Introduction

Investigations on energy harvesting have become a mainstream in materials science, integrating different research disciplines such as physics, chemistry, electronics, and engineering1,2,3,4. This trend is motivated partly by the global requirement to develop advanced energy saving technologies5,6, aiming to reduce the current considerable carbon emission. Simultaneously, there is an urgent need to provide micro IoT (Internet of Things) sensors with stand-alone power systems7,8, the number of which will be skyrocketing in the coming decades. Thermoelectric energy conversion, which is a transformation of waste heat into electricity, has manifested itself as a promising CO2-free technology for generating electricity from environment9. In order to realize more useful and efficient thermoelectric energy conversion, new strategies have been proposed, such as flexible devices10,11, nano structures12,13, and ionic thermoelectric materials14,15,16. Furthermore, besides the continuous experimental efforts, computer-based approaches have emerged and been transforming traditional processes of experimental researches.

For the past decade, high-throughput calculations and machine learning have been advancing dramatically, demonstrating the indisputable utility of these approaches for the exploration of thermoelectric materials17,18. Computer-based materials research deductively or statistically investigate materials, structures, and compositions, to predict promising thermoelectric materials that should be evaluated experimentally. Thus, both experimental and computer-based approaches for exploration of thermoelectric materials should proceed together in tandem and feedback the results between them. However, considering the fast growth of computer-based science, experimental side would be required to accelerate; in general, evaluation of the thermoelectric properties requires great experimental effort because the figure of merit or power factor depends on mutually related parameters of the Seebeck coefficient S and the electrical conductivity σ, which drastically change with the carrier density n.

Recently, electrolyte gating techniques have been applied to a variety of materials and were found to be very effective for systematically controlling n19. By just applying several volts of the gate voltage VG, highly insulating electronic states are converted into conducting metallic states and even superconducting phases in oxides20,21,22, two dimensional materials23, and inorganic semiconductors24. Furthermore, continuous carrier doping using liquid electrolytes was found applicable to materials in which conventional methods such as bulk chemical doping were ineffective25,26,27. Thus, gate scanning of the thermoelectric properties in a wide temperature T and VG (i.e., n) space would definitely help accelerate thermoelectric researches from an experimental side.

In this study, we present the gate control of thermoelectric properties in WO3, a functional oxide semiconductor that has been attracting attention28,29,30,31,32,33,34. Although thermoelectric measurements in WO3 and related materials have been conducted especially in the last decade28,35, basic characteristics such as n and T dependences have not been investigated systematically, which makes WO3 the best choice to study the effect of electrolyte gating. We synthesized WO3 thin films with the thickness of 30 nm on a substrate of yttria stabilized zirconia (YSZ). The synthesis process and the X-ray diffraction (XRD) data (see Supplementary Figure S1) were reported elsewhere36. An electric double layer transistor structure was fabricated by depositing a drop of an ionic liquid on the WO3 thin film, as shown in Fig. 1a and b. We systematically investigated the thermoelectric power factor of WO3 and mapped the data for a wide range of T and VG. The results demonstrate how the thermoelectric properties develop against carrier doping, providing a solid guideline for the investigation of much higher thermoelectric performance in WO3.

Figure 1
figure 1

source current ID versus VG). The values of ID increases with increasing VG, showing that the electron carriers are doped under positive VG. The drain voltage VD was 0.1 V.

Structure and typical transistor operation of ion gated device based on WO3 thin film. (a) Schematic device structure to measure the Seebeck effect under gate voltage VG. In this configuration, the resistance and the Seebeck coefficient are simultaneously measured. Here, Th1(Th2), T, and ΔV stand for the resistive thermometer, temperature, and thermoelectric voltage, respectively. A drop of an ionic liquid was deposited to cover the WO3 channel and the gate electrode. See the “Experimental section” for more details of the device structure. (b) Optical image of WO3 device with ionic liquid on top. (c) Schematic diagram of alignment of ions under VG. When a positive VG is applied to the gate electrode, the cations align on the surface of the WO3 thin film to form the electric double layer, which could trigger both the electrostatic carrier accumulation and electrochemical doping. (d) Typical transfer characteristics (drain-

Full size image

Results and discussion

The details of the device structure and the concept behind its fabrication are summarized in Fig. 1. The device was patterned using the standard photolithography technique. As seen in Fig. 1a, b, and Supplementary Figure S2, a drop of the ionic liquid covered the WO3 thin film and the gate electrode, forming the electric double layer on the surface of WO3. Figure 1c schematically shows that the application of the positive gate bias causes the cations align on the surface of WO3, inducing a strong electric field on the interface. A typical transfer characteristic (VG dependence of the drain source current ID) at 297 K is given in Fig. 1d, suggesting that electron carriers were induced under the positive gate bias. The hysteresis was observed possibly due to the slow re-formation of ions against VG. Two resistive thermometers and a heater were prepared by the metal evaporation on the same substrate. The heater was placed in the immediate vicinity of WO3 to induce a thermal gradient along the channel. The local temperatures were monitored using the two thermometers, Th1 and Th2, and the thermoelectric voltage ΔV was measured using the electrodes connected to both ends of the channel (See Supplementary Notes). Therefore, the device structure prepared for this study allows for the evaluation of thermoelectric properties under the application of gate electric field (See “Experimental section”)37.

The thermoelectric effect was measured in this on-chip device structure (Fig. 1a and b) and systematically controlled under the gate bias. Figure 2a shows the temperature difference ΔT between Th1 and Th2 (see Supplementary Figure S2) for different values of the heater current IH. The values of ΔT showed a stepwise increase with increasing IH; ΔT was stable for a fixed value of IH. It was also confirmed from Fig. 2b that ΔT was proportional to IH2, validating that the thermal gradient on the WO3 thin film was induced solely by the Joule heating of the heater. Figure 2c shows the ΔV − ΔT plot of the WO3 thin film for different VG at 295 K. The values of ΔV linearly increased with ΔT, indicating that the thermoelectric effect was correctly measured. The slope of the ΔV − ΔT plot was continuously suppressed with increasing VG, which shows that the absolute values of S were suppressed. It is worth noting that the change of the thermoelectric response seen in Fig. 2c is consistent with the gate induced ID modulation seen in Fig. 1d. Generally, in semiconductors, |S| is suppressed by the shift of the Fermi level to higher energy regions38. Thus, both results in Figs. 1d and 2c are the direct outcome of the gate-induced electron doping into the WO3 thin film.

Figure 2
figure 2

On-chip thermoelectric measurements of WO3 thin film. (a) Temperature difference ΔT for different heater current IH. When IH was applied to the heater in Fig. 1a, ΔT was induced between the edges of the channel due to the Joule heating of the heater. The values of ΔT increased with increasing IH. (b) Heater power dependence of ΔT. The values of ΔT linearly increased against IH2, confirming that the temperature gradient on the sample is solely attributed to the Joule heating of the heater. (c) Thermoelectric voltage ΔV under ΔT in WO3 thin film. The measurements were performed at 295 K. The slope of the ΔV − ΔT plot was systematically suppressed with increasing VG.

Full size image

The gate-induced carrier doping is much clearly illustrated in the T dependence measurements of the electrical and thermoelectric transports. Figure 3a shows the T dependence of the sheet resistance Rs of the WO3 thin film for different VG. When VG = 0 V, Rs showed a large value of ~ 0.5 MΩ at 295 K and semiconducting behavior against T. With increasing VG, Rs systematically decreased, accompanied by a transition from an insulating to a metallic state33,36. The lowest value of Rs obtained here was ~ 200 Ω for VG = 3.9 V, which is comparable to the previous reports on ion-gated WO330,31,36. As shown in Fig. 3b, we simultaneously evaluated the thermoelectric response of WO3 with the measurement of Rs. The values of S were gradually suppressed with increasing VG in the entire measured T range. For low VG values, S was obtained only at high temperatures because the resistance between the voltage probe for ΔV measurements including the contact resistance became too high (> ~ MΩ), which made it difficult to reliably measure ΔV at low temperatures. For large VG values, S approached zero with decreasing T, which is typical behavior in metallic semiconductors38. Here we note that the systematic modulation of S in bulk WO3 has never been realized by conventional chemical doping method; the electric double layer doping is a unique approach to explore the thermoelectric properties of semiconductors.

Figure 3
figure 3

Gate induced transition from insulator to metal in WO3 thin film. (a) Temperature T dependence of sheet resistance Rs under gate bias. The values of Rs are suppressed by the application of gate voltage VG. Rs showed insulating behavior at VG = 0 V, whereas a flat T dependence was observed at VG = 3.9 V. (b) T dependence of Seebeck coefficient S under gate bias. The values of S were gradually suppressed with increasing VG in the entire measured T region. For low VG values, S was obtained only at high temperatures because the resistance between the voltage probe for ΔV including the contact resistance was too high (> ~ MΩ). For large VG values, S approached zero with decreasing T, which is typical behavior in metallic semiconductors. The error bars correspond to the standard error in the linear fitting of ΔV–ΔT plot.

Full size image

The overall thermoelectric property of the WO3 thin film is electrically visualized in a wide T and VG space. Figure 4a shows the contour plot of the thermoelectric power factor S2σ, using the data from Fig. 3. The electrical conductivity σ was estimated as σ  = 1/(d × Rs), where d is the thickness of the thin film, ~ 30 nm. Here, we assume that the carrier accumulation layer thickness is the same with d. Generally, in field effect transistors, only the topmost layer as thin as several nanometers is affected by the gate electric field. However, it has been reported that electrolyte gating can uniformly dope electrons into the whole WO3 thin film, even for thick films having a thickness of 70 nm30,33. This suggests that the electron doping does not proceed electrostatically but occurs electrochemically in the ion-gated WO3 thin film30,33.

Figure 4
figure 4

Optimization of thermoelectric power factor in WO3 thin film. (a) Mapping of thermoelectric power factor S2σ against temperature T and gate voltage VG. The value of S2σ increased with increasing VG and showed a maximum at VG ~ 3.1 V. (b) Evolution of S against electrical conductivity σ. The values of S change linearly with ln(σ), which holds for a variety of semiconductors45. (c) Optimization of S2σ of WO3 at room temperature.

Full size image

The systematic modulation of the thermoelectric property in Fig. 4a suggests the continuous change in the electronic structure for the WO3 thin film. In the low VG region, Rs increased with decreasing T, as shown in Fig. 3a, suggesting that the WO3 thin film is nondegenerate with an activation energy Ea39,40. The relationship between σ and T can be expressed with an Arrhenius type equation,

$$\sigma = \sigma_{0} {\text{exp}}\left( { - \frac{{E_{{\text{a}}} }}{{k_{{\text{B}}} T}}} \right)$$
(1)

where σ0 is a pre-exponential factor, and kB is the Boltzmann constant. We applied Eq. (1) to the experimental curve of Rs for the most resistive state with VG = 0, according to the analysis conducted by Mattoni et al40. The estimated value of Ea was ~ 106 meV (see Supplementary Figure S5), which is one order of magnitude smaller than the insulating gap of WO3, ~ 3.0 eV28,41,42,43. This suggests that, at VG = 0 V, a shallow donor level exists below the conduction band bottom due to oxygen vacancies in the film40. On the other hand, in the large VG region, Rs was drastically suppressed and showed a flat T dependence, as shown in Fig. 3a. A systematic trend can be also seen in the thermoelectric response in Fig. 3b. The absolute value of S was gradually suppressed with increasing VG due to the increase in the carrier doping level. These suggest that the WO3 thin film was highly doped and degenerate.

Our electrochemical approach modifies S of WO3 in a similar manner with the bulk chemical doping on other inorganic semiconductors. The relationship between S and σ in doped semiconductors is described by the following expression44,45,46 as

$$S = - \frac{{k_{{\text{B}}} }}{e} \times \left( { - \ln \sigma + \ln \left[ {N_{{\text{c}}} e\mu } \right] + \frac{5}{2} + r} \right)$$
(2)

where Nc, μ, and r, are the effective density of states for the conduction band, the carrier mobility, and the scattering parameter, respectively. Equation 2 suggests that S is proportional to ln σ when the n dependence of μ is moderate. This condition would hold in transition metal oxides with low carrier mobility at room temperature47,48. Actually, the linear relationship between S and ln σ has been confirmed for a variety of semiconductors although the slope of the plot often varies from kB/e for each material45. Figure 4b shows the change in S as a function of σ for the WO3 thin film, which demonstrated that a good linearity between S and ln σ reasonably held in WO3 as well. When we plot S2σ as a function σ, a peak was observed at σ  = 585 S cm-1, as shown in Fig. 4c. The value of S2σ was modulated and optimized through the continuous carrier modulation from an insulating to a metallic region, which has never been found in the studies of bulk WO3. Here, it would be noted from Fig. 4a that the highest S2σ would exist above room temperature. Expanding this approach toward much higher temperatures would be more useful for the investigation of novel thermoelectric materials.

Conclusions

In conclusion, we revealed the systematic evolution of thermoelectric properties of WO3 against electrochemical carrier doping. Ion gating is a versatile technique in carrier doping and continuously modulated n of the WO3 thin film. The value of S2σ was optimized to have ~ 0.16 μW K-2 cm-1 in a highly electron-doped region, where WO3 exhibited the metallic electrical and thermoelectric transports. These findings could serve as an important guideline for exploring much higher thermoelectric performance in bulk WO3 and its related materials such as nanocrystalline WO3 and WO3-based composites28,49,50. The current approach, the T − VG (i.e., n) mapping of S2σ, is applicable to various functional materials other than WO3 as well. Its synergetic combination with other approaches such as computation and machine learning would make that a powerful tool, leading to the emergence of high performance multifunctional thermoelectric materials hidden in unexplored materials groups.

Experimental section

Thin film growth

The hexagonal WO3 film were grown on a YSZ(111) substrate by evaporating tungsten via RF sputtering with a RF power of 30 W. We followed the same procedure that was reported elsewhere36. The O2 partial pressure was controlled at ~ 2.2 mTorr in the growth chamber, and the Ar partial pressure was maintained at ~ 8 mTorr. Supplementary Figure S1 in Supplementary Information shows the XRD curves of out-of-plane and in-plane configurations, which confirmed that the hexagonal structure of WO3 was realized. The thickness of the film was controlled by the deposition time and was estimated with X-ray reflectometry. The film thickness was estimated to be ~ 30 nm.

Fabrication of ion gated device

To control the electron carrier density in the WO3 thin film, an electric double layer transistor structure was fabricated, which is schematically shown in Fig. 1a. The optical image of the device is shown in Fig. 1b and Supplementary Figure S2. The channel length and width were 300 μm and 100 μm, respectively. The electrodes and the channel were patterned using standard photolithography techniques. All the electrodes required for the ion gating experiments were prepared by the evaporation of Cr/Au with the thickness of 3/35 nm. In addition, the resistive thermometers, Th1 and Th2, and a heater were also prepared using the same procedure. A small amount of an ionic liquid N,N-dimethyl-N-(2-methoxyethyl)-N-methylammonium bis-(trifluoromethylsulfonyl)-imide (DEME-TFSI) was deposited to cover both the channel and the gate electrode in order to conduct gating experiments with a side gate configuration. DEME-TFSI has the glass transition temperature at 182 K, below which the polarization of ions is fixed and cannot be modulated by VG51,52. Throughout the experiments, VG was applied and changed at 300 K.

Thermoelectric measurements

As schematically shown in Fig. 1a, we prepared a heater in close vicinity of one end of the WO3 channel; it produced a thermal gradient −∇T from the heater to the other edge of the channel37. Two resistive thermometers, Th1 and Th2, monitored the T differences, which was induced by the Joule heating of the heater (See Fig. 2). The values of S were estimated as −E/|∇T |, where E is the electric field induced through the Seebeck effect, by measuring the thermoelectric voltage between both ends of the WO3 channel. The calibration of resistive thermometers is described in detail in Supplementary Notes.