E. A. Marquis, Department of Materials, University of Oxford, 16 Parks Road, Oxford OX1 3PH, U.K. Tel.: +44-1865-273711; fax: +44-1865-273789; e-mail: email@example.com
Atom-probe tomography analysis of complex multilayer structures is a promising avenue for studying interfacial properties. However, significant artefacts in the three-dimensional reconstructed data arise due to the field evaporation process. To clarify the origin and impact of these artefacts for a FeCoB/FeCo/MgO/FeCo/IrMn multilayer, tip shapes were observed by transmission electron microscopy and compared to those obtained by finite difference modelling of electric fields and evaporation processes. It was found that the emitter shape is not spherical and its surface morphology evolves during successive evaporation of the different layers. This evolving morphology contributes to the artefacts generally observed in the reconstructed atom-probe data for multilayer structures because algorithms for three-dimensional reconstruction are based on the assumption that the shape of the emitter during field evaporation is spherical. Some proposed improvements to data reconstruction are proposed.
Thin films and multilayer structures are important in microelectronics and data storage applications. Interfaces control the electronic and magnetic properties of materials, and as such, measuring interfacial roughness and chemistry has been the object of numerous studies using the combination of atom-probe tomography (APT) and high-resolution electron microscopy (Larson et al., 1999a, b, 2000, 2004; Petford-Long et al., 2005; Cerezo et al., 2006; Chiaramonti et al., 2008; Pinitsoontorn et al., 2008). Indeed, the combination of these two techniques offers the promise of simultaneous atomic-scale chemical and structural characterization. However, we will show that interpretation of APT data from multilayer structures is often nontrivial.
APT is based on atom-by-atom field evaporation from sharply pointed specimens with end diameters of the order of 100–200 nm. The evaporated atoms are detected as ions on a position-sensitive detector and their identities are determined by time-of-flight mass spectrometry. Spatial reconstruction of atomic positions relies on the sequential evaporation of atoms from the tip surface and utilizes a simple projection law based on a spherical tip shape approximation (Bas et al., 1995; Geiser et al., 2009b). The difficulties in analysing multilayer structures by APT stem from the combination of challenging specimen preparation and complex artefacts in the reconstructed three-dimensional (3D) data. It was recently shown that reproducible specimens can be produced by focused-ion-beam (FIB) milling methods (Larson et al., 1999a, b). Spatial positioning in 3D reconstructed data, however, remains a challenge due to constraints in the spherical-shape approximation. It is well known that crystallographic facets (Waugh et al., 1975) and differences in evaporation fields between different phases (Miller & Hetherington, 1991) lead to nonspherical morphologies. It is difficult to assess the actual shape adopted by the evaporating surface, although it would be an important input parameter into a reconstruction algorithm. Previous attempts have been made at assessing local changes of curvature for small precipitates using field evaporation modelling (Vurpillot et al., 2000). Experimentally, field-ion microscopy (FIM) observations and ring counting methods (Drechsler & Wolf, 1958) can provide accurate tip shape measurements, but remain a tedious task limited to specific systems that provide interpretable FIM images. In cases where crystallography is not observable by FIM, tip shapes can be obtained via electron microscopy approaches such as that described in (Loberg & Norden, 1969; Petersen & Ringer, 2009). However, these measurements remain challenging due to (1) potential surface modification of the samples during transfer from the atom-probe microscope to the electron microscope, (2) the very small changes of curvature being measured, and (3) unless performing electron tomography, the requirement that the feature, whose radius of curvature needs to be measured, be positioned on the equatorial plane of the tip apex with respect to the incident electron beam.
The consequences of complex tip morphologies on the accuracy of 3D reconstructions were first suggested by Vurpillot et al. (2004b), who modelled the evaporation of a multilayer structure composed of two alternating metal layers with different evaporation fields. As evaporation proceeds through the layers, the different evaporation fields of the layers produce a nonspherical specimen shape, which produces variations in hit densities on the detector due to magnification variations. These effects result in spatial positioning errors in 3D reconstructions, which have been observed experimentally (Vurpillot et al., 2004a).
Most previous studies were performed using small-field-of-view instruments, where the extent of the analysed region was limited to ∼20 nm × 20 nm. In these instruments, the small angle approximation commonly used in the derivation of reconstruction algorithms (Bas et al., 1995) and the spherical-shape approximation for the specimen surface lead to reasonably accurate reconstructions. However, the problem is more pronounced in wide-angle atom-probe instruments because of the increased probability that the analysed surface contains a highly nonspherical region. In these cases, the spherical approximation can result in obvious deviations in position well beyond the instrumental resolution. Focusing on tunnelling magnetoresistance (TMR) multilayer structures, it is shown here that the evolution of the tip shape can be followed both experimentally and theoretically via modelling. The detailed understanding of field evaporation artefacts not only provides a basis for developing more accurate 3D reconstruction approaches, but also allows confident data analysis and interpretation of local concentration measurements.
Materials and methods
Films of Ta(10)/IrMn(40)/CoFeB(0.35)/CoFe(5)/Mg(0.8)/MgO(2.7)/CoFe(2)/CoFeB(30) (numbers in parenthesis correspond to nominal film thickness in nanometres) were grown by sputter deposition onto Si/SiO2 substrates. Nominal concentrations of the alloy layers were Co–21Fe–30B, Co–30Fe and Mg–50O with the concentrations given in atomic percent. Further details of the growth and properties of this structure can be found in (Larson et al., 2010). All analyses performed in this study were on as-grown specimens. Atom-probe specimens were prepared using standard FIB lift-out methods (Thompson et al., 2007) and mounted on W needles swaged to Cu half disks using the Omniprobe Shortcut™ (Gorman et al., 2008; Prosa et al., 2009). The end-form of the W needles were FIB-milled flat to a ∼2 μm diameter to accept the lift-out wedge. Final Pt-capped wedge dimensions approximated a 2.5 μm equilateral triangle with a lateral dimension of 1.5 μm. A four-step annular milling procedure was adopted to produce a repeatable final end-form. The first three annular milling patterns were performed with 30 kV Ga ions at 0.28 nA beam current. Each milling pattern had a 4 μm outer diameter with blanked inner diameters of decreasing size (1.6, 0.7 and 0.3 μm) to produce a tapered 300 nm end-form. A 5 kV clean-up step was applied last to remove any 30 kV damage (Thompson et al., 2007), remove any capping material remaining from the previous preparation steps and position the multilayer films near the surface of the tip with a consistent tip dimension (∼150 nm diameter). Backside specimens, where the layer sequence is reversed to the atom-probe analysis direction becoming IrMn3/FeCo/MgO/FeCo/FeCoB along the tip axis (Fig. 1a) were also analysed to investigate the effect of the layer sequence on the evaporation process. These specimens were prepared following the method described in (Prosa et al., 2009). A JEOL 3000FX microscope was used for the observation of tip shapes by transmission electron microscopy (TEM). Field evaporation was performed using a LEAP® 3000XHR instrument operated in laser mode. The sample was maintained at 50K in UHV conditions and evaporated at constant detection rate (0.002 atom per pulse). Laser pulsing was done at 200 kHz with a wavelength of 532 nm, a spot size of ∼10 μm (4σ) and a spot energy between 0.3 and 0.7 nJ.
The evaporation behaviour of the multilayer structure was modelled using a finite difference method similar to that described in Vurpillot et al. (1999) and Geiser et al. (2009a). In these simulations, a tip is constructed with an initial spherical geometry for simplicity. The crystal lattice size was 0.4 nm and the resolution of the simulation lattice was twice that of the crystal lattice. The simulation lattice itself was 512 cells on a side, for a simulation volume of 102.4 nm on a side. The initial height of the tip was 51.2 nm. The crystal lattice of all the specimen layers in the simulation was FCC with the 001 direction aligned to the vertical axis of the simulation volume. The FCC structure was chosen for simplicity even though the FeCo layers have a BCC structure. The overall results of the simulations are however independent of the exact crystal structure. Note that these choices allowed for a match of the crystal lattice into the simulation lattice with no aliasing errors.
The simulation is 3D sourceless electrostatic potential solver. Potentials are specified at the walls and on the atoms that compose the tip. Because the atoms in the tip are all assumed to be at the same potential, it is equivalent to the simulation of an ideal metal tip. See Vurpillot & Bostel (2001) for a description of a very similar simulation system. In our software, the electrostatic solution is computed using standard multigrid techniques with a Gauss–Seidel smoother at the coarsest scale (Hackbusch, 2003).
To create hit data, atoms are propagated through the computed potential field using a Runge–Kutta integration technique with adaptive step-size (Cash & Karp, 1990). When trajectories reach the top of the simulation volume, they are linearly projected to the detector plane. At the detector plane, ions may be discarded according to an efficiency function or if they miss the detector. Positions are converted to equivalent time to digital converter (TDC) counts. No degradation of the registered positions is simulated, other than time digitization.
The voltage assigned to an evaporated atom is determined by the ratio of the specified evaporation field for the material to the observed local field at the point of evaporation. The final assigned voltage is smoothed over a sequential group of atoms. Some evolution of the voltage curve should be expected to be due to the shrinking of the tip in a relatively small simulation volume. Note that the limitation of a nonequilibrated initialization for the top layer of FeCoB is reduced by increasing the depth of the region of interest. This allows the spherical shape to equilibrate before the deeper layers are exposed.
The ratios of the different layer thicknesses are very similar to the experimental thicknesses (Fig. 1). The evaporation fields of Co and Fe were set equal, considering the theoretical values of Fe = 33 V nm−1 and Co = 37 V nm−1 (Tsong, 1978). The evaporation fields of B and MgO (modelled as one molecule) were set to five-thirds and two-thirds of the Fe/Co evaporation field, respectively. Modelling the effect of a dielectric material on the exact field distribution at the surface of an emitter is a complex problem that should be addressed elsewhere. To simplify the problem, the behaviour of the thin oxide layer was modelled using an effective lower evaporation field, therefore treating it as a metal with MgO as the evaporating species. This simplification focuses the analysis on the effect of varying evaporation fields across layers, on tip morphology and resulting reconstruction quality. The evaporation field of Ir and Mn was set equal to that of Fe and Co, although the theoretical value for Ir deviates significantly (Ir = 44 V nm−1; Tsong, 1978). Note that the evaporation fields of alloys (such as FeCo and IrMn3) are generally unknown.
Results and discussion
Reproducibility of experimental initial tip shape
In previous APT analyses of structures containing oxide layers, poor yield often constrained attempts to study repeatability of results. In this work, the combination of reproducible specimen preparation methods (Thompson et al., 2007) and laser pulsed APT (Bunton et al., 2007) has resulted in very good yield (typically greater than two in every three specimens). It is important to note that the experimental results on tip shape and field evaporation behaviour presented hereafter are therefore reproducible within the constraints of identical experimental conditions, as illustrated in Fig. 2.
A typical voltage curve during field evaporation is shown in Fig. 3(a) and is accompanied by a nonuniform hit density observed on the detector (Fig. 3b). During the evaporation of the multilayer stack, the evolution of the applied voltage (adjusted in real time to maintain constant evaporation rate across the entire detector) systematically reveals a decrease when the MgO layer completely evaporates from the tip surface. The MgO layer in particular first appears during its evaporation as a very dense shrinking ring (Fig. 3b) on the periphery of the region defined by the atoms evaporating from the top FeCo layer. Figures 3(c) and (d) show a theoretical voltage curve along with a calculated desorption map taken during the simultaneous evaporation of the FeCoB, MgO and FeCo layers. Comparing the evaporation behaviour of the experimental data versus the simulated data, the differences between the starting experimental (Fig. 2) and simulated tip shapes (Fig. 1b) do not appear to affect the overall evaporation behaviour. Moreover, the drop in voltage observed both experimentally and theoretically validates the choice of a lower evaporation field for MgO as compared to that of the surrounding metals.
Tip shape during evaporation
Considering a simple projection law, variations in tip curvature produce variations in magnification which result in heterogeneous ion hit densities at the detector. In the case of the multilayer structure, an inhomogeneous hit density is observed on the detector, particularly during the evaporation of the MgO layer. This layer evaporates simultaneously (although at different rates) with the top of the FeCo layer, causing the specimen to be nonspherical. A similar effect was shown by Vurpillot et al. (2004a), who simulated the evaporation of Cu/Co multilayers, and showed that the specimen shape changes significantly depending on whether the Cu or the Co layer is at the specimen apex. To demonstrate the nonspherical nature of the specimen end form during evaporation, we interrupted the field evaporation analysis after partially evaporating the multilayer structure to observe the tip shape by TEM. Note that the delicate handling required to transfer specimens from one instrument to another and possible oxidation make this experiment very challenging. Figure 4 shows that when the MgO layer is nearing the apex of the specimen, the tip shape has a complex morphology with a small radius of curvature in the centre and much larger effective radii for the FeCo/IrMn underlying layers.
The experimentally observed shapes were modelled and reproduced by finite-difference calculations of the field evaporation process (Geiser et al., 2009a). Figure 5 shows the evolution of the tip shape during the evaporation of the different layers. It is particularly evident that the tip shape is aspherical and is controlled by the different evaporation field of the different layers. The top FeCo layer is present with a high local radius of curvature at the very apex, as shown in Fig. 5e) and once it has completely evaporated, the MgO layer evaporates very quickly (transition from shape shown in Figs. 5(e)–(f)) due to its small radius of curvature and lower evaporation field.
A spherical tip morphology during the evaporation of the entire layer stack would yield constant magnification over the entire field of view. It would then be possible to reconstruct the entire dataset with flat layers and accurate thicknesses. In this case, however, because of the different evaporation field of the different layers, the tip shape is constantly evolving with varying local radii of curvature. This results not only in varying magnification over the surface of the specimen, but also out of sequence of evaporation of the atoms when compared to a tip that maintains a spherical end form. Because the z-component of the reconstructed atomic positions is dictated by the sequence of evaporation, edges of some of the layers are reconstructed with z-coordinates that are erroneously high. Under the constraint of a spherical tip shape, the reconstruction can thus exhibit an artificial bowl shape and inaccurate layer thicknesses as seen in Fig. 6(a) for the simulated data and Fig. 6(b) for the experimental data. Density variations also result from this nonsequential evaporation: the MgO layer appears compressed towards the top FeCo layer whereas a low-density region is observed at the top of the lower CoFe layer and on the sides of the reconstructed volumes. Although correction methods such as a z-coordinate rescaling were suggested to improve the reconstruction of multilayers (Vurpillot et al., 2004b), these methods alone are not sufficient to entirely correct the curvature and scaling of this complex set of layers for a wide field of view data set.
In the context of the significant differences in evaporation field and projection patterns of the different phases, let us consider their effects on the measured concentration profiles. A representative 1D profile is shown in Fig. 7(a). Certain artefacts are evident when comparing this profile to an electron energy loss spectroscopy profile obtained on the same structure.
First, according to the nominal growth conditions, there is a thin B-free layer at the top of the MgO layer that is nearly absent in the APT data. The B-free layer is absent in the experimental (Fig. 7a, black line) and significantly compressed in the simulated data (Fig. 7c). This apparent absence can be understood in terms of evaporation field differences and therefore variations in evaporation rate. As shown by the simulations, the evaporation shape of the top FeCo is greatly affected by the evaporation of the preceding FeCoB layer with a higher evaporation field. The effective reduction of the radius of curvature as evaporation proceeds from the FeCoB layer to the FeCo layer (shown by Figs. 5c and d) implies a magnification increase (which is not taken into account in the current global reconstruction) and thus the thin B-free layer will appear significantly compressed in the depth direction. Another smaller, but possible, contribution not taken into account in the field simulations could come from the high evaporation field of B as compared to that of Fe and Co, leading to some retained B atoms on the surface. In the top-down geometry, the B atoms of the top FeCoB layer may evaporate at the same time as the B-free FeCo layer underneath. The FeCo layer would then appear thinner than its original thickness. In the backside geometry, the surface Fe and Co atoms of the FeCoB layer will evaporate first causing B atoms to be reconstructed slightly below their original positions and therefore the B-free layer may appear thicker than its true thickness. The analysis of atom-probe data (Fig. 7) and evaporation simulations (Fig. 8) of specimens in a backside configuration confirm the importance understanding the effect of specimen shape on the evaporation behaviour. In the backside geometry, the thin B-free FeCo layer is underneath (with respect to the order of field evaporation) the MgO layer (which has a lower evaporation field), but above the FeCoB layer (which has a high evaporation field). Both experimental and simulated concentration profiles obtained from the backside geometry (Figs. 7b and d) show the presence of the B-free FeCo layer. Because of the reverse sequence of layers with increasing evaporation fields, the FeCo layer is less magnified and therefore the reconstruction is stretched in depth.
Secondly, the atom-probe data indicate a relatively sharp upper MgO interface and mass spectra suggest the a transition-metal oxide layer at the bottom of the MgO layer (Fig. 7a) with the presence of FeO, FeCoO, MnO and CoO peaks. The relative amount of oxygen present in the transition metal oxide complexes is shown in Fig. 7(a) and appears at the bottom of the MgO layer. However, the presence of oxidised states of Fe and Co is not observed by electron energy loss spectroscopy analysis (Larson et al., 2010). If an atom-probe specimen is prepared in a ‘backside’ orientation, the top (with respect to the growth direction of the layers) MgO interface that was previously sharp now appears broader (Fig. 7b) and the oxygen associated with the transition metal oxide peaks are now located at that interface. Although the profiles in Figs. 7(a) and (b) might be expected to be mirror images of each other, this is not observed to be the case for the selected datasets. A potential reason for this is the variation of interface integrity (sharpness) and variation of transition metal oxides regions with respect to the ‘XY’ plane (e.g. within the regions used to calculate composition profiles in the plan normal to the profile direction). The fact remains that the transition metal oxide layer systematically appears ‘below’ the MgO in upside specimens and above in backside specimens.
There are several possibilities for this extra layer appearing at the MgO/FeCo interface when MgO evaporates before the FeCo layer. Adsorption (and subsequent field evaporation) of residual vacuum gases would lead to apparent high oxygen concentration. Adsorption would occur onto the low-field (and thus flattened) region that is formed near the specimen shank both due to FIB milling of differing sputter rate species (such as the shapes experimentally shown in Figs. 2 and 4) or due to the evaporation from a material of low evaporation field species (MgO) into a higher field species (CoFe). However, the 3D distribution of hydrogen does not correlate with this interface, and therefore the adsorption of various residual gas species appears unlikely. Another possibility is preferential diffusion of (electronegative) oxygen away from the field-evaporated surface subsequent to an ionization event that results in evaporation of Mg prior to oxygen due to the polarization of the MgO bonded pair. It has been reported in previous APT studies (Chiaramonti et al., 2008) and TEM and X-ray diffraction methods (Meyerheim et al., 2002) that the bottom FeCo layer may be partially oxidized under certain growth conditions. The atom-probe analysis in one direction alone cannot prove unequivocally whether the extra oxidized layer results from growth or from evaporation artefacts. In this work, analysing the multilayer stack in the opposite direction straightforwardly answers this question.
Considering the large error bars (55 ± 10 at.% Mg–45% O), it is difficult to assess whether the MgO layer is under- or overoxidized. When taking into account the oxygen atoms in the transition metal oxide peaks, the concentration becomes 44 ± 10 at.% Mg–57% O. Unfortunately, the large error bars prevent the mechanisms of oxygen retention or adsorption to be unequivocally determined.
The two artefacts discussed earlier result from the convolution of differences in evaporation fields (possible B retention), physical processes of evaporation leading to apparent short-range diffusion (oxygen), as well as variations in evaporation rate due to shape changes. Through field evaporation modelling, tip shape observations and backside analysis, the nature of these artefacts can be understood and solutions proposed.
This work clearly shows that the assumptions of spherical morphology and steady-state specimen shape used in traditional reconstruction algorithms are not satisfied for the complex morphologies observed in the current multilayer specimens. This fact must not be ignored when evaluating the results of the standard reconstruction methods and the composition measurements that arise from such reconstructions. We observe that it may be necessary to support sizeable variation in reconstruction parameters based on the lateral position on the detector and evaporation sequence. In particular, correlating launch positions of simulated atoms with their final detected positions relative to the high-density and low-density regions may be a useful tool in predicting proper projection parameters. We anticipate that if a strictly projection-based reconstruction is performed, variations in both surface profile and projection-point will be required to optimize the result. Although the global reconstruction is obviously inaccurate, it is of note that subnanometre spatial resolution is observed in some parts of the data, particularly near the centre of the reconstruction, the only region analysed with small field of view instruments. Figure 6(c) shows the atomic planes (∼0.14 nm spacing) are resolved in the bottom FeCo layer.
Multilayer structures were analysed by APT. As anticipated, some artefacts were observed in the reconstruction quality (density variations) and compositional profiles (B-free thin layer), which are direct consequences of the reconstruction algorithm used on the data. Electron microscopy images of specimens revealed that shapes that are far from the spherical morphology assumed in the standard reconstruction algorithm. This work demonstrates the necessity to improve the accuracy of reconstructions for the atomic-scale analysis of materials. Electron microscopy observations of tip shapes along with modelling of field evaporation appear to be critical to the development of new algorithms, being employed, for instance, as major components of a more sophisticated reverse projection approach. It is important to note that although the starting specimen end form produced by FIB can be systematically reproduced for this study, it may not be the optimum shape for the most accurate 3D reconstructions. Depending on the approach adopted for reconstructing the APT data, different end form shapes may be needed. We also suggest that the combination of the information contained in the detector density maps together with simulation could be used to provide initial and evolving tip shape information, resulting in a more accurate final APT reconstruction.
The authors thank Drs. Philip Rice and Stuart Parkin, IBM, Almaden, CA, for supplying materials and for helpful discussions. EAM acknowledges the Royal Society for financial support, and the authors acknowledge Drs. Thomas F. Kelly, Jesse D. Olson and colleagues at Imago Scientific for valuable discussions.