoO3 via IL gating, we explored whether lithium cations could also be exchanged. Thus, we employed LiClO4−PEO, a lithiumcontaining ionic liquid (LIL; see the Experimental Section for the synthesis details). The samples with LIL were characterized similar to the BMIM/Tf2N. By applying multiple gate sweeps using the LIL, a similar metallization process was observed as shown in Figure S3. However, devices showed much higher sensitivity when constant bias voltage was applied; Id increased 22625

DOI: 10.1021/acsami.8b05577 ACS Appl. Mater. Interfaces 2018, 10, 22623−22631

Research Article

ACS Applied Materials & Interfaces

Figure 4. ToF-SIMS mapping of a MoO3 device after applying +1 V bias through LIL for 1 min. (a) shows the top down view (xy plane) of the device which is the sum of the Mo counts in the entire depth profile which shows the geometry of the flake. The bright regions denote the metal contact regions which appear artificially bright perhaps due to a charging artifact. The cross-sectional scanning images (xz plane) along the blue dotted line are shown in (b) showing the Li, Mo, and O intensity and a colored overlay, denoting green for Li, blue for Mo, and red for O. (c) shows the xy scan of the three kinds of atoms at different depths in the device, as indicated by red dashed lines. The oxygen intensity from outside the flake comes from the SiO2 layer. The thickness of the flake is 60 nm.

very fast, and microstructural changes in the MoO3 flake tended to break down the device after a very short constant positive gating time (see Figure S3). Thus, we studied the electrical characteristics of the LIL via a “pulsed bias” method. Figure 3 shows the Id as a function time with various ±2 V gate pulses, with on/off points indicated. The pristine device is initially insulating with an Id in the pA current level. After biasing the gate with +2 V, the MoO3 quickly metalizes to an Id current level of greater than 1 mA. This 9 orders of magnitude increase happens in around 50 s, which is significantly shorter than the BMIM/Tf2N without lithium. When the gate voltage is turned off, the Id drops immediately by an order of magnitude (60 s) and gradually decays to ∼10 μA in 1 h. A +2 V gate was again applied to the MoO3, and the current again increased to 1 mA, but at a slightly slower rate; the second pulse required 90 s for Id to increase from 10 μA to 1 mA. Interestingly, the subsequent decay in the conductivity is less after the second pulse as the steady state value approaches 170 μA. A −2 V pulse was then applied on the device, and its Id was suppressed from 170 μA to 1 nA in ∼45 min. After the pulse was removed, the Id of the device recovered and saturated at the μA level. Figure 4 insets are magnified linear plots of specific regions. The electrical measurement result shows that the MoO3 device has a much stronger response to LIL gate bias compared to the BMIM/Tf2N IL. The metallization of the MoO3 is attributed to Li+ intercalation during the gating process in addition to the possibility of O2− migration; of course, the polarity difference in the O2− anion and the Li+ cation would both increase the conductivity with positive gate bias (O2− vacancies and Li+ doping, respectively) and decrease the conductivity with negative gate bias (O intercalation and Li+ extraction, respectively). ToF-SIMS was employed to

investigate the ion migration after the LIL gating. Figure 4 shows the 3D ion maps of a MoO3 device after applying +1 V bias through LIL for 1 min. Figure 4a is the plan view of an ∼60 nm thick flake with the intensity corresponding to the sum of Mo atoms in the entire sputter depth profile in the xy plane. The bright regions denote the top metal contact regions which appear bright, perhaps due to a charging artifact. The upper left “panhandle” of the flake appears consistently darker, which could be a result of the MoO3 flake fracturing and thus another charging artifact. Three monochrome images in Figure 4b represent the distribution of Li, Mo, and O along the depth of the device, collected by plotting the xz plane along the blue dotted line in Figure 4a. The bottom right panel is the RGB overlay of the three atoms, with Mo in red, Li in green, and O in blue. The black regions on the flake result from the Au/Ti contact region, and thus the signals are delayed during the simultaneous sputtering through this layer. Compared to the Mo and O, the Li distribution shows an obvious inhomogeneity: most of the Li aggregates near the sample surface and exhibits a decay along the depth direction of the sample. The migration of Li from the liquid to the flake is clearly observed by this concentration change, which is also supported by the higher intensity at the edges of the flake. It is also clear that in this brief pulsing the Li has not diffused under the metal contact region. This Li migration is also observed from the xy plane images at different depths, as shown in Figure 4c, where each set of images is collected along the corresponding red dashed line indicated in Figure 4b. At the surface of the device, the planar distributions of all three kinds of atoms are homogeneous. As the sputtering depth increases, the intensity of Li starts to decrease, as shown in the middle panels, with the largest contrast change among three of them. When scanning near the bottom of the sample, the Li counts 22626

DOI: 10.1021/acsami.8b05577 ACS Appl. Mater. Interfaces 2018, 10, 22623−22631

Research Article

ACS Applied Materials & Interfaces

Figure 5. Quantitative analysis of the Li distribution along the depth of three devices after different LIL biasing process. Sample A was partially charged at +1 V for 1 min. Sample B was charged at +2 V for 20 min. Sample C was initially charged at +2 V for 20 min, and then Li was extracted via a −2 V for 1 h. The Li/Mo atom ratios was obtained from the intensity of 3D ion mapping images using ToF-SIMS. Thickness of sample A, B, and C is approximately 60, 50, and 100 nm, respectively. Sample A is the same device presented in Figure 4, with Li highly concentrated at the top surface of the sample, while other two devices exhibit comparatively uniform Li distribution along depth.

Figure 6. (a) Two energy-favorable Li+ sites in α-MoO3 and band structures of intralayer Li+ ion in a (b) 3 × 1 × 3, (c) 2 × 1 × 2, and (d) 1 × 1 × 1 α-MoO3 supercell. The Fermi energy is set to 0 eV. PBE+DFT-D2 was employed to obtained the relaxed ionic configurations, based on which GGA+U scheme was imposed to plot the band structures (see Computational Section for details). (b), (c), and (d) have the Li/Mo ratio of 0.028, 0.063, and 0.25, respectively. At low Li/Mo ratios, supercells containing Li+ interstitials are insulating; as the ratio increases, the band gap is gradually reduced and the system transitions to a metallic state.

ratio along depth changes very little within the first 60 nm and then slowly decays. The xz Mo distribution image is also shown, demonstrating the shape of the flake. It is easier to see from the atom ratio mapping that Li/Mo remains comparatively homogeneous throughout most of the MoO3. The average Li/Mo value is clearly reduced compared to that of in sample B, close to 0.8 within 60 nm from the surface. It is interesting that unlike the intercalation process, the removal of Li+ seems to occur throughout the MoO3 flake. A 3D spatial distribution of Li/Mo ratio in these three samples is shown in the Supporting Information (Figure S4) to give a more intuitive view. Similar analysis for Li/O was also conducted ToF-SIMS, but no clear evidence for O migration was observed in LIL series samples. Thus, we attribute the significant electrical modulation of the LIL gating to the Li+ intercalation.

are nearly gone except at the edge, which illustrates that the intercalation occurs on all the edges of the sample. Additional ToF-SIMS study was conducted on devices biased differently to further investigate the ion migration mechanism. MoO3 devices were treated with different LIL biasing recipes, and the atom ratios of Li/Mo are plotted in Figure 5 as a function of the device depth. Sample A corresponds to that presented in Figure 4, which was biased at +1 V for 1 min. The Li/Mo ratio reaches ∼2 at its top surface and decays toward zero approximately midway through the flake. Sample B was subjected to a +2 V bias for 20 min, and the averaged Li/Mo ratio reaches ∼3 and is evenly distributed through the entire thickness. From the xz plane image it appears that the Li intercalates laterally under the electrodes in the flake and distributes evenly also at the edge. Sample C was first biased similar to sample B (+2 V, 20 min) and then biased at −2 V for 1 h to show the profile after Li+ removal. Its Li/Mo 22627

DOI: 10.1021/acsami.8b05577 ACS Appl. Mater. Interfaces 2018, 10, 22623−22631

Research Article

ACS Applied Materials & Interfaces

Figure 7. LIL short pulse measurement results under drain voltage of (a) 50 mV and (b) 1 V. 2 V pulses through LIL side gate were applied on MoO3 devices sequentially with increasing pulse widths. With Vsd = 50 mV drain voltage drain current increases instantly with the pulse and quickly decays; with an increased Vsd at 1 V, postpulse drain current saturates at a higher compared to that measured before the pulse, leading to a longterm change similar to a biological synaptic system. Inset in (a) zooms in a short time region near pulses, Id after different pulses follows the same exponential decay. (c) plots the Id incensement as a function of pulse width with Vd at 1 V, where ΔId = Id|t=20 s − Id|t=−20 s.

orders of magnitude; after the pulse, Id decays quickly and saturates to near the initial current. We attribute this shortterm property change to either an electrostatic effect in the MoO3 or a thin accumulated Li+ layer that forms under the gate bias; in this regime the Id enhancement is only affected by the magnitude of gate voltage applied, regardless of the pulse width. The decay curves after the gate pulses follow the same exponential trend, as shown in the inset of Figure 7a, which is attributed to ion redistribution after the gate voltage turned off. At Vsd = 1 V, the Id also decays quickly, but apparently longterm plasticity relative to the pristine material has been observed, as shown in Figure 7b. By sequentially applying pulses with increasing pulse width up to 5 s, the residual drain current was increased to ∼10 μA. This long-term current enhancement indicates a significant memory behavior of conductivity change, which is attributed to electrochemical doping where the Li+ migrates from liquid to MoO3 devices. The reason for the different long-term Id change measured at high and low Vsd is not clear, though Joule heating at higher Vsd could facilitate the Li+ intercalation. Our results have shown that increasing Vsd simply increases the current as expected by Ohm’s law as shown in Figure 7; for instance, the Id ratio at 1 V versus 50 mV Isd(1 V)/Isd(50 mV) is about 20. Another report shows that applying a Vsd pulse can induce long-term change in MoO3 solid state cells, though the magnitude of the change is small and the device is a vertical rather than lateral geometry.34 Further dynamic studies need to be conducted to fully understand these observations. Biologically, neurons tune their synapse weight depending on the stimulation in order to store information and neuromorphic computing concepts leverage “synaptic plasticity” to mimic short- and long-term memory processes.34,39 In our study we used gate pulses as the stimulation to modulate Id which acts similar to a synaptic memory process. Short- and long-term memory can be switched by tuning the drain voltage, and the intensity of long-term memory signal can be tuned by gate pulse width. With 1 V Vsd, the Id change after each pulse, marked as “excitatory postsynaptic current” in neuromorphic computing,35 is plotted as a function of pulse width in Figure 7c. These MoO3 devices with LIL can realize a long-term Id change from nA to μA level through short pulses,

We further carried out density functional theory (DFT) calculations to examine the underlying metallization mechanism. An Li+ ion was introduced as an interstitial and the Li/ Mo ratio was tuned by varying the size of α-MoO3 supercell. The 3 × 1 × 3, 2 × 1 × 2, and 1 × 1 × 1 supercells were employed to reach a Li/Mo ratio of 0.028, 0.063, and 0.25, respectively, with the second axis of the supercells being the van der Waals stacking dimension. The relative defect stabilities of different Li interstitial sites were evaluated, and two energy-favorable Li+ sites were identified, namely, an intralayer site and an interlayer site (as illustrated in Figure 6a). The former’s defect formation energy (DFE) is 0.05 eV lower than the latter’s (using the PBE+DFT-D2 method combined with Ueff = 0 eV settings; more details are given in the Computational Section). Therefore, we considered that Li+ can occupy both sites under the experimental conditions. These findings are consistent with the previous study by Ding et al.36 However, it should be noted that DFE calculations may depend on the exchange-correlation functionals and +U parameters.36 Figure 6b−d shows the band structures of intralayer defects with different Li+ concentrations. It indicates that an increase in Li+ concentration gradually reduces the bandgap and eventually leads to an insulator−metal transition, which is consistent with our experimental observation of the metallization process. Interestingly, interlayer Li+ does not make the system conducting with the Li/Mo ratio as high as 0.25 (band structures demonstrated in Figure S6). When Li/ Mo reaches 2:1, the Li intercalation was previously reported to result in a change of crystal structure to space group R3̅m,37 which was revealed to be metallic in the reference.38 In addition, we also find these band structures calculations are sensitive to the charge states of lithium interstitials. As shown above, the electrical properties of MoO3 devices can be significantly changed with LIL gate pulses due to the rapid migration of Li+. These devices thus have a potential to be developed into artificial synaptic systems, mimicking the biological memory, which inspired us to conduct a series of short-pulse tests. Two V pulses through an LIL side gate were applied on MoO3 devices sequentially with increasing pulse widths. Figure 7a shows the Id change as a function of time during the gate pulsing under 50 mV Vsd. Id increases by over 2 22628

DOI: 10.1021/acsami.8b05577 ACS Appl. Mater. Interfaces 2018, 10, 22623−22631

Research Article

ACS Applied Materials & Interfaces

been done in an ultrahigh-vacuum chamber with pressure ∼2 × 10−9 bar. Liquid metal Bi+ ion gun (30 keV energy, 500 pA current) has been used as a primary source with a focused ion beam spot size of ∼120 nm and a mass resolution of Δm/m ∼ 200/500. Cs+ ion source operated at 1 keV energy, and 60 nA current was used for the depth profiling. The images were collected with a resolution of 256 × 256 points across a 65 μm square imaging areas. Cs2O+ clusters have been used to identify spatial resolution of negative oxygen ions. The thickness of the flakes is estimated with the ratio to known thickness of deposited metals through ToF-SIMS imaging.

which is remarkable compared to previous studies on artificial synaptic devices.34,35 Each of these devices can work as a basic unit and is able to be developed into a network to store in formation in a nonvolatile manner. It makes ionic liquid gated solid state materials and the MoO3-LIL system specifically is a promising candidate for future neuromorphic computing elements. It is worth mentioning that we found MoO3 flakes fractures after several charge−discharge cycles; however, only small linear changes in conductivity are necessary for neuromorphic applications, and thus further studies are needed to investigate the stability of these ionic liquid gated devices under short pulse modulation.



COMPUTATIONAL SECTION The Kohn−Sham scheme 42,43 DFT calculations were performed using Vienna Ab initio Simulation Package (VASP)44,45 with projected augmented wave (PAW) pseudopotentials.46 The Perdew−Burke−Ernzerhof (PBE) version47of the exchange-correlation functional was employed together with the DFT-D2 method developed by Grimme et al.48 to better capture the van der Waals interactions. The structures were relaxed until the force for any give atom was smaller than 0.01 eV/ Å; the optimized lattice constants of MoO3 are a = 3.9347 Å, b = 13.9934 Å, and c = 3.7112 Å (a, b, and c are in the convention of space group Pbnm), which are close to experimental measurements (see Figure S5). The energy cutoff for plane-wave basis was set to 520 eV based on convergence tests. A 7 × 5 × 7 k-point mesh of Gammacentered Monkhorst−Pack grid was employed for the structural optimizations in the conventional unit cell of αMoO3; a 4 × 3 × 4 mesh of the same scheme was used for the 2 × 1 × 2 supercell and 3 × 3 × 3 for the 3 × 1 × 3 supercell. In order to account for the strongly correlated d-electrons of Mo, the Dudarev approach49 of DFT+U approximation was used in the band-structure and density-of-state (DOS) calculations with Ueff = 6 eV for Mo.36 In fact, our results are robust against Ueff value as it did not alter the metallic/ insulating state of the systems containing lithium interstitials (see Supporting Information, e.g. Figure S7, for band structures without +U). A homogeneous background charge is assumed when dealing with the charged defects.



CONCLUSION In summary MoO3 devices were exfoliated and successfully metalized by applying bias voltage through two different ILs where the Id enhanced by 9 orders of magnitude. Metallization can be realized by sweeping gate voltages in a number of cycles or by applying a fixed bias over time. This process is reversible as the devices can be tuned back to an insulator by applying negative bias. We attribute this controllable electrical property change to the migration of oxygen and lithium ions for the groups using normal IL gating and lithium IL gating, respectively. The migration of lithium is much faster than oxygen, and thus the current modulation occurs much faster. The ToF-SIMS result confirms the control of the ion migration in gating process. DFT calculations have also been carried out, and the Li metallization results are consistent with an intralayer Li+ interstitial. Results of short-pulse tests also illustrate the synaptic plasticity for neuromorphic computing elements.



EXPERIMENTAL SECTION

MoO3 single crystals were grown using chemical vapor transport with oxygen gas flow. An alumina boat containing 1 g of 99.9995% MoO3 powder, purchased from Alfa Aesar, was placed in the middle of a 19 mm × 4 ft open-ended fused quartz tube, which was then centered in a tube furnace. Oxygen gas was flowed into the tube through a rubber stopper by 3/16 in. tubing and then passed out through another rubber stopper at 2−3 bubbles/min into water. The tube furnace was set to 700 °C at 2 °C/min heating rate and allowed to sit for 2−3 weeks, until large crystals formed on the quartz tube near the opening of the furnace toward the oxygen gas outlet. Pale yellow transparent crystals of flat rectangular shapes were obtained up to 1 in. × 0.5 in. size along with many smaller needle-like crystals. X-ray diffraction was conducted on ground crystals, and the results are shown in Figure S5, in good agreement with the literature.40,41 Few-layer MoO3 flakes were exfoliated by the “Scotch tape” micromechanical cleavage technique onto Si substrate which has a 290 nm SiO2 dielectric layer on top. Electron beam lithography (FEI 600 Nova with Raith software) followed by electron beam evaporation (at the rate of 1 Å s−1) was then employed to fabricate the source/drain and side gate electrodes for electrical characterizations (Figure 1b inset). The contacts consisted of Ti/Au (5/25 nm) metals deposited and patterned through a standard lift-off process. The electrical measurement for pristine MoO3 devices was conducted in a cryogenic probe system with a semiconductor parametric analyzer (Agilent Tech B1500A). An optical micrograph of a MoO3 side gate device is shown in Figure 2a inset. A small drop of IL was subsequently applied to the device using a micropipet, covering the device and side gate areas. Samples were measured in the vacuum of ∼1 × 10−6 Torr. Lithiumcontaining IL gel was prepared by mixing 0.3 g of LiClO4 and 1 g of PEO into 15 mL of methanol and stirring overnight. The scanning speed of gate voltage during the measurement was fixed as 1 V/s. Time of flight−secondary ion mass spectrometry (ToF-SIMS) measurements were performed in positive ion detection mode using TOF SIMS-5 (ION-TOF GmbH) instrument. Investigations have



ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acsami.8b05577.



Supplementary figures and the corresponding descriptions, including microscopic images, Raman spectra results, XRD results, and additional ToF-SIMS and DFT analysis (PDF)

AUTHOR INFORMATION

Corresponding Author

*E-mail [email protected] (P.D.R.). ORCID

Cheng Zhang: 0000-0001-6531-4703 Akinola D. Oyedele: 0000-0001-5596-5238 Anton V. Ievlev: 0000-0003-3645-0508 Kai Xiao: 0000-0002-0402-8276 Olga S. Ovchinnikova: 0000-0001-8935-2309 Notes

The authors declare no competing financial interest. 22629

DOI: 10.1021/acsami.8b05577 ACS Appl. Mater. Interfaces 2018, 10, 22623−22631

Research Article

ACS Applied Materials & Interfaces



Volume Metallization of Tungsten Trioxide Films. Adv. Mater. 2016, 28 (26), 5284−5292. (14) Lu, N.; Zhang, P.; Zhang, Q.; Qiao, R.; He, Q.; Li, H.-B.; Wang, Y.; Guo, J.; Zhang, D.; Duan, Z.; Li, Z.; Wang, M.; Yang, S.; Yan, M.; Arenholz, E.; Zhou, S.; Yang, W.; Gu, L.; Nan, C.-W.; Wu, J.; Tokura, Y.; Yu, P. Electric-Field Control of Tri-State Phase Transformation with a Selective Dual-Ion Switch. Nature 2017, 546 (7656), 124−128. (15) Wong, A. T.; Noh, J. H.; Pudasaini, P. R.; Wolf, B.; Balke, N.; Herklotz, A.; Sharma, Y.; Haglund, A. V.; Dai, S.; Mandrus, D.; Rack, P. D.; Ward, T. Z. Impact of Gate Geometry on Ionic Liquid Gated Ionotronic Systems. APL Mater. 2017, 5 (4), 042501. (16) Pudasaini, P. R.; Noh, J. H.; Wong, A. T.; Ovchinnikova, O. S.; Haglund, A. V.; Dai, S.; Ward, T. Z.; Mandrus, D.; Rack, P. D. Ionic Liquid Activation of Amorphous Metal-Oxide Semiconductors for Flexible Transparent Electronic Devices. Adv. Funct. Mater. 2016, 26 (17), 2820−2825. (17) Leng, X.; Pereiro, J.; Strle, J.; Dubuis, G.; Bollinger, A. T.; Gozar, A.; Wu, J.; Litombe, N.; Panagopoulos, C.; Pavuna, D.; Božović, I. Insulator to Metal Transition in WO3 Induced by Electrolyte Gating. NPJ. Quantum Mater. 2017, 2 (1), 35. (18) Black, J. M.; Come, J.; Bi, S.; Zhu, M.; Zhao, W.; Wong, A. T.; Noh, J. H.; Pudasaini, P. R.; Zhang, P.; Okatan, M. B.; Dai, S.; Kalinin, S. V.; Rack, P. D.; Ward, T. Z.; Feng, G.; Balke, N. Role of Electrical Double Layer Structure in Ionic Liquid Gated Devices. ACS Appl. Mater. Interfaces 2017, 9 (46), 40949−40958. (19) White, R. T.; Thibau, E. S.; Lu, Z.-H. Interface Structure of MoO3 on Organic Semiconductors. Sci. Rep. 2016, 6, 21109. (20) Kanno, H.; Holmes, R. J.; Sun, Y.; Kena-Cohen, S.; Forrest, S. R. White Stacked Electrophosphorescent Organic Light-Emitting Devices Employing MoO3 as a Charge-Generation Layer. Adv. Mater. 2006, 18 (3), 339−342. (21) Janssen, A.; Riedl, T.; Hamwi, S.; Johannes, H.-H.; Kowalsky, W. Highly Efficient Organic Tandem Solar Cells Using an Improved Connecting Architecture. Appl. Phys. Lett. 2007, 91 (7), 073519. (22) Chang, C.-C.; Chen, J.-F.; Hwang, S.-W.; Chen, C. H. Highly Efficient White Organic Electroluminescent Devices Based on Tandem Architecture. Appl. Phys. Lett. 2005, 87 (25), 253501. (23) Rahmani, M. B.; Keshmiri, S.-H.; Yu, J.; Sadek, A.; Al-Mashat, L.; Moafi, A.; Latham, K.; Li, Y.; Wlodarski, W.; Kalantar-Zadeh, K. Gas Sensing Properties of Thermally Evaporated Lamellar Moo 3. Sens. Actuators, B 2010, 145 (1), 13−19. (24) Lupan, O.; Cretu, V.; Deng, M.; Gedamu, D.; Paulowicz, I.; Kaps, S. r.; Mishra, Y. K.; Polonskyi, O.; Zamponi, C.; Kienle, L.; et al. Versatile Growth of Freestanding Orthorhombic A-Molybdenum Trioxide Nano-and Microstructures by Rapid Thermal Processing for Gas Nanosensors. J. Phys. Chem. C 2014, 118 (27), 15068−15078. (25) Balendhran, S.; Walia, S.; Alsaif, M.; Nguyen, E. P.; Ou, J. Z.; Zhuiykov, S.; Sriram, S.; Bhaskaran, M.; Kalantar-zadeh, K. Field Effect Biosensing Platform Based on 2d A-MoO3. ACS Nano 2013, 7 (11), 9753−9760. (26) Xie, W.; Su, M.; Zheng, Z.; Wang, Y.; Gong, L.; Xie, F.; Zhang, W.; Luo, Z.; Luo, J.; Liu, P.; et al. Nanoscale Insights into the Hydrogenation Process of Layered A-MoO3. ACS Nano 2016, 10 (1), 1662−1670. (27) Meduri, P.; Clark, E.; Kim, J. H.; Dayalan, E.; Sumanasekera, G. U.; Sunkara, M. K. MoO3‑x Nanowire Arrays as Stable and HighCapacity Anodes for Lithium Ion Batteries. Nano Lett. 2012, 12 (4), 1784−1788. (28) Hanlon, D.; Backes, C.; Higgins, T. M.; Hughes, M.; O’Neill, A.; King, P.; McEvoy, N.; Duesberg, G. S.; Mendoza Sanchez, B.; Pettersson, H.; et al. Production of Molybdenum Trioxide Nanosheets by Liquid Exfoliation and Their Application in HighPerformance Supercapacitors. Chem. Mater. 2014, 26 (4), 1751− 1763. (29) Xiang, D.; Han, C.; Zhang, J.; Chen, W. Gap States Assisted MoO3 Nanobelt Photodetector with Wide Spectrum Response. Sci. Rep. 2015, 4, 4891.

ACKNOWLEDGMENTS P.D.R., C.Z., P.R.P., and M.G.S acknowledge support by U.S. Department of Energy (DOE) under Grant DE-SC0002136. D.G.M. and A.V.H. acknowledge funding by the Gordon and Betty Moore Foundation’s EPiQS Initiative through Grant GBMF4416. T.Z.W. acknowledges support by U.S. DOE, Office of Basic Energy Sciences (BES), Materials Sciences and Engineering Division. A.T.W. acknowledges support by Laboratory Directed Research and Development Program of Oak Ridge National Laboratory, managed by UT-Battelle, LLC, for the U.S. Department of Energy. A portion of this research was conducted by A.V.L. and O.S.O. at the Center for Nanophase Materials Sciences, which is a DOE Office of Science User Facility, and using instrumentation within ORNL’s Materials Characterization Core provided by UTBattelle, LLC under Contract DE-AC05-00OR22725 with the U.S. DOE. H.X. and L.X. acknowledge the support of Organized Research Unit Program (ORU-IMHM-18) at the University of Tennessee.



REFERENCES

(1) Nakano, M.; Shibuya, K.; Okuyama, D.; Hatano, T.; Ono, S.; Kawasaki, M.; Iwasa, Y.; Tokura, Y. Collective Bulk Carrier Delocalization Driven by Electrostatic Surface Charge Accumulation. Nature 2012, 487 (7408), 459−462. (2) Jeong, J.; Aetukuri, N.; Graf, T.; Schladt, T. D.; Samant, M. G.; Parkin, S. S. P. Suppression of Metal-Insulator Transition in VO2 by Electric Field-Induced Oxygen Vacancy Formation Field−Induced Oxygen Vacancy Formation. Science 2013, 339 (6126), 1402−1405. (3) Ueno, K.; Nakamura, S.; Shimotani, H.; Ohtomo, A.; Kimura, N.; Nojima, T.; Aoki, H.; Iwasa, Y.; Kawasaki, M. Electric-FieldInduced Superconductivity in an Insulator. Nat. Mater. 2008, 7 (11), 855−858. (4) Kötz, R.; Carlen, M. Principles and Applications of Electrochemical Capacitors. Electrochim. Acta 2000, 45 (15), 2483−2498. (5) Misra, R.; McCarthy, M.; Hebard, A. F. Electric Field Gating with Ionic Liquids. Appl. Phys. Lett. 2007, 90 (5), 052905. (6) Ye, J. T.; Zhang, Y. J.; Akashi, R.; Bahramy, M. S.; Arita, R.; Iwasa, Y. Superconducting Dome in a Gate-Tuned Band Insulator. Science 2012, 338 (6111), 1193−1196. (7) Yamada, Y.; Ueno, K.; Fukumura, T.; Yuan, H. T.; Shimotani, H.; Iwasa, Y.; Gu, L.; Tsukimoto, S.; Ikuhara, Y.; Kawasaki, M. Electrically Induced Ferromagnetism at Room Temperature in Cobalt-Doped Titanium Dioxide. Science 2011, 332 (6033), 1065− 1067. (8) Ueno, K.; Nakamura, S.; Shimotani, H.; Yuan, H. T.; Kimura, N.; Nojima, T.; Aoki, H.; Iwasa, Y.; Kawasaki, M. KawasakiM. Discovery of Superconductivity in KTaO3 by Electrostatic Carrier Doping. Nat. Nanotechnol. 2011, 6 (7), 408−412. (9) Ye, J. T.; Inoue, S.; Kobayashi, K.; Kasahara, Y.; Yuan, H. T.; Shimotani, H.; Iwasa, Y. Liquid-Gated Interface Superconductivity on an Atomically Flat Film. Nat. Mater. 2010, 9 (2), 125−128. (10) Schladt, T. D.; Graf, T.; Aetukuri, N. B.; Li, M.; Fantini, A.; Jiang, X.; Samant, M. G.; Parkin, S. S. P. Crystal-Facet-Dependent Metallization in Electrolyte-Gated Rutile TiO2 Single Crystals. ACS Nano 2013, 7 (9), 8074−8081. (11) Li, M.; Han, W.; Jiang, X.; Jeong, J.; Samant, M. G.; Parkin, S. S. P. Suppression of Ionic Liquid Gate-Induced Metallization of SrTiO3 (001) by Oxygen. Nano Lett. 2013, 13 (10), 4675−4678. (12) Jeong, J.; Aetukuri, N. B.; Passarello, D.; Conradson, S. D.; Samant, M. G.; Parkin, S. S. P. Giant Reversible, Facet-Dependent, Structural Changes in a Correlated-Electron Insulator Induced by Ionic Liquid Gating. Proc. Natl. Acad. Sci. U. S. A. 2015, 112 (4), 1013−1018. (13) Altendorf, S. G.; Jeong, J.; Passarello, D.; Aetukuri, N. B.; Samant, M. G.; Parkin, S. S. Facet-Independent Electric-Field-Induced 22630

DOI: 10.1021/acsami.8b05577 ACS Appl. Mater. Interfaces 2018, 10, 22623−22631

Research Article

ACS Applied Materials & Interfaces (30) Tan, Z.-H.; Yin, X.-B.; Guo, X. One-Dimensional Memristive Device Based on MoO3 Nanobelt. Appl. Phys. Lett. 2015, 106 (2), 023503. (31) Lee, S.-H.; Kim, Y.-H.; Deshpande, R.; Parilla, P. A.; Whitney, E.; Gillaspie, D. T.; Jones, K. M.; Mahan, A. H.; Zhang, S.; Dillon, A. C. Reversible Lithium-Ion Insertion in Molybdenum Oxide Nanoparticles. Adv. Mater. 2008, 20 (19), 3627−3632. (32) Greiner, M. T.; Helander, M. G.; Tang, W.-M.; Wang, Z.-B.; Qiu, J.; Lu, Z.-H. Universal Energy-Level Alignment of Molecules on Metal Oxides. Nat. Mater. 2012, 11 (1), 76−81. (33) Guo, Y.; Robertson, J. Origin of the High Work Function and High Conductivity of MoO3. Appl. Phys. Lett. 2014, 105 (22), 222110. (34) Yang, C.-S.; Shang, D.-S.; Chai, Y.-S.; Yan, L.-Q.; Shen, B.-G.; Sun, Y. Electrochemical-Reaction-Induced Synaptic Plasticity in Moox-Based Solid State Electrochemical Cells. Phys. Chem. Chem. Phys. 2017, 19 (6), 4190−4198. (35) Yang, C. S.; Shang, D. S.; Liu, N.; Shi, G.; Shen, X.; Yu, R. C.; Li, Y. Q.; Sun, Y. A Synaptic Transistor Based on Quasi-2d Molybdenum Oxide. Adv. Mater. 2017, 29 (27), 1700906. (36) Ding, H.; Lin, H.; Sadigh, B.; Zhou, F.; Ozolins, V.; Asta, M. Computational Investigation of Electron Small Polarons in AlphaMoO3. J. Phys. Chem. C 2014, 118 (29), 15565−15572. (37) Takahashi, Y.; Kijima, N.; Hayakawa, H.; Awaka, J.; Akimoto, J. Single-Crystal Synthesis and Structure Refinement of Li2MoO3. J. Phys. Chem. Solids 2008, 69 (5−6), 1518−1520. (38) Baldoni, M.; Craco, L.; Seifert, G.; Leoni, S. A Two-Electron Mechanism of Lithium Insertion into Layered Alpha-MoO3: A Dft and Dft+U Study. J. Mater. Chem. A 2013, 1 (5), 1778−1784. (39) Seok Jeong, D.; Kim, I.; Ziegler, M.; Kohlstedt, H. Towards Artificial Neurons and Synapses: A Materials Point of View. RSC Adv. 2013, 3 (10), 3169−3183. (40) Toby, B. H.; Von Dreele, R. B. Gsas-Ii: The Genesis of a Modern Open-Source All Purpose Crystallography Software Package. J. Appl. Crystallogr. 2013, 46 (2), 544−549. (41) Negishi, H.; Negishi, S.; Kuroiwa, Y.; Sato, N.; Aoyagi, S. Publisher’s Note: Anisotropic Thermal Expansion of Layered MoO3 Crystals [Phys. Rev. B 69, 064111 (2004)]. Phys. Rev. B: Condens. Matter Mater. Phys. 2004, 69 (10), 109902. (42) Hohenberg, P.; Kohn, W. Inhomogeneous Electron Gas. Phys. Rev. 1964, 136 (3B), B864. (43) Kohn, W.; Sham, L. J. Self-Consistent Equations Including Exchange and Correlation Effects. Phys. Rev. 1965, 140 (4A), A1133. (44) Kresse, G.; Furthmüller, J. Efficiency of Ab-Initio Total Energy Calculations for Metals and Semiconductors Using a Plane-Wave Basis Set. Comput. Mater. Sci. 1996, 6 (1), 15−50. (45) Kresse, G.; Hafner, J. Ab Initio Molecular Dynamics for Liquid Metals. Phys. Rev. B: Condens. Matter Mater. Phys. 1993, 47 (1), 558. (46) Blöchl, P. E. Projector Augmented-Wave Method. Phys. Rev. B: Condens. Matter Mater. Phys. 1994, 50 (24), 17953. (47) Perdew, J. P.; Burke, K.; Ernzerhof, M. Generalized Gradient Approximation Made Simple. Phys. Rev. Lett. 1996, 77 (18), 3865. (48) Grimme, S. Semiempirical Gga-Type Density Functional Constructed with a Long-Range Dispersion Correction. J. Comput. Chem. 2006, 27 (15), 1787−1799. (49) Dudarev, S.; Botton, G.; Savrasov, S.; Humphreys, C.; Sutton, A. Electron-Energy-Loss Spectra and the Structural Stability of Nickel Oxide: An Lsda+ U Study. Phys. Rev. B: Condens. Matter Mater. Phys. 1998, 57 (3), 1505.

22631

DOI: 10.1021/acsami.8b05577 ACS Appl. Mater. Interfaces 2018, 10, 22623−22631