Introduction

Water has not been utilized as an active electronic material though water has been widely applied in industry as a coolant (radiators), solvent (batteries) and pressure medium (hydroelectric power generation). We have previously aimed to exploit the electrolysis of water1. Although the electrical conductivity of pure water is extremely low (~0.055 μS cm−1 at 25°C)2, ionization of H+ and OH ions occurs when the bias voltage (>1.23 V) is applied between two metallic electrodes immersed in water, as shown in Figure 1a. The ions are then attracted to the cathode and anode. Finally, H2 and O2 gases are generated on the cathode and anode, respectively, via the electron transfer on the electrode surface. As H+ and OH ions, which are strong reducing/oxidizing agents for most oxide semiconductors3,4,5,6,7,8, are simultaneously produced in water electrolysis, one may expect that the electrical conductivity of an oxide semiconductor, which is strongly dependent on oxygen non-stoichiometry, can be modulated by utilizing the redox reaction between H+/OH ions and the oxide surface. However, water electrolysis and the redox reaction do not take place, because no electric field can be applied on the insulating oxide surface in the first place. Thus, the surface of the oxide must be conductive, as schematically shown in Figure 1b.

Figure 1: Field-induced water electrolysis switches an oxide semiconductor from an insulator to a metal.
figure 1

(a) Simple water electrolysis with two Pt electrodes as the cathode and anode immersed in water. H+ and OH ions, which are generated by the electrolysis, become H2 and O2 gases on the anode and cathode, respectively. (b) Water electrolysis with an insulating oxide MOx, with a slightly conductive surface MOx–δ. Similar to a, H+/OH ions are attracted to the MOx–−δ, leading to the redox reaction between H+/OH ions and the MOx–δ surface. GND, ground.

We have found that water-infiltrated nanoporous glass overcomes this problem and switches an oxide semiconductor from an insulator to a metal. Electron accumulation, water electrolysis and redox take place successively on an oxide surface at room temperature (RT), leading to the formation of a thin metal layer on the oxide. We have fabricated a field-effect transistor (FET) structure with source, drain and gate electrodes on an insulating oxide, using water-infiltrated nanoporous glass as the gate insulator. First, the insulating oxide surface becomes slightly conductive by applying the gate voltage because of electrostatic charge accumulation. Then, water electrolysis occurs between the gate and the oxide surface. Finally, a redox reaction takes place between H+/OH ions and the oxide surface. As a result, a thin metal layer is formed on an insulating oxide.

The key material to make the best use of the electron activity of water is nanoporous glass. We chose amorphous 12CaO·7Al2O3 (a-C12A7) with a nanoporous structure for this purpose. C12A7 is an abundant and environmentally benign material. Crystalline C12A7 becomes semiconducting9 or metallic10,11 when the clathrated free oxygen ions (O2−), which are incorporated into the cage structure (~0.4 nm in diameter), are removed by chemical reduction treatment. On the other hand, 'amorphous' C12A7 is a good electrical insulator, because amorphous C12A7 does not have a cage structure, we therefore ruled out the possibility of electrical conductivity of an amorphous C12A7 film. As C12A7 can be hydrated easily12, it is used commercially as a major constituent of aluminous cement. In 1987, Hosono and Abe13 found that a large amount of bubbles was generated in an a-C12A7 glass, when C12A7 melt was quenched under high oxygen pressure. In the present study, we develop a method to fabricate a-C12A7 film with nanoporous structure (CAN, hereafter) by pulsed laser deposition.

In this study, we show that water-infiltrated CAN electrically switches an oxide semiconductor from insulator to metal, using a CAN-gated FET structure on a SrTiO3 single crystal (Fig. 2a) as a proof of concept. Although SrTiO3 is a wide bandgap (~3.2 eV) insulator, it becomes n-type conducting SrTiO3−δ by appropriate reducing treatments5. Conducting SrTiO3, especially when it is a two-dimensional (2D) conductor14,15, is of great importance as an active material for future electronic devices16, because it has several potential advantages over conventional semiconductor-based electronic materials, such as transparency17, giant magnetoresistance18 and giant thermopower19. On applying positive gate voltage to the CAN-gated SrTiO3 FET, electron accumulation, water electrolysis and electrochemical reduction occur successively on the SrTiO3 surface at RT. This leads to the formation of a thin (~3 nm) metal layer with an extremely high electron concentration (1015–1016 cm−2), which exhibits exotic thermoelectric behaviour.

Figure 2: The CAN-gated SrTiO3 FET.
figure 2

(a) A schematic illustration of the CAN-gated SrTiO3 FET. (b) Cross-sectional TEM image of the CAN-gated SrTiO3 FET (left panel). Scale bar is 100 nm. Trilayer structure composed of Ti/CAN/SrTiO3 is observed. Large amount of light spots are seen in the whole CAN region. Broad halo is observed in the selected area electron diffraction pattern of CAN, and diffraction pattern from SrTiO3 single crystal is also shown below (right panel). (c) Z-contrast, high-angle, annular dark-field STEM image of the CAN/SrTiO3 interface. Scale bar is 20 nm. Nanopores with diameter <10 nm appear dark. (d) TDS spectrum of water (m/z=18 H2O) in the CAN film (blue). The amount of H2O up to 400°C was estimated to be 1.4×1022 cm−3 (=0.41 g cm−3, ~41%). TDS spectrum in the dense a-C12A7 film is shown for comparison (red, 0.009 g cm−3). The water concentration increases monotonically with an increase in the porosity of the CAN films (the inset).

Results

Water-infiltrated nanoporous glass (CAN)

The trilayer structure composed of Ti (20 nm)/CAN (200 nm)/SrTiO3 is clearly observed in the cross-sectional transmission electron microscopic (TEM) image of the CAN-gated SrTiO3 FET (Fig. 2b). Many brighter contrasts (diameter ~10 nm) are seen throughout the CAN region. Furthermore, several dark contrasts with a diameter <10 nm are observed in the Z-contrast, high-angle, annular dark-field scanning TEM (STEM) image of the CAN film (Fig. 2c). Judging from these TEM/STEM images, high-density nanopores with a diameter of <10 nm are incorporated into the CAN film.

We subsequently measured thermal desorption spectrum (TDS) of the CAN films to detect weakly bonded chemical species in the nanopores. Most of the desorbed species were H2O (m/z=18, where m/z indicates the molecular mass to charge ratio; Fig. 2d). The amount of H2O up to 400°C was estimated to be 1.4×1022 cm−3, which corresponds to 41%. The bulk density of the CAN film was ~2.1 g cm3, evaluated by grazing incidence X-ray reflectivity (Supplementary Fig. S2), which corresponds to 72% of fully dense a-C12A7 (2.92 g cm3)20. From these results, we judged that moisture in the air (humidity 40–50%) would infiltrate the CAN film most likely due to the capillary effect, hence, nanopores in the CAN film were filled with water.

Field-induced water electrolysis

We then measured the electron transport properties of the CAN-gated SrTiO3 FET at RT. Figure 3 summarizes (a) drain current (Id) versus gate voltage (Vg) curves, (b) gate current (Ig) versus Vg curves, and (c) capacitance (C) versus Vg curve of the CAN-gated SrTiO3 FET. Corresponding properties of the dense a-C12A7-gated SrTiO3 FET21 (d, e, f) were also measured for comparison. Both the channel length, L, and the channel width, W, were 400 μm. Dielectric permittivity (ϵr) of a-C12A7 was 1221. The gate voltage sweeps were performed in numerical order (for example: 1:−5 V→0 V→+5 V→0 V→–5 V). The Id versus Vg curves (a) show large anticlockwise hysteresis, indicating movement of mobile ions22, though very small clockwise hysteresis (~0.5 V) is seen in the dense a-C12A7-gated SrTiO3 FET (d). Although very small, Ig (~2 pA) is observed in the dense a-C12A7-gated SrTiO3 FET (e), whereas the Ig of the CAN-gated SrTiO3 FET (b) increases exponentially up to 20 nA with Vg, suggesting that mobile ions transport electronic charge. The C versus Vg curve (c) of the CAN-gated SrTiO3 FET shows large anticlockwise hysteresis loop. The maximum C is ~160 pF, ~76% of the value for the dense a-C12A7-gated SrTiO3 FET ((f) ~210 pF), consistent with the fact that the volume fraction of dense a-C12A7 part in the CAN film is ~72%.

Figure 3: Electron transport properties of the CAN-gated SrTiO3 FET at RT.
figure 3

(a) Id versus Vg, (b) Ig versus Vg and (c) C versus Vg, Id versus Vg (d), Ig versus Vg (e), and C versus Vg curves (f) of the dense a-C12A7-gated SrTiO3 FET are also shown for comparison. Both the channel length, L, and the channel width, W, are 400 μm. The gate voltage sweeps were performed in numerical order (for example: 1: −5 V→0 V→+5 V→0 V→–5 V). (a) Id versus Vg curves (Vd=+2 V) show large anticlockwise hysteresis, although very small clockwise hysteresis is seen in the dense one (d, Vd=+1 V). (b) Ig increases exponentially up to 20 nA with Vg, which is ~104 greater than that of the dense a-C12A7-gated SrTiO3 FET (e, red: observed, grey: smoothed). (c) The C versus Vg curve shows a large anticlockwise hysteresis loop. The maximum C (frequency: 20 Hz) of the nanoporous a-C12A7-gated SrTiO3 FET is ~160 pF, ~76% of that of the dense a-C12A7-gated SrTiO3 FET ((f) ~210 pF).

We also observed a clear pinch-off and saturation in Id at low Vg region (see Supplementary Fig. S3), indicating that the operation of this FET conformed to standard FET theory at low gate voltage. Thus, we concluded that first the insulating SrTiO3 surface became slightly conductive with gate voltage because of the electron charge that was accumulated at the SrTiO3 surface by pure electrostatic effect.

To clarify the role of water in the electrical transport properties, we measured the Id versus Vg characteristics of the CAN-gated SrTiO3 FET at 0°C using a Peltier cooler, because H+ and OH+− ions cannot move through ice (Supplementary Fig. S4). The device does not show such large anticlockwise hysteresis at 0°C. Although the maximum C increased from ~160 pF (25°C) to ~240 pF (0°C), Id at Vg=+10 V decreased from ~5 μA (25°C) to ~1 μA (0°C), most likely because of the fact that water in the CAN acts as a simple gate dielectric at 0°C (ϵr 25°C ~78, ϵr 0°C ~88)23. The field-effect mobility (μFE) value of the FET at 0°C is ~0.8 cm2 V−1 s−1, which is comparable to that of the dense a-C12A7-gated SrTiO3 FET (~2 cm2 V−1 s−1), obtained from μFE=gm((C×Vd)/(W×L))−1, where gm was the transconductance ∂Id/∂Vg. We also measured Hall mobility (μHall) of a CAN-gated SrTiO3 FET after several Vg applications (metallic state) and obtained μHall values of 2.3–2.5 cm2 V−1 s−1, which is approximately three times larger than the ìFE value (~0.8 cm2 V−1 s−1). These results clearly indicate that H+ and OH+− ions in the CAN are the main contributors to electron transport at RT.

Insulator-to-metal switching

We then observed the electrochemical redox reaction of SrTiO3. Figure 4a shows the changes of Id and Ig during Vg sweep from +10 to +40 V, as a function of retention time. The Id gradually increases with retention time. The rate of Id increase grows with Vg. The Ig also increases with Vg. The Id reaches ~1 mA at Vg=+40 V. The Id decreases drastically when negative Vg of −30 V is applied.

Figure 4: A redox reaction switches an insulating SrTiO3 to metal.
figure 4

(a) Retention time dependences of Id and Ig for the CAN-gated SrTiO3 FET at several Vg at RT (Vd=+2 V). The Id increases gradually with retention time at constant Vg. Ion density, which is the retention time integral of Ig, reaches ~9×1016 cm−2 when Vg=+40 V is applied. (b) Rxx versus ion density for the state A–E marked in a at RT. The grey line (slope=–1) indicates Rxx=(e·nxx·μ)−1, where nxx and μ are ion density, and μFE (0.8 cm2 V−1 s−1), which was obtained as in Supplementary Figure S4b. (c) Temperature dependence of Rxx for the state A–E marked in Figure 4a.

The sheet resistance (Rxx) at several points (A–E) was plotted as a function of ion density, which was obtained by Ig×t (Fig. 4b). The observed values were close to the grey line, which is Rxx=(e·nxx·μ)−1, where nxx and μ are ion density, and μFE (0.8 cm2 V−1 s−1) was obtained as in Supplementary Figure S4b, suggesting that carrier electrons were generated as a result of the electrochemical reduction of SrTiO3. Although samples A–C exhibit insulating RxxT behaviour, D and E exhibit metallic RxxT behaviour (Fig. 4c).

Exotic thermopower

We have also shown that the metal layer on the SrTiO3 in the FET exhibits novel thermoelectric behaviour: V-shaped turnaround of S is seen, although Rxx decreases gradually as Vg increases (Fig. 5). As we have reported previously, |S| value enhancement of electron-doped SrTiO3 can be observed when the conducting layer thickness is <~3 nm19, probably because of 2D effect24, where the thermal de Broglie's wavelength of conduction electrons in SrTiO3 is ~6 nm25. We therefore expected that 2D |S| can be observed at high Vg, because the layer thickness may become thinner.

Figure 5: Thermopower and sheet resistance as a function of applied Vg.
figure 5

Novel thermopower behaviour: V-shaped turnaround of S is seen, though Rxx (circle: before S measurement, square: after S measurement) decreases gradually as Vg increases, probably due to transition of electronic nature from 3D to 2D in the interface region.

When the Vg was <+26 V, |S| value gradually decreased with Vg. Similar |S| behaviour, which can be analysed using simple three-dimensional (3D) electron diffusion theory, was also observed in the dense a-C12A7-gated SrTiO3 FET, as shown in Supplementary Figure S5. Thus, we used the 3D electron diffusion theory26 to analyse the layer thickness. The 3D electron concentrations (n3D) at Vg=+16, +20 and +26 V were estimated to be ~1.5×1019, 1.5×1020 and ~1.1×1021 cm−3, respectively. The nxx values at Vg=+16, +20 and +26 V, which can be calculated using Rxx and μFE (0.8 cm2 V−1 s−1), are 1.4×1013, 9.0×1013 and 3.3×1014 cm−2, respectively, suggesting that the SrTiO3–δ thicknesses (~nxx/n3D) are ~9, 6 and 3 nm, respectively, confirming that the observed S obeys simple 3D electron diffusion theory. On the contrary, when the Vg>26 V was applied, the |S| value increased. The conducting layer thickness may be <3 nm, the observed upturn at Vg>26 V is most likely due to 2D effect19,24.

Discussion

The present CAN-gated SrTiO3 FET has several merits compared with established methods such as thermal reduction5, ion irradiation17,27, thin film growth14,15,17,18 and simple FET21,28, which are often utilized to switch an insulating SrTiO3 to a metal. This is because the current method allows a metal layer to be easily fabricated with extremely low energy. First, required electricity is extremely less (~50 μW cm−2 for 'D' in Fig. 4) compared with that of thermal reduction (~kW) and ion implantation (~W cm−2). Second, an extremely thin (~3 nm) metal layer, which exhibits exotic thermopower (Fig. 5), can be fabricated. Normally, such thin layers must be fabricated through complicated vapour phase epitaxy methods, such as pulsed laser deposition and molecular beam epitaxy. Third, our CAN-gated SrTiO3 FET exhibits a nonvolatile metal or highly conductive/insulator transition (Figs 3a and 4a and Supplementary Fig. S6), because a reversible redox reaction is utilized in addition to the field effect.

Although several efficient gating methods using liquid electrolytes have been proposed very recently28,29,30,31, we would like to argue that the present water-infiltrated nanoporous glass 'CAN' is truly superior to the liquid-based gate dielectrics. Liquid electrolytes including 'gel' would be very useful to electrostatically accumulate carriers at the transistor channel by applying rather low Vg (a few volts) using their huge capacitance. However, they would not be suitable for practical applications without sealing because of liquid leakage problem. Our 'CAN' is a chemically stable rigid glassy solid with a higher decomposition voltage, showing excellent adhesion with chemically robust oxide such as SrTiO3 surface and no liquid (water) leakage occurs. Carrier injection/discharge can be controlled by Vg as redox reaction, which occurs at the interface of H+ or OH/semiconductor, is utilized for carrier injection/discharge, though rather high Vg (several ten volts) should be required for Redox reaction. Further, we observed novel thermopower behaviour: V-shaped turnaround, probably due to transition of electronic nature from 3D to 2D in the interface region, as the FET could be operated at high Vg. These clearly indicate the effectiveness of our 'CAN'.

In summary, we have demonstrated that water, when it infiltrates nanoporous glass, can switch an insulating oxide to metal. As an example, we have built a FET on an insulating oxide, SrTiO3, using water-infiltrated nanoporous glass, amorphous 12CaO·7Al2O3 with nanoporous structure 'CAN' as the gate insulator. First, the insulating SrTiO3 surface became slightly conductive with gate voltage because of electrostatic charge accumulation. Then, H+/OH ions were generated due to water electrolysis occurring between the gate and the SrTiO3 surface. Subsequently, a redox reaction took place between H+/OH ions and the SrTiO3 surface. As a result, a thin metal (~3 nm) layer with extremely high electron concentration of 1015–1016 cm−2 was formed on the insulating SrTiO3. The electron activity of water, as it infiltrates nanoporous glass 'CAN', may find many useful applications in electronics or energy storage.

Methods

Fabrication of the CAN-gated SrTiO3 FET

The FET structures (Fig. 2a) were fabricated on the (001)-face of SrTiO3 single crystal plate (10×10×0.5 mm, SHINKOSHA), treated in NH4F-buffered HF (BHF) solution32. First, 20-nm-thick metallic Ti films, used as the source and drain electrodes, were deposited through a stencil mask by electron beam evaporation (base pressure ~104 Pa, no substrate heating/cooling) onto the SrTiO3 plate. Ohmic contact between the Ti and SrTiO3 surface was confirmed by conventional IV characteristics (output characteristics), as shown in Supplementary Figure S3b. Then, a 200-nm-thick CAN film was deposited through a stencil mask by pulsed laser deposition (KrF excimer laser, fluence ~3 J cm2 per pulse) at RT using dense polycrystalline C12A7 ceramic as target. During the CAN deposition, the oxygen pressure in the deposition chamber was kept at 5 Pa. Finally, a 20-nm-thick metallic Ti film, used as the gate electrode, was deposited through a stencil mask by electron beam evaporation.

Analysis of the CAN films

The bulk density and thickness of the CAN films were evaluated by grazing incidence X-ray reflectivity (ATX-G, Rigaku). Microstructures of the CAN films were observed by using TEM (JEOL JEM-2010, acceleration voltage=200 kV) and aberration-corrected STEM (JEOL JEM-2100F, acceleration voltage=200 kV). Water concentration in the CAN films was analysed by TDS measurements (TDS1200, ESCO). The TDS measurements were carried out in a vacuum chamber with the background pressure of ~10−7 Pa at varied temperatures from 60 to 700°C at a heating rate of 60°C per min.

Electron transport properties

Electrical properties of the FETs were measured by using a semiconductor device analyser (B1500A, Agilent). The capacitance of the CAN layer on the FETs was measured using an LCR meter (4284A, Agilent). The thermopower (S) values were measured using two Peltier devices under the FET to give a temperature difference between the source and drain electrodes (Supplementary Fig. S7). Two thermocouples (K-type) located at both ends of the channel were used for monitoring the temperature difference (ΔT, 0–5 K). S values were measured after each Vg sweep (for example: Vg application 0 V→+16 V→0 V→S measurements).

Additional information

How to cite this article: Ohta, H. et al. Field-induced water electrolysis switches an oxide semiconductor from an insulator to a metal. Nat. Commun. 1:118 doi: 10.1038/ncomms1112 (2010).