FM - 3rd International Conference
Emerging Materials, Technologies and Applications for Non-volatile Memory Devices

Session FM-1 - Magnetic, Ferroelectric and Multiferroic Materials for Memory Devices

FM-1:IL01  NVM Technologies Based on Ferroelectric Hafnium Oxide
T. MIKOLAJICK^1,2, T. SCHENK¹, T. MITTMANN¹, M. HOFFMANN¹, B.MAX², C. RICHTER¹, M. PESIC¹, F. FENGLER¹, H. MULAOSMANOVIC¹, M.-H. PARK¹, S. SLESAZECK¹, U. SCHROEDER¹, J. MÜLLER³, P. POLAKOWSKI³, S. MÜLLER⁴, R. MATERLIK⁵, A. KERSCH⁵, ¹NaMLab gGmbH, Dresden, Germany; ²Chair of Nanoelectronic Materials, TU Dresden, Dresden, Germany; ³Fraunhofer IPMS-CNT, Dresden; ⁴FMC GmbH, Dresden, Germany; ⁵Munich University of Applied Sciences, Munich, Germany

Ferroelectric memories are in the market since almost a quarter of a century [1]. However, they are currently restricted to niche markets due to the limited scalability of such devices using complex perovskite or layered perovskite based material systems. Fueled by the discovery of ferroelectricity in hafnium oxide based materials [2], the interest in ferroelectric memories for broader markets has significantly increased again in the last three to four years. In this talk the current knowledge on ferroelectricity in hafnium oxide based materials and the possibilities arising for nonvolatile memories will be described. Recent findings on the physical basis and the controlling influences from film composition, film fabrication and post treatments as well as the actual status of the integration into memory cells and arrays will be presented.
References [1] D. Bondurant, Ferroelectronic RAM Memory Family for Critical Data Storage, Ferroelectrics 112 (1990), 273-282. [2] T.S. Boescke, J. Mueller, D. Braeuhaus, U. Schroeder, and U. Boettger, Ferroelectricity in hafnium oxide thin films, Appl. Phys. Lett. 99 (2011), 102903.



FM-1:L02  Magnetoelectric Coupling at Ferromagnet/Ferroelectric-HfO2 Interface
A. ZENKEVICH¹, Y. MATVEYEV¹, V. MIKHEEV¹, R. MANTOVAN², A.I. CHUMAKOV³, ¹Moscow Institute of Physics and Technology, Dolgoprudny, Moscow Region, Russia; ²CNR-IMM Laboratorio MDM, Agrate Brianza (MB), Italy; ³ESRF-The European Synchrotron CS40220, Grenoble Cedex, France

Composite multiferroic materials exhibiting coupling between ferromagnetism (FM) and ferroelectricity (FE) are of great scientific interest and can be potentially used in multifunctional electronic device concepts. Previously, magnetoelectric effects have been observed at the interface of the model FM and FE materials, such as Fe/BaTiO3. However, classical perovskite ferroelectrics have poor compatibility with the Si-based CMOS technology. Meanwhile, the ferroelectricity was discovered in thin (3-15 nm) polycrystalline films of doped HfO2 stabilized in the non-centrosymmetric orthorhombic phase, thus opening an opportunity to develop fully CMOS-compatible multiferroic devices. In this work, we have performed in operando study of functional capacitors comprising FE‑Hf0.5Zr0.5O2 layer in contact with Fe57/Pt electrode by synchrotron-based Mössbauer spectroscopy supplemented by XMCD, HAXPES, HR TEM and electrical characterization. Following in situ switching of polarization in FE‑Hf0.5Zr0.5O2, we observed statistically reliable changes in Mössbauer spectra, which we take as the first manifestation of magnetoelectric coupling at the FM/ FE‑HfO2 interface. These results are explained in terms of charge redistribution at the interface with few ML thick Fe57 capped with Pt.


FM-1:L03  Effect of Polarization Reversal on the Potential Distribution Across Ferroelectric HfO2 based Capacitors Revealed in Operando by Hard x-ray Photoemission Spectroscopy
Y. MATVEYEV, D. NEGROV, V. MIKHEEV, A. CHERNIKOVA, A. ZENKEVICH, Moscow Institute of Physics and Technology, Dolgoprudny, Moscow Region, Russia; A. Gloskovskii, Deutsches Elektronen-Synchrotron, Hamburg, Germany

Among various novel non-volatile memory concepts, the one based on the remnant polarization in ferroelectric (FE) films (FeRAM) is advanatgeous due to very low power consumption and theoretically unlimited endurance. In addition, such architectures as FE MOSFET and ferroelectric tunnel junction (FTJ) can offer a non-destructive readout. The recent discovery of FE properties in doped and alloyed HfO2 films has opened an opportunity to develop a fully CMOS-compatible FERAM. However, the realization of both FE MOSFET and FTJ based on HfO2 films require careful optimization of electronic properties in functional stack. Therefore, the determination of true potential distribution across the FE stack including electronic band line-up at interfaces is critically important. In this work, we present the results of the standing-wave hard X-ray photoemission spectroscopy in operando study of functional FE Hf0.5Zr0.5O2 based capacitors modelling memory devices. In order to switch polarization and monitor the actual polarization values in the device, we used parallel electrical characterization. By analyzing the set of spectroscopic data obtained at different exciting x-ray incident angles, we reconstruct full potential profile in the device and its changes following the polarization reversal.


FM-1:IL05  Ultrafast MRAM Strategies for Cache Applications and Beyond
I.L. PREJBEANU¹, A. TIMOPHEEV¹, M. MIRON¹, G. GAUDIN¹, B. LACOSTE¹, T. DEVOLDER², M. MARINS DE CASTRO¹, R.C. SOUSA¹, L.D. BUDA-PREJBEANU¹, S. AUFFRET¹, U. EBELS¹, B. RODMACQ¹, B. DIENY¹, ¹Univ. Grenoble Alpes, CEA, CNRS, INAC-Spintec, Grenoble, France; ²Univ. Paris-Sud, Orsay, France

The ongoing Big Data revolution faces major issues associated to energy cost and time delay of transferring data between the processor cores and the multiple levels of memory. To speed up operations, cache memories, mostly composed of SRAM, very fast but having a large footprint and volatile, are inserted between the processor and DRAM. STT-MRAM are considered as one of the most promising technology for non-volatile RAM due to their non-volatility, quasi-infinite endurance, speed and high-density. In standard STT-MRAM the switching dynamics of the storage layer magnetization is characterized by an incubation delay that can take up to a few nanoseconds. We present in this study new MRAM writing strategies allowing to greatly improve the write speed down to the sub-nanosecond range. First, we demonstrate that sub-ns switching with final state determined by the current polarity through the stack can be achieved in STT-MRAM cells comprising two spin-polarizing layers having orthogonal magnetic anisotropies. Second, we have shown that the write stochasticity can be almost completely suppressed and the writing speed greatly increased by inducing an easy-cone anisotropy in the storage layer. Finally, we will discuss recent writing strategies based on SOT-MRAM or all-optical switching.


FM-1:IL06  Sub-ns Current-induced Magnetization Switching Driven by Spin-orbit Torques
P. GAMBARDELLA, Department of Materials, ETH Zürich, Switzerland

Current-induced spin-orbit torques (SOTs) represent one of the most effective ways to manipulate the magnetization in spintronic devices [1-3]. The orthogonal torque-magnetization geometry and the large domain wall velocities inherent to materials with strong spin-orbit coupling make SOTs especially appealing for fast switching applications in nonvolatile memory and logic units [4]. Here, we report the direct observation of SOT-driven magnetization dynamics in Pt/Co/AlOx dots during current pulse injection [5]. Time-resolved x-ray images with 25 nm spatial and 100 ps temporal resolution reveal that switching is achieved within the duration of a sub-ns current pulse by the fast nucleation of an inverted domain at the edge of the dot and propagation of a tilted domain wall across the dot. Our measurements reveal how the magnetic symmetry is broken by the concerted action of both damping-like and field-like SOT and show that reproducible switching events can be obtained for over 10E12 reversal cycles.
[1] Miron et al., Nature 476, 189 (2011). [2] Garello et al., Nat. Nanotech. 8, 587 (2013). [3] Garello et al., Appl. Phys. Lett. 105, 212402 (2014). [4] Prenat et al., IEEE Trans. on Multi-Scale Comp. 2, 149 (2016). [5] Baumgartner et al., Nat. Nanotech., in press (2017).


FM-1:IL07  Electric-field Controlled Nucleation of Magnetic Skyrmions at Room Temperature
T. SRIVASTAVA, M. SCHOTT, A. HALLAL, M. CHSHIEV, S. AUFFRET, C. BARADUC, H. BEA, Univ. Grenoble Alpes, CEA, CNRS, Grenoble INPi, INAC, SPINTEC, Grenoble, France; M. SCHOTT, A. BERNAND-MANTEL, L. RANNO, V. KRIZAKOVA, S. PIZZINI, D. GIVORD, Institut NEEL/CNRS/UGA-Grenoble/ Grenoble-INP, France

Skyrmions are chiral magnetic textures with immense potential for spintronic devices [1] especially in ultra-thin heavy metal(HM)/ferromagnet(FM)/insulator(I) trilayers, where their formation is governed by interfacial Dzyaloshinskii-Moriya interactions. In these structures, a lower current density is necessary to move the skyrmions as compared to standard domain walls. For storage or logic applications, their nucleation should also be controllable with low power and localized means, such as electric field. In unpatterned HM/FM/I films with perpendicular magnetic anisotropy, we have formed skyrmions by applying a small out-of-plane magnetic field on a demagnetized sample presenting labyrinth domains. We have applied a gate voltage on the skyrmions while observing them with polar Kerr microscopy. The application of –0.33V/nm (+0.33V/nm) switches skyrmions on (off) [2]. We have realized this skyrmion switch in two systems, namely Pt/Co/AlOx and Ta/FeCoB/TaOx. This switch is compatible with energy efficient skyrmion-based applications and constitutes an important milestone towards the use of skyrmions for memory or logic.
[1] N.S. Kiselev et al., J. Phys. D. Appl. Phys. 44, 392001 (2011); A. Fert et al., Nat. Nanotech. 8, 152 (2013). [2] M. Schott et al., Nano.Lett., 17, 3006 (2017)


FM-1:IL10  Negative Capacitance: Theory, Practice and Limitations
Y.J. KIM, M.H. PARK, CHEOL SEONG HWANG, Department of Material Science & Engineering and Inter-University Semiconductor Research Center, Seoul National University, Seoul, South Korea

The negative capacitance (NC) effects in ferroelectric materials have emerged as the possible solution to low-power transistor devices and high-charge-density capacitors. This is fundamentally based on the total energy argument of dielectric/ferroelectric (DE/FE) stacked film using the Landau-Ginzburg-Devonshire (LGD) theory. Although the subthreshold swing


FM-1:L12  Self-assembled Network of Nanostructures in BiFeO3 Thin Films
B. COLSON, V. FUENTES, C.FRONTERA, F. SANDIUMENGE, LL. BALCELLS, B. MARTINEZ, A. POMAR, ICMAB-CSIC, Campus UAB, Bellaterra, Spain; D. COLSON, M. VIRET, A. FORGET, SPEC/IRASMIS/ DSM, CEA-Saclay, Gif-sur Yvette, France; J. SANTISO, ICN2, CSIC, BIST, Campus UAB, Bellaterra, Spain; Z. KONSTANTINOVIC, N. LAZAREVIC, M. SCEPANOVIC, Z.V. POPOVIC, CSSPNM, Institute of Physics Belgrade, University of Belgrade, Serbia

Well defined structures at nanometric scale of multiferroic materials present an increasing interest due to their unique physical properties and potential applications. Fabrication of artificial nanostructures requires sophisticated technology and has been recognized as a hard-attainable issue. For these reasons the fabrication of ordered nanostructures, via spontaneous self-organization, is a topic of major relevance. Complex oxide thin films are often elastically strained and this lattice strain can, in some cases, select preferential growth modes leading to the appearance of different self-organized morphologies. In this work we report on the controlled fabrication of a self-assembled network of nanostructures (pits and grooves) in highly epitaxial BiFeO3 thin films. As previously shown in the case of manganite thin films [1], the remarkable degree of ordering is achieved using vicinal substrates with well-defined step-terrace morphology. Nanostructured BiFeO3 thin films show mixed-phase morphology, exhibiting the giant ferroelectric polarization close to the theoretical limit. These particular microstructures open a huge playground for future applications in multiferroic nanomaterials.
[1] Z. Konstantinovic et al., Small 5, 265 (2009); Nanoscale, 5, 1001 (2013)

 
Session FM-2 - Resistance Switching (RRAM) and Phase Change (PCM) Memories

FM-2:IL01  Simplified Resistive Memory for CMOS Integration
M.N. KOZICKI, School of Electrical, Computer and Energy Engineering, Arizona State University, Tempe, AZ, USA

Switching in metal-oxide-metal structures has been studied for decades, particularly in the context of memory. Several materials have been exploited but it is the resistive switching of copper-silica-metal structures that is the current focus of our work. Such switching involves the transport of copper ions and their subsequent reduction to form a conducting metallic pathway in the insulating oxide. A bias opposite to that used for writing is used to dissolve the filament to return the device to its high resistance state. Non-volatile memory devices that rely on the same mechanism have been successfully commercialized but they typically utilize materials or processes that are not common in integrated circuit manufacturing, or introduce back-end-of-line (BEOL) processing restrictions that semiconductor foundries find unacceptable. This has reduced the adoption of these technologies and forced them into specialized applications. On the other hand, the materials and processes of the copper-silica system are widely used in the semiconductor industry, particularly in the BEOL which is the ideal place to integrate such simple devices. This talk will highlight the materials, processes, and performance of our copper-silica devices in the context of their integration with CMOS logic.


FM-2:L02  Impact of the Transistor Current Control on the Multiple Resistive Switching Properties in 1T1R RRAM Devices
E. PEREZ, M.K. MAHADEVAIAH, Ch. WENGER, IHP, Frankfurt (Oder), Germany; C. Zambelli, P. Olivo, Università degli Studi di Ferrara, Ferrara, Italy

Resistive random access memories (RRAM) based on hafnium oxides with 1-transistor-1-resistor (1T1R) architecture are one of the most promising candidates for the next generation of non-volatile memory applications. In order to add multi-bit capability to such memory devices, accurate control of multiple resistive states is required. Incremental step pulse with verify algorithm (ISPVA) has shown to be an effective way to reduce the cell-to-cell variability by establishing current targets (Ith): a top value in high resistive state (HRS) and a bottom value in low resistive state (LRS). One strategy to perform multi-bit operation is to define a HRS and several LRS. To ensure discrete LRS levels not only bottom current values must be defined but also top current values by using the current compliance control through the transistor, effectively modulated by gate voltage (Vg). Therefore, to find suitable Ith-Vg pairs it is crucial in order to define multiple LRS levels. The performance issues detected by using this programming scheme in such 1T1R configuration, caused by the interaction between the resistor and transistor, have not been well understood. In this work, the impact of several Ith-Vg pairs on the switching properties and their performance issues have been investigated.


FM-2:L03  Atomic Layer Deposition of Oxygen Deficient TaOx Dielectrics for Resistive Switching Memory Applications
A.M. MARKEEV¹, K.V. EGOROV¹, D.S. KUZMICHEV¹, D.I. MYKOTA¹, V.A. GRITSENKO², T.V. PEREVALOV², C.S. HWANG^1,3, ¹Moscow Institute of Physics and Technology, Dolgoprudny, Moscow region, Russia; ²Rzhanov Institute of Semiconductor Physics SB RAS, Novosibirsk, Russia; ³Department of Materials Science and Engineering and Inter-University Semiconductor Research Center, Seoul National University, Seoul, South Korea

According to the general point of view it is favorable for functional oxide ReRAM layer to have a defined oxygen deficiency. Therefore, it is of a great interest to find a way to control oxygen content in metal oxide ALD films. In this work a robust ALD process of oxygen-deficient TaOx film was developed basing on plasma-activated hydrogen as the reactant and alkoxide compound Ta(OC2H5)5 as Ta precursor, which already has a Ta−O bond. In this approach, the critical idea is to remove the C2H5 by forming volatile molecules via reaction with plasma-activated H* without breaking Ta−O bond in the chemisorbed Ta precursor molecules. Another critical feature is a novel reaction pathway of partial reduction of the growing oxide in a controllable manner due to the attack of H* on Ta−O bonds. As a result, the new way to control the oxygen vacancy concentration in PEALD TaOx films was confirmed by the in situ XPS measurements of Ta 4f core and the valence band photoelectron lines. The oxygen-deficient PEALD TaOx layer played a fluent role as the oxygen vacancy source for the resistance switching. Thus, the PEALD TaOx based stack demonstrated a reliable performance up to 5 × 106 switching cycles.


FM-2:L04  Tuning the Switching Properties of ZnO Thin Film Memristors by Al Doping via ALD
C. GIOVINAZZO, C. RICCIARDI, S. PORRO, Politecnico di Torino, Department of Applied Science and Technology, Torino, Italy

The properties of a Al:ZnO (AZO) thin films can be tuned over the range of values defined for the two pure oxides: ZnO films via ALD are poly-crystalline and conducting, while Al2O3 films are amorphous and insulating. In this work, the tuning of ZnO memristive properties by Al doping via ALD is explored. Pt/AZO/Cu thin film memristors were fabricated by ALD doping process. During the deposition, a single cycle of Al2O3 was periodically inserted in a given number of ZnO cycles. The result is an alloy film, where ZnO locally contains Al2O3 partial layers, which intrinsically change its structural and electrical properties. Structure, composition and morphology of AZO films were investigated, revealing a gradual decrease of crystallinity with doping. While Hall measurements show that AZO films are naturally n-type doped with low resistivity; mobility decreases adding Al partial layers that act as local barriers. AZO memristors were tested in voltage sweep mode, showing a stable I-V characteristic and good endurance. We observed a trend in setting voltages depending on Al content of the film and HRS/LRS ratio variation due to the change in resistivity of the thin films.


FM-2:IL05  Resistive Memory Technology and Applications
H.J. BARNABY, School of Electrical, Computer and Energy Engineering Arizona State University, Tempe, AZ, USA

Resistive memory (RRAM) devices exhibit resistance switching through bias dependent ion transport through solid-state ion conductors and reduction/oxidation reactions. RRAM technologies are categorized in two major classes: valence change memory (VCM) and electrochemical memory (ECM). A VCM cell consists of a transition metal oxide switching layer sandwiched between reactive and inert electrodes. The switching mechanism in VCM is due to the modulation of conductivity in a channel formed and dissolved by the movement of oxygen anions through the channel under the forces of electric field. Resistive switching in ECM RRAM (or CBRAM) is facilitated by metallic filament formation and dissolution through the drift of metal cations (typically Ag or Cu) in solid state electrolytes The electrolytes used in ECM devices are typically oxide or chalcogenide glass films. RRAM devices can be used in a range of applications such as nonvolatile memory, threshold logic, neuromorphic systems, and cyber-security applications. This talk provides an overview RRAM operation and discusses designs that utilize the technology for applications including radiation tolerant nonvolatile memory, novel computational architectures, and physical unclonable functions to protect the internet of things.


FM-2:L06  Dynamics of the Electroforming Process of Valence Change Memory Cells
S. MENZEL¹, A. MARCHEWKA², T. HEISIG¹, C. BÄUMER¹, R. DITTMANN¹, R. WASER^1,2, ¹Forschungszentrum Jülich, Peter Grünberg Institut (PGI-7), Jülich, Germany; ²RWTH Aachen, Institut für Werkstoffe der Elektrotechnik (IWE 2), Aachen, Germany

Redox-based resistive switching devices based on the valence change mechanism (VCM) have attracted great attention due to their potential use in future non-volatile memories, logic-in-memory or neuromorphic applications. A VCM cell consists of a simple metal-oxide-metal structure. An initial electroforming step is required before the cell can be switched repeatedly between different resistance states. In this work the dynamics of the forming process of VCM cells are investigated by means of experiments and simulation. In the experimental part, the electroforming process of Pt/SrTiO3/Nb-doped SrTiO3 VCM cells is studied using voltage sweeps and voltage pulses. The measured current transients upon application of voltage pulses show that a gradual current increase precedes a rapid increase in current. Moreover, the forming time decreases with increasing voltage. A numerical drift-diffusion model combining bulk ion transport with electrode kinetics for oxygen exchange at the anode will be presented. It will be shown that this model reproduces qualitatively the experimental results. Moreover, the simulation results show that the filament growth direction during electroforming depends on the rate of oxygen exchange compared to the rate of ionic transport inside the oxide layer.


FM-2:L07  Effect of Heavy Ion Radiation on Resistive Switching in HfOx based RRAM Devices Grown by MBE
S. PETZOLD¹, S.U. SHARATH¹, J. LEMKE¹, E. HILDEBRANDT¹, C. TRAUTMANN², L. ALFF¹, ¹Institute of Materials Science, Technische Universität Darmstadt, Darmstadt, Germany; ²Materials Research Department, Gesellschaft für Schwerionenforschung (GSI), Darmstadt, Germany 

Classical FLASH technology shows limited radiation tolerance making it sensible to radiation induced errors, e.g., single event upsets. Charge based memories are becoming more and more sensitive to radiation with further downscaling leading to a lack of radiation hard memories beyond Mbit storage capacities. There is, thus, great demand for new intrinsically radiation hard NVM technologies. Since the storage of information in Resistive Random Access Memory (RRAM) is ascribed to a conductive filament of oxygen vacancies, the information is not based on charge but on a physical microstructure related state within the device, providing high resistance towards ionizing radiation, as shown for high energy protons, γ-radiation and X-ray-radiation. This makes RRAM based on hafnium oxide interesting for applications in harsh environments, such as energy plants or (aero) space applications. For such applications, the effect of heavy ion radiation on the switching behaviour needs to be investigated. Therefore, hafnium oxide based RRAM (TiN/HfOx/Pt/Au) stacks[1] were irradiated with 1.1 GeV Au-ions with fluences up to 10^12 ions/cm^2 and evaluated regarding pristine resistance, forming voltage, and data retention.
[1] S. U. Sharath, Adv. Funct. Mater. 27, 1700432 (2017)


FM-2:L08  Nonvolatile Impedance Switching in Electroforming-free BFO Memristors
M. KIANI^1,4, NAN DU^1,4, N. MANJUNATH¹, D. BÜRGER^1,4, I. SKORUPA^1,3, S.E. SCHULZ^4,6, O.G. SCHMIDT^1,2, H. SCHMIDT^1,4,5, ¹Materials systems for Nanoelectronics, Chemnitz University of Technology, Chemnitz, Germany; ²Institute for Integrative Nanosciences, IFW Dresden, Dresden, Germany; ³Helmholtz-Zentrum Dresden-Rossendorf, Institute of Ion Beam Physics and Materials Research, Dresden, Germany; ⁴Fraunhofer-Institut für Elektronische Nanosysteme, Abteilung Back-End of Line, Chemnitz, Germany; ⁵Leibniz-Institut für Photonische Technologien e.V., Jena, Germany; ⁶Chemnitz University of Technology, Center for Microtechnologies, Chemnitz, Germany

Materials revealing non-volatile impedance and resistance switching, allow to fashion slimmer, smaller and smarter electronic devices. BiFeO3 (BFO) based memristors demonstrate reliable non-volatile hysteretic memristive characteristics and have been considered as promising candidate in applications in the field of modular light detection [1], neuromorphic computing [2] and data analyzing [3]. Frequency dependent capacitance (C) and conductance (G) data have been modeled. In an electrical equivalent circuit (EEC) a combination of constant phase element (CPE) with series resistor and inductor has been utilized to model BFO in low resistance state (LRS), and a second CPE has been added in the EEC for BFO in high resistance state (HRS). Impedance observed on the Nyquist plot suggests that BFO memristors do not form a perfect capacitor, hence CPEs are used in modelling. To validate the EEC model, circuit elements namely, capacitance, bulk resistance and contact impedance have been determined from the model. Temperature dependent impedance data have been analyzed to extract the activation energy Ea for hopping transport in BFO memristors in both LRS and HRS.
[1] T. You, L.P. Selvaraj, N. Du, Dr. D. Bürger, I. Skorupa, Prof. O.G. Schmidt, Dr. H. Schmidt, Adv. Electron. Mater, 10.1002/aelm.201500352 ( 2016). [2] N. Du, M. Kiani, C. Mayer, D. Bürger, I. Skorupa, Prof. O.G. Schmidt, Dr. H. Schmidt Front. Neurosci., 9:227 (2015). [3] T. You, N. Du, Dr. D. Bürger, I. Skorupa, Prof. O.G. Schmidt, Dr. H. Schmidt , Adv. Funct. Mater.,10.1002/adfm.201303365 (2014)​.


FM-2:L09  Resistive Switching Modes and Dynamics in Defect Engineered Polycrystalline HfOx based RRAM Devices
S.U. SHARATH¹, S. PETZOLD¹, E. HILDEBRANDT¹, J. KURIAN¹, P. KOMISSINSKIY¹, C. WENGER², T. SCHROEDER^2,3, L. ALFF¹, ¹Institute of Materials Science, TU Darmstadt, Darmstadt, Germany; ²IHP, Frankfurt (Oder), Germany; ³Brandenburgische Technische Universität, Cottbus, Germany

Hafnium oxide (HfO2) based RRAM devices have huge potential as next generation non-volatile memories and in neuromorphic electronics. Defect engineering approaches offer potential towards improving performance and yield of HfO2-RRAM via lowered forming voltages and tuning the switching dynamics to achieve analog switching synapses. Here, defects have been introduced intrinsically in polycrystalline HfOx thin films, via stoichiometry engineering and doping using reactive molecular beam epitaxy (MBE) at CMOS compatible temperatures. Oxygen vacancy stabilized modifications of tetragonal-HfO1.5 and cubic-La:HfO2-x were compared to monoclinic-HfO2 in a Pt/HfOx/TiN device configuration. While a bipolar switching occurs in all the devices irrespective of defect concentration, switching modes like unipolar and threshold switching is favored only at higher oxygen stoichiometry.[1] This suggests suppression of thermochemical switching via higher heat dissipation and lowered concentration gradient of oxygen vacancies in oxygen deficient HfO1.5. Moreover, defect altered interplay between Joule heating/electric field stabilizes intermediate quantized conductance states during set/reset processes as well as a gradual reset behavior.
[1]S. U. Sharath et al. Adv. Funct. Mater. 2017, 27, 1700432.


FM-2:IL10  Mechanisms and Nanoscale Processes in Resistive Switching Memories
I. VALOV, Research Centre Juelich, Electronic Materials (PGI-7), Juelich, Germany

Redox-based memristive devices are intensively studied in respect application as non-volatile memories, selector devices and building units for neuromorphic computing. Despite the huge progress on developing and utilizing these devices in the recent years, one of the main unresolved issue is to define a clear relation between materials properties/physical processes and device performance. Thus, the stat-of-the-art design rules are predominantly phenomenological. In this contribution, the latest achievements on revealing the relation materials-performance will be presented. The focus is set on the importance of the interfaces, interface interactions, formation of intermediate films and the importance of suboxides. The influence/necessity of moisture will be presented as well as the set/reset mechanism, based entirely on electrochemical passivation, reduction. The fundamental mechanism in terms of sequence of electrochemical processes at the nanoscale will be highlighted.


FM-2:L11  MIS Structures with Interfacial Graphene for RRAM Applications: A Nanoscale and Device Level Characterization
S. CLARAMUNT, QIAN WU, M. PORTI, M. NAFRIA, X. AYMERICH, Electronic Engineering Dept., Universitat Autònoma de Barcelona, Bellaterra, Spain  

Resistive Random Access Memory (RRAM), because of its scalability, non-volatility and high performance [1] is a promising alternative for future memories. RRAM devices are based in the Resistive Switching (RS) phenomenon, which can be associated to the formation/destruction of a conduction filament (CF) through an insulator [2], but the physical phenomenon behind this process and its dependence with the electrode material is still unclear. Some authors have proposed the use of novel materials like graphene to improve the device performance. In this work, the electrical properties and variability (at device level and at the nanoscale with Conductive Atomic Force Microscope, CAFM) of MIS structures with and without graphene as an interfacial layer between the top electrode and the dielectric are studied. When graphene is present several resistive switching cycles can be measured meanwhile the standard MIS structures cannot be switched. CAFM analysis showed that the graphene layer prevents the complete structural damage of the material during a forming process, confirming the protective role of graphene in a MGIS structure.
[1] J.Y. Seok et.al., Adv. Func. Mater., vol. 24, no. 34, pp. 5316-5339, 2004. [2] R. Waser et. al. Nat. Materials, vol. 6, no. 11, pp. 833-840, 2007.


FM-2:L12  An Electrochemical Metallization Memory Cell Based on a Single ZnO Nanowire
G. MILANO^{1, 2}, S. PORRO¹, C. RICCIARDI¹, ¹Politecnico di Torino, Department of Applied Science and Technology, Torino, Italy, ²Istituto Italiano di Tecnologia, Center for Sustainable Future Technologies, Torino, Italy

Metal oxide nanowire (NW) based devices are a promising platform to localize and investigate resistive switching mechanisms in Electrochemical Metallization memory cells. This work reports the resistive switching properties of CVD-grown wurtzite (hexagonal) single-crystalline ZnO NWs. Isolated single ZnO NWs were dispersed on a SiO2 substrate and electrically contacted, by Electron-Beam Lithography patterning, with an electrochemically active electrode (Ag) and an inert counter electrode (Pt). The electrical characterization by I-V measurements revealed the typical hysteresis loop (conversely, resistive switching was not observed in similar devices fabricated using Pt symmetric inert electrodes). After the forming process, the devices exhibited high OFF/ON ratio (up to 10^3), low SET and RESET voltages (< 5 V) and endurance of at least 60 cycles. The switching mechanism was found to be based on the electromigration of Ag adatoms on the NW surface, as confirmed by the direct observation of Ag nanoclusters on the NW surface, whose density depended on the current compliance applied to the device. In addition, we found that the switching properties are strongly affected by ambient moisture, and resistive switching was not observed in case of vacuum conditions.


FM-2:L14  Multi-level Resistive Switching in Core-Shell ZnO Nanowires Exhibiting Tunable Surface States
S. PORRO, F. RISPLENDI, G. MILANO, G. CICERO, C. RICCIARDI, Politecnico di Torino, Department of Applied Science and Technology, Torino, Italy

Surface and quantum confinement effects in one-dimensional systems such as ZnO nanowires (NW) are responsible for novel electrical properties, and can be exploited to tune the electrical transport at the nanoscale. In this work, core-shell structures of CVD-grown ZnO NW arrays coated by a thin plasma-processed polyacrylic acid film exhibited Resistive Switching characterized by multiple resistance states. These were ascribed to the electrical interaction of the polymer with ZnO surfaces during the application of electrical bias voltage. Differently from previous literature, where the formation of a conductive filament is the preferential switching mechanism, we propose that the resistance can be additionally tuned introducing surface states induced by redox reactions, which can induce depletion (reset) and accumulation (set) of electrons in the n-type ZnO. The mechanism of switching between multiple steps, as probed by DFT calculations, was associated to redox reactions involving species at the interface, each characterized by a given redox potential. Therefore, as shown experimentally, the transition from HRS to LRS occurred in multiple steps, induced by specific and stable threshold voltages, each associated to a redox reaction involving a given species at the interface.


FM-2:IL15  Exploiting Nanoscale Effects in Phase Change Memories
M. SALINGA, RWTH Aachen University, Aachen, Germany

Phase Change Memory (PCM) has been developed into a mature technology capable of storing information in a fast and non-volatile way. In addition, the potential of employing this technology for neuromorphic computing has recently been demonstrated. However, whether PCM will constitute a useful element of the electronics ecosystem for many years to come, depends crucially on how the requirements arising from continued scaling to higher device densities can be met by the materials at the core of the technology. Continued miniaturization of PCM devices is not only prescribed in order to achieve memories with higher data density and neuromorphic hardware capable of processing larger amounts of information. Smaller PCM elements are also incentivized by the prospect of increased power efficiency per operation as less material needs to be heated up for switching. Here, it will be described how miniaturization lets interface effects become increasingly important and as a consequence the crystallization kinetics of phase change materials, when confined into nanometer sized structures, can change significantly depending on the material it is in contact with. Based on this analysis, the implications of such nanoscale effects will be discussed and possible ways of exploiting them proposed.


FM-2:IL16  Ovonic Threshold Switching Selector: From Material Engineering to Device Performance Improvement
G. NAVARRO, A. VERDY, V. SOUSA, M. BERNARD, G. BOURGEOIS, F. FILLOT, N. CASTELLANI, C. SABBIONE, P. NOE, J. GARRIONE, L. FELLOUH, G. MOLAS, M.C. CYRILLE, E. NOWAK, CEA-LETI, Grenoble, France

Crossbar Resistive Memory is considered a good candidate for the next generation of memory architectures. It allows low access and writing time, elemental memory cell scaling and 3D integration to target higher memory density [1]. However, the maximum array size and the memory operations reliability are limited by problems related to the sneak-paths and to the cell-to-cell programming disturbs. Hence, the selector device engineering becomes fundamental to achieve a reliable crossbar array, in particular targeting a high non linearity, low leakage current and high cycling endurance. We show here how Ovonic Threshold Switch (OTS) selectors based on GeSe materials, known for their high temperature stability and high band-gap energy, can be improved in terms of switching performance by Sb and N doping. Moreover, we present how material engineering, in order to achieve a high sub threshold non-linearity, can allow a new non-switching reading strategy enhancing the device programming endurance.
[1] DerChang Kau et al., 2009 IEEE IEDM, pp. 1-4, 2009.


FM-2:IL17  Atomistic Simulations of Crystallization and Aging of GeTe Nanowires
S. GABARDI, E. BALDI, E. BOSONI, D. CAMPI, S. CARAVATI, M. BERNASCONI, Dipartimento di Scienza dei Materiali, Università di Milano-Bicocca, Milano, Italy; G.C. SOSSO, Department of Physics and Astronomy, University College London, UK; J. BEHLER, Institut fuer Physikalische Chemie, Theoretische Chemie, Universitaet Goettingen, Germany

Nanowires made of chalcogenide alloys are of interest for application in phase change nonvolatile memories. For this application insights into their thermal properties and in particular into the crystallization kinetics at the atomic level are crucial. To this aim, we have performed large scale atomistic simulations of ultrathin nanowires (9 nm in diameter) of the prototypical phase change compound GeTe. We made use of an interatomic potential generated with a neural network fitting of a large ab initio database to compute the thermal properties of the nanowires. By melting a portion of the nanowire, we investigated the velocity of recrystallization as a function of temperature. The simulations show that the melting temperature of the nanowire is about 100 K below the melting of the bulk which yields a reduction by about a factor of 2 in the maximum crystallization speed. Further, the analysis of the structural properties of the amorphous phase of the nanowire suggests a possible origin of the reduction of the resistance drift observed experimentally in nanowires with respect to the bulk. This analysis is based on our previous study of the microscopic origin of the resistance drift in the bulk.


FM-2:IL19  Van der Waals Gap Reconfiguration and Switching Mechanism in Ge-Sb-Te Superlattices
A.V. KOLOBOV, P. FONS, Y. SAITO, J. TOMINAGA, Nanoelectronics Research Institute, National Institute of Advanced Industrial Science & Technology, Tsukuba Central 5, Tsukuba, Ibaraki, Japan

Interfacial phase-change memory (iPCM) in the form of GeTe-Sb2Te3 superlattices (SLs) is superior in performance compared to composite material of the same average composition. In early work, switching in SLs was attributed to a change in the structure of GeTe blocks such as a transition between the symmetric Petrov or inverted Petrov phases and a ferro phase. Subsequent TEM studies of such superlattices indicated strong Ge/Sb intermixing and challenged the initial models based on Ge switching. In this work we propose an alternative model for switching in SLs based on reconfiguration of van der Waals gaps. We demonstrate that such reconfiguration, leading to a local change in stoichiometry, which predominantly depends on the switching of Sb atoms, is responsible for a drastic change in the density of states between the SET and RESET states of iPCM.


FM-2:L20  Unipolar Resistive Switching in Pt/MgO/TaOx/Ta/Ru Thin Films
C. DIAS, L.M. GUERRA, B.D. BORDALO, J. VENTURA, IFIMUP-IN and Department of Physics and Astronomy, Faculty of Sciences, Porto, Portugal; HUA LV, S. CARDOSO, P.P. FREITAS, INESC-MN and IN - Institute of Nanoscience and Nanotechnology, Lisboa, Portugal; A.M. FERRARIA, A.M. BOTELHO DO REGO, Centro de Química-Física Molecular and IN, IST, Universidade de Lisboa, Lisboa, Portugal 

The downscaling of conventional nonvolatile memories is reaching its fundamental limit, thus existing a tremendous demand for novel scalable approaches. Among the most interesting technologies are metal-insulator-metal memristive devices that exhibit nonvolatile resistive switching (RS) between low (ON) and high resistive states (OFF). RS comprises Set (OFF to ON) and Reset (ON to OFF) processes. RS is being studied for a large range of applications such as memory (RRAMs) or neuromorphic. We obtained unipolar RS in Pt/MgO/Ta/Ru devices, after a forming step, with a good resistance separation for more than 14h. This behavior was observed with all voltage polarity combinations and the devices can also be arbitrarily switched between modes. We found that the operation is better (lower variability and operating voltages) for positive Set and negative Reset voltages (Pt grounded). The XPS analysis revealed that there is a TaOx interfacial layer resulting from the migration of oxygen from the MgO, given the high affinity of Ta to oxygen. The RS behavior can then be explained by the formation and rupture of oxygen vacancies filaments from the Ta to the Pt electrode, with contribution from joule heating. The shape of the filament was also inferred to be related with the operation mode.


FM-2:L21  Epitaxial Stabilization of Single Crystalline Semiconducting and Metallic NbO2
J.E. BOSCHKER, S. BIN ANOOZ, T. MARKURT, M. ALBRECHT, J. SCHWARZKOPF, Leibniz Institute for Crystal Growth, Berlin, Germany; S. BIN ANOOZ, Physics Department, Faculty of Science, Hadhramout University, Mukalla, Yemen; P. PETRIK, B. KALAS, Institute of Technical Physics and Materials Science, Budapest, Hungary; M. RAMSTEINER, Paul-Drude-Institut für Festkörperelektronik, Berlin, Germany

Niobium oxides have attractive electrical properties that can be exploited in electronic devices. For example, its hysteretic and non-linear I-V characteristics are of interest for selector and memory devices. Especially, NbO2 has attracted scientific interest as it exhibits a semiconductor-metal transition at 1080K. Fundamental studies on the properties of niobium oxides are however hindered by the wide range of possible polymorphs and compositions of niobium oxide and the lack of high quality materials. In this paper, we report on single crystalline NbO2 thin films that are epitaxially grown on TiO2 substrates by pulsed laser deposition. The anisotropic structural and optical properties of thin films grown at different substrate temperatures were determined by x-ray diffraction, transmission electron microscopy, Raman spectroscopy and spectroscopic ellipsometry. Furthermore, we demonstrate the epitaxial stabilization of the semiconducting NbO2 and metallic NbO2 at room temperature by an adequate choice of the deposition parameters. Possible implications for resistive switching in niobium based devices will be discussed with regard to film structure and composition.


FM-2:L22  Magnetism as a Probe of the Origin of Memristive Switching in Oxide Semiconductors
X.L. WANG, A. RUOTOLO, Dept. of Materials Science and Engineering, City University of Hong Kong, Kowloon, Hong Kong SAR, China

We show that resistive switching in n-type ferromagnetic Mn-ZnO and p-type antiferromagnetic NiO coexists with a switching of the magnetic phase. Thin films of these oxides were sandwiched between two metallic electrodes and resistive switching was induced. We found that a switching of the resistance corresponds to a switching of the magnetic phase in the film [1, 2]. By measuring the magnetic properties of the devices in the two resistive states, we can draw important conclusions on the underlying switching mechanism. For instance, in NiO we could exclude that switching was due to the formation of Ni-ion filaments across the device. We have demonstrated [2] that the switching is due to the formation and rapture of oxygen-vacancy filaments.
[1] X. L. Wang et al., Appl. Phys. Lett. 104, 062409 (2014). [2] X. L. Wang et al., Appl. Phys. Lett. 103, 223508 (2013).


FM-2:IL23  Random Telegraph Noise in Resistive Switching Memory Devices
F.M. PUGLISI, University of Modena and Reggio Emilia, Modena, Italy

RRAM devices are of great interest nowadays for many applications. Besides being excellent non-volatile memory (NVM) devices, they can be employed in logic-in-memory architectures, neuromorphic computing, physical unclonable functions, and random number generators. However, these devices still exhibit a number of issues, among which a rather important one is Random Telegraph Noise (RTN). It consists in the abrupt and random switch of the measured current/voltage among (two or more) discrete values. RTN has been observed in all RRAM devices, despite differences in materials and process steps, and is a detrimental factor for many applications, e.g., causing synaptic weight fluctuations in neuromorphic networks or decreasing the noise margin in NVM arrays. Therefore, it is important to develop appropriate tools to correctly characterize RTN. In this talk, we devise a methodology to self-consistently measure, analyze, and interpret RTN. This methodology is then successfully applied to the case of HfO2 RRAM devices. Understanding RTN in these devices allows developing a physics-based compact model for RTN in RRAMs. This model can be readily plugged in existing RRAM compact models to allow circuit designers accounting for RTN during the design stage.


FM-2:L24  Characterization of Low Frequency Noise in Oxygen Engineered Hafnium Oxide-based RRAM Devices
E. PIROS¹, M. LONSKY², S. PETZOLD¹, S.U. SHARATH¹, E. HILDEBRANDT¹, B. KRAH¹, J. MÜLLER², L. ALFF¹, ¹Technische Universität Darmstadt, Darmstadt, Germany; ²Goethe-Universität Frankfurt, Germany

Resistive random access memory (RRAM) cells are suited candidates for the next generation of non-volatile memory devices. The simple metal-insulator-metal structure, very good scalability, CMOS compatibility, and low production costs offer a possible solution to the current challenges in data storage. In the wide field of RRAM devices the most investigated insulator materials are transition metal oxides, among which hafnium oxide-based materials attract special attention due to their established use as high-k dielectric in CMOS technology. An appealing question regarding RRAM devices is their reliability and cycle-to-cycle variability, where noise analysis can be implemented as a powerful investigation tool. Here we analyzed the low frequency noise behavior in stoichiometric and oxygen-deficient hafnia based RRAM devices [1]. The noise power spectra were compared for the different resistance states of the devices.
[1] S. U. Sharath, Adv. Funct. Mater. 27, 1700432 (2017)


FM-2:IL25  Anionic and Protonic Carriers for Oxide-based Neuromorphic Computing
J.L.M. RUPP, Electrochemical Materials, Massachusetts Institute of Technology, MIT, USA

The next generation of information memories and neuromorphic computer logics in electronics rely largely on solving fundamental questions of mass and charge transport of oxygen ionic defects in materials and their structures. Here, understanding the defect kinetics in the solid state material building blocks and their interfaces with respect to lattice, charge carrier types and interfacial strains are the prerequisite to design new material properties beyond classic doping. Through this presentation basic theory1 and model experiments for solid state oxides their impedances and memristance2, electro-chemo-mechanics and lattice strain3-5 modulations is being discussed as a new route for tuning material and properties in ionic conducting oxide film structures up to new device prototypes based on resistive switching. Central are the making of new oxide film materials components, and manipulation of the charge carrier transfer and defect chemistry (based on ionic, electronic and protonic carriers), which alter directly the resistive switching property and future computing performances. A careful study on the influence of microstructure and defect states vs. the materials` diffusion characteristics is in focus. For this, we suggest novel oxide heterostructure building blocks and show in-situ spectroscopic and microscopic techniques coupled with electrochemical micro-measurements to probe near order structural bond strength changes relative to ionic, protonic and electronic diffusion kinetics and the materials integration to new optimized device architectures and computing operation schemes. 

 
Session FM-3 - Emerging Applications for Non-volatile Memories

FM-3:IL01  Learning in Spiking Neural Networks using Phase Change Memory Synapses
B. RAJENDRAN, Department of Electrical and Computer Engineering, New Jersey Institute of Technology, Newark, NJ, USA

Inspired by the computational efficiency of the brain, there are several research efforts worldwide to build information processing systems that can surpass the performance of today's von Neumann computing systems for automated data analysis and decision making. In this talk, we present some recent results from our research on using nanoscale phase change memory (PCM) devices for event-driven supervised learning in spiking neural networks. A comprehensive phenomenological model of MLC PCM that accurately captures the statistics of programming inter-cell and intra-cell variability was developed and used to study the performance of exemplary learning tasks using an event-driven supervised learning algorithm. Our studies suggest that learning tasks for spiking neural networks could be implemented efficiently using PCM synapses, in spite of their non-ideal operating characteristics. Further technology improvements to minimize programming energy, inter-cell and intra-cell programming variability, resistance drift etc., as well as architectural solutions to mitigate some of these issues will be required to realize learning systems approaching the integration density and power budget of the human brain.


FM-3:IL03  RRAM Based New Computing Paradigms
JINFENG KANG, P. HUANG, R.Z. HAN, C. LIU, Y.N. JIANG, Z. ZHOU, Y.C. XIANG, L.F. LIU, X.Y. LIU, Peking University, Beijing, China

With the development of information technologies such as AI, IOT, Big Data etc., it is challenge to process the high volume of and complicated data in more efficiency, low power and smart ways. Meanwhile, the CMOS scaling is approaching the fundamental limits and the traditional processor architecture is suffering from the von Neumann Bottleneck due to the communication between memory and CPU. Therefore, it is imperative to innovate new computing paradigms beyond traditional von Neumann architecture based on new function devices. Resistive-switching random access memory (RRAM) is one of the promising candidates for the new function devices, which was initially proposed for memory application. Based on RRAM, we developed the novel computing paradigms to boost the efficiency and speed of information processing with low power and data communication between memories and CPU. In this talk, we will present the novel computing paradigms based on RRAM, which will benefit to future AI, IOT, big data applications.


FM-3:IL04  Diffusive Memristor as a Building Block for a Novel True Random Number Generator
QIANGFEI XIA, HAO JIANG, University of Massachusetts Amherst, MA, USA

One of the biggest obstacles for memristive devices to be widely used as memories is the intrinsic variability in device switching parameters. The non-uniformity in cycle-to-cycle switching, on the other hand, may be a source of randomness for hardware security application such as true random number generator (TRNG). We propose and demonstrate a novel TRNG based on a newly invented device, diffusive memristor – a volatile device that relies on the diffusion dynamics of metal atoms in the memristive layer. When an electric pulse is applied across the two terminals of a diffusive memristor, the device switches to a low resistance state after a random delay time. We utilized the randomness in the delay time and built a TRNG with a diffusive memristor, a comparator, an AND-gate and a counter. Because of the self-OFF switching behavior and small size, our new TRNG has advantages in energy efficiency and circuis complexity. More impotantly, the random bits generated from the diffusive memristor TRNG passed all 15 NIST randomness tests without any post-processing, first of its kind among TRNGs exploiting memristive switching. This work opens the avenue for memristors in hardware security applications in the era of Internet of Things (IoT).


FM-3:IL05   Design and CMOS Co-integration of ReRAM Devices and Crossbar Arrays for Neuromorphic Applications 
Y. LEBLEBICI, EPFL, Switzerland

Resistive RAM (ReRAM) elements based on transition-metal oxide layers are rapidly becoming viable options for nonvolatile multi-level information storage as well as for the implementation of memristive synaptic functions in neuromorphic operations, allowing easy integration with conventional CMOS technologies. In this talk, we will review the ongoing research at EPFL on the realization of various ReRAM elements based on TiOx, TaOx, WOx and HfOx layers tailored for low voltage operation, as well as the design and co-integration of the CMOS peripheral circuitry for the read/write operations. The fabrication and characterization of ReRAM devices are explored using wafer-scale post-processing as well as using individual samples. In particular, the chip embedding platform enabling post-processing of diced samples for fabrication of memristive elements will be discussed, and examples will be provided for potential neuromorphic functions such as spike-timing-dependent-plasticity (STDP) and back-propagation algorithms implemented on cross-bar arrays.


FM-3:L06  Specific Switching Algorithms for Emerging Applications of RRAM based Memories
E. PEREZ, M.K. MAHADEVAIAH, Ch. WENGER, IHP, Frankfurt (Oder), Germany; C. ZAMBELLI, P. OLIVO, Università degli Studi di Ferrara, Ferrara, Italy; F.M. PUGLISI, P. PAVAN, Università di Modena e Reggio Emilia, Modena, Italy; M. ZIEGLER, H. KOHLSTEDT, Kiel University, Kiel, Germany

Recent advances in the performance of resistive random access memory (RRAM) have led to a significant interest in CMOS technologies. Although RRAM based memory arrays demonstrated excellent performance parameters, the intercell and intracell variability still a critical issue. While the intracell variability can be optimized for particular memory cells, the intercell variability must be minimized by using specific programming routines like incremental step pulses with an additional read and verify algorithm. In addition, major concerns are cycling variability, and resistance distributions degradation. Controlling these phenomena requires employing program-verify schemes. In this talk, the dispersion-aware PV (DAPV) scheme to minimize resistance dispersion and achieve reliable multi-bit operation is evaluated. However, statistical variations can be tolerated in computing applications like neuromorphic networks. Recently, Hebbian learning as an important biological concept has been realized with single RRAM devices. The synaptic behaviour of RRAM devices can be evaluated by applying successive algorithms consisting of set or reset pulses with different lengths and amplitudes. These algorithms can be used to study the synaptic functionality of single cells.


FM-3:L07  Evolution of a-IGZO Thin-film Transistor Memory: From Incapability of Electrical Erase to Achievement of Multi-level Cell
SHI-JIN DING, School of Microelectronics, Fudan University, Shanghai, China

Amorphous IGZO TFT nonvolatile memories have been studied recently for flexible electronics and transparent panel systems because a-IGZO has many advantages of high mobility, low process temperature, and visible-light transparency etc. However, IGZO is a natural n-type semiconductor, thus the memory cannot be electrically erased by injection of holes. In this talk, I will talk about several types of a-IGZO TFT memories. (1) The a-IGZO TFT memory with Zn-doped Al2O3 charge storage layer. It exhibited fast programming and very good electron retention, but it could not be erased electrically because of very deep traps in the charge trapping layer. Further, the light-erase was explored. (2) The a-IGZO TFT memories with ALD metal nanocrystals. The devices exhibited a high programming efficiency and incapability of electrical erase. The UV erasing or light-assisted electrical erasing was thus studied. (3) Multi-level TFT memory cells with a specially processed oxide semiconductor charge trapping layer. The fresh devices could be programmed with positive or negative gate bias, and the programmed devices could be erased electrically under negative or positive bias. Moreover, the memories exhibited fast program/erase characteristics even under low voltages, and multi-level cell as well.


FM-3:IL08  Spintronics Memories for Bio-inspired Computing
D. QUERLIOZ, A.F. VINCENT, A. MIZRAHI, D. VODENICAREVIC, T. HIRTZLIN, J.S. FRIEDMAN, N. LOCATELLI, J. GROLLIER, University Paris-Sud, Orsay, France

Neuromorphic hardware holds the promise of low-energy, intelligent and highly adaptable computing systems. However, one of the major challenges of fabricating such hardware is building ultra-high density networks out of complex processing units interlinked by tunable connections. Nanometer-scale devices exploiting spin electronics (or spintronics) can be a key technology in this context. In particular, magnetic tunnel junctions are well suited for this purpose because of their multiple tunable functionalities. One such functionality, non-volatile memory, can provide massive embedded memory in unconventional circuits, thus escaping the von-Neumann bottleneck arising when memory and processors are located separately. Other features of spintronic devices that could be beneficial for bio-inspired computing include tunable fast non-linear dynamics, controlled stochasticity, and the ability of single devices to change functions in different operating conditions. In this talk, I will introduce several neuromorphic designs based on these ideas, exploiting both standard and prospective spintronic devices. I will review the recent advances in this direction, and the challenges towards fully integrated spintronics-CMOS bio-inspired hardware.


FM-3:IL09  Spintronic Analog Memory for Neuromorphic Computing
SHUNSUKE FUKAMI^1,2,3,4, W.A. BORDERS¹, A. KURENKOV¹, C. ZHANG^1,2, S. DUTTAGUPTA^1,3, H. OHNO^1,2,3,4,5, ¹Laboratory for Nanoelectronics and Spintronics, RIEC, Tohoku University, Japan; ²Center for Spintronics Integrated Systems, Tohoku University, Japan; ³Center for Innovative Integrated Electronic Systems, Tohoku University, Japan; ⁴Center for Spintronics Research Network, Tohoku University, Japan; ⁵WPI-Advanced Institute for Materials Research, Tohoku University, Japan

Neuromorphic computing has attracted great attention because of its capability to execute complex tasks that the conventional von Neumann computers cannot readily complete. Here, we present an analog spintronic device based artificial neural network. Spintronic devices, in general, offer non-volatility and virtually infinite endurance, showing promise for realization of low-power “edge” neuromorphic computing hardware with online learning capability. An antiferromagnet-ferromagnet heterostructure operated by spin-orbit torque is employed, which allows us to control the magnetization state in an analog manner and thus can be used as an artificial synapse [1]. Using the developed artificial neural network with the analog spintronic device, we show a proof-of-concept demonstration of an associative memory operation based on the Hopfield model, which is a representative model of neuromorphic computing [2].
This work is partly supported by R&D Project for ICT Key Technology of MEXT, JSPS KAKENHI No. 17H06093, JST-OPERA, and ImPACT Program of CSTI.
[1] S. Fukami et al., Nature Materials, vol. 15, 535 (2016). [2] W. A. Borders et al., Appl. Phys. Express, vol. 10, 013007 (2017).


FM-3:L10  Understanding Organic Spintronic Devices and their Applications to Neuromorphic Computing
A. RIMINUCCI¹, ZHI-GANG YU2, M. CALBUCCI¹, R. CECCHINI¹, P. GRAZIOSI¹, M. PREZIOSO3, I. BERGENTI¹, A. DEDIU¹, ¹Institute for the Study of Nanostructured Materials, CNR, Bologna, Italy; ²ISP/Applied Sciences Laboratory, Washington State University, Spokane, WA, USA; ³Department of Electrical and Computer Engineering, University of California at Santa Barbara, Santa Barbara, CA, USA

Despite the extensive research efforts devoted to the understanding of spin transport in organic semiconductors, key results, such as the Hanle effect, are still missing. To understand these devices, that show magnetoresistance and resistive multistability, we have expanded on the impurity band model introduced by Yu. In our devices we envisage oxygen to be the impurity and bistability is attributed its migration within the organic semiconductor/AlOx bilayer. This model provides a comprehensive explanation of the phenomena observed in the low resistance devices, including the interplay between the resistive state and the magnetoresistance. These devices can be used for neuromorphic computing: with positive voltage pulses of fixed duration and amplitude it was possible to increase gradually the conductance of the device, while with negative voltage pulses it was possible to decrease it. This type of synapse can be used in several types of neural networks, from a simple single layer perceptron to spiking neural networks. The presence of magnetoresistance adds a unique tool to effect parallel, selective changes on the weight of synapses. This possibility provides a novel tool for neural network training.


FM-3:L12  Emerging Applications for Electroforming-free Perovskite Memristors
H. SCHMIDT^1,2,3,5, NAN DU^1,4, D. BÜRGER^1,4, I. SKORUPA³, R. ECKE⁴, S.E. SCHULZ⁴, ¹Materials systems for Nanoelectronics, Chemnitz University of Technology, Chemnitz, Germany; ²Faculty of Physics, Friedrich-Schiller University of Jena, Jena, Germany; ³Leibniz-Institut für Photonische Technologien e.V. (IPHT), Jena, Germany; ⁴Helmholtz-Zentrum Dresden-Rossendorf, Institute of Ion Beam Physics and Materials Research, Dresden, Germany; ⁵Fraunhofer-Institut für Elektronische Nanosysteme, Abteilung Back-End of Line, Chemnitz, Germany

The most essential synaptic modification rule of competitive Hebbian learning, spike timing dependent plasticity (STDP), has been achieved mainly in resistive switching (RS) cells, e.g in BFO-based RS cells, using a short and simplified single pairing potentiating and depressing spike sequence [N. Du, H.S. et al., Front. Neurosci. 9 (2015)]. The main problems of RS cells which need further improvements include mainly the issue of electroforming step, reproducibility in RS characteristics and/or RS material, so-called “voltage-time” dilemma and sneak paths. Electroforming-free bipolar BiFeO3 [T. You, H.S. et al., Adv. Funct. Mat. 24 (2014)] and unipolar YMnO3 [A. Bogusz, H.S. et al., AIP Advances 4 (2014)] perovskite memristors have been developed. Furthermore, it has been demonstrated that the large non-linearity of memristors can be used to realize the so-called memristor-based encryption [N. Du, H.S. et al., J. Appl. Phys. 115 (2014)]. In cryptography a physical unclonable function (PUF) is an entity that derives secret information from physical properties of an object. The challenge in building a PUF is to find entities which reliably produce an unpredictable device-unique secret. One interesting candidate to be used as a PUF is the perovskite memristor BiFeO3.


FM-3:IL15  Interfacing Organic Memristors with Neurons in a Bio-hybrid Network
S. IANNOTTA, S. BATTISTONI, V. EROKHIN, CNR-IMEM, Parma, Italy

Organic biosensing and memristive devices are paving the way to novel perspectives both in mimicking and interfacing natural systems while representing an ideally suitable platform for applications in bio-electronics and bio-medicine. They also represent a very promising playground for neuromorphic devices and systems. Our contribution to the field including applications to bioelectronics will be discussed together with the recent achievements in developing memristive devices based both on PANI/PEO and PEDOT::PSS polymers. The evolution from simple logic elements up to the first organic based Perceptron will be discussed envisaging the perspectives. The results and potential of the approach based on organic electrochemical devices will be discussed together with the great potential of these devices in the framework of other methods and technologies already established in the field. The novel approach based on interfacing memristive devices with biological cells and systems will be discussed together with the demonstration of a memristive organic-bio-hybrids that will be proposed and discussed as a potential for novel very promising applications. The first demonstration of an artificial synapsis interfacing, stimulating and reacting to a living brain will be reported.


FM-3:L16  A RRAM-based Self-organizing Neural Network
M. PEDRO, J. MARTIN-MARTINEZ, R. RODRIGUEZ, M. NAFRIA, Departament d'Enginyeria Electrònica, Universitat Autonoma de Barcelona (UAB), Cerdanyola del Valles, Barcelona, Spain 

Plasticity in RRAM devices opens a wide range of possibilities for testing unsupervised learning algorithms in RRAM neuromorphic crossbar arrays. A popular artificial neural network is the Self-Organizing Map (SOM). SOMs are widely used for pattern recognition, clustering applications, etc. and are able to process large amounts of data without external assistance. SOMs are inspired in the topographic maps found in the brain cortex regions, related to sensory processing. However, the software-based algorithm for achieving a SOM is quite distant to the biological mechanism of such topographic organization, often leading to high computational requirements, being unable to be implemented electronically with the current available technology. In this work, a SOM is implemented in a simulated RRAM crossbar array, where an adapted compact model of the devices based on experimental data, which includes device-level variability, is considered for the synapse update. An unsupervised learning algorithm based in plasticity and lateral communication between output neurons is applied. In order to test the device-level variability impact on the RRAM-based SOM performance, an image-processing application is presented. Results show that simulated SOM presents high tolerance to variability.


FM-3:IL17  Computational Memory: The First Step Towards Non-von Neumann Computing?
A. SEBASTIAN, IBM Research - Zurich, Rueschlikon, Switzerland

The emergence of data-centric cognitive computing and end of CMOS scaling laws are posing key challenges to the field of information technology. It is becoming increasingly clear that to build efficient cognitive computers, we need to transition to non-von Neumann architectures. A first step in this direction could be computational memory where certain tasks such as bit-wise operations and arithmetic operations are performed in place in memory. Resistive memory devices, in particular, phase-change memory (PCM) devices could play a key role. It is shown that if data is stored in PCM devices, then the physical attributes of these devices can be exploited to achieve in-place computation. PCM devices when organized in a cross-bar array can be used to perform matrix-vector multiplications with very low computational complexity. An application of this concept for the problem of compressed sensing is presented. Next, I will present the concept of mixed-precision computing to counter the lack of precision arising from such matrix-vector multiplication operations. The example of solving systems of linear equations is presented. Finally, I will present an example of computational memory where the state dynamics of PCM devices is used to detect temporal correlations.


FM-3:IL19  Dynamics of HfO2-based Resistive Memory for Neuromorphic Computation
S. BRIVIO, J. FRASCAROLI, E. COVI, S. SPIGA, Laboratorio MDM, IMM-CNR, Agrate Brianza, Italy

Architectures supporting efficient running of deep neural networks (DNNs) algorithms, as graphic or tensor processing units, are already a reality in present electronics. Non-volatile resistive memories (RRAMs) allow the design of new DNNs accelerators with improved speed and energy consumption. A real breakthrough is the realization of spiking neural networks (SNNs), which, beyond parallelism and co-localization of logic and memory, allow asynchronous operation, real time data analysis and adaptation. SNNs need to leverage the switching dynamics of RRAMs operated in an unconventional way. To this aim, we study and describe the analog switching dynamics of filamentary-type HfO2-based RRAMs in relation to the kinetic and stochasticity of the switching mechanism. The emerging picture highlights the limits and advantages of using the switching dynamics for alternative computing schemes. SNN simulations are performed including the modeling of the RRAMs. The network, whose fabrication in silicon has already been proven, relies on a generalization of spike time and rate dependent plasticity [Brader Neur. Comp. 2007]. The learning dynamics and the performances of the network are analyzed in relation to the RRAM switching dynamics and tested against the classification of MNIST digits.