Recently, a thin film of Bi4Ti3O12 (BTO) has been widely studied as one of candidate materials for applications of ferroelectric random access memory due to its large remnant polarization at room temperature. The crystal structure of BTO consists of pervoskite type units, Bi2Ti3O10 (An−1BnO3n+1 for n = 3), and interleaved Bi2O2 layers along the c-axis. In the Bi2Ti3O10 units, Ti ions are surrounded by oxygen octahedrons, and Bi ions occupy the outside of TiO6 octahedrons [1, 2]. Although the crystal structural information of BTO is essential to understand the remnant polarization at room temperature, the structural study of BTO has not been performed sufficiently because the difference between two kinds of crystal systems, monoclinic and orthorhombic, is subtle. This present work is focused on the structural refinement of BTO at room temperature using neutron powder diffraction data. The Rietveld refinement of Bi4Ti3O12 was carried out using neutron powder diffraction data, which provide a better refined results for a number of oxygen atoms in the unit cell, though Ti atom has a negative scattering length (−0.3438 × 10−12 m) for neutron radiation. The sample of BTO was prepared by a normal solidstate reaction using Bi2O3 and TiO2 as a starting material. The mixture was annealed at 1050 ◦C for 17 h. The second annealing process was carried out at 1150 ◦C for 48 h followed by intermediate regrinding. Neutron powder diffraction data were collected over scattering angles 0 ◦–160 ◦ using 1.8348 A neutron on the High Resolution Powder Diffractometer (HRPD) at Hanaro Center of Korea Atomic Energy Research Institute in Korea. The General Structure Analysis System (GSAS) program was used to do the Rietveld refinement [3]. The pseudo-Voigt function was used as a profile function among profile ones in GSAS program [4]. According to the report of the single crystal X-ray experiment [5], the phase of BTO at room temperature belongs to the monoclinic system with B1a1 space group. However, Hervoches et al. reported that BTO at room temperature has orthorhombic crystal system from the Rietveld refinement using neutron powder diffraction. Also, they mentioned that the deviation between orthorhombic and monoclinic system was too small to be detectable with neutron diffraction data [6]. When the reflections by the b-glide in the orthorhombic system with B2cb space group are not observed, it is difficult to discern the difference between the two crystal systems quantitatively because the final diffraction patterns for both crystal systems are almost identical. In order to start the Rietveld refinement, a reasonably approximated model of BTO for the actual structure based on the monoclinic and orthorhombic crystal system is required. As a starting model of BTO, two kinds of models based on the monoclinic and orthorhombic systems were used. The model based on the B1a1 space group (M-model) was built with crystallographic data taken from the single crystal X-ray diffraction [5]. The crystallographic data reported by Hervoches et al., were used to make the other one (O-model) [6]. The Rietveld refinement of BTO was performed by neutron powder diffraction data, which is powerful to determine atomic positions and occupancies factors of oxygen atoms. Because BTO contains a large number of oxygen atoms in the unit cell. The neutron scattering lengths for Bi, Ti and O atoms are 0.853 × 10−12 m, −0.3438 × 10−12 m and 0.5805 × 10−12 m, respectively. The initial refinement for the O-model was done by the unit-cell, the zero-point shift and background parameters. The zero-point shift was correct by neutron powder diffraction data of α-Al2O3 as a standard sample. A good match of the peak positions was achieved from the preliminary refinement, and then the peak profile parameters including the peak asymmetry were refined. A pseudo-Voigt function in this study was used as a profile function. It assumes that there is no defect for oxygen sites because the samples were prepared under air atmosphere. The final weighted and profile R-factors, Rwp and Rp, obtained from the O-model through the Rietveld refinement procedure showed 8.61% and 6.31%, respectively. Fig. 1 illustrates the Rietveld refinement patterns for the O-model. In the case of the same Rietveld refinement procedure applied to the M-model, all R-factors for M-model were lower than those for O-model. Among R-factors, the weighted R-factor, Rwp, was considerably decreased from 8.61% to 5.59%. The profile factor, Rp, which is a true quantity based on the discrepancies between observed and calculated intensities, was 4.17%. Also, the goodness-of-fit indicator, S(= Rwp/Re), was decreased from 2.15 to 1.40. The Rietveld refinement patterns of the M-model are shown
Read full abstract