Introduction

Nanoscale resistive random-access memory devices (RRAM) based on metal-oxides, also known as memristors, have great prospect in becoming a mainstream memory technology, due to their infinitesimal dimensions, fast switching and capacity to store multiple bits of information per element1,2,3,4,5,6. Such devices have the ability to toggle their resistance between high resistive state (HRS, or ON) and low resistive state (LRS, or OFF) in either a digital or analogue fashion7. TaOx8 and HfOx9 are of particular interest for RRAM due to their high-k properties and compatibility with deep submicron complementary metal-oxide semiconductors (CMOS) technologies and their simple reduction dynamics; the ability to toggle between two phases only, makes them easy to control while at the same time enhances their state stability. On the other hand, TiOx, one of the most celebrated metal oxides, provides wider opportunities for multi-state memory capacity due to the intrinsic variety of possible chemical phases. This complexity however makes its control more difficult, also prohibiting the rigorous study of the unrelying physical mechanisms. Whilst the electrical behaviour of TiOx based memristors has been widely investigated, much less is known about the atomic-scale changes that occur in the film as a consequence of switching10. It has been suggested that the formation and migration of defects within the oxide active layer induce the formation of reduced phases in form of conductive filaments that render resistive switching. However, direct localization and chemical characterization of such conductive phases remains a challenge11.

Previous work by Kwon et al.12 based on high resolution transmission electron microscopy (HRTEM), identified Ti4O7 as the structure of conductive filaments in TiO2 based-devices by interpretation of electron diffraction patterns. However, HRTEM requires the use of high energy electron beam (200–300 kV) which could induce sample crystallization and effectively alter the physical-chemical state of the film. Furthermore, electron diffraction requires the crystals to be aligned in specific orientations with respect to the beam for unambiguously capturing distinct diffraction patterns12. In this work, we employed soft X-ray spectromicroscopy for measuring Near-Edge X-ray Absorption Fine Structure (NEXAFS) spectra at high spatial resolution in a full-field Transmission X-ray Microscope (TXM). This technique allowed us to perform simultaneously imaging and spectroscopy for investigating morphological changes in the film as well as performing chemical analysis at nanometric scale of localized regions, respectively. Our set-up differs substantially from previous transmission X-ray microscopy studies of TiO2-based13,14 or SrTiO3-based15 RRAM devices, which are based on irradiating the device from the top electrode, alleviating the challenges imposed for accessing the TiOx film through the capping top electrode. When compared to electron energy loss spectroscopy (EELS), TXM-NEXAFS provides more analytically useful information per unit radiation damage16. Moreover, identification of distinct phases using NEXAFS is based on fingerprints methods, avoiding the difficulties imposed via multiple scattering effects and/or the presence of diffraction spots from other TiO2 phases and/or metallic electrodes when employing electron diffraction.

This technique has allowed us to spatially resolve distinct TiOx phases that macroscopically assemble in a conductive localized area within a RRAM prototype that has been switched to a LRS. Presumably, this conductive spot can be regarded as a conductive filament. However, as reversibility of switching in this region cannot be proved after cutting the lamella for the experiment, dielectric breakdown or Joule heating could also be responsible for the formation of the reduced area. We prove that the conductive path is located right underneath a protrusion of the top electrode and it is composed by a reduced TiOx phase with a O/Ti ratio close to 1.37. Crystalline TiO2 rutile phase and an additional orthorhombic-like TiO2 phase regions were also identified in a localized area between the TiOx reduced phase and the top electrode, suggesting that the temperature due to Joule heating increases up to 1000 K in this area17. A Ti-containing phase similar to the TiOx reduced phase was also identified in both the top and bottom Pt electrodes18.

Results and Discussion

The studied prototypes are Cr(3 nm)/Pt(30 nm)/TiOx(50 nm)/Pt(30 nm) stacks (Fig. 1a), fabricated as stand-alone devices with a junction of 40 × 40 μm2 (Fig. 1b,c) (see Methods). The as-deposited TiOx film was determined to be slightly reduced by X-ray photoelectron spectroscopy (Supplementary, Fig. S1). Prior the TXM-NEXAFS investigation, all devices were electrically characterized using voltage sweeping. In Fig. 1d, the current-voltage (I-V) characteristic of a pristine state device (PRI) is shown; note the extremely low conductance level. The device is subsequently electroformed (inset of Fig. 1e) before being switched to LRS (Fig. 1e main).

Figure 1
figure 1

Device architecture and electrical characterisation of solid-state RRAM devices.

(a) Schematic of the device stack structure. (b,c) Optical images of TiOx RRAM stand-alone devices. (d,e) Electrical characterization of devices before TXM measurement: (d) I-V curve of pristine device showing an initial high resistive state; (e) LRS state device (Inset: electroforming step).

Full size image

TXM-NEXAFS measurements were performed on a cross-section of each of the two devices cases, i.e. PRI and LRS. An outline of the method is briefly described in Fig. 2. The lamella is extracted from the red region highlighted in Fig. 2a using the Focused Ion Beam (FIB) lift-out technique (see Methods). The thin lamella (Fig. 2b) is positioned perpendicular to the X-ray beam direction for absorption measurements and imaging as shown in Fig. 2c. A schematic of the TXM experiment is shown in Fig. 2d.

Figure 2
figure 2

Experimental set-up.

(a) SEM image of pristine device active junction. (b) SEM images of final thin lamella. (c) Lamella orientation for the TXM-NEXAFS experiment. (d) Schematic of the TXM for NEXAFS experiment.

Full size image

For each of the two cases, PRI and LRS, two sequences (stacks) of X-ray images were acquired at closely spaced photon energies using a transmission X-ray microscope, in the Ti 2p (450–485 eV) and O 1 s (525–555 eV) energy range. Each stack, consisting of 1053 images over the Ti 2p energy range and 351 images over the O 1 s energy range, was carefully aligned using a cross-correlation iteration process until the image shift was less than ±0.6 pixels (±3 nm) across the entire energy range (see Methods). After conversion of all stacks into optical density (absorbance), a detailed analysis involving extraction of the X-ray absorption spectra of specific regions of interest and conversion to component maps using spectral fitting or multivariate statistical analysis procedures was performed19,20.

X-ray images of the lamellae for the PRI case at 450 eV (below the Ti 2p edge) and 465 eV (on the strongest Ti 2p absorption peak) are presented in Fig. 3. The individual layers comprising the device stack i.e. Si wafer, SiO2, Pt bottom electrode (BE), TiOx active film and Pt top electrode (TE) can be clearly distinguished in Fig. 3a; the TiOx active layer appears dark as Ti does not absorb at this energy. At 465 eV, the TiOx layer appears bright due to the strong absorption of Ti (Fig. 3b). Spatially localized NEXAFS spectra at the Ti 2p and at the O 1 s were then extracted from the TiOx film area. Two representative spectra are reported in Fig. 3c,d, respectively. The fine structure of Ti 2p spectrum is complex but at a coarse level can be considered to consist of four main peaks (two doublets). The first doublet (2p3/2) occurs within 457–462 eV while the second doublet (2p1/2) appears in the range 462–468 eV (see Supplementary, TXM-NEXAFS section for details)21. The Ti 2p spectrum exhibits distinct features for the different polymorphs phase of crystalline TiO2 and for amorphous TiO222. In particular, peaks of amorphous TiO2 are quite broad due to structural disorder caused by a range of bond angles and lengths, in contrast to spectra of crystalline samples which are characterized by sharp peaks. The Ti 2p spectrum shown in Fig. 3c is consistent with amorphous TiOx22,23. This result is in agreement with a TiOx thin film prepared via reactive sputtering and indicates damage/crystallisation was not caused by the sample preparation process. In agreement with our observations at the Ti 2p, the O 1 s spectrum shown in Fig. 3d, possesses typical signatures of disordered systems22, demonstrating that the TiOx film is amorphous all along the cross-section.

Figure 3
figure 3

Optical density TXM micrographs of the PRI case obtained using X-ray photon energy of 450 eV (a) and 465 eV (b) and Ti 2p (c) and O 1 s (d) NEXAFS spectra extracted from the TiOx film. Scale bar = 200 nm.

Full size image

Notable differences are however observed in the film after switching to LRS (Fig. 4). Optical images (Fig. 4a and Supplementary, Fig. S2) and AFM images (Fig. 4b) viewed from the Pt TE show the formation of a protrusion about 140 nm high at the top left rim of the TE. Several studies have shown that protrusions of the TE could indicate possible critical regions of the film responsible for the RS switching24,25. The LRS lamella cut across the TE defect (Fig. 4c), presents some electron transparent areas in the damaged region (Fig. 4d) indicating the presence of a very low absorption area. Under the identified protrusion, the TE is bulged and discontinuous, having the shape of a dome (Fig. 4e). The region below this dome is formed by a main open cavity with two additional smaller ones towards the region outside the junction. The TiOx layer is dislocated towards the region where the TE is discontinuous. The dome-like empty area suggests that the TE protrusion was caused by O2 gas escaping through the weakest point of the TE, the rim, in agreement with previous observations24,25. These features are similar to those reported in ref. 11.

Figure 4
figure 4

(a) Optical and (b) AFM images viewed from the TE. Inset in (b) line profile of the protrusion. (c) SEM image of lamella cut location and in inset detail of lamella extraction. (d) Thinned lamella with details of the damaged region. (e) Cross-section SEM image of the damaged region. Scale bar (a–c) 10 μm; (d) 1 μm; (e) 200 nm.

Full size image

The X-ray cross-section image of the LRS case recorded in the region underneath the TE protrusion at the energy of 450 eV is reported in Fig. 5a. The Pt/TiOx/Pt stack can be clearly distinguished on the right side of the image (Pt BE/TE are bright and TiOx is black) and appear unaffected. In the defect region, the BE remains unaffected whereas the TE is discontinuous and delaminated from the TiOx film. At 450 eV, the Ti containing regions cannot be distinguished from void zones as both appear dark. By examining the Ti 2p image sequence at different energies, we can spatially resolve the location of regions containing Ti species as the contrast of the image changes in correspondence with changes in the Ti absorption peak intensities. The image at 458 eV (Fig. 5b) shows highly absorbing regions containing TiOx appearing with brighter contrast; the regions that appear dark at 450 eV but bright at 458 eV correspond to Ti containing areas. Regions that remain dark at 458 eV are clearly defined as voids. It is interesting to note that three highly localized (filamentary) Ti containing regions, (indicated as A, B and C – see Fig. 5b) link the BE and TE. In particular, the magnified image of Fig. 5c, shows that the TiOx layer aggregates in a ~100 nm wide localized region underneath the highest deformation of the TE (C). These observations are consistent with the X-ray coloured image shown in Supplementary, Fig. S3 where the areas containing only Ti species are shown in red.

Figure 5
figure 5

Physicochemical characterization of a RRAM device programmed in LRS.

Optical density TXM micrographs of LRS device obtained with photon energy of 450 eV (a) and 458 eV (b,c). NEXAFS Ti 2p (d) and O 1 s (e) spectra extracted from the regions circled in the X-ray images b and c. Scale bar (a,b) 150 nm; (c) 100 nm.

Full size image

By extracting NEXAFS spectra in the regions where Ti species exist, chemical identification of the species can be performed as NEXAFS spectra are specific for each TiOx phase. 6 regions of interest (ROIs) were identified, indicated in Fig. 5b,c by numbered circles. The Ti 2p and O 1 s X-ray absorption spectra extracted from each of these localized regions are reported in Fig. 5d,e, respectively. There are significant variations in the Ti 2p spectral shapes clearly indicating significant changes in the chemical structure of TiOx. Peak splitting and changes in relative intensities of the Ti 2p peaks can be attributed to changes in symmetry around Ti. In particular, the (2p3/2, eg) orbitals at the Ti 2p point directly towards the 2p orbitals of the surrounding O and are therefore very sensitive to the titanium local environment (Fig. S4a)22,26.

These spectral differences were then employed to spatially resolve different TiOx phases across the TiOx film.

Spectra extracted from undamaged areas of the film far away (ROI_1) and close (ROI_2) to the identified defect, are similar to each other. They show broad and smooth spectral features, higher intensity of the (2p3/2, eg) peak compared to that of the (2p3/2, t2g) peak27 and absence of splitting of the (2p3/2, eg) peak around 460 eV; all these features indicate a high structural disorder as expected for amorphous TiOx22,23. The spectra extracted in the defect regions, ROI_3, ROI_4 and ROI_6, are quite different and characteristic of crystalline TiO2. A notable finding is the splitting of the (2p3/2, eg) peak, which indicates distortion of the TiO6 octahedra28. The relative intensities of these two peaks depend on the particular type of octahedral distortion, therefore on the particular polymorphic form of TiO2 (see Supplementary, Fig. S4a for details)29.

In the Ti 2p spectrum of ROI_3, the intensities of these two peaks are comparable. This particular (2p3/2, eg) profile has been reported for orthorhombic-like phases such as TiO2-II30 and brookite31. The profile of ROI_3 region is very similar to that of the TiO2-II phase reported in30, an usually high pressure phase which has recently been grown at ambient pressure by atomic layer deposition as 50 nm thin films32. In the spectrum of ROI_4, the relative intensities of the split (2p3/2, eg) peak are different, being higher for the peak at higher energy. This is typical of TiO2 rutile phase which has tetragonal distortion of the TiO6 octahedra22,31,33. Assignment of this spectrum to the rutile phase is also supported by the fact that the crystal-field splitting of the 2p1/2 peak is 2.4 eV28. Finally, in the spectrum of ROI_6, the relative intensity of the split (2p3/2, eg) peak is opposite to that in rutile, being the intensity higher for the peak at lower energy. This profile is typical of anatase. Moreover, the crystal-field splitting of the 2p1/2 peak is 1.9 eV, lower than the one observed in rutile and typical of the anatase phase30,28. Two additional low intensity peaks appear in ROI_3, ROI_4 and ROI_6 at 456.6 and 457.2 eV (indicated as a and b in Supplementary, Fig. S4a) are also observed in rutile, anatase22 and TiO2-II30. They are atomic multiplet features related to exchange interaction between 3d electrons and the 2p core hole21 which tend to be very similar in different species, but may be modified in lower symmetry environments or when there are other interactions affecting the 3d partial density of states34.

The spectrum of ROI_5 is quite peculiar and represents the predominant species observed underneath the TE defect. The intensity of (2p3/2, t2g) peak in the Ti 2p spectrum extracted from ROI_5 is lower than that in the spectrum from ROI_3, ROI_4 and ROI_6, indicating lower local symmetry at point 5. The low energy shoulder around 456 eV is consistent with the presence of a partially reduced phase containing Ti3+ 13,27. This is also confirmed by the fact that the peaks are slightly broader compared to ROI_3, ROI_4 and ROI_6, as inferred from the reduced 2p3/2 and 2p1/2 peak to valley ratio (see Supplementary, Fig. S4a for details)35. Spectra of phases containing Ti3+ show broader features compared to the pure Ti4+ phases because of the binding energies of the Ti3+ levels and overlapping and increasing numbers of allowed transitions in the Ti3+ spectra36,37. Formation of Ti3+ species underneath the defect can be explained by the following reaction:

O2 gas accumulates below the TE and erupts from the weakest part of the film, causing the physical dome-like deformation observed at the rim of the TE38. The edge of the TE has been suggested recently to be the preferred reduction site since it is a three-phase contact site (oxide, metal and ambient atmosphere)39.

Formation of Ti3+ upon switching has been previously reported by Strachan et al.13. In this work, Scanning Transmission X-ray Microscopy (STXM) was used for phase identification of nanoscale channels underneath TE defects of Pt/TiO2/Pt devices. It is interesting to note that whereas the Ti 2p spectrum measured underneath a defect in a region near the centre of the device shows features similar to Ti4O7, the corresponding spectrum underneath a defect located at the rim of the top electrode of a LRS device, is quite similar to the one observed in the present work, with a low energy shoulder around 457 eV and a minor broadening of the 2p3/2 and 2p1/2 doublets.

The corresponding O 1 s X-ray absorption spectra extracted from the same localized regions used to extract the Ti 2p edge spectra are reported in Fig. 5e and Supplementary, Fig. S4b. Variations between the selected ROI_s are less evident at the O 1 s as the change in structure of TiOx has less influence on O 1 s than Ti 2p spectra40. Nevertheless, our analysis of O 1 s spectra is in good agreement with that of Ti 2p. O 1 s spectra of undamaged areas ROI_2 (and the similar ROI_1) show broad and smooth spectral features typical of amorphous TiOx23. Moreover, the dip between the (O 1 s, t2g) and (O 1 s, eg) peaks is shallow compared to the spectra of regions from ROI_3 to ROI_6 and their energy difference, which is often used to evaluate crystal field splitting, is smaller and typical of disordered materials (2.0 eV)26. In the spectra extracted from ROI_3 to ROI_6, the eg and t2g peaks appear more defined and intense. The energy gap between eg and t2g in ROI_4 and ROI_6 is ~3.0 eV, a value reported for both anatase and rutile phases22. This gap is slightly smaller in ROI_3.

In order to better visualise the chemical distribution of the main TiOx phases across the LRS case, the aligned Ti 2p and O 1 s image sequences on an optical density scale were converted to chemical components maps by means of singular value decomposition (SVD) procedures41. The black and white chemical maps at the Ti 2p (Fig. 6a,b) and O 1 s (Fig. 6c,d) were obtained using the two predominant components in the film, namely amorphous (corresponding to the spectrum of ROI_1 and ROI_2) and reduced (corresponding to the spectrum of ROI_5) TiOx phases. The brightest regions correspond to TiOx amorphous in Fig. 6a,c and TiOx reduced in Fig. 6b,d. The fit of the two components gives consistent results at the Ti 2p and O 1 s edges. In the unaffected regions, TiOx remains amorphous in between the electrodes whereas the reduced TiOx phase is localized in a well-defined region underneath the TE protrusion (C) and in the two thinner filamentary regions connecting TE and BE on the left side of the protrusion (A and B).

Figure 6
figure 6

Chemical maps depicting regions of amorphous and reduced TiOx.

Components maps at the Ti 2p (a,b) and at the O 1 s (c,d). Scale bar = 100 nm.

Full size image

The spatial correlation of amorphous and reduced TiOx phases is better visualized by merging the two components in single color-coded composition maps. Spectra of these components at the Ti 2p and O 1 s are respectively shown in Fig. 7a,b and were used to generate the color-coded composition maps reported in Fig. 7d,e, with blue regions corresponding to amorphous and green to reduced TiOx.

Figure 7
figure 7

Physicochemical characterization of a RRAM device programmed in LRS

. Color-coded composition maps at the Ti 2p and O 1 s generated independently (d,e) and simultaneously (f) by using the absorption spectra of the two components, TiOx amorphous (blue) and TiOx reduced (green) shown in (a–c). Scale bar = 100 nm.

Full size image

Localization of the two phases is apparent with the TiOx amorphous in the unaffected film areas whereas the reduced TiOx is localized mainly below the highest protrusion of the defect (C) and in the regions A and B. As reduced TiOx has increased conductivity compared to the amorphous state, this is a direct observation of the area responsible for the LRS state of the device. Our observations are further validated by appending the Ti 2p and O 1 s stacks of the same ROI area (Supplementary, Fig. S5) and by appending the Ti 2p and O 1 s spectra into a single combined spectrum for each of the amorphous and reduced TiOx regions (Fig. 7c). The spatial distributions of the amorphous and reduced TiOx regions shown in Fig. 7f are in excellent agreement with those derived from considerations of the Ti 2p or O 1 s edges independently, supporting the validity of our analysis and giving further evidence of the highly reduced TiOx percolation.

Further analysis was performed by generating chemical components maps exploring various combinations of reference spectra extracted from the image sequences. For example, the colour-code composite map at the O 1 s shown in Supplementary, Fig. S6b, was obtained by including in the fitting the O 1 s spectrum of the SiO2 layer below the BE (Supplementary, Fig. S6a). A satisfactory final fit of the Pt(BE)/TiOx/Pt(TE) cross-section was obtained by adding a third component in addition to amorphous and reduced TiOx (Supplementary, Fig. S7). The Ti 2p spectrum of the third component (represented in red in Supplementary, Fig. S7a and S7b) is similar to that of the reduced phase but exhibits a smaller 2p1/2 crystal field splitting and it has a significant lower intensity compared to the reduced TiOx. The O 1 s spectrum of the third component is quite different from that of reduced TiOx. Combined color-coded maps obtained independently at the Ti 2p and O 1 s (Supplementary, Fig. S7c and S7d), show that the third phase, represented in red, is clearly present only in the TE and BE regions. This also explains the observed lower intensity in comparison to the other two components. Combined fitting at Ti 2p and O 1 s further validates this finding (Supplementary, Fig. S7e). The identification of the phase observed in BE and TE is not straightforward. The O 1 s spectrum suggests the formation of a new phase, perhaps a Pt alloy. The presence of Ti in both Pt TE and BE could be caused by interdiffusion or by different bonding situations of Ti and O at the TiOx-electrodes interfaces. Other causes such as misalignment due to sample tilting or image shifting are very unlikely as careful alignment procedures (of sample and images) were performed before analysis (Supplementary, Fig. S8).

The combined approach allows us to estimate the O/Ti ratio of the three species used for the fitting by performing an atom-edge jump analysis (Supplementary, Table S1). Our results indicate that the O/Ti ratio is 1.85±0.15 for the amorphous film, 1.37 ± 0.15 for the reduced and 5±1 for the reduced phase in the Pt BE and TE. Chemical mapping results combined with point spectra analysis of selected ROI_s show that the core of the conductive region is formed by a reduced phase with a O/Ti ratio close to 1.37 adjacent to a crystalline phases such as TiO2 rutile and brookite or TiO2-II in the region between the reduced area and the delaminated TE area. It is also interesting to note that the profile of region ROI_3 is very close to that of orthorhombic TiO2-II which has been recently prepared by ALD42 and found to be an intermediate phase in rutile formation by ball milling43. Its presence could support the argument of local stress manifested due to the switching. The presence of crystalline phases adjacent to the reduced TiOx area confirms that the current passes through the reduced localized area and the heating is mainly localized in this region. The increase in temperature necessary for the growth of crystalline TiO2 phases is induced by the localized Joule heating produced during the switching cycles. It was previously shown that by annealing a TiO2 amorphous film, the anatase phase is formed first because of energy surface minimization around 650 K44,45. Formation of rutile only occurs after further annealing at higher temperatures, around 1000 K17. In our film, the anatase phase is only observed in a region slightly away from the core of the conductive region, where a small reduced area is observed. However the rutile phase was identified in the region adjacent to the main reduced area. This suggests that the Joule heating temperature reached in the main conductive region as a consequence of the bias reaches 1000 K.

By recording a sequence of X-ray images of the cross-section of the TiOx thin film at closely spaced photon energies, we have directly visualized the location of the reduced area and characterized it chemically at a very fine spatial scale by extracting Ti 2p (450–485 eV) and O 1 s (525–555 eV) X-ray absorption spectra. Our results demonstrate that the conductive localized region is composed by a reduced TiOx phase with a O/Ti ratio close to 1.37 and that it is located right underneath a protrusion of the top electrode. Crystalline TiO2 rutile phase and an additional orthorhombic-like TiO2 phase regions were identified for the first time, in a localized area between the TiOx reduced phase and the top electrode of the device. This provides the first experimental indication that the temperature in the core of the conductive region increases up to 1000 K. A Ti-containing phase similar to the TiOx reduced phase was also identified in both the top and bottom Pt electrodes, giving evidence of Ti diffusion into the Pt contact electrodes due to the switching. This technique can be adapted for other RRAM materials, such as NiO46, Cu2O47 and VO248, enabling more in-depth studies in reduction dynamics.

Methods

Device Preparation

The Cr/Pt/TiOx/Pt based memristor devices were fabricated on an oxidised (200 nm SiO2) 6-inch Si wafer. The bottom electrode (BE) and top electrode (TE) were fabricated using conventional optical lithography, electron-beam evaporation followed by a lift-off process, with BE and TE composed of Cr/Pt (3 nm/30 nm) and Pt (30 nm) successively. The Cr film in the BE served as adhesive layer for Pt. A 50 nm TiOx layer was then deposited using reactive sputtering from a Ti metal target with the following settings: 8 sccm O2, 35 sccm Ar, 2 kW at the cathode and 15 sccm O2 and 2 kW at the additional plasma source.

Electrical switching of devices

Electrical biasing of the devices has been carried out through voltage sweeping in order to assess or modify their resistive states as required (2 V maximum). Passive resistive state assessment of the Device Under Test (DUT) was carried out via low voltage, non-invasive voltage sweeps whilst resistive switching was induced through the application of more aggressive, high voltage sweeps. All voltage biasing was carried out under current compliance protection. The compliance current level was determined on ad hoc basis. All sweeping was performed using a Keithley 4200 electrical characterisation instrument.

FIB-SEM

A dual-beam focused ion beam/scanning electron microscope system (Zeiss NVision 40 FIB/FEGSEM) equipped with a gas injection system (GIS) was used to record SEM images and for cutting FIB cross-sections. SEM images were recorded at an accelerating voltage of 5 kV. Prior to performing FIB cross-sections, an electron beam-induced tungsten protective layer was deposited on the top of the electrodes in order to minimize damage caused by the gallium ions in the subsequent ion beam-induced tungsten deposition step. After extraction, the thickness of the lamella is further decreased to allow X-ray transmission by low energy ion polishing at a low incident angle until a thickness of 40–70 nm is achieved.

TXM-NEXAFS

The TXM-NEXAFS study was performed at the undulator beamline U41-FSGM at the BESSY II electron storage ring operated by the Helmholtz-Zentrum Berlin, Germany. The beamline itself and also the environment of the sample is kept under high vacuum conditions. The optical design of TXM has a spatial resolution of 25 nm and spectral resolution E/ΔE = 1000049.

Samples alignment was checked using a test specimen with defined distances between structures employing the same sample holder used for the actual experiment. The alignment along the beam direction was done by having the correct focus for each photon energy. The Ti 2p images were acquired from 450 to 485 eV in 0.1 eV steps of photon energy. In order to increase the signal to noise ratio of Ti 2p spectra, three separate images were acquired for each photon energy giving a total of 1053 images. The O 1 s images were acquired from 525 to 540 eV in 0.1 eV steps and from 540 to 555 eV in 0.2 eV steps giving a total of 351 images. Images size is 1340 × 1300 pixels size corresponding to 5 nm per pixel. Each image was taken with the sample at the proper focus position to optimize spatial resolution50. First, a stack of all images was created followed by automatic alignment using Fourier cross correlation techniques in aXis2000 and in Stack_Analyze. Careful alignment of images was performed using the Zimba procedure to correct for lateral motion of the X-ray beam on the sample50. Alignment was iterated until the x,y deviations were less than 0. 6 pixels (±3 nm). The Io for each stack was obtained from an internal region of each stack (off the FIB section) and used to convert the aligned stack from transmission to optical density (OD = −log[I(E)/Io(E)]). Ti 2p and O 1 s spectra were then extracted from the stack. Spectra can be obtained from regions as small as the spatial resolution of the microscope (25 nm). Data analysis was performed using software aXis200051.

AFM

AFM maps were acquired with a MultiMode Nanoscope V AFM (Veeco Metrology Group) in contact mode using Pt/Ir coated Si tips with a cantilever spring constant of 0.2 N/m and nominal radius of 12 nm (Bruker, SCM-PIC).

Additional Information

How to cite this article: Carta, D. et al. Spatially resolved TiOx phases in switched RRAM devices using soft X-ray spectromicroscopy. Sci. Rep. 6, 21525; doi: 10.1038/srep21525 (2016).