Atomistic modeling of Cu doping in the lead-free ferroelectric potassium sodium niobate
Sabine Korbel¨
Dissertation zur Erlangung des Doktorgrades der Fakultat¨ fur¨ Mathematik und Physik der Albert–Ludwigs–Universitat¨ Freiburg im Breisgau Dekan: Prof. Dr. Michael Ru˚ziˇ ckaˇ Erstgutachter: Prof. Dr. Christian Elsasser¨ Zweitgutachterin: Prof. Dr. Elizabeth von Hauff Tag der mundlichen¨ Prufung:¨ 12.12.2012 Atomistic modeling of Cu doping in the lead-free ferroelectric potassium sodium niobate
Dissertation zur Erlangung des Doktorgrades der Fakultat¨ fur¨ Mathematik und Physik der Albert–Ludwigs–Universitat¨ Freiburg im Breisgau
vorgelegt von Sabine Korbel¨ aus Munster¨
Freiburg 2012
Parts of this work have already been published:
• E. Erunal,¨ R.-A. Eichel, S. Korbel,¨ C. Elsasser,¨ J. Acker, H. Kungl, and M, J. Hoffmann. ”Defect structure of copper doped potassium niobate ceramics“. Functional Materials Letters, Vol. 3, pp. 19–24, January 2010. •S.K orbel,¨ P. Marton, and C. Elsasser.¨ “Formation of vacancies and copper substitutionals in potas- sium sodium niobate under various processing conditions“. Phys. Rev. B, Vol. 81, pp. 174115-1– 174115-11, May 2010. •S.K orbel¨ and C. Elsasser.¨ “Cu substitutionals and defect complexes in the lead-free ferroelectric KNN”. In: W. E. Nagel, M. Resch, and D. B. Kroner¨ (eds.), “High Performance Computing in Science and Engineering ’10”. Springer-Verlag Berlin Heidelberg, pp. 181–188, 2011. •S.K orbel¨ and C. Elsasser.¨ “Ab initio and atomistic study of ferroelectricity in copper-doped potas- sium niobate”. Phys. Rev. B, Vol. 84, pp. 014109-1–8, July 2011. • E. Erunal,¨ P. Jakes, S. Korbel,¨ J. Acker, H. Kungl, C. Elsasser,¨ M.J. Hoffmann, and R.-A. Eichel. “CuO-doped NaNbO3 antiferroelectrics: Impact of aliovalent doping and nonstoichiometry on the defect structure and formation of secondary phases”. Phys. Rev. B, Vol. 84, pp. 184113-1–11, November 2011.
• M. J. Hoffmann, H. Kungl, J. Acker, C. Elsasser,¨ S. Korbel,¨ P. Marton, R.-A. Eichel, E. Erunal,¨ and P. Jakes. “Influence of the A/B stoichiometry on defect structure, sintering, and microstructure in undoped and Cu-doped KNN”. In: S. Priya and S. Nahm (eds.), “Lead-Free Piezoelectrics”. Springer-Verlag New York, pp. 209–254, 2012. vi Contents
List of Figures ix
List of Tables xi
List of Acronyms xiii
List of Symbols xv
Summary xvii
1 Introduction 1
2 Computational Methods 7 2.1 Density-functional-theory calculations ...... 7 2.2 Atomistic simulations with shell-model potentials ...... 8 2.3 The Nudged-Elastic-Band Method (NEB method) ...... 8
3 Ab initio thermodynamics of isolated Cu and Fe substitutionals in KNN 9 3.1 Introduction ...... 10 3.2 Thermodynamic formalism ...... 10 3.3 Cu substitutionals and vacancies in KNN 50/50 ...... 14 3.4 The Virtual Crystal Approximation for KNN 50/50 ...... 26 3.5 Effect of the alkali stoichiometry on the substitution site of Cu ...... 27 3.6 Fe substitutionals in KNN 50/50 ...... 29 3.7 Discussion ...... 33
4 Ferroelectricity in KNO with isolated Cu substitutionals 37 4.1 Introduction ...... 38 4.2 Method ...... 38 4.3 Results ...... 40 4.4 Discussion ...... 49
5 Ab initio thermodynamics of defect complexes in KNO and KNN 51 5.1 Introduction ...... 52 5.2 Formation of defect complexes in Cu-doped KNO and KNN ...... 52 5.3 Atom diffusion in KNO ...... 58 5.4 Ferroelectricity in KNO with Cu-defect complexes ...... 61 5.5 Discussion ...... 68
6 Conclusion 71
Bibliography 73
Acknowledgement 81 viii Table of Contents
Curriculum Vitae 83 List of Figures
1.1 The perovskite structure ...... 2 1.2 Phase diagrams of PZT and KNN ...... 3 1.3 Ferroelectric “hardening“ and the “large-strain effect” ...... 5
3.1 Phase stability diagram of KNN ...... 13 3.2 40-atom and 80-atom supercells of KNN ...... 14 3.3 Defect formation energies of Cu in KNN as a function of the Fermi level ...... 17 3.4 Defect formation energies of Cu in KNN as a function of the chemical potentials . . . . . 18 3.5 Defect formation energies of vacancies in KNN as a function of the Fermi level ...... 19 3.6 Defect formation energies of vacancies in KNN as a function of the chemical potentials . . 21 3.7 Density of states of CuK-doped KNN ...... 23 3.8 Density of states of CuNb-doped KNN ...... 24 3.9 Density of states of KNN with vacancies ...... 25 3.10 VCA-DOS of KNN ...... 27 3.11 Effect of the alkali stoichiometry on the most stable site for Cu in KNN ...... 28 3.12 Defect formation energies of Fe substitutionals in KNN ...... 31 3.13 Phase stability diagram of Fe doped KNN ...... 32
4.1 Ferroelectric energy profile in undoped KNbO3 ...... 43 4.2 Ferroelectric energy profile in CuK-doped KNbO3 ...... 43 4.3 Ferroelectric instabilities in CuK-doped KNbO3 as a function of the Cu content ...... 44 4.4 Histogram of atomic displacements in CuK-doped KNbO3 ...... 45 4.5 Atomic displacements and ionic dipole moments in CuK-doped KNbO3 ...... 46 4.6 Ferroelectric energy profile in CuNb-doped KNbO3 ...... 47 4.7 Histogram of atomic displacements in CuNb-doped KNbO3 ...... 47 4.8 Atomic displacements and ionic dipole moments in CuNb-doped KNbO3 ...... 48
5.1 Defect complex configurations ...... 53 5.2 Formation energies of defect complexes in KNO ...... 54 5.3 Electronic DOS of KNO with defect complexes ...... 54 5.4 Binding energies of defect complexes in KNO ...... 55 5.5 Binding energies of defect complexes in KNN ...... 55 5.6 Binding energies of defect complexes in KNN with a bandgap correction ...... 56 5.7 Energy barriers for atom diffusion in KNO ...... 59 5.8 Defect complex configurations in ferroelectric KNO ...... 61 5.9 Energies of defect complexes in orthorhombic KNO ...... 63 5.10 Polarization around defect complexes in orthorhombic KNO ...... 64 5.11 NEB relaxation with the SMP ...... 65 5.12 Energy and polarization along transition paths in KNO:CuNb −VO ...... 66 5.13 Energy and polarization along transition paths in KNO:VO−CuNb −VO ...... 67 5.14 Energy and polarization along transition paths in KNO:VO−CuNb −VO ...... 68 x List of Figures List of Tables
3.1 Chemical potential of O as a function of pressure and temperature ...... 13 3.2 Lattice constants and formation energies of elements and compounds ...... 15 3.3 Atomic chemical potentials for special points in the phase diagram ...... 16 3.4 Displacements of atoms surrounding Cu substitutionals ...... 20 3.5 Displacements of atoms surrounding vacancies ...... 20 3.6 Differences of spin-up and spin-down electrons of Cu substitutionals ...... 22 3.7 Numbers of electrons and holes generated by Cu substitutionals ...... 22 3.8 Numbers of electrons and holes generated by vacancies ...... 26 3.9 VCA for KNN 50/50 ...... 26 3.10 Magnetic spin moments of Fe substitutionals on K and Nb sites in KNN ...... 30
4.1 Ferroelectric displacements in undoped KNbO3 ...... 40 4.2 Ferroelectric instabilities in undoped KNbO3 ...... 41 4.3 Shell-model potential parameters for CuK-doped KNbO3 ...... 41 4.4 Phonon frequencies in undoped KNbO3 ...... 42 4.5 Energies of ferroelectric and rotational modes in CuK-doped KNbO3 ...... 45
5.1 Defect formation energies, binding energies and concentrations in KNO ...... 56 5.2 Defect formation energies, binding energies and concentrations in KNN ...... 57 5.3 Defect formation energies, binding energies and concentrations in KNN after a bandgap correction ...... 57 5.4 Energy barriers and diffusion times of K and O vacancies in undoped and Cu-doped KNO 60 5.5 Shell-model potential parameters for CuNb-doped KNbO3 ...... 62 xii List of Tables List of Acronyms
CBM Conduction-Band Minimum DFT Density Functional Theory DOS Density Of States EPR Electron Paramagnetic Resonance fcc face-centered cubic GGA Generalized-Gradient Approximation GULP General Utility Lattice Program LDA Local Density Approximation
KNN Potassium sodium niobate, (K,Na)NbO3
KNO Potassium niobate, KNbO3 MBPP Mixed-Basis PseudoPotential (program) MD Molecular Dynamics MEP Minimum Energy Path MPB Morphotropic Phase Boundary NEB Nudged Elastic Band (method)
NNO Sodium niobate, NaNbO3
PTO Lead titanate, PbTiO3
PZT Lead zirconate titanate, Pb(Zr,Ti)O3 PE ParaElectric sc simple cubic SC SuperCell SMP Shell-Model Potential TST Transition state theory VBM Valence-Band Maximum VCA Virtual Crystal Approximation xiv List of Acronyms List of Symbols
A 12-fold coordinated cation (A) site in the perovskite ABO3 structure ...... 11 a lattice constant ...... 15 B 6-fold coordinated cation (B) site in the perovskite ABO3 structure ...... 10 B bulk modulus ...... 41 EB binding energy ...... 53 E f defect formation energy ...... 10 0 ∆E f formation energy ...... 15 EF Fermi energy ...... 10 EG band-gap energy ...... 22 G Gibbs free energy ...... 12 0 ∆Hf formation enthalpy ...... 11 kB Boltzmann’s constant ...... 52 P ionic dipole moment ...... 39 P polarization ...... 39 p, p0 pressure ...... 12 q defect charge ...... 10 S entropy ...... 53 S strain ...... 53 T temperature ...... 52 Z ionic charge ...... 40 ∆µ relative chemical potential (of atoms) ...... 11 µ chemical potential ...... 10 × × + 0 Kroger-Vink¨ notation for a neutral defect charge, e.g. CuK ≡(CuK+ ) ...... 16 • • 2+ + Kroger-Vink¨ notation for a positive defect charge, e.g. CuK ≡(CuK+ ) ...... 16 0 0 − Kroger-Vink¨ notation for a negative defect charge, e.g. VK ≡ (VK+ ) ...... 16 xvi List of Symbols Summary
Ferroelectrics are part of many piezoelectric applications, such as actuators and sensors. A commonly used ferroelectric material is lead zirconate titanate (PZT), but efforts are made to develop or improve lead- free alternatives. A possible way to optimize material properties is doping, for example with transition metals. One candidate for future lead-free ferroelectrics is potassium sodium niobate ((K,Na)NbO3, KNN). According to the literature, Cu doping in KNN improves the density of the sintered KNN ceramics. In this work, Cu and for comparison also Fe doping in KNN were investigated on the atomic and electronic level in order to obtain a better understanding by which microscopic mechanisms the dopants influence the macroscopic ferroelectric properties of KNN. In this work, two possible doping mechanisms were investigated. Some dopants can cause a polymor- phic or a morphotropic phase transition (a phase transitions with temperature or composition), or they can shift them to a desired position. A phase boundary between two compositions is called a morphotropic phase boundary (MPB). Such an MPB exists, for example, in PZT and in Li-doped KNbO3. Another doping effect in ferroelectrics is ferroelectric “hardening”. In “hard” ferroelectrics the fer- roelectric domain walls are less mobile than in “soft” ferroelectrics, which is sometimes desired because in this case the hysteresis is small and the electromechanical strain depends almost linearly on an applied voltage. The domain wall mobility is related to the strain response of the ferroelectric to an applied voltage. Defect complexes consisting of dopants and lattice vacancies possibly contribute to ferroelectric “harden- ing” by impeding domain wall motion. In this work, density-functional theory (DFT) in the local-density approximation (LDA) and molecular dynamics (MD) simulations with classical interatomic potentials were combined to search for MPB’s in Cu-doped KNN and to clarify if and how defect complexes consisting of Cu dopants and oxygen vacancies increase the energy needed for ferroelectric domain wall motion. We found a MPB in the CuNbO3-KNbO3 system at a few mol% Cu, similar to the MPB in Li-doped KNbO3, and a hardening effect of the defect complexes consisting of Cu substitutionals and oxygen vacan- cies. xviii Summary 1 Introduction
Motivation and background Motivated by the vision of an environment-friendly industry, in 2003 the European Union adopted the “Directive on the restriction of the use of certain hazardous substances in electrical and electronic equipment” (RoHS directive), which restricts the use of six elements and com- pounds, including lead [EU 03]. Piezoelectric devices containing lead are so far excluded from the RoHS regulation [EU 03] because the lead-free alternatives cannot yet compete in performance with the currently best material lead zirconate titanate (PZT). One possible substitute for PZT is the lead-free piezoelectric material potassium sodium niobate ((K,Na)NbO3, KNN) [Sait 04]. This is investigated in this work.
Piezoelectrics If a material responds to mechanical stress with a voltage, this phenomenon is called piezoelectricity. It results from changes in dipole moments in the crystal due to stress, e.g. when sublat- tices of positive and negative ions are shifted relative to each other. The occurrence of strain in response to an electric field is called the converse piezoelectric effect. Hence, piezoelectric materials can trans- form electrical into mechanical energy and vice versa. A well-known piezoelectric material is quartz [Curi 82]. There is a wide range of technological applications for piezoelectrics, such as displacement actuators [Damj 05], force generators, ultrasonic devices, microphones, underwater sound sources, ac- celerometers, and electric wave filters [Scot 89, Jaff 58]. Except for quartz, most of the piezoelectric mate- rials used for piezoelectric applications are ferroelectrics [Damj 05]. Ferroelectrics belong to the family of piezoelectric materials, but not every piezoelectric is ferroelectric. Ferroelectrics were named in analogy with ferromagnets [Jaff 58]. Like ferromagnets exhibit a spontaneous magnetic polarization, ferroelectrics exhibit an electric one (ferroelectricity is discussed more extensively in the next paragraph). Some ferro- electrics, especially PZT, have much higher piezoelectric constants with respect to other, non-ferroelectric piezoelectrics [Damj 05, Jaff 58]. This is why they play such an important role among the piezoelectrics. In addition to the piezoelectric applications mentioned above, ferroelectrics are used in, e.g., ferroelec- tric capacitors, in particular ferroelectric DRAM’s (FRAM’s) [Scot 89, Wiki 11]. Among the industrially relevant ferroelectrics the ones with a perovskite structure are the most common ones [Damj 05]. The prominent piezoelectric PZT is one of them, and it is a strong ferroelectric [Jaff 58]. In order to find a suitable substitute for PZT, it is obvious to begin the search among compounds with the same or a similar crystal structure. KNN is such a compound. Like PZT, it is ferroelectric and crystallizes in the perovskite structure.
Ferroelectric perovskites Perovskite is the name of the mineral calcium titanate (CaTiO3). Several other oxides with the formula ABO3 (also some non-oxides) also crystallize in the perovskite structure, which is depicted in Fig. 1.1. Cations (A and B ions) occupy the cube corners (A sites) and the cube centers (B sites), oxygen anions sit on the centers of the cube faces. There are three oxygen anions per unit cell, each with a formal charge of −2. The formal charges of the cations add up to +6 so that the unit cell is altogether neutral. Typically the formal charge of the cation at the cube corner is +2 (e.g. in BaTiO3, SrTiO3, PbTiO3, PbZrO3), +1 (KNbO3, KTaO3, NaNbO3), or +3 (e.g. in LaAlO3). While some perovskites remain unpolar also at low temperatures (SrTiO3, KTaO3), others undergo one or several transitions to ferroelectric or antiferroelectric phases (BaTiO3, PbTiO3, PbZrO3, KNbO3, NaNbO3). 2 1. Introduction
A A O B B x O ↑ P ↓
↑
Figure 1.1: The cubic, paraelectric unit cell of an oxide with perovskite structure (left), and a tetragonally distorted unit cell in which the negative O ions and the positive B site ions are shifted in opposite directions, resulting in a spontaneous ferroelectric polarization P (right). The atomic displacements and unit cell distortion are those of tetragonal KNbO3 [Hewa 73b].
Above the Curie temperature, the unit cell of an oxide with perovskite structure is cubic and paraelec- tric. Below the Curie temperature of a ferroelectric perovskite, the positive and negative ions are displaced from their cubic positions into different directions, so that the unit cell is spontaneously electrically polar- ized. The unit cell is elongated along the direction of the ferroelectric polarization, which is usually either a h001i, h011i, or h111i direction. The magnitudes of both the ionic displacements and the elongation are typically of the order of one or few percent of the lattice constant, which is about 4 A.˚ Some perovskite oxides, e.g. NaNbO3, have an antiferroelectric ground state. Instead of or in addition to the ferroelectric or antiferroelectric displacements of the ions, the oxygen octahedra can be tilted [Glaz 72], e.g. in NaNbO3 and KxNa1−xNbO3 up to x ≈ 0.5 [Ahte 78] or in SrTiO3 [Glaz 72]. The origin of ferroelectricity is mainly ascribed to hybridization between electronic states of the B and the O atoms [Post 94, Migo 76, Cohe 92]. The cubic, paraelectric crystal can lower its energy by shifting the atoms off center and reducing the distance between B and O atoms. In some cases, like PbTiO3, also hybridization between the A and the O states contributes to ferroelectricity [Cohe 92].
Lead-free ferroelectrics Environment-friendly lead-free ferroelectrics are currently investigated in order to substitute lead zirconate titanate [Pb(Zi,Ti)O3, PZT]. There are two main candidates for possible lead-free substitutes for PZT: Bismuth-based perovskite oxides ((Bi,Na,K)TiO3 − BaTiO3, BNKT [Take 05]) and alkali niobates ((K,Na)NbO3, KNN [Sait 04]). Both BNKT and KNN crystallize in the perovskite structure like PZT. In a specially textured KNN sample with Li, Sb, and Ta additives [Sait 04], piezoelectric characteristics similar to those of PZT were measured. In this work, only KNN is investigated. The most commonly used ferroelectric, PZT, is a mixture of approximately 50% PbTiO3 and 50% PbZrO3. At this composition, PZT exhibits enhanced piezoelectric properties due to a morphotropic phase boundary [Zhan 07]. At a morphotropic phase boundary (MPB), a phase transition with composition takes place. Near a composition of K:Na=50:50 a morphotropic phase boundary exists also in KNN. Figure 1.2 shows the phase diagrams of PZT and KNN, including the MPB’s around PZT 50/50 and KNN 50/50 (K0.5Na0.5NbO3). Below the Curie temperature of PZT, which lies roughly between 600 and 700 K, PZT is ferroelectric. At the MPB in PZT with a Zr:Ti ratio of about 50:50 (PZT 50/50), a transition between a tetragonal phase (polarization in a h001i direction) and a rhombohedral phase (polarization in a h111i direction) occurs. KNN is ferroelectric below about 400° C. In this work the composition region between 0 and 50 mol% Na is investigated. In this composition range, the order of the ferroelectric phases is the same as in the end member KNbO3, whose cystal structure changes upon cooling from cubic and paraelectric to tetragonal and ferroelectric with a polarization along a h001i direction at about 435° C, to orthorhombic with a polarization along a h011i direction at about 225° C, and to rhombohedral with a polarization in a h111i direction at about −10° C (this is analoguous to BaTiO3) [Hewa 73a, Shir 54a]. KNN undergoes a transition between two different orthorhombic structures at at K:Na ratio of about 50:50 (KNN 50/50), one with and one without tilted oxygen octahedra [Ahte 76]. In a more recent study, a 3 coexisting monoclinic phase was found at this composition [Bake 09]. Near the MPB the two structures are approximately degenerated in energy, which facilitates structural transformations and therefore the reorientation of the ferroelectric polarization [Zhan 07]. Other than polymorphic phase transitions (phase transition with temperature), which also facilitate switching, phase transitions at MPB’s allow operation at various temperatures, if the MPB depends only weakly on the temperature. For both PZT and KNN this is the case. Possibly the MPB in KNN 50/50 leads to enhanced ferroelectric properties like the one in PZT 50/50, but according to [Zhan 07] the enhanced piezoelectric coefficients measured in KNN at this composition are caused by the polymorphic phase transition (phase transition with temperature).
PC
FT
FO
MPB FR
0 . 0 5
Figure 1.2: Phase diagrams of PZT (left, from [Nohe 00]) and KNN (right, from [Bake 09]). In the phase diagram of PZT, PC is the paraelectric, cubic phase, and FT,FM,FO, and FR are the cubic, tetragonal, monoclinic, orthorhombic, and rhombohedral ferroelectric phases. In the phase diagram of KNN the superscripts denote the sense of oxygen tilting (cf. section 4.3 for details) and the subscripts the cation displacements.
Tailoring ferroelectric properties by doping Analoguous to the magnetization of ferromagnets, the po- larization in ferroelectrics as a function of the external electric field follows a hysteresis loop. Whereas in ferroelectric memory applications a strong hysteresis (a large remanent ferroelectric polarization and a large coercive field, hence a long-time stable ferroelectric memory effect) is desirable, one wants the displacement of a piezoelectric actuator to increase approximately linearly with the applied voltage. The amount of hysteresis can be tailored to a large extent by doping a standard ferroelectric material like PZT or KNN. Typically, in analogy to ferromagnets, ferroelectrics are devided into “hard” and “soft” ferroelectrics. “Soft” ferroelectrics exhibit a high compliance, permittivity and piezoelectric coefficients, accompanied by strong hysteresis and nonlinearity, “hard” materials have low values of the piezoelectric coefficients but a reduced hysteresis, nonlinearity, conductivity and frequency dispersion [Damj 05], and they have smaller losses [Dibb 95]. “Hard” and “soft” ferroelectrics can be obtained from the same base material, e.g. PZT, by acceptor doping (lower valent ions, e.g. Fe3+ on a Ti4+ site) or by donor doping (higher valent ions, e.g. Nb5+ on a Ti4+ site), respectively [Damj 05]. While the origin of “softening“ is still unclear [Damj 05], one possible ”hardening“ mechanism is the interaction of defect complexes with ferroelectric domain walls [Damj 05]. Ferroelectric domains are regions in which the ferroelectric polarization has the same orienta- tion in each unit cell. Normally a single crystal or a grain of a polycrystal of a ferroelectric material consists of many domains with different orientations of the polarization [Dibb 95]. These differently oriented do- mains form in order to lower the electrostatic and elastic energy of the system [Damj 05]. The domains are separated by domain walls. In a tetragonal ferroelectric, the domain walls can separate domains with polar- ization directions that differ by 90° (90° domain walls) or by 180° (180° domain walls). In orthorhombic and rhombohedral phases, additional angles between the polarization directions of adjacent domains are possible. A newly fabricated ceramic is macroscopically isotropic because there are many domains with different polarization directions. Piezoelectricity is achieved by ”poling“, i.e. orienting the ferroelectric domains 4 1. Introduction along one direction in an electric field on the order of 1 kV/mm, usually at elevated temperatures (e.g. 150°C) [Dibb 95]. The remanent polarization after ”poling“ is the working point of the ceramic [Dibb 95]. For small electric fields, the strain in the poled ceramic is approximately proportional to the electric field [Dibb 95].
The motion of ferroelectric domain walls through the crystal or through a grain usually is the domi- nant switching mechanism in a ferroelectric, although in thin films or under irradiation with strong electric pulses [Qi 09] homogeneous switching (the polarization in a whole domain changes its direction instan- taneously) may occur. According to [Damj 05], it is ”a reasonable assumption in most cases encountered in practice“ that ”domain-wall displacement is the main source of the dielectric, elastic and piezoelec- tric hystereses“. Figure 1.3(a) shows schematically the motion of a 90° domain wall inside a tetragonal perovskite in an electric field. One possible ”hardening“ mechanism is the pinning of domain walls by electric dipoles that originate from acceptor doping (cf. Fig. 1.3 (b)). An Fe3+ or Cu2+ substitutional on 4+ − 2+ a Ti site in PbTiO3, for example, may trap an oxygen vacancy and form a defect dipole FeTi −VO or 2− 2+ CuTi −VO [Erha 07b] which may pose an obstacle to domain wall motion [Mart 11]. In [Erha 07b], the relative energies of the nonequivalent orientations of the defect dipoles CuTi −VO and FeTi −VO parallel, antiparallel, and perpendicular to the surrounding bulk polarization in PbTiO3 are reported. According to [Erha 07b], the alignment of defect and bulk polarization is energetically favored by 1.21 eV (Cu doping) and 0.45 eV (Fe doping), respectively, compared to the antiparallel orientation. In [Mart 11], the energy barrier for reorienting the FeTi −VO defect complex from the parallel (low energy) to the antiparallel (high energy) configuration was determined to be approximately 1 eV, so that altogether about 1.5 eV have to be overcome to reorient the defect complex. The mechanism underlying ferroelectric hardening is believed to be the following [Damj 05]: In each ferroelectric domain atomic defect dipoles and the ferroelectric polar- ization tend to align. As a consequence it becomes hard to change the ferroelectric polarization direction of the domain by an applied electric field, and the motion of domain walls is impeded. Polarization switching can in general occur via the motion of non-180° or of 180° domain walls. However, in the following the focus lies on the motion of non-180° domain walls (e.g. 90°domain walls). For tetragonal KNbO3 (this work, not shown) and for PbTiO3 it has been found that non-180° domain walls (in this case 90°domain walls) have a lower formation energy than 180°domain walls [Meye 02] and should therefore be more abundant. In a ferroelectric domain the crystal structure is elongated along the polarization direction, i.e., a ferroelectric polarization along the (positive or negative) z direction is accompanied by an elongation along z. If a ferroelectric crystal was poled along the z direction and an electric field in the same direction (±z) is applied, the field-induced changes in ferroelectric polarization and in ferroelectric strain are also oriented along z. The ferroelectric strain obtained this way is approximately reversible, since the ferroelectric re- turns to its original poled state once the electric field is turned off. An especially large ferroelectric strain is obtained when the ferroelectric polarization is switched, e.g., from the z to the x direction by an electric field along x. In this case the strain along x and z is very large, but this large strain normally is obtained only once, since after the polarization has been switched into the new (x) direction, there is no mechanism that restores the original polarization direction (z) [Ren 04]. However, defect dipoles can provide such a restoring force that allows reversible switching by, e.g., 90°, by the following mechanism [Ren 04]: When the ferroelectric crystal or ceramic is produced at a temperature above the Curie temperature, the material is paraelectric, hence the defect dipoles are initially orientated randomly along all possible directions. After cooling down below the Curie temperature, ferroelectric domains form and give rise to preferential orien- tations of the defect dipoles in each domain. If the material is given enough time to equilibrate, the defect dipoles orient along the ferroelectric polarization in their respective domain. If an electric field is applied to the equilibrated material, some of the randomly oriented domains will have a polarization perpendicular to the electric field, such that their polarization is switched by 90°. However, since the new polarization direction is no longer aligned along the defect dipoles in the domain, the polarization will switch back once the electric field is turned off (cf. Fig. 1.3 (c)), since in typical applications the switching cycles are too short to allow the defect dipoles to reorient. This reversible large strain effect was first proposed and observed in [Ren 04] for acceptor doped BaTiO3 and PbTiO3. It was also observed in acceptor doped KNN [Feng 09], which is orthorhombic at room temperature with a polarization in a h011i direction. 5
E
time (a) P S DW (b)
PD
(c)
Figure 1.3: Ferroelectric “hardening“ and the “large-strain effect” caused by defect dipoles. (a): In a ferroelectric crystal region without defect dipoles the spontaneous ferroelectric polarization P is switched by 90° by an applied electric field E. Because different polarization directions are equivalent, the domains remain in their new polarization state after the electric field approaches zero. The large strain S accompanying this switching process is irreversible. (b): A defect dipole with a dipole moment PD parallel to the surrounding spontaneous polarization impedes switching in a hard ferroelectric. The domain wall (DW) gets stuck at the defect. A small strain occurs. (c): The polarization around a defect dipole is switched temporarily by an electric field, but as the electric field approaches zero, the defect dipole restores the original polarization direction of its domain. This is the reversible “large-strain effect”.
Cu doping in KNN Piezoelectric ceramics are commonly produced via a carbonate and oxide processing solid state powder technology [Li 08, Herb 07]. Metal oxide and carbide powders, like KCO3 and Nb2O5, are mixed and solidified and densified by calcining and sintering. In industrial production the ceramics are preferentially sintered at atmospheric pressure [Mats 05]. Pure KNN ceramics fabricated in this way are less dense than desired [Mats 05]. This problem can be overcome using advanced sintering techniques, such as hot pressing or the field assisted sintering technique (FAST, also called spark plasma sintering) [Acke 10], but these techniques are unfeasible for industrial mass production [Acke 10]. A cheaper way to obtain denser KNN ceramics is liquid phase sintering using a sintering aid. Adding CuO as a sintering aid improves the densification of the ceramic KNN [Sait 04, Mats 04], possibly by forming a liquid phase, which can act as a vehicle for atom transport and therefore facilitate grain growth [Mats 05]. Combining CuO as a sintering aid with a small Nb excess (on the order of 1%), densities close to those obtained with hot-pressing were obtained [Mats 05]. Cu doping is therefore a possible approach to improve the quality of KNN ceramics in industrial production processes. Besides its positive influence on the grain and pore structure of the KNN ceramics, Cu may be incorporated into the KNN lattice, where it acts as a dopant and may affect the electrical and ferroelectric properties of KNN. But it is not as obvious which lattice site Cu occupies in (K,Na)NbO3 as it is for Ta dopants, for example. If one compares formal ionic charges and ionic radii of host and dopant atoms, in the case of Ta doping it is very likely that Ta substitutes on Nb sites. Ta is isovalent with Nb, and the ionic radii of Ta5+ and Nb5+ are equal (0.64 A)˚ [Shan 76]. In the case of Cu doping the substitution site is less obvious. The typical formal charges of Cu ions in oxides + 2+ + + + are Cu (as in Cu2O) and Cu (as in CuO). Cu is isovalent with the A site ions in KNN (Na and K ), but the ionic radius of Cu+ is smaller (Cu+: 0.77 A;˚ Na+: 1.02 A;˚ K+: 1.38 A˚ in 6-fold coordination). Cu2+ (0.87 A˚) has a similar ionic radius as Nb5+ (0.64 A)˚ [Shan 76], but its ionic charge is much smaller. Therefore it is hardly possible to determine the substitution site just by means of ionic radius and charge. As far as ionic charge and radius are concerned, A and B sites are the only plausible substitutions sites, but neither of them is ideally suited for Cu dopants. Cu might therefore substitute on A or B sites, or even on both. Depending on its substitution site, Cu can act either as a donor (Cu2+ on A sites) or as an acceptor (Cu2+ on B sites), or it may be incorporated as an isovalent dopant (Cu+ on A sites). As a donor Cu may lead to softening, as an acceptor to hardening of KNN ceramics. As an isovalent dopant it may alter the piezoelectric properties in a way similar to Li doping, which causes an additional MPB with enhanced 6 1. Introduction piezoelectric coefficients [Guo 04]. Therefore it is important to know the substitution site of Cu dopants in KNN in order to understand how Cu dopants modify the ferroelectric properties of KNN.
Goals and structure of this work This work is a theoretical atomic- and electronic-level study of transition-metal doping in the lead-free ferroelectric perovskite KNN. Cu and Fe were chosen as repre- sentative doping elements. In addition, Cu and Fe are EPR-active (EPR: electron paramagnetic resonance, also called electron spin resonance) so that the theoretical findings can be compared to EPR experiments. The focus lies on Cu doping, which has been more intensively investigated experimentally than iron doping because of its positive effect on the sintering behavior of KNN ceramics. This work is divided into two parts: In the first part isolated Cu and Fe substitutionals, in the second part defect complexes of Cu substitutionals and vacancies are investigated. After this introduction, a short gen- eral description of the methods applied here follows in chapter 2, “Computational Methods”. The first goal of this work is to determine whether Cu and Fe dopants substitute on A (alkali) or B (Nb) sites, or on both. The results are compiled and discussed in Chapter 3, “Ab initio thermodynamics of isolated Cu and Fe substitutionals in KNN”. The second goal is to find or refute a possible MPB in Cu-doped KNN. The results are given and discussed in Chapter 4, “Ferroelectricity in KNO with isolated Cu substitutionals”. The third goal is to gain a better understanding if and how defect complexes in Cu-doped KNN contribute to ferroelectric hardening. The results are compiled and discussed in Chapter 5, “Ab initio thermodynamics of defect complexes in KNO and KNN”.
Methods For the thermodynamics of point defects an accurate method is required which is transfer- able to different chemical environments. Especially if point defects are charged, the electronic structure should be taken into account. A suitable method for such cases is density-functional theory (DFT). DFT was developed by Hohenberg and Kohn [Hohe 64]. Its main statements are that the quantum-mechanical ground-state properties of a system of electrons, like valence electrons in a crystal, are a functional of the electron density, and that the total energy follows a variational principle with respect to the electron density: the ground-state density minimizes the total energy. Therefore DFT allows to obtain electronic ground-state properties for atoms, molecules and solids, such as energies and structures. In this work DFT is applied in order to obtain defect thermodynamics and electronic and structural properties of undoped and doped KNN. Density-functional theory calculations are regularly applied in materials modeling because of their accuracy, but they are computationally demanding compared to other methods and therefore limited to sys- tems that contain on the order of a few hundred to thousand atoms, whereas classical atomistic simulation using interatomic potentials is much faster and therefore applicable to much larger systems. The simplest form of such a potential a two-body interaction between two atoms each. Then the potential energy of the crystal is obtained by superposing the interactions between all pairs of atoms. One example for the analytical form of the atom-atom interaction is the Buckingham potential. In this model the atoms interact via Coulomb forces between ions, van der Vaals forces, an short-range repulsive forces, which mimic the repulsion of overlapping electronic wave functions of neighboring atoms according to the Pauli exclusion principle. In this work, atomistic simulations with Buckingham potentials were combined with DFT cal- culations to obtain structural and energetical properties of undoped and doped KNO. By this combined approach it is possible to take advantage of both the accuracy of DFT and the speed of classical atomistic simulations. Details of the method and the computational approach are given in chapter 2, “Computational Methods“, and in the chapters 3 to 5. 2 Computational Methods
In this chapter it is described shortly how the density-functional theory and the classical atomistic simu- lations in this work were performed. For the theoretical background of the DFT and its numerical imple- mentation used here the reader is referred to Ref. [Meye 99], and for details of the program GULP, which was used for the classical atomistic simulations, to, e.g., Ref. [Gale 97], and references therein.
2.1 Density-functional-theory calculations
Density functional theory (DFT) in the local (spin-)density approximation (L(S)DA) was employed to determine defect formation energies of Na, K, Nb and O vacancies, substitutional Cu and Fe atoms, and Cu- defect complexes in KNbO3,K0.5Na0.5NbO3, and NaNbO3. The defect formation energies were calculated for atomistic supercell models containing 2 × 2 × 2 sc (40 atoms) or 2 × 2 × 3 sc (60 atoms) or 2 × 2 × 2 fcc (80 atoms) unit cells. Optimally smooth norm-conserving pseudopotentials as proposed by Vanderbilt [Vand 85], the Ceper- ley-Alder [Cepe 80] LDA exchange-correlation functional as parametrized by Perdew and Zunger [Perd 81], and Chadi-Cohen [Chad 73] k-point meshes that are equivalent to an 8 × 8 × 8 k-point mesh for the sim- ple cubic 1 × 1 × 1 unit cell with Gaussian broadening [Fu 83] of 0.2 eV were applied. The calculations were performed with the mixed-basis pseudopotential (MBPP) method [Meye, Elsa 90, Ho 92, Meye 95, Lech 02], employing a basis of plane waves up to a maximum energy of 340 eV combined with atom- centered basis functions for alkali metal s + p semicore states, oxygen p valence states, Nb s + p semicore and d valence states, and Cu and Fe d valence states. The atomic positions were relaxed until all forces were smaller than 10 meV/A.˚ The differences in total energies between successive relaxation steps were typically of the order of 10−6 eV per formula unit. In the case of charged defects a compensating homoge- neous background charge density was included in the calculations of electrostatic energies and potentials. The LDA bandgap of bulk KNN obtained with a band structure calculation is about 1.7 eV, whereas the experimental bandgap of KNbO3 is about 3.3 eV [Wies 74]. Because this serious underestimation of the bandgap by LDA can affect the gap states and thus the defect formation energies, in some cases the LDA bandgap was corrected by rigidly shifting the conduction band states upwards by the difference between the experimental and the LDA bandgap. This crude method was chosen because there is no preferential first-principles method to correct for LDA bandgap errors, which is computationally efficient enough to be applicable to rather large defect supercells. Two potentially applicable approaches are LDA+U (see, e.g., Refs. [Mose 07, Cart 08, Lech 04]) or SIC-LDA (see, e.g., Refs. [Korn 10] and [Voge 97]). But an application to the KNN system was postponed to a future work.
The Virtual Crystal Approximation for KNN 50/50 Modelling arbitrary compositions of KNN is hardly possible using the supercell approach because large supercells are required. E.g., the ratio K0.51Na0.49NbO3 requires a supercell consisting of 100 unit cells (51 KNbO3 and 49 NaNbO3 unit cells). The virtual crystal approximation (VCA) overcomes this difficulty by replacing the two atom types of a binary system by one virtual atom type which is created by mixing the properties (in this case the pseudopotentials that represent the atoms) in the required ratio [Nord 31]. Sev- 8 2. Computational Methods eral mixing procedures were employed for PZT by other authors [Rame 00, Bell 00]. In parts of this work pseudopotentials of K and Na were linearly mixed to model KNN. Further details are given in section 3.4.
2.2 Atomistic simulations with shell-model potentials
In the shell model [Dick 58], the atoms are represented as point-like ionic cores and spherical electronic shells, which enables atoms to be electrically polarized, in contrast to the so-called rigid-ion potentials, in which each atom or ion is represented by one particle only. The cores and shells of different atoms interact via the electrostatic Coulomb potential,
Coul qiq j Vi j (ri j) = , (2.1) 4πε0ri j where qi is the charge of particle i and ri j the distance between particle i and particle j. The shells addi- tionally interact via a two-body Buckingham potential,
r − i j C Buck ρi j i j Vi j (ri j) = Ai je − 6 , (2.2) ri j where ri j is the distance between two shells. The first term is the short-range repulsion, the second the Van-der-Waals interaction. The core and shell of an atom interact via a spring potential, 1 1 V spring = k r2 + k r4 , (2.3) i,cs 2 2 i,cs 24 4 i,cs where ri,cs is the distance between core and shell of atom i. The Buckingham interaction was truncated at a distance of 6.5 A,˚ the spring interaction at 1 A.˚
2.3 The Nudged-Elastic-Band Method (NEB method)
In order to obtain energy barriers for switching the spontaneous polarization, the ”nudged-elastic-band“ (NEB) method [Henk 00] was applied. The NEB method is a method to obtain the ”minimum energy path“ (MEP) between two structures (initial and final structure, e.g. two different polarization directions) that correspond to local minima on the potential energy landscape of the system which are separated by an energy barrier. The energy barrier (together with entropy differences) determines how probably or at which rate the system changes from one state to the other. In the NEB method a chain of structures are set up which interpolate the initial and final structure in some way, for example linearly. These structures are called images. The energy of the images is then minimized subject to the constraint that the images may not move into the direction of the initial or final structure. This constraint is implemented by subtracting from the forces that act on the particles the force component along the NEB path. In order to keep the images in approximately equal distances from each other, the images are connected by an artificial spring force. At the end of the NEB calculation the images lie along the MEP, and the energy barrier is the energy maximum along the MEP. In this work the NEB was used along with the DFT in Chapter 5.3 and with SMP calculations in Chapter 5.4. 3 Ab initio thermodynamics of isolated Cu and Fe substitutionals in KNN 10 3. Ab initio thermodynamics of isolated Cu and Fe substitutionals in KNN
The content of this chapter, with the exception of the sections 3.4, 3.5, and 3.6, has been published in [Korb 10] (Korbel,¨ Marton, Elsasser,¨ Phys. Rev. B, Vol. 81, pp. 174115-1–11, May 2010. Copyright 2010 by the American Physical Society).
In this chapter the most stable substitution sites of Cu and Fe dopants in KNN are determined for dif- ferent thermodynamic conditions.
3.1 Introduction ([Korb 10], Section I)
In this chapter, the substitution site of Cu dopants in KNN is determined as a function of the processing conditions. In Ref. [Zuo 08] it is assumed that Cu atoms substitute Nb atoms (on B sites in the perovskite structure ABO3), and oxygen vacancies are formed as a charge compensation mechanism. In Ref. [Li 07] Cu is assumed to substitute both alkali and Nb atoms (on A or B sites, respectively). While Cu2+ on a Nb5+ site can act as an acceptor dopant and lead to ferroelectric hardening, it can act as a donor on a Na+ or K+ site and cause ferroelectric softening. In order to determine the substitution site of Cu dopants and the prevailing vacancy type as function of the chemical processing conditions, first-principles density functional theory (DFT) calculations were carried out to obtain the thermodynamic stability of defect configurations and charge states of vacancies and Cu and Fe substitutionals in KNN. The defect formation enthalpy, which determines the substitution site of the dopant, depends on the chemical potentials of the atomic reservoirs [Wall 04], which for gases (oxygen in the case of KNN) correspond to the temperature and the partial pressures of the elemental constituents during processing [Reut 01]. In the present work the influence of the processing conditions is investigated according to the thermodynamic first-principles formalism outlined in, e.g., Ref. [Erha 07a]. First-principles calculations have already been applied to other ferroelectric materials by other authors, e.g., copper and iron doping in lead titanate [Erha 07b], vacancies in alkali niobates [Shig 04, Shig 05], and iron impurities in potassium niobate [Post 98] have been investigated. No such studies of Cu doping of alkali niobates have been reported so far, and comparable published studies of vacancies in the KNN system have been limited to neutral vacancies.
3.2 Thermodynamic formalism ([Korb 10], Section II)
The thermodynamically most stable lattice site for a dopant is governed by the defect formation enthalpy, which for most cases of inorganic solid state systems can be approximated by the defect formation energy [Wall 04]. The defect formation energy E f for a defect X in a charge state q is given by [Wall 04]
f q q E [X ] = Etot[X ] − Etot[bulk] − ∑µini + q · (EF + EVBM + ∆V), i
q where Etot[X ] is the total energy of the supercell containing a defect X with charge q, and Etot[bulk] is the total energy of a perfect cell of the same size. µi is the chemical potential of atom species i, ni is the number of atoms of species i that is exchanged with a reservoir in the defect formation reaction (ni > 0 for species that are added to, ni < 0 for species that are removed from the host crystal), and EF is the Fermi energy relative to the energy of the valence band maximum, EVBM. At zero temperature the Fermi energy is identical with the chemical potential of an electron, and the total energy per atom of a species is identical with the chemical potential of the species. ∆V is a correction term which aligns the energy zero of the crystal with a defect to that of the perfect crystal [Wall 04]. For charged defects, −q is the number of excess electrons assigned to the defect. Assuming chemical equilibrium conditions, the chemical potentials of the constituents can vary in ranges that are given by requiring the perovskite phase to be stable and the fact that the chemical potentials cannot exceed those of the composing elements in their most stable phases (see Tab. 3.3 for the numerical values). 3.2 Thermodynamic formalism 11
In the following, relative chemical potentials ∆µi are used:
0 ∆µi = µi − µi , (3.1) 0 where µi is the chemical potential of species i in its most stable elemental phase. In equilibrium, the alkali niobate perovskite ANbO3 is stable if
0 ∆µA + ∆µNb + 3∆µO = ∆Hf (ANbO3), (3.2) 0 where ∆Hf (ANbO3) is the formation enthalpy of ANbO3 from its elemental constituents (metallic A, metal- lic Nb, and gaseous oxygen). The upper boundary of the chemical potentials is
∆µi ≤ 0, (3.3) since otherwise the elemental phase of component i would precipitate. The outer triangle in Fig. 3.1 represents the allowed range of the chemical potentials for the alkali (A) and Nb atoms according to Eqs. (3.2) and (3.3). The allowed region is further confined by the competing oxide phases A2O, A2O2, NbO, NbO2, and Nb2O5:
0 2∆µA + ∆µO ≤ ∆Hf (A2O), (3.4) 0 2∆µA + 2∆µO ≤ ∆Hf (A2O2), 0 ∆µNb + ∆µO ≤ ∆Hf (NbO), 0 ∆µNb + 2∆µO ≤ ∆Hf (NbO2), 0 2∆µNb + 5∆µO ≤ ∆Hf (Nb2O5).
In the case of a mixed perovskite, where the A sites are occupied partly by K and partly by Na atoms, there is an additional degree of freedom for the two A species. However, because metallic Na and K crystallize in the same structure, and the formation energies for the two alkali metals and their oxides are of similar magnitude, in the calculation of the chemical potentials it was assumed that ∆µK = ∆µNa = ∆µA and 1 ∆H0(bccA) = ∆H0(bccK) + ∆H0(bccNa), (3.5) f 2 f f 1 ∆H0(A O) = ∆H0(K O) + ∆H0(Na O), f 2 2 f 2 f 2 1 ∆H0(A O ) = ∆H0(K O ) + ∆H0(Na O ). f 2 2 2 f 2 2 f 2 2 The allowed ranges of the chemical potentials that remain after taking into account Eqs. (3.4) are indicated by the gray area in Fig. 3.1. Different processing conditions correspond to different combinations of the chemical potentials. The points indicated by the numbers 1 to 8 in Fig. 3.1 cover all extremes of the possible conditions. These extremes correspond to the equality in two of the Eqs. (3.3) and/or (3.4). For each of the points 1 to 8, the ANbO3 perovskite and two reservoir materials are in thermodynamical equilibrium. For instance, at point 2, ANbO3,O2, and A2O2 are in equilibrium. Together with Eq. (3.2), which must hold for all points, three equations for each point determine the three chemical potentials of A, Nb, and O:
Point 1:
0 2∆µA + ∆µO = ∆Hf (A2O), 0 2∆µA + 2∆µO = ∆Hf (A2O2). 12 3. Ab initio thermodynamics of isolated Cu and Fe substitutionals in KNN
Point 2:
∆µO = 0, 0 2∆µA + 2∆µO = ∆Hf (A2O2).
Point 3:
∆µO = 0, 0 2∆µNb + 5∆µO = ∆Hf (Nb2O5).
Point 4: 0 2∆µNb + 5∆µO = ∆Hf (Nb2O5), 0 ∆µNb + 2∆µO = ∆Hf (NbO2).
Point 5: 0 ∆µNb + 2∆µO = ∆Hf (NbO2), 0 ∆µNb + ∆µO = ∆Hf (NbO).
Point 6:
∆µNb = 0, 0 ∆µNb + ∆µO = ∆Hf (NbO).
Point 7:
∆µNb = 0, ∆µA = 0.
Point 8:
∆µA = 0, 0 2∆µA + ∆µO = ∆Hf (A2O).
In the following, the chemical potentials µ were approximated by the total energies per atom E for the 0 0 crystalline materials and the formation enthalpies ∆Hf were approximated by the formation energies ∆E f . When calculating the defect formation energies E f , a temperature and pressure correction was applied for the oxygen gas as follows. The chemical potential of oxygen is related to the oxygen partial pressure and the temperature via the ideal gas equation [Reut 01]:
1 1 µ (T, p) = G (T, p) = (H (T, p ) − TS (T, p )) O 2 O2 2 O2 0 O2 0 1 p + kBT ln (3.6) 2 p0 1 p = µO(T, p0) + kBT ln , 2 p0 where p0 is a reference oxygen partial pressure (the standard atmospheric pressure, about 1 bar), T is the temperature, G is the Gibbs free energy per molecule, and H and S are enthalpy and entropy per oxygen molecule, which can be found in thermochemical tables and were taken from Ref. [Lide 08]. For a deriva- tion of Eq. (3.6) see e.g. Ref. [Reut 01]. The oxygen chemical potentials were calculated (cf. Table 3.1) that correspond to air at the standard atmospheric pressure (pO2 ≈ 0.2 bar) and to air at 1 µbar (pO2 ≈ 0.2 µbar), indicated by diagonal lines in Fig. 3.1, at about room temperature (300 K) and at 1300 K, the latter being in the range of typical sintering temperatures for KNN [Acke 10]. 3.2 Thermodynamic formalism 13
300 K 1300 K 0.2 bar 0.2 µbar 0.2 bar 0.2 µbar -0.295 -0.473 -1.576 -2.350
Table 3.1: Relative chemical potential of oxygen, ∆µO, in eV for oxygen partial pressures of 0.2 bar and 0.2 µbar, for the temperatures 300 K and 1300 K, respectively. ([Korb 10] Tab. I)