This action might not be possible to undo. Are you sure you want to continue?
S P E C I A L I S S U E PA P E R
U. Breuer · W. Krumpen · F. Fitsilis
XRF and SIMS/SNMS analyses of BaxSr1–xTiO3 dielectrics
Received: 9 October 2002 / Revised: 8 January 2003 / Accepted: 16 January 2003 / Published online: 27 February 2003 © Springer-Verlag 2003
Abstract The development of analytical tools and procedures for process control is a prerequisite for the integration of high permittivity and/or ferroelectric materials in CMOS devices. The thickness and composition of perovskite oxide films were determined by wavelength dispersive X-ray fluorescence analysis (XRF) with special emphasis on the ratio of the group-II elements to the Ti content, and a precision of 0.5% was achieved for a typical film thickness of 20–30 nm. Secondary ion mass spectrometry (SIMS) and sputtered neutrals mass spectrometry (SNMS) was used for depth profiling to determine film homogeneity and elemental interdiffusion at heterointerfaces. Examples are given for BaxSr1–xTiO3 and SrTiOx thin films which were grown in a prototype MOCVD production tool. No interdiffusion was observed for films grown at 600 °C on Pt electrodes in contrast to films grown directly on Si. Keywords XRF · SIMS · SNMS · MOCVD · BST · DRAM
ium–strontium–titanate, BST, which is considered to be a promising candidate to enter the DRAM manufacturing process as a replacement for the conventional oxide/nitride/oxide (ONO) dielectric layers. Produced in thin films, BST has a high permittivity of 100–400, compared to 3.9 for pure SiO2, thus allowing the design of the cell in a simple stack structure even for the future Gbit DRAMs Special emphasis is given to the exact determination of the group-II/Ti ratio. Bulk ceramics of these perovskites are exactly stoichiometric and a possible surplus of metallic elements precipitate at grain boundaries and triple points. For thin films grown at lower temperatures, such precipitates are not identified up to rather large deviations from the stoichiometric composition. The surplus metals are incorporated in the films and affect the electrical properties . Hence, a perfect control of the group-II/Ti ratio is necessary, and we applied XRF as a nondestructive tool to determine the average composition of the films. These investigations are supplemented by studies of the homogeneity of the film and especially the interfacial reactions on different substrates by SIMS/SNMS.
Introduction MOCVD thin films
Electroceramic materials can be integrated into novel devices with unique functionality: multi-Gbit memory modules, nano-sized transistors, high-performance sensors and nano-actuators are examples of applications incorporating ceramic materials which are likely to come to the market in the near future. Perovskite oxides are prime candidates for high-k  as well as for ferroelectric applications  and BST, PZT and SBT were analysed in detail. As an example we concentrate here on the high-k material barMOCVD was used for thin film deposition which allows the fast deposition on large area wafers as required for a mass production technique. Thin films were deposited on commercial p-type Si(100) substrates and on substrates with sputter-deposited Pt films as electrodes. These platinized films generally include additional adhesion layers (i.e. the film stacks used are Pt/TiO2/SiO2/Si or Pt/ZrO2/ SiO2/Si). The precursors for the MOCVD were supplied by a liquid delivery system which mixes solutions from three sources: 0.35 M solutions of Ba(thd)2 and Sr(thd)2 and a 0.4 M solution of Ti(O-i-Pr)2(thd)2 dissolved in butyl acetate. The AIXTRON Planetary Reactor of the 2600G3 series can handle five 6-inch wafers simultaneously. The wafers are placed on a coated graphite susceptor that rotates typically at 8 rpm and carries five smaller plates
U. Breuer (✉) · W. Krumpen ZCH, Research Centre Jülich, 52425 Jülich, Germany e-mail: email@example.com F. Fitsilis IFF, Research Centre Jülich, 52425 Jülich, Germany
907 Table 1 Typical deposition conditions for BST films Substrate temperature (°C) Reactor pressure (mbar) O2/N2O flow rate (sccm) Liquid source feeding rate (mL s–1) Vaporization temperature (°C) Film thickness (nm) 500–750 1–3 50–500 0.08 225–245 10–120
(satellites) which rotate at 40 rpm by gas foil rotation. The gas inlet nozzle is placed centrally above the reactor providing a pure horizontal gas flow direction which makes this reactor a radial flow system. The oxidizer gases (O2/ N2O) enter the reactor separately from the precursors just below the nozzle in order to avoid premature reactions. The reactor operates under low pressure at approximately 2 mbar to increase the gas diffusivity and prevent pre-reactions. The deposition temperatures were varied 500– 700 °C. The process conditions are summarized in Table 1 [3, 4].
high energy Sr-Kα line. A second important background contribution appearing at high energies, corresponding to low theta angles comes from the Compton tail of the characteristic Kα-line of the Rh tube. In the case of our thin films the penetration of the X-rays can be several micrometers (µm) and is consequently much deeper than the film thickness (<10 nm and 150 nm). That means that not only the elements in the desired film but also sublayers must be taken into account. In order to achieve a reliable background subtraction the background must be investigated for every film substrate combination. XRF calibration strategy We will concentrate on the cation composition by XRF and especially the group-II/Ti ratio. Figure 1 depicts the scan profile of a BST standard near the interesting lines. The lines are characteristic for the concentration of the
Wavelength dispersive XRF
Tool description The XRF tool is a RIGAKU ZSX-100e with a 3-kW Rhtube which was generally operated at its maximum tube power of 50 kV–60 mA. The wavelength dispersive element characterization is based on Bragg’s law:
where n is the order, λ the wavelength, d is the lattice parameter of the analyser crystal and θ is the position of the diffraction maximum. The analysing crystal and the detector are rotated in the θ/2θ mode. For the analysis of thin films which contain the elements Ba, Sr, Ti, Pb and Ta the standard analysing crystal LiF200 was used with a lattice parameter 2d=4.0276 Å. For Bi, which was overlapped by the Pt Lβ lines, the M line and a Ge crystal with a lattice parameter of 2d=6.532 Å was chosen. The beam diameter was set to 25 mm, the sample size was 1×1-inch square, and a stainless steel sample holder mask restricted the sample area to a defined size of A=4.91 cm2.Additionally, the samples were rotated during the measurement in order to average any inhomogeneity in the composition and thickness. Analysis was performed at a base pressure of 3 mbar in order to avoid absorption of the fluorescent X-rays as well as air scattering. In addition to the fluorescent radiation, a disturbing diffraction background may pass the analysing crystal of the spectrometer. As we use single crystalline Si substrates, such effects may not be very reproducible even with a rotating sample holder. These diffraction effects become more serious with decreasing wavelength and can in principle be avoided by taking a fluorescent wavelength which is longer than the diffraction limit of the sample. In the case of BST, diffraction effects become significant for the
Fig. 1a,b Line spectrum of fluorescence analysis with background fitting and peak deconvolution a Sr-Ka, b Ti-Ka and Ba-La line
specific elements in the sample, while the background intensity level depends on the substrate (e.g. platinized wafers, typically Pt/adhesion-layer/SiO2/Si, have a higher background than SiO2/Si wafers). For the quantitative evaluation of the integral intensity the peak form can be considered as a pure Gaussian distribution. The intensity calculation starts by fitting the background using a linear, multi-point or polynomial fit. In case of overlapping lines, peak deconvolution is necessary and Gaussian distributions are fitted in the line profile, as shown in Fig. 1. Background calculation is different for the high energetic peaks (Sr) than for the low energetic ones (Ba, Ti) at the other end of the spectrum, where the background remains almost constant. In the first case, a 4-point fitting is used to find a mean value for background extraction in the high-noise region (±50 cps), in contrast to the 2-point linear fit for the Ti and Ba peaks. The fluorescence intensity is directly proportional to the mass of an element in an unknown sample; however, in order to obtain absolute values this mass must be determined from a comparison to the standards: mE E E = cst I N (2) A where mE is the mass of the element in the irradiated surE face A, I N is the net intensity of the examined peak in the XRF spectrum, and c=(mE/A)/IE and is the calibration factor derived from the calibrated standards. Such standards with known stoichiometry and thickness were prepared by chemical solution deposition (CSD). During CSD deposition there is no loss of material and the final composition of the film is therefore identical to the original solution, which was additionally controlled by ICP–AES. Hence, a comparison of the intensity ratios yields directly the cation stoichiometry of the films with high precision. For the determination of the areal mass density and the film thickness, the thickness of the calibration standards was determined by profilometry using a
Table 2 Extracted parameter from the linear fits in Fig. 2 Element Sr Ti Ba C 4.84027 13.0588 44.7132 R2 0.99973 0.99966 0.99973 Reproducibility (%) 0.34 0.21 0.28
VEECO DEKTAK system, and the mass density was deduced assuming the ideal density of the corresponding perovskite. Hence the stoichiometry (e.g. the group-II/Ti ratio) can be determined with a precision of the order of 0.5%, whereas the thickness includes the additional uncertainties of the profilometer and possible deviations from the ideal density. In order not to rely on one calibration sample only, the average over a thickness series was used. Figure 2 shows an intensity versus mass (µg)/illuminated area plot prepared from the data of the calibration standards. We observe a linear increase of the intensity with the mass for all elements. The proportionality factor for each element is derived from the slope of the curves. The parameters of the linear regression model, X=C×I+D, for the BST system are as shown in Table 2. Intercept D was set to zero. The precision of the calibration is depicted in the quality of the fits, as the correlation factor R2 is always near unity. The reproducibility of the measurements is excellent and is below 0.35% for the Sr, Ba, and Ti elements (average of 10 measurements). As the intensities for the single elements in the sample increases, within the given uncertainties, proportionally to the areal mass density, influences of matrix effects and inter-element effects can be neglected for these film thicknesses. The layer thickness, t, can be derived from Eq. (3) if the density of the film is known additionally; we generally use the theoretical density of the crystalline bulk, 5.758 g cm–3 for Ba0.7Sr0.3TiO3 as an approximation.
ρ f ilm
Application For the measurement of the unknown BST samples we use the same measurement parameters as used for calibration samples, and especially the same sample size. Analysis of films grown on Si or oxidized wafers are straightforward; however, layer stacks need special consideration. Our standard platinized wafers have a TiOx, TiNx or TiAlN adhesion layer, which corroborate the Ti stoichiometry. Therefore separate depositions on wafers without adhesion layer or replacement of the Ti by Zr in the adhesion layer was necessary. Under these conditions films of typical thickness 30 nm could be analysed within 5 min (average of 3 measurements) with a precision of 0.5%. Homogeneity tests over 6-inch wafers could be performed after cutting into 1’’×1’’
Fig. 2 Mass versus fluorescence intensity plot for CSD deposited calibration samples. Both stoichiometric (filled symbols) and offstoichiometric (open symbols) samples are included
Fig. 3 Dependence of the average lattice parameter of B70S30T films on the stoichiometry
pieces and very good homogeneity good be demonstrated . In addition Fig. 3 shows as an example the variation of the lattice parameter of the films as a function of the stoichiometry, which indicates a different incorporation of the excess group-II elements and of the Ti (for details see ). Finally, as the XRF analysis presents an average of the elemental distribution within the whole layer stack, interdiffusion may corroborate the results for the ‘bulk’ of the layer of interest. Therefore interdiffusion on different substrates is investigated by SNMS and/or SIMS. For these examples standard TiO2 adhesion layers were used as in this case the Ti can be well separated from the film.
Method For depth profiling of the layer system secondary ion mass spectrometry (SIMS) and sputtered neutral mass spectrometry (SNMS) are suitable techniques. The sample is bombarded with primary ions (Ar, O2, Cs or Ga) and sputtered atoms or ions are detected using a mass spectrometer. Monitoring the mass intensities as a function of sputter time gives the raw dataset for conversion into concentration versus depth . The presented SIMS data were taken from a CAMECA ims4f and the SNMS data were measured using a VG SIMSLAB 410 (quadrupole based). For SNMS an electron beam is used for postionisation of sputtered neutrals. In general the SNMS is about 3 orders of magnitude less sensitive compared to SIMS because of geometry effects (limited collection efficiency). On the other hand SIMS is more sensitive but suffers from strong matrix effects when profiling heterolayer structures. Matrix effects are absent or very small in SNMS. Nevertheless the required precision in the determination of the group-II to Ti ratio is not possible with these methods. The study of the interface and its role in the nucleation of the films as well as its final electrical properties is of primary interest. Some information about the interface quality was already found in terms the HRTEM picture that showed a sharp transition between the Pt substrate and the BST film for the high-Tgrowth deposition [3, 4]. SIMS was performed at different deposition temperatures
Fig. 4a,b SIMS depth profiles for BST grown at different temperatures: a a 32-nm slight group-II-rich sample (group-II/Ti=1.02) grown at 655 °C, b a stoichiometric 37-nm BST sample grown at 565 °C. No sharp transition at the BST–Pt interface
in order to examine any possible interdiffusion of single atoms into the sublayers. All films were grown on standard textured (111) Pt-coated Si substrates with an additional TiOx adhesion layer below the 100-nm Pt film. SNMS investigations are presented for studying diffusion when deposition is done on Si or SiO2 surfaces without a Pt top layer. SIMS/SNMS results In Fig. 4 the SIMS depth profiles of the major constituents of the layer system are presented for two different growth temperatures. Cs+ was used as a primary ion. The following isotopes were used: 86Sr+, 137Ba+, 47Ti+, 194Pt+ and 30Si+. For the sample with the highest growth temperature (655 °C) the depth profiles are shown in Fig. 4a. At the surface is a small transient region which indicates the build up of sputter equilibrium. In the BST film the signals are constant indicating a homogeneous distribution and entering the interface region to the Pt there is a small
hump for all elements indicating an increase in sputter rate. The Ba, Sr and Ti profiles exhibit a sharp transition into the Pt film. Also the increase of the Pt at that interface indicates that this interface is well defined. Following the Pt profile it is from first sight not directly explicable that the Pt intensity increases from leaving the Pt layer entering the TiOx. This is the outcome of the matrix effects. The ion yield from an oxidized surface is by orders of magnitude higher than from a clean metal. Therefore the decrease in Pt concentration is overcompensated by the increase of the ion yield in the oxide. This results in a broad transition in the measured depth profile; of course this effect also makes the drop of intensities (BST/Pt) more pronounced. The depth profiles of the film with the lower deposition temperature (565 °C) are depicted in Fig. 4b. We observe a good homogeneity of the element concentration in the growth direction. However, the transition to the Pt interface appears very broad, in contrast to the samples processed at higher temperatures. This is attributed to the increased surface roughness of the film, as well as to its polycrystalline structure. Thus, some of the ions may have hit the underlying Pt, while some BST rest still remains. Polycrystalline grain supports this effect, since the sputter rates may differ significantly for different orientations. The increased roughness of this sample (565 °C) was also observed by SEM. The SIMS results can to some extend help in the characterization of the layer structure but there are many problems associated that makes SNMS more favourable. In SIMS we have seen the matrix effects, to some extent charging effects when approaching the thick oxide (300-nm SiO2). Figure 5 depicts the SNMS profile of the complete layer structure under Ar bombardment. The profile shows in the STO layer the presence of C to a certain amount, which is attributed to a reduced concentration of oxygen during deposition of this special sample. The Pt and TiOx layers are well resolved and the transition from SiO2 to Si is also found in the raw data. This profile indicates to some extent a diffusion of Pt and Ti into the SiO2 but this diffusion
Fig. 6a,b SNMS depth profiles of deposited STO on different substrates a SrTiOx deposited on Si and b SrTiOx deposited on SiO2; *=STO
does not penetrate the insulating layer deteriorating its functionality. SrTiO3 was deposited on different substrates and afterwards characterized by SNMS depth profiling. The depth profiles of the deposited layers on Si and SiO2 are depicted in Fig. 6. The deposition on Si (Fig. 6a) shows a minor carbon contamination of the SrTiOx layer. The interface between STO and Si is relatively broad but there is a difference between Ti and Sr indicating a stronger diffusion of Ti into Si. The deposition on SiO2 indicates a stronger diffusion of both Ti and Sr into the oxide. It should be noted that the differences in the Si intensity is due to different concentrations of Si in Si and in SiO2.
Summary and conclusion
The combination of XRF and SIMS/SNMS analysis enables the investigation of a broad spectrum of critical film properties. We developed XRF methods for the quick,
Fig. 5 SNMS depth profile of a BST film deposited on Pt/TiOx/ SiO2/Si; *=STO
non-destructive characterization of BST films. We could determine the ratio of the group-II elements to the Ti content with a precision of 0.5% for typical film thicknesses of 20–30 nm. These methods were also successfully applied on the PZT and SBT material systems. The stoichiometry and thickness of the films was determined using CSD-manufactured calibration standards. Additionally, SIMS and SNMS were used to characterize the vertical element homogeneity and the quality of the dielectric–electrode interface. If SIMS or SNMS confirms that group-II elements and Ti are in the same depth interval, the stoichiometry can be derived with high precision from the XRF data. As a result we could speed up process development of a prototype MOCVD reactor combining both analytical methods.
Acknowledgements The authors wish to thank S. Regnery for supplying the STO samples and P. Ehrhart for helpful discussions.
1. Summerfelt SR (1997) In: Ramesh R (ed) Thin film ferroelectric materials and devices. Kluwer, Boston, pp 1–42 2. Jones RE Jr, Desu SB (1996) Process integration for nonvolatile ferroelectric memory fabrication, MRS Bulletin 21:6 3. Fitsilis F et al (2001) J Eur Cer Soc 21:1547 4. Ehrhart P et al (2001) Mat Res Soc Symp Proc 655:CC9.4 5. Williams P (1992) Quantitative analysis using sputtering techniques: secondary ion and sputtered neutral mass spectrometry. In: Briggs D, Seah MP (eds) Practical surface analysis, 2nd edn, vol 2: ion and neutral spectroscopy. Wiley, New York, pp 177– 228
This action might not be possible to undo. Are you sure you want to continue?