Spectromicroscopy of Tantalum Oxide Memristors ͒ John Paul Strachan,1,A Gilberto Medeiros-Ribeiro,1 J

APPLIED PHYSICS LETTERS 98, 242114 ͑2011͒

Spectromicroscopy of tantalum oxide memristors ͒ John Paul Strachan,1,a Gilberto Medeiros-Ribeiro,1 J. Joshua Yang,1 M.-X. Zhang,1 ͒ ͒ Feng Miao,1 Ilan Goldfarb,1,b Martin Holt,2 Volker Rose,3 and R. Stanley Williams1,c 1nanoElectronics Research Group, HP Labs, 1501 Page Mill Road, Palo Alto, California 94304, USA 2Center for Nanoscale Materials, Argonne National Laboratory, Argonne, Illinois 60439, USA 3Advanced Photon Source, Argonne National Laboratory, Argonne, Illinois 60439, USA ͑Received 16 April 2011; accepted 17 May 2011; published online 17 June 2011͒ We report experiments to measure material changes in tantalum oxide-based memristive devices. The high endurance and low power demonstrated in this material system suggests a unique mechanism for the switching, which we investigated using x-ray based spectromicroscopy and nanospectroscopy. Our study nondestructively identiﬁed a localized ͑Ͻ150nm diameter͒ Ta-rich phase surrounded by nano- or polycrystalline Ta2O5.© 2011 American Institute of Physics. ͓doi:10.1063/1.3599589͔

Thin film metal oxides exhibit a dynamical resistance extending across the substrate and was sputter-deposited switching, or memristive, behavior with the application of an from a tantala target with an Ar pressure of about 1.5mTorr electrical bias across a metal/oxide/metal ͑MOM͒ device at 270 °C substrate temperature. Transmission electron mi- structure. This phenomenon is pervasive across many mate- croscopy measurements of an identically grown film re- rial systems and has been studied1–6 particularly for develop- vealed a uniform amorphous structure, while core-level x-ray ing a high density nonvolatile memory. These motivations photoelectron spectroscopy analysis indicated that the tanta- have driven significant recent advances in device develop- lum was present in multiple oxidation states: 92% Ta5+,4% 7–12 ment and materials research. One promising material sys- Ta4+, and 4% Ta0. Thus, the film is essentially amorphous 13,14 ͑ ͒ tem is tantalum oxide, which has demonstrated over tantalum pentoxide a-Ta2O5 with a high concentration of 9 10 cycles of write/erase endurance, relatively low power oxygen vacancies. A top electrode of Pt, rather than Ta,13 operation, and no required electroforming step. As yet, there was used here to ensure that the only Ta present in the device is little understanding of the microscopic switching mecha- was from the switching oxide layer, thus eliminating a po- nism, including, for example, the presence and material com- tential source of ambiguity in the following materials char- position of any conductive channels. acterization. Generally speaking, the chemical and structural charac- The devices with Pt top electrodes were highly resistive terization of resistive switching centers is challenging, be- in the virgin state and required an initial electroforming step cause of the extremely small volume of material involved to enable bipolar resistive switching. This was performed and the desire to perform analyses nondestructively. Many Ϫ 15–19 with a 10 V bias applied to the top electrode with the resistance switching systems are now known to operate ͓ ͑ ͒ ͔ ͑ bottom electrode grounded Fig. 1 a , inset While this elec- based on the formation of localized anion-deficient e.g., troforming step can be eliminated by using a Ta top electrode oxygen vacancy rich͒ channels. However, many of the tech- 13 and a thinner tantalum oxide switching layer, the bipolar niques applied require special device preparation, are de- switching operation following the electroforming step ͓Fig. structive, or operate on structures substantially different from 1͑a͔͒ is qualitatively very similar in both types of devices. the usual MOM devices. Additionally, none of the heretofore Following electroforming and switching, the memristive studied systems has exhibited the high endurance of tantalum device was studied at the Advanced Photon Source ID-26 oxide, thus making a material understanding of this system Hard X-ray Nanoprobe beamline.20 A Fresnel zone plate fo- particularly important to guide future device improvements. ϳ With this goal, we performed a synchrotron-based micros- cused the x-rays to a 70 nm full width at half maximum copy and spectroscopy study after the electrical cycling of a spot size at the sample. X-rays were incident nearly perpen- tantalum oxide resistance switching device without using any dicular to the plane of the sample while being scanned later- ally using an optomechanical nanopositioning system based destructive sample preparation. We discovered a localized 21 ͑Ͻ150 nm diameter͒ Ta-rich channel within a 7.5 on laser doppler interferometry. The element-specific fluo- ϫ7.5 ␮m2 device. The Ta-rich channel was surrounded by a rescence was detected at each point in the scan allowing modified phase of the oxide, which had a spectral signature simultaneous mapping of the different material components distinct from the as-grown layer. We discuss the most likely in the device as well as more detailed material information interpretation for the observed spectroscopy and microscopy. by sweeping the incident x-ray energy. In this case, the pho- Our memristive device had a bottom electrode of Pt ton energy was scanned around the Ta L3 absorption edge ͑15 nm͒/Ti ͑5nm͒, a top electrode of Pt ͑30 nm͒ and a ͑ϳ10 keV͒ with a fine structure that is sensitive to micro- switching layer of tantalum oxide ͑22 nm͒ fabricated on a scopic structure, valence state, and chemical bonding infor- / SiO2 Si substrate. The tantalum oxide film was unpatterned, mation. The probing depth of the fluorescence measurements was longer than the thickness of the entire device material ͒ a Electronic mail: [email protected]. stack. No pre- or postpreparation of the sample was required, ͒ b On sabbatical leave from Tel Aviv University, Israel. making this technique well-suited for characterizing and im- ͒ c Electronic mail: [email protected]. aging all material layers within a standard resistance switch-

0003-6951/2011/98͑24͒/242114/3/$30.0098, 242114-1 © 2011 American Institute of Physics Downloaded 17 Jun 2011 to 146.137.70.71. Redistribution subject to AIP license or copyright; see http://apl.aip.org/about/rights_and_permissions 242114-2 Strachan et al. Appl. Phys. Lett. 98, 242114 ͑2011͒

9885 eV 9889 eV (a) ON switching (a) (b) (c) 100 10-5 Forming -8 ) 80 I(A) 10 50 (c) A) -11 10 a.u. (

μ 100 nm -8 -6 -4 -2 0 60 Ta L (d) (e) 0 (b) 3 9893 eV 9985 eV (d) -50 Current ( 40 -100 OFF switching (e) (f) 20 Ta-rich -0.8 -0.4 0.0 0.4 New

Fluorescence Yield TaO Device Voltage (V) x phase (b) (c) 0 a-Ta O Pt Ta 9880 9920 9960 2 5 X-ray energy (eV)

͑ ͒͑͒ FIG. 2. Color a Ta L3 x-ray absorption spectroscopy of the as-grown a-Ta2O5 film. Dashed vertical lines indicate the x-ray energies at which microscopy was performed. ͓͑b͒–͑e͔͒ High resolution x-ray fluorescence microscopy zoomed in on the region found in Fig. 1͑c͒. 2μm istry or structure.23 Thus, this spectrum served as a guide and FIG. 1. ͑Color͒ Electroforming and switching of a tantalum oxide-based Figs. 2͑b͒–2͑e͒ show high resolution scans taken at different crossbar memristor. ͑a͒ Device was cycled ON and OFF, after a forming step x-ray energies near and within this absorption line. The im- ͑inset͒, exhibiting bipolar operation. X-ray fluorescence microscopy follow- ͑ ͒ ing electrical operation showing the Pt ͑b͒ and Ta ͑c͒ elemental signal. ͑c͒ age at 9985 eV, Fig. 2 e , is in the postedge energy region, Within junction, a Ͻ400 nm region with enhanced signal was observed and therefore, revealing elemental Ta concentration levels, and is indicated by an arrow. shows a spatial region ͑brighter, near center of image͒ with a higher Ta concentration by up to 8.9% ͑Ϯ0.7%͒.Itisob- served in Figs. 2͑b͒–2͑d͒ that surrounding this Ta-rich spot is ing cross-bar device, even buried below thick contacting a region showing another material change. This is evident by electrodes. the different absorption contrast in this surrounding region, Figures 1͑b͒ and 1͑c͒ show imaging of the tantalum ox- evolving from brighter than neighboring regions at 9885 eV ide device, detecting fluorescence from Pt and Ta, respec- to darker ͑a 6.1% reduced intensity͒ at 9889 and 9893 eV. tively, in response to incident x-rays at an energy of 9885 eV. This semicircular region in the device had an absorption The x-ray energy was chosen to be within the steep absorp- spectrum that differed from the as-grown film in Fig. 2͑a͒, tion edge of the Ta L3 resonance, allowing spatially localized yet has an unchanged Ta concentration, thus, showing a more material changes in the tantalum oxide layer to be more eas- subtle and localized material change. Figure 2͑f͒ is a com- ily detected. The vertical and horizontal bright regions in posite image formed by an overlay of the regions for this Fig. 1͑b͒ show the bottom and top Pt electrodes, respectively, ͑ ͒ ͑ ͒ new phase red , the Ta-rich region blue , and the a-Ta2O5 with the bottom electrode less bright due to a reduced thick- ͑black͒ obtained from Figs. 2͑b͒–2͑d͒. ness as well as partial absorption of the fluorescence by the The stability of the Nanoprobe beamline allowed the oxide layer. The measured Ta fluorescence in Fig. 1͑c͒ might performance of nanospectroscopy: sweeping the incident be expected to be spatially uniform since the oxide layer was x-ray energy while holding the position of the x-ray focus unpatterned but attenuation by the top electrode caused the fixed within a 70 nm spot of the sample to measure the appearance of a darker horizontal stripe. A bright spot near localized absorption spectrum. The nanospectroscopy results the bottom of the junction area was observed in Fig. 1͑c͒, indicated with an arrow. This spatially nonuniform Ta fluo- 9000 a-Ta O as-grown rescence signal was less than 400 nm in diameter, consistent 2 5 New TaOx phase with previously observed conductive channels in titanium 8000 oxide memristors with similar device size and power applied 7000 during operation.22 The brighter contrast of this spot is consistent with ͑1͒ an increased Ta concentration, ͑2͒ a chemi- 6000 ͑ ͒ 5000 cally or structurally altered local region of the oxide, or 3 100 nm an enhanced fluorescence signal due to reduced top electrode 4000 thickness or another enhancement mechanism. Possibility ͑3͒ Fluorescence Yield (a.u.) 3000 is unlikely as there was no indication of less Pt in this region, as evident from Fig. 1͑b͒ and additional higher resolution 2000 scans. To investigate further, high resolution 1 ␮mϫ1 ␮m 9880 9890 9900 9910 images of this region were acquired at several important X-ray energy (eV) ͑ ͒ x-ray energies Fig. 2 and nanospectroscopy was performed FIG. 3. ͑Color͒ Nanospectroscopy of two different Ͻ70 nm spots within as well ͑Fig. 3͒. the device. Inset shows the different material phase regions found in Fig. 2 Figure 2 shows spectromicroscopy measurements of the and one of the spatial locations for nanospectroscopy is indicated by a circle. device region of interest, starting with a reference Ta L The second location is roughly two micrometers away. Within the region of 3 interest, the spectrum ͑red͒ shows a reduced absorption intensity and a 0.2 spectrum of the as-grown, primarily a-Ta2O5 layer in Fig. eV energy shift in the peak compared to the as-grown film ͑black͒, suggest- 2͑a͒. Changes in the fine structure of this absorption spec- ing both a subtle composition change and increasing short-range order of the trum can reveal regions in the layer having an altered chem- initially amorphous film. Downloaded 17 Jun 2011 to 146.137.70.71. Redistribution subject to AIP license or copyright; see http://apl.aip.org/about/rights_and_permissions 242114-3 Strachan et al. Appl. Phys. Lett. 98, 242114 ͑2011͒

are shown in Fig. 3 in which two spots in the device, one here, and the present analysis should provide a framework within the modified phase and another roughly two microme- for understanding this operation. ters away, as indicated in the Fig. 3 inset, were analyzed. The observed spectrum was consistent with the microscopy of We thank J. Borghetti, M. D. Pickett, and W. Yi for Fig. 2; the absorption intensity matched the as-grown region helpful discussions. Work at HP is sponsored by the U.S. in the pre-edge and postedge energies while showing reduced Government’s Nano-Enabled Technology Initiative. Use of the Advanced Photon Source and the Center for Nanoscale absorption within the Ta L white line, resulting in the darker 3 Materials was supported by the U.S. Department of Energy, contrast observed within this region in Figs. 2͑c͒ and 2͑d͒. Office of Science, Office of Basic Energy Sciences, under Moreover, there was an overall reduced intensity through the Contract No. DE-AC02-06CH11357. white line from 9880–9900 eV. The peak intensity was re- ͑ duced by 6.8% while the integrated area proportional to the 1G. Dearnaley, A. M. Stoneham, and D. V. Morgan, Rep. Prog. Phys. 33, unoccupied valence states͒ was reduced by 3.1%. There was 1129 ͑1970͒. also an observable shift in the peak to lower energy by ap- 2R. Waser, R. Dittman, G. Staikov, and K. Szot, Adv. Mater. ͑Weinheim, proximately 0.2 eV. Prior studies24 comparing the Ta L Ger.͒ 21, 2632 ͑2009͒. 3 3A. Sawa, T. Fuji, M. Kawasaki, and Y. Tokura, Appl. Phys. Lett. 85,4073 x-ray absorption spectra in Ta2O5 amorphous and crystalline ͑2004͒. films have shown similar spectral features, specifically a de- 4C. Schindler, S. C. P. Thermadam, R. Waser, and M. N. Kozicki, IEEE creased white line absorption in the crystalline film which Trans. Electron Devices 54, 2762 ͑2007͒. 5 was attributed to reduced covalent ͑increasingly ionic͒ Ta–O S. Q. Liu, N. J. Wu, and A. Ignatiev, Appl. Phys. Lett. 76, 2749 ͑2000͒. 6D. B. Strukov, G. S. Snider, D. R. Stewart, and R. S. Williams, Nature bonding. Thus, a likely interpretation is that this nanoscale ͑London͒ 453,80͑2008͒. 7 region has undergone a change from an initially a-Ta2O5 to a K. Szot, W. Speier, G. Bihlmayer, and R. Waser, Nature Mater. 5,312 ͑ ͒ nano- or polycrystalline Ta2O5. Our nanospectroscopy work 2006 . 8B. P. Andreasson, M. Janousch, U. Staub, and G. I. Meijer, Appl. Phys. thus suggests that a crystallization of the Ta2O5 has occurred Lett. 94, 013513 ͑2009͒. in this region of the device. 9 S. C. Chae, J. S. Lee, S. Kim, S. B. Lee, S. H. Chang, C. Liu, B. Kahng, The exact location of the nano- or polycrystalline Ta2O5 H. Shin, D.-W. Kim, C. U. Jung, S. Seo, M.-J. Lee, and T. W. Noh, Adv. phase surrounding the more metallic, Ta-rich channel ͓Fig. Mater. ͑Weinheim, Ger.͒ 20, 1154 ͑2008͒. ͑ ͔͒ 10A. Odagawa, Phys. Rev. B 70, 224403 ͑2004͒. 2 e suggests that the crystallization was driven by Joule 11 heating during electrical operation. Prior studies of initially M. J. Rozenberg, I. H. Inoue, and M. J. Sánchez, Appl. Phys. Lett. 88, 033510 ͑2006͒. amorphous films of Ta2O5 having a comparable thickness 12D. S. Jeong, H. Schroeder, and R. Waser, Phys. Rev. B 79, 195317 ͑2009͒. showed that crystallization occurred at temperatures greater 13J. J. Yang, M.-X. Zhang, J. P. Strachan, F. Miao, M. D. Pickett, R. D. than 600 °C.25,26 Such structural changes surrounding a Kelley, G. Medeiros-Ribeiro, and R. S. Williams, Appl. Phys. Lett. 97, 232102 ͑2010͒. conductive channel have also been observed in TiO2-based 14 19,22 27 J. H. Hur, M.-J. Lee, C. B. Lee, Y.-B. Kim, and C.-J. Kim, Phys. Rev. B memristors with similar temperature levels attained 82, 155321 ͑2010͒. from the Joule heating in a nanoscale channel. Since oxygen 15O. Heinonen, M. Siegert, A. Roelofs, A. K. Petford-Long, M. Holt, K. was not probed in this study, we were unable to determine d’Aquila, and W. Li, Appl. Phys. Lett. 96, 103103 ͑2010͒. 16 the precise composition of the Ta-rich phase, which will re- T. Fujii, M. Arita, K. Hamada, H. Kondo, H. Kaji, Y. Takahashi, M. Moniwa, I. Fujiwara, T. Yamaguchi, M. Aoki, Y. Maeno, T. Kobayashi, quire different techniques. and M. Yoshimaru, J. Appl. Phys. 109, 053702 ͑2011͒. Our work highlights that resistance switching, especially 17K. Fujiwara, T. Nemoto, M. J. Rozenberg, Y. Nakamura, and H. Takagi, the electroforming step, in tantalum oxide memristors in- Jpn. J. Appl. Phys. 47, 6266 ͑2008͒. 18 volves a bulk material change in the oxide layer itself, elimi- D.-H. Kwon, K. M. Kim, J. H. Jang, J. M. Jeon, M. H. Lee, G. H. Kim, X.-S. Li, G.-S. Park, B. Lee, S. Han, M. Kim, and C. S. Hwang, Nat. nating several mechanisms, for example, the migration of Nanotechnol. 5, 148 ͑2010͒. electrode ͑Pt͒ atoms, or purely interface effects. Instead, we 19J. P. Strachan, M. D. Pickett, J. J. Yang, S. Aloni, A. L. D. Kilcoyne, G. directly observe ͑1͒ a Ta-rich channel surrounded by ͑2͒ an Medeiros-Ribeiro, and R. S. Williams, Adv. Mater. ͑Weinheim, Ger.͒ 22, altered material phase which was identified as Ta O with 3573 ͑2010͒. 2 5 20J. Maser, R. Winarski, M. Holt, D. Shu, C. Benson, B. Tieman, C. Preiss- additional short-range order, both of which evolved from the ner, A. Smolyanitskiy, B. Lai, S. Vogt, G. Wiemerslage, and G. B. initially amorphous film, which was multivalent, but mostly Stephenson, Proceedings of the Eighth International Conference on X-ray ͑ Ta2O5. This provided evidence for local heating to a tem- Microscopy, edited by S. Aoki, Y. Kagoshima, and Y. Suzuki, IPAP, Himeji, Japan, 2005͒, pp. 26–29. perature greater than 600 °C. The only thermodynamically 21 ͑ ͒ D. Shu, J. Maser, M. Holt, R. Winarski, C. Preissner, A. Smolyanitskiy, B. stable phases in the Ta–O system are a metallic Ta O solid Lai, S. Vogt, and G. B. Stephenson, AIP Conf. Proc. 879, 1321 ͑2007͒. 22 solution and an insulating Ta2O5. Thus, it is likely that any J. P. Strachan, J. J. Yang, R. Münstermann, A. Scholl, G. Medeiros- metastable phases evolve and segregate in favor of these two Ribeiro, D. R. Stewart, and R. S. Williams, Nanotechnology 20, 485701 phases, particularly, during electrical operation and the asso- ͑2009͒. 23J. Stöhr, NEXAFS Spectroscopy ͑Springer, Berlin, 1992͒. ciated Joule heating. Growth of a metallic Ta-rich channel 24G. G. Long, A. G. Revesz, and M. Kuriyama, J. Non-Cryst. Solids 70, within the insulating Ta2O5 matrix will increase the conduc- 271 ͑1985͒. tivity ͑ON switching͒, while reoxidation ͑OFF switching͒ 25G. S. Oehrlein, F. M. d’Heurle, and A. Reisman, J. Appl. Phys. 55, 3715 ͑1984͒. will decrease it. While more work is needed to understand 26 the kinetics and electric field dependence of the observed E. Atanassova, N. Novkosvski, A. Paskaleva, and M. Pecovska- Gjorgjevich, Solid-State Electron. 46, 1887 ͑2002͒. material changes during ON and OFF switching, such a 27J. P. Strachan, D. B. Strukov, J. Borghetti, J. J. Yang, G. Medeiros-Ribeiro, study is possible with the nondestructive technique employed and R. S. Williams, Nanotechnology 22, 254015 ͑2011͒.

Downloaded 17 Jun 2011 to 146.137.70.71. Redistribution subject to AIP license or copyright; see http://apl.aip.org/about/rights_and_permissions