Impedance Spectroscopy of Undoped BaTiO, Ceramics Department of Chemistry, University of Aberdeen, Aberdeen, AB 2UE ‘The electrical properties of ceramic BaTiO, were invest gated by ac impedance spectroscopy over the ranges 25° 330°C and 0.03 Hz—1 MHz. Results are compared with those ‘obtained from fixed-frequeney measurements, at 1 kHz and 100 kHz. Fixed-frequency Curie-Weiss plots show devia- tions from linearity at temperatures well above T,. The ac ‘measurements show that grain boundary impedances influ- ence Curie-Weiss plots in two ways: at high temperatures, they increasingly dominate the fixed-frequency permitti ties; at lower temperatures, closer to 7, the high-frequency ittivity contains a contribution from grain boundary effects. Methods for extraction of bulk and grain boundary capacitances from per and electric modulus com- plex plane plots are discussed. The importance of selecting the appropriate equivalent to model the impedance response is stressed. A constriction impedance model for the grain boundary in BaTiO, ceramics is proposed: the grain boundary capacitance is neither temperature-independent, nor shows Curie-Weiss behavior. The grain boundary ferroelectric, similar to the grains, but its impedance is modified by either air gaps or high-impedance electrical inhomogeneity in the region of the necks between grait the activation energy of the constriction grain boundary impedance differs from that of the bulk, suggesting differ- ences in defect states or impurity levels. L ARM TITANATE, BaTiO,, is a ferroelectric material with the perovskite structure. It finds widespread applications as a capacitor material because of its high permittivity, e” ~ 10°10". There is an enormous literature on the properties of BaTiO, and the effect of dopants and microstructure on the properties it is not proposed to review this literature here. In common with most studies on the dielectric properties of clectroceramics, the permittivity of BaTiO, is usually reported at fixed frequency, typically at 1 kHz. While such data give information that is clearly relevant to the practical applications ‘of BaTiO,, a great deal of additional information is lost by not ‘considering the variation of electrical properties with frequency. In the area of ionic conduction in solids, fixed-frequency measurements were commonplace several decades ago, but nowadays variable-frequency methods are used almost exc sively:** Partly this is because it is necessary to avoid the effect ‘on the electrical properties of interfacial polarization at t sample-electrode interface. In addition, however, polycry talline materials may show a variety of frequeney-dependent effects associated with heterogeneities, such as grain bound- aries or surface layers, as well as frequency dependences associ- ated withthe intrinsie properties of the materials concerned. In such cases, fixed-frequency measurements give little informa- tion om the frequency dispersions or their origins, and interpre- tation of fixed-frequency data may be ambiguous. ‘A number of variable frequency measurements on BaTiO, and related materials have been reported, particularly on semicon- Introduction TTI ion —contbating er Mira No, 193282. Resin Sete 7.1998 aproved Agus 2 Mentor Ancian Cems Soy Sports by ten Company La pan, 1633 Naohiro Hirose’ and Anthony R. West’ cotland, UK. ducting doped materials that show the PTCR (positive tempera- ture coefficient of resistance) effect” but also on ferroelectric thin films," pure and doped ferroelectrics," and including ‘measurements at microwave frequencies. From these, itis possible to separate the effects of electroactive components such as grain boundary and grain (= bulk) and to determine the temperature dependences of the relevant resistances and permittivities. Using variable-frequeney measurements, and irrespective of the nature of the material under study, the analysis of ac data requires considerable care in order 10 (a) assigy the different electrical impedances to appropriate regions of the sample and (b) extract the maximum amount of useful information from the data available. To achieve this, a consideration of possible equivalent circuits that can represent the data is necessary, together with the optimization of ways to present data graphi- cally. There are three variables, real and imaginary component of the electrical response and frequency, together with four inter-related electrical response formalisms: impedance, admit- tance, relative permittivity, and electric modulus. Consequently, data can be presented in many formats; in practic, it is often found that different formats may highlight different electrical characteristics of a sample. The purposes of the present work are as follows: (1) Apply the variable-frequency techniques of impedance spectroscopy to dielectric barium titanate ceramies. (2) Compare the results obtained from fixed-frequency and variable-frequency measurements (3). Characterize each electroactive component and investi- gate its possible dependence on microstructural features. (4) Determine the most appropriate equivalent circuit which can model the experimental data and therefore, deter- ‘mine meaningful values for the component resistance and capac- itance parameters L BaTiO, was prepared by conventional solid-state reaction of| BaCO, (99.98%, Aldrich) and THO, (anatase, 999%, Aldrich). ‘The raw materials in a 1:1 mole ratio (Cotal weight ~5 g) were mixed with acetone, dried, and heated in Pt ervcibles, fist at 900°C, 5h to decarbonate. After grinding, the mixture was hneated at 1250°C, 72 h, with daily regrinding, to obtain a homogeneous final product. Phase purity and latice constants of the BaTiO, product were assessed by X-ray powder diffrac tion (Sive SiadiP_ system, linear position sensitive detector, Cukey radiation, 1.54059"), The Sample was tttagonal with parameters a = 3.998, ¢ = 4.038 A, in good azreement with those of JCPDS card No. 5.0626 (a ~ 3.994, ¢ = 4,038 A). For electrical property measurements, 2 disk was pressed uniaxially at 150 MPa, for2 min, with 3 w/v! PVA aqueous solution added as a binder and sintered at 1300°C, for 2h. The isk was then slow cooled, at 1C/min, to 800°C followed by air quenching, The disk was about 0.3 g, 8 rm dia, 1 mm thick. Disk density estimated approximately frem its external dimensions was ~90%. lnvGa electrodes. were rubbed. onto ‘opposite disk faces; the disk was mounted in a conductivity jig, whose leads were shielded separately inside a noninductively ‘wound tube furnace, in air. The detailed jig design is similar to that described in Ref. 27. Impedance measurements were car- ried out over the range 0.03 Hz-~1 MHz, using Solartron Experimental Procedure1634 Journal ofthe American Ceramic Society—Hirose and West 1250/1286 (0.03 Hz-65 kHz) and Hewlett Packard 41924 (100 Hz-I MHz) instrumentation and analyzed using in-house software. Over the range 100 Hz~65 kHz, overlapping data sets from the two instruments were obtained: this provided a useful ‘check on the correct functioning of the measurement arrange- ments, The temperature range was ~25°-330°C. The sample was allowed to equilibrate for 1-2 h at each temperature prior {o measurements; temperatures were accurate to £1°C, Data shown are for the heating eycle only but spot tests showed that, at least up to 220°C, data above the Curie temperature, T., were fully reversible on cooling. Above 220°C, deterioriation of the In/Ga electrode occurred over a period of afew hours and, there= fore, fresh InjGa was applied prior to cach set of impedance ‘measurements, Tt was found unnecessary to correct the experimental data for stray impedances. The open-circuit “parallel” capacitance of the sample holder was only ~5 pF, whereas the corresponding value with the sample present was always much greater, 500 pF. The lead impedances (<1) were always insignifi- cant in comparison with sample impedances and could be ignored. Data collection was limited to frequencies <1 MHz, since it was found that > MHz, inductive effects increasingly influenced the impedance data ‘At the conclusion of the electrical measurements, the pellet used was polished and analyzed by EPMA (electron probe microanalysis, CAMECA, SX31). with the results. given in ‘Table I, The main phase was BaTiO, accurately stoichiometric to within two estimated standard deviations. It was uncontami- nated by any impurities. Diffusion of In/Ga from the electrodes to the interior of the sample did not occur, although slight diffusion into the surface of the sample was possible, since this ‘would have been removed by polishing and would be unde- tected by EPMA. A small amount of a second phase was detected, as small (2-3 jum) isolated particles. This was a Ba-Ti-Si-O phase whose exact nature is not presently known. "The microstructure of a polished and etched section of the same sample was also examined using scanning electron ‘microscopy (ISI-SS40 SEM). The ceramic was rather porous with an approximate grain size of 10-20 jam, Il, Results and Discussion As a first step in data analysis, relative permittivity values, e’, at | kHz were extracted from the impedance data and are shown in traditional fixed-frequeney format in Fig. 1 as a fune- tion of temperature. Classical behavior is seen in which e” increases from ~1500~2000 just below the Curie temperature, passes through a sharp maxirium >6000 at the Curie tempera: ture, Te, 127°C, decays gradually 10 ~1100 by ~220°C, and then staris(o rise again, more gradually, at higher temperatures. ‘The data above Tin Fig. 1 are replotted in Curie—Weiss format in Fig. 2; over the temperature range ~130°~200°C, the plot is essentially linear and extrapolates to an apparent T, value of “114°C, but at higher temperatures, gross departure from lin- arity occurs. In arder to obtain improved Curie-Weiss plots, higher frequency measurements may be used.'” The response at 100 kHz, Fig. 2 is linear up to ~270°C, but gives a slightly ‘Table I. Composition of the Main Phase and a ‘Minor Second Phase Determined with EPMA (Analyzed ‘Sample Is the Same One Used for Electrical Measurement) Thescad Mainphsc(ase) Senda ausey EET Bement Tenge Sudan AvwageSicdew U3) Ba 1983008 1539009 20 T «2008 005103 20 Si 0020011303035 In 000 0.000.000 Ga 001.91 0.009.00 0 6005 003 oss 003 Tat aan ere nomial ge wal fetal caged oer pa) an 98 fart second ae esd ‘ver pois Vol. 79,No.6 + ecco soo - f° 5 seo : & : +4000 5 seo Pe oa pe 200), . [2000 100 : 1000 ° 280 ido 14a 80220280300 Tic Fig. 1. Temperature dependence of capacitance for BaTiO, at 1 kHz higher extrapolated Tq value of ~116°C and still deviates slightly from linearity at higher temperature. “The above analyses, especially those below 200°C, are fairly standard and typical of many reported in the literature. Differ- ‘ences between the values of T, and 7, have been noticed many times; the main reason is often ascribed to the characteristic nature of the first-order transition of BaTiO, as given by the phenomenological theory of Devonshire.” Sometimes, as addi- tional factors for ceramic materials, the differences have been attributed to the presence of grain size, porosity, and grain ‘boundary impedances” in the samples, but itis probably true to say thatthe influence of these additional effects is not clearly defined. Deviations from linear Curie-Weiss plots of e-' vs T fat high temperatures, well above Te, have also been noticed previously and were atributed to the influence of grain bound- ry effects of impedance data!" Tn order to carry out a more thorough analysis of permittivi- ties and their temperature dependence, itis of course necessary to separate the contributions of the various grain and grain ‘boundary components, The importance of making this separa- tion is seen from Fig. 3 in which capacitance data obtained by processing the experimental impedance data, at several temper- tures, are shown as a function of frequency. The data generally show two plateaus separated by a dispersion over a range of intermediate frequencies; with inereasing temperature, the spectrum is displaced toward higher frequencies. If we make the initial assumption that the high-frequency plateau refers to the bulk permittivity (in fact, this is only approximately correct, 10°C" prim Fig. 2. Curie-Weiss behavior with fxed-frequency measurements at Land 100 KHz for BaTiO,June 1996 see below), whereas the low-frequency one! is dominated by the grain boundaries, then the consequences of making fixed- frequency measurements can be clearly seen: at | kHz, arrowed, data at, é.g., 186°C, are on the high-frequency plateau, but at higher temperatures tie in the dispersion region. For a more complete analysis of such permittivity data, We have adopted a general analysis procedure which requires selecting the most appropriate formalism and data preset method for the particular materials under study linked to the evaluation of plausible equivalent circuits for modeling the experimental data, Electrical ac data may be presented in any of four inter- related formalisms:* Impedance: Z* w Admittance: ¥* ‘ Q) Relative permittivity: e* = [jaC,Z*]"' eB) Electric modulus: M* JaCoZ* 4) where j= «/=T, w isthe angular frequency, 2nfand Cy = GAC! (5) in which Cy is the vacuum capacitance of the cell without the sample, € is the permittivity of free space, 8.854 10-" From”, (and A are the thickness and area of the pellet. There is no a priori method for assigning the correct equiva- lent circuit to a sample or data set. Much of the uncertainty associated with data analysis arises because, for a given data set, itis always possible to find >I equivalent circuits that can, in principle, be fitted to the data, The methodology, therefore, requires the consideration of various plausible circuits and then testing them to see which provides the most realistic representa- tion of the system under study. Various tests can be used, such as repeating the measurements on samples with widely different geometry, £0 aid this process. Here, we investigate whether the parameters (R, C) that are extracted show reasonable tempera- ture dependences. Using this approach, we were able to identify ‘an equivalent circuit which gave a reasonable representation of the sample behavior and to exclude other possible circuits as being inappropriate. It is assumed that the impedance data can be represented ideally by an equivalent circuit consisting of two parallel RC ‘elements in series, as shown in the Fig. 4 inset. This was shown to be correct by subsequent analysis; see later. This circuit is used very widely with materials whose properties are some TAC Tow frequencies and high temperatures, deviation from the fow-srequeney pnea sce rota ths due some kind of surface laser formed by the Feattusien af Ga tom the led. No suc ow-requney anomaly wa cbse shen pd cleus were uss over ssn mesuing temperate Tange nl gt a6 10! ad 10 108 wo 10° 10 Frequency / Hz Fig. 3. Frequency dependence of the real part of the complex per- miltvity,e” for BaTiO, at three temperatures. Impedance Spectroscopy of Undoped BaTiO , Ceramics 1635 ‘combination of bulk and grain boundary impedances connected in series. The impedance for this circuitis Ze = (Ri + joy) * + [Ret + FOC] © where the subscripts b and gb refer to bulk and grain boundary. In cases where C, < C,, and R, << Ry (we will show later, in Figs. 7 and 13, that this is correet for the material studied here), the real part of e*, e', calculated from Eqs. (3) and (6), is represented by: En + (RCFE, Ee FUORI ‘The frequency dependence of e” for this circuit has the follow- ing characteristic +A low-frequency plateau (€” when w — 0) given by fw @) +A dispersion centered on a frequency, f,,, which also corre- sponds fo a maximum in e" and for which fy = (2rre(€5! + €GDRCAI™ Rak IC, + CIT! ° + A high-frequency plateau (when « — ) given by fu (10) ‘The high- and low-frequency plateaus in Fig. 3 are labeled accordingly. It is assumed that, in a ferroelectric material above Te, Cy and €, obey the Curie-Weiss law: oy e Cy = Coby = CoAT = 1)" + Cyt. CoAT ~ To) ay where A is the Curie-Weiss constant, and T is temperature. €. is the limiting high-frequency relative permittivity; usually, this value is small and can be ignored for temperatures that are not to0 far above T. At temperatures well above To, however, €. becomes increasingly significant>® we will no: consider this term initially inthe discussion which follows. Itis also assumed that €,, represents a nonferroelectric component and is indepen dent of temperature, although, as shown later, this may not be the case. +The permittivity value of the high-frequency plateau, €, which contains contributions from both e, and ¢,, varies with temperature because of the presence of fy, but tat of the low- frequency plateau, e,, i temperature independent, + Both the high-frequency plateau and the dispersion move to higher frequencies with increasing temperature. This is because GC, decreases with temperature above Te and (ii) Ry is ‘expecied to be thermally activated and decrease with tempera- ture with an Arrhenius dependence of the form Ry = Bexp(ABIkr) «2 Hence, f, in Eq, (9) increases at the same time as the value of «, decreases with increasing temperature. Returning now to fixed-frequency measurements of eit can be seen that, if grain boundary impedances are present, then at no frequencies does the measured permittivity correspond to &,, Even in the high-frequency plateau region, the permittivity &, is modified by the grain boundary capacitance, Fg. (10) ‘The influence of the ever-present grain boundary capacitance ‘on fixed-frequency Curie-Weiss plots may be simulated as shown in Fig. 4. €” is shown for both low f (e.g. 1 kHz) and high f(e.g., 100 kHz), together with e,, dotted, The e,' data are linear and extrapolate to To, by definition (Eq. (11)). The simu- lated e*”' data are linear at low temperatures, because combin- ing Eqs. (10) and (11) gives + eg = ATT) + eg AMD = (Ty ~ Ale] (13) However, this equation shows that the apparent Curie-Weiss temperature is smaller than the true value, Ty, by an amount Al. Undoped BaTiO, has typical values of A'~10°C and eat