FH - 4th International Conference
Emerging Materials, Technologies and Applications for Non-volatile Memory and Memristive Devices


FH-1.1:L02  New Insight into the Origin of the Resistive Switching Mechanism and the Low Power Consumption of GeTe/Sb2Te3 Superlattices in Phase-change Memory Devices
D. TEREBENEC1, N. Castellani1, N. Bernier1, V. Sever1, P. Kowalczyk1, M. Bernard1, M.-C. Cyrille1, N.-P. Tran1, J.-Y. Raty1, 2, F. d’Acapito3, F. Hippert4, P. Noé1, 1Université Grenoble Alpes, CEA, LETI, Grenoble, France; 2CESAM-Physics of Solids Interfaces and Nanostructures, B5, Université de Liège, Belgium; 3CNR-IOM-OGG c/o ESRF - The European Synchrotron, Grenoble, France; 4Université Grenoble Alpes, CNRS, Grenoble INP, LMGP, Grenoble, France

Van der Waals-layered GeTe/Sb2Te3 superlattices (SLs) have demonstrated outstanding performances for use in Phase-Change Memory (PCM). However, the physical mechanism behind the resistance change and performance improvements in SL devices compared with standard PCM is still highly debated. [(GeTe)2/(Sb2Te3)m]n SLs are made by periodically stacking ultra-thin GeTe and Sb2Te3 crystalline layers. In this contribution, we will describe by means of advanced XRD, EXAFS and HAADF-STEM experiments the complex structure of prototypical [(GeTe)2/(Sb2Te3)m]n SLs deposited by magnetron sputtering. Then, we will highlight the origin of the resistive switching in our [(GeTe)2/(Sb2Te3)8]4 SL by means of TEM images of a device programmed in the High Resistive State. Finally, we will give new insights on the correlation between the crystalline structure and the improved programming performances of [(GeTe)2/(Sb2Te3)m]n SLs devices.

FH-1.1:L03  Analysis of Ge-incorporation and Crystallization Study in GexSbyTez Phase Change Alloys for Automotive Applications
A. Diaz Fattorini, F. De Nicola, M. Bertelli, S. De Simone, V. Mussi, R. Calarco, M. Longo, CNR-IMM Unit of Rome, Rome, Italy

Phase change memory (PCM), based on chalcogenide materials, are good candidates for embedded memories in the automotive industry, where devices must be able to operate reliably at high temperatures (higher than 160 °C) for at least ten years. Chalcogenide alloys like Ge2Sb2Te5 (GST225) are already a standard to realize PCM devices. However, one of the major disadvantages of GST225 single layers is the low crystallization temperature (Tc), that results in low thermal stability and limited data retention. An alternative way to doping the material to increase its Tc is to grow Ge-rich GST alloys. In this work, several GexSb2Te5 and GexSb2Te3 layers were grown by RF-sputtering and their thermal stability was studied by X-ray diffraction (XRD) and Raman spectroscopy as a function of temperature. The compositional analysis was performed by X-ray Fluorescence (XRF). The results showed that both the Tc and the Ge-segregation amount grow linearly with the Ge-content in the alloys.
This project has received funding from the European Union's Horizon 2020 research and innovation program under Grant Agreement No. 824957 (“BeforeHand:” Boosting Performance of Phase Change Devices by Hetero‐ and Nanostructure Material Design).

FH-1.1:L04  Electronic Properties of Phase Change Material Heterostructures based on Ge-rich Ge-Sb-Te Alloys
F. RIGHI RIVA1, C. Chèze1, E. Placidi2, G. Di Bella2, S. Prili1, A. Diaz Fattorini3, S. Cecchi4, V. Mussi3, S. De Simone3, M. Longo3, R. Calarco3, M. Bernasconi5, O. Abou El Kheir5, F. Arciprete1, 1Department of Physics, University of Rome Tor Vergata, Roma, Italy; 2Department of Physics, Sapienza University of Rome, Rome, Italy; 3CNR Institute for Microelectronics and Microsystems-IMM, Consiglio Nazionale delle Ricerche, Roma, Italy; 4Paul-Drude-Institut für Festkörperelektronik, Berlin, Germany; 5Department of Materials Science, University of Milano-Bicocca, Milan, Italy

Ge2Sb2Te5 (GST225) is the prototype alloy used for non volatile optical and electrical memories in most practical uses. We investigate here combinations of different PCM in multilayered heterostructures for applications requiring higher crystallization temperatures (e. g. automotive smart devices). Among the chalcogenide alloys suitable for automotive applications, Ge-rich Ge-Sb-Te (GST), In-Sb-Te and In-Ge-Te alloys have been proposed. The combination of PCM layers with different physical properties might help their use in embedded devices with the best compromise between reducing latency and power consumption, increasing endurance, retention and storage capability. In this context, we present here experimental investigations on chemical state and composition, electronic and structural properties of Sb2Te3/(Ge-rich GST) and GST225/(Ge-rich GST) heterostructures grown by co-evaporation of the constitutive elements from solid source Knudsen cells in ultra-high vacuum conditions. The evolution of the electronic properties of the heterostructures is studied by a combination of in-situ X-ray (XPS) and Ultraviolet photoemission spectroscopy (UPS) for increasing thicknesses of the Ge-rich GST layer focusing on the interface between the two PCM films.

FH-1.1:IL05  The Role of (dis)Order on the Structural-electronic Interplay in Amporphous GexSe1-x: A Microscopic Investigation
F. Tavanti, A. Slassi, A. Calzolari, CNR-NANO Istituto Nanoscienze, Modena, Italy

Amorphous chalcogenides, such as GexSe1-x, have been proposed as ovonic switching materials for volatile memories and selectors. Their structural and electrical properties are related to the presence of short- and medium-range structures in the amorphous phase [1]. In order to understand the local geometry-dependent properties of these systems, we employed a combined approach based on classical molecular dynamics and DFT calculations. We employed cutting-edge techniques to analyse the short- and medium-range order from both a chemical-physical and topological point of view. This allows for a thorough understanding of the disorder at the microscopic level. Our results [2] indicate that a small difference in the stoichiometry affects not only the atomic structure (e.g. Se-rich systems are less ordered than Ge-rich ones), but also the mobility band-gap and the trap states, which are mainly responsible for the transport properties of GexSe1-x.
[1] Tavanti, et al. ACS Appl. Electron. Mater. 2, 2961 (2020). [2] Tavanti, et al. submitted (2021).

FH-1.1:L06  Selective MOCVD Growth of Sb-Te and In-Ge-Te Nanostructures on Templated Substrates for Phase Change Memories
R. Cecchini1, 2, M. LONGO1, 3, C. Martella1, A. Lamperti1, S. Brivio1, F. Rossi4, L. Lazzarini4, E. Varesi5, 1CNR-IMM, Unit of Agrate Brianza, Agrate Brianza (MB), Italy; 2CNR-IMM, Bologna, Italy; 3CNR-IMM, Rome, Italy; 4CNR-IMEM, Parma, Italy; 5Micron Technology Inc., Vimercate (MB), Italy

Chalcogenide compounds of the Sb-Te and In-Ge-Te systems are attractive materials for the realization of advanced devices, ranging from electronics to photonics and sensors. In particular, they can be used for non-volatile Phase Change Memories (PCMs) and neuromorphic applications. Improved efficiency and higher storage/processing density in such devices can be gained by employing high-aspect-ratio chalcogenide nanostructures, for the realization of which bottom-up growth processes are inherently better suited than top-down ones. In this work, it is shown that selective, bottom-up growth of chalcogenide nanostructures can be obtained by Metal-Organic Chemical Vapor Deposition (MOCVD) on substrates patterned with CMOS technology-compatible processes and materials. Namely, Sb-Te and In-Ge-Te nanostructures with high crystal quality were grown at the bottom of pores (~130 nm width) fabricated on a SiO2 masking layer and using a CoSi2 layer both as the catalytic element for the growth and the bottom electrode for electrical measurements. The morphological, compositional, and structural properties of the nanostructures, as well as the selectivity of the growth as a function of the different MOCVD process parameters, were investigated.

FH-1.1:L07  Shape Controlled Self-assembly of Core Shell Ge-Sb-Te/Sb2Te3 Nanowires by MOCVD
A. Kumar, C. Wiemer, CNR-IMM Unit of Agrate Brianza, Agrate Brianza, Italy; R. Cecchini. CNR-IMM Unit of Bologna, Bologna, Italy; M. Scuderi, G. Nicotra, CNR-IMM Unit of Catania, Catania, Italy; V. Mussi, S.D. Simone, R. Calarco, M. Longo, CNR-IMM Unit of Rome, Rome, Italy

Phase change memories have emerged as a good candidate for non-volatile memory for the Internet of Things (IoT), due to their high performances and scalability. In the present scenario, we are investigating the chalcogenide nanowires (NWs), which represent a good direct-access device for the functionality study of highly nanoscaled PCM cells. Moreover, the core-shell NW geometry, not obtainable by top-down methods, enables the investigation of size effects with possibility of multi-level PCM. Here, it will be shown that the synthesis of core-shell Ge-Sb-Te/Sb2Te3 nanowires (NWs) was investigated by metalorganic chemical vapor deposition (MOCVD) on Si(100) and SiO2/Si(100) substrates, coupled to the use of nano Au catalyst for the vapour-liquid-solid (VLS) mechanism, yielding NWs with core diameters down to 10 nm. Different metalorganic precursors (such as Lewis/Alkali) and deposition parameters (such as temperature, pressure) were used for the core NW self-assembly growth and for the shell overgrowth. SEM, TXRF, XRD, TEM and Raman analysis were carried out for assessing the morphological, compositional and microstructural properties of the NWs.
This project has received funding from the European Union's Horizon 2020 research and innovation program under Grant Agreement No. 82495

FH-1.1:IL09  Deep Learning Inference and Training using Computational Phase-change Memory
M. LE GALLO, A. Sebastian, IBM Research Europe, Rueschlikon, Switzerland

We are at the pinnacle of a revolution in artificial intelligence (AI) and cognitive computing. The computing systems that run today’s AI algorithms are based on the von Neumann architecture, which is inefficient at shuttling huge amounts of data back and forth at high speeds. Thus, to build efficient cognitive computers, we need to transition to novel architectures where memory and processing are better collocated. In a first level of inspiration, the idea would be to build computing units where memory and processing co-exist in some form. In-memory computing is one such approach where the physical attributes of memory devices are exploited to perform computational tasks with very high areal and energy efficiency. In this talk, I will present our latest efforts in employing such a computational memory architecture for performing inference and training of deep neural networks. First, the phase-change memory technology we use will be described. Next, the application of computational memory to neural network inference will be explained, and experimental results will be presented based on a state-of-the-art fully-integrated computational phase-change memory core. Finally, our latest efforts in employing such an architecture also for training neural networks will be elaborated.

FH-1.1:L11  C-based Phase-change Material Nanocomposites for Improved Phase-change Memory
J. PATERSON1, M. Tomelleri1, 2, D.Térébenec1, R.R. Chahine1, M. Bernard1, N. Bernier1, N. Castellani1, M.-C. Cyrille1, N.-P. Tran1, R. Cravero3, O. Bourgeois3, V.M. Giordano4, J.-Y. Raty1, 5, F. d’Acapito6, D. Benoit2, F. Hippert7, P. Noé1, 1Univ. Grenoble Alpes, CEA, LETI, Grenoble, France; 2STMicroelectronics, Crolles, France; 3Univ. Grenoble Alpes, CNRS, Grenoble INP, Institut Néel, Grenoble, France; 4ILM, UMR5306 Université Lyon 1-CNRS, Villeurbanne, France; 5Physics of Solids Interfaces and Nanostructures, B5, Université de Liège, Sart-Tilman, Belgium; 6CNR-IOM-OGG c/o ESRF - The European Synchrotron, Grenoble, France; 7Univ. Grenoble Alpes, CNRS, Grenoble INP, LMGP, Grenoble, France

Phase-change (PC) memories are now considered as one of the best candidates for Storage Class Memory or neuromorphic devices [1] but they still have to overcome several challenges, in particular the reduction in programming power [2]. In this context, we show how tailoring the PC material (PCM) polycrystalline structure at the nanoscale by introducing C atoms improves the performances of memory devices using such nanocomposites. First, the impact of C on the PCM polycrystalline structure is studied by means of XRD, FTIR and STEM-EELS in C-doped GeTe thin films. We propose that the reduction in thermal conductivity of C-doped GeTe polycrystalline thin films, compared with undoped films, results from an increase in phonon scattering at multiple PCM/C interfaces. The latter is thus responsible for the decrease in programming current of PC memory devices [3, 4]. Then, we show how another type of nanocomposites consisting of PCM/C multilayers (MLs) with a well-controlled nanostructure (with a PCM layer, GeTe or Ge2Sb2Te5 of a few nm and C layers as thin as 0.5 nm) can be obtained by industrial magnetron co-sputtering. The introduction of GeTe/C MLs into PC memory devices allows for the reduction of RESET current, demonstrating that PCM nanocomposites are a key material for improving the performance of PC memories.
[1] P. Noé, B.J. Kooi, and M. Wuttig, Phys. Status Solidi RRL 15, 2100078 (2021).
[2] P. Noé et al., Semicond. Sci. Technol. 33, 013002 (2018).
[3] Q. Hubert et al., in Extended Abstracts of the 2013 International Conference on Solid State Devices and Materials (The Japan Society of Applied Physics, 2013), pp. 550–551.
[4] G. Betti Beneventi, et al., Solid-State Electronics 65–66, 197 (2011)

FH-1.1:L12  Thermal Engineering Targeting Low Power Consumption for Next Generation Phase-change Memory
C. DE CAMARET, Y. Le-Friec, STMicroelectronics, Crolles, France; G. Bourgeois, O. Cueto, V. Meli, V. Beugin, N. Castellani, M.C. Cyrille, F. Andrieu, J. Arcamone, G. Navarro, CEA-LETI, Grenoble, France

Phase Change Memory (PCM) has demonstrated to fulfill both Storage Class Memory and more strict automotive market requirements. Nevertheless, programming current reduction remains a challenge to target ultra-low power applications in next generation technology nodes. Being the adiabatic limit still far away [1], the thermal engineering of the device demonstrated to be a valid strategy to achieve a higher programming reliability in PCM [2]. In this work, we demonstrate the high thermal efficiency enhancement achieved thanks to the engineering of the dielectrics surrounding the heater element and the chalcogenide layer in state of the art 4kb PCM arrays based on “Wall” structure. Moreover, we highlight the correlation between the thermal properties and the electrical characteristics of devices based on both common Ge2Sb2Te5 and optimized Ge rich GeSbTe. These results are supported by electro thermal TCAD simulations performed in Synopsys Sentaurus Device simulator. Thanks to these results, we demonstrate the strong benefit of thermal engineering to target low power consumption in next generation PCM.
[1] F. Xiong et al., IEDM 2016. [2] A.L. Serra et al., Solid State Electron., 186, 108111 (2021).

FH-1.1:L13  Overcoming the Thermal Stability Limit of Chalcogenide Phase-change Materials for High-temperature Applications
M. TOMELLERI1, 2, F. Hippert3, A. Albanese1, F. d’Acapito4, T. Farjot1, C. Sabbione1, M. Tessaire1, D. Terebenec1, N. Castellani1, V.M. Giordano5, D. Benoit2, P. Noé1, 1Univ. Grenoble Alpes, CEA, LETI, Grenoble, France; 2STMicroelectronics, Crolles, France; 3Univ. Grenoble Alpes, CNRS, Grenoble INP, LMGP, Grenoble, France; 4CNR-IOM-OGG c/o ESRF - The European Synchrotron, Grenoble, France; 5ILM, UMR 5306 Univ. Lyon 1-CNRS, Villeurbanne, France

Thanks to their unique properties, Phase-Change Materials (PCMs) have enabled the development of the most promising resistive memory technology. Despite their well-established performances, PCM memories still have technological limits to overcome for embedded applications, for which a high data retention at high temperature becomes critical. The most studied approach so far to increase the thermal stability of PCM amorphous phase has been to use Ge-Sb-Te alloys enriched in Ge. However, while the required out-diffusion of Ge excess strongly increases the crystallization temperature (Tx), its segregation risks compromising the reliability of device fabrication. As a result, it is mandatory to find new PCM compounds with a high thermal stability and with limited or without phase separation. In this presentation, we will show that Se-rich GeSe1-xTex thin films exhibit an exceptionally high Tx and a huge electrical contrast between their amorphous and crystalline states, thanks to a unique bonding mechanism recently renamed as “Metavalent Bonding”. This outstanding combination of properties in a homogeneous alloy is unprecedented among all the PCMs materials studied so far, making these compounds extremely promising for integration in memory devices requiring high data retention

FH-1.1:L14  A Novel Sb2Te3/Ge2Sb2Te5/Ge Heterostructure with Enhanced Stability for PCM Application
A. Diaz Fattorini, F. De Nicola, M. Bertelli, S. De Simone, V. Mussi, M. Longo, R. Calarco, CNR-IMM Unit of Rome, Rome, Italy; G. D’Arrigo, I. Lopez Garcia, G. Maida, S.M.S. Privitera, CNR-IMM Headquarters, Catania, Italy; M. Borghi, A. Redaelli, STMicroelectronics, Agrate B.za (MB), Italy; M.-C. Cyrille, CEA, LETI, Univ. Grenoble Alpes, Grenoble, France

The interest in Phase change Memory (PCM) devices that can withstand higher temperatures is increasing, with a search for a retention capability for 10 years > 150°C, and control in programming currents and switching speed. Among the different possibilities, our approach combines different phase change materials with opposite properties (Sb2Te3 and Ge2Sb2Te5), also introducing a confinement Ge layer to realize a novel PCM cell. The Sb2Te3/Ge2Sb2Te5/Ge heterostructures with a total thickness of 110 nm were deposited by RF sputtering and their thermal stability was analyzed by X-ray diffraction and Raman spectra, evidencing a slower crystallization dynamic (by 50°C) than in single Ge2Sb2Te5 layers. After optimization, the heterostructures were deposited onto proper single-cell vehicles prepared on Si(001) and metal contacts were nanolithographically defined. The subsequent technical analysis (I-V, R-V, R-cycles graphs for the SET/RESET states) showed that the cell has a programming current of 1.2 mA and an endurance of 2x10^5 cycles.
This project has received funding from the European Union's Horizon 2020 research and innovation program under Grant Agreement No. 824957 (“BeforeHand:” Boosting Performance of Phase Change Devices by Hetero‐ and Nanostructure Material Design).

FH-1.2:IL01  Memristors for Artificial Neural Networks: From Device Compact Modelling to Circuit and System Level Simulations
F.L. AGUIRRE, Universitat Autònoma de Barcelona, Spain

Memristive-based neuromorphic computing is nowadays in the spotlight as it is a possible candidate to overcome the so-called “memory-wall” that severely limits the traditional Von-Neumann architectures. In this regard, the hybrid memristor-CMOS integration is expected to allow the design and fabrication of dense neural networks for deep learning algorithms. A critical tool for speeding up the research in this field is the reliable electrical simulation of memristive neuromorphic systems. Following a bottom-up approach, this talk discusses the memristor compact modelling and how to use it for simulating the response of memristive artificial neural networks. The first part of this talk presents a brief summary of the available memristor models and sets the focus on the so-called memdiode model. Such model follows Prof. Chua’s memristive device theory and represents the transport equation with a double exponential expression, which resembles a diode with memory in series with a resistor. This model (in its different variants) provides a versatile yet accurate behavioural compact model for resistive switching devices of a wide range of materials with different electrical characteristics (memory resistive switching, complementary resistive switching, threshold resistive switching). These qualities turn the memdiode model into a suitable candidate for the simulation of large-scale neuromorphic circuits based on the vector-matrix multiplication method. This is precisely the topic of the second part of the talk and summarizes some of the tools for simulating artificial neural networks. The case of pattern recognition is considered for benchmarking and used to evaluate networks of up to tenths of thousands of memristors under different simulation platforms, training methods and image datasets. Nonetheless, special emphasis is placed on the SPICE electrical simulation, as when combined with the memdiode model, it allows to account for the electrical details of the devices as well as the circuit parasitics in hybrid CMOS-memristor circuits. Since the performance of memristive artificial neural networks is still hindered by plenty of non-idealities (latency, device parasitics, power dissipation, area, etc.) the realistic electrical simulation is fundamental for the development of memristive artificial neural networks with higher performance and a reduced power consumption.

FH-1.2:IL02  Evaluation Framework Assessing Memristor Technologies for Neural Network Implementations
G. BERSUKER, J. Farmer, M. Luengo-Kovac, D. Veksler, The Aerospace Corporation, Los Angeles, USA; D.Z. Gao, A.-M. El-Sayed, T. Durrant, A. Shluger, Nanolayers Research Computing LTD, London, UK; T. Rueckes, L. Cleveland, H. Luan, R. Sen, Nantero Inc., Woburn, USA

Memristor technology enabling low power-high performance neuromorphic computing must deliver well-controlled multi-level memory updates to meet the requirements of NN algorithms. To reach this goal, a complete in-depth memristor technology evaluation approach is developed that combines atomic-level material modeling (DFT, mesoscopic force fields) and statistical simulations of currents via paths through the intrinsically stochastic structure of material fabrics. Verification of the proposed physical model including electrical measurements of cell operations performed under circuitry-relevant sub-nsec pulse durations, and assessment of the impact of hardware non-idealities, such as memory-update variability, etc., on NN learning characteristics by employing TensorFlow Probability simulations has been conducted. This approach is applied to the evaluation of metal oxides (OxRAM), carbon nanotubes (CNT NRAM), and 2D material-based memristors. In particular, the developed switching model for CNT fabrics allows identifying structural features responsible for CNT cell operations and outlining possible fabrication improvements. In HfO2-based memristors, our study points to operating conditions triggering switching instability, as well as options to mitigate variability.

FH-1.2:L03  Resistive Switching in Sputtered MoS2 Memristive Devices
A. LINKENHEIL1, 3, T. Scheler2, 3, S. Park1, 3, P. Schaaf2, 3, F. Schwierz1, M. Ziegler1, 3, 1Micro- and Nanoelectronic Systems, Department of Electrical Engineering and Information Technology, TU Ilmenau, Ilmenau, Germany; 2Materials for Electrical Engineering and Electronics, Department of Electrical Engineering and Information Technology, TU Ilmenau, Ilmenau, Germany; 3Institute of Micro- and Nanotechnologies MacroNano®, TU Ilmenau, Ilmenau, Germany

Transition metal dichalcogenides (TMDCs) are promising materials for memristive devices. In particular, few-layer-TMDCs have advantageous properties such as excellent scaling behavior and the potential for an easy integration into a planar wafer technology. In this contribution, devices based on the TMDC molybdenum disulfide (MoS2) are presented, which were fabricated in a 4-inch wafer thin film technology and allow a systematic investigation of the switching mechanisms and contribution of electrodes to the device operation. In a first step, Au/MoS2/Au layer stacks were realized with sputtered 10 nm thin MoS2, and their encapsulation by high-quality silicon oxide (SiO2) was evaluated. Non-covered devices show pronounced memristive behavior with low switching voltages, which are comparable to published devices with only one or a few MoS2 layers. However, resistive switching is absent when individual cells are encapsulated. To tailor reliable devices, layer characteristics and interfacial behavior are modelled. Planar deposition methods for TMDCs and different electrode materials are applied. The electronic properties of material stacks Cu/MoS2/Al and Ag/MoS2/Al are compared to further comprehend the effects of impurity atoms and metal ion intercalation.

FH-1.2:L04  Resistive Switching Behavior of Lateral TMDC Devices
Z. GENG1, C. Zhang1, S. Park1, C. Ziebold1, S. Sharma1, F. Schwierz1, K. Rossnagel2, M. Ziegler1, 1Micro- and Nanoelectronic Systems, Department of Electrical Engineering and Information Technology, Technische Universität Ilmenau, Germany; 2Institute of Experimental and Applied Physics, Kiel University, Germany, and Ruprecht Haensel Laboratory, Deutsches Elektronen-Synchrotron DESY, Hamburg, Germany

Memristive devices (memristors) have attracted considerable attention in the electronic devices community since they possess an inherent memory effect and are able to emulate the function of biological synapses. While the majority of memristors are metal-insulator-metal structures with active regions consisting of metal oxides, in the recent past memristive switching has also been observed in two-dimensional materials, most notably in TMDCs (transition metal dichalcogenides) such as MoS2. In the present contribution, we report on the fabrication and electrical characterization of lateral TMDC memristors. For our devices, mechanically exfoliated MoS2, HfS2, and WSe2 flakes were transferred to SiO2/Si substrates and serve as active device regions. All fabricated devices show pronounced memristive behavior. Worth mentioning is (i) that our MoS2 devices exhibit memristive behavior already at very low sweep voltages (down to 1 V), much lower than the typical switching voltages of 5 … 80 V of lateral TMDC memristors reported by other groups, and (ii) that several of our WSe2 devices show high on-off ratios exceeding 1000. Moreover, the devices underwent a dynamic characterization and pulsed measurements clearly revealed a synaptic behavior of our MoS2, HfS2, and WSe2 devices.

FH-1.2:L05  Multiple Physical Time Scales in Few-nanometers Sized Graphene-SiOx-graphene Memristors
L. Pósa, Department of Physics, Budapest University of Technology and Economics and MTA-BME Condensed Matter Research Group, Budapest, Hungary, Institute for Technical Physics and Materials Science, Centre for Energy Research, Budapest, Hungary; T. Török, G. Fehérvári, Z. Balogh, A. Halbritter, Department of Physics, Budapest University of Technology and Economics and MTA-BME Condensed Matter Research Group, Budapest, Hungary

Characterization and understanding phenomena arising at single resistive switching elements is essential for utilizing their particular traits for computation. Herein we present the detailed electrical study of sub-10 nm sized graphene-SiOx-graphene phase change memory devices [1,2]. During our work we focused on the following aspects: (i) Evolution of the subthreshold electrical transport during the structural relaxation, with special regard to the period of the dead time, while the device is blocked in the high resistance state [2]. (ii) Temperature dependence of the subthreshold transport and its theoretical description. (iii) Identification of the physical processes which account for the variations in the distribution of the set time – the timespan before the onset of the transition from a high resistance OFF state to a low resistance ON state (iv) Tunability of the set time statistics via changing the reset amplitude parameter in sequential pulsed measurements. The latter phenomenon could prove useful for controlling stochasticity in memristor-based probabilistic computing applications.
[1] L. Pósa et al. npj 2D Materials and Applications 5, 57 (2021). [2] L. Pósa et al. Nano Letters 17(11), 6783, 2017

FH-1.2:IL06  The Role of Materials Design for the Transition from Digital to Analog Memories
L. Alff, Institute of Materials Science, Technische Universität Darmstadt, Darmstadt, Germany

Memristive devices are in the center of an ongoing revolutionary development in computer architectures for advanced applications in the field of data analysis, memristive FPGA design, neuromorphic computing and processing-in-memory. The different applications call for different types of memory, from digital to multi-bit and quasi-continuous analog memristive systems. It is a challenging task for materials science to design and tune the functional materials and material combinations within these memristive systems to fulfill the desired electronic functions. Here, the focus is on point defect and grain-boundary engineering in oxide materials which allow to tune forming voltage levels [1] and switching modes [2] of resistive RAM from digital to analog [3]. To correlate electronic device properties with atomic microstructure, advanced characterization tools such as operando TEM, fluctuation spectroscopy and physical picture based modelling of current voltage characteristics [4] are absolutely essential. Only by tightly connecting materials science and electrical device engineering, memristive devices will exploit their full potential.
[1] Adv Electron Mater 5, 1900484, 2019 [2] Adv Funct Mater 27, 1700432, 2017 [3] Adv Electron Mater 2000439, 2020 [4] J Appl Phys 125, 234503, 2019

FH-1.2:L07  Oxygen Engineering in Yttrium Oxide-based RRAM Devices: Suppressed Noise, Digital-to-Analog Switching Transition and Conductance Quantization for NVM, Multibit and Neuromorphic Applications
E. Piros1, S. Petzold1, R. Eilhardt2, A. Zintler2, M. Lonsky3, N. Kaiser1, T. Vogel1, E. Jalaguier4, E. Nolot4, C. Charpin4, C. Wenger5, J. Müller3, E. Miranda6, L. Molina-Luna2, L. Alff1, 1Advanced Thin Film Technology, Institute of Materials Science, Technische Universität Darmstadt, Darmstadt, Germany; 2Institute of Physics, Goethe-University Frankfurt, Frankfurt am Main, Germany; 3Advanced Electron Microscopy Division, Institute of Materials Science, Technische Universität Darmstadt, Darmstadt, Germany; 4CEA, LETI, Grenoble, France; 5IHP, Leibniz-Institut fuer innovative Mikroelektronik, Frankfurt (Oder), Germany; 6Departament d'Enginyeria Electrònica, Universitat Autònoma de Barcelona, Cerdanyola del Valles, Barcelona, Spain

With the ever increasing demand for computational capacity the need for high-density, low-power and reliable memory devices is currently at an utmost high, also motivating research for alternative computing methods, such as neuromorphic computing. Yttrium oxide-based resistive random-access memory (RRAM) can be a good choice for both application areas: we show that by oxygen engineering the switching dynamics can be tuned from two-level/digital to multilevel/analog characteristics.[1] Decreasing the oxygen content of the functional layer leads to an increased accessibility of intermediate current levels that exhibit conductance quantization, even in the set process. These states can be exploited to tune the resistance in an analog manner, as required for multi-level data storage and neuromorphic applications. Evaluating low-frequency noise in stoichiometric yttria-based RRAM reveals a universal 1/f-type behavior in the LRS, at intermediate resistive states and in the HRS.[2] The decreasing noise in the HRS with DC cycling, explained by filament stabilization, highlights the exciting possibility of suppressing noise via DC endurance training.
[1] Petzold and Piros et al., Adv. Electron. Mater. 2000439 (2020) [2] Piros et al., Phys. Rev. Applied 14, 034029 (2020)

FH-1.2:L08  Substoichiometric Hafnium Oxide Polymorphs with Semiconducting Properties
N. KAISER1, T. Vogel1, A. Zintler2, S. Petzold1, A. Arzumanov1, E. Piros1, R. Eilhardt2, L. Molina-Luna2, L. Alff1, 1Advanced Thin Film Technology Division, Institute of Materials Science, TU Darmstadt, Darmstadt, Germany; 2Advanced Electron Microscopy Division, Institute of Materials Science, TU Darmstadt, Darmstadt, Germany

Hafnium oxide is one of the most promising candidates as the functional material for next generation nonvolatile memory like OxRAM and FeRAM. The functionality of such hafnium oxide-based devices dependents crucially on the oxygen defect density. Also, various sub-oxide phases of hafnium oxide have been suggested as the oxygen-depleted conducting filament in OxRAM. However, so far, the investigation of oxygen deficient hafnium oxide is extremely scarce and no conclusive evidence of such a conducting sub-oxide has been reported in literature. To answer these open questions, we grow oxygen deficient hafnium oxide thin films over a broad range of oxidation conditions. By utilizing XRD, XPS, HRTEM, UV/Vis transmission spectroscopy, Resistivity and Hall-effect measurements, we identify two novel hafnium oxide polymorphs, namely the low temperature phase (LTP) c-HfO1.7 and hcp-HfO0.7 with semiconducting properties. Finally, the results are transferred into a comprehensive band-structure model covering the complete range from stoichiometric hafnia to metallic hafnium. The discussed results have impact on the design and the understanding of hafnium oxide based next generation non-volatile memory.

FH-1.2:L09  Statistical Evaluation of Tailored Memristive Characteristics in TiOx-HfOx Bilayer System
S. Park, C. Ziebold, S. Klett, T. Ivanov, M. Ziegler, Department of Electrical Engineering and Information Technology, & Institute of Micro and Nanotechnologies MacroNano, TU Ilmenau, Germany

Memristive device is a promising candidate for future neuromorphic systems. Here, different device requirements are needed for the multitude of computation schemes; linear and time-independent conductance modulation for machine learning, non-linear and time-dependent properties for neurobiological learning schemes. Recently, two-layer oxide systems have proven to be a suitable structure due to their possibility of tailored memristive characteristics with low power consumption and uniformity of the performance. However, technological solutions are required for a precise adjustment of layer thicknesses, defect densities in the oxide layers, and suitable area sizes of the active part of the devices. In this talk, we report on a systematic investigation of TiOx-HfOx bilayer memristive devices exploiting a fabrication process at 4’’ wafer level with respect to the thickness of the HfO2 layer, the size of the active area of the device, and the HfO2 deposition condition. Statistical evaluation of the electrical properties is discussed in terms of the variability, tailored I-V non-linearity, number of resistance states, electroforming, and operating voltage. Experimental results are supported by numerical simulations that show the contribution of the HfO2 film in the bilayer system.

FH-1.2:L11  Selective Activation of Memristive Interfaces in TaOx-based Devices by Controlling Oxygen Vacancies Dynamics at the Nanoscale
M.J. Sánchez1, C. Ferreyra2, M. Aguirre3, 4, 5, C. Acha6, S. Bengió2, J. Lecourt7, U. Luders7, D. Rubi2, 1INN-Centro Atómico Bariloche and Instituto Balseiro, San Carlos de Bariloche, Argentina; 2GIyA and INN-CONICET, CNEA, San Martín, Buenos Aires, Argentina.2INN-CONICET; 3Depto de Fíısica de Materia Condensada, Universidad de Zaragoza, Zaragoza, Spain; 4Lab. de Microscopías Avanzada (LMA), Instituto de Nanociencia de Aragón (INA)-Universidad de Zaragoza, Zaragoza, Spain; 5Instituto de Ciencias de Materiales de Aragón (ICMA), Universidad de Zaragoza, Zaragoza, Spain; 6Depto. de Física, FCEyN, Universidad de Buenos Aires and IFIBA,UBA-CONICET, Pab I, Ciudad Universitaria, Buenos Aires, Argentina.; 7CRISMAT, CNRS UMR 6508, ENSICAEN, Caen, France

The development of novel devices for neuromorphic computing and non-traditional logic operations largely relies on the fabrication of well controlled memristive systems with functionalities beyond standard bipolar behavior and digital ON-OFF states. In the present work we demonstrate for Ta2O5-based devices that it is possible to selectively activate/deactivate two series memristive interfaces in order to obtain clockwise or counter-clockwise multilevel squared remanent resistance loops, just by controlling both the electroforming process and the (a)symmetry of the applied stimuli, and independently of the nature of the used metallic electrodes. Based on our thorough characterization, analysis and modeling, we show that the physical origin of this electrical behavior relies on controlled oxygen vacancies electro-migration between three different nanoscopic zones of the active Ta2O5−xlayer. Our devices fabrication process is rather simple as it implies the room temperature deposition of only one CMOS compatible oxide -Ta-oxide- and one metal, suggesting that it might be possible to take advantage of these properties at low cost and with easy scalability.
Ref: C Ferreyra et al 2020 Nanotechnology 31 155204

FH-1.2:L12  Investigation of the Ultrafast Resistive Switching
B. SANTA, D. Molnár, A. Halbritter, Budapest University of Technology and Economics and MTA-BME Condensed Matter Research Group, Budapest, Hungary; M. Csontos, Empa, Swiss Federal Laboratories for Materials Science and Technology, Dübendorf, Switzerland

One of the key features of devices based on resistive switching is the switching speed which reached the range of few nanoseconds for various materials [1] However, the further reduction is tremendously challenging and therefore only rarely presented in the literature [2-3]. We developed a custom-designed pulsed microwave setup to examine resistive switching behaviour in the subnanosecond regime [4]. By analyzing the transmitted and reflected pulses our method is capable to resolve the evolution of the whole switching process during the switching pulse. The operation of this setup is demonstrated with metallic Ag-AgI-PtIr resistive switching junctions, verifying a less than 500 ps set event [5].
[1] W. Chen et al., J. Electroceramics 39, 109 (2017) [2] A. C. Torrezan et al., Nanotechnology 22, 485203 (2011) [3] U. Böttger et al., Sci. Rep. 10, 16391 (2020) [4] D. Molnár et al., Nanoscale 10, 19290 (2018) [5] B. Sánta et al., Beilstein J. Nanotechnol. 11, 92 (2020)

FH-1.2:IL13  Oxide ReRAM Technology NVM Solution
A. Regev, Weebit-nano, Hod Hasharon, Israel

In recent years the great interest in emerging memories is increasing as Flash memory cannot scale towards advanced technology nodes below 28nm. It is considered that the technology node of 28nm would be the last generation where e-Flash can be implemented. ReRAM technology is maturing fast demonstrating an extremely fast and energy efficient NVM solution in technologies smaller than the available embedded Flash memory can offer. Oxide ReRAM is an extremely cost competitive solution, easy to integrate into any existing CMOS fab and is cheap to manufacture. Our ReRAM is demonstrating highly reliable memory introducing high endurance and data retention tolerance with low Bit Error Rate and maturing towards commercial application usage. As it only requires very few process steps ReRAM resulting in highly cost effective solution with a potential for automotive-grade reliability features such as high-temperature tolerance and long retention.

FH-1.2:IL14  Structural Changes and Conductive Filament Formation in Silicon Oxide during Resistance Switching
A.J. Kenyon1, A. Mehonic1, M. Buckwell1, L. Montesi1, M. Singh Munde1, 2, D. Gao3, S. Hudziak1, R.J Chater4, S. Fearn4, D. McPhail4, M. Bosman2, A.L. Shluger3, 1Department of Electronic & Electrical Engineering, UCL, London, UK; 2Institute of Materials Research and Engineering, Singapore; 3Department of Physics and Astronomy and London Centre for Nanotechnology, University College London, London, UK; 4Department of Materials, Imperial College London, South Kensington Campus, London, UK

Silicon oxide has for many years provided engineers with an ideal insulator. Silicon microelectronics still relies on its physical, chemical and, above all, electrical durability; modern devices incorporate few-nanometre thick oxide layers in which the electrical stress can be extreme. Here, we report the highly dynamic structural and electrical behaviour of thin silicon oxide films under voltage stress. We show, using a combination of electrical measurements, structural measurements, in situ ion detection and mass spectroscopy, along with DFT and Monte Carlo models, that realistic device voltages can generate major changes to the oxide that are reflected in high contrast resistance switching. In some cases these changes are reversible; in others they are permanent precursors to dielectric breakdown. We also show conductive atomic force microscopy tomography of conductive filaments. These reveal the internal structure of conductive filaments, which is likely to result from the inhomogeneous nature of the amorphous oxide matrix. Our results have major implications for the use of silicon oxide in electronics and photonics – rather than a passive, stable insulator prior to breakdown it is instead a highly dynamic electrically manipulated system.

FH-1.2:IL15  Prospects and Challenges of Area-dependent Memristive Devices for Neuromorphic Computing
R. Dittmann, Peter-Grünberg-Institute (PGI-7), Research Center Jülich, Jülich, Germany

For the most common type of memristive devices, the non-volatile change of the resistance is induced by the movement of oxygen vacancies along nanosized filaments. In contrast to this, in area-dependent memristive devices the ionic motion as well as the change of the resistance occurs along the whole devices area. We will give an overview over area-dependent memristive devices based on the p-conducting Pr0.7Ca0.3MnO3 (PCMO) and different tunnel barriers and eludidate the underlying switching mechanisms by operando spectroscopy. We will discuss the differences of these devices to the common filamentary devices with respect to switching kinetics, reliability and analogue operation. Based on the very gradual type of switching, the resistance can be tuned in an analogue fashion by repeated switching with voltage pulses of the same amplitude and polarity. We investigate in detail the impact of different pulse heights and pulse lengths on the shape of the stepwise SET and RESET curves. We use these measurements as input for the simulation of a multilayer perceptron for pattern recognition and identify certain trends for the impact of the applied voltages and pulse length on the resulting shape of the measured curves and on the learning rate and accuracy of the multilayer perceptron.

FH-1.2:L16  Memristor-based Artificial Neural Network (ANN) Hardware based on Flexible Semiconducting Technology
A. Kiazadeh, J. Deuermeier, M. Pereira, R. Martins, E. Fortunato, i3N/CENIMAT, Department of Materials Science, NOVA School of Science and Technology, Universidade NOVA de Lisboa and CEMOP/UNINOVA, Campus de Caparica, Portugal

Memristors are known to present spatial and temporal variations (SV, TV respectively). SV originate in the process of device fabrication are referred to as device-to-device variations and TV are due to stochastic creation of conductive filaments, resulting in cycle-to-cycle variation. Both non-idealities negatively impact classification accuracy in ANN hardware. However, by design of an area-scaling resistive switching (RS) device, these variations can be minimized. For this purpose, RS active layers based on amorphous oxide semiconductors such as Indium-gallium-Zinc-oxide (IGZO) is employed into development of low power flexible self-service ANN platform. We present the capabilities of flexible electronic technology, IGZO-memristors combined with TFTs of the same material as support electronics for ANN machine to perform cognitive tasks. Analog control of multi-level resistance states associated with linear accumulative current-voltage characteristics in dynamic range between 2 to 100 is demonstrated. The impact of SV and TV is discussed regarding active cross bars. We present accuracy, and complexity of the support circuit for cognitive task performance. Simulation reveals the accuracy of over 94%, for online training, promising for an efficient memristor-based ANN system.

FH-1.2:L17  Advanced Analysis of Current Noise in SiOx-based Phase-change Memories
z. balogh, A. Halbritter, Dept. of Physics, Budapest University of Technology and Economics, and MTA-BME Condensed Matter Research Group, Budapest, Hungary; P. Balázs, D. Krisztián, G. Fehérvári, Dept. of Physics, Budapest University of Technology and Economics, Budapest, Hungary; L. Pósa, Dept. of Physics, Budapest University of Technology and Economics, Budapest, Hungary and Institute for Technical Physics and Materials Science, Centre for Energy Research, Hungarian Academy of Sciences, Budapest, Hungary

The investigation of electronic noise in resistive switching memories can reveal fundamental information about the microscopic processes governing the device operation [1]. In our former measurements, we have investigated the scaling of the relative current noise during the resistance and voltage tuning of graphene interconnects and demonstrated that the noise diagnostic can serve information about the microscopic conduction mechanism and the origin of the noise [2]. Here, I present the same detailed analysis of 1/f-type noise in SiOx-based phase-change memories. In these measurements, we tune the resistance of the memristor in a broad range from low to high resistance state and measure the current noise at different voltages up to the nonlinear I(V) regime. To obtain the possible noise sources, we compare the resistance scaling and the nonlinear current noise with model calculations. Furthermore, we extend our noise analysis to the proximity of the switching threshold voltages and show the precursors of the switching from the noise spectra. Finally, we demonstrate how the noise tailoring of a memristor can be used in a Hopfield neural network as a tunable noise source.
[1] Z. Balogh et al. Nano Futures, 5, 042002 (2021) [2] L. Pósa et al. 2D Mat. and App., 5, 57 (2021)

FH-1.2:IL18  Reliability of Redox-based Memristive Elements
R. Waser1,2, D. Wouters1, S. Menzel2, 1 JARA-FIT & IWE2, RWTH Aachen University, Aachen, Germany; 2JARA-FIT & PGI-7/-10, Forschungszentrum Jülich, Jülich, Germany

As for any new technology, reliability of redox-based memristive elements needs to be considered carefully. The major reliability issues of memristive elements are: data retention (i.e. the ability to store data for extended periods of time), the cyclability or endurance (i.e. the number of rewrite cycles possible), the variability of all parameters (such as LRS and HRS values, switching voltages and currents), and the read disturb (i.e. the shift of the resistance state upon multiple read operations). In the case of the variability one has to distinguish between cycle-to-cycle (C2C) variability and device-to-device (D2D) variability. In addition, various noise issues have to be considered. All these reliability characteristics have to be considered as a function of the temperature. This is linked to the question of thermal crosstalk in dense arrays of memristive elements. We will focus in our talk on devices based on the Valence Change Mechanism (VCM) with the subclasses of conducting filament and area-dependent VCMs cell of both switching polarities. We will mainly describe the physics behind the reliability characteristics and report the state of art.
And we will show to what extent this work is embedded in our joint project NEUROTEC.

FH-1.2:L20  Influence of Different SiO2 Matrices on the Properties of Cu/SiO2/W Devices
F. MAUDET1, H. Amari2, D. Marlina1, 3, V. Deshpande1, C. Dubourdieu1, 3, 1Institute Functional Oxides for Energy-Efficient Information Technology (QM-IFOX), Helmholtz-Zentrum Berlin für Materialien und Energie, Berlin, Germany; 2Leibniz-Institut für Kristallzüchtung (IKZ), Berlin, Germany; 3Freie Universität Berlin, Physical Chemistry, Berlin Germany

Conductive-bridge random access memory (CBRAM) devices are promising candidates for low-power consumption non-volatile memory applications owing to their dimensional scalability and fast switching speed. Among the proposed materials, Cu/SiO2/W stacks have demonstrated high performance and allow easy integration with CMOS technology. Because of their filamentary nature, the devices are very sensitive to local inhomogeneities in the SiO2 dielectric or interface roughness. We investigated the impact of the deposition process of SiO2 on the electrical properties of cross-point Cu/SiO2/W devices. The SiO2 film was grown either by plasma-enhanced chemical vapor deposition, by sputtering or by solgel. A statistical analysis of the electrical behavior of the devices show that the SiO2 process conditions have a major impact on the resistive switching. Several characterization methods were used to correlate the physico-chemistry of the SiO2 electrolyte to the electrical properties of the devices. Electron energy loss spectroscopy performed in a scanning transmission electron microscope will be discussed to highlight the differences between the SiO2 films. Possible routes to engineer the SiO2 matrix will be proposed in order to enhance the retention and reliability of CBRAM devices.

FH-1.2:L22  Swift Heavy Ion Irradiation Induced Effects on Emerging Memories - Correlation of Structural and Electrical Properties
T. VOGEL1, T. Kämpfe2, A.L. Serra3, N. Kaiser1, M. Lederer2, G. Lefèvre3, N. Guillaume3, R. Olivo2, T. Ali2, D. Lehninger2, E. Piros1, G. Navarro3, C. Charpin-Nicolle3, S. Petzold1, C. Trautmann4, L. Alff1, 1Advanced Thin Film Technology Division, Institute of Materials Science, TU Darmstadt, Darmstadt, Germany; 2Fraunhofer IPMS, Dresden, Germany; 3CEA LETI, Grenoble, France; 4GSI Helmholtzzentrum, Darmstadt, Germany & Institute of Materials Science, TU Darmstadt, Darmstadt, Germany

Emerging memory classes such as oxide-based resistive random-access memory (OxRAM), ferroelectric random-access memory (FeRAM) and phase-change random-access memory (PCM) are discussed as the successors of flash technology for highly-scaled device technology. Thereby, radiation hardness is of particular interest, enabling applications e.g. in space conditions. As their information storage mechanisms are not charge based, such emerging memory-based devices are promising for applications in harsh radiation environments due to their superior data retention upon ionizing radiation [1], especially compared to charge-based flash technology. In this study, we compare the effect of heavy ion irradiation, the most hazardous kind of ionizing radiation, on the structural and electrical properties of different emerging memory classes: Phase Change Memory (PCM/PCRAM) based on GST, ferroelectric random-access memory (FeRAM) and OxRAM based on HfO???? [2,3]
[1] S. Petzold et al., IEEE Trans. Nucl. Sci. 66, 1715 (2019). [2] T. Kämpfe, T. Vogel et al., 2020 Joint Conference of the IEEE International Frequency Control Symposium and International Symposium on Applications of Ferroelectrics (IFCS-ISAF). IEEE, 1-3 (2020). [3] T. Vogel et al.,IEEE Trans. Nucl. Sci. 68.8, 1542-1547 (2021).

FH-1.2:L23  Quantum Transport Phenomena in Transition Metal Oxide Memristors
T.N. Török, P. Makk, A. Halbritter, Department of Physics, Budapest University of Technology and Economics, Budapest, Hungary; M. Csontos, Empa, Swiss Federal Laboratories for Materials Science and Technology, Transport at Nanoscale Interfaces Laboratory, Switzerland

In transition metal oxide based resistive switching filaments the determination of the junction diameter is an especially challenging task. The total conductance of a quantum point contact is given by the Landauer formula, GN=G0∙∑τi. One can map the individual quantum conductance channels by determining the τi transmission eigenvalues with the help of superconducting subgap spectroscopy, i.e. by contacting the junction by superconducting electrodes, and fitting the I(V) characteristics. This analysis revealed truly single-atom-sized active region of the resistive switching in Nb/Nb2O5/Nb junctions [1]. The subgap analysis can be extended to greater conductance values as well. Random matrix theory (RMT) predicts a universal, bimodal transmission density for diffusive nanowires [2]. This density function is modified if an extended defect is introduced into the wire: the number of transparent channels are reduced [3]. Pure Ta nanowires are well described by a bimodal density originating from RMT, however for Ta/Ta2O5/Ta memristor states, an extended defect is found to affect the shape of transmission densities.
[1] T. N. Török et al. Nano Lett. 20, 1192-1200, 2020 [2] C. W. J. Beenakker Rev. of Modern Physics 69, 731-808, 1997 [3] Yu. V. Nazarov Phys. Rev. Lett. 73, 134-137, 1994

FH-2:IL01  Magneto-ionics: using Ionic Motion to Control Magnetism in Spintronics Devices
L. Herrera Diez, CNRS - Université Paris-Saclay, Palaiseau, France

Reliable and dynamic control of magnetic properties in technologically relevant magnetic materials is at the heart of a variety of emerging practical applications in spintronics. In particular, modulating the perpendicular magnetic anisotropy (PMA) and the Dzyaloshinskii Moriya interaction (DMI) as well as controlling the velocity and pinning of magnetic domain walls are key aspects for the development of memory applications. I will present a short overview of this exciting field, what it means for practical applications and discuss the physical mechanisms involved. In this context, I will show our recent results [1-3] on magneto-ionic control of PMA, magnetic domain wall motion and the DMI. Electric fields induce the migration of mobile oxygen-rich ionic species present in HfO2 across Co/HfO2 and CoFeB/HfO2 magnetic stacks inducing a gentle oxidation. The variation in oxygen content in the magnetic layers, and in particular at their interfaces with oxides and heavy atom buffer layers, produces a spin reorientation transition between in-plane and perpendicular anisotropy accompanied by a significant effect on DMI and domain wall motion. Non-volatile reversible and irreversible regimes have been identified which are attributed to a complex distribution and binding of the mobile oxygen species in the magnetic layers.
[1] L. Herrera Diez et al, ‘Nonvolatile Ionic Modification of the Dzyaloshinskii-Moriya Interaction’, Phys. Rev. Applied 12, 034005 (2019). [2] R. Pachat et al. ‘Multiple Magnetoionic Regimes in Ta/Co20Fe60B20/HfO2’, Phys. Rev. Applied 15, 064055 (2021). [3] R.Pachat et al. ‘Impact of annealing on magneto-ionic reversibility in W/CoFeB/HfO2’, submitted (2021).

FH-2:IL02  Improved Dynamical Switching Properties in Perpendicular Shape Anisotropy Magnetic Tunnel Junctions
N. Caçoilo, B.M.S. TEIXEIRA, B. DIENY, R.C. SOUSA, L.D. BUDA-PREJBEANU, O. FRUCHART, I.L. PREJBEANU, Univ. Grenoble Alpes, CEA, CNRS, Grenoble INP, SPINTEC, Grenoble, France

The perpendicular Spin-Transfer-Torque Magnetic Random-Access Memory (p-STT-MRAM) is one of the most promising emerging non-volatile memory technologies. However, these devices are limited by their thermal stability factor at technological nodes smaller than 20 nm [1, 2]. A promising solution to this problem relies on providing a perpendicular shape anisotropy of the storage layer by increasing its thickness. In this case, the magnetostatic energy adds up as a volume contribution to the interfacial anisotropy, resulting in a much larger and easily tunable effective perpendicular anisotropy than in a conventional p-STT-MRAM. This concept allows to extend the downsize scalability of the STT-MRAM at sub-10 nm technological nodes [3-4]. As the diameter decreases, the aspect-ratio of the storage layer needs to be increased so that the total magnetostatic energy is enough to meet the thermal stability factor requirements. In this work, it is shown that, above a certain thickness threshold, the magnetization reversal follows a non-coherent behavior, detrimental in terms of the switching voltage and switching time [5]. A proposed approach to circumvent this limitation is make use of additional anisotropy sources in the storage layer, which allows for the total thickness of the storage layer to be reduced, while keeping an appropriate thermal stability, leading to an overall lower switching voltage, and switching times (figure 1). This approach leads to an improvement in the PSA-MTJ technology, opening the possibility to faster and coherent reversals at sub-10 nm technological nodes.
References: [1]    C. Yoshida et al, Jpn. J. Appl. Phys. 58, SBBB05 (2019).
[2]    H. Sato et al, Jpn. J. Appl. Phys. 58, 0802A6 (2017).
[3]  N. Perrissin et al, Nanoscale 10, 12187-12195 (2018)
[4]    K. Watanabe et al, Nat. Com. 9, 663 (2018).
[5]    N. Caçoilo et al., Phys. Rev. Appl. 16, 024020 (2021)

FH-2:L04  A High-coercivity Non-volatile Thin-film Magnet based on the Shell-FM Effect
A. ÇAKIR, Department of Metallurgical and Materials Engineering, Mugla University, Mugla, Turkey; M. Acet, M. Farle, Physics, Duisburg-Essen University, Duisburg, Germany

Shell-ferromagnetism is a maroscopic monopolar effect in addition to having temperature- and magnetic-field-resistant properties useful for non-volatile magnetic memory applications.. It occurs when off-stoichiometric Heuslers in the form Ni50Mn45X5 (X: Al, Ga, In Sn, Sb) are annealed at 650 K in a magnetic-field, and the material decomposes resulting in the formation of 2-5 nm full-Heusler cubic ferromagnetic Ni50Mn25X25 laminar precipitates embedded coherently in an antiferromagnetic tetragonal Ni50Mn50 matrix. The spins at the interface shell-layer become pinned in the direction of the annealing field. The shell has a coercivity exceeding 15 T, but comprises at most 10 % of the material so that the magnetization remains low with a value of 8 Am2kg-1. To attain the full magnetization of the full-Heusler, namely 80 Am2kg-1, while maintaining the high coercivity, we prepare Ni50Mn50/Ni50Mn25X25/Ni50Mn50 multilayers simulating the shell-ferromagnet in two-dimensions. The initially incoherent interfaces are made to be coherent by annealing at 650 K in a magnetic field, just as in the case for preparing shell-ferromagnetic precipitates. The spins at the interface then become pinned leading to a two-dimensional magnet.

FH-2:IL05  In-memory Computing with Ferroelectric Germanium Telluride
C. RINALDI, L. Nessi, F. Fagiani, M. Cantoni, R. Bertacco, Politecnico di Milano, Milano, Italy; S. Varotto, M. Bibes, CNRS, Thales, Palaiseau, France; S. Cecchi, Paul-Drude-Institut für Festkörperelektronik, Berlin, Germany; R. Calarco, CNR-IMM, Roma, Italy; J. SławiNska, University of Groningen, Groningen, Netherlands; M. Buongiorno Nardelli, University of North Texas, Denton, TX, USA; S. Picozzi, CNR-SPIN, Chieti, Italy; P. Noël, J.-P. Attané, L. Vila, Université Grenoble Alpes, CEA, CNRS, Grenoble, France

Information and communication technology will use 20% of the world’s total electricity by 2030, which is simply unsustainable. To face this issue, Intel proposed a conceptual architecture allowing for ultra-low power electronics beyond-CMOS. The so-called magneto-electric spin-orbit (MESO) logic [1] comprises several stacks divided in two functional elements: a magnetoelectric unit to drive a ferromagnetic memory, and a read-out unit which converts the magnetic state into an electrical signal via spin-to-charge current conversion. Here we propose a ground-breaking alternative based on the ferroelectric Rashba semiconductor germanium telluride. The ferroelectric polarization of epitaxial thin films of GeTe can be reversed by electrical gating and used to switch the spin-to-charge current conversion [2]. Thus, information can be stored in the ferroelectric polarization of the material, while computing can be performed exploiting spin-to-charge conversion, within one single material compatible with silicon. The so-called ferroelectric spin-orbit (FESO) device paves the way to spin-based in-memory computing with ferroelectric Rashba semiconductors.
[1] S. Manipatruni, Nature 565, 35 (2019) [2] S. Varotto et al., Nature Electronics (2021). DOI: 10.1038/s41928-021-00653-2

FH-2:IL06  Resistive Switching in Ferroelectric Perovskites
I. Fina, Institut de Ciència de Materials de Barcelona (ICMAB-CSIC), Campus UAB, Bellaterra, Catalonia, Barcelona, Spain

Resistive switching consists in the switch of the resistance state by electric stimuli. If the device can show multiple intermediate resistance state, it is called a memristor. A memristor is the building block of neuromorphic computing and thus of high interest for applications. Several mechanisms can lead to resistive switching: thermochemical, valance change or electrochemical metallization. These mechanisms are related to ionic motion, and electrical current crosses the material. Thus, there is an important amount of energy consumed by the current injection and the associated Joule dissipation. In a metal/ferroelectric/metal structure, resistance can also change after electric stimuli, due to the change of the band diagram shape upon polarization switching. This electronic mechanism allows resistance change in open-circuit conditions (without current injection). Thus, the energy consumed and the dissipated power should be lower compared with ionic motion related mechanisms. However, ferroelectric perovskites also show ionic motion, thus it is of the highest relevance distinguish electronic and ionic processes in ferroelectric-based resistive switching devices, which will be the main topic of my talk.

FH-2:IL07  Hafnia-based Ferroelectric Devices: A Singular Type of Switching
P. Nukala1, 2, D. Carbone3, M. Ahmadi1, Y. Wei1, 4, S. Matzen5, B. Noheda1, 6, 1Zernike Institute for Advanced Materials, University of Groningen, Groningen, The Netherlands; 2Center for Nanoscience and Engineering, Indian Institute of Science, Bengaluru, India; 3MAX IV Laboratory, Lund University, Lund, Sweden; 4Nanoelectronic Devices Laboratory (NANOLAB), École Polytechnique Fédérale de Lausanne, EPFL, Lausanne, Switzerland; 5Center for Nanoscience and Nanotechnology, University Paris-Saclay, CNRS, Palaiseau, France; 6CogniGron Centre, University of Groningen, Groningen, The Netherlands

Hafnia-based films are changing the way we think of ferroelectric switching. Ferroelectricity in these materials arises from metastable phases stabilized at the smallest dimensions. In addition they present low leakage and CMOS compatibility 1], making them ideal candidates for memory and logic devices. Their switching takes place, quite uniquely, without involving domain wall motion, which allows experimental access to negative capacitance states[2]. Multiferroic tunnel junctions (MTJs) fabricated with (La,Sr)MnO3 electrodes and ferroelectric Hf0.5Zr0.5O2 barriers show both tunneling magnetoresistance (TMR) and tunneling electroresistance effect (TER), displaying four non-volatile resistance states[3]. Moreover, under electric field cycling, the TER effect can reach values as large as 10^6%. Experiments indicated that polarization switching alone cannot be responsible for those changes[4]. In this talk we will show direct evidence, by means of operando transmission electron microscopy and synchrotron x-ray diffraction experiments with in-situ application of electric fields[5], of the mechanisms that come into play during electric field switching in hafnia-based devices, as well as their relative importance determining the properties of these devices.

FH-2:L11  Ferroelectric Devices on CAAC-OSLSI
S. NUMATA, Y. Egi, F. Isaka, Y. Jimbo, T. Hamada, H. Baba, K. Ohshima, T. Murakawa, S. Tezuka, H. Kunitake, S. Yamazaki, Semiconductor Energy Laboratory, Co, Ltd., Atsugi-shi, Kanagawa, Japan

Recently, ferroelectric memories have attracted attention as emerging memories [1], and HfOx-based materials (e.g., HfZrOx[2]) and AlScN[3] are actively researched and developed. With a great demand for memory size reduction, a downscalable ferroelectric memory is desired. A thin ferroelectric reduces the switching field, but might reduce remanent polarization and memory window. However, it is unnecessary to make the ferroelectric too thin when a field-effect transistor (FET) with high breakdown voltage is used as a selector in a ferroelectric memory. We reported that LSI based on c-axis-aligned crystalline oxide semiconductor (CAAC-OS) maintains its high breakdown voltage even when being scaled [4], suggesting that the combination of CAAC-OSLSI and ferroelectrics is promising for reducing the size of ferroelectric memories. In this work, we fabricated ferroelectric devices over CAAC-OSLSI to reduce the total area of the memory device. Here, we report the fabrication process and the measurement results of CAAC-OSLSI performance and ferroelectric characteristics.

[1]T.S.Böscke et al., Applied Physics Letters 99.10 (2011). [2]T.Francois et al., IEDM 2019. [3]S.Fichtner et al., J. Appl. Phys. 125, 114103 (2019). [4]H.Kunitake et al., SSDM, pp.206-306, 2019.

FH-2:L12  Evaluation of Ferroelectric Memory using CAAC-OSLSI Technology
K. OHSHIMA, K. Furutani, T. Matsuzaki, K. Tsuda, S. Numata, F. Isaka, Y. Jimbo, H. Kunitake, T. Onuki, S. Yamazaki, Semiconductor Energy Laboratory Co., Ltd., Atsugi-shi, Kanagawa, Japan

It is urgent to develop new storage memory like a cross between SRAM, DRAM, and flash memory. Memory using ferroelectrics is especially promising new memory because it is nonvolatile and has advantages of lower power consumption and higher read/write speed than flash memory. In 2011, it was reported that an ultrathin film (10 nm) of hafnium oxide (HfO2) with impurities exhibits ferroelectricity[1]. Since then, the expectation for highly integrated HfO2-based ferroelectric memory has been growing. Potential ferroelectric memory is achieved by a combination of ferroelectrics with c-axis-aligned crystalline oxide semiconductor LSI (CAAC-OSLSI) technology[2]. The CAAC-OSLSI technology whose process temperature is 450C or lower can be employed in the back end of line of bulk silicon CMOS. A prototype of ferroelectric memory with CAAC-OSLSI technology was fabricated to demonstrate the memory operation. The prototype memory cell consists of one selector transistor and one ferroelectric device with our CAAC-OSLSI technology. Unlike DRAM, such a prototype memory need no refreshing and enables more than five layers of the memory cell to be stacked.
[1]T.S.Böscke et al., Applied Physics Letters 99.10 (2011). [2]H.Kunitake et al., SSDM, pp.206-306, 2019.

FH-3:IL01  Ferroelectric Memories for Neuromorphic Computing
S. Slesazeck, NaMLab gGmbH, Dresden, Germany

Ferroelectricity is already known for a long time, however its adoption in electronic circuits was confined to niche-applications so far due to the incompatibility of the perovskite materials to modern CMOS technology. The discovery of ferroelectricity in doped HfO2 that was firstly published in 2011 by Böschke et al. strongly increased the interest in ferroelectricity. The polarization reversal in ferroelectric HfO2 films can be adopted to store information in three distinct device classes. Depending on the material stack composition different devices can be constructed from the very same ferroelectric layer - ferroelectric capacitors (FeCAP), ferroelectric field effect transistors (FeFET) and ferroelectric tunnel junctions (FTJ). The electrical characteristics of this devices are strongly influenced by the whole material stack, rather than being dictated by the properties of the ferroelectric layer itself. In this talk the three different device concepts will be investigated. Based on the current state of the art in their development and electrical characteristics a ranking of their suitability for memory and for beyond-memory application e.g. in neuromorphic computation and AI will be discussed.

FH-3:IL04  Organic Spintronic Multilevel Resistive Switching Devices as Synapses for Neuromorphic Computing
A. Riminucci, CNR-ISMN, Bologna, Italy

Molecular spin valves use molecular materials, such as tris(8-hydroxyquinoline)aluminum (Alq3) to transport spin and charge between two ferromagnetic electrodes, that provide a spin polarized current. The hallmark of a spin valve is its magnetoresistance: its electrical resistance depends on the relative orientation of the magnetization of the two ferromagnetic electrodes. These devices also have a memristive behaviour: their electrical resistance can be changed, in a reversible and non-volatile fashion, by the application of voltage. Interestingly, there is an interplay between the memristive behavior and the magnetoresistance. The combination of memristive behavior with magnetoresistance makes them unique candidates as synapses for neuromorphic computing. The presence of magnetoresistance provides a tool to effect global, selective changes on the weight of synapses: all high weight (i.e. all high conductance) synapses are affected by the magnetic field, whose effect is that of increasing the weight further. Low weight synapses, on the contrary, are left untouched.

FH-3:IL07  Neuromorphic Computing with Redox-based Memristive Elements - Capacitive and Resistive Concepts
T. ZIEGLER, D.J. Wouters, R. Waser, JARA-FIT and IWE2, RWTH Aachen University, Aachen, Germany; S. Menzel, R. Waser, JARA-FIT and PGI-7/-10, Forschungszentrum Jülich GmbH, Jülich, Germany

Memristive elements have been proposed as artificial synapses in artificial neural networks (ANN) and a variety of these concepts are currently investigated. We will recapitulate our concept of Complementary Resistive Switches (CRS) and explore the use of these cells in the field of Neuromorphic Computing (NC) in an associative capacitive network and a resistive network. We will present concrete examples of computation-in-memory (CIM) concepts for binary multiply-accumulate (bMAC) operations. By exploiting the bitwise Boolean exclusive OR (XOR) operation, the Hamming distance (HD) can be computed in memory. This HD is encoded in the voltage drop of the common electrode, and from it, the result of a bMAC operation can be calculated. For the resistive network, a small demonstration is performed experimentally and the feasibility of the CIM concept is confirmed. The application as a potential hardware accelerator for the inference step of binary neural networks (bNN) is also investigated. To this end, a single-layer, fully connected neural network is trained on a binarized version of the MNIST dataset and the inference step of the test dataset is simulated.

FH-3:L08  Nonlinear Dynamics of Chua’s Circuits with Physical Memristor Model
F. CORINTO, Politecnico di Torino, Torino, Italy; M. Di Marco, M. Forti, R. Moretti, L. Pancioni, University of Siena, Siena, Italy

The paper investigates nonlinear dynamics and bifurcations in nonlinear oscillators with real non-volatile memristor devices. In particular, the focus is on a Chua’s circuit with a memristor described by the physics-based Stanford model. The main idea in the work is to show how memristor devices can be utilized as a programmable nonlinear resistor and then tune complex oscillatory and chaotic dynamics. With this in mind, two main operational modes are considered for the Memristor Chua’s Circuit (MCC): 1) Analogue transient phase: MCC is designed so that in the transient oscillations the voltage on the memristor is below threshold, hence the main memristor state variable, i.e., the gap g of the insulating material, is almost constant and the memristor behaves as a static nonlinear resistor. 2) Programming phase: the nonlinear characteristic of the memristor, which depends on g, can be changed via the application of voltages above threshold. The paper studies nonlinear oscillations of MCC in the transient phase for a fixed gap and also the bifurcations phenomena displayed by MCC when the gap is varied. Most importantly, the study makes clear the effect of physical parameters in influencing nonlinear dynamics of memristor-based oscillators.

FH-3:L09  Resistive Switching-based Invariant Manifold Tunable Chua's Circuit Design
M. ESCUDERO1, L. Pancioni2, S. Spiga1, S. Brivio1, 1CNR-IMM, Unit of Agrate Brianza, Agrate Brianza, Italy; 2Università degli Studi di Siena, Siena, Italy

Chaotic oscillator circuits are getting interest as part of bio-inspired analog computing systems. However, the lack of compact integration keeps these alternatives away from being attractive to adopt; new device technologies are key to provide better or diverse functionalities. In this direction, a design process for a nonvolatile resistive switching-based Chua circuit, the simplest chaotic circuit known, is presented. In this implementation, the nonlinear part of the circuit consists of the parallel connection of a negative impedance converter, enabling local negative resistance, and a Pt/HfO2/TiN device, providing nonlinear characteristics, tunable properties and an overall more compact circuit. Pt/HfO2/TiN device i-v characteristics in its non-switching regime are acquired and fitted into a model. In function of the device parameters, the design approach scales the impedances present on the Chua circuit while guaranteeing the oscillating circuit behavior. Measurements on device samples and on a built Chua circuit supports our work, and simulations of the designed circuit confirm that different invariant manifolds -including complex ones- are reproducible by using different device programmable states.
This work is partially supported by the project COSMO (Prot. 2017LSCR4K)

FH-3:L10  Storing and Retrieving Data in Mem-Processors
A. Ascoli1, R. Tetzlaff1, I. Messaris1, S.-M. Kang2, L.O. Chua3, 1Faculty of Electrical and Computer Engineering, Technische Universitat Dresden, Dresden, Germany; 2Jack Baskin School of Engineering, University of California Santa Cruz, Santa Cruz, CA, USA; 3Dept. of Electrical Engineering and Computer Sciences, University of California Berkeley, Berkeley, CA, USA

The Dynamic Route Map (DRM) is a powerful system-theoretic technique, lying at the basis of a systematic approach to design standard space-invariant Cellular Nonlinear Networks (CNNs) with first-order cells. However, the investigation of a memristive variant of these cellular arrays, in which each processing element is endowed with an additional dynamical state due to the non-volatile resistance switching memory, it accommodates, calls for the extension of the DRM graphical tool to second-order dynamical systems. The novel analysis method, named Second-Order Dynamic Route Map (DRM2), is employed to gain a thorough understanding of the mechanisms, by which a class of Memristor CNNs (M-CNNs) with second-order memristive cells store and retrieve information locally, within the same elements designated for data processing, providing an innovative exemplar of a non-von Neumann mem-processing architecture. Drawing a comprehensive picture of the nonlinear dynamics of these second-order mem-computing elements enables the development of a systematic procedure, based upon the DRM2 graphical tool, for the synthesis of M-CNNs, in which the time evolution of the cells’ dynamical states toward predefined equilibria lies at the basis of a universal mem-processing paradigm.

FH-3:L11  Autonomous Decision Making in Noisy Environment
D. MOLNAR, T.N. Török, L. Pósa, A. Halbritter, Department of Physics, Budapest University of Technology and Economics and MTA-BME Condensed Matter Research Group, Budapest, Hungary

We explore the possibility of utilizing the complex temporal dynamics and the highly voltage dependent switching speed of Ta2O5 resistive switching units for autonomous decision making tasks. Relying on this complex dynamical operation we demonstrate well-tunable short-term and long-term memory features facilitating the autonomous detection of small spikes in a highly noisy environment.

FH-3:IL12  From Resistive Switching in Mott Insulators to Mott Memory and Artificial Neuron
E. Janod, B. Corraze, J. Tranchant, L. Cario, M.-P. Besland, Nantes University, UMR CNRS, Institut des Matériaux Jean Rouxel (IMN), Nantes, France

The application of short electrical pulses on narrow gap Mott insulators induces a new phenomenon of resistive switching (RS). This insulator-to-metal transition (IMT) is volatile above threshold electric fields of a few kV/cm and becomes non-volatile and reversible (RS) for higher electric fields. This phenomenon is driven by the electric field, which triggers an electronic avalanche through injection of hot carriers. It induces the breakdown of the Mott insulating state at the nanoscale and creates the formation of granular conductive filaments. Such properties have been demonstrated in several Mott insulators, including oxides (V2O3:Cr) and chalcogenides (NiS2-xSex or AM4Q8 family (A=Ga, Ge; M=V, Nb, Ta, Mo; Q=S, Se)). Both volatile and non-volatile transitions can occur at room temperature and are therefore suitable for several applications. The non-volatile transition can be used for memory and memristor applications. Indeed, the (V1-xCrx)2O3 based devices realized by magnetron sputtering exhibit very competitive memory performances. Moreover, the volatile transition allows to implement the Leaky-Integrate-and-Fire (LIF) functionalities of an artificial neuron. Our results stand as basis for a new electronics based on the Electric Mott transition: Mottronics.

FH-3:IL13  VO2 Metal-insulator Devices for Energy Efficient Neuromorphic Computing
S. KARG, E. Corti, O. Maher, IBM Research, Zurich, Switzerland

Machine Learning is virtually penetrating every aspect of society with its ability to process natural language or images. However, what appears a simple task for the brain brings even the most powerful computer to its limit. The underlying von-Neumann architecture with its separated processor and memory turns out as a major bottleneck. Recently developed computing schemes allow to move specific tasks into the analog domain and process data inherently parallel and power efficient. The development of oxides with non-linear properties such as memristive or metal-insulator transition (MIT) materials offer new routes to brain-inspired data processing. We will demonstrate a novel computing paradigm based on oscillating neural networks (ONN). The MIT material VO2 provides the main building block for stable and efficient oscillators. Tunable memristors act as synaptic weights coupling the individual oscillators. We will cover fabrication of VO2 oscillators on silicon, device scaling to nm-dimensions in crossbar and planar configuration as well as characterization of electronic properties. The computational power of the phase-modulated ONN will be demonstrated with experiments and simulations. Key questions such as variability, compute performance and power consumption will be addressed.

FH-3:L14  Innovative V2O3:Cr Mott Insulator-based Resistive Memory
L. LABORIE, N. Castellani, G. Bourgeois, C. Castellana, T. Magis, E. Nowak, G. Molas, E. Jalaguier, CEA, LETI, Univ. Grenoble Alpes, Grenoble, France; G. Lefevre, S. David, C. Vallée, Univ. Grenoble Alpes, CNRS, CEA-LETI Minatec, LTM, Grenoble, France; J. Tranchant, B. Corraze, M.-P. Besland, E. Janod, L. Cario, Université de Nantes, CNRS, Institut des Matériaux Jean Rouxel, IMN, Nantes, France

Resistive Random Access Memories (ReRAMs) have already demonstrated their potential as a candidate for next generation non-volatile memories. Among ReRAM, Oxide Random Access Memories (OxRAMs) based on redox reactions and atoms diffusion are the most studied so far. However, one major drawback of this technology remains the variability of its resistance states, directly linked to their working principle. This work focuses on another type of ReRAM based on alternative materials, i.e. Mott Insulators. Those materials exhibit electric field-driven resistive transitions of electronic nature that do not involve atom diffusion [1]. We report for the first time integrated nano-sized memory cells made of V2O3:Cr Mott Insulators and compare their variabilities with the one of OxRAMs. In addition to having very good performances in terms of programming speed and consumption, they show lower high resistance state variability than conventional OxRAMs. This might be understood in the framework of SET and RESET mechanisms proposed in previous works and based on the formation and relaxation of a granular metallic filament made of compressed metallic domains [2].
[1] E. Janod et al., Adv. Funct. Mater., 2015 [2] M. Querré et al., Physica B: Condensed Matter, 2018

FH-3:L15  Silicon-integrated La2NiO4+d-Based Valence-change Memristive Devices with Neuromorphic Programming Capabilities
KHANH KHUU, C. Jimenez, M. Burriel, Univ. Grenoble Alpes, CNRS, Grenoble INP, LMGP, Grenoble, France; G. Lefèvre, A. Bsiesy, Univ. Grenoble Alpes, CNRS, CEA/LETI Minatec, LTM, Grenoble, France; S. Blonkowski, E. Jalaguier, Univ. Grenoble Alpes, CEA, LETI, Grenoble, France

Resistive switching (RS) devices have attracted much attention for their application as memories and neuromorphic computing systems. This work uses La2NiO4+δ (L2NO4), a mixed ionic electronic oxide, as memristive layer deposited on a Si-based substrate (Pt/TiO2/SiO2/Si), where Pt acts as bottom electrode, and with an oxidizable metal as top electrode. The L2NO4 films were grown by Metal-Organic Chemical Vapour Deposition, leading to highly dense and crystalline layers. The film growth and device fabrication were fully characterized by combining structural, microstructural and chemical techniques. In this work, an analogue bipolar RS behaviour with multiple resistance states is demonstrated in vertical TiN/L2NO4/Pt devices for the first time. Furthermore, the RS mechanisms were studied by operando X-ray absorption spectroscopy using synchrotron radiation, tracking the changes in the Ni-K edge position under polarization. In addition, gradual changes in conductance by the application of repetitive DC bias, which can be regarded as the evolution of the synaptic weight between neurons, were successfully obtained. These promising results open the door to the use of L2NO4-based memristors as artificial synapses for neuromorphic computing.

FH-3:IL16  Oscillator Computing based on Memristive Devices
C. LENK, L. SEEBER, S. DURSTEWITZ, S. PARK, T. IVANOV, M. Ziegler, Micro- and Nanoelectronic Systems, Electrical Engineering and Information Technology, Ilmenau University of Technology, Ilmenau, Germany

The human brain is able to process a myriad of information in a flexible and stable way and is therefore able to react adequately to continuously changing environmental influences. The processing heavily relies on nonlinear dynamics. This enables to integrate the multitude of information in an enormous and massively parallel network of neurons that are divided into functionally specialized regions such as the visual cortex, auditory cortex, or dorsolateral prefrontal cortex. Each of these regions participates as a context-dependent, self-organized, and transient subnetwork. Even if the underlying mechanisms are only partially understood, the interaction between dynamics and topology has been identified as one of the essential building blocks of information processing in the brain in recent years. In this context, memristive devices integrated in oscillator networks offer a new degree of freedom for the investigation of this concept: a local memory that supports transient connectivity patterns in oscillator networks. This talk will give an overview of this topic by presenting different examples. We show how resistance changes of memristive devices affect the dynamics of networks, but also how network dynamics influence the network connectivity. Furthermore, important requirements for memristive devices will be discussed and it will be shown how a new way of information processing beyond current approaches can be enabled to opens a new pathway toward the construction of cognitive electronics.

FH-3:L17  Implementation of an Electrical Set-up to enable RRAM-based Neural Network Operations
E. Perez-Bosch Quesada, E. Perez, M. Kalishettyhalli Mahadevaiah, C.A. Chavarin, C. Wenger, IHP-Leibniz-Institut für innovative Mikroelektronik, Frankfurt (Oder), Germany; F. Zahari, H. Kohlstedt, Nanoelectronic, Faculty of Engineering, Kiel University, Kiel, Germany; M. Ziegler, Department of Micro- and Nanoelectronic Systems, TU Ilmenau, Ilmenau, Germany; C. Wenger, BTU Cottbus-Senftenberg, Cottbus, Germany

The RRAM technology is in a maturity level that leads to its integration in CMOS-compatible NVM arrays. Moreover, the use of this technology in the implementation of hardware-based neural networks has gained momentum. Due to their resistive switching properties, memristors have been considered a suitable candidate to mimic the synaptic activity of the biological learning process. Therefore, synaptic plasticity can be emulated by memristive elements integrated within RRAM arrays. In order to move the devices in the different resistive states, several AC and DC signals are involved in their program and read-out operations. Thus, it is an accurate and coordinated process to generate and apply the electrical signals, achieving a correct behavior of the RRAM devices within a hardware-based artificial neural network. In this work, an electrical set-up is implemented in order to access and program the HfO2-based RRAM devices integrated in CMOS-compatible 4 kbit 1T-1R memory arrays. Moreover, an 8x8 VMM hardware accelerator has been developed and assessed. Operated by a virtual instrument implemented in LabVIEW, the set-up based on a Keithley 4200-SCS and an Arduino Mega 2560, is exposed and analyzed in order to show its functionality in terms of RRAM-based neural networks control.

FH-3:IL18  Nanowire based Memristive Devices
G. Milano, C. Ricciardi, Advanced Materials Metrology and Life Science Division, INRiM (Istituto Nazionale di Ricerca Metrologica), Torino, Italy; Department of Applied Science and Technology, Politecnico di Torino, Torino, Italy

In recent years, the interest of the research community for nanowire based memristive devices is becoming more and more relevant [1,2]. On one hand, single nanowires were reported to exhibit all-in one memristive and neuromorphic functionalities, thanks to their unique ability of uncoupling ionic and electronic contributions in resistive switching phenomena [3]. On the other hand, self-organizing nanowire networks very recently showed that their structural plasticity through reconnection and rewiring [4], coupled with nonlinear dynamics and fading memory properties, can be fruitfully exploited for the implementation of brain-inspired computing paradigms, such as reservoir-computing [5]. This talk will review the principal characteristics of nanowire-based memristive and neuromorphic devices, in the perspective of in materia physical computing.
[1] Milano G. et al., Adv. Elect. Mater., https://doi.org/10.1002/aelm.201800909 (2019) [2] Kuncic Z. et al., Adv. Phys-X, https://doi.org/10.1080/23746149.2021.1894234 (2021) [3] Milano G. et al., Nat. Commun., https://doi.org/10.1038/s41467-018-07330-7 (2018) [4] Milano G. et al., Adv. Intell. Syst. https://doi.org/10.1002/aisy.202000096 (2020) [5] Milano G. et al., Nat. Mater., https://doi.org/10.1038/s41563-021-01099-9 (2021)

FH-3:L19  A Graph Theory-based Approach for Memristive Nanowire Networks
G. MILANO1, K. Montano2, E. Miranda3, C. Ricciardi2, 1Advanced Materials Metrology and Life Science Division, INRiM (Istituto Nazionale di Ricerca Metrologica), Torino, Italy; 2Department of Applied Science and Technology, Politecnico di Torino, Torino, Italy; 3Departament d’Enginyeria Electrònica, Universitat Autònoma de Barcelona (UAB), Cerdanyola del Valles, Spain

Self-organized memristive nanowire (NW) networks represent promising platforms for the implementation of unconventional computing paradigms. Contrary to conventional memristive architectures based on crossbar arrays where each element of the network is individually addressed, these networks are complex systems where an emergent behavior arises from collective synergy without the need of fine tuning of network elements. Here, we report on a graph theory-based approach for analyzing the interplay between topology and functionality of NW memristive networks. Besides the strong connection between graph metrics and geometrical aspects, the onset of percolation was observed in correspondance of the appearance of a giant component spanning the graph as predicted by percolation theory. In agreement with experimental observations, the emergent behavior of the graph model is characterized by short-term memory, heterosynaptic plasticity and by the emergence of a self-selected conductive pathway connecting stimulated areas. Results show that a graph theory-based approach can be exploited for modelling nanonetworks and for exploring the implementation of unconventional computing paradigms within the realm of biologically inspired neuromorphic architectures.

FH-3:L20  Long-term Probing of Memristive Ag-based Nanostructures via an Unconventional cAFM Approach
N. CARSTEN, R. Gupta, T. Strunskus, F. Faupel, A. Vahl, Institute for Materials Science, Chair for Multicomponent Materials, Kiel University, Kiel, Germany; A. Hassanien, Department of Condensed Matter Physics, J. Stefan Institut, Ljubljana, Slovenia

Biological neural systems easily outperform modern computation architecture in a broad range of problems like e.g. pattern-recognition through shaping of adaptive and self-organized networks. Recent approaches from neuromorphic engineering aim to create similarly organized networks of memristive switches based on fundamental memristive building units like metallic nanoparticles or -wires, which ultimately shall provide a physical basis for concepts like reservoir computing. The design and understanding of emergent network dynamics rely on experimental insights into the corresponding fundamental switching units. Here, we utilize an unconventional cAFM approach to probe the long-term memristive dynamics of individual Ag-filaments evolved from a continuous active electrode and of nanoscale statistical assemblies of AgPt nanoparticles. The long-term response from both systems are representative for the local switching action in larger networks of memristive switches and are evaluated with respect to short-term memory, temporal correlation and inhibitory dynamics. Such insights, representing the local level within networks of memristive switches, are key requirements to understand the emergent network behavior and progress in technical device implementation.

FH-3:L21  Resistive Switching in Nanoparticle-based Systems: From Diffusive Switching towards Collective Dynamics
N. Carstens1, B. Adejube1, M. Noll2, T. Birkoben2, M. Mirigliano3, R. Gupta1, T. Strunskus1, F. Faupel1, P. Milani3, H. Kohlstedt2, A. Vahl1, 1Institute for Materials Science - Chair for Multicomponent Materials, Faculty of Engineering, Kiel University, Kiel, Germany; 2Institute for Electrical Engineering and Information Engineering - Chair for Nanoelectronics, Faculty of Engineering, Kiel University, Kiel, Germany; 3Dipartimento di Fisica & Cimaina, Universita degli Studi di Milano, Milano, Italy

Nanoparticles (NPs), due to their high surface-to-volume ratio, small size and high number of intrinsic defects, offer the potential to show unique properties. Further, NP assemblies differ in many aspects from their atom-assembled counterparts. Owing to the capability of electric field confinement and hence localized memristive action, NPs are promising building units for resistive switching devices and electronic devices with neuron-inspired functionalities. Here, diffusive memristive switching as well as collective critical dynamics are discussed to showcase the application potential of NPs. Metal-insulator-metal structures with sparsely embedded noble metal alloy NPs inside a dielectric matrix show diffusive memristive switching with distinct, well-separable resistance states. In contrast, dynamic resistance changes are observed in a broad variety of highly interconnected NP networks from various material systems and are described in terms of criticality. Resistance change avalanches as well as inter-event intervals follow power laws, whose exponents have been successfully tailored upon addition of a ceramic capping layer. Upon coupling of oscillators with multi-terminal NP network devices the translation between power-law dynamics and spiking output is demonstrated.

FH-3:L22  Correlation between the Threshold Kinetics and Relaxation Behavior of Ag/HfO2-based Diffusive Memristors
S.A. CHEKOL, S. Menzel, R. Waser, S. Hoffmann-Eifert; W.R. Ahmad, R. Waser, JARA-FIT and Peter Gruenberg Institute (PGI 7 & 10), Forschungszentrum Juelich GmbH, Juelich, Germany; JARA-FIT and Institute of Materials in Electrical Engineering and Information Technology II, RWTH Aachen University, Aachen, Germany

Diffusive memristors are attractive for memory and neuromorphic applications due to their thresholding and self-relaxation behavior. The set and relaxation times of these devices can be modulated over several orders of magnitude depending on the material system, programming condition, and external circuitry. However, implementation into emerging concepts such as temporal coding requires a physical understanding of the set and relaxation dynamics. Here, we investigate the switching dynamics of Ag/HfO2/Pt volatile electrochemical metallization (v-ECM) cells. Depending on the amplitude of applied voltage pulses, different mechanisms are identified as the rate-limiting steps for filament formation. Consequently, the set kinetics is simulated based on a modified physical model introduced for the SET behavior of non-volatile ECMs earlier. With this, the relaxation behavior is analyzed depending on the programming conditions such as voltage amplitude and pulse width. The combination of the threshold and relaxation kinetics study enabled us to achieve a deeper understanding of the significance of the filament formation and growth process on its relaxation time. This knowledge can be directly transferred into the optimization of the operation of v-ECM devices in neuromorphic circuits.

FH-3:L23  Short Time Dynamics in SiO2 and HfO2-based Threshold Switching Devices
M. DUTTA, S. Brivio, M. Alia, S. Spiga, CNR-IMM, Unit of Agrate Brianza, Agrate Brianza, Italy

Spiking neural networks are especially suited for processing of temporal information. They employ synapses and neurons with short-term dynamical behaviour. In state-of-the-art neuromorphic chips emulating SNNs the short-term dynamics is implemented by capacitors. The capacitor footprint does not allow extending the short-term dynamics to biological times. Resistive memory devices with oxide switching layer and Ag or Cu as the active electrode, can show the threshold switching (TS) behavior, i.e. voltage-driven temporary activation of a conductive state. The retention time of such conductive state is supposedly programmable up to few ms and s range. Here we demonstrate the TS characteristics and the tunable retention in various time-scales by engineering the Pt/SiO2/Ag and TiN/HfO2/Ag stacks. The devices show TS in the range of 100 nA to 1 mA and long endurance for more than 1000 cycles. Moreover, we present data on the device retention time as a function of the programming conditions for exploring the retention time dynamics.
This research is supported by H2020 EU project MeM-Scales- (grant agreement No. 871371) [1] “Memristive devices for brain inspired computing”, S. Spiga, A. Sebastian, D. Querlioz, B. Rajendran, Editors, 1st ed., Woodhead Publishing, 2020.

FH-3:L24  Voltage-time Dilemma of Current-driven Silver Single-atom Resistive Switches
A. NYARY, B. Sánta, A. Halbritter, Budapest University of Technology and Economics, MTA-BME Condensed Matter Research Group, Budapest, Hungary

Common resistive switching devices mostly rely on electrochemical metallization cells. However, as the active volume approaches the atomic size-limit, resistive switching due to the current-induced rearrangement of the central atoms contributes more and more, which demonstrates solely pure Ag single-atom contacts suffice for resistive switching [1]. Current-driven atomic switches have also been reported to be capable of trainable, extremely stable memory operation [2]. The goal of our work is the detailed characterization of a single atom memory. Common silver-based memristors exhibit the so-called voltage–time dilemma [3,4]. Here, we present the thorough study of the frequency dependence of the threshold voltage and the multilevel programming possibilities of the current-induced atomic switching of Ag-based current-induced single-atom resistive switching using mechanically controllable break junction (MCBJ) devices. Additionally, we present the theoretical approaches aiming to describe the frequency dependence of the threshold voltage in current-induced single-atom resistive switches.
[1] Geresdi et al., Nanoscale 3, 1504, 2011 [2] Schirm et al., Nat. Nanotechnol. 8, 645, 2013 [3] Gubicza et al., Nanoscale 7, 4394, 2015 [4] Sánta et al., Beilstein J. Nanotechnol. 11, 92, 2020

FH-3:IL25  Physic Based Simulation of ReRAM for Neural Network Applications
S. Menzel, Forschungszentrum Jülich, Peter Grünberg Institut (PGI-7),  Jülich, Germany

Redox-based resistive switching devices (ReRAM) based on the valence change mechanism (VCM) and the electrochemical mechanism (ECM) have attracted great attention due to their potential use in neuromorphic applications due to the possibility to program many different resistance levels in one device. In general, filamentary-switching VCM devices show an abrupt SET transition from a high resistive state (HRS) to a low resistive state (LRS), but a rather gradual RESET transition. For neuromorphic applications, however, an analog switching for RESET and SET is favored. Based on a physics-based simulation model, it will be shown that analog SET and RESET can be obtained in any filamentary switching VCM cell. This theoretical analysis is validated by experimental data on HfO2-based VCM devices. Moreover, the gradual nature of the set transition will be demonstrated on a sub-100 picosecond regime using dedicated coplanar wave guide devices. Scaled ECM and VCM devices may show quantized conduction steps. This can lead to severe issues when using backpropagation during learning. In this talk, a 1-layer neural network simulation using a quantized ECM model is presented. Quantized steps are also observed while reading a VCM cell.The impact of read noise on MAC operations will be discussed.

FH-3:L27  Impact of Memristive Switching Dynamics on Spiking Neural Network Operation
S. Brivio, S. Spiga, CNR-IMM, Unit of Agrate Brianza, Agrate Brianza, Italy; D.R.B. Ly, E. Vianello, Université Grenoble Alpes, Leti, Grenoble, France

Hardware implementations of Spiking Neural Networks (SNNs) are a possible solution for real-time information extraction in edge environments. Online training of hardware SNNs require a re-thinking of the conventional computing architecture and technology. Memristive devices are a promising solution to build electronics synapses modulating the connection between neurons over a continuum of conductance states. The number of conductance steps and their distribution over the entire range affects the SNN performances, but a general and quantitative analysis on technologically plausible memristive SNNs is still lacking. In our research, we perform both development of memristive devices and analysis of SNNs operation including their synaptic functions. In this talk, on one side, we will present the latest results on the characterization of the switching dynamics of filamentary metal-oxide memristors. On the other side, we will present the analysis of the operation of SNNs simulated according to CMOS-plausible circuits and realistic memristive models,[1] as a first step towards the combined development of networks and memristive devices.
This work is partially supported by the project Horizon 2020 EU MeM-Scales (grant No. 871371). [1] Brivio et al. Front. Neurosci. 15 580909 (2021)

FH-3:L28  Modelling the Switching of RRAM Devices for Bio-inspired Computing
F. VACCARO, S. Brivio, S. Spiga, CNR-IMM, Unit of Agrate Brianza, Agrate Brianza, Italy; S. Perotto, Dipartimento di Matematica, Politecnico di Milano, Milano, Italy; A.G. Mauri

RRAM devices have been recently identified as the most promising candidates in diverse alternative approaches to computation, e.g., in-memory and bio-inspired computing. In bio-inspired schemes, the switching dynamics and the stochastic properties of RRAMs are involved in the functionality of the overall computing system. In this context, it is therefore crucial to model the device behavior during the switching, with its inherent variability. In this work, the variable switching behavior of TiN/HfO2/Ti/TiN RRAMs has been extracted from an experimental characterization and reproduced by a physics-based compact model. The model consists of a system of ordinary differential equations describing the evolution of the ion concentration and the temperature in the oxide. The model succeeds in replicating the experimental behavior of the devices over different time scales of voltage application (in both dynamic and quasi-static conditions), and for different operational modes (for analog and digital transitions). In particular, the simulations allow us to correctly reproduce the dynamics in presence of variability, the difference between abrupt SET and gradual RESET in the I-V hysteresis loop, the memory window in weak and strong regimes, and the kinetics over several orders of magnitude.

FH-3:L30  1/f Noise Spectroscopy and Noise Tailoring of Resistive Switching Devices
A. HALBRITTER, Z. Balogh, B. Sánta, L. Pósa, T.N. Török, D. Molnár, Department of Physics and MTA-BME Condensed Matter Research Group, Budapest University of Technology and Economics, Budapest, Hungary; M. Csontos, Transport at Nanoscale Interfaces Laboratory, Empa, Swiss Federal Laboratories for Materials Science and Technology, Dübendorf, Switzerland

In resistive switching memories the functionality is confined to a tiny volume, where local atomic fluctuations play a crucial role. The resistance, voltage and frequency dependent 1/f-type noise characteristics serve as rich, device specific fingerprints of the relevant transport and noise-generating mechanisms in such structures [1]. We demonstrate a comparative noise analysis of transition metal oxide and silver based resistive switching filaments as well as phase change memory systems, demonstrating orders of magnitude material specific differences in the overall noise floors, and orders of magnitude tunability of the noise levels along the analog tuning of the resistance states. This behavior is analyzed in the framework of a point-contact noise model highlighting the possibility for the disorder-induced suppression of the remote fluctuators’ contribution [2]. These findings promote the design of multipurpose resistive switching units, which may serve as analog-tunable noise sources in probabilistic computing machines [1,2].
[1] Z. Balogh et al. 1/f noise spectroscopy and noise tailoring of nanoelectronic devices (topical review), Nano Futures 5, 042002 (2021) [2] B. Sánta et al. Noise Tailoring in Memristive Filaments, ACS Applied Materials and Interfaces 13, 7453 (2021)


Cimtec 2022

Copyright © Techna Group S.r.l.
C.F.-P.I. 03368230409
Privacy Policy - Cookie Policy - Software Commercio Elettronico by Pianetaitalia.com