Letter pubs.acs.org/NanoLett
Three-Dimensional Observation of the Conductive Filament in Nanoscaled Resistive Memory Devices Umberto Celano,*,†,‡ Ludovic Goux,† Attilio Belmonte,† Karl Opsomer,† Alexis Franquet,† Andreas Schulze,† Christophe Detavernier,†,§ Olivier Richard,† Hugo Bender,† Malgorzata Jurczak,† and Wilfried Vandervorst†,‡ †
IMEC, Kapeldreef 75, B-3001 Heverlee (Leuven), Belgium Department of Physics and Astronomy (IKS), KU Leuven, Celestijnenlaan200D, 3001 Leuven, Belgium § Universiteit Gent, Krijgslaan 281 (S1), 9000 Gent, Belgium ‡
S Supporting Information *
ABSTRACT: The basic unit of information in filamentary-based resistive switching memories is physically stored in a conductive filament. Therefore, the overall performance of the device is indissolubly related to the properties of such filament. In this Letter, we report for the first time on the threedimensional (3D) observation of the shape of the conductive filament. The observation of the filament is done in a nanoscale conductive-bridging device, which is programmed under real operative conditions. To obtain the 3Dinformation we developed a dedicated tomography technique based on conductive atomic force microscopy. The shape and size of the conductive filament are obtained in three-dimensions with nanometric resolution. The observed filament presents a conical shape with the narrow part close to the inert-electrode. On the basis of this shape, we conclude that the dynamic filament-growth is limited by the cation transport. In addition, we demonstrate the role of the programming current, which clearly influences the physical-volume of the induced conductive filaments. KEYWORDS: Conductive filament, resistive switching, C-AFM, CBRAM, AFM-tomography hanks to tremendous effort in engineering the flash NAND device, this technology has recently entered the sub-20 nm node.1 The prospects to continue the scaling of these memory devices beyond the 15 nm node are uncertain, mainly due to reliability issues and cross-talk limitations.1 For these reasons, the community is constantly looking for a valuable alternative for future nonvolatile storage. Because of its fast operation, low power consumption, high endurance, and high scalability, filamentary-based resistive switching memory (RRAM) is considered as a promising candidate for next generation nonvolatile memories.2−5 RRAM is based on materials exhibiting an abrupt change in electrical resistance when submitted to a voltage pulse. In RRAM, the information is stored using two/more very distinct resistance levels assumed by the device. In general, the different resistance states are induced by the formation or dissolution of a highly conductive path, that is, filament, into a poorly conductive medium.2,4,6,7 Among RRAMs, conductive bridging memories (CBRAM) are a class of devices which use a solid-state electrochemical reaction with Ag or Cu, to create or dissolve conductive filaments (CFs) into a solid electrolyte.4,8 The memory element is based on a metal−insulator−metal (MIM) structure, where a thin dielectric is sandwiched between an active electrode and an inert counterelectrode. Ag or Cu are commonly used as the active electrode in view of their property to inject metallic
T
© XXXX American Chemical Society
cations within the solid electrolyte.8−10 A variety of oxide, chalcogenide, and sulfide thin films have been proposed for the solid electrolyte.4 The inert counter-electrode is generally made of Pt, W, or TiN. When a positive voltage is applied to the active electrode, a field-assisted injection and transport of cations begins. This leads to the creation of the CF inside the switching layer. The presence of the CF lowers dramatically the resistance of the device thereby defining a low resistive state (LRS). The CF can be dissolved by applying a negative voltage to the active electrode and thus restoring a high-resistance-state (HRS). The two different resistance states are used as the logic values 1 or 0 for data storage applications.4 CBRAM devices have shown good cycle stability (>108), long data retention (104 s), very large RON/ROFF ratio (>106), multilevel capability, small operation voltages ∼2 V, nanosecond switching speed, and small operating current.3,11−15 All these attractive properties rely on the nanoscaled CF embedded in the solid electrolyte when the device is programmed. The nucleation and growth of the CF is based on stochastic processes that occur in a highly confined volume, for example, few hundreds of cubic nanometers.4,16 As these CFs are embedded in Received: January 6, 2014 Revised: April 2, 2014
A
dx.doi.org/10.1021/nl500049g | Nano Lett. XXXX, XXX, XXX−XXX
Nano Letters
Letter
nanoscaled devices, their physical characterization is generally further complicated by the presence of many surrounding layers (conducting and insulating). These obstacles have hampered the observation of CFs in integrated devices with most of the known techniques. Though the morphology and chemical composition of the conductive filament has been observed using scanning electron microscopy (SEM) and transmission electron microscopy (TEM),17−21 a complete observation in three-dimensions of the CF in a scaled device is still missing. Nevertheless the full characterization of the conductive filament in real device would be beneficial for three main reasons. First, it would dramatically enhance our understanding of the filament growth dynamics. Second, more insight would become available to locate the critical interface during filament cycling, which is essential in device optimization. Finally the information about the shape and size of the CF would clarify the underlying physical mechanisms and contribute to advancing filamentary-based memory technology toward a reliable mature technology. In this work, we present for the first time a three-dimensional characterization of the CF in a nanoscaled conductive-bridging memory device. In order to perform this three-dimensional (3D) characterization, we developed a dedicated tomography technique based on a slice-and-view approach using conventional conductive atomic force microscopy (C-AFM) as the observing methodology. The device under investigation is a Cu/Al2O3/TiN-based memory, integrated in a one-transistor-one-resistor (1T1R) configuration (Figure 1a). The device is placed at the crosspoint between the bottom and top electrode (BE and TE). The memory element is based on a 5 nm amorphous Al2O3 sandwiched between 10 nm TiN (BE) and 40 nm Cu (TE), as shown in Figure 1b. In Figure 1c, a cross-sectional TEM image clearly shows the integration scheme (1T1R). The inset of Figure 1c shows the details of the Cu/Al2O3/TiN memory element. The memory cell (100 × 200 nm) is created by electron beam patterning as shown in the AFM-topographical image in Figure 1d. The BE is directly connected to the drain of the control transistor which is used to limit the current during the device operations (Figure 1a). This type of nonvolatile memory device shows attractive characteristics such as high ON/OFF resistance ratio (>106), good retention (104 s), high endurance (>106 cycles), and multilevel capability, similarly to stacks we reported elsewhere.13,14 Figure 2 shows the I−V behavior during the electrical programming of the cell, prior to the C-AFM tomography. The as-deposited device is normally in the OFF-state showing high resistance. The pristine cell requires an initial electroforming to activate the device. After electroforming, the formation and rupture of the CF can be triggered respectively by means of a positive or negative voltage on the Cu electrode. The Cu electrode is positively biased for the filament formation (SET), and negatively biased for the filament rupture (RESET). In our studies, the memory element is programmed using a semiconductor parameter analyzer. The control transistor delivers a current compliance of 10 μA during all the programming phases. In Figure 2, the forming voltage is about ∼3.5 V and the set and reset voltages are respectively ∼1.5 and −2 V. In our studies the device was subjected to few tens of cycles and then was placed in the AFM system for the C-AFM tomography. Beyond its application for oxide characterization, C-AFM has also been used in RRAM studies whereby the tip was used to act as a nanosized virtual-top-electrode on the sample’s surface.22−24 In this work however we used the tip to
Figure 1. (a) Schematic of the one-transistor-one-resistor configuration. (b) Details of the resistive-memory structure. (c) Crosssection TEM image of our device. The inset shows details of the crossbar memory element. (d) Topographical AFM image of the crosspoint area 100 × 200 nm.
Figure 2. Electrical programming of the 100 × 200 nm2 memory device, under DC voltage sweep. Note, the current is limited to 10 μA.
probe the local variations in conductivity. Figure 3a presents the topographical and current map (i.e., C-AFM analysis) taken on the complete device after filament formation. The presence of the top electrode shields the observation of the CF as clearly visible by C-AFM in Figure 3a. In this configuration, traditional C-AFM is completely ineffective for CF characterization. Our C-AFM tomography is based on a slice-and-view approach using traditional C-AFM technique (Supporting Information). In essence, we use the AFM tip also to remove material from the sample surface in a very controlled manner. The latter is B
dx.doi.org/10.1021/nl500049g | Nano Lett. XXXX, XXX, XXX−XXX
Nano Letters
Letter
Figure 3. (a) Planar 2D C-AFM, performed on our cross-point memory element (bottom inset) in SET-state. The C-AFM is completely ineffective due to the presence of the TE shielding the CF observation. (b) Schematic of the C-AFM tomography procedure, the diamond tip is exploited to collect several slices at different heights of the CF after the removal of the TE. (c) Over imposition of the collected 2D C-AFM slices, prior to the 3D interpolation. Note, the average space between each slice is ∼0.5 nm. (d) Collection of 2D slices constituting the data set for the 3D interpolation (scale bar 80 nm). The CF appears in the middle of the active area after top electrode removal. The highly conductive features on the top-left and bottom-right corners are the exposed parts of the TiN BE, which is progressively exposed during the removal of Al2O3.
enabled by the precise force-control of AFMs and the hardness of the diamond-tips we use for this application.25 Basically in each planar C-AFM scan, we probe the local variations in conductivity and partially erode a controlled amount of the sample surface. Hence, in essence we use the conductivediamond tip as a scalpel for a controlled material removal26,27 by physical scraping, and as a conductive probe to record the spatial variations of the tip−sample current. When applied to our memory device, we thus collect two-dimensional (2D) current maps of the CF at different depths (Figure 3b). Finally, (Figure 3c) the 2D images are combined into a 3D representation (tomogram) by a specialized software interpolation (Supporting Information). The C-AFM tomogram starts after the removal of the top electrode.27 Once the top electrode is removed and the Al2O3 surface is exposed, we record the first C-AFM slice of the collection (Figure 3d1). The CF is now clearly visible as a highly conductive spot in the middle of the active area, in contrast with the rest of the Al2O3 surface, which is still highly insulating (red background in Figure 3e). Different slices are collected at constant load force as in Figure 3d1−7. Using a removal rate of ∼0.5 nm/scan, the final collection constitutes 10 slices for the 5 nm thick switching layer. We stop the acquisition when the switching layer is completely removed and the highly conductive bottom electrode, for example, TiN, becomes visible. An analysis of the consecutive images clearly indicates that in contrast with the first C-AFM slice the last one before the bottom electrode presents a CF with a much smaller size, indicating the shrinkage of the CF from the active electrode (Cu) to the inert counter electrode (TiN). The conical shape, ending at the bottom electrode, becomes very clear from the full 3D-representation in Figure 4. In Figure 4a−c, the CF shows a shrinkage in the shape moving from TE to BE, the is, from Cu/oxide-interface to oxide/TiN. The area of the CF shrinks from ∼493 nm2 on the Cu-side down to ∼200 nm2 on the side of the inert-electrode. The morphology
Figure 4. (a) Schematic of the CF position, note the section axes for 3D observation. (b) Observation of the reconstructed 3D tomogram for the CF under investigation in 5 nm thick Al2O3. (c) The lowcurrent contribution in the tomogram is suppressed to enhance the contrast of the highly conductive features.
of the CF clearly presents two branches evolving through the Al2O3 toward the inert-electrode following a conical shape. Of the two branches, the major is fully connecting the electrodes, while the small one may be not fully connected. This cannot be detected by C-AFM due to the presence of a non-negligible tunneling current flowing through a possible nonconnected C
dx.doi.org/10.1021/nl500049g | Nano Lett. XXXX, XXX, XXX−XXX
Nano Letters
Letter
filament’s branch. The same effect imposes some limitations in the characterization of the reset (OFF) state using this technique, as presented later. In Figure 4c, we subtract all the low-current contribution in the 3D tomogram in order to show only the highly conductive features. This is likely representative of the shape of the CF, which is supposed to be the lower resistive path for the current inside the Al2O3. More 3D images and the animated cross-section of the tomogram are available in Supporting Information. A similar experiment is carried out for the characterization of the reset (OFF) state. To this end one memory cell has been formed and cycled using 50 μA current compliance. Higher operative currents are needed in order to induce more stable OFF states and to avoid the formation of weak filaments that could not be detected by C-AFM owing to the sensitivity of the current amplifier. Prior to the C-AFM tomography the device undergoes a reset voltage sweep that induces high resistance to the memory element. Figure 5a shows a partial 2D-set of slices
the OFF state. In Figure 5d, a cross-section of the gap region is shown, and the gap size can be estimated between 0.5 and 0.7 nm. The shape observed by C-AFM tomography is opposite to the one predicted for chalcogenide-based ECM cells.4,28 The difference emphasizes the critical role of the cation-transport through the electrolyte and thus the need for an optimized material-selection when targeting resistive switching. For chalcogenides, sulphites, or ionic-conductors, the cation mobility is much higher as compared to the case of amorphous dielectrics. Therefore, the cations can easily reach the counterelectrode, where they are reduced thereby initiating the nucleation of the CF. In this scenario the reduction and nucleation of the filament are the rate-limiting processes for the filament growth, and the CF starts to grow from the inertelectrode to the active electrode. Such a shape has experimentally been observed by Choi et al. in Cu-doped GeTe using TEM.18 In this case, the CF present a conical shape with the filament constriction close to the active electrode interface. On the other hand, our 3D tomogram unambiguously reveals that the CF has the narrow part close to the inert-electrode. To explain this different shape we suggest that the cation transport is the rate-limiting process for the CF formation in our devices. In such condition, the injected cations get reduced within the electrolyte nearby the injection interface and start forming the CF from the active anode side. Our observations are in agreement with TEM observations by Yang et al.17 in amorphous-silicon and by Liu et al. in ZrO2.19 In those cases we can expect solubility and diffusion coefficients similar to the case of amorphous-Al2O3. Those works have raised debates in the community about the possible mechanisms for reversed filament growth.29 To explain our results in more detail we describe the CF filament formation in our oxide-electrolytebased CBRAMs as a field-assisted ion transport inside the Al2O3 (Figure 6). Under the effect of positive bias, Cu atoms are ionized and diffuse into the switching layer Figure 6a. At the same time, the high electric field, ∼109 V m−1, might generate oxygen vacancies within the Al2O3. These might contribute to electronic current leakage throughout the electrolyte and also have an impact on the Cu migration due to the ionic transport along chains of O-sites of the Al−O bonding in the amorphous matrix.30 However, given the low cation mobility in the Al2O3, the Cu cations may only travel a short distance inside the dielectric before they are reduced by capturing electrons injected in the electrolyte under the high electric field. The Cu+ are reduced back to Cu near the active electrode and become an extension of it. The Cu+ ions that are subsequently injected will preferentially become reduced at the end of the already existing filament, as its end constitutes the point with the highest electric field. Hence the filament growth can be viewed as based on this repetition of ionization and injection, slow (limited) migration and reduction processes. Under these assumptions, the filament grows from the active electrode, for example, Cu, toward the inert-electrode. Note, our C-AFM tomography does not provide any compositional analysis of the filament, leaving still opened the debate on its chemical composition. The final conductive filament can be constituted of different conductive species such as Cun+ ions, Cu0, or Al− Cu−O phases. During the reset operation when an opposite polarity is applied, the rupture of the CF takes place due to an electrochemical and Joule-heating assisted process, restoring
Figure 5. (a) Collection of 2D slices constituting the data set for the 3D interpolation of the OFF state (scale bar 200 nm). (b) Electrical programming of the reset state, under negative DC voltage sweep. Note, the current is limited to 50 μA during forming and cycling. (c) Observation of the reconstructed 3D tomogram for the CF in the OFF state, in 5 nm thick Al2O3. (d) The 2D cross section extracted in the gap region (the low-current contribution is suppressed to enhance the contrast) the gap size is ∼0.5−0.7 nm.
collected for a device in the OFF state imposed after the negative voltage sweep in Figure 5b. During slicing, the current leakage in the Al2O3 shows a gradual recover in proximity of the inert-electrode interface (Figure 5a4−6) until the bottom electrode is exposed (Figure 5a7). The entire 2D set of C-AFM images (16 slices in total) is interpolated in three dimension. The 3D tomogram in Figure 5c shows the presence of an unconnected filament between the two electrodes in the case of D
dx.doi.org/10.1021/nl500049g | Nano Lett. XXXX, XXX, XXX−XXX
Nano Letters
Letter
Figure 6. Illustration of the electrochemical processes during resistive switching. (1) First, the Cu oxidizes and Cu+ ions are injected in the Al2O3. Second, the high electric field might lead to the formation of oxygen vacancies in the dielectric layers (white balls in the cartoon). (2) The slow migration of Cu+ ions in the switching layer implies that a reduction reaction occurs before the Cu+ reaches the inert-electrode. (3) The filament growth continues and the CF eventually shorts the two electrodes thereby creating the low resistive state. (4) When the bias is reversed, a Jouleheating assisted electrochemical reaction is responsible for the rupture of the CF in the point of max power dissipation, that is, CF constriction.
Figure 7. (a) Typical reset sweep for three different programming currents. The resistance of the CF tends to decrease as the programming current is increased. (b) Correlation between the resistance value and the CF physical-volume as induced by different programming currents the values for the CF physical volumes are extracted using C-AFM tomography.
the high resistance state for the device.31 It is clear that the CF will rupture at the position where the current density is the highest. The capability of storing multiple resistance-states in a single memory cell is one of the most important requirements for nonvolatile RRAMs because it can dramatically enhance the memory density. Specifically, different programming currents (Figure 7a) can be utilized to induce multiple resistance-states and therefore, multiple logical bits.14,32 In order to elucidate the nature of this multilevel bit capability; we studied the physical volume of the CF in relation to the differences in CF resistance as induced by different programming currents. Three different devices are programmed while limiting the programming current to 10−50−100 μA. The corresponding reset sweeps (Figure 7a) clearly indicate that filaments with different resistances have been formed. We measured the resistance of
the CF’s after the electrical programming and then performed the C-AFM tomography. From the obtained 3D tomograms we calculated the physical volume of the CF in the Al2O3 for each case. The correlation shown in Figure 7b indicates very clearly that a larger programming current induces a larger physical volume and that the larger volume leads to a lower resistance value. Hence by controlling the programming current, the volume of the CF and thus the resistance can be modulated. In conclusion, we systematically studied CFs in conductivebridging memory devices. For that purpose, we developed a dedicated slice-and-view 3D tomography technique, based on the combination of C-AFM with material removal using wearresistant, conductive diamond tips. The uniqueness of our concept is that it provides the CF observation in a normal, complete memory cell. We consistently find a conical shape with the narrowest part of the CF near the dielectric/inertE
dx.doi.org/10.1021/nl500049g | Nano Lett. XXXX, XXX, XXX−XXX
Nano Letters
Letter
electrode interface. Our findings suggest that the dynamics of the CF growth are not limited by the electrochemical dissolution of Cu and the reduction/nucleation of the Cu+ at the anode interface but rather by the limited mobility of the Cu in the electrolyte. This provides ample time for reduction and nucleation at the active electrode. In the light of this study, it is clear that this is the most sensitive interface to engineer for improved device performances. Furthermore, our study demonstrates the close correlation between the physical volume and the resistance (programming current) of the CF, as an explanation for the multilevel capability of CBRAM devices. This study broadens the fundamental understanding of filamentary resistive switching and introduces a novel, threedimensions characterization concept, which can be further explored for other possible applications as well.
■
(13) Goux, L.; Sankaran, K.; Kar, G.; Jossart, N.; Opsomer, K.; Degraeve, R.; Pourtois, G.; Rignanese, G.; Detavernier, C. Presented at Symposium on VLSI Technology (VLSIT)m Honolulu, HI, 2012; pp 69−70. (14) Belmonte, A.; Kim, W.; Chan, B.; Heylen, N.; Fantini, A.; Houssa, M.; Jurczak, M.; Goux, L. In 2013 5th IEEE International Memory Workshop (IMW), May 26−29, 2013, Monteray, CA; pp 26−29 (15) Yang, Y. C.; Pan, F.; Liu, Q.; Liu, M.; Zeng, F. Nano Lett. 2009, 9, 1636−43. (16) Yang, Y.; Lu, W. Nanoscale 2013, 5, 10076−10092. (17) Yang, Y.; Gao, P.; Gaba, S.; Chang, T.; Pan, X.; Lu, W. Nat. Commun. 2012, 3, 732. (18) Choi, S.-J.; Park, G.-S.; Kim, K.-H.; Cho, S.; Yang, W.-Y.; Li, X.S.; Moon, J.-H.; Lee, K.-J.; Kim, K. Adv. Mater. 2011, 23, 3272−7. (19) Liu, Q.; Sun, J.; Lv, H.; Long, S.; Yin, K.; Wan, N.; Li, Y.; Sun, L.; Liu, M. Adv. Mater. 2012, 24, 1844−9. (20) Aulin, C.; Karabulut, E.; Tran, A.; Wågberg, L.; Lindström, T. ACS Appl. Mater. Interfaces 2013, 3−10. (21) Xu, Z.; Bando, Y.; Wang, W.; Bai, X.; Golberg, D. ACS Nano 2010, 4, 2515−2522. (22) Szot, K.; Speier, W.; Bihlmayer, G.; Waser, R. Nat. Mater. 2006, 5, 312−320. (23) Qi, J.; Olmedo, M.; Zheng, J.-G.; Liu, J. Sci. Rep. 2013, 3, 2405. (24) Lee, M. H.; Hwang, C. S. Nanoscale 2011, 3, 490−502. (25) Hantschel, T.; Demeulemeester, C.; Eyben, P.; Schulz, V.; Richard, O.; Bender, H.; Vandervorst, W. Phys. Status Solidi A 2009, 206, 2077−2081. (26) Schulze, A.; Hantschel, T.; Dathe, A.; Eyben, P.; Ke, X.; Vandervorst, W. Nanotechnology 2012, 23, 305707. (27) Celano, U.; Goux, L.; Opsomer, K.; Iapichino, M.; Belmonte, A.; Franquet, A.; Hoflijk, I.; Detavernier, C.; Jurczak, M.; Vandervorst, W. Microelectron. Eng. 2013, DOI: 10.1016/j.mee.2013.06.001. (28) Guo, X.; Schindler, C.; Menzel, S.; Waser, R. Appl. Phys. Lett. 2007, 91, 133513. (29) Valov, I.; Waser, R. Adv. Mater. 2013, 25, 162−4. (30) Sankaran, K.; Goux, L.; Clima, S.; Mees, M.; Kittl, J.; Jurczak, M.; Altimime, L.; Rignanese, G.; Pourtois, G. ECS Trans. 2012, 45, 317−330. (31) Kim, S.; Kim, S.-J.; Kim, K. M.; Lee, S. R.; Chang, M.; Cho, E.; Kim, Y.-B.; Kim, C. J.; Chung, U.-I.; Yoo, I.-K. Sci. Rep. 2013, 3, 1680. (32) Russo, U.; Kamalanathan, D.; Ielmini, D.; Lacaita, A. L.; Kozicki, M. N. IEEE Trans. Electron Devices 2009, 56, 1040−1047.
ASSOCIATED CONTENT
S Supporting Information *
Operation principles of C-AFM tomography, the details on the sample preparation, and the software used in the work. Other observations of CF in 3D including a movie of the 3D filament evolution are also shown. This material is available free of charge via the Internet at http://pubs.acs.org.
■
AUTHOR INFORMATION
Corresponding Author
*E-mail:
[email protected];
[email protected]. Tel.: 0032 16 28 8656. Author Contributions
U.C., L.G., A.B., and W.V. designed this work and U.C. prepared the manuscript. The experiments were carried out by U.C., L.G., A.B., A.S., A.F., O.R., and K.O., and H.B., C.D., and M.J., contributed in sample deposition, results analysis, and discussion of the manuscript during the preparation. All authors have given approval to the final version of the manuscript. Notes
The authors declare no competing financial interest.
■
ACKNOWLEDGMENTS Research funded by a Ph.D. grant of the Agency for Innovation by Science and Technology (IWT), we acknowledge the partial funding by IMEC’s Industrial Affiliation program on RRAM.
■
REFERENCES
(1) Chandrasekaran, N. IEDM Tech. Dig. 2013. (2) Strukov, D. B.; Snider, G. S.; Stewart, D. R.; Williams, R. S. Nature 2008, 453, 80−83. (3) Waser, R.; Dittmann, R.; Staikov, G.; Szot, K. Adv. Mater. 2009, 21, 2632−2663. (4) Valov, I.; Waser, R.; Jameson, J. R.; Kozicki, M. N. Nanotechnology 2011, 22, 254003. (5) Jeong, D. S.; Thomas, R.; Katiyar, R. S.; Scott, J. F.; Kohlstedt, H.; Petraru, a; Hwang, C. S. Rep. Prog. Phys. 2012, 75, 076502. (6) Yang, J. J.; Pickett, M. D.; Li, X.; Ohlberg, D. A. A.; Stewart, D. R.; Williams, R. S. Nat. Nanotechnol. 2008, 3, 429−433. (7) Waser, R.; Aono, M. Nat. Mater. 2007, 6, 833−40. (8) Jo, S. H.; Lu, W. Nano Lett. 2008, 8, 392−7. (9) Tsuruoka, T.; Terabe, K.; Hasegawa, T.; Aono, M. Nanotechnology 2010, 21, 425205. (10) Tappertzhofen, S.; Mündelein, H.; Valov, I.; Waser, R. Nanoscale 2012, 4, 3040−3. (11) Jo, S. H.; Kim, K.-H.; Lu, W. Nano Lett. 2009, 9, 870−4. (12) Zhu, X.; Su, W.; Liu, Y.; Hu, B.; Pan, L.; Lu, W.; Zhang, J.; Li, R.-W. Adv. Mater. 2012, 24, 3941−6. F
dx.doi.org/10.1021/nl500049g | Nano Lett. XXXX, XXX, XXX−XXX
