Controlling mixed conductivity in Na 1/2 Bi 1/2 TiO 3 using A-site non-stoichiometry and Nb-donor doping of Materials Chemistry C

a Precise control of electronic and/or ionic conductivity in electroceramics is crucial to achieve the desired functional properties as well as to improve manufacturing practices. We recently reported the conventional piezoelectric material Na 1/2 Bi 1/2 TiO 3 (NBT) can be tuned into a novel oxide-ion conductor with an oxide-ion transport number ( t ion ) 4 0.9 by creating bismuth and oxygen vacancies. A small Bi-excess in the nominal starting composition (Na 0.50 Bi 0.50+ x TiO 3+3 x /2 , x = 0.01) or Nb-donor doping (Na 0.50 Bi 0.50 Ti 1 (cid:2) y Nb y O 3+ y /2 , 0.005 r y r 0.030) can reduce significantly the electrical conductivity to create dielectric behaviour by filling oxygen vacancies and suppressing oxide ion conduction ( t ion r 0.10). Here we show a further increase in the starting Bi-excess content (0.02 r x r 0.10) reintroduces significant levels of oxide-ion conductivity and increases t ion B 0.4–0.6 to create mixed ionic/electronic behaviour. The switch from insulating to mixed conducting behaviour for x 4 0.01 is linked to the presence of Bi-rich secondary phases and we discuss possible explanations for this effect. Mixed conducting behaviour with t ion B 0.5–0.6 can also be achieved with lower levels of Nb-doping ( y B 0.003) due to incomplete filling of oxygen vacancies without the presence of secondary phases. NBT can now be compositionally tailored to exhibit three types of electrical behaviour; Type I (oxide-ion conductor); Type II (mixed ionic-electronic conductor); Type III (insulator) and these results reveal an approach to fine-tune t ion in NBT from near unity to zero. In addition to developing new oxide-ion and now mixed ionic/electronic NBT-based conductors, this flexibility in control of oxygen vacancies allows fine-tuning of both the dielectric/piezoelectric properties and design manufacturing practices for NBT-based multilayer piezoelectric devices. NBT phase analysed cation EDX the ratio of A- to B-site elements 1:1. TEM grinding Ar-beam shows that both A-site non-stoichiometry via excess Bi 2 O and B-site donor-doping in fine-tuning in NBT. for applications r x r 0.10) induces mixed ionic/electronic behaviour by reintroducing higher oxide-ion conductivity with t ion B 0.4–0.6. For Nb-doping, mixed conduction is observed at low doping levels ( y r 0.003). In contrast to increasing the Bi-excess content, increasing the Nb-donor doping level to y = 0.050, 0.100 systematically suppresses oxide-ion conductivity with t ion r 0.05. The results show NBT compositions can be tuned smoothly to exhibit t ion from near unity to zero. This study provides guidelines on fine-control of oxygen vacancies in NBT-based materials to tune electrical, dielectric and piezoelectric properties and to improve manufacturing practices.


Introduction
It has been long known that low levels of non-stoichiometry in functional oxides, either associated with intentional element doping or unintentional element loss/gain during processing or impurities in raw materials can induce significant changes in functional properties. [1][2][3] Understanding the defect chemistry and controlling non-stoichiometry in functional oxides is a powerful and proven approach to improve the desired electronic, ionic, mixed ionic/electronic and/or polar dielectric properties of existing materials, design new materials and optimise manufacturing practices for various modern electronic devices. In particular, the presence of low levels of oxygen vacancies can play an important role in the functional properties.
BaTiO 3 -based MLCCs with base-metal electrodes require sintering at low oxygen partial pressure (pO 2 ) to avoid oxidation of metal electrodes. 4 Such sintering conditions will lead to titanium reduction, generating undesired n-type electronic conduction. Acceptor doping with the creation of oxygen vacancies is useful to limit reduction of titanate-based materials under the reducing processing conditions. On the other hand, the level of oxygen vacancies has to be minimised to control the insulation resistance and tan d at elevated temperatures. Elaborate elemental doping schemes in BaTiO 3 are required to achieve balanced oxygen vacancy concentration, which enables manufacturing of MLCCs using base-metal electrodes at low pO 2 without titanium reduction while maintaining device reliability and lifetime.
A-site non-stoichiometry linked to oxygen vacancies is not an issue in titanate-based perovskite dielectrics such as (Ba,Sr)-TiO 3 that contain only alkaline earth A-site cations but is a known issue in others such as PZT due to PbO-loss. In the processing of ceramics that contain volatile A-site species, this is often circumvented by either; (i) adding an excess of the volatile species to the starting composition; or (ii) donor-doping on the A-or B-site, e.g. La-doping for Pb in PZT. [4][5][6][7][8][9][10] In both cases, the additive and/or dopant concentration level(s) need to be well balanced to control the oxygen vacancy concentration required for the creation of pinning defects that influence domain wall motion. 10,11 This can be employed to manufacture 'hard' and 'soft' piezoelectric ceramics to tune functional properties (e.g., piezoelectric constant, coercive field, depolarisation temperature, quality factor, relative permittivity and dielectric loss) to suit different applications. 10 The ferroelectric perovskite Na 1/2 Bi 1/2 TiO 3 (NBT) is known to exhibit interesting and diverse structure-composition-property relationships. [12][13][14][15][16][17][18][19][20][21] A combination of A-site (Na, Bi) disorder, Bi-and Ti-displacements and in-phase and out-of-phase octahedral tilting ensures the crystal chemistry and polymorphism of NBT remain a challenging subject, especially below ca. 520 1C where distortions from the ideal 'cubic' cell are reported to occur. 13,14,[22][23][24][25][26] The electrical conductivity of undoped NBT materials are known to be critically dependent on low levels of A-site non-stoichiometry, commonly AE0.02 from the nominally stoichiometric 0.50Na : 0.50Bi ratio. 5,27 Hiruma et al. were the first to show the room temperature dc resistivity of NBT ceramics based on 0.01 Bi-excess (e.g. 0.50Na : 0.51Bi) or 0.01 Na-deficiency (e.g. 0.49Na : 0.50Bi) to exceed those based on 0.01 Na-excess or 0.01 Bi-deficiency by up to three orders of magnitude. 28 Furthermore, it is widely reported that undoped and nominally A-site stoichiometric NBT ceramics can be leaky with tan d 4 0.10 at 4300 1C whereas others report excellent dielectric behaviour with tan d o 0.02 at B600 1C. [29][30][31] Such significant variations in electrical behaviour from nominally similar starting compositions demonstrate the importance of controlling the loss of any Na and/or Bi-oxides as volatile species during the processing of NBT-based materials. Detecting these small variations in A-site stoichiometry is challenging for conventional analytical methods (see ESI in ref. 27 for more details) and electrical property measurements are sensitive tools to probe the compositionproperty relationships of NBT-based materials.
We have recently used a combination of Impedance Spectroscopy (IS), O 18 Time-of-Flight Secondary Ion Mass Spectrometry (ToF SIMS) and oxygen concentration cell (electromotive force, EMF) measurements to show nominally Na-rich (e.g. Na 0.51 Bi 0.50 TiO 3.005 ) or Bi-deficient NBT (e.g. Na 0.50 Bi 0.49 TiO 2.985 ) ceramics and nominally stoichiometric NBT to be excellent oxide-ion conductors with oxide-ion transfer numbers (t ion ) exceeding B0.9 at 600 1C whereas nominally Na-deficient (e.g. Na 0.49 Bi 0.50 TiO 2.995 ) or Bi-excess NBT (e.g. Na 0.50 Bi 0.51 TiO 3.015 ) ceramics are electrically insulating but retain t ion B 0.1 at 600 1C. 5,27 This showed undoped NBT ceramics to be mixed ion-electron conducting materials where the magnitude of the bulk conductivity, s b , and t ion are heavily dependent on the nominal Na : Bi ratio in the starting composition. We attribute the source of oxygen vacancies in nominally stoichiometric NBT to be associated with low levels of Bi 2 O 3 -loss during ceramic processing. The conduction properties of nominally A-site non-stoichiometric compositions rely on a deficiency of Bi 2 O 3 to promote bulk oxide-ion conduction or an excess of Bi 2 O 3 to compensate for Bi 2 O 3 -loss during processing to ensure bulk oxide-ion conduction is suppressed and dielectric behaviour is obtained.
Chemical doping with Mg for Ti (acceptor-type, i.e. Ti 4+ + O 2À -Mg 2+ ) can significantly increase the oxide-ion conductivity in NBT with t ion approaching unity to promote solid electrolyte behaviour whereas Nb-doping for Ti (i.e. donor-type, Ti 4+ -Nb 5+ + 1/2O 2À ) can suppress the oxygen vacancy concentration and oxide-ion conductivity such that excellent dielectric behaviour can be obtained with tan d o 0.02 at 600 1C. 9,27 Such behaviour is desirable for NBT-based applications, e.g. as Pb-free piezoelectrics and/or as a solid solution member in temperature stable, high permittivity MLCCs operating at 4175 1C.
Here we report the electrical properties of NBT-based ceramics with a wider range of nominal Bi nonstoichiometry (Na 0.50 Bi 0.50+x TiO 3+3x/2 , 0.00 r x r 0.10) and Nb-doping (Na 0.50 Bi 0.50 Ti 1Ày Nb y O 3+y/2 , 0.000 r y r 0.100). The solid solutions limits are low with x o 0.01 and y o 0.030 but dramatic changes in the bulk conduction mechanisms are reported. We show an apparently peculiar behaviour that, as compared to the resistive Bi-excess composition (x = 0.01) reported recently, a further increase of Bi content (e.g., x Z 0.02) reintroduces significant levels of oxide-ion conductivity and increases t ion to intermediate levels of B0.4-0.6 to create mixed ionic/ electronic behaviour. This mixed conduction behaviour with t ion B 0.5 is also observed in single-phase samples with low levels of Nb-doping (y = 0.002, 0.003). The results reveal, in addition to conducting NBT compositions with t ion close to unity and insulating NBT compositions with t ion close to zero shown in our previous reports, NBT compositions can now be fine-tuned to give intermediate t ion values of B0.4 to 0.6. This has important implications for the compositional control of NBT-based materials for various dielectric, piezoelectric, oxideion and now mixed ionic/electronic conducting applications.

Experimental
Ceramics of Na 0.50 Bi 0.50+x TiO 3+3x/2 (for x = 0.00, 0.01, 0.02, 0.05 and 0.10) and Na 0.50 Bi 0.50 Ti 1Ày Nb y O 3+y/2 (for y = 0.002, 0.003, 0.030, 0.050 and 0.100) were prepared by the conventional solid state route. Na 2 CO 3 (99.5%, Sigma-Aldrich), Bi 2 O 3 (99.9%, Acros Organics), TiO 2 (99.9%, Sigma-Aldrich) and Nb 2 O 5 powders (99.9%, Sigma-Aldrich) were used as raw materials. The powders were pre-dried (300 or 800 1C for 2 h, as appropriate) before weighing. The mixtures of raw materials were ball milled in isopropanol using yttria-stabilized zirconia milling media for 6 h, then dried at 80 1C, sieved, and calcined at 800 1C for 2 h in air. The resultant powders were then ball milled for 4 h followed by drying, sieving and a 2 h calcination at 850 1C and another 6 h ball milling. Green bodies were first formed by a uniaxial steel die and then isostatically pressed at 200 MPa. Prior to sintering at 1050-1150 1C for 2 h, pellets were covered in calcined powder of the same composition to minimise volatilisation.
The crystalline phases were determined by X-ray powder diffraction (XRD) analysis (high-resolution STOE STADI-P diffractometer, STOE & Cie GmbH, Darmstadt, Germany) using CuK a1 radiation on sintered and crushed samples. A combination of scanning electron microscopy (SEM, JEOL 6400, JEOL Ltd, Tokyo, Japan), energy dispersive X-ray spectroscopy (EDX, Oxford Link ISIS, Oxford Instruments Ltd, Oxfordshire, UK), and transmission electron microscopy (TEM, JEOL 2010F STEM/TEM) were employed to examine the phase purity and microstructure. Ceramic samples for SEM were polished or polished and thermally etched at 90% of the sintering temperature for 1 h before being coated with carbon. EDX analysis was performed with an average of ten points being collected for each identified phase. An accelerating voltage of 20 kV and an effective spot size of B2 to 3 mm was used in this procedure. This spot size was generally comparable or slightly larger than the size of secondary phase(s) and therefore EDX data from secondary phases may contain a contribution from the matrix phase. The chemical composition of the NBT perovskite phase was estimated from the analysed relative cation at% from the EDX data with the ratio of A-to B-site elements fixed as 1 : 1. TEM specimens were prepared by mechanical grinding followed by Ar-beam ion-milling.
Impedance spectroscopy was performed using a combination of an Agilent E4980A, a Solartron 1260 system (Solartron Analytical, United Kingdom) and a Solartron Modulab. Au paste (fired at 800 1C for 2 h) or Pt paste (fired at 900 1C for 2 h) electrodes were coated on both major faces of the pellets. IS data were corrected for sample geometry (thickness/area) and, additionally, high frequency data were corrected for instrumentalrelated inductance by performing a short circuit measurement to assess the lead resistance and associated inductance of our set-up.
Oxygen ion transport number measurements were performed on a ProboStat system (NorECs Norwegian Electro Ceramics AS, Fig. 1 Room temperature X-ray powder diffraction data for (a) Na 0.50 Bi 0.50+x -TiO 3+3x/2 and (b) Na 0.50 Bi 0.50 Ti 1Ày Nb y O 3+y/2 series of crushed ceramics. Oslo, Norway). Dense ceramics were coated with Pt paste (fired at 900 1C for 2 h) electrodes on both major faces and sealed to one end of a YSZ tube by a commercial glass frit (KeraGlass ST K02). The voltage between the inner and outer Pt electrodes on the YSZ tube was measured to monitor the pO 2 gradient created by flowing N 2 inside the tube while keep the outside of tube in air. The generated voltage was measured by a Keithley 182 sensitive digital voltmeter (Keithley, Ohio, USA). More details can be found in Fig. S1 (ESI) in ref. 27.

Results
Laboratory XRD on crushed powders of Na 0.50 Bi 0.50+x TiO 3+3x/2 sintered ceramics showed x = 0.00 to 0.02 to be phase pure whereas additional reflections were clearly observed for x = 0.05 and 0.10, Fig. 1(a). Na 0.5 Bi 4.5 Ti 4 O 15 (A 5 B 4 O 15 phase) was identified by XRD as one of the secondary phases (filled triangular symbols in Fig. 1(a)) in x = 0.1 and was also confirmed in x = 0.02 and 0.05 by a combination of SEM/EDX, Fig. 2(a), (c) and Table 1. XRD data associated with the remaining additional reflections (open triangular symbols in Fig. 1(a)) were consistent with a tetragonal (b) Bi 2 O 3 (space group P% 42 1 c). However, it was not easy to detect by SEM/EDX due to its small size, as shown by the small white spots in Fig. 2(b) and (d) for x = 0.02 and 0.05, respectively. The presence of Bi 2 O 3 -rich pockets of secondary phase at triple point junctions between grains and along selected grain boundaries was detected in x = 0.01 by high angle annular dark field scanning transmission electron microscopy (HAADF-STEM), as shown by the bright regions in Fig. 2(g) and the elemental mapping of such a region in Fig. 3. TEM of x = 0.10 revealed a higher volume fraction of the Bi 2 O 3 phase but again it was localised at triple point junctions and in the grain boundary regions, as shown by the bright regions in Fig. 2(h). EDX analysis performed on a sample of x = 0.10 using an electron probe of 1 nm did not detect the presence of any significant level of Na and/or Ti in this phase (not shown).
XRD and SEM/EDX data were consistent for Na 0.50 Bi 0.50 -Ti 1Ày Nb y O 3+y/2 ceramics for 0.000 r y o 0.030 and showed them to be single-phase. Although XRD data suggested y = 0.030 to be phase-pure, SEM/EDX revealed the presence of an A 5 B 4 O 15 secondary phase, Fig. 2 Fig. 1(b). SEM/EDX on y = 0.050 revealed the A 2 B 2 O 7 pyrochlore-type phase to contain Na and Bi on the A-sublattice and Ti and Nb on the B-sublattice with an average composition of Na 0.34 Bi 1.59 Ti 1.63 Nb 0.37 O 6.74 and the average matrix composition for the perovskite phase to be Na 0.53 Bi 0.47 Ti 0.95 Nb 0.05 O 3.00 , Table 1 and Fig. 2(f). The average grain size in NBT is B5 mm and is substantially reduced with increasing x or y, Fig. 4.
Complex impedance plane (Z*) plots for the x-and y-series of ceramics at B650 1C are shown in Fig. 5. With the exception of Table 1 Chemical composition (relative cation at%) and associated error (standard deviation) by SEM/EDX of the main and secondary phases present in x = 0.02, x = 0.05, y = 0.030 and y = 0.050 ceramics. Each value is obtained from an average of 10 analysed points. A summary of phases observed by XRD data is included for completeness. * Finely dispersed Bi 2 O 3 secondary phase was detected by EDX element mapping of HAAD-STEM (Z-contrast) images, see Fig. 3 and text for details. Some totals are slightly greater than 100.0% due to rounding of values  x = 0.00, all Z* plots were dominated by a single, large arc with the presence of a small, low frequency (o1 Hz) electrode-type response. The large and dominant arc could be modelled, to a first approximation, on a single parallel resistor-capacitor (RC) element. The associated capacitance value of the arc was consistent with a bulk (grain) response, e.g. C b B 60-80 pF cm À1 which corresponds to a relative permittivity (e r ) value of B600 to 800 for paraelectric NBT at this high temperature. The Z* response for x = 0.00 contained bulk, grain boundary and electrode contributions and has been reported previously. 27  Here we show only the high frequency Z* response of x = 0.00 at 650 1C, inset to Fig. 5(a), to indicate R of the bulk response (R b ) to be exceptionally low, B360 O cm as compared to all other ceramics where R b was a factor of B50 to 500 higher, Fig. 5. In the x-series, R b increases from B360 O cm for x = 0.00 to B100 kO cm for x = 0.01 before decreasing to an intermediate value of B10-30 kO cm for x Z 0.02, Fig. 5(a). The variation of R b with pO 2 for x = 0.05 is consistent with n-type electronic behaviour, Fig. 5(b). For the y-series, R b increases systematically to B15 kO cm for y = 0.002, B55 kO cm for y = 0.003 and a maximum R b of B200 kO cm for y = 0.050. It then decreases to B160 kO cm for y = 0.100 which is similar in magnitude to x = 0.01 and an order of magnitude larger compared to x = 0.05 and y = 0.002 and 0.003, Fig. 5(c).
The low frequency electrode response was most prominent in x = 0.00 (not shown, see ref. 27 for more details) but was also readily observed in x-series, inset Fig. 5(b) and (d), and y r 0.003 ceramics. The presence of a low frequency electrode arc was difficult to resolve for y = 0.050 and 0.100, see inset of  Table 2 together with some previously reported values. 27 The behaviour is subdivided into three categories based on the magnitude of t ion . Type I,

Discussion
The XRD and SEM results show the limits of x and y to be low, with x o 0.01 and y o 0.03, Fig. 1 and 2 but they have a dramatic  effect on s b , tan d and t ion , Fig. 6, 7 and Table 2, respectively. In the case of simple additions of excess Bi 2 O 3 (x-series), nominally stoichiometric NBT (x = 0.00) transforms from

View Article Online
Type I with high s b and t ion (600 1C) B 0.93 to Type III with low s b and t ion (600 1C) B 0.07 for x = 0.01, Fig. 6. This has very little influence on the e r -T profile but transforms NBT into an excellent high temperature dielectric material with tan d o 0.02 in the range T m to B650 1C, Fig. 7. Further additions of Bi 2 O 3 excess, e.g. x Z 0.02, however exhibit Type II behaviour, with a significant increase in s b and t ion (600 1C) 4 0.45, Fig. 5(b), 6 and Table 2, and has a deleterious influence on the high temperature dielectric properties with tan d rising steeply above B500 1C and exceeding 10% at 600 1C, Fig. 7. At this stage it is unclear why starting compositions with x Z 0.02 exhibit mixed conduction and higher conductivity but it must be linked to the presence of secondary phases, both of which are Bi-rich compared to NBT, i.e. Na 1/2 Bi 9/2 Ti 4 O 15 and a Bi 2 O 3 phase. Two possible explanations are; (i) a change in the Bi-content in the NBT main phase, and (ii) a space-charge model. We outline both below.
Firstly, the Bi-rich secondary phases may deplete the NBT main phase of Bi in x 4 0.01, therefore reducing its Bi-content resulting in an increase in oxide-ion conduction. Previous studies have shown that conventional analytical methods such as EDX are unable to distinguish the low level of A-site nonstoichiometry (Na : Bi ratio) required to switch from Type I (x = 0.00) to III (x = 0.01) behaviour in NBT-based ceramics. 27 As a consequence, it is not possible to unequivocally prove a change in stoichiometry associated with the main NBT phase in the ceramics, especially due to the fine scale distribution of the Bi 2 O 3 -type phase in the x-series, Fig. 2(b) and (d). The level of x required to obtain Type III as opposed to Type I or II behaviour will be dependent on the ceramic processing conditions employed as they control the extent of thermodynamic equilibrium within the ceramics and this will influence the chemical potential and therefore concentration (strictly, the activity) of Bi between the three co-existing phases. In our studies, NBT powders are prepared at B850 1C and ceramics are sintered at 1100 to 1150 1C for 2 h in air, covered in sacrificial powder of the same composition. This results in x = 0.00 being Type I and the requirement for a small excess of starting Bi 2 O 3 (x B 0.01) to compensate for losses during processing prior to the formation of Bi-rich secondary phases to obtain Type III behaviour.
x values are likely to be significantly different for processing conditions that don't use sacrificial powders to limit Na, Bi volatilisation and/or for higher/longer sintering periods as these factors will influence the final phase assemblage, i.e. composition and weight percentage of the phases present.
Secondly, enhanced ionic conductivity based on a space charge model has been observed in mixtures of two phases, based either on one ionic conductor and one insulator or mixtures of two ionic conductors. 32 For example, additions of Al 2 O 3 to lithium iodide LiI led to an increase in ionic conductivity by a factor of B50 at room temperature 33 and additions of ZrO 2 nano-powder to AgI increased ionic conductivity by more than three orders of magnitude at room temperature. 34 The interfacial space charge effect is widely considered as a driving force for the enhancement of ionic conductivity. 32 This effect may account for the enhanced oxide-ion conduction and the t ion of B0.4-0.6 in starting compositions with higher Bi-excess (x Z 0.02) where NBT, Bi 2 O 3 -type and Na 0.5 Bi 4.5 Ti 4 O 15 -type phases coexist. The fine scale distribution of the Bi 2 O 3 -type phase with the NBT matrix, Fig. 2(g) and (h), combined with the fact that many Bi 2 O 3 -type materials exhibit high levels of oxide-ion conductivity would appear to be the most likely source of any space charge model in these ceramics. In addition, t ion of these ceramics may be overestimated if the Bi 2 O 3 -type phase present is a good oxide-ion conductor. Further study is needed to clarify the composition and properties of this Bi 2 O 3 -type phase to elucidate its influence on the measured electrical properties of these ceramics and in attempts to try and distinguish between the two possible explanations outlined above.
The trend in behaviour is different for Nb-doping (y-series) compared to excess Bi 2 O 3 . Very low levels of Nb-donor doping, e.g. y = 0.002 and 0.003 transform Type I into II, and Type III behaviour is observed for y = 0.010 to 0.100, albeit the Nb-donor doping solid solution limit is y o 0.030, Fig. 1 and 2. Nb-doping is systematic and effective in 'mopping-up' and compensating for any oxygen deficiency associated with Bi 2 O 3 loss during processing; t ion (600 1C) decreases from 0.93 (undoped) to 0.47 for y = 0.003 and is o0.06 for 0.010 r y r 0.100, Table 2. As a consequence, s b and tan d remain low for y Z 0.010. All ceramics with y Z 0.01 retain Type III behaviour and exhibit excellent, high temperature low dielectric loss behaviour above T m to B650 1C, Fig. 7(b). The e r -T profile for y = 0.020 remains similar to undoped NBT, however there is a significant decrease in e r,max for y = 0.050, Fig. 7(a) indicating Nb-doping for Ti influences the lattice polarisability/ferroelectric domain wall motion. Again, the level of y required to suppress oxide-ion conduction in NBT ceramics to promote, high temperature, low dielectric low behaviour will be dependent on the ceramic processing conditions employed.
This study shows that both A-site non-stoichiometry (via excess Bi 2 O 3 ) and B-site donor-doping (Nb 5+ + 1/2O 2À -Ti 4+ ) can be effective in fine-tuning the oxide-ion conductivity and t ion in NBT. These findings are important for technical applications of NBT-based ceramics. For dielectric applications, some caution is required with the level of excess Bi 2 O 3 in the starting composition as an over excess can lead to a switch from Type III to II behaviour in the presence of Bi-rich secondary phases. Excess Nb-doping therefore offers a more reliable method to retain Type III behaviour to control the level of oxygen vacancies in NBT to ensure low dielectric loss ceramics above T m to B650 1C. The ferroelectric/piezoelectric properties including d 33 , depolarisation temperature and coercive field can be adjusted by tuning the concentration of oxygen vacancies. Further studies are in progress to establish the ferroelectric and piezoelectric properties of these ceramics. In addition, manufacturing of multilayer actuators with base metal electrodes will require a balanced concentration of oxygen vacancy to increase resistance to reduction for titanium under lower pO 2 to avoid n-type semiconductivity as well as to minimise performance degradation associated with oxygen vacancies.
It is worth noting that NBT-based ceramics can be mixed conducting, albeit with a low electronic carrier concentration, e.g. x = 0.02, 0.05, y = 0.002 and 0.003, Fig. 6 and Table 2. Although s b is low, the associated activation energy, E a , for bulk conduction is high, e.g. 40.9 eV and tan d remains low below B500 1C, see Type II in Fig. 7(b), there remains a level of mobile oxygen vacancies that contribute appreciably to s b , as readily detected by EMF measurements, Table 2. An indication of the presence of ionic conduction in the x-series ceramics (where t ion can be as low as 0.07 for Type III) can be obtained from low frequency (o1 Hz) IS measurements at elevated temperatures, as shown in the Z* plots in the insets of Fig. 5(b) and (d) for Types II and III, respectively. In the case of y-series ceramics with Type III behaviour with t ion o 0.05, the presence of ionic conduction was not readily observed by low frequency IS data, see inset of Z* plot for y = 0.050 in Fig. 5(d). This highlights the value of performing EMF measurements in A-site nonstoichiometric perovskites that contain volatile species, such as Na, K, Bi and Pb, where modest levels of oxide-ion conduction (e.g. B5% of s b ) may be present.

Conclusions
We recently showed that the oxygen vacancies in the nominal starting NBT composition associated with Bi 2 O 3 -loss generated during sample processing can be suppressed by a small Bi-excess in the nominal starting composition (Na 0.50 Bi 0.50+x TiO 3+3x/2 , x = 0.01) or by Nb-donor doping (Na 0.50 Bi 0.50 Ti 1Ày Nb y O 3+y/2 , 0.005 r y r 0.030). In this work we show that further increasing the starting Bi-content (0.02 r x r 0.10) induces mixed ionic/electronic behaviour by reintroducing higher oxide-ion conductivity with t ion B 0.4-0.6. For Nb-doping, mixed conduction is observed at low doping levels (y r 0.003). In contrast to increasing the Bi-excess content, increasing the Nb-donor doping level to y = 0.050, 0.100 systematically suppresses oxide-ion conductivity with t ion r 0.05. The results show NBT compositions can be tuned smoothly to exhibit t ion from near unity to zero. This study provides guidelines on fine-control of oxygen vacancies in NBTbased materials to tune electrical, dielectric and piezoelectric properties and to improve manufacturing practices.