Fermi level limitation in Na_1/2Bi_1/2TiO₃–BaTiO₃ piezoceramics by electrochemical reduction of Bi Open Access

The solid solution Na_1/2Bi_1/2TiO₃–BaTiO₃ (NBT–BT) is a candidate material to replace Pb-containing Pb(Zr,Ti)O₃ in sensors, transducers, and energy harvesting devices.^1–4 At room temperature, there is a morphotropic phase boundary (MPB) at 6% BaTiO₃ between the rhombohedral and tetragonal phases. This enables NBT–6BT to exhibit high piezoelectric coefficients, which is favorable for its integration into technological applications.^5,6 Multilayer ceramic/electrode stacks are often used for applications in capacitors or actuators.⁷ During co-sintering of these stacks, the (electro)chemical stability of the ceramic/electrode interfaces can become a crucial issue, in particular, when expensive noble metal Ag–Pd electrodes are replaced by Ni or Cu base metals because sintering in reducing atmosphere is required to avoid oxidation of the metal electrodes.^8–10 

Reduction of the ceramic at the electrode interface can be induced either chemically or electrochemically. Chemical reduction occurs when the chemical boundary condition, i.e., the oxygen chemical potential or partial pressure, becomes too reducing and reaches the thermo-dynamic stability limit of the dielectric. The range of chemical stability can be described by Ellingham diagrams.¹¹ While the chemical boundary conditions typically determine the Fermi level in the bulk of the material, the Schottky barrier controls the Fermi level at the electrode interfaces depending on the electrode material. As Fermi level pinning is weak at interfaces of oxides, the work function of the electrode material determines the Fermi level at the interface as long as the interfaces are not reactive.¹² 

An electrochemical reduction or oxidation corresponds to the change of the oxidation state of one of the species of the compound. It occurs when the electrostatic boundary condition, i.e., the electron chemical potential or Fermi level, is raised or lowered, respectively. Such valence changes do not occur in classical semiconductors but can substantially limit the Fermi level inside the energy gap of polar materials such as oxides^13–16 and are essential for a proper description of the defect properties of materials.¹⁷ Therefore, it is important to determine the range of possible Fermi levels, for which a material is thermodynamically stable, i.e., the range of Fermi levels where all species have their nominal oxidation states.

The limits of the Fermi level can be determined by means of photoelectron spectroscopy, which simultaneously records the chemical composition, bonding, and the Fermi level of a sample. For example, Lohaus et al. observed a reduction of Fe³⁺ to Fe²⁺ in Fe₂O₃ when the Fermi level was raised to ≈0.5 eV below the conduction band minimum during the formation of an interface with 10% Sn-doped In₂O₃ (ITO).¹³ Deposition of ITO onto BiFeO₃ led to a reduction of Bi and Fe.¹⁴ Reduction of Bi and Fe in BiFeO₃ can also be induced by the adsorption of water.¹⁴ It can be assigned to an electrochemical origin, as it occurs at the same Fermi level of 1.7 eV above the valence band maximum for ITO deposition and H₂O adsorption.

It might be straightforward to study the range of the Fermi level by depositing metals with different work functions. However, deposition of metals on oxide surfaces often results in a reduction of the oxide surface due to the release of the metal's heat of condensation, which results in a defect-related Fermi level pinning at the interface.¹² The reaction can be avoided if conducting oxides, such as ITO or RuO₂, are used as contact material. ITO is a degenerate n-type semiconductor, whose carrier concentration and the Fermi level depend on its oxygen content.¹⁸ Due to its high electron concentration, the Fermi level in ITO is also determining the Fermi level in NBT–BT at the interface. Therefore, it is possible to vary the Fermi level at an oxide/ITO interface by changing the oxygen content in the ITO layer. Raising the Fermi level at the oxide/ITO interface can be achieved in two ways: (i) annealing in vacuum will result in the removal of oxygen from ITO to the gas phase as depicted in Fig. 1(a)^,19 and (ii) by using ITO as the cathode in an electrochemical cell illustrated in Fig. 1(b). If a voltage is applied across the cell at a high temperature, oxygen can be pumped from the cathode to the anode, thereby changing the oxygen content in the electrode depending on voltage polarity. Electrochemical cells have been employed in photoelectron spectrometer systems to study the changes in electrode materials upon oxidation and reduction.²⁰ If the electrode material has a thickness comparable to the inelastic mean free path of the photoelectrons, not only the electrode but also changes in the chemical composition, bonding, and the Fermi level at the surface of the electrolyte material can be monitored. Using this approach, Huang and co-workers have employed such a setup to study ferroelectric and antiferroelectric undoped and La-doped Pb(Zr,Sn,Ti)O₃ [P(L)ZST].¹⁶ They could demonstrate that raising the Fermi level at the interface upon electrochemical polarization results in the reduction of Pb at a Fermi level ≈1 eV below the conduction band minimum.

A prerequisite for such experiments is the ability of the “electrolyte” used in the cell to conduct oxygen ions. For NBT, it was found that acceptor doping or Bi nonstoichiometric results in very high oxygen ion conductivity.^21–24 The NBT–BT system, which is studied in this work, has also been shown to exhibit a mixed electronic/ionic conductivity.²⁵ At elevated temperatures, pure NBT, but also acceptor-doped NBT–BT, is very good oxygen ion conductors due to the presence of oxygen vacancies $( V O ⋅ ⋅)$ or defects associated with acceptor dopants $( A ′)$ with the induced oxygen vacancies $( [ A − V O ] ⋅)$⁠.²⁶ Therefore, NBT–6BT has sufficient oxygen ionic conduction at evaluated temperatures, which offers the possibility to use it as an electrolyte to observe the stability of the interface and the limits of the Fermi level.

In this work, the connection between the Fermi level and chemical properties at interfaces between ITO and NBT–6BT is studied by means of in situ x-ray photoelectron spectroscopy. The Fermi level in the ITO electrode is varied either by heating in vacuum or by electrochemical polarization using the setup depicted in Fig. 1. It will be shown that a reduction of Bi is observed when the Fermi level in NBT–6BT is raised to 2.23 ± 0.10 eV above the valence band maximum, corresponding to a Fermi level ≈1 eV below the conduction band minimum. The effect is accelerated by an increase of oxygen vacancy concentration upon acceptor doping of NBT–6BT.

Sample preparation: undoped and 1.0 mol. % Zn doped (substituted on the B-site according to the effective ionic radius)²⁷ NBT–6BT were prepared by solid-state synthesis. Powders with given purities were utilized: Na₂CO₃ (99.5%), BaCO₃ (99.8%), Bi₂O₃ (99.975%), TiO₂ (99.6%), and ZnO (99.99%) (all from Alfa Aesar GmbH and Co. KG, Germany). After weighing, powders were milled with yttria-stabilized zirconia balls in ethanol for 24h at 250 rpm (Zn doped: 6h, 250 rpm), then dried and homogenized, and finally calcined at 900 °C for 3h using a heating rate of 5 °C/min. After calcination, the samples were re-milled using the same conditions as above. Cylindrical samples of either 10 or 15 mm in diameter were pressed at a uniaxial pressure of 40 MPa followed by an isostatic pressure of 200–350 MPa. The samples were placed in a closed alumina crucible with sacrificial powder with the same composition and were sintered at 1150 °C for 3h using a ramp rate of 5 °C/min (Zn doped: 1100 °C, 1 h).²⁸ 

In situ x-ray photoelectron spectroscopy (XPS): the sintered ceramic samples were first ground with sandpaper (#800, #1200) to 0.45 mm thickness and then cut into a rectangular shape of 3 × 4 mm². The samples were subsequently annealed in air at 450 °C for 1.0h to relieve the residual stress introduced by machining. The bottom electrodes of Pt with a thickness of 50 nm were deposited with a sputter coater (Quorum Q300T D, Quorum Technologies Ltd., UK). Before the deposition of the ITO top electrode, a surface cleaning was carried out inside the deposition chamber of the Darmstadt Integrated System for Materials Research (DAISY-MAT; Fig. S1 in the supplementary material)¹² by heating in oxygen (0.5 Pa, 400 °C, 0.5 h) to get rid of adventitious carbon species.²⁹ Top electrodes of 10% Sn-doped In₂O₃ with a thickness of 2 nm were subsequently deposited by radio frequency magnetron sputtering either at room temperature for subsequent annealing or at 400 °C. Finally, the samples were mounted onto stainless-steel sample holders allowing for separate electrical contacts to the bottom and top electrode. The ITO electrode was connected to the ground, ensuring that the Fermi energy at the top of the sample is aligned with that of the spectrometer, which serves as a binding energy reference for the spectra. XPS measurements were then executed by a Physical Electronics PHI 5700 spectrometer system (Chanhassen, MN) with monochromated Al Kα radiation. Binding energies were calibrated using a sputter cleaned Ag foil. XPS measurements were performed either in the course of heating the samples inside the XPS chamber or by applying a positive voltage to the Pt electrode at elevated temperatures.

X-ray photoelectron survey spectra of nominally undoped NBT–6BT with 10% Sn-doped In₂O₃ top electrodes recorded in the course of vacuum annealing and electrochemical polarization are displayed in Figs. S2 and S3 in the supplementary material. Apart from carbon adsorbates, which are present on the sample used in the electrochemical cell setup as a result of air exposure of the sample, only emissions from the NBT–BT substrate and the ITO film are present. Due to the surface sensitivity of XPS, the spectra are dominated by the signals from the ITO layer despite its low thickness of 2–3 nm. No pronounced changes in the spectra are observed in the course of the treatments, confirming the absence of significant changes in the sample composition and ITO film morphology.

Enlarged Bi 4f_7/2 and In 3d_5/2 regions of the x-ray photoelectron spectra of a nominally undoped NBT–6BT sample coated with a 2–3 nm thick ITO film are presented in Fig. 2. The ITO films were purposely deposited at room temperature in order to start the experiment with an ITO film having a lower conductivity and the Fermi level.¹⁸ Heating the sample in vacuum is expected to increase the conductivity by partial removal of oxygen from the ITO film as it has been observed for a 10 nm thick ITO film on the glass substrate by in situ conductivity measurements.¹⁹ Consequently, the Fermi level in the ITO and at the NBT–6BT/ITO interface should rise when the sample is heated inside the XPS system. Raising the Fermi level will correspond to an increase in the binding energies, which are measured with respect to the Fermi level. The increase of binding energies can be observed in Fig. 2. However, heating the sample to 250 °C first results in a lowering of the binding energies of the Bi 4f and In 3d core levels. This shift can be assigned to the resistance of the room temperature deposited thin ITO layer, which may be too high to completely eliminate sample charging by the photoemission process. Heating the sample in vacuum results in an increase in the conductivity of the ITO.¹⁹ In addition, the conductivity of NBT–6BT will also increase with temperature.³⁰ Therefore, charging will be reduced at elevated temperatures. Further heating to 350 °C does not result in further changes in the binding energy. While keeping the sample at 350 °C, the binding energies of both core levels increase. The parallel shift of the binding energies of the Bi 4f and In 3d core levels indicates that the shift is induced by a change in the Fermi level (oxygen content) of the ITO layer. The ITO layer has the higher carrier concentration and is, therefore, determining the Fermi energy at the interface. After 5 h of annealing, no more binding energy shifts are observed. After cooling to room temperature, the binding energy of the Bi 4f_7/2 peak amounts to 159.51 ± 0.05 eV. This value is ≈1.2 eV higher than that observed at the interface between pure NBT and RuO₂.³¹ The difference of 1.2 eV corresponds well to the difference in work functions of ITO and RuO₂ and to the difference in interface Fermi energies observed for other dielectric perovskite oxides,^12,32 providing substantial credit for the absolute value of the binding energy.

Apart from the binding energy shifts and a slightly enhanced peak broadening at higher temperatures, the Bi 4f emission exhibits no changes in peak shape in the course of the experiment. Changes in the peak shape of the In 3d emission are also barely noticeable. Such changes are expected due to the partial crystallization of the ITO layer and the related increase of the carrier concentration of the ITO, which has been observed for ITO films on glass substrates.¹⁹ The removal of oxygen from the ITO should not only result in a higher binding energy of the peak but also in an asymmetric broadening toward higher binding energies.^29,33 The latter is related to the interaction of the photoelectrons with the free electron gas of the degenerately doped ITO layer. Apparently, the ITO film deposited on glass at room temperature is more reduced than those deposited on the NBT–BT substrates. The difference is assigned to a partial compensation of the oxygen loss from the ITO film on NBT–BT into vacuum by migration of oxygen from the NBT–BT substrates into the ITO film.

The experiment performed with electrochemical polarization of the nominally undoped NBT–6BT sample is displayed in Fig. 3. The binding energies of the Bi 4f and In 3d core levels after deposition are similar to those observed in the other measurements (see Figs. 2 and S4 in the supplementary material). In contrast to the experiment described above, the binding energies are reduced less after reaching 350 °C. The binding energies are rather comparable to those reached after 9h heating in vacuum. The different behavior upon heating of the two undoped samples is likely related to a difference in the conductivity of the ITO. Unfortunately, a higher but also a lower conductivity might explain the difference. In the first case, the ITO would have a higher Fermi level and, therefore, also a higher binding energy, in the second case, the conductivity may be too low, and the binding energies increase by sample charging during the measurement. The conductivity of the ITO layer is affected by the deposition conditions but also by exchange of oxygen with the substrate. While it is desirable to understand the details of these relations, they do not affect the conclusions of this manuscript.

Applying an electric field across the sample with cathodic polarization of the ITO top electrode results in a reduction of the ITO layer by migration of oxygen. This reduction results in an increase of the binding energies of the core levels (see Fig. 3). During each step of the electric field application, only the Bi 4f and In 3d core-level spectra were measured, with each measurement taking about 20 min before proceeding to the next, higher electric field. Once the electric field reaches a value of 10 V/mm, corresponding to an applied voltage of 4.5 V, a very small additional peak emerges in the Bi 4f spectra at a binding energy of ≈157 eV. This peak can be assigned to metallic Bi.³⁴ The behavior is analogous to that observed for (anti)ferroelectric PLZST, where the cathodic electrochemical polarization induced a reduction of Pb.¹⁶ By comparing observations for different samples, Huang et al. found that the reduction of Pb always occurred when the Pb 4f binding energy reached a certain value, which corresponded to a Fermi energy of ≈2.45 eV above the valence band maximum. As indicated by Fig. 3(c), the reduction of Bi occurred in the present case when the binding energy of the Bi³⁺ component of the Bi 4f core level reached a value of 159.6 eV. This value cannot be used directly, however. The reason is that the binding energies of the top electrodes are affected by a voltage drop across the ITO layer when the current through the cell becomes too high. The voltage drop across the ITO layer cannot be avoided but may be too small to be recognized if the current through the cell is low.¹⁶ The voltage drop is related to the resistance of the ITO layer and, therefore, to the low thickness. The latter is required for the experiment in order to be able to monitor the changes induced in the NBT–6BT layer. The uncertainty of extracting the binding energy at which the reduction of Bi occurs can be avoided if the value is taken at the end of the electric field sequence when the sample is still at high temperature, but the electric field has been removed. The binding energy at this step of the experiment amounts to 159.52 eV. This is 0.14 eV higher than the highest binding energy reached in the first experiment, where only heating in vacuum is used to reduce the ITO layer and raise the Fermi energy. The higher Fermi energy reached in the electrochemical reduction experiment is consistent with the observation of metallic Bi, while no metallic Bi is observed during reduction by vacuum annealing.

The two experiments described above have also been performed with samples containing different concentrations of acceptor doping. Both Mg and Zn have been used as acceptors. In principle, the results for the acceptor-doped NBT–6BT are the same as those described above for the nominally undoped material. No formation of metallic Bi is observed if the reduction is performed by heating inside the XPS system and metallic Bi is observed if an electrochemical polarization is carried out. As an example, the experiments performed with 1 mol. % Zn-doped NBT–6BT are presented in the supplementary material. There are some differences compared to the undoped sample: (i) the binding energy of the oxide Bi 4f_7/2 component reaches 159.49 eV already during vacuum annealing without inducing metallic Bi and (ii) during electrochemical polarization, metallic Bi is already observed for electric fields as low as 2 V/mm (0.9 V) and the binding energy of oxide Bi 4f_7/2 component is only 159.46 eV. Considering the uncertainty of the binding energies of 50 meV, there is no contradiction to the statement that an electrochemical reduction of Bi occurs when the binding energy reaches 159.5 eV. The lower field required to induce the formation of metallic Bi is consistent with the very high oxygen ion conductivity of acceptor-doped NBT and NBT–6BT.^24,30

In contrast to the case of BiFeO₃, where raising the Fermi level resulted in a simultaneous reduction of Bi and Fe,¹⁴ and to the case of BiVO₄, where Bi and V were reduced,¹⁵ no reduction of Ti was observed in the case of NBT–6BT. This is confirmed by the Ti 2p spectra, which are displayed for the electrochemical polarization experiment in Figs. 4(a) and 4(b). The Ti 2p spectra for the two vacuum annealing experiments are provided in the supplementary material. The shape of the Ti 2p peak does not change in any of the experiments. Therefore, a reduction of Ti from its nominal oxidation state Ti⁴⁺ to Ti³⁺ can be excluded, as this would give rise to an additional Ti 2p emission at binding energy below 458 eV.³⁵ 

The evolution of the Na 1s spectra during the electrochemical polarization of the samples is displayed in Figs. 4(c) and 4(d). Those recorded in the course of vacuum annealing are presented in the supplementary material. No change of peak shape is observed. There are also no substantial changes in the peak intensity apart from those induced by the temperature change. The latter can be explained by a slightly varying sample position due to the thermal expansion of the sample holder. The absence of pronounced changes in the Na intensity is different from the substantial Na enrichment observed at the surface of NBT surrounding the anode region after resistance degradation, which should be related to a reduction of the Na content inside the material at the anode.²⁶ The main difference here is the use of the ITO electrode as a cathode. If Na ions are mobile in the presence of an electric field, it has to be expected that Na ions are attracted toward the cathode with polarization. However, the applied electric fields are probably not sufficient to induce such a migration. Indeed, the fields used in the present experiments are more than one order of magnitude lower than those used in the study of resistance degradation.²⁶ This is particularly the case for the acceptor-doped sample, where fields of only 2 V/mm raise the Fermi level and induce the formation of metallic Bi [see Fig. S4 in the supplementary material). The magnitude of the field which can be applied depends on the oxygen ion conductivity of the electrode. Acceptor-doped oxides such as the Zn-doped NBT–BT have a rather high oxygen ion conductivity, which is confirmed by the lower electric fields required for reducing Bi, while donor-doped materials have low oxygen vacancy concentration and low oxygen ion conductivity. For example, donor-doped Pb-based perovskites could sustain fields of up to 2 kV/mm at 250 °C.¹⁶ The undoped NBT–BT samples used in this study exhibit a low electrical conductivity and n-type behavior,³⁶ which is induced by some background donor doping. Therefore, the undoped NBT–BT exhibits a lower oxygen ion conductivity and either higher voltages or higher temperatures are required to induce a reduction of the ITO and the desired increase of the Fermi level. The experiments were initially conducted at 250 °C, as this temperature was expected to induce sufficient mobility of oxygen vacancies. However, it appeared that this temperature was inadequate for achieving the desired mobility. Consequently, we increased the temperature to 350 °C to ensure sufficient activation of oxygen vacancies and to facilitate the observation of electrochemical reduction under applied electric fields. We then kept the temperature for the Zn-doped sample for better comparability. As expected, the voltage needed to induce the reduction (oxygen ion mobility) of the Zn-doped sample was much lower than for the undoped one.

As described above, the metallic Bi appears when the binding energy of the oxide Bi 4f_7/2 emission reaches a binding energy of 159.5 ± 0.05 eV. This value corresponds to a reduction potential (charge transition level) of Bi^3+/0 and this transition level is considered to be the upper limit of the Fermi level of NBT–6BT. A direct reduction of Bi from Bi³⁺ to Bi⁰, which is indicated by the XPS measurements, is consistent with the lack of intermediate oxidation states of Bi. In order to relate the binding energy of the Bi 4f_7/2 emission to the Fermi energy in the bandgap, the distance between the Bi 4f_7/2 core level and the valence band maximum is required. These data can be derived from room-temperature XPS data acquired on NBT bulk ceramic samples as $BE VB Bi 4 f=157.27 eV$⁠.³¹ Since the energy gap varies with temperature, the Fermi level position is taken from the Bi 4f binding energies after cooling to room temperature without applied field, yielding E_F–E_VB = 2.23 ± 0.10 eV. This level is also calculated by the binding energy of the Ti 2p core level. With an energy gap of 3.3 eV,¹² the Fermi energy in NBT–6BT will be restricted to values ≈1 eV below the conduction band minimum. Because of this upper limit of the Fermi energy, free electrons in the conduction band will not substantially contribute to the electrical conductivity of this material. We like to emphasize that this is only valid for electrons in the conduction band, not for holes in the valence band. The latter will be relevant for acceptor doped, i.e., oxygen ion conducting NBT-based materials.

The reduction potentials, i.e., the Fermi levels at which a species is reduced, are compared for different piezoelectric materials in Fig. 5. The reference energy is taken to be the valence band maximum of BaTiO₃, and the energy bands of NBT, BiFeO₃, and Pb(Zr,Ti)O₃ are arranged according to the Fermi level positions of their interfaces with RuO₂, which is 1.05 eV for NBT,³¹ 0.7 eV for BiFeO₃,³¹ 1.0 eV for PZT,³⁷ and 1.7 eV for BaTiO₃.³⁸ According to this graph, the upper limit of the Fermi level in NBT–6BT and BiFeO₃, which are both related to a reduction of Bi, is at a similar energy. The reduction potential of Bi may, therefore, also be transferable between compounds, similar to what has been reported for a transition metal, hydrogen, and oxygen vacancy related defect states.^39–41 The limit of the Fermi level determined here for NBT–6BT is furthermore similar to that observed in PZST, 2.45 eV above the valence band maximum, which is caused by the reduction of Pb.¹⁶ According to Fig. 5, the limits of the Fermi level in NBT–6BT, BiFeO₃, and PZT are still lower than those of BaTiO₃. The latter has energy bands, which are substantially lower than those of the Bi and Pb-containing compounds due to the lack of occupied 6s orbitals.³¹ BaTiO₃ is furthermore known to become an electronic conductor upon reduction or donor doping, which is, for example, the basis for the positive temperature coefficient resistor (PTCR) effect,⁴² and Fermi levels close to the conduction band have been measured by XPS on reduced poly- and single crystalline samples.^38,43

The electrochemical reduction of Pb reported for (anti)ferroelectric PLZST has been observed for materials containing 75% of Zr on the B-site of the perovskite lattice.¹⁶ For Zr-rich Pb(Zr,Ti)O₃, the minimum of the conduction band is formed from Pb 6p orbitals.⁴⁶ The observed reduction of Bi, instead of the reduction of Ti, may, therefore, be related to a dominant contribution of Bi 6p states to the conduction band. Band structure calculations reported in the literature indicate a similar contribution of Bi 6p and Ti 3d states at the conduction band minimum.⁴⁸ However, these calculations do not take different potential orderings of the A-site cations into account. Figure 6 depicts the calculated density of states of pure NBT for different A-site orderings. Four polymorphs, namely, the 111-ordered rhombohedral R3 [Fig. 6(a)] phase, the tetragonal $I 4 ¯2 m$ phase [Fig. 6(d)], the cubic $Fm 3 ¯ m$ phase [Fig. 6(c)], and the 001-ordered orthorhombic structure Pmc2₁ [Fig. 6(b)], are considered.⁴⁹ At first sight, the density of states of all symmetries appears to be similar. In all cases, we have a substantial Bi 6s overlapping with the oxygen 2p states band, while a relatively small contribution of Bi 6s can be found in the upper valence band, as already pointed out by Gröting⁵⁰ for the cubic aristotype. Bi 6p states are primarily at the lower valence band and in the conduction band. The electronic states at the conduction band minimum are dominated either by Ti 3d orbitals or by Bi 6p states in the cubic $Fm 3 ¯ m$ phase [Fig. 6(c)]. Total energy calculations including unit cell relaxation indicated that the 001 ordering is energetically favorable against 111 ordering.⁵¹ This could indicate that Ti 3d states are occupied (reduced) before Bi 6p states. However, it is not clear at all whether a direct correlation between the orbital contributions to the conduction band states and the preference for (electrochemical) reduction exists. An electrochemical reduction corresponds to a valence change of an ion, i.e., to the localization (trapping) of electrons on the ion. Such an occupation, which is equivalent to the formation of a polaron, will induce a significant lattice distortion and thereby a change in the electronic structure. It is also not clear how much the partial substitution of Na and Bi by Ba is affecting the A-site ordering. Despite all the uncertainties, a contribution of Bi 6p states to the conduction bands is evident from the calculations, making a reduction of Bi upon the addition of electrons feasible.

Apart from a clear preference for the reduction of Bi compared Ti, the experiments further reveal no difference between nominally undoped and 1% Zn-doped NBT–BT. While Zn is expected to have a nominal oxidation state of +2 for most values of the Fermi level, a reduction of Zn²⁺ to Zn⁰ might occur for high Fermi levels. A tentative estimation suggests a reduction of Zn²⁺ near the conduction band minimum of BaTiO₃.¹⁷ Following the energy band positions given in Fig. 5, this should be above the reduction potential of Bi, although a lower level cannot be excluded due to the uncertainty of the estimate. If the reduction of Zn would already occur at a lower energy than that of Bi, the presence of Zn acceptors could prevent the reduction of Bi as Zn would have to be reduced first.¹⁷ Such a confinement of the Fermi level has been observed after partial substitution of Fe in BiFeO₃ by Co.¹⁴ However, the concentration of Co in BiFeO₃, which prevented the reduction of Bi, has been 10% of the total B-site content, which is ten times higher than the Zn doping concentration in the samples used in this study. It can, therefore, not be excluded that Zn has the potential to prevent the reduction of Bi. Higher Zn concentrations would be required to verify this.

The electrochemical instability at the electrode interface will result in a decomposition of NBT–6BT, whenever the Fermi level is raised to ≈2.3 eV above the valence band maximum. Such a situation may occur at an interface to an electrode having a low work function. The work function of ITO under reducing conditions (high Fermi level) amounts to 4.2–4.3 eV.⁵² Metals typically used as electrodes in multilayer ceramic capacitors, Ag, Ag–Pg, Cu, Ni, or Pt, exhibit work functions of 4.5 eV or higher. An electrochemical decomposition at the electrode interfaces is, therefore, not expected. Experimentally, this is difficult to prove, as the deposition of metals onto NBT-based materials is expected to result in a reduction of the oxide surface even in the case of very noble metals such as Pt or Au due to the release of the condensation energy.¹² Nevertheless, it is desirable to perform the electrochemical experiment also with metal electrodes. In such experiments, the interface Fermi level should not vary upon electrochemical polarization. Therefore, the experiments could reveal the influence of field-induced oxygen ion migration on the interface without changing the electrostatic boundary condition. This scenario is closer to the operating conditions of multilayer ceramic capacitors or actuators and will contribute to the understanding of resistance degradation. In acceptor-doped SrTiO₃ and BaTiO₃, resistance degradation is explained by the field-induced migration of oxygen vacancies toward the cathode, which leads to oxidation of the anodic and a reduction of the cathodic region.^53,54 Although the darkening of the anode region, which is caused in Fe-doped SrTiO₃ by the oxidation of the Fe acceptors,⁵⁵ is also induced by resistance degradation of NBT–BT,²⁶ a microscopic description of degradation must be different due to the different mechanisms of electronic conduction in the reduced cathode region. Band conduction by free electrons is possible for SrTiO₃ and BaTiO₃ but not for NBT-based materials.

Metal electrodes could also be used as anode materials in electrochemical cell experiments, which is not possible with ITO as it becomes highly resistive upon anodic polarization. The changes induced at the anode upon electrochemical polarization would be quite different from those at the cathode. For NBT–6BT, this may include the reduction of the Na content²⁶ or an oxidation of Bi. The latter would be analogous to hole trapping on Pb, which has been reported for Pb(Zr,Ti)O₃.^56,57 For SrTiO₃, a delamination of Pt electrodes due to excessive extraction of oxygen has also been reported,⁵⁸ indicating that the effective oxygen partial pressure can be at the limit of the material's stability in the anode region. Under such conditions, it should be possible to determine the lower limit of the Fermi energy in the material.

The upper limit of the Fermi level in NBT–6BT and acceptor (Zn) doped NBT–6BT has been determined using in situ photoelectron spectroscopy. Polycrystalline bulk ceramic materials have been coated with 2–3 nm thick 10% Sn-doped In₂O₃ layers and exposed to high temperature inside the ultrahigh vacuum environment of the spectrometer system. Annealing in vacuum raises the Fermi level due to the removal of oxygen from the ITO layer. Even higher Fermi levels can be obtained when a second Pt electrode is applied to the bottom of the NBT–6BT. Applying a voltage across the cell inside the spectrometer allows us to monitor the changes in the ITO electrode and the NBT–BT surface below the electrode. Upon cathodic polarization of the ITO electrode, oxygen vacancies are attracted and lead to a reduction of the ITO electrode, which is more pronounced than that obtained during vacuum annealing. The effect is much faster and occurs already at electric fields of only 2 V/mm in the case of acceptor doped, which is a consequence of a very high oxygen ion conductivity. Metallic Bi is observed when the Fermi energy at the NBT–6BT surface is raised to E_F–E_VB = 2.23 ± 0.10 eV. No reduction of the oxidation states of Ti, In, or Sn are observed.

The situation is analogous to that observed in BiFeO₃ and Zr-rich Pb(Zr,Sn,Ti)O₃, where a reduction of Bi and Pb has been observed, when the Fermi level is raised to E_F–E_VB = 1.7 and 2.45 eV, respectively. Due to the limits of the Fermi energies in NBT–BT and Zr-rich PZST, the Fermi energy is restricted to values ≈1 eV below the conduction band, preventing high concentrations of free electrons. The reduction of Bi and Pb can, therefore, be considered as an intrinsic mechanism to suppress n-type electronic conductivity in these materials. By comparing with Zr-rich PZT, it has been discussed whether the preferred reduction of Bi instead of Ti in NBT–6BT might be related to a dominant contribution of Bi 6p orbitals to the electronic states at the conduction band minimum. DFT calculations suggest that the composition of the electronic states at the conduction band minimum depends on the cation ordering on the A-site. However, understanding whether the observed preference for the reduction of Bi is related to a specific cation ordering would require more systematic studies on this topic. The reduction potential of Bi in NBT–BT is within the experimental uncertainty identical to that in BiFeO₃, indicating that the reduction potential can be transferred from one compound to another.

A quantitative consideration of the electrode work function, at which NBT–BT would be reduced, suggests that the electrochemical reduction will not occur at the electrode interface for typical electrode materials used in co-sintering multilayer ceramic/electrode stacks. Nevertheless, it would be desirable to perform such in situ XPS experiments not only with oxide but also with metal electrodes. Such studies could provide more insights into resistance degradation and help to further discern between the effects of electrostatic and chemical boundary conditions and also to evaluate the lower limits of the Fermi level reached the anode side of the cell. This is particularly relevant for acceptor-doped NBT–BT, in which the oxygen ion conductivity is so high, that migration of oxygen vacancies occurs already at very low electric fields.

The supplementary material contains additional information and illustrations of the experimental setup, additional x-ray photoelectron survey and core-level spectra, and sample currents/conductivities recorded during the XPS measurements.

The presented work has been conducted within the collaborative research center FLAIR (Fermi level engineering applied to oxide electroceramics), which is supported by the German Research Foundation (DFG) (Project-ID No. 463184206–SFB 1548) and the Fonds zur Förderung der wissenschaftlichen Forschung (FWF, Austrian Science Fund), Project No I 6450-N. Pengcheng Hu also acknowledges support from the China Scholarship Council (CSC) (Award No. 202106220039).

The authors have no conflicts to disclose.

Pengcheng Hu: Data curation (lead); Investigation (lead); Visualization (lead); Writing – original draft (lead); Writing – review & editing (equal). Binxiang Huang: Methodology (equal); Supervision (supporting); Writing – review & editing (equal). Daniel Bremecker: Investigation (equal); Validation (supporting); Writing – review & editing (supporting). Jurij Koruza: Funding acquisition (equal); Validation (equal); Writing – review & editing (equal). Karsten Albe: Funding acquisition (equal); Methodology (equal); Software (lead); Validation (equal); Visualization (supporting); Writing – review & editing (supporting). Andreas Klein: Conceptualization (lead); Funding acquisition (equal); Resources (lead); Supervision (lead); Validation (equal); Visualization (equal); Writing – review & editing (equal).

The data that support the findings of this study are available from the corresponding author upon reasonable request.