reversible migration of silver on memorized pathways in ag ge40s60 films

Reversible migration of silver on memorized pathways in Ag-Ge40S60 films , , J Orava , M N Kozicki, S N Yannopoulos, and A L Greer Citation: AIP Advances 5, 077134 (2015); doi: 10.1063/1.4927006 View online: http://dx.doi.org/10.1063/1.4927006 View Table of Contents: http://aip.scitation.org/toc/adv/5/7 Published by the American Institute of Physics AIP ADVANCES 5, 077134 (2015) Reversible migration of silver on memorized pathways in Ag-Ge40S60 films J Orava,1,2,a M N Kozicki,3 S N Yannopoulos,4 and A L Greer1,2,a Department of Materials Science & Metallurgy, University of Cambridge, 27 Charles Babbage Road, Cambridge CB3 0FS, United Kingdom WPI-Advanced Institute for Materials Research (WPI-AIMR), Tohoku University, 2-1-1 Katahira, Aoba-ku, Sendai 980-8577, Japan School of Electrical, Computer, and Energy Engineering, Arizona State University, Tempe, Arizona 85287-6206, USA Foundation of Research and Technology Hellas - Institute of Chemical Engineering Sciences (FORTH/ICE-HT), Patras, P O Box 1414, Greece (Received 26 May 2015; accepted July 2015; published online 14 July 2015) Reversible and reproducible formation and dissolution of silver conductive filaments are studied in Ag-photodoped thin-film Ge40S60 subjected to electric fields A tip-planar geometry is employed, where a conductive-atomic-force microscopy tip is the tip electrode and a silver patch is the planar electrode We highlight an inherent “memory” effect in the amorphous chalcogenide solid-state electrolyte, in which particular silver-ion migration pathways are preserved “memorized” during writing and erasing cycles The “memorized” pathways reflect structural changes in the photodoped chalcogenide film Structural changes due to silver photodoping, and electrically-induced structural changes arising from silver migration, are elucidated using Raman spectroscopy Conductive filament formation, dissolution, and electron (reduction) efficiency in a lateral device geometry are related to operation of the nano-ionic Programmable Metallization Cell memory and to newly emerging chalcogenide-based lateral geometry MEMS technologies The methods in this work can also be used for qualitative multi-parameter sampling of metal/amorphous-chalcogenide combinations, characterizing the growth/dissolution rates, retention and endurance of fractal conductive filaments, with the aim of optimizing devices C 2015 Author(s) All article content, except where otherwise noted, is licensed under a Creative Commons Attribution 3.0 Unported License [http://dx.doi.org/10.1063/1.4927006] I INTRODUCTION The structure and physico-chemical properties of silver-doped thin-film amorphous chalcogenides have been extensively studied in connection with their application as optical elements,1 inorganic resists,2 ion-selective electrodes,3 thin-film ionic batteries,4 and photonic devices.5 In the present work, the main focus is on the use of amorphous chalcogenides in the solid-state memory variously known as nano-ionic,6 programmable metallization cell, PMC,7 or conductivebridging random-access memory, CB-RAM.8 In such devices, the memory cell typically has top silver and bottom inert (Pt, W, Au, Ni ) electrodes Writing is by formation of conducting filament(s) (CF) of metallic silver within the cell on application of a voltage (silver anode, inert cathode) leading to the ON state Erasing is by dissolution of the metal on reversal of the potential difference (OFF state) The ON state stability depends on the smallest lateral size of the CF Writing can already be achieved at low voltage but then the ON state can have poor stability if the CF has a very small diameter or is discontinuous The width of the CF has been estimated to be in the order of a Authors to whom correspondence should be addressed Electronic mail: jo316@cam.ac.uk (JO); alg13@cam.ac.uk (ALG) 2158-3226/2015/5(7)/077134/10 5, 077134-1 © Author(s) 2015 077134-2 Orava et al AIP Advances 5, 077134 (2015) tens of nanometers (assuming a cone-shaped CF, with growth limited by compliance current),9 and it is supposed that only one CF is formed in each memory cell in the ON state.10 The morphology of CFs is explored in more detail in this paper Amorphous Ag-photodoped Ge x S100−x and Ge x Se100−x (x = 25–40 at.%) are the usual choice for PMC.11 The electrodeposit growth rate in PMC is estimated to be m s−1.7 The resistance ratio ROFF/RON is typically ∼105.12 The PMC is a low-power device, requiring switching energy ∼10−15 J, three orders of magnitude lower than for example in phase-change memory (also chalcogenide-based), and the ionic switching takes ∼50 ns.13 An attraction is that memory based on structural changes should be non-volatile: ON-state data retention in PMC is estimated to be >10 yrs at room temperature;8 and recently retention up to 105 minutes at 200◦C has been demonstrated for Ge-S-based PMC.14 On the other hand, damage accumulation can limit the endurance; beyond 1011 cycles is claimed with 20% decrease in ON current after 1016 cycles.15 PMC memory is highly scalable, including 3-D porous alumina templating.16 Contact sizes down to 20 nm have been achieved,8 but technology at this scale is not yet applicable because of poor endurance and data-retention Given these characteristics, chalcogenide-based ionic memory is attractive for portable devices, and has found a niche application, because of its good resistance to X-rays and short-term high-temperature stability, in medical environments with strict sterilization requirements PMC also shows promise beyond von-Neumann computing.17 Still early in its development, and with clear prospects for scaling down, PMC performance is likely to improve significantly in next-generation devices In the present work, we examine some of the issues associated with the reversible growth of CFs in Ag-photodoped Ge40S60 films in a lateral configuration The dissolution of silver must involve volume changes and high local stresses How these affect the structure of the electrolyte, the nature of the reversibility (for example, does the CF follow a different path for growth in each successive ON state?) and the endurance of the device? In the writing cycle, there is a leakage current through the electrolyte that does not contribute to Ag+ migration and metal deposition When the CF is established, further current is superfluous In erasing, similarly, there are currents along the CF (at the start) and through the electrolyte (later) that not contribute to the dissolution of the Ag filament With the energy of device operation in mind, it is clearly of interest to study what may be termed the electron efficiency of writing and erasing Such issues of electrolyte structure and efficiency are difficult to examine in commercially relevant memory cells We use conductive atomic-force microscopy (C-AFM)18 to study micrometer-scale Ag dendrites in Ge40S60 films, and Raman scattering to understand structural changes in Ag-photodoped films and samples with filaments The lateral growth in these films, similarly to Refs 19–21, contrasts with the vertical growth in PMC, but the configuration in the present study does permit detailed and direct examination of the CF C-AFM studies on vertical samples show I–V characteristics relevant for PMC memory operation, but such studies not provide direct access to the full CF morphology, which is typically inferred from topographic swelling on the surface caused by CF growth.22,23 The role of stresses in interdiffusion in an amorphous phase has been analyzed for example by Greer et al.24 In the present work, the larger length-scales and the longer time-scales (s rather than ns), imply a reduced role of stresses, but the observations still have points of relevance for understanding the operation of commercial chalcogenide-based PMC memory The studied sample geometry is of direct relevance for a newly emerging Ag-photodoped chalcogenide lateral devices, of the similar micrometer length-scales, used as visible light and radiofrequency switches,25 high-Q tunable MEMS resonators, microvalves11 and omnidirectional microphone for tunable hearing aids suiting individual patients’ needs.26 A lateral device geometry may sound arbitrary chosen just for its simplicity, but for example radiofrequency switch has to have lateral configuration with only electrode edges facing one another, otherwise the (OFF state) isolation would be very poor at GHz frequencies due to high capacitance.25 All these devices rely on the same basic ionic-switching mechanism as in PMC with alike issues such as meagre repeatability and ON-like state stability The reversible CFs growth/dissolution is exploited in bringing new functionalities MEMS have been demonstrated using thin-film Ag-Ge-Se, compositions similar to 077134-3 Orava et al AIP Advances 5, 077134 (2015) those used in PMC, as the main solid-state electrolyte These devices are at early stages of development beyond the established memory technology The present work not only takes advantage of the experience with PMC, but also attempts to establish a basis for the evolving lateral technologies C-AFM may be helpful in mass screening of different samples to observe trends and compare electrolyte performance in a simple but effective lateral geometry II EXPERIMENTAL SECTION Amorphous Ge40S60 films (thickness ∼300 nm) were deposited by magnetron sputtering on Si3N4 (50 nm)/silicon wafers at argon pressure 2.5 Pa and power 10 W, the substrates, initially at room temperature, heated by < 10 K during deposition Silver was deposited on top of as-prepared Ge40S60 films by thermal evaporation, under a background pressure of ∼10−4 Pa, at a rate of ∼0.2 nm s−1 The multilayer stack was illuminated by a halogen lamp to optically induce silver diffusion into the amorphous film.4 The residual silver on top of the film was removed by dissolving in Fe(NO3)3.9H2O (8 g/100 ml H2O) solution The AFM (Veeco 3100), equipped with an electronic tunneling module (TUNA) to extend current sensitivity, was used to induce electrodeposit formation and dissolution under ambient conditions The cantilevers, of standard n-type Si (spring constant ∼0.2 N m−1, maximum tip radius 25 nm), coated with a 20 nm film of Pt/Ir, had to be replaced regularly due to significant wear The topographic images were recorded immediately after the induced dendrite growth/dissolution using the same tip in contact mode The AFM data were processed using Gwyddion software.27 While the C-AFM tip itself served as one electrode, the other electrode was a silver conductive pad painted on the shorter side edge of the 1.0×0.5 cm cut sample (Fig 1) with a typical tip to planar electrode distance of 200–500 µm CF growth was achieved by scanning down a 50×50 nm area of the sample at a scan rate of Hz while applying a constant DC voltage from +2 V up to +10 V, typically for 130 s No detectable change in topography was observed for voltage biases lower than +2 V The entire dendritic shape of the deposited CFs was evaluated using a fractal dimension (Df ) analysis.28,29 The built-in FracLac plugin for the ImageJ 30 was used for Df calculations via local-connected-fractal-dimension (LCFD) analysis This method was tested on a standard Koch snowflake giving a reproducible value of Df = 1.25; the LCFD does not allow evaluation of lacunarity Outlined images were used for fractal analysis, therefore ignoring the internal morphology of the dendrites Compositional analysis used a FEG SEM CamScan MX2600 (20 kV) equipped with silicondrift detector (Oxford Instruments), and X-ray data were collected using a Bruker D8 diffractometer equipped with position-sensitive detector Micro-Raman spectra of the films on Si substrates were FIG Schematic of a C-AFM configuration for tip-sample DC biased current flow measurements during writing (not to scale) 077134-4 Orava et al AIP Advances 5, 077134 (2015) collected using a Renishaw Ramascope-1000 spectrophotometer (cut-off at ∼200 cm−1) equipped with a He-Ne laser (λ = 633 nm); a photon energy of 1.96 eV (bandgap) and power output ∼7 mW were used (Ge40S60 bulk has an optical bandgap energy ∼2 eV) No photo-induced changes in the films were observed during the micro-Raman measurements The off-resonant Raman spectrum of the Ge40S60 bulk glass was recorded with a Fourier-transform Raman set-up, (excitation at λ = 1064 nm; the photon energy of 1.165 eV) The scattered intensity below 120 cm−1 is attenuated by the instrument notch filter III RESULTS AND DISCUSSION The amorphous structure of the as-prepared and the silver photo-doped thin films was confirmed by X-ray diffraction The composition of the as-prepared thin films was Ge40S60 (±0.5 at.%) The doped chalcogenide film was found to contain ∼16 at.% Ag (Ge/S atomic ratio 2:3) A C-AFM - Ag CF growth and dissolution The electric-field induced migration of silver in Ag-G40S60 films leads to formation of dendrites on the surface with complex branched morphology that depends on the applied voltage (Figs 2(a) FIG AFM topography images showing dendritic growth on an amorphous Ag-Ge40S60 film after application of a DC voltage: (a) +2 V, no appreciable growth was observed below this voltage; and (b) +10 V, upper limit of the set-up (c) The surface between two main branches in (b) corresponding to the topography of the as-prepared sample after removal of residual Ag The original tip position was always at the center-point of the dendritic growth 077134-5 Orava et al AIP Advances 5, 077134 (2015) and 2(b)) There are small “electrodeposits” up to ∼20 nm in height (Fig 2(c)), between the main branches, corresponding to the topography of the as-prepared sample after removal of residual Ag There is no detectable change in topography outside the effective area of dendritic growth It is therefore assumed that silver ions originate from the surface and sub-surface layers adjacent to the propagating dendrites Hsu et al.31 have also shown, in a similar lateral device configuration, that silver CFs in Ag2S solid electrolyte are predominantly formed close to the surface and also propagate along it The Ag+ ions in the electrolyte, migrating in response to the electric field, deposit on the growing dendrite with a sticking coefficient ∼1 The dendrites thus grow outwards from the C-AFM tip (cathode) through the reduction reaction Ag+ + e− → Ag at the newly formed CF tip/electrolyte interface When the first silver deposits are formed, at tip-film contact, the electric field is higher at that point and the silver migration and electrodeposition are accelerated, with the first Ag particles serving as nucleation centers for further growth There has been no detectable role of irreversible processes, such as germanium oxidation,32 as any grown dendrite could be fully dissolved by exposure to light The fractal dimension Df increased linearly with increasing voltage from 1.20 to 1.30 ± 0.02, at +2 V and +10 V, respectively The Df neglects the internal morphology of the dendrites, which was found to be independent of the applied voltage (Figs 2(a) and 2(b)) It is clear that the morphology of the dendrites depends on the applied voltage: at higher voltage, the dendrite arms grow initially thinner and then are more branched There is a loose analogy here with dendritic growth in other contexts, such as the solidification of metallic alloys; at greater melt supercooling (corresponding to greater voltage in the present case) dendrites grow faster, have sharper tips and thinner arms, and are more branched.33 In solidification dendritic growth can occur in steady state, but that is not possible in the present case because of the complex roles of chemical diffusion, and mixed ionic and electronic conductivities The interpretation of Df in a lateral device and its relationship with PMC memory are not straightforward The growth kinetics and final shape of the dendrites must depend on the composition of the solid-state electrolyte and not only on silver content For exploring a wide range of systems, however, a fast and simple technique such as C-AFM, could be particularly attractive Mitkova et al.21 observed the morphology of Ag dendrites in Ge x Se100−x films (x = 20, 30, 33 and 40 at %.) by sweeping the applied voltage from to +5 V As the Ge concentration in the matrix increased, the dendrites became more fragmented This was attributed to local Ag-rich particles/clusters serving as nucleation centers Fractal analysis of the dendritic growth in lateral samples could provide insights on dendrite sharpening and branching, dependent on the sample and its initial structure, and on the electrical conditions driving dendrite growth The degree of branching, related to dendrite-arm thickness, is likely to be important for memory applications: thinner arms grow faster, but are less stable.34 In co-planar type lateral devices, further apart the electrodes are, the longer the time is, and the more branched dendrites may grow for a given voltage For example, the tunable reflectance of the visible-light frequency switch,11 via the surface metallic electrodeposit, can be in first approximation, taken as an effective-medium approximation (percentage of the surface coverage), and therefore branching and dendrites width will have pronounced influence on the complex dielectric function The electron efficiency ηe of the electrodeposition process relates the actual amount of Ag metal in the CFs to the theoretical value according to Faraday’s law assuming that all the charge transport corresponds to reduction of Ag+ Calculation of η e was possible in the regime of constant current flow during stable dendrite growth (writing), where measured currents were in the range 10−8 to 10−6 A for +2 V to +10 V, respectively Two limiting cases of η e were calculated The lower limit was calculated by taking electrodeposit volume above the surface only, neglecting any complex internal morphology and sub-surface deposit, which cannot be imaged by AFM The η e was ∼9% for +2 V and dropped below 1% at higher voltages The upper limit could be estimated by assuming dendritic growth through the entire thickness of the film (the projected area of the dendrite taken throughout the entire sample thickness ∼300 nm), then η e was ∼25% for +2 V and dropped below 5% at higher voltages The low η e of CF bridge formation has also been noted for PMC memory.35 Some losses originate from the entire circuit, but most of the current just flows through the low-resistance electrolyte The η e will depend on device size, and must be higher for 077134-6 Orava et al AIP Advances 5, 077134 (2015) FIG AFM topography images of reversible: (a) writing at +7 V; (b) erasing at −7 V; and (c) rewriting at +7 V The arrows show the directions of growth (a,c) and dissolution (b) Once the conductive filament is established in the first cycle, then growth along the same (memorized) pathways is facilitated in subsequent cycles devices in which the electrodeposited CF occupies a larger fraction of the volume and cross-section of the memory cell After the CF touches the top electrode, the rest of the current is superfluous, short-circuiting through the CF; the electrodeposition then slows down and further deposition is at very low values of η e Reproducibility and reversibility of growth (writing) and dissolution (erasing) have been tested over 10–20 cycles by switching the applied voltage between +7 V and −7 V (Fig 3) In the dissolution phase, the dendrites dissolve from the branch tips towards the center, in other words in what is a close approximation of a time-reversal of the growth process Dissolution leaves behind some fragments of undissolved Ag In detail, two types of dissolution have been observed: (i) usually, there is partial dissolution of the dendrite (Fig 3(b)), similar to that seen in oxide-based devices;36 and (ii) more rarely, there is a complete dissolution of the dendrite when the voltage bias is reversed An analogous distinction was found in oxide electrolytes,37 where CF dissolution is mostly partial It should be noted, though, that the origin and mechanism of reset are different in this case (being thermally assisted and therefore dependent on power density, unlike in PMC) A dual dissolution characteristics, resulting in a different morphology of broken- and continuous-shaped CF, has been observed in the oxide-based CB-RAM.38 The particular CF shape could be assigned to different I–V erasing curve Again, highlighting the importance of CF morphology on the device performance In the present set-up, the growth ∼0.3–0.6 µm s−1 at +7 V was 6–7 times slower than the dissolution, and the current ratio Igrowth/Idissolution was ∼102 In PMC devices, the ratio of the writing/erasing currents is 1–10 at maximum because the limiting compliance current is used to prevent later growth of CF during the writing cycle.7 The growth and dissolution of dendrites occurs along the same pathways (“memory effect”) during cycling, with the last metallic electrodeposit to grow being the first to dissolve It is tempting to assume that a similar mechanism operates in real PMC memory The reproducible re-growth cannot be simply because of the sample geometry, which lacks any 2-D directionality, or any substrate templating The dendritic deposits slightly resemble the ‘four-fold’ symmetry (Fig 2) that would be associated with crystallography, but in fact the entire shape is rather random, without defined orientations That the same shape is re-formed (Fig 3(c)), in the absence of the geometrical effects and any substrate templating, provides evidence for the memorized pathways in the films Modeling predicts that a CF dissolves first at the contact with the Ag top electrode,39 and the memory immediately switches to the OFF state When the bias is reversed, the Ag electrode being positive, the CF can quickly be restored in a repeatable and reproducible way due to “memorized” pathways in the electrolyte The successful formation of CFs in PMC memory could also be indicated by the threshold initialization voltage being higher than subsequent writing voltages, e.g +0.20 V and +0.15 V, respectively for a bottom electrode in an amorphous Ag-Ge-Se thin film.40 The difference between the threshold initialization and subsequent writing voltage is even more pronounced for oxide-based memory (Cu-WO3).41 077134-7 Orava et al AIP Advances 5, 077134 (2015) Apart from PMC devices, the “memorized” pathways could potentially be beneficial for quantitative liquid pumping in microvalve technology,11 if the same volume of the liquid can always be restored inside a microchannel Unlike in the PMC technology, ionic transport enhanced on cycling via structurally memorized pathways can have pronounced negative effects in some contexts This is likely in the case of lithium batteries, where swelling and shrinking of graphite electrode anode associated with (de)lithiation leads to battery degradation and failure through dendrite formation.42 B Micro-Raman spectroscopy – elucidating structural changes in the thin films Micro-Raman spectroscopy was employed to reveal the origin of the structural differences between the as-deposited films (Fig 4(a)), the Ag photo-doped films (Fig 4(b)) and films in the area of a dendrite (Fig 4(c)) The Raman spectra include sharp peaks of the Si substrate located at 300 cm−1 and 520 cm−1; the former was subtracted from the presented normalized Raman spectra, after proper scaling, to achieve more reliable fitting.43 Despite the technological interest about Ge-rich chalcogenides, the absence of consensus on peak assignment makes the understanding of structure quite uncertain.44–46 Ag photodoping induces structural modifications evidenced from changes in the Raman spectra The breadth of the peaks makes the quantitative analysis of the Ag-Ge-S spectra dubious The off-resonant Raman spectrum of the Ge40S60 bulk glass is added in Figure 4(a) for comparison This spectrum bears a close resemblance to that recorded from the sputtered film, confirming FIG Normalized deconvoluted Raman spectra from: (a) the as-deposited amorphous thin film, and bulk glass for comparison; (b) the Ag photo-doped thin film; and (c) the region around a dendrite The open symbols show experimental data after background and Si substrate Raman bands subtraction The dashed-blue lines in (a) show the Gaussian peaks and the solid-red line through the experimental data is the best fit result See text for further discussion on the importance of the high-frequency modes 077134-8 Orava et al AIP Advances 5, 077134 (2015) that the sputtered film not only has a composition that is close to nominal but also its structure is very similar to the bulk glass structure as no sharp bands, representative of molecular species, appear in the film spectrum The bands in the high-frequency envelope (300–450 cm−1) represent predominantly Ge-S stretching vibrations Considering that the chemically ordered network model applies here (namely, assuming the satisfaction of the 8–N rule) the structure of the nonstoichiometric Ge40S60 glass must be composed of a mixture of interconnected GeSn Ge4−n (n = 0, 1, 2, 3, 4) tetrahedra The relative proportions of these tetrahedra at the S:Ge ratio = 1.5 are: 31.64% GeS4; 42.18% GeS3Ge; 21.09% GeS2Ge2; 4.69% GeSGe3; and 0.39% GeGe4.47 Only one third of the tetrahedra are GeS4 units with full tetrahedral, Td, symmetry The reduction of the Td symmetry to C3v and C2v in GeS3Ge1 and GeS2Ge2 units lifts the degeneracy and causes the appearance of bands in addition to those expected for the GeS4/2 tetrahedra, complicating the interpretation The main stretching Ge-S vibrations of the corner sharing (CS) GeS4/2 tetrahedra in the stoichiometric glass GeS2 are at ∼342 cm−1 (symmetric stretching, v1(A2))47 and at ∼394 cm−1 (antisymmetric stretching, v3(F2)).45 Edge-sharing (ES) tetrahedra are also structural motifs in Ge-S glasses with their main band at 372 cm−1, known as the companion line, Ac1 The origin of the band at ∼435 cm−1 is discussed below The best-fit deconvolution of the spectrum of the as-deposited Ge40S60 film was performed with six Raman peaks The four bands in the high-frequency envelope (300–450 cm−1) resemble the four bands noted above in connection with stoichiometric amorphous GeS2, and are highlighted by arrows The band at ∼347 cm−1 is assigned to the symmetric stretching Ge-S modes of GeS4/2 and GeS3Ge tetrahedral units, which compose the majority of tetrahedral and tetrahedral-like species for this composition The band at ∼405 cm−1 is near the 395 cm−1 band of the stoichiometric GeS2 compound, which, using scaling analysis, has been attributed to the asymmetric stretching mode of GeS4/2 and GeS3Ge tetrahedral units.45 The band at 379 cm−1 originates from ES (companion line Ac1) GeS4/2 tetrahedra.45,47 Finally, the fit has revealed a band at ∼435 cm−1 whose interpretation is still controversial Several studies (Ref 44 and references therein) consider this to arise from S-S vibrations (short chains or cluster terminations) This assignment can be excluded based on the Raman spectrum of crystalline and liquid S where the most intense bands of both molecular (S8 rings) and polymeric (Sn chains) sulfur species, not expected in Ge-rich glasses, lie at 474 and 463 cm−1 respectively.48 In an alternative explanation, the 435 cm−1 mode has been associated with the highest mode in the ES clusters using ab initio calculations.49 The bending modes of the tetrahedral units and the Ge-Ge homonuclear bonds form a rather structureless continuum of vibrational modes in the range 200–300 cm−1 The band associated with Ge-Ge bonds, at ∼230 cm−1, appears as soon as the Ge content exceeds even slightly the 33.33% stoichiometric level in GeS2, and becomes dominant for Ge40S60.45 Lucovsky et al.47 suggested that around 230 cm−1, symmetric Raman modes of GeS3Ge and GeS2Ge2 mixed tetrahedral units are expected The introduction of silver makes the interpretation of Raman spectra even more difficult Deconvoluting the Raman spectrum is, then, highly uncertain as a multiplicity of Gaussian lines can result in equally good fits, and the spectral changes can be discussed only qualitatively Figure 4(b) shows an appreciable change of the relative band intensities between the high (300–450 cm−1) and low (300 cm−1) frequency ranges The intensity of the high-frequency band (Ge-S stretching modes) decreases after Ag addition in relation to the low frequency intensity Ag seems to cause network degradation through the rupture of Ge-S-Ge bridges The formation of Ag-S bonds, which bear more ionic character than the covalent Ge-S bonds, exhibit low Raman cross-sections Indeed, Ag-S modes appear in Raman spectra at frequencies below 300 cm−1 as broad weak bands.50 Therefore, the shape of the Raman spectrum of the Ag-doped Ge-S glass (Fig 4(b)) provides evidence for significant network degradation caused by silver atoms This ‘opening’ of the structure may provide channels for fast migration of Ag+ ions The spectrum recorded from the electrodeposited dendrite region (Fig 4(c)), where Ag concentration is much higher than the surrounding area, can be accounted for along the same lines The increased Ag concentration is responsible for Ge-S network damage evidenced by the weakening of the high-frequency spectral area, corresponding to Ge-S stretching modes As a result, the intensities of low-frequency bands increase over the high-frequency Raman bands 077134-9 Orava et al AIP Advances 5, 077134 (2015) IV CONCLUSIONS While the length-scales, time-scales, geometry and reduced role of stresses in conductivefilament growth in lateral thin-film samples are different from those in PMC memory, the C-AFM observations in the present work are useful in permitting the imaging of the formation/dissolution of conductive paths in metal-doped thin-film amorphous chalcogenides The growth of metallic silver dendrites in Ge40S60 films is found to show an effect, where Ag migration pathways are “memorized” after erasing, allowing easier re-growth in subsequent cycles The open structure of the Ge-rich sulfides further promotes fast silver diffusion under applied bias Whereas the PMC has been so far the most promising commercial technology, the “memorized” effect and lateral sample geometry is directly important for the new emerging thin-film MEMS technologies based on Ag-photodoped chalcogenides, which are specifically designed to operate in a lateral geometry Description of dendritic structures via fractal analysis is promising for assessment and comparison of chalcogenide solid-state electrolytes for potential use in both vertical, PMC, and lateral devices A lateral geometry such as that in the present work, is useful in developing and understanding of how the shape of conducting filaments depends on the solid-state electrolyte and on the electrical conditions for their growth The lateral geometry, unlike the fabrication of PMC devices, would allow mass (i.e multi-parameter) screening to guide the selection of chalcogenide electrolyte and metal electrode for optimized memory-device performance ACKNOWLEDGMENTS This work was supported by World Premier International Research Center Initiative (WPI), MEXT, Japan J.O and A.L.G acknowledge support from the Engineering and Physical Sciences Research Council (EPSRC, UK) C Huang and D Hewak, Thin Solid Films 500, 247 (2006) J Orava, T Wagner, M Krbal, T Kohoutek, Mil Vlcek, P Klapetek, and M Frumar, J Non-Cryst Solids 354, 533 (2008) J M Conde Garrido, F Macoretta, M A Urena, and B Arcondo, J Non-Cryst Solids 355, 2079 (2009) M Frumar and Wagner, Curr Opin Solid State Mat Sci 7, 117 (2003) B J Eggleton, B Luther-Davies, and K Richardson, Nat Photonics 5, 141 (2011) R Waser and M Aono, Nat Materials 6, 833 (2007) M N Kozicki, M Balakrishnan, C Gopalan, C Ratnakumar, and M Mitkova, in Programmable Metallization Cell Memory Based on Ag-Ge-S and Cu-Ge-S Solid Electrolytes: Non-Volatile Memory Technology Symposium, Dallas, USA, 10–10 November 2005 (The Institute of Electrical and Electronics Engineers, Inc., Piscataway, 2005) pp 83–89 R Symanczyk, R Dittrich, J Keller, M Kund, G Muller, B Ruf, P.-H Albarede, S Bournat, L Bouteille, and A Duch, in Conductive Bridging Memory Development from Single Cells to 2Mbit Memory Arrays: Non-Volatile Memory Technology Symposium, Albuquerque, USA, 10–13 November 2007 (The Institute of Electrical and Electronics Engineers, Inc., Piscataway, 2007) pp 71–75 R Waser and I Valov, ECS Transactions 5, 431 (2009) 10 C Schindler, M Meier, and R Waser, in Resistive Switching in Ag-Ge-Se with Extremely Low Write Currents: Non-Volatile Memory Technology Symposium, Albuquerque, USA, 10–13 November 2007 (The Institute of Electrical and Electronics Engineers, Inc., Piscataway, 2007) pp 82–85 11 M Kozicki and M Mitkova, J Non.-Cryst Solids 352, 567 (2006) 12 S Dietrich, M Angerbauer, M Ivanov, G Gogl, H Hoenigschmid, M Kund, C Liaw, M Market, R Symanczyk, L Altimime, S Bournat, and G Muller, IEEE J Solid-State Circuit 42, 839 (2007) 13 C An, in Ionic Memories: Status and Challenges: 9th Annual Non-Volatile Memory Technology Symposium, Pacific Groove, USA, 11–14 November 2008 (The Institute of Electrical and Electronics Engineers, Inc., Piscataway, 2008) pp 1–5 14 J R Jameson, P Blanchard, C Cheng, J Dinh, A Gallo, V Gopalakrishnan, C Gopalan, B Guichet, S Hsu, D Kamalanathan, D Kim, F Koushan, M Kwan, K Law, D Lewis, Y Ma, V McCaffrey, S Park, S Puthenthermadam, E Runnion, J Sanchez, J Shields, K Tsai, A Tysdal, D Wang, R Williams, M N Kozicki, J Wang, V Gopinath, S Hollmer, and M Van Buskirk, in Conductive-Bridge Memory (CBRAM) with Excellent High-Temperature Retention: IEEE International Electron Devices Meeting, Washington, USA, 9–11 December 2013 (The Institute of Electrical and Electronics Engineers, Inc., Piscataway, 2013) pp 30.1.1–30.1.4 15 M N Kozicki, M Park, and M Mitkova, IEEE Trans Nanotechnol 4, 331 (2005) 16 J Kolar, J M Macak, K Teraba, and T Wagner, J Mater Chem C 2, 349 (2014) 17 D Mahalanabis, H J Barnaby, Y Gonzalez-Velo, M N Kozicki, S Vrudhula, and P Dandamudi, Solid-State Electron 100, 39 (2014) 18 M H Lee and C S Hwang, Nanoscale 3, 490 (2011) 19 M N Kozicki, M Yun, S.-J Yang, J P Aberouette, and J P Bird, Superlattices Microstruct 27, 485 (2000) 077134-10 20 Orava et al AIP Advances 5, 077134 (2015) M N Kozicki, M Mitkova, and J P Aberouette, Physica E 19, 161 (2003) M I Mitkova, M N Kozicki, and J P Aberouette, J Non-Cryst Solids 326–327, 425 (2003) 22 A Pradel, N Frolet, M Ramonda, A Piarristeguy, and M Ribes, Phys Status Solidi A 208, 2303 (2011) 23 C Schindler, K Szot, S Karthauser, and R Waser, Phys Status Solidi–Rapid Res Lett 2, 129 (2008) 24 A L Greer, N Karpe, and J Bøttiger, J Alloys Comp 194, 199 (1993) 25 J A Nessel, R Q Lee, C H Mueller, M N Kozicki, M Ren, and J Morse, in A Novel Nanoionics-Based Switch for Microwave Applications: IEEE MTT-S International Microwave Symposium Digest, Atlanta, USA, 15–20 June 2008 (The Institute of Electrical and Electronics Engineers, Inc., Piscataway, 2008) pp 1051–1054 26 S.-S Je, J Kim, J C Harrison, M N Kozicki, and J Chae, J Appl Phys Lett 93, 123501 (2008) 27 http://gwyddion.net/ 28 T G Smith, Jr., G D Lange, and W B Marks, J Neurosci Methods 69, 123 (1996) 29 K Kudo, IEEE Trns Dielectr Electr Insul 5, 713 (1998) 30 http://imagej.nih.gov/ij/ 31 K Hsu, P Ferreira, and N Fang, Appl Phys Lett 96, 024101 (2010) 32 A Kovalskiy, H Jain, and M Mitkova, J Non.-Cryst Solids 355, 1924 (2009) 33 D M Herlach, R F Cochrane, I Egry, H J Fecht, and A L Greer, Int Mater Rev 38, 273 (1993) 34 C.-P Hsiung, H.-W Liao, J.-Y Gan, T.-B Wu, J.-C Hwang, F Chen, and M.-J Tsai, ACS Nano 4, 5414 (2010) 35 R Waser, V Rana, S Menzel, and E Linn, in Energy-Efficient Redox-Based Non-Volatile Memory Devices and Logic Circuits: Third Berkeley Symposium on Energy Efficient Electronic Systems, Berkeley, USA, 28–29 October 2013 (The Institute of Electrical and Electronics Engineers, Inc., Piscataway, 2013) pp 1–2 36 Q Liu, J Sun, H Lv, S Long, K Yin, N Wan, Y Li, L Sun, and M Liu, Adv Mat 24, 1844 (2012) 37 C S Hwang, in Collective Motion of Nano-Filaments in Pt/n-Type T iO /p − T y pe NiO/Pt Stacked Resistance Switching Memory: 1st International Workshop on Resistive RAM, Leuven, Belgium, 20–21 October 2011, p 10 38 U Celano, L Goux, A Belmonte, K Opsomer, R Degrave, C Detavernier, M Jurczak, and W Vandervorst, J Phys Chem Lett 6, 1919 (2015) 39 X Guo, C Schindler, S Menzel, and R Waser, Appl Phys Lett 91, 133513 (2007) 40 M N Kozicki, C Gopalan, M Balakrishnan, M Park, and M Mitkova, in Non-Volatile Memory Based on Solid Electrolytes: Non-Volatile Memory Technology Symposium, Stanford, USA, 15–17 November 2004 (The Institute of Electrical and Electronics Engineers, Inc., Piscataway, 2004) pp 10–17 41 M N Kozicki, C Gopalan, M Balakrishnan, and M Mitkova, IEEE Trans Nanotechnol 5, 535 (2006) 42 S J Harris, A Timmons, R D Baker, and C Monroe, Chem Phys Lett 485, 265 (2010) 43 P A Temple and C E Hathaway, Phys Rev B 7, 3685 (1973) 44 I P Kotsalas and C Raptis, Phys Rev B 64, 1252101 (2001) 45 T Petkova, B Monchev, O Kostadinova, P Petkov, and S N Yannopoulos, J Non-Cryst Solids 355, 2063 (2009) 46 A Chrissanthopoulos, P Jovari, I Kaban, S Gruner, T Kavetskyy, J Borc, W Wang, J Ren, G Chen, and S N Yannopoulos, J Solid State Chem 192, (2012) 47 G Lucovsky, F L Galeener, R C Keezer, R H Geils, and H A Six, Phys Rev B 10, 5134 (1974) 48 A G Kalampounias, K S Andrikopoulos, and S N Yannopoulos, J Chem Phys 118, 8460 (2003) 49 K Jackson, A Briley, S Grossman, and D V Porezag, Phys Rev B 60, R14985 (1999) 50 F Kyriazis, A Chrissanthopoulos, V Dracopoulos, M Krbal, T Wagner, M Frumar, and S N Yannopoulos, J Non-Cryst Solids 355, 2010 (2009) 21