X-ray spectromicroscopy investigation of soft and hard breakdown in RRAM devices

The Pt/TiO_x/Pt/Cr-based devices were fabricated on an oxidized (200 nm SiO₂) 6 inches Si wafer. A bottom electrode (BE) and TE were fabricated using conventional optical lithography, and electron-beam evaporation followed by a lift-off process, with the BE and TE composed of Cr/Pt (3 nm /30 nm) and Pt (30 nm), respectively. The Cr film served as the adhesive layer for the Pt BE. A 50 nm TiO_x 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. The active layer area was patterned using optical lithography and the lift-off process.

Electrical biasing of the devices has been carried out through voltage sweeping both for assessing and modifying their resistive states. Passive resistive state assessment of the device under test was carried out via low voltage (within +/−2 V), non-invasive voltage sweeps, whilst RS was induced through the application of more aggressive, high-voltage sweeps. All voltage biasing was carried out under current compliance protection. The compliance current levels were determined on an ad hoc basis. All sweeping was performed using a Keithley 4200 electrical characterization instrument.

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⁻¹ and a nominal radius of 12 nm (Bruker, SCM-PIC).

A dual-beam BFIB-SEM system (Zeiss NVision 40 FIB/FEGSEM) equipped with a gas injection system 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 cutting steps. 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.

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. The TXM has a spatial resolution of 25 nm and spectral resolution of E/ΔE = 10 000 [28]. Sequences (stacks) of x-ray images were acquired at closely spaced photon energies (0.1 eV steps) using a TXM, in the Ti 2p (450–485 eV) and O 1s (525–555 eV) energy range. Image size is 1340 × 1300 pixels corresponding to 5 nm per pixel. Each image was taken with the sample at the proper focus position to optimize the spatial resolution [29]. The stack of all images was created followed by automatic alignment using Fourier cross-correlation techniques in aXis2000 and in Stack_Analyze [30]. Each stack, consisting of 1053 images over the Ti 2p energy range and 351 images over the O 1s 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. The alignment of images is required to correct for lateral motion of the x-ray beam on the sample [29]. The incident x-ray intensity (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 1s spectra were then extracted from the stack. Spectra can be obtained from regions as small as the spatial resolution of the microscope (25 nm). A detailed analysis involving extraction of the x-ray absorption spectra of specific regions of interest was performed [31, 32] using aXis2000 software[30].

Several transition-metal oxides are of particular interest for RRAM, in particular TaO_x [33] and HfO_x [34] due to their simple reduction dynamics. On the other hand, TiO_x provides wider opportunities for multi-state memory capacity due to the intrinsic variety of possible chemical phases. A previous work by Kwon et al [4] based on high-resolution transmission electron microscopy (HRTEM) study and interpretation of electron diffraction patterns, identified Ti₄O₇ as the structure of CFs buried underneath a TE deformation in a TiO₂ based RRAM device. 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 during irradiation. Furthermore, electron diffraction requires crystals to be aligned in specific orientations with respect to the beam for unambiguously capturing distinct diffraction patterns [4]. Previous works have shown that synchrotron radiation-based techniques are powerful tools for nanoscale characterization of changes in RRAM devices [7, 10, 11, 15–17, 35, 36]. TXM-NEXAFS allowed us to simultaneously perform imaging and spectroscopy for investigating morphological changes in the film as well as performing chemical analysis at nanometer scale of localized regions, respectively. Identification of distinct TiO_x phases formed in localized regions underneath the TE damage has been achieved using Ti 2p and O 1s NEXAFS spectra. The identification is based on fingerprint 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.

Two devices were electrically characterized using voltage sweeping and the corresponding current–voltage (I–V) curves are reported in figure 2. The first device was electroformed at +5.7 V (inset of figure 2(a), compliance current 10 mA) and subsequently underwent a single switching cycle (figure 2(a)). First, a negative sweep caused switching from an LRS to an HRS at approximately −5.0 V. Next, a positive voltage sweep caused the device to switch back to the original LRS. Despite the high voltages employed, the device successfully switched between ohmic and non-linear regimes reversibly, as shown in the I–V curve of figure 2(a) and it will be hereafter called Dev_SB. The second device was formed at 12.0 V and subsequently cycled 20 times when breakdown occurred as shown in the I–V curve in figure 2(b). The device will be hereafter called Dev_HB. Figure 2 b) displays only four loops out of the 20: electroforming (inset), 1st, 8th and the 20th switching cycles where breakdown occurs. It has to be noted that the voltages/currents used to reach HB are somewhat higher than the usual values used in our commonly used operational devices due to the higher thickness of the TiO_x film (50 nm), chosen considering the spot size of the TXM-NEXAFS beam. Significant cycle-to-cycle variations in switching voltages (+3.5 , −2.5 V and +9.0, −15.0 V for cycles 1 and 8 respectively) were observed. A compliance current of 10 mA was used during forming and raised to 100 mA during switching; a value that was never reached. Throughout the electrical characterization, voltages as high as ±15 V were employed in order to account for the 50 nm active layer thickness employed in this experiment. Even if quite different electroforming voltages were required to form the Dev_SB and Dev_HB, the qualitative aspects of the switching behavior were broadly maintained.

Zoom In Zoom Out Reset image size

Several studies have shown that protrusions of the TE (both in the center [8, 16, 37–39] and at the rim [16, 38]) could indicate possible critical regions of the film responsible for the RS. The AFM image of the Dev_SB viewed from the TE, reported in figure 3(a), shows two defects: a 300 nm high isolated protrusion in the center of the active area and an extended defect at the rim of the Pt TE in the top right region. We have only investigated the extended defect at the rim of the TE as this study is mainly focused on the understanding of device failure. Therefore, the most damaged area is likely to give more detailed information on the breakdown mechanism of devices. Moreover, it has been recently reported that this position could give significant information on the RS being a preferred reduction site since it is a three-phase contact site (oxide, metal and ambient atmosphere) [36]. The detailed AFM image of the extended defect at the edge of the electrode, reported in figure 3(b), shows that the Pt TE forms several protrusions up to 400 nm high. The corresponding SEM images are reported in figures 3(c) and (d). The detailed SEM image (figure 3(d)), shows that the TE is protruded, forming several rounded features, suggesting melting of the Pt. It has to be noted that damage at the rim of the TE has been previously observed in similar devices [12].

Zoom In Zoom Out Reset image size

Optical images taken before and after electrical switching (figure S.I.1), prove that central and edge deformations were not present in the pristine device and are, therefore, due to the RS. In particular, it has been suggested that these irreversible morphological deformations are caused by the development of oxygen gas at the interface TiO_x/TE due to the formation of reduced TiO_x phases which aggregate in CFs according to the following reaction [40, 41]:

$$\begin{eqnarray}&&{{\rm{O}}}^{2-}+2\;{{\rm{Ti}}}^{4+}\to 1/2\;{{\rm{O}}}_{2}({\rm{g}})+2\;{{\rm{Ti}}}^{3+}\end{eqnarray} \tag{ 1 }$$

The temperature in the localized conductive nanofilaments increases significantly due to Joule heating causing the Pt to melt [12, 40].

In order to visualize and characterise the area buried underneath the damaged TE, an FIB lamella was extracted along the red line indicated in figure 3(a) using the lift-out technique as shown in figure S.I.2(a). Details of the cross-sectional lamella (figure S.I.2(b)), show that under the identified protrusion, the TE bulges and partially aggregates in rounded features connecting the BE and TE. The TE is lifted from its original position, supporting the argument that the protrusions were caused by O₂ gas escaping through the weakest point of the TE, the rim, in agreement with previous observations [40]. The SEM image of the lamella thinned for electron transmission, reported in figure S.I.2(c), shows that the damaged region is also lifted from the support as indicated by the bright area in the inset image.

In order to investigate chemical changes across the TiO_x layer as a consequence of SB switching, TXM-NEXAFS measurements were performed on the FIB cross-section of the device. In figure 4, x-ray images of the lamellae at 450 eV (below the Ti 2p edge), 465 eV (on the strongest Ti 2p absorption peak), 525 eV (below the O 1s edge) and 531 eV (on the strongest O 1s absorption peak) are presented.

Zoom In Zoom Out Reset image size

The stack of layers i.e. Pt/TiO_x/Pt/SiO₂/Si are highlighted in figure 4(a). By examining the Ti 2p and O 1s image sequences at different energies, we can spatially resolve the location of the regions containing Ti and O species, respectively, as the contrast of the image changes in correspondence with changes in the Ti and O absorption peak intensities. At 450 eV (figure 4(a)), the TiO_x layer appears dark as Ti does not absorb at this energy. At 465 eV (figure 4(b)), the same layer appears bright due to the strong absorption of Ti. Therefore, the regions that appear dark at 450 eV, but bright at 465 eV, correspond to Ti-containing areas. Similarly, for the O 1s at 525 eV (figure 4(c)), the TiO_x layer appears dark as O does not absorb at this energy and bright at 531 eV (figure 4(d)), due to the strong absorption of O. Therefore, the regions that appear dark at 525 eV, but bright at 531 eV, correspond to O-containing areas. Different from what is observed in the cross-section of functioning devices in the LRS [27] a main Pt agglomeration is observed underneath the defect, exactly underneath the rim of the TE, as indicated in figure 4(a). Electrochemical dissolution of Pt in oxides under high electric fields in RRAM cells has already been reported by Yang et al [42]. Spatially localized NEXAFS spectra at the Ti 2p and at the O 1s absorption edges were then extracted from various regions of the TiO_x film area in both sides of the Pt agglomeration. Five regions of interest (ROI) were identified, indicated in figures 4(b) and (d) by numbered circles. The Ti 2p and O 1s x-ray absorption spectra extracted from each of these localized regions are reported in figures 5(a) and (b), respectively. Different spectral features at both Ti and O edges appear for the different ROI, indicating changes from the pristine structure (as-deposited TiO_x).

Zoom In Zoom Out Reset image size

The Ti 2p spectrum presents distinct features for the different polymorph phase of crystalline TiO₂ and for amorphous TiO₂ [43]. The Ti 2p spectrum of the pristine sample (figure 5(a)) shows broad peaks, and no splitting of the (2p_3/2, e_g) peak. These features are consistent with a high structural disorder caused by a range of bond angles and lengths as expected for amorphous TiO_x. Ti 2p spectrum of region 5, on the right side of the Pt agglomeration, exhibits broad peaks as an amorphous structure. In addition, an energy shoulder appears at around 456 eV, which is consistent with the presence of a partially reduced phase containing Ti³⁺ [38]. The presence of a mixture of Ti³⁺ and Ti⁴⁺ could contribute to the fact that the peaks are slightly broader compared to the pristine sample, as inferred from the reduced 2p_3/2 and 2p_1/2 peak-to-valley ratio. In addition, 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⁺ spectra [44]. Spectra from regions 1–4, all extracted from the TiO_x areas underneath the damaged TE show sharper features with significant variation in shape, clearly indicating significant changes in the chemical structure of TiO_x. In particular, all spectra show splitting of the (2p_3/2, e_g) peak, which indicates distortion of the TiO₆ octahedra [45]. The relative intensities of these two peaks depend on the particular type of octahedral distortion, and therefore, on the particular polymorphic form of TiO_2. In the spectrum of region 4, very close to the Pt agglomeration, the relative intensities of the split (2p_3/2, e_g) peak are different, being higher for the peak at higher energy. This is typical of the TiO₂ rutile phase, which has tetragonal distortion of the TiO₆ octahedra [43, 46]. In the spectra of regions 1–3, the relative intensity of the split (2p_3/2, e_g) peak is the opposite to that in rutile, with the intensity being higher for the peak at lower energy. This profile is typical of anatase [45, 47]. The peak-to-valley ratio of the 2p_1/2 peak in regions 2 and 3, where the TiO_x film is delaminated from the Pt BE, are particularly high, indicating a high degree of crystallinity. In contrast, 2p_1/2 peaks of region 1, where the TiO_x layer is not yet delaminated, are much broader, indicating a lower degree of crystallinity.

If Ti 2p spectra give a clear indication of the phases formed, while O 1s spectra, shown in figure 5(b), are more difficult to interpret. Variations between the selected regions of interest are less evident at the O 1s, as the change in structure of TiO_x has less influence on O 1s than Ti 2p spectra. O 1s spectra of the pristine device show broad and smooth spectral features typical of amorphous TiO_x. Moreover, the dip between the (O 1s, t_2g) and (O 1s, e_g) peaks is shallow compared to the spectra of all the other regions and their energy difference, which is often used to evaluate crystal field splitting, it is smaller and typical of disordered materials (2.0 eV) [48]. In the spectra extracted from regions 1–4, the e_g and t_2g peaks appear more defined and intense. The energy gap between e_g and t_2g in regions 1–4 is ∼3.0 eV, a value reported for both anatase and rutile phases [43]. In region 5, the dip between e_g and t_2g is less intense, which is in agreement with the Ti edge that shows broad peaks, as an amorphous structure. It has to be noted that no signal due to the presence of oxygen vacancies is observed beneath the O 1s jump, as expected for a reduced TiO_x. This could be ascribed to the fact that only a small amount of Ti⁴⁺ is reduced to Ti³⁺ so the amount of Ti³⁺ might not be high enough to generate an observable peak. In support of this statement, in the work of Strachan et al [38], where a shoulder at 456 eV of similar intensity corresponding to Ti³⁺ was observed in the Ti 2p spectrum, no signal due to defect state was observed in the O 1s spectrum.

We now consider the sample switched for 20 cycles until HB. Figure 6(a) depicts the SEM image of Dev_HB viewed from the TE taken using a conventional secondary electron detector (EHT = 5 kV).

Zoom In Zoom Out Reset image size

Major damage of the TE occurred, with tree branch-like patterns spreading over the junction area and peeled-off metal as a result of heating and oxygen pressure. A similar branch-like pattern of local defects has already been observed as a result of electrical conduction through thin sandwich structures [18]. A complete detachment of the TE from the active area is observed on the right rim of the junction. Very similar morphological changes with the formation of dendrite-like structures were observed by Skaja et al [49] in the TE of Pt/Ta₂O₅/Ta devices. As evident from the magnified SEM (figure 6(b)) and optical (figure 6(c)) images, several parts of the Pt TE are delaminated. However, the BE is still mainly unaffected as proved by the dark blue regions observed in the optical image underneath the delaminated Pt TE. The AFM images reported in figure 6(d) and figure S.I.3(a), show that the Pt protrusions are mostly 100–150 nm high with a few features up to about 400 nm. The deflection error image, reported in figure S.I.3(b), reveals additional details of the defects, such as the 2 μm elongated crater shown in the inset.

An FIB lamella was cut in the region highlighted by the red line in figure 6(a). Details of the lamella extraction are reported in figure S.I.4(a). Different from Dev_SB, cross-section views reported in figure S.I.4(b) and figure S.I.4(c), show several Pt inclusions lying on the BE. In particular, a main Pt aggregate about 200 nm wide protrudes across the oxide film, marked in figure 7(a). It has to be noted that in this device, the number of Pt protrusions across the active oxide is much higher than in Dev_SB.

Zoom In Zoom Out Reset image size

TXM micrographs at the Ti 2p (450 and 465 eV) and at the O 1s (525 and 531 eV) are reported in figure 7. As described previously for the Dev_SB, regions that appear dark at 450 and 525 eV and bright at 465 and 531 eV, correspond to Ti- and O-containing areas, respectively. NEXAFS spectra at the Ti 2p and the O 1s were extracted from the TiO_x film area. Due to the extent of the damage, ten ROI were identified, indicated in figures 7(b) and (d) by numbered circles.

The Ti 2p and O 1s x-ray absorption spectra extracted from each of these localized regions are reported in figures 8(a) and (b), respectively.

Zoom In Zoom Out Reset image size

Let us first consider the Ti 2p spectra. As observed in the case of Dev_SB, significant spectral variations are observed in different regions. Region 1, extracted from an unaffected area, shows mainly amorphous features. Regions 2, 3, 6, 7, 8 and 10 show crystalline features typical of anatase phase as the first peak of the 2p_3/2 eg splitting has higher intensity than the second. Region 4, very close to the main Pt agglomeration, again shows crystalline features, but the intensity of the 2p_3/2 eg splitting is comparable. This particular (2p_3/2, e_g) profile has been reported for orthorhombic-like phases such as TiO₂-II [47] and brookite [46]. However, the authors cannot exclude the possibility that such a profile could be due to the simultaneous contribution of anatase and rutile. Finally, in regions 5 and 9, which are close to the main defects, the first peak of the 2p_3/2 eg splitting has lower intensity than the second, a feature typical of rutile phase. Spectra at the O 1s edge are more difficult to interpret. However, it is confirmed that the pristine sample has broader features compared to the highly crystalline phases.

It has been previously shown that the main TE defects (such as craters or protrusions) are typically surrounded by broad areas with scattered surface residue, which has moderate conductivity [25, 50]. We have observed the presence of a halo surrounding the tree-like defect by imaging the TE using a SEM in-lens detector (figures 9(a) and (b)).

Zoom In Zoom Out Reset image size

The conventional secondary electron detector (used for the image shown in figure 6(a)) is positioned outside the lens system and it is more bulk sensitive. The in-lens detector is positioned in the optical axis in the SEM column and has high surface sensitivity as the image is formed by low-energy electrons. Therefore, the in-lens image contains direct information of the sample surface. The contrast difference between images taken using a conventional secondary electron detector and an in-lens one can be clearly distinguished and different features are observed. In figures 9(a) and (b), two areas with different brightness can be distinguished, the brightest area forming a halo around the defect. Different brightness could be related to difference in the work function (e.g. electronic variations) of the Pt TE in that area. Kelvin probe force microscopy (KPFM) measurement, reported in figure S.I.5, indicates that there is a difference in work function between the heavily damaged area and the undamaged electrode. However, the change in work function in the halo area is not clear. Further analysis is required in order to clarify this point. It has to be noted that other factors affecting the Pt TE in that area could cause differences in brightness such as recrystallization or a cleaner surface due to Joule heating. Chemical characterization of the oxide film underneath the two areas was performed using TXM-NEXAFS chemical mapping. In the inset of figure 9(b), the FIB lamella cut along the green (outside the halo) and red (inside the halo) regions is presented, showing the cross-sections of the film underneath both areas. Chemical maps were generated using the components corresponding to the optimal fits, TiO_x amorphous and TiO₂ anatase, represented as white regions in the inset in figure 9(b). While the region outside the halo corresponds to amorphous TiO_x, as expected, the region underneath the halo has crystallized to anatase.