Bi₄Ti₃O₁₂ is an attractive ferroelectric material for use in nonvolatile memories because of its demonstrated fatigue resistance.^1,2 Additionally, its highly layered perovskite structure makes it ideally suited for study using the atomic-scale layering capabilities of molecular beam epitaxy (MBE). Epitaxial Bi₄Ti₃O₁₂ has been grown on various substrates by sputtering,³ laser ablation,⁴ and by laser MBE.⁵ We have used adsorption-controlled growth conditions to accurately and reproducibly grow stoichiometric films.

The EPI 930 MBE system⁶ used for this study is described in detail elsewhere.⁷ The (001) SrTiO₃ wafers utilized in this study are etched with a buffered-HF solution prior to growth⁸ exposing the TiO₂-terminated surface. During growth, bismuth and ozone are supplied to the surface of the film continuously while the substrate temperature is maintained at 600­660 °C. The ozone background pressure used during growth is 2 ? 10^­5 Torr, which represents an incident flux of ozone hundreds of times greater than the flux incorporated into the growing film. Similarly, bismuth is supplied at an incident flux 2 to 5 times greater than the average titanium incident flux. Titanium is supplied in shuttered doses each containing three monolayers of titania which make up the Bi₂Ti₃O₁₀ perovskite sheets. Adsorption-controlled growth is achieved under these conditions. The average growth rate we have investigated is ~ 0.5 Å/s.

Using the growth conditions described above it is possible to use a range of bismuth incident fluxes and substrate temperatures and still obtain phase-pure material. Rutherford backscattering spectrometry (RBS) measurements indicate that the Bi₄Ti₃O₁₂ films are stoichiometric within the error of the measurements. Results obtained from film thicknesses calculated from x-ray diffraction peak widths and thickness fringes of ultrathin films, RBS composition measurements, counting RHEED half-order intensity oscillations during growth, and in situ flux measurements using a quartz crystal thickness monitor and AA spectroscopy all indicate that the incident titanium flux determines the growth rate of the films. The excess bismuth and oxygen desorb from the surface. A similar adsorption-controlled growth mechanism was shown to be operative by de Keijser and Dormans⁹ for Pb(Zr,Ti)O₃ thin films grown by organometallic chemical vapor deposition (OMCVD). They found that by using a range of lead precursor partial pressures, it was possible to obtain stoichiometric films.

Adsorption-controlled growth was first successfully utilized for the MBE synthesis of epitaxial GaAs thin films nearly 30 years ago.¹⁰ This growth mechanism relies on the volatility of the group V component. In the case of GaAs, the equilibrium vapor pressure of gaseous As_4,(g) over pure, solid As_(s) is about ten orders of magnitude higher than the equilibrium vapor pressure of gaseous As_2,(g) over a gallium-rich GaAs surface at 600 °C.¹¹ As a result, stoichiometric films are easily grown by supplying an excess of arsenic to the surface of the depositing film.

For the case of bismuth titanate its heat of formation is unknown, so we cannot make a thermodynamic assessment of its adsorption-controlled growth. But we have demonstrated the ability to synthesize stoichiometric films under adsorption-controlled conditions. At the incident bismuth flux and growth temperatures employed, elemental bismuth will not condense on the substrate surface¹² and must first be oxidized. Although Bi₂O_3,(s) does not evaporate congruently, the primary oxide species present in the vapor phase under growth conditions that we typically employ are BiO, Bi₄O₆, and Bi₂O₃.¹³ It is the volatility of these oxides that must be considered to take the place of the group V molecules during adsorption-controlled growth. Our achievement of adsorption-controlled growth indicates that the equilibrium vapor pressure of these bismuth oxide complexes over pure Bi₂O_3,(s) is much higher than over a titania-rich Bi₄Ti₃O_12,(s) surface.

Prior to the initiation of growth, the continuous exposure of ozone and bismuth on the substrate surface is established. This aids in the cleaning of the substrate surface.¹² No change in the RHEED pattern is observed until the titanium flux is initiated. Below we describe the temporal evolution of the RHEED pattern during the deposition of one-half unit cell where we have utilized changes in surface symmetry as a means of controlling shutters.

Figure 1(a) shows the RHEED pattern of the bare etched SrTiO₃ substrate at ~ 640 °C during exposure to 2 ? 10^­5 Torr of ozone. Figures 1(b)­1(f) show the evolution of the RHEED pattern during the deposition of one-half-unit cell of Bi₄Ti₃O₁₂ oriented with its c-axis normal to the substrate surface (c-axis oriented). The half-order streaks present along the substrate [110] azimuth [Fig. 1(b)] are due to the epitaxial relationship between the cubic substrate and pseudoorthorhombic film² where Bi₄Ti₃O₁₂ [100]||SrTiO₃[110]. With each opening of the titanium shutter, the half-order streaks disappear after a period of time that corresponds to the deposition of 1.5 monolayers of titania as shown in Fig. 1(c). After the deposition of 3 monolayers of titania, the half-order streaks again reappear, as shown in Fig. 1(d), and the titanium shutter is closed. The disappearance of the half-order streaks halfway through the titanium burst of each formula unit could be due to the reduced distortion of the centermost layer of TiO₆ octahedra in the Bi₄Ti₃O₁₂ crystal structure. The octahedra in this central layer, unlike those on either side of the Bi₂O₂ layers, are not rotated about the c axis of Bi₄Ti₃O₁₂.^1,14 The square surface mesh of the central TiO₆ layer has a perovskite lattice spacing (a_p) like the substrate and thus produces no half-order streaks. However, due to the ~ 7° antiphase rotation of the TiO₆ octahedra in the other two layers,^2,18 they have a lattice spacing of ~ a_p and a surface mesh that yields half-order streaks.
Figure 1.

If the titanium shutter is not closed following the initial reappearance of the halforder streaks [Fig. 1(d)], the film surface quickly roughens and the RHEED pattern becomes spotty. The spotty RHEED pattern observed is consistent with the formation of anatase (TiO₂), which has been conclusively identified in subsequent four-circle x-ray diffraction measurements.

Although we argue that the sticking coefficients of bismuth oxide complexes to pure Bi₂O_3,(s) are negligible under adsorption-controlled conditions, one Bi₂O₂ double layer is incorporated with each Bi₂Ti₃O₁₀ perovskite sheet to form Bi₄Ti₃O₁₂ during growth. This behavior is consistent with maintaining charge neutrality—one (Bi₂O₂)^{2 +} is incorporated to neutralize each (Bi₂Ti₃O₁₀)^2­ sheet.

The adsorption-limited incorporation of bismuth was also recently demonstrated for the growth of the superconductor Bi₂Sr₂CuO₆ by MBE.¹⁵ Migita et al.¹⁵ found that under adsorption-limited growth conditions bismuth incorporation was limited to two BiO layers per formula unit of Bi₂Sr₂CuO₆, despite their flooding the film surface with excess bismuth. This observation is also consistent with our hypothesis of adsorption-limited incorporation to achieve change neutrality.

In Fig. 2(a) the -2 four-circle x-ray diffraction scan (using Cu K radiation) of a 1000 Å thick Bi₄Ti₃O₁₂ film grown on (001) SrTiO₃ is shown. Intense 00 peaks indicate that the film is pure c-axis oriented Bi₄Ti₃O₁₂. The rocking curve full width at half maximum (FWHM) of the Bi₄Ti₃O₁₂ 00 reflection is measured to be 0.3° in  and 0.25° in 2 showing minimal out-of-plane misalignment. Figure 2(b) shows the azimuthal scan (-scan) of the 117 reflections of this same film. The peaks indicate a Bi₄Ti₃O₁₂ [110] || SrTiO₃ [010] orientation relationship with a FWHM of 0.4° in , indicating little variation in the in-plane alignment. These peak widths are all comparable to the instrumental resolution of our Picker four-circle diffractometer. RBS channeling results for this film showed a minimum channeling yield (_min) of 0.20 for the bismuth signal behind the surface peak, which is the lowest reported value for Bi₄Ti₃O₁₂ thin films.
Figure 2.

An atomic force microscopy (AFM) image of the surface of this same film is shown in Fig. 3. Clearly visible on the surface are micron-sized mounds that protrude approximately 100­200 Å out of the film. The terraces making up these mounds have step heights that are predominantly integral multiples of a half-unit cell (the height of a Bi₄Ti₃O₁₂ formula unit). The surface morphology revealed by AFM is reminiscent of that of layered perovskite superconductor films,^16,17 although it is unclear if oppositely-signed screw dislocations are present within each mound or if the mounds arise due to limited surface diffusion across Ehrlich­Schwoebel step-edge barriers.^18,19
Figure 3.

Figure 4 is a high resolution transmission electron microscopy (HRTEM) micrograph of this same film viewed along the [100] Bi₄Ti₃O₁₂ zone axis. Figure 4(b) is an enlarged view from the middle of Fig. 4(a). This image was taken at a defocus value of about ­40 nm with a JEOL-4000EX high resolution electron microscope. Figure 4 shows the lattice image corresponding to the projected structure of Bi₄Ti₃O₁₂ along the [100] direction. From Fig. 4(b) it is seen that the Bi₂O₂ double layers (indicated by two white arrows) and the three perovskite sheets (marked by the black arrows) alternate along the [001] direction of the Bi₄Ti₃O₁₂ film. No stacking faults or other defects were found along the entire film studied with TEM.
Figure 4.

In conclusion, we have grown epitaxial Bi₄Ti₃O₁₂ films by reactive MBE under conditions of ozone background pressure and temperature where an adsorption-controlled growth mechanism dominates, i.e., the excess bismuth, bismuth oxides, and ozone desorb from the surface leaving behind a phase pure, stoichiometric crystal.

The authors gratefully acknowledge the financial support of the Office of Naval Research through Grant No. N00014-94-1-0690, the Department of Energy through Grant No. DE-FG02-97ER45638, and Dr. Larry McIntyre of the University of Arizona for the RBS analysis. One of the authors (C.D.T.) gratefully acknowledges the support of the IMAPS Educational Foundation.