Purely electronic mechanism of electrolyte gating of indium tin oxide thin films

Epitaxial indium tin oxide films have been grown on both LaAlO3 and yttria-stabilized zirconia substrates using RF magnetron sputtering. Electrolyte gating causes a large change in the film resistance that occurs immediately after the gate voltage is applied, and shows no hysteresis during the charging/discharging processes. When two devices are patterned next to one another and the first one gated through an electrolyte, the second one shows no changes in conductance, in contrast to what happens in materials (like tungsten oxide) susceptible to ionic electromigration and intercalation. These findings indicate that electrolyte gating in indium tin oxide triggers a pure electronic process (electron depletion or accumulation, depending on the polarity of the gate voltage), with no electrochemical reactions involved. Electron accumulation occurs in a very thin layer near the film surface, which becomes highly conductive. These results contribute to our understanding of the electrolyte gating mechanism in complex oxides and may be relevant for applications of electric double layer transistor devices.

doping that can extend throughout the entire film thickness. Such devices are referred to as the electrochemical transistors (ECTs) and explored for various electronics and sensor applications 1 . In the recent EGT-based research on complex oxides, originally it was supposed that the charging process is purely electronic, i.e., accumulation or depletion of electrons occurs near the surface of the gated material. The problem with this scenario is that it is at variance with the observation that, in most cases, the entire gated film turned metallic upon electrolyte gating, even if it was more than 1,000 Å thick [9][10][11][12] . However, because in metals the density of mobile charge carriers is very high (in the 10 22 to 10 23 cm −3 range), the screening length is typically just a few Angstroms. In this case, one indeed expects that the field-induced effects would be restricted to a very thin (< 1 nm thick) surface layer.
For this reason, uniform charging of thick films was rather attributed to some electrochemical reaction, such as massive electromigration and intercalation of oxygen ions or vacancies, as indeed substantiated for VO 2 and SrRuO 3 by incisive experimentation 10,12 . An implicit assumption has been that predominantly electrochemical mechanisms may be widespread and perhaps ubiquitous in complex oxides. It is thus of interest to test this hypothesis and search for a counter-example: a complex oxide resilient to electrochemical reactions, and susceptible to purely electronic charging when gated using an ionic liquid or a polymer electrolyte.
To differentiate between the two competing scenarios, note that electrochemical charging mechanism has several telltale characteristics that can be used as litmus tests. First, chemical intercalation is generally a slow process; under the typical experimental conditions it takes many minutes or even hours to reach the saturation charge density. Second, the charging process tends to be hysteretic; to discharge the sample one must apply the opposite voltage (and wait). In some cases, the intercalation process is irreversible, and can even be destructive, causing film delamination or complete chemical decomposition. Yet another litmus test, that we have devised, is to fabricate lithographically several devices ('pixels") for 4-point contact transport measurements next to one another. Then we apply electrolyte gating to one of them, say Pixel 1, and measure simultaneously the transport in this pixel, as well as in the first-neighbor Pixel 2, the next-nearest neighbor Pixel 3, etc. Note that the Pixels 2 and 3 are not covered by the electrolyte and hence not exposed to the large electric field. The two clear-cut limiting behaviors that this measurement can show are either (i) the transport properties of Pixel 1 show large changes that scale with the gate-voltage V G , while Pixels 2 and 3 remain unaffected, or else (ii) Pixel 2 also shows concomitant (although probably smaller) changes that scale with those in Pixel 1 and with V G . In the first case, we infer that the doping is predominantly electronic. In the latter case, there must be a substantial ionic diffusion and electro-migration, which is not limited to just the gated Pixel 1 but spreads into the neighbor pixel(s) because of the (small) lateral probe current.
In this paper, we report to have identified a complex oxide, indium tin oxide (ITO), in which electrolyte gating causes very fast charging, on the scale of a couple seconds, shows no hysteresis, and is completely reversible, suggestive of a predominantly electronic mechanism. Moreover, the changes in resistivity are only observed in the device that is covered by the gated electrolyte, with no signs of ionic diffusion into the neighbor devices.
ITO has been extensively studied due to its optical transparency in the visible range coexistent with high conductivity (up to 10 4 S/cm at room temperature) [13][14][15][16][17][18] . The carrier concentration of ITO is in the range of 10 20 cm −3 -10 21 cm −3 , low enough to be amenable to substantial changes upon electrolyte gating 15 . For the field-effect experiments, electrolyte or otherwise, one needs high quality thin films with atomically smooth surfaces. Among various methods that have been used to prepare ITO thin films, magnetron sputtering stands out as having been proven capable of producing high quality films 18 , so this is the technique that we have adopted for the present study. So far, only a few studies focused on ultrathin ITO films 16,17 . The main conclusion was that the structure and the physical properties of ITO thin films are strongly dependent on the deposition conditions, so we have explored various growth protocols and substrates.

Results
Synthesis of epitaxial ITO thin films. In the present study, ITO thin films were synthesized on both single-crystal LaAlO 3 (LAO) and yttria-stabilized zirconia (YSZ) substrates using RF magnetron sputtering (see methods). Prior to electrolyte gating experiments, every film was characterized by atomic force microscopy (AFM), X-ray diffraction (XRD), and transport measurements.
AFM and XRD data show that ITO films grown on either LAO or YSZ can have epitaxial alignment with respect to the substrate. Figure 1a shows XRD patterns for ITO films deposited at various growth temperatures on LAO substrates polished with the surface perpendicular to the crystallographic [001] direction. All these films were grown at the same radio-frequency (RF) power (P = 60 W) for the same time (t = 30 minutes), and in each case the measured thickness was about d = 30 nm. In the range 25° < 2θ < 40°, the XRD patterns of the films grown at T s = 600 °C and T s = 650 °C show a single peak (labeled as ITO 004 in Fig. 1), suggesting that the films are epitaxially alligned to the substrate. For films grown outside of this temperature window, i.e., at T s = 550 °C or T s = 700 °C, the XRD patterns show extra peaks. As the substrate temperature drops to T s = 500 °C, only the ITO 222 peak appears in the XRD pattern, indicating that this film is also epitaxial but its orientation is different.
In Fig. 1b we present a typical AFM image of an ITO film grown on LAO substrate at T s = 600 °C. One can clearly see steps and terraces and the average root mean square (rms) roughness of the surface is only 2.2 Å, indicating that the film surface is indeed atomically flat. However, the LAO substrate is generally heavily twinned, and this may affect the transport properties of ultrathin films grown on it 19 . Moreover, the lattice mismatch between LAO and ITO is substantial. In LAO a 0 = 3.821 Å, while the lattice constant in ITO calculated from the XRD pattern is a 0 = 10.156 Å; normalizing both to a reduced pseudo-cubic unit cell the calculated strain is + 6.4% (tensile).
Thus, we speculated that YSZ may be a better choice of the substrate, because it does not twin, and since it has a 0 = 5.12 Å it is much better lattice-matched to ITO, which would be under + 1% tensile strain. Our experimental results confirmed this expectation. In Fig. 2a we show XRD patterns of ITO films grown on YSZ substrates polished with the surface perpendicular to the crystallographic [001] direction. These films were grown at T s = 600 °C, with the remaining parameters the same as above, except that we also varied the film thickness. We can clearly see the so-called finite thickness (or Laüe) fringes, a clear sign that in a single-crystal film the surface and the substrate-film interface are perfect and parallel to each other on the atomic scale. In Fig. 2b we present the AFM image of a 10 nm thick ITO film on YSZ substrate. The rms surface roughness is only 1.5 Å, indicating that the film surface is atomically flat -as flat as the substrate itself.
Electrolyte gating of ITO films. After the structural characterization, the ITO/YSZ films were patterned into a Hall bar configuration for transport measurements. In our gating experiments, we have used both an ionic liquid N,N-diethyl-N-methyl-N-(2-methoxyethyl)ammonium bis(trifluoromethylsulphonyl)imide (DEME-TFSI) and a polymer electrolyte (sodium fluoride salt dissolved in polyethylene glycol, PEG-NaF), and the results turned out to be quite similar. In Fig. 3 we show the dependence of the sheet resistance (R S ) on the temperature, for various values of the V G . As the gate voltage increases from V G = 0 to V G = 2.5 V, the room temperature resistance drops by more than 50%, and then saturates. The sample remains insulating down to the lowest temperature. However, at higher temperatures (T > 120 K), the temperature coefficient of resistance changes from negative (semiconductor-like) to positive (metallic-like), as the gate voltage increases.
In the previous electrolyte gating experiments with various complex oxides and correlated electron materials, the usual charging time (needed to reach saturation at a fixed V G ) was reported to be in the range of 10 to  30 minutes for each gate voltage step 2-12 . However, we have found that the charging time in ITO is extremely short. In Fig. 4a, we can see that for ITO the resistance drops immediately after we change the gate voltage, and then it saturates at each gate voltage. The charging time for each voltage step is about 1 minute, but most (> 80%) of the resistance drop occurs within the first second or so. For comparison, in Fig. 4b we also show the results of gating WO 3 films that were deposited in the same RF sputtering system 20 and using the same ionic liquid electrolyte, DEME-TFSI. One can see that in WO 3 (and other materials), the resistance keeps dropping even after half an hour after we change the gate voltage, and it never saturates at each voltage step. This striking difference indicates that the charging mechanisms in ITO and WO 3 are most likely different. Since electrochemical reactions usually take a long time to proceed while electrostatic charging occurs essentially immediately, we infer the electrolyte gating in ITO is probably a purely electronic effect while in WO 3 (and many other materials) electrochemical processes are involved.
In Fig. 5, we can see that the charging/discharging loop for ITO shows no hysteresis. Again, we contrast this with WO 3 , in which a large conductivity change persists after the gate voltage is removed. To restore the initial state in WO 3 (before the positive gate voltage was applied), one has to apply the gate voltage of opposite (negative) polarity, and this process is also slow. Thus, this is another piece of evidence pointing to an electronic mechanism in ITO, in contrast to an electrochemical process in WO 3 .
In Fig. 6 we show the result of a new type of experiment that we have devised to conclusively test whether electrolyte gating causes massive ion intercalation into or removal from the gated material. We pattern the film into a long bar with multiple contact leads, as shown in Fig. 6a. This enables simultaneous measurements of the  As an ITO film is charged using DEME-TFSI as an electrolyte, the resistance drops immediately after we change the gate voltage, and then it saturates at each voltage. The dashed lines indicate the time when the gate voltage is changed and each voltage step is 0.1-0.2 V. (b) For WO 3 , with the same gate voltage step and the same ionic liquid, the resistance keeps dropping even after half an hour after we change the gate voltage, and it never saturates at each voltage step.
Scientific RepoRts | 6:31239 | DOI: 10.1038/srep31239 longitudinal resistance and/or the Hall effect in a number of segments of the bar ("pixels"). We have used this type of parallel transport measurement extensively in our experiments on films with combinatorial spread in the chemical composition 21 . The innovation here is that we pick one pixel (call it Pixel 1), cover it with electrolyte and gate it under varying conditions (temperature, gate voltage) while measuring the electrical transport simultaneously in that pixel and in the neighboring pixels (Pixel 2, Pixel 3, etc.) that are not covered by the electrolyte. In this case, only Pixel 1 could experience the extremely high electric field (up to 10 8 V/cm) within the Helmholtz double layer. Given the small lateral probe current (typically 100 μ A), the lateral electric field is 6-8 orders of magnitude smaller. Moreover, the electron screening length is 5 orders of magnitude shorter than the pixel length, and therefore any electron depletion or accumulation in Pixel 2 would be confined to a minuscule fraction of its volume. Altogether, if electrolyte gating causes only electronic processes, there could be no observable changes of the longitudinal or Hall resistance in Pixel 2, Pixel 3, etc. Conversely, if any such changes are detected, they could only come from electromigration of ions. If the ionic conductivity of the gated material is non-negligible under the actual experimental conditions (temperature, gate voltage), the (small) lateral probe current could drive some ions into the Pixels 2, 3, etc., which could produce a measurable change in electrical resistance.
In Fig. 6b we show that the later effect is in fact quite dramatic in thin films of WO 3 . Upon gating Pixel 1, as its resistance drops with time, there is a concomitant drop of resistance of Pixel 2 and Pixel 3 (although smaller for Pixel 3 by about a factor of 10), and the two track one another in time. Pixel 4, the furthest from the electrolyte-gated Pixel 1, shows the least change. The charging process in all of the pixels is slow, which is yet another indication of the electrochemical nature of the doping process. Figure 6c shows a complete absence of such an effect in ITO; Pixel 2 shows no changes whatsoever. The charging process is much faster, in particular the initial precipitous drop in Pixel 1 that occurs on a scale of seconds or less. Figure 6d also shows charging of Pixel 1 in ITO as a function of gate voltage, and the absence of any observable charging of Pixel 2. Altogether, these are incisive experiments that clearly demonstrate a dramatic difference in the charging mechanisms in WO 3 and ITO; it is predominantly electrochemical (i.e., ionic) in WO 3 and predominantly electronic in ITO.

Discussion
If our inference is correct, and the observed modifications of the electrical resistance of ITO films originate from electron accumulation, then the affected layer must be very thin, just a few Angstroms thick. In this case, since the conductivity of the underlying layer remains the same, the change in the conductivity of the modified top layer must be much larger than the apparent change in the entire film. To make this more concrete, consider the data in Fig. 3 at some fixed temperature, e.g., T = 200 K. Upon electrolyte gating, the sheet resistance drops from R S = 2.7 kΩ/□ to R S = 1.15 kΩ/□ , i.e., by about a factor of 3. If we assume that only the top layer about 5 Å thick is modified, while the rest of the film (95 Å thick layer) is unaffected, this would imply that the conductivity of the top layer has increased by a factor of about 30, to more than 2 × 10 5 S/cm -comparable to those in good metals (say, nickel). While this is just a crude estimate, it ensures that the electron accumulation in the top layer must be massive, with the saturated electron density in the 10 23 cm −3 range.
A question left open is why even at our highest gate voltage the resistivity slope changes sign at low temperature to negative (ρ increases as T is decreased). The two most likely explanations are as follows. First, the surface charge density induced by electrolyte gating may not be perfectly uniform, and hence one could imagine that (relatively large) metallic domains are embedded in a semiconducting matrix; as the temperature is decreased, these semiconducting barriers become more resistive and eventually dominate the resistivity of that surface layer. The other possibility is that rather than in real space, the 'phase separation' occurs in the k-space, e.g. by formation of  3 , the charging/discharging time at each voltage step is over half-an-hour. A large hysteresis appears and one has to apply a negative gate voltage to fully discharge it. a charge-density wave that partially gaps the Fermi surface along some azimuthal sectors that grow as the temperature is lowered. To decide between these scenarios one would need additional experiments that are outside the scope of the present study. Nevertheless, the fact that we can induce a highly metallic surface state by electric field provides an encouraging perspective for potential applications. Greatly increased charging/discharging speed (compared to WO 3 and the like) is certainly another positive factor.
It is of interest to make a brief comparison with somewhat more broadly related experiments in the literature. One study similar in spirit to the present one was done on well-characterized, single-crystal SrTiO 3 samples with atomically smooth surfaces 22 . Systematic transport measurements were made on lithographically defined devices with the Hall bar geometry. The four-point contact resistance and the Hall coefficient were measured, providing accurate values of the sheet conductance σ s , the sheet carrier density n s , and the Hall mobility μ. EGT experiments were made using a polymer electrolyte (polyethylene oxide with KClO 4 ). (Dis)charging was done at T = 320 K, where the electrolyte was in a liquid state. The SrTiO 3 sample is initially an insulator with a band gap of 3.15 eV. Electrolyte gating can turn its surface metallic, and a two-dimensional superconducting state emerges 23 below the critical temperature T c = 0.4 K. The doping mechanism apparently depends on the voltage bias. In the low-bias regimes, V G < 3.7 V, the sample charging and discharging occurred within a second, and it is fully reversible. The maximum induced carrier density reaches n s ≈ 10 14 cm −2 . This changes dramatically for V G > 3.75 V. A leakage current, I G , shows a slow decay upon turning V G on, and a negative value upon V G removal. This high-bias process is irreversible; persistent conduction is observed even after removal of the gate bias. In this regime, the sheet carrier density can reach n s ≈ 10 15 cm −2 . However, the thickness of the conductive region was deduced to be around 10 μ m, which was supported by transmission electron microscopy (TEM). This was ascribed to some electrochemical process. The two candidate reactions are intercalation of K + cations into SrTiO 3 and oxygen removal. Secondary ion mass spectrometry (SIMS) showed no presence of K + in SrTiO 3 , so it was concluded that the conductive layer is formed by oxygen removal from a few micrometers deep near-surface region of SrTiO 3 . Reflecting now on our present study focused on ITO, the main difference is that SrTiO 3 is known to be prone to oxygen vacancy formation, and thus may be more analogous to VO 2 . This warrants some additional experimental study, but this is outside of the scope of the present paper.
In another interesting recent paper 24 , electrolyte gating of WO 3 was studied by electro-chemical techniques. Experiments were largely complementary to ours, in several respects. First, there is a big difference in the film morphology. The WO 3 films were grown by sol-gel technique at room temperature on conducting glass substrates and post-annealed in flowing oxygen at 400-550 °C for 30 min, and described as meso-porous 24 . From XRD patterns shown, they appear polycrystalline. While no morphological information has been provided in ref. 24, scanning electron microscope (SEM) images in an earlier report on similar films 25 show disconnected grains, with the size typically on ~10 nm scale. Two ionic liquids were used, 1-methyl-1-butylpyrrolidinium bis-(trifluoro-methylsulfonyl)imide or [PYR 14 ][TFSI] and 1-ethyl-3-methylimida-zolium bis(tri fluoromethylsulfonyl) imide or [EMIM] [TFSI]. They have different ion sizes; the radius of [PYR 14 ] is 3.9 Å, and for [EMIM] it is 3.5 Å. The key data shown are cyclic voltammetry, the dependence of source-drain current on V G , and X-ray diffraction. The main conclusion is that despite the large size of the ions, the doping mechanism in the films cannot be purely electrostatic since (a) the cyclic voltametry shows a clear Faradaic behavior, (b) the charging process is relatively slow (the tail in the source-drain current I sd extends for several minutes), and (c) the doping charge estimated from the specific capacitance of the ionic liquid and the surface area of the WO 3 exposed to the electrolyte is much smaller (only about 10%) than the measured charge. On the other hand, estimating the electrochemical doping charge needed to dope W from the oxidation state + 6 to + 5, the authors conclude that only 3% of the WO 3 volume is doped if [EMIM][TFSI] is used as the electrolyte, and even less, 1.2% in the case of [PYR 14 ][TFSI]. This is attributed to the large size of the cations of the ionic liquid that cannot be inserted in the relatively small cages of WO 3 . This led to speculation that a third mechanism may be at play as well for electrolytes made up of ions with relatively large size that cannot be inserted in the oxide lattice: nonconventional electrochemical doping, confined at the electrolyte/metal oxide interface. In summary, the authors proposed that doping of meso-porous WO 3 films includes all three contributions -an electrostatic process, a conventional electrochemical process (protons from water traces are inserted into the WO 3 ), and nonconventional interface-confined electrochemical doping, in which the injected electrons are compensated by the large cations of the ionic liquid packing at the interface.
Turning back to ITO, we note that there is a possibility of some electrochemical process, likely limited to the interface, and perhaps responsible for the tail in I sd that extends for about a minute. (Alternatively, this could come from changes in the ionic liquid itself.) But we stress that most of the change occurs very fast and should be of electronic nature; this is the dominant effect in ITO.
In terms of the material investigated, the previous work most directly related to this one is a study of EGTs made of ITO films 15 . It is also complementary to the present work, in several respects. (a) The first major difference is in the film quality, structure, and morphology. ITO films studied there were amorphous, porous, and granular. Note that the material of which the grain boundaries are composed has a different (and unknown) structure and chemical composition, while it is precisely this material that is in direct contact with the ionic liquid. (b) They used a different ionic liquid, 1-ethyl-3-methylimidazolium bis(trifluoromethylsulfonyl)imide (EMI-Beti). (c) They performed two-terminal ac resistance measurements, using a frequency response analyzer to measure the dependence of the complex impedance on V G . Since both the capacitance and the resistance of the ionic liquid change with the gate voltage, it is conceivable that in these ac measurements the changes in properties of the ionic liquid play a role. (d) The doping mechanism was not discussed. However, since for positive V G , which causes electron accumulation in the ITO channel, equilibrium was reached within ~100 ms, we take this a signature of purely electronic process. (e) In contrast, for V G < 0 (electron depletion) the time to reach equilibrium was orders-of-magnitude longer, ~100 s; this appears as an indication for some electrochemical process, but regrettably no further details were reported.
Comparing this with the present study, we conclude that there is a basic consistency, at least insofar as our basic conclusion about the doping mechanism being predominantly electronic in nature (for V G > 0) is concerned. Our samples are high-quality, epitaxial, single-crystal, stoichiometric ITO films, and moreover, before the measurements of the temperature dependence of the resistivity we clean the surface by fast cycling. Thus, our results should be more characteristic of the intrinsic behavior of ITO crystals. Next, we have performed four-point-contact dc resistance measurements as a function of temperature down to T = 4 K, checking for possible superconductivity (none was observed down to this temperature), which of course was the primary motivation for the present as well as for almost all other recent studies of EGTs in complex oxides. Next, we have compared the behavior of ITO to that of single-crystal WO 3 films under the same charging conditions, and showed that they have quite different intrinsic behavior. We have shown that there is substantial lateral ionic diffusion in WO 3 while none is observed in ITO by a new type of experiment, where we lithographically pattern the film into a series of devices for 4-point contact resistivity measurements, expose one to electrolyte gating and monitor its resistivity simultaneously with that in neighboring devices. In the case of WO 3 , we have found that neighboring devices, even though they are not covered with electrolyte, show changes in resistivity that track those in the gated device, while no such effect is observed in ITO. Thus the charging mechanism is apparently material-dependent and different in these two cases, predominantly electrochemical in WO 3 and predominantly electronic in ITO. Finally, we infer that in ITO the (electronic) charging must be restricted to a very thin layer near the film/electrolyte interface, implying that the local change in mobile carrier density is quite large, saturating at ~10 23 cm −3 .
Apart from the structure of the electrolyte itself, the crystal structure of the gated material plays a role and determines which doping mechanism will dominate upon electrolyte gating. In very open oxide structures with high propensity for oxygen vacancy formation, and high ionic mobility of oxygen, of which VO 2 may be the Scientific RepoRts | 6:31239 | DOI: 10.1038/srep31239 best-studied example, the dominant mechanism can be oxygen electro-migration. WO 3 is apparently less susceptible to this mechanism of ionic conduction because the vacant B site at the center of the perovskite unit cell is surrounded by six O 2− ions as nearest neighbors, so it readily accepts small cations, such as Li + . Indeed, H + (i.e., a proton) is the smallest cation and, if present in sufficient concentration, dominates the ionic conductivity of WO 3 (X . Liang et al., unpublished). Apparently, neither of these mechanisms is operative in ITO, by virtue of its 'more compact' structure.
In summary, we have grown high-quality epitaxial ITO films on both LAO and YSZ substrates by rf magnetron sputtering. Electrolyte gating triggers a large resistance drop under a small gate voltage. The process is very fast; a drop by over 80% occurs within a couple of seconds after the gate voltage is applied). No hysteresis is observed in the charging/discharging loop and the device next to the one that is electrolyte gated shows no changes in resistivity. Collectively, these three findings indicate that in ITO the primary effect of electrolyte doping is of electronic nature. This contrasts with previous studies of electrolyte gating of various complex oxides where the predominant effect is an electrochemical reaction.