Chemical insight into electroforming of resistive switching manganiteheterostructures†

Francesco Borgatti,∗a Chanwoo Park,∗b Anja Herpers,b Francesco Offi,c Ricardo Egoavil,d Y.Yamashita,e Anli Yang,e M. Kobata,e K. Kobayashi,e Jo Verbeeck,d Giancarlo Panaccionef andRegina Dittmannb

Received Xth XXXXXXXXXX 20XX, Accepted Xth XXXXXXXXX 20XX

First published on the web Xth XXXXXXXXXX 200X

DOI: 10.1039/b000000x

We have investigated the role of the electroforming processin the establishment of resistive switching behaviour forPt/Ti/Pr0.5Ca0.5MnO3/SrRuO3 layered heterostructures (Pt/Ti/PCMO/SRO) acting as non-volatile Resistance Random AccessMemories (RRAMs). Electron spectroscopy measurements demonstrate that the higher resistance state resulting from electro-forming of as-prepared devices is strictly correlated to the oxidation of the top electrode Ti layer through field-induced electro-migration of oxygen ions. Conversely, the PCMO exhibits oxygen depletion and downward change of the chemical potentialfor both resistive states. The impedance spectroscopy analysis, supported by the detailed knowledge of these effects,providesan accurate model description of the device resistive behaviour. The main contributions to the change of resistance from theas-prepared (low resistance) to the electroformed (high resistance) states are respectively due to reduced PCMO at theboundarywith the Ti electrode and to the formation of an anisotropicn-p junction between the Ti and the PCMO layers.

Complex transition metal oxides such as manganites1–5,titanates and zirconates6,7 may exhibit large electrically-driven changes of resistance, thereby attracting considerableattention for the development of novel electronic conceptssuch as non-volatile Resistance Random Access Memories(RRAMs)8–10. The perovskite manganites are particularlypromising for the development of multifunctional memristordevices, owing to their largely tunable electric and magneticproperties11,12. Moreover, high and low resistive state currentsof several manganites films scale with the electrode area13,which might be advantageous for circuit design since it of-fers another degree of freedom to adjust the device parameters,while for mostn-conducting oxides, e.g. TiO2, TaOx, SrTiO3,the resistive switching (RS) behaviour is determined by area-independent distribution of sparse filamentary regions14–16. It

† Electronic Supplementary Information (ESI) available. See DOI:10.1039/b000000x/a CNR – Istituto per lo Studio dei Materiali Nanostrutturati (ISMN), via P.Gobetti 101, I-40129 Bologna, Italy. Fax: +39 051 6398540; Tel: +39 0516398518; E-mail: francesco.borgatti@cnr.itb Peter Gruenberg Institute (PGI-7) and JARA-FIT ForschungszentrumJuelich GmbH, 52425 Juelich, Germany. E-mail: park@IWE.RWTH-Aachen.dec CNISM and Dipartimento di Scienze, Università Roma Tre, Rome, Italy.d EMAT, Electron Microscopy for Materials Science, University of Antwerp,Groenenborgerlaan 171, 2020 Antwerp, Belgium.e NIMS beamline Station at Spring-8, National Institute for Materials Science,Sayo, Hyogo 679-5148, Japan.f CNR – Istituto Officina dei Materiali (IOM), Laboratorio TASC, S.S.14, Km163.5, I-34149 Trieste, Italy.

is widely recognized that thebipolar RS behaviour exhib-ited by manganite compounds is most frequently connectedwith voltage-driven oxygen vacancy movement, therefore rel-evant redox processes may occur at the electrode/oxide inter-faces17,18. In fact, the choice of oxidizable electrodes like Al,Ti or Ta and the occurrence of rectifying current-voltage (I-V)characteristics look as essential prerequisite for RS in a typ-ical electrode/manganite/electrode heterostructure19–22, whileohmic contact and no resistive switching is generally observedfor noble metal electrodes like Pt or Au23. The chemical statesof the boundary regions can differ significantly from the bulkand are susceptible of a crucial role for the establishment (orthe enhancement) of the RS since the initialelectro-formingprocess. This treatment, consisting of strong voltage stressapplication, is evenly essential to set and/or improve the repro-ducibility of the RS also for perovskite compounds exhibitingnative switching behaviour. However the local effects withinthe device structure, and in particular at the electrode/oxideboundary regions, are still not properly understood on a gen-eral basis24. In the perspective of tuning the desired func-tionalities of these heterostructures in a ”materials-by-design”approach for the development of reliable RS-based RRAMdevices, the detailed identification of the chemical statesatthe interfaces and their relationship with electroformingandswitching processes is an open key question.

In this work we have approached this topic forPt/Ti/Pr0.5Ca0.5MnO3/SrRuO3 layered heterostructures(Pt/Ti/PCMO/SRO), for which the electro-forming treatment

1–7 | 1

establishes the RS behaviour by modifying the system frominitially low to high resistance state. PCMO is one of thebasic elements for reversible RS behaviour, and PCMO-baseddevices exhibit the scalability of the RS effect with the area,thereby representing a prototypical system of large interest.On the basis of the electrical and resistive behaviour ofthe devices, we have focused on the study of the top Tielectrode and the underlying PCMO oxide by combiningchemical and structural sensitive spectroscopies such as HArdX-ray PhotoElectron Spectroscopy (HAXPES), high angleannular dark-field scanning transmission electron microscopy(HAADF-STEM), electron energy loss spectroscopy (EELS),and impedance spectroscopy (IS). Whereas EELS analysisprovides us with elemental and chemical analysis up toatomic spatial resolution, the bulk sensitivity provided bycore-level HAXPES25–27 allows for nondestructivestudiesof deep regions and buried interfaces for all the elements ofinterest28–32. As a result, we have been able to discriminatethe role of the different interface contributions to the electricalproperties of this system, and to present a consistent devicemodel inclusive of the chemical changes produced by theelectro-forming mechanism.

Pr0.5Ca0.5MnO3 (PCMO) and SrRuO3 (SRO) thin filmswere epitaxially grown on SrTiO3(001) single crystals bypulsed laser deposition, subsequently covered by a Ti/Pt bi-layer deposited by RF sputtering (the Pt acting as protectivelayer), and then patterned by photolithography†. Referencesamples were made by depositing PCMO and SRO thick filmson SrTiO3(100) substrates. The HAXPES measurements wereperformed at the BL15XU beam-line of the Spring-8 syn-chrotron radiation source (Japan). The experimental geome-try was chosen to maximize the bulk sensitivity, with the an-gle between the incident x-ray beam and the electron analyzerset to 90◦, and the x-ray beam grazing incidence at 2◦ on thesamples. The beam footprint on the sample surface was about3000x100µm2. The photon energy was set at 5950.14 eV andwas periodically verified by measuring the Fermi edge spec-trum of a reference Au sample, showing energy fluctuationslower than 10 meV. The overall resolution was 0.25 eV. All themeasurements were performed at room temperature. Typicalthickness of the layers for the HAXPES measurements is 4 nm(Pt), 7-10 nm (Ti), 20 nm (PCMO), 30 nm (SRO). The photo-electron emission from the buried regions of interest is ensuredby the large electron inelastic mean free path (IMFP) typicalof the hard x-ray excitation regime, of the order of 8 nm33.High-angle annular dark-field (HAADF)-scanning transmis-sion electron microscopy (STEM) and electron energy-lossspectroscopy (EELS) was performed on a JEOL 3000 trans-mission electron microscope operated at 300 kV equippedwith a Gatan 2000 imaging filter. Two TEM cross-sectionelectron transparent lamellas were prepared using a FEI NovaNanolab 200 DualBeam SEM/FIB system using the so called

Fig. 1 (a) schematic of the layered Pt/Ti/PCMO/SRO switchingcells. The substrate is SrTiO3(100); (b)I-V characteristics measuredby using a two-point probe method of structures with different topelectrode (SRO, Pt and Ti) and a SRO/PCMO/Nb:STO structure.The bias is applied to the top electrode; (c) the formingI-V curveand the subsequent strong asymmetric switching hysteresis; (d)resistance distribution of 135 pads before (filled black boxes) andafter forming the high resistance state (hollow red boxes).Athreshold DC forming automatic system was used, in which constant10 V is applied and the measurement is terminated when theresistance reaches a threshold resistance (5 kΩ). The electrical datashown in (c), (d) are for the HAXPES samples, which had a 4 nm Ptand a 7 nm Ti layer.

”in-situ lift-out” technique†. A STEM convergence angle of∼7 mrad and a collection angleβ ∼21 mrad were used. Theenergy resolution is estimated to be∼1eV. The observed thick-ness of the layers for the devices dedicated to the TEM/EELSanalysis is 20 nm (Pt), 15 nm (Ti), 20 nm (PCMO), 30 nm(SRO). The EELS spectra were analysed using Digital Micro-graph. To improve the signal to noise ratio in the quantitativeelemental profiles, principal component analysis (PCA) wasapplied to the EELS data sets34,35.

The structure of the Pt/Ti/PCMO/SRO stacked switchingcells is schematically shown in Fig. 1(a). A first indica-tion about the role of layers and interfaces in the chargetransport is provided by the comparison in Fig. 1(b) of thecurrent-voltage (I-V) characteristics for stacks with differentbottom/top electrodes (TEs), namely three TE/PCMO/SROstructures (TE=SRO, Pt and Ti) and a SRO/PCMO structurefabricated on a Nb-doped SrTiO3 (Nb:STO) substrate with0.5 wt% Nb doping concentration. The devices with epitax-ial SRO and Pt, both being high work function metal elec-trodes, show a linearI-V behavior indicating an ohmic contact

2 | 1–7

to the PCMO (RSRO≈ 126Ω, RPt ≈ 93 Ω). Conversely, theSRO/PCMO contact on the Nb:STO substrate, which is ann-type semiconductor, shows the rectifying behavior expectedby ap-n junction, confirming that PCMO behaves as ap-typesemiconductor. The case of the Ti top electrode, a shallowwork function metal, differs from the others and shows a sym-metric nonlinearI-V curve. This is not the behavior expectedfor a genuine Ti/PCMO Schottky-contact, thereby suggestingthe presence of a more complex interface structure5. The ap-plication of a positive threshold voltage of the order of 8-10 V,shown in Fig. 1(c), induces the electro-forming process whichbrings this system from the inital resistance state (IRS) tothehigh resistance state (HRS), which is larger by about one or-der of magnitude. After forming, theI-V curves exhibit a re-versible hysteresis which is more pronounced in the negativebranch, while no permanent change of the electrical proper-ties could be obtained when a negative voltage is applied tothe top electrode (”negative forming”). Fig. 1(d) summarizesthe resistance of 135 devices in the IRS and HRS condition,the latter being achieved after forming with a dc voltage of 10V. Although some spread in the device resistances is evident,we demonstrate that all devices could be formed to the HRSquite reliably with the same electrical treatment. The stabilityof the formed state was measured in some devices over morethan two months repeatedly without observing any significantchange in the resistance. This offers the opportunity to inves-tigate the chemical changes in the formed state weeks or evenmonths after electrical programming.

The microscopic changes of the chemical states and com-position induced by the electro-forming process have beenprobed by HAADF-STEM and EELS for both resistive states.The TEM images of the samples, depicted in Fig. 2(a) withthe normalized cross-sectional elemental distributions acrossthe PCMO/Ti layers for Ti, Ca, Mn, Pr and O, show that bothIRS and HRS systems at the PCMO/Ti boundary contain anintermixing region of about 8.5 nm. For the IRS sample theoxygen is distributed within this intermixing region, indicatingan oxidation of the Ca-doped TiOx to x = 2 while the rest ofthe Ti layer remains low in oxygen content. The HRS systemon the other hand, has a much wider oxidized region, reach-ing most of the Ti layer thickness. This result suggests thatthe Ti/PCMO interface is largely involved in the HRS form-ing process through the migration of oxygen ions from thePCMO towards the Ti layer. The evolution of the chemicalstate across the PCMO/Ti interface is indicated by the Energy-Loss Near-Edge Structure (ELNES) spectra reported in Fig.2(b,c) for Ti and Mn L2,3 absorption edges detected at differ-ent positions within the samples. The Mn L2,3 edges mea-sured closer to the intermixed PCMO/Ca-TiOx region showa slight shift with respect to the PCMO bulk, indicative of asmall variation towards a larger electron charge density attheMn sites36,37. This is more evident for the IRS case, confirm-

Fig. 2 (a) on the right: cross-section TEM images of theas-prepared (IRS) and high-resistance formed (HRS) samples. Onthe left, the associated EELS quantification of Ti, Ca, Mn, Prand O.The intensities are normalized to the maximum of each curve.TheELNES spectra for Ti L2,3 and Mn L2,3 edges in panels (b) and (c),respectively, are collected at the sample positions indicated in panel(a) by the correspondent vertical colored arrows. Spectra arenormalized to maximum edge intensity and to the pre-edge.

ing that most of the oxygen ions involved in the HRS formingprocess belong to the PCMO and Ca-TiOx layers. The shift-ing of about 2 eV between the Ti L2,3 edges measured at themixed PCMO/Ca-TiOx region and nearby the Pt overlayer isdue to different valence states, namely Ti4+ (titanium dioxide)and Ti0+ (metallic). Therefore the Pt layer is thick enough toefficiently protect from external oxidation by the surroundingatmosphere38,39. For both resistive states the Ti absorptionedges of the TiOx regions are slightly shifted from the metallicTi of about 0.5 eV, suggesting that Ti-O chemical interactionoccurs via formation of intermediate TiOx compounds.

To further disentangle the chemical variations correlatedtothe change of the resistance state, core-level HAXPES mea-surements of Ti and PCMO have been performed for deviceseither in the IRS or HRS condition. The electronic structureof the Ti top electrode has been probed by the Ti 2p spectra,shown in Fig. 3. Samples with a thin Pt capping layer (4nm) and two different Ti layer thicknesses (7 and 10 nm) havebeen investigated to vary the depth sensitivity to the bulk.Be-cause of the partial overlap of the Ti 2p1/2 with the Ru 3p3/2

1–7 | 3

peak, only the spectral region of the Ti 2p3/2 components isclearly distinguishable and exhibits two distinct features at454 eV and 459 eV which are associated to Ti0 (metallic) andTi4+ ion valence states, respectively40,41. The relative inten-sity of these two peaks is very different in the IRS and HRS,demonstrating the strong sensitivity of the HAXPES spectrato changes of the Ti chemical states due to the electroform-ing process. The detailed fit analysis of the Ti 2p spectra, re-ported for the sample with a Ti layer of 7 nm, demonstratesvery clearly that both spin-orbit components include interme-diate valence terms, confirming that even for the as-preparedIRS systems there is a large chemical interaction with the un-derlying PCMO. In particular, the intensity of the Ti0 term dueto the metallic Ti present at the top of the layer is always lessthan 15% of the overall Ti 2p photoelectron yield. Howeverthe Ti4+ spectral term for the IRS case cannot be exclusivelyrelated to formation of TiO2 but could be partially ascribedto Ti4+ ions distributed within the PCMO-Ti mixed region,with the formation of a Ca-doped TiOx compound (Ca-TiOx)or substitution in the PCMO lattice, thereby contributing to thedegradation of the system metallicity42–44. After the formingof the HRS state the Ti0/Ti4+ peak intensity ratio is reducedfrom 0.32 (0.56) to 0.17 (0.42) for the sample with a Ti layerof 7 (10) nm, while the intensity of the components associ-ated with the intermediate valence states is reduced just by2-6%, showing that formation of TiO2 through the diffusion ofoxygen ions plays a crucial role for the change of the resistiv-ity 45. In fact, TiO2 is essentially ann-type semiconductor withoxygen vacancy and/or excess Ti ions at the interstitial site asdonors. For the IRS devices the doping of TiO2 with Ca inthe mixed boundary region changes the electrical conductionfrom n-type for un-doped TiO2 towardsp-type for Ca-dopedTiO2 46–49.

The core-level spectra for all the PCMO elements are re-ported in Fig. 4. All the peaks exhibit a shift of the bindingenergy (BE) with respect to the spectra of the bare materi-als, towards higher BE for Pr 3d (0.6 eV), Ca 2p (0.7 eV)and lower BE for Mn 2p (0.2 eV). Being very similar forboth resistive states, they denote a stable change of the PCMOelectronic structure determined by the lowering of the PCMOchemical potential at the Ti/PCMO boundary, in close analogyto what has been observed for PCMO compounds as a func-tion of the Ca doping level50,51. The opposite direction of theMn 2p shift results is interpreted as the progressive changeofthe Mn4+ species to lower valence states (Mn3+) due to therelease of surrounding oxygen atoms. This behaviour differsfrom the rigid BE shifts observed for the core-level spectraofTi/HfO2/TiN devices in the low and high resistance states31.In our case, it is the fingerprint of oxygen depletion inside thePCMO acting as an effective compensation of the hole dopingby the Ca ions, thus raising up the PCMO resistivity. Most ofthe core level spectra do not exhibit spectral features indica-

Fig. 3 Ti 2p core-level spectra of the IRS (red) and HRS (blue)samples. Spectra are normalized to the background intensity. The fitresults for the Ti layer of 7 nm is shown at bottom for both IRS (red)and HRS (blue) spectra. Both Ti 2p3/2 and 2p1/2 components arereported. The dashed black line is the background adopted for the fitanalysis.

tive of strong chemical interaction with Ti52,53. The slightlylarger Mn 2p3/2 tail asymmetry at the high BE side of the HRSspectra and the lower intensity of feature at about 664 eV, re-lated to charge transfer with the surrounding ions37,54, supportfor a slightly different amount of oxygen vacancies within theprobed volume. On the other hand, the O1s spectra clearlydiffer from the reference curves of the bare oxides, with thebroadening of the peaks and the appearing of at least threewell defined features. The complexity of these lineshapes,being composed of several contributions, makes a reliable fitanalysis difficult. On the basis of their intensity behaviour,we guess that the small and stable feature ”A” is associatedto the surface oxidation of the Pt capping layer, while the ”B”and ”C” features, modified in opposite way by the formingprocess, could be related to Ti oxide states and the PCMO,respectively. We can therefore conclude that the PCMO un-dergoes minor changes during electroforming and the mainchemical changes in our device structure occur in the TiOxinterface layer.

4 | 1–7

The knowledge of the mechanisms involved in the electro-forming step provides valuable information to analyze the es-tablishment of the RS condition by impedance spectroscopyof the devices. Fig. 5(a) and (b) show Cole-Cole plots of theimpedance for IRS and HRS states. The impedance depen-dence on the lateral device size, shown in Fig. 5(c,d), con-firms that the current is transported homogeneously across theentire electrode area. Both contain two semi-circles whichhave been fitted by two parallel Resistance (R)-Capacitance(C) components in series. For the IRS sample, one of theseR-C circuits would represent the Ca-doped TiOx (Ca-TiOx)interface layer, while the other could be the Schottky barrierat the interface†. However the symmetric nonlinearI-V curveshown in Fig. 1(b) for the as-prepared Ti/PCMO device doesnot exhibit the rectifying behavior which would be expectedfor a Schottky-contact. Accordingly to the large carrier con-centration of the PCMO, the small depletion width enableselectron tunneling across this thin barrier in both currentdi-rections, hence the second R-C circuit is most likely related tothe oxygen-deficient PCMO layer which is naturally formedduring the deposition of the Ti top electrode5. Considering thethickness of 8.5 nm for the intermixed Ca-TiOx layer, achievedby the EELS measurements, and the resistance values R1 ∼2.8kΩ and R2 ∼51 kΩ from the numerical simulation, the relativedielectric constants for the two R-C contributions are aboutε1=20 andε2=110. Since the relative dielectric constant ofthe Ca-TiOx layer, which contains a large amount of defects,should not exceed the TiO2 literature value (about 100), wesuggest that R1 and R2 are associated with the Ca-TiOx layerand to the reduced PCMO interface, respectively. We noticethat R2 ≫ R1, thereby the resistance in the IRS samples isdominated by the reduced PCMO interface.

The electro-forming process increases R1 and R2 by one or-der of magnitude up to R′1 (18 kOhm) and R

′

2 (429 kOhm),correspondingly to the chemical changes recognized by theelectron spectroscopy measurements. On this basis, the mech-anism for the change from IRS to HRS is the diffusion of oxy-gen ions inside the Ti layer and the formation of oxygen va-cancies in the intermixed Ca-TiOx region, which transformsthis layer from insulator-like to ann-type semiconductor, andmakes the Ti/TiOx/Ca-TiOx structure a kind of metal/n-type/n-type junction. As a consequence, in the HRS sample theTi/PCMO interface evolves towards an anisotropicp-n junc-tion consisting of thep-type PCMO and then-type TiOx,respectively†. This picture is fully consistent with the rectify-ing I-V behavior reported in Fig. 1(c) after the electro-formingprocess. Although the complexity of this interface model doesnot allow to correlate unambiguously one of the two semicir-cles to thep-type depletion width, our simulations suggest thatthe entire reduced PCMO interface in the HRS could belongto thep-type depletion layer. Therefore the two parallel R-Ccircuits of the impedance spectra in the HRS case can be as-

Fig. 4 PCMO-related core-level spectra of the IRS (red) and HRS(blue) samples and the bare PCMO (black) and SRO (brown).Photon energy was about 5.95 keV. The O 1s and Pr 3d spectra arenormalized to the background intensity, while a second-orderpolynomial background has been subtracted to the Mn 2p and Ca2pspectra before normalizing the peak intensities.

sociated with the two depletion layers on the two sides of theanisotropicn-p junction and imply the energy band-diagramdepicted in Fig. 5(f). Since the formation of the TiOx layer andthe resulting space charge region at the PCMO interface is thedominating effect during electroforming, one could speculatethat the electrical changes in the PCMO itself will also playaminor role for the later resistive switching process. In order toconfirm this assumption, detailed investigations of the switch-ing between high and low resistive states will be performed inthe future.

In conclusion, we have investigated by electron andimpedance spectroscopy techniques the influence of theelectroforming process on the Ti/PCMO interface forPt/Ti/PCMO/SRO heterostructures exhibiting resistiveswitching behaviour. The analysis performed for samplesin the initial (as-prepared) and high resistance conditionsshows that the PCMO undergoes minor changes duringelectroforming whilst the main chemical changes in ourdevice structure occur in the TiOx interface layer. The EELSmeasurements indicate that as-prepared devices present acomplex interface structure with a mixed PCMO/Ca-TiOx-Tiregion and partial oxidation of the Ti electrode which wasstrongly modified during electroforming. The comparisonof the Ti 2p HAXPES spectra for IRS and HRS samplesfurther demonstrate that oxidation of titanium is correlated

1–7 | 5

Fig. 5 (a,b) Semi-circle combinations of Cole-Cole plot of the IRSand the HRS after forming. The impedance and resistancedependence on the lateral device size are shown in panels (c)and(d), respectively. (e,f) Energy band-diagrams of the IRS and theHRS after forming.

to the field-induced electromigration of oxygen ions insidethe Ti layer. Conversely, the PCMO layer exhibits oxygendepletion and downward change of the chemical potentialfor both resistive states. Supplementing these investigationsby impedance spectroscopy, we are able to discriminate theinfluence of the electroforming process on the electricalproperties of the devices through the variation of chemicalstates. Based on the comparison between chemical analysisand electrical data, we have recognized the reduced PCMOinterface layer as the dominant resistance contribution intheIRS and thep-n junction between the TiOx and the PCMO thedominating resistance contribution in the HRS.

Acknowledgments

This work was funded by the European Union Council un-der the 7th Framework Program (FP7) grant nr. 246102IFOX. J.V. acknowledges funding from the European Re-search Council under the 7th Framework Program (FP7),ERC grant number 278510 VORTEX. J.V. and ERC grantnumber 246791-COUNTATOMS. The titan microscope waspartly funded by the Hercules fund from the Flemish Govern-

ment. The synchrotron radiation experiments were performedat the BL15XU of SPring-8 with the approval of the JapanSynchrotron Radiation Research Institute (JASRI) (ProposalNo. 2011A4900). This work was partially supported by theNanotechnology Network Project, the Ministry of Education,Culture, Sports, Science and Technology (MEXT), Japan.The authors are grateful to HiSOR, Hiroshima Univ. andJAEA/SPring-8 for the development of HX-PES at BL15XUof SPring-8. We would like to thank C. Lenser for his assis-tance and valuable discussions.

References

1 A. Asamitsu, Y. Tomioka, H. Kuwahara and Y. Tokura,Nature, 1997,388, 50–52.

2 H. Oshima, K. Miyano, Y. Konishi, M. Kawasaki and Y. Tokura,AppliedPhysics Letters, 1999,75, 1473–1475.

3 Y. B. Nian, J. Strozier, N. J. Wu, X. Chen and A. Ignatiev,Physical ReviewLetters, 2007,98, 146403.

4 M. Quintero, P. Levy, A. G. Leyva and M. J. Rozenberg,Physical ReviewLetters, 2007,98, 116601.

5 A. Sawa, T. Fujii, M. Kawasaki and Y. Tokura,Applied Physics Letters,2004,85, 4073–4075.

6 A. Beck, J. G. Bednorz, C. Gerber, C. Rossel and D. Widmer,AppliedPhysics Letters, 2000,77, 139–141.

7 Y. Watanabe, J. G. Bednorz, A. Bietsch, C. Gerber, D. Widmer, A. Beckand S. J. Wind,Applied Physics Letters, 2001,78, 3738–3740.

8 M. H. Lee and C. S. Hwang,Nanoscale, 2011,3, 490–502.9 R. Waser, R. Dittmann, G. Staikov and K. Szot,Advanced Materials,

2009,21, 2632.10 R. Waser and M. Aono,Nature Materials, 2007,6, 833–840.11 Y. V. Pershin and M. Di Ventra,Advances in Physics, 2011,60, 145–227.12 M. Prezioso, A. Riminucci, P. Graziosi, I. Bergenti, R. Rakshit,

R. Cecchini, A. Vianelli, F. Borgatti, N. Haag, M. Willis, A.J.Drew, W. P. Gillin and V. A. Dediu,Advanced Materials, 2012, doi:10.1002/adma.201202031, n/a–n/a.

13 A. Sawa,Materials Today, 2008,11, 28–36.14 J. P. Strachan, J. J. Yang, R. Munstermann, A. Scholl, G. Medeiros-

Ribeiro, D. R. Stewart and R. S. Williams,Nanotechnology, 2009,20,485701.

15 F. Miao, J. P. Strachan, J. J. Yang, M. X. Zhang, I. Goldfarb, A. C. Tor-rezan, P. Eschbach, R. D. Kelley, G. Medeiros-Ribeiro and R.S. Williams,Advanced Materials, 2011,23, 5633.

16 R. Muenstermann, T. Menke, R. Dittmann and R. Waser,Advanced Ma-terials, 2010,22, 4819.

17 M. J. Rozenberg, M. J. Sanchez, R. Weht, C. Acha, F. Gomez-Marlascaand P. Levy,Physical Review B, 2010,81, 115101.

18 Y. S. Chen, B. Chen, B. Gao, L. P. Chen, G. J. Lian, L. F. Liu, Y. Wang,X. Y. Liu and J. F. Kang,Applied Physics Letters, 2011,99, 072113.

19 K. Shono, H. Kawano, T. Yokota and M. Gomi,Applied Physics Express,2008,1, 055002.

20 S. Asanuma, H. Akoh, H. Yamada and A. Sawa,Physical Review B, 2009,80, 235113.

21 H. Kawano, K. Shono, T. Yokota and M. Gomi,Applied Physics Express,2008,1, 101901.

22 T. Yamamoto, R. Yasuhara, I. Ohkubo, H. Kumigashira and M.Oshima,Journal of Applied Physics, 2011,110, 053707.

23 K. Tsubouchi, I. Ohkubo, H. Kumigashira, M. Oshima, Y. Matsumoto,K. Itaka, T. Ohnishi, M. Lippmaa and H. Koinuma,Advanced Materials,2007,19, 1711.

6 | 1–7

24 F. Gomez-Marlasca, N. Ghenzi, M. J. Rozenberg and P. Levy,APPLIEDPHYSICS LETTERS, 2011,98, 042901.

25 G. Panaccione and K. Kobayashi,Surface Science, 2012,606, 125–129.26 G. Panaccione, F. Offi, M. Sacchi and P. Torelli,Comptes Rendus

Physique, 2008,9, 524–536.27 C. S. Fadley,Journal of Electron Spectroscopy and Related Phenomena,

2010,178, 2–32.28 K. Kobayashi, M. Yabashi, Y. Takata, T. Tokushima, S. Shin,

K. Tamasaku, D. Miwa, T. Ishikawa, H. Nohira, T. Hattori, Y. Sugita,O. Nakatsuka, A. Sakai and S. Zaima,Applied Physics Letters, 2003,83,1005–1007.

29 F. Borgatti, I. Bergenti, F. Bona, V. Dediu, A. Fondacaro,S. Huotari,G. Monaco, D. A. MacLaren, J. N. Chapman and G. Panaccione,AppliedPhysics Letters, 2010,96, 043306.

30 H. Tanaka, I. Satoh, T. Kanki, E. Ikenaga, M. Kobata, J. J. Kim, S. Uedaand K. Kobayashi,Applied Physics Letters, 2011,98, 133505.

31 M. Sowinska, T. Bertaud, D. Walczyk, S. Thiess, M. A. Schubert,M. Lukosius, W. Drube, C. Walczyk and T. Schroeder,Applied PhysicsLetters, 2012,100, 143501.

32 A. M. Kaiser, A. X. Gray, G. Conti, B. Jalan, A. P. Kajdos, A.Gloskovskii,S. Ueda, Y. Yamashita, K. Kobayashi, W. Drube, S. Stemmer andC. S.Fadley,Applied Physics Letters, 2012,100, 261603.

33 S. Tanuma, C. J. Powell and D. R. Penn,Surface and Interface Analysis,1993,20, 77–89.

34 M. Watanabe, E. Okunishi and K. Ishizuka.,Microscopy and Analysis,2009,23, 5–7.

35 M. R. Keenan and P. G. Kotula,Surface and Interface Analysis, 2004,36,203–212.

36 H. Y. Tan, J. Verbeeck, A. Abakumov and G. Van Tendeloo,Ultrami-croscopy, 2012,116, 24–33.

37 F. de Groot and A. Kotani,Core Level Spectroscopy of Solids, Taylor &Francis CRC press, 2008.

38 M. T. Otten, B. Miner, J. H. Rask and P. R. Buseck,Ultramicroscopy,1985,18, 285–289.

39 B. P. Cao, K. Suenaga, T. Okazaki and H. Shinohara,Journal of PhysicalChemistry B, 2002,106, 9295–9298.

40 NIST X-ray Photoelectron Spectroscopy Database, Version 3.5 (NationalInstitute of Standards and Technology, Gaithersburg, 2003); see also therefs. of the Suppl. Inf.

41 The assignment is unambiguous, because the spin-orbit separation of theTi 2p1/2-2p3/2 levels is about 6 eV, hence the 2p1/2 peaks (from the metaland all the other components) of each 2p doublet occur at larger bindingenergies and do not overlap with the observed 2p3/2 features.

42 D. N. H. Nam, L. V. Bau, N. V. Khiem, N. V. Dai, L. V. Hong, N. X.Phuc,R. S. Newrock and P. Nordblad,Physical Review B, 2006,73, 184430.

43 V. Kulkarni, K. R. Priolkar, P. R. Sarode, R. Rawat, A. Banerjee andS. Emura,Journal of Physics-Condensed Matter, 2008,20, 075203.

44 H. L. Ju, J. Gopalakrishnan, J. L. Peng, Q. Li, G. C. Xiong, T. Venkatesanand R. L. Greene,Physical Review B, 1995,51, 6143–6146.

45 The different Ti0/Ti4+ intensity ratio for the thicker Ti layer is due tothe smaller bulk sensitivity to the buried PCMO/Ti region dictated by theelectron IMFP.

46 S. Fujitsu and T. Hamada,Journal of the American Ceramic Society,1994,77, 3281–3283.

47 T. Bak, J. Nowotny, M. Rekas and C. C. Sorrell,Journal of Physics andChemistry of Solids, 2003,64, 1043–1056.

48 T. Bak, J. Nowotny, M. Rekas and C. C. Sorrell,Journal of Physics andChemistry of Solids, 2003,64, 1057–1067.

49 A. R. Bally, E. N. Korobeinikova, P. E. Schmid, F. Levy and F. Bussy,Journal of Physics D-Applied Physics, 1998,31, 1149–1154.

50 K. Ebata, H. Wadati, M. Takizawa, A. Fujimori, A. Chikamatsu, H. Ku-migashira, M. Oshima, Y. Tomioka and Y. Tokura,Physical Review B,2006,74, 064419.

51 A. Fujimori, A. Ino, J. Matsuno, T. Yoshida, K. Tanaka and T. Mizokawa,Journal of Electron Spectroscopy and Related Phenomena, 2002, 124,127–138.

52 R. Yasuhara, T. Yamamoto, I. Ohkubo, H. Kumigashira and M.Oshima,Applied Physics Letters, 2010,97, 132111.

53 S. Jeon and H. Hwang,Applied Physics Letters, 2002,81, 4856–4858.54 A. E. Bocquet, T. Mizokawa, K. Morikawa, A. Fujimori, S. R.Barman,

K. Maiti, D. D. Sarma, Y. Tokura and M. Onoda,Physical Review B,1996,53, 1161–1170.

1–7 | 7

Supplementary Information on: Chemical insight into

electroforming of resistive switching manganite heterostructures

F. Borgatti,1 C. Park,2 A. Herpers,2 F. Offi,3 R. Egoavil,4 Y. Yamashita,5 A. Yang,5

M. Kobata,5 K. Kobayashi,5 J. Verbeeck,4 G. Panaccione,6 and R. Dittmann2

1CNR – Istituto per lo Studio dei Materiali Nanostrutturati (ISMN),

via P. Gobetti 101, I-40129 Bologna, Italy∗

2Peter Gruenberg Institute (PGI-7) and JARA-FIT

Forschungszentrum Juelich GmbH, 52425 Juelich, Germany†

3CNISM and Dipartimento di Scienze, Università Roma Tre, Rome, Italy

4EMAT, Electron Microscopy for Materials Science,

University of Antwerp, Groenenborgerlaan 171, 2020 Antwerp, Belgium

5NIMS beamline Station at Spring-8,

National Institute for Materials Science, Sayo, Hyogo 679-5148, Japan

6CNR – Istituto Officina dei Materiali (IOM),

Laboratorio TASC, S.S.14, Km 163.5, I-34149 Trieste, Italy

1

Figure S 1: Sketch of the sample preparation. (1) The STO substrate. (2) SRO and (3) PCMO

deposition by PLD. (4) Metal deposition by DC sputtering. (5) TE patterning by photolithography

and etching. (6) The completed Ti/PCMO/SRO stack structure.

I. SAMPLE PREPARATION FOR I-V CHARACTERIZATION

Resistive switching Ti/PCMO/SrRuO3 (SRO) cells with stack structures with metal-

oxide-metal (MOM) configuration have been fabricated. A Pr0.5Ca0.5MnO3 (PCMO) thin

film was deposited on an epitaxial SRO layer on a single crystal SrTiO3 (STO) substrate by

reflection high energy electron diffraction (RHEED) assisted pulsed laser deposition (PLD).

The STO substrate has been TiO2-terminated by a treatment with buffered HF and anneal-

ing for 2 hours in air at 950 C resulting in a surface roughness of 0.16 nm (RMS). The 30 nm

SRO has been grown epitaxially in step-flow growth mode. During the deposition in oxygen

atmosphere of 0.133mbar, the substrate temperature was kept at 650 C and the laser energy

density was set to 1 J/cm2. The PCMO layer was subsequently deposited in a layer-by-layer

growth mode using the same oxygen atmosphere, but a fluency 2.8 J/cm2 and a temperature

of 700 C. The layer stack was cooled down in 1 bar oxygen. In the topography of the PCMO

thin film, the step-terrace structure of the substrate is reproduced with a roughness of 0.2

nm (RMS). Both films have a X-ray diffraction rocking curves width smaller than 0.045

(FWHM). To complete the MOM structure, a Pt/Ti was deposited on the PCMO/SRO

thin film by DC magnetron sputtering and the layered structure was patterned by optical

lithography and a dry etching process. An ohmic contact to the SRO layer is provided by

using Al wire bonding to a chip carrier with Pt pads.

2

II. HOMOGENEOUS VIRGIN AND FORMING STATES

Impedance measurements are very useful tools to separate different resistance contribu-

tions by using a Cole-Cole plot with a simulation of an equivalent circuit. Fig. S2(a) shows

impedance spectra of virgin states with four different pad sizes. Each impedance spectrum

shows two semi-circle combinations which prove that two parallel RC components exist in

series. The interface resistance and the bulk resistance were obtained by equivalent circuit

simulation. The interface resistance is one order higher that the bulk resistance and both

of them clearly show area dependence. Therefore, the resistance of as prepared Ti/PCMO

contacts with mixed layers is distributed homogeneously over the entire electrode area. The

reliability of the electro-forming process can be demonstrated by using an automatic mea-

surement system with a patterned mask map able to probe more than 100 devices patterned

on the same substrate.

Figure S3 shows impedance spectra of the formed states of samples with four different

pad sizes. One semi-circle was observed because the bulk resistance is too small compared to

the resistance after forming. The resistance after the forming process has increased of about

one order of magnitude for all different pad sizes. In Fig. S3(c) the log resistance is in inverse

proportion to the log area in which the resistance after forming is nearly homogeneous over

the entire electrode area as well.

III. COMMENTS ON THE IMPEDANCE ANALYSIS

A. IRS model

In general, the impedance of a metal-semiconductor junction can be modeled as two paral-

lel R-C circuits associated with the Schottky barrier at the interface and a semiconductor-like

behaviour in the bulk. However, in our system, the resistance of the PCMO bulk can be

ignored in the simulations according to its low resistivity, which we could confirm in our

devices by its ohmic contact to Pt and SRO.

3

Figure S 2: Homogeneous resistance of virgin states. (a) Cole-cole diagrams of virgin states with

four different pad sizes. (b) The equivalent circuit. (c) Area dependence of the interface resistance

and the bulk resistance from simulation of the equivalent circuit.

Figure S 3: Homogeneous resistance of forming states. (a) Cole-cole diagrams of forming states

with four different pad sizes. (b) Area versus resistance measured at 100 mV of virgin and forming

states.

B. IRS model

The total impedance of the anisotropic p-n junction should be a series combination of

the depletion layer capacitance on the two sides of the junction and the remaining n-type

TiOx layer and the reduced PCMO interface, but we are unable to simulate our data with

4 semicircles. The possible explanation is that (a) the TiOx layer could be considered as

a highly doped n-type conducting semiconductor and could therefore be ignored for the

simulations, and (b) if the entire reduced PCMO interface belongs to the p-type depletion

4

layer, its contribution would also not be visible in the impedance spectra.

IV. SAMPLE PREPARATION FOR TEM/EELS

The TEM samples preparation were performed by Focused Ion Beam (FIB) under exactly

the same conditions for devices in the IRS and HRS state fabricated on the same substrate.

The devices are indicated by label M8 and M9 for IRS and HRS, respectively. Each slice

was chosen horizontally (i.e. parallel to the X direction) as is shown in the figure. No

deformation was observed after electroforming.

Figure S 4: SEM images of the FIB lamella for TEM measurements.

V. SAMPLE PREPARATION FOR HAXPES

Due to the large footprint of the Sychrotron beam, single devices were not suitable for

HAXPES measurements and therefore device arrays were fabricated. Fig. S5(a) shows a

sketch of the sample layout as seen from above. It consists of a large number of 100x100

µm2 pads with 10 µm spacing between each other, and organized in stripes about 7 mm

long and 0.8 mm wide, where every stripe consist of pads in the same resistive state. In this

way the X-ray footprint illuminate about 30 pads in the same resistive state. Samples with

a thickness of the Ti layer of 10 nm and 7 nm, respectively, were investigated in order to

probe different depth regions of the heterostructures. Reference spectra for the SRO were

obtained from the part of the sample not covered by the heterostructures, while for the

5

Figure S 5: (a) A schematic layout of the samples and (b) the experimental geometry.

Figure S 6: The threshold DC forming example.

PCMO were acquired from a bare bulk PCMO thick film (50 nm) deposited on STO.

To program the pads for the first forming states an automatic measurement was used

with a patterned mask map. A threshold DC forming program was used, in which constant

10 V is applied and the measurement is terminated when the resistance reaches a threshold

resistance (5 K ohm), as shown in Fig. S6.

VI. HAXPES CORE-LEVEL FITTING

The Ti 2p core level spectra have been fitted in the binding energy region 450 – 472 eV.

They are partially overlapped to the Ru 3p3/2 peak, hence the lineshape of the Ru 3p3/2

peak, measured for the bare SrRuO3, has been used in the fit analysis. The experimental

spectra and the corresponding fit of the Ti 2p spectral region are shown in Fig. S7 where

each panel corresponds to the spectrum from a sample with definite Ti layer thickness (7

6

450 452 454 456 458 460 462 464 466 468 470 472450 452 454 456 458 460 462 464 466 468 470 472

Ru 3p3/2

Ti1+Ti2+Ti0

Ti4+

Ti2+Ti2+

Ti1+Ti0

data fit Ti 2p

3/2

Ti 2p1/2

Ru 3p back.

AH058 Virgin

Ti0

Ti4+

Ti3+

AH058 HRS

Binding energy (eV)

AH060 Virgin

AH060 HRS

Binding energy (eV)

Figure S 7: Fit of the Ti 2p core level spectral region from the samples as indicated in the panels.

or 10 nm) and in a definite resistive state (initial or high resistive state). Open circles are

experimental data and red solid line is the best fit to the data, while the dashed line is the

integral background. In the fitting curve we have the following components:

• Ru 3p3/2 component shown as green line. This curve changes only for a multiplicative

factor

• Ti0 : metallic titanium

• Ti1+: non stoichiometric titanium oxide or TiC or TiN. This second supposition is

drawn from the binding energy difference between this component and the Ti0 com-

ponent, as inferred from Ref. 1

• Ti2+: titanium monoxide (TiO)

• Ti3+: titanium sesquioxide (Ti2O3)

• Ti4+: titanium dioxide (TiO2)

Every peak used to fit the Ti 2p levels is a Voigt profile, whose Gaussian part shall

represents the experimental energy resolution and the Lorentzian width shall be a measure

of the lifetime. The Gaussian full width at half maximum (FWHM) was fixed to be equal for

all the peaks and free to vary. Intensity and energy position of the components associated

7

Ti 10 nm IRS Ti 10 nm HRS Ti 7 nm IRS Ti 7 nm HRS

(eV) (eV) (eV) (eV)

Ti0 453.95 453.98 454.10 454.13

Ti1+ 454.67 454.66 454.77 454.82

Ti2+ 455.66 455.43 455.79 455.79

Ti3+ 456.75 456.69 457.32 456.99

Ti4+ 458.95 458.95 458.90 458.96

∆Ti1+ 0.72 0.68 0.67 0.69

∆Ti2+ 1.51 1.45 1.68 1.66

∆Ti3+ 2.79 2.71 3.22 2.86

∆Ti4+ 5.00 4.97 4.80 4.83

S.O. 6.17 6.16 6.06 6.17

Table I: Binding energy position of the peaks associated to the different Ti 2p3/2 components. The

obtained spin-orbit splitting is also indicated (SO) and the energy difference between the different

Tii components (i=1+,. . . ,4+) and the Ti0 component is also calculated and indicated as ∆Tii.

to the Ti 2p3/2 level were free parameters, while the energy position of the components

associated to the Ti 2p1/2 level was fixed to literature results2, and their intensity fixed to

the statistical ratio of 3:2. For what concern the energy position we obtained the results

reported in Table I. Note that the results reported are in good agreement with literature

results2,3.

∗ Electronic address: francesco.borgatti@cnr.it

† Electronic address: park@IWE.RWTH-Aachen.de

1 S. V. Didziulis, Langmuir 1995, 11, 917

2 J. T. Mayer, U. Diebold, T. E. Madey, and E. Garfunkel, J. Electron Spectrosc. Relat. Phenom.

1995, 73, 1

3 E. McCafferty and J.P. Wightman, Appl. Surf. Sci. 1999, 143, 92

8