Track H
Avances in Materials and Devices Research for Digital, Neuromorphic and Unconventional Computing

ABSTRACTS

Session H-1.1 Memristive materials and devices for brain inspired computing

H-1.1:IL01 Memristive Devices for Bio-inspired Information Pathways
A. Linkenheil¹, Z. Geng¹, K. Nikiruy¹, B. Spetzler¹, J. Schneegaß², T. Ivanov^{1, 2}, F. Schwierz¹, M. Ziegler^1, 2, ¹Micro- and Nanoelectronic Systems, Dept of Electrical Eng. and Information Technology, TU Ilmenau, Germany; ²Institute of Micro- and Nanotechnologies MacroNano®, TU Ilmenau, Germany

Memristive devices have attracted considerable attention in the electronic device community due to their inherent memory effect, which allows them to mimic the function of biological synapses. As such, they are a key building block for realizing artificial neural computing schemes in hardware, so-called neuromorphic systems. However, there are various neuromorphic architectures with different requirements for memristive devices, requiring optimization of electrical properties and materials as well as the technological framework for each specific application. This talk will discuss the challenges and prospects of memristive devices for neuromorphic computing in general and several selected examples. Valence-change-based memristive devices are presented, and it will be discussed how their properties can be tailored by systematic design variations for applications in neuromorphic computing architectures. Furthermore, it is shown how the resistance change of memristive devices affects the dynamics of networks and how network dynamics influence network connectivity. Important requirements for memristive devices will be discussed, and it will be shown how a new way of information processing beyond current approaches can open a new bio-inspired pathway toward the construction of cognitive electronics.
This work was partially funded by the Carl-Zeiss Foundation via the Project MemWerk and the German Research Foundation (DFG) through the Collaborative Research Centre CRC 1461 "Neurotronics – Bio-Inspired Information Pathway".

H-1.1:IL02 Oxide Materials for Artificial Neurons
M. Salverda, M. van den Broek, R. Hamming-Green, P. Nukala, B. NohedA, University of Groningen, Zernike Institute for Advanced Materials, Groningen, Netherlands

Compared to the relatively large number of materials and device concepts that are reported to display synapse-like behaviour, compact neuron-like devices have been demonstrated in a limited number of materials. Most often, these are simple transition metal oxides that undergo a metal-insulator transition, giving rise to a current-controlled negative differential resistance (NDR) regime, by which self-oscillatory behaviour can be attained. However, these materials do not offer enough endurance, due to the different oxidation states that take place upon local heating. We show current-controlled negative differential resistance and self-oscillatory behaviour in thin films of the perovskite SmNiO3, NdNiO3 and TbMnO3. While the nickelates undergo a metal-insulator transition, that is not the case for TbMnO3, for which the NDR regime is accessible thanks to the promotion of electrons across the band gap of this semiconducting material. This opens possibilities to create compact artificial neurons (devices that emulate the firing of action potentials) using materials that are robust upon intense Joule heating and can endure a large number of cycles.

H-1.1:IL03 Leveraging Ferroelectric Technologies for Neuromorphic Computing
E. COVI, University of Groningen, Zernike institute for Advanced Materials and Groningen Cognitive Systems and Materials Center (CogniGron), Groningen, Netherlands

Brain-inspired architectures, particularly Spiking Neural Networks (SNNs), have emerged as promising models for achieving low-latency computation, simultaneous internal-state storage, and energy-efficient operation. The integration of ferroelectric technologies holds significant promise for the development of neuromorphic hardware. Ferroelectric devices such as Ferroelectric Capacitors and Ferroelectric Tunnel Junctions offer fast and energy-efficient tuneable volatile and non-volatile storage, making them well-suited for storing SNN parameters. Their unique properties enable efficient computation, neural dynamics, and synaptic plasticity, essential for emulating the brain's functionality in hardware. This talk sheds light on the challenges and opportunities associated with achieving a neuromorphic-ferroelectric hardware system, offering insights into their integration challenges and optimization perspectives. We discuss the pivotal role of ferroelectric technologies in supporting the storage and computation requirements of SNNs, emphasizing their ability to meet the power efficiency, response time, and online learning demands of edge systems.

H-1.1:IL04 Materials Design and Defect Engineering towards Quantum Conductance and Neuromorphics in Memristive Devices
L. ALFF, Materials Science, Technische Universität Darmstadt, Darmstadt, Germany

Here we show how multidimensional defects and their interactions govern memristive device behaviour. A key effect in oxides is that oxygen vacancies enhance electric conductivity in the conducting filament and thermal conductivity in the surrounding dielectric matrix. Furthermore, the controlled induction of grain boundaries has a dramatic effect on the forming voltages showing that complex defect interactions have to be taken into account for a full understanding of memristive devices [1]. We investigate the occurrence of substoichiometric phases as a result of the oxygen dynamics [2,3]. We show how defect engineering can be applied to implement analogue depression and potentiation of the resistance, and how the results serve as guideline for smart materials selection. We suggest modified HfO2 and Y2O3 as most suited neuromorphic /multi resistive state materials due to their specific materials properties [4]. In addition, we show how artificial intelligence can be applied to detect correlations between materials properties and device behaviour in huge data sets [5].
[1] Adv. Sci. 9, 2201806 (2022).
[2] ACS Appl. Electron. Mater. 5, 754 (2023).
[3] ACS Appl. Mater. Interfaces 14, 1290 (2022).
[4] Adv. Electron. Mater. 6, 2000439 (2020).
[5] Micromachines 13, 2002 (2022).

H-1.1:L05 Effect of the La2NiO4+s Deposition Temperature on the Memristive Properties of the TiN/La2NiO4+s/Pt Devices
A. KOROLEVA^{1, 2}, N.A. Nguyen^{1, 2}, C. Ternon², M.a Burriel², E.-I. Vatajelu¹, ¹Université Grenoble Alpes, CNRS, Grenoble INP, TIMA, Grenoble, France; ²Université Grenoble Alpes, CNRS, Grenoble INP, LMGP, Grenoble, France

Resistive switching (RS) devices are considered amongst the best candidates for the emulation of synaptic connections in neuromorphic architectures. Recently, promising analog devices were built based on the La2NiO4+δ (L2NO4) films, which exhibit oxygen storage capability along with high oxygen mobility [1], [2]. However, the L2NO4 deposition temperatures of 600-650°C used in the previous studies are not compatible with the BEOL fabrication process, so the decrease of L2NO4 deposition temperature is very desirable. In this work, we explore the effect of the L2NO4 deposition temperature on the RS behavior of the TiN/L2NO4/Pt devices. We show that at the BEOL limit temperature of 450°C, the device exhibits high resistivity, which gives rise to the low-current self-rectifying RS. At the same time, at the temperature of 550-570°C, the RS behavior is similar to the 600°C L2NO4-based devices but a decrease of operation current is observed, which allows reducing power consumption. Therefore, as shown in this study, the fine control of the deposition temperature is crucial for L2NO4-based device engineering.
[1] Maas et al. 2020 J Mater Chem C Mater
[2] Khuu et al. 2022 Adv Mater Technol

Session H-1.2 Phase change materials and applications

H-1.2:IL01 Phase Change Materials for Rreliable Flexible Memories
S. CALVI¹, M. Bertelli², S. De Simone², F. Maita², F. De Matteis³, S. Prili^{1, 2}, F. Righi Riva¹, V. Mussi², A. Diaz Fattorini¹, F. Arciprete^{1, 2}, M. Longo^{2, 4}, R. Calarco², ¹Department of Physics University of Rome Tor Vergata, Rome, Italy; ²Institute for Microelectronics and Microsystems (CNR-IMM), Rome, Italy; ³Department of Industrial Engineering University of Rome Tor Vergata, Rome, Italy; ⁴Department of Chemistry University of Rome Tor Vergata, Rome, Italy

Memories based on phase-change materials (PCMs) have the potential to be a solution to overcome the Von Neumann architecture, by enabling the development of a non-volatile memory and computational unit integrated in a single device. PCM-based memories can ensure low power consumptions, fast operations and high endurance. Remarkably, they can be fabricated with low cost technologies on flexible and recyclable substrates, towards the development of green electronics and circular economy. The possibility of implementing memories directly on a flexible substrate would allow an easy integration with photovoltaic harvesting and conformable sensors, already demonstrated on such substrates. In fact, this integration would improve scalability and portability of electronics. However, currently the potential of PCM materials in flexible devices is far from being fully exploited, as their behavior on plastic substrates is still not clear. The specific issues that PCM-based devices are expected to overcome in large-area flexible electronics applications have not been thoroughly investigated and their behavior is yet to be understood. The proposed work is intended to lay the groundwork for the development of low-cost and large-area compatible processes for high performance PCM-based memories.

H-1.2:L03 The Influence of Sb/Te Ratio on the Crystallization Kinetics of GeSbTe Alloys
O. DAOUDI¹, E. Nolot¹, F. Fillot¹, J. Li¹, M. Bernard¹, N. Bernier¹, V.-H. Le¹, H. Renevier², G. Navarro¹, ¹Univ. Grenoble Alpes, CEA, Leti, Grenoble, France; ²Univ. Grenoble Alpes, Grenoble-INP, LMGP, Grenoble, France

Phase-Change Memory (PCM) has proven to meet the demands of different applications such as Storage Class and automotive market. This was possible in particular thanks to the engineering of the stoichiometry of GeSbTe materials. The alloys based on a stoichiometry close to the GeTe-Sb2Te3 tie line of the ternary diagram, show a “nucleation dominated” crystallization with multiple phase crystallization steps, while Sb-rich alloys show a single-phase “growth dominated” crystallization [1]. In this study, we investigate the transition from the amorphous to crystalline states in several materials based on Ge, Sb and Te, varying the Sb/Te ratio from lower to higher than one. We provide insights into the dependency of the atomic vibrations on Sb/Te ratio in GeSbTe system. To achieve this, we combine Raman spectroscopy measurements and ab initio simulations. This study is supported with the measurements of the lattice parameters of the crystalline structures by X-ray diffraction. Furthermore, we follow the amorphous to crystalline transition of each material, by performing resistivity measurements as a function of temperature. In summary, this work highlights the key role of Sb/Te ratio into the crystallization kinetics of GeSbTe system.
[1]G. D’Arrigo et al., J.MSSP, v.65, 100-107, 2017

H-1.2:L04 Valence Transition in SmTe Films Enabling Non-volatile Resistive Change without Structural Transition
Shogo HATAYAMA¹, S. MORI², Y. SAITO¹, P. FONS³, Y. SHUANG², Y. SUTOU², ¹National Institute of Advanced Industrial Science and Technology (AIST), Tsukuba, Japan; ²Tohoku University, Japan; ³Keio University, Japan

Phase change materials (PCMs), which show a non-volatile resistive change between high-resistive amorphous and low-resistive crystalline phases, are used as the recording layer of emerging non-volatile memory. In the write/erase processes, PCMs are heated above the transition temperature, such as crystallization temperature and melting point, by Joule heating caused by an electrical pulse. In these processes, the drastic atomic rearrangement upon phase transition is required with a large density change, leading to the delamination of the PCM layer at surrounding interfaces, which is the plausible origin of failure operation. To improve this concern for PCMs, our group has developed SmTe films showing non-volatile resistive change by valence transition (VT). The SmTe film was found to show the resistive switching caused solely by the valence state change. Although a resistivity contrast with 5 orders of magnitude is obtained upon VT, the crystal structure remains almost intact before and after the transition. Therefore, VT is expected as a new principle for non-volatile recording. The resistive change behavior in thin films as well as devices will be presented.

Session H-2 Advances in memory and memristive devices: devices, mechanism, and applications for computing

H-2:IL01 Architectures and Materials for Storage Class Memories
P. FANTINI, Micron Semiconductor, Vimercate, Italy

Semiconductor technology definitively entered in the 3D era to overcome some physical and technological limits associated with the planar scaling process that continued to bring amazing and surprising results. This 2D to 3D inflection is also called 3D Manhattan solution, based on the analogy with the urban architecture that discovered the unexplored space of the vertical dimension introducing the skyscrapers in many big cities, and it is bringing a lot of challenges and opportunities to develop new memory technologies with the promise to continue to follow the scaling roadmap dictated by the Moore’s law. Aim of this talk is to describe the interlaced challenges among novel architectures, the exploration of new materials and deposition techniques to fulfil the general memory market trend and to open new opportunities in the storage class memory area that 5G and AI applications can unleash.

H-2:IL02 Integration Aspects of Hafnium Oxide-based Memristive Devices
E. Perez-Bosch Quesada, A. Baroni, E. Perez, K. Dorai Swamy Reddy, Ch. Wenger, IHP - Leibniz Institute for High Performance Microelectronics

Due to its advantages of massive parallelism, high energy efficiency, and cognitive functions, brain-inspired neuromorphic computing is attracting immense interest. As the basic unit cell for learning algorithms, the implementation of synaptic behavior into memristive devices is the key step toward neuromorphic computing. Recent advances in the performance of resistive random access memory (RRAM) acting as memristive devices have led to a significant interest in neuromorphic applications. Although RRAM based memory arrays demonstrated excellent performance parameters, the variability is still a critical issue. Controlling this intrinsic phenomenon requires employing integration aspects and program-verify schemes. In this talk, an optimized scheme to minimize resistance state dispersion and to achieve reliable multi-bit operation is evaluated.

H-2:IL03 CMOS Compatible Materials and Devices for beyond von Neumann
V. BRAGAGLIA, IBM Research Europe, Zurich, Switzerland

Brain inspired computing is a promising paradigm of artificial intelligence (AI) systems that aims at developing an efficient computing architecture that resembles the biological brains. The development of novel materials and devices with neural and synaptic functions incorporated into unique architectures will allow the implementation of a computing system that can efficiently perform the heavy vector - matrix manipulation inherent to AI workloads with O(1) time complexity[1]. Memristors are key building blocks for the realization of the artificial neural and synaptic function in neuromorphic computing. [2] These devices rely on diverse physical mechanisms and materials and their understanding via experimental and theoretical means is pivotal to the device optimization and coupling to the higher layers of the computer architecture. We will see examples of how material and device engineering can lead to breakthroughs in device performance. Nonetheless, device variability among other non-idealities still hinders the hardware scaling to large networks required to solve more complex AI tasks. Challenges at device and hardware level may also be overcome through a complementary research effort to develop more robust, hardware-friendly algorithms and computational models that could compensate for the variability issues of devices[3].
[1] A. Sebastian et al., "Memory devices and applications for in-memory computing", Nature Nanotechnology, 15, 529-544, 2020.
[2] From: https://spectrum.ieee.org/analog-ai, 2021
[3] N. Gong, et al., “Deep learning acceleration in 14nm CMOS compatible ReRAM array: device, material and algorithm co-optimization” IEEE IEDM 33.7. 1-33.7. 4, 2022

H-2:L04 Resistive Memory Window Enhanced through Bandgap Tuning in V-substituted Cr2O3 Thin Films
J. TRANCHANT, M. Rodriguez Fano, M. Haydoura, B. Corraze, E. Janod, M.-P. Besland, L. Cario, CNRS, Institut des Matériaux de Nantes Jean Rouxel, (IMN), Nantes, France

Mott memories are promising to overcome the limitations of other non-volatile technologies. However, resistive switching (RS) has been demonstrated so far only in narrow gap Mott insulators such as (V0.95Cr0.05)2O3 with a tight memory window. In this study, we have investigated the whole range of composition of (Cr1-xVx)2O3 compounds (0

H-2:L05 Exploring, Tailoring, and Harnessing Electrical Noise in Resistive Switching Memories
Z. BALOGH, A. Nyáry, B. Sánta, J.G. Fehérvári, S.W. Schmid, L. Pósa, A. Halbritter, Department of Physics, Institute of Physics, Budapest University of Technology and Economics, Budapest, Hungary

Exploring noise phenomena and understanding the underlying physics is of great importance for the development of novel computing devices. This is particularly true for memristive systems, where the very small active volumes result in highly system-specific noise characteristics. Therefore, comprehensive noise analysis, i.e. the detailed investigation of the dependence of the noise level on experimental parameters provides a unique tool to gain information about the device under study: (i) the resistance dependence of the relative noise levels represents transitions between different transport regimes; (ii) the frequency-dependent noise spectrum separates the contribution of individual nearby fluctuators and the ensemble of more remote fluctuators; (iii) noise measurements in the nonlinear current-voltage regime reveal nonlinear noise phenomena that are strongly dependent on the relevant fluctuation parameter of the transport model and are able to identify the source of the fluctuations. Furthermore, a thorough noise analysis not only provides a fundamental understanding of the mechanisms involved, but also lays the foundation for noise reduction strategies in high accuracy applications and enables the development of controlled noise sources for advanced probabilistic computing.

H-2:IL07 Brain-inspired Computing with Nonlinear Dynamical Materials
R.S. WILLIAMS, Department of Electrical and Computer Engineering, Texas A&M University, College Station, TX, USA

We have entered an era to find new modes of computation that will continue to advance exponentially with time even though transistor circuits only improve modestly. Much of the inspiration for new ways of computing comes from what little we understand about the brain. Since the brain itself is a highly nonlinear dynamical system, an appropriate area to investigate is the Principle of Local Activity, which provides a basis for inventing and building new generations of nanoscale oscillators and amplifiers that emulate the integrate and fire dynamics of neurons for signal processing, learning and computation. We need to design neuromorphic circuits that are biased at the Edge of Chaos, where the emergence of complex patterns and behavior in a homogeneous medium are found. However, this is not sufficient. We also need to understand how thermal fluctuations and entropy production minimization contribute to the optimization of brain-like computation. We are just now starting to understand these issues work together, and I will provide an overview of these concepts.

H-2:L08 Noise Diagnostics of Nanoscale Memristor Devices
S. SCHMID¹, Z. Balogh^{1, 2}, B. Sánta^{1, 2}, L. Pósa^{1, 3}, A. Halbritter^1, 2, ¹Department of Physics, Institute of Physics, Budapest University of Technology and Economics, Budapest, Hungary; ²ELKH-BME Condensed Matter Research Group, Budapest, Hungary; ³Institute of Technical Physics and Materials Science, Centre for Energy Research, Budapest, Hungary

The voltage-induced volatile resistive switching phenomenon of vanadium-dioxide memristors is widely utilized to emulate artificial neuronal functionalities. The operation of these devices fundamentally relies on the Mott-type insulator-to-metal transition of the nanoscale active volume, where fluctuation phenomena are expected to play a significant role. We perform a thorough noise-diagnostics of VO2 resistive switches with an ultrasmall active volume, comparing the noise characteristics in the high and low resistance states, contrasting the measured relative noise amplitudes to other types of resistive switching systems, explaining the non-obvious voltage dependence of the relative noise amplitude in the metallic low resistance state, and resolving the noise evolution in the verge of the local phase transition.

H-2:IL09 Development of Ferroelectric Tunnel Junctions and Field-effect Transistors Compatible with Back-end-of-line Integration for Neuromorphic Computing
T.L. Phan¹, K.S. Nair^{1, 2}, M.H. Raza¹, V. Deshpande¹, W. Hamouda¹, C. Dubourdieu^{1, 2}, ¹Helmholtz-Zentrum Berlin für Materialien und Energie, Berlin, Germany; ²Freie Universität Berlin, Physical Chemistry, Berlin, Germany

Ferroelectric tunnel junction (FTJ) and ferroelectric field-effect transistor (FeFET) memories based on the ferroelectric compound Hf0.5Zr0.5O2 (HZO) are well suited for emerging neuromorphic applications due to their low power consumption, non-volatility, and potential to attain multiple resistance states. Moreover, HZO is compatible with back-end-of-line integration. We will discuss the fabrication and characterization of HZO-based FTJs and FETs as stand-alone devices as well as integrated in the back-end-of-line of CMOS chips. The FTJ devices consist of HZO-Al2O3 bilayer. We will discuss the optimization of the whole stack and of the electrical wake-up process. After integration in the BEOL, multiple well-separated resistance states are demonstrated. For the FET devices, HZO is combined with amorphous gallium oxide as the semiconducting channel in a back gated configuration. We demonstrate transistor operation with multiple threshold voltages associated to partial ferroelectric switching in HZO. The FeMFET devices exhibit a large memory window of ~ 3.2 V, a high on/off ratio of 105 and an ultra-small subthreshold swing of 68 mV/dec. The combination of HZO and a-GaOx for FeFETs holds promise for the development of BEOL-compatible neuromorphic hardware based on ferroelectrics.

H-2:IL10 Domain Switching Dynamics in the Ferroelectric AlScN Thin Film Capacitors
A. GRUVERMAN, Department of Physics and Astronomy, University of Nebraska, Lincoln, NE, USA

Recently discovered ferroelectricity in the aluminum scandium nitride (AlScN), a piezoelectric with wurtzite crystal structure, has triggered a surge of interest in the mechanism of polarization reversal in the emerging III-V semiconductor based ferroelectrics. Here, we investigate the field-induced domain structure evolution in the AlScN thin film capacitors using a combination of the macroscopic pulse measurement and piezoresponse force microscopy (PFM). This approach provides a unique insight into the details of nanoscopic domain growth dynamics allowing direct measurements of such parameters as the nucleation rate and domain wall velocity. It is found that the polarization switching kinetics strongly depends on the applied switching field, where several orders of magnitude increase in the switching speed is achieved by doubling the applied field over the coercive one. The polarization switching dynamics experiences a gradual transition from a mechanism described by the nucleation limited switching (NLS) model at a low-field regime toward the one described by the Kolmogorov-Avrami-Ishibashi (KAI) model of more uniform switching in a high-field regime.

H-2:IL11 Tuning the Switching Speed of Valence Change-based Memristive Devices by Thermal Enhancement Layers
A. Sarantopoulos, S. Menzel, R. Dittmann, Forschungs-zentrum Jülich GmbH, Germany; K. Lange, IWE II, RWTH Aachen University, Germany, F. Rivadulla, CIQUS, Universidad de Santiago de Compostela, Spain

Due to their wide dynamic range of resistance with analogue tunability and complex switching dynamics, redox-based memristive devices enable revolutionary new functions and computing paradigms. Switching speed is one of the critical parameters, and while there have been efforts to accelerate the switching kinetics of memristive devices, this usually comes at a cost in terms ofenergy cost. Here we present an approach to accelerate the switching kinetics of the memristive model system SrTiO3 by up to 1000 times, or reduce the operating voltage by ≈ 30% to maintain the switching speed. Our approach is to introduce a low thermal conductivity layer inside the active electrode of the active electrode of the devices, which blocks the heat dissipation caused by Joule heating during switching. Our method leaves the switching layer and its interfaces with the electrodes unaffected. The use of HfO2 and TaOx as the heat blocking layers ensures ease of fabrication and CMOS compatibility. We will demonstrate that this approach is transferable to other more common material systems.

H-2:L12 Correlation between Electronic Structure and Microstructure of Al2O3/TiOx-based Memristive Cells Switched in Filamentary- and Area-mode
s. Hoffmann-Eifert, S. Aussen. F. Cüppers, C. Funck, S. Menzel, R. Dittmann, R. Waser, Peter Grünberg Institut (PGI 7 and 10) and JARA-FIT, Forschungszentrum Jülich GmbH, Jülich, Germany; S. Werner, C. Pratsch, Helmholtz-Zentrum für Materialien und Energie GmbH, Department X-ray Microscopy, Berlin, Germany; J. Jo, R. Dunin-Borkowski, Ernst Ruska-Center (ERC-1 / PGI-5) and JARA-FIT, Forschungszentrum Jülich GmbH, Jülich, Germany

Memristive devices based on the valence change mechanism (VCM) are promising candidates for emerging memories and neuromorphic applications. Aggressive scaling requirements provoke a competition about optimized performance between two operational modes of redox-based resistive random-access memory (ReRAM). These are the standard filamentary-mode and the area-mode characterized by an abrupt and a gradual SET transition from the high to low resistance state, respectively. TiO2-x/Al2O3-based crossbar devices enable stable resistive switching in the filamentary- and area-mode for the same physical device under different programming conditions. By means of operando spectromicroscopy experiments and the analysis of the electronic conduction behavior, the two switching modes are clearly correlated to differences in the microstructure and electronic structure. [1] The results are transferable to other bilayer stacks and provide experimental guidance for proper device selection depending on the constraints of the intended application.
[1] S. Aussen et al., Adv. Electron. Mater. 2023, 2300520.

H-2:L13 Resolving the Dynamics of Picosecond Time-scale Resistive Switching
M. CSONTOS¹, S.W. Schmid², L. Pósa^{2, 3}, T.N. Török^2, 3, Y. Horst¹, N.J. Olalla¹, U. Koch¹, I. Shorubalko⁴, J. Leuthold¹, J. Volk³, A. Halbritter^{2, 5}, ¹Institute of Electromagnetic Fields, ETH Zurich, Switzerland; ²Department of Physics, Budapest University of Technology and Economics, Hungary; ³Institute of Technical Physics and Materials Science, Centre for Energy Research, Budapest, Hungary; ⁴Transport at Nanoscale Interfaces Laboratory, Empa, Switzerland; ⁵ELKH-BME Condensed Matter Research Group, Hungary

Characterizing the dynamical aspects of the resistance change in memristor devices is essential to better understand the underlying microscopic mechanisms. Furthermore, the exploitation of the dynamical aspects of resistive switching together with the static resistance states is expected to facilitate various technology breakthroughs from telecommunication frequency applications of memristors to the realization of adaptive neuromorphic circuits [1,2]. Here we present a state-of-the-art experimental approach to monitor fast resistive switching in real-time, at 70 GHz analog bandwidth and picosecond time-resolution. The measurement technique was recently used to achieve picosecond time-scale set switching in Ta/Ta2O5/Pt memristors [3] as well as to demonstrate comparably short set switching times and sub-nanosecond reset delays in nanometer-scale Au/VO2/Au phase change cells [4].
[1] S. Kumar et al. Nat. Rev. Mater. 7, 575-591 (2022).
[2] D. Molnár et al. arXiv:2307.13320 (2023).
[3] M. Csontos et al. Adv. Electron. Mater. 9, 2201104 (2023).
[4] L. Pósa et al. ACS Appl. Nano Mater. 6, 9137-9147 (2023).

H-2:IL14 Solution Processing of Metal Oxide Memristors: from Coating to Printing
E. CARLOS, R.A. Martins, M. Franco, J. Deuermeier, R. Martins, A. Kiazadeh, i3N/CENIMAT, Department of Materials Science, NOVA School of Science and Technology, Universidade NOVA de Lisboa and CEMOP/UNINOVA, Caparica, Portugal

Memristor devices are a crucial component for the requirements of the Internet of Things (IoT) which demands ultra-low power and high density with new computing principles, exploiting low-cost products and technologies. Most of the reported memristors use conventional methods based on vacuum techniques which are quite expensive and non-sustainable. As an alternative, solution-processing methods to produce these devices have been studied over the years, being now a reliable technology that offers many advantages for memristor devices, such as high versatility, large area uniformity, and transparency at lower fabrication costs. Solution-based metal oxide memristors are emergent and promising non-volatile memories for future electronics. However, some challenges remain in printing efficient and stable eco-devices using a low thermal budget and high-throughput compatibility. In this work, we review the present status of solution-based metal oxide memristor devices as well as discuss the challenges to pass from coating to printing techniques. Also, we report suitable processes to reduce the associated thermal budget of some metal oxides (IGZO, MoO3, Al2O3) memristors, such as solution combustion synthesis and photonic curing to assure their compatibility with low-cost substrates.

Session H-3 Neuromorphic and unconventional computing: devices, algorithms, circuits, theory

H-3:IL01 Silicon Oxide Memristors: Low-cost, CMOS Compatible, High-density Emerging Memory Technology
F. AGUIRRE¹, W.H. Ng^{1, 2}, M. Schormans¹, M. Dickinson¹, A.J. Kenyon^{1, 2}, B. Jones¹, A. Mehonic^{1, 2}, ¹Intrinsic Semiconductor Technologies Ltd., Madrid, Spain; ²University College London, UK

To overcome the challenges that e-Flash memories currently face, we proposed Silicon-oxide as the switching layer of ReRAM devices, which helps simplifying the structure and thus allowing for a low-cost memory technology fully compatible with the conventional CMOS processes. Here we show the electrical performance of on-chip integrated silicon-oxide 1T1R memristor structures while putting special emphasis in their compact modelling, which is critical to allow reliable electrical simulations of circuits including memristors. We propose a model comprising a system of coupled equations, that in its different variants provides a versatile yet accurate behavioural compact model for resistive switching. Then, given their relevance in the field of hybrid CMOS-memristor circuits, the versatility of the model is exploited to demonstrate the suitability of our silicon-oxide memristors for two critical tasks: Following a bottom-up approach, we first deepen into the analogue world and study the electrical behavior of the circuitry used to write and read the memristor´s memory state. Secondly, we move up towards a mixed-signal approach and show our device applicability for Artificial Intelligence hardware-accelerators based on the memristor Vector-Matrix-Multiplication.

H-3:IL02 Multi-input Logic-in-Memory and Neural Inference Accelerators with RRAM Devices
T. ZANOTTI, P. Pavan, F.M. Puglisi, Università degli studi di Modena e Reggio Emilia, Modena, Italy

The rapid increase in the number of smart connected devices is promoting the creation of a distributed intelligence ecosystem at the edge the network. Still, to enable such transition, the development of novel ultra-low power computing solutions is needed. Among these solutions logic-in-memory (LiM) frameworks based on Resistive Memory technologies enable the energy efficient computation of logic operations directly in-memory, thus removing the main source of inefficiencies of conventional computing architectures, i.e., the data movement between memory and processing units. In LiM frameworks, complex logic operations are computed as a sequence of a small set of core logic operations, resulting in increased delays. In this talk, we present a solution that enables multi-input operations, reducing the computing delay and improving the energy efficiency of the smart IMPLY (SIMPLY) LiM framework. Circuit-level simulations enabled by the UNIMORE RRAM physics-based compact model were used to evaluate the reliability and performance of such solution. In addition, an hardware accelerator for binarized neural networks inference based on the SIMPLY framework was designed and simulated, demonstrating considerable improvements in performance with respect to an embedded system implementation.

H-3:IL03 Memristor Prototyping Platforms for Material, Device and Neural Network-level Integration and Benchmarking
G.C. ADAM, Electrical and Computer Engineering Department, George Washington University, Washington, DC, USA

Artificial intelligence systems are expected to consume increasing amounts of computing resources in the coming decades at significant financial and environmental costs. New hardware alternatives are necessary to keep up with the increasing demand in complexity and energy efficiency. Brain-inspired hardware based on emerging analog devices, like memristors (or resistive switches or ReRAM), show promise for dense energy efficient systems given their ultra-scalable footprint and better energy/bit consumption. However, challenges in terms of yield, uniformity and integration require careful co-design and prototyping to demonstrate potential for system-level adoption. Building prototypes using emerging materials and devices has many challenges in mixed signal data acquisition, hyperparameter optimization, hardware co-processing and packaging that require significant research and development efforts. This talk will describe our interdisciplinary work on designing, fabricating and testing prototyping platforms for the next generation of memristor-based computing hardware. Our vision is to provide an end-to-end system capable of supporting, material-level, device-level and network-level benchmarking of these technologies.

H-3:L04 An Optical Neuromorphic Device for Classification and Pattern Recognition
P. MILANI, B. Paroli, M.A.C. Potenza, CIMAINA and Dipartimento di Fisica, Università di Milano, Milano, Italy

Here we will show that the receptron model ( a generalization of the perceptron) can be used as a starting point for the implementation of an all-optical device that exploits the non-linearity of optical speckle fields produced by a solid scatterer. By encoding these speckle fields we can generate a large variety of target Boolean functions without the need for time-consuming machine learning algorithms. We demonstrate that by properly setting the model parameters, different classes of functions (including reversible logic gates) with different multiplicity can be solved efficiently. Our approach have been successfully used to recognize high definition handwritten patterns. The optical implementation of the receptron scheme opens the way for the fabrication of a completely new class of optical devices for neuromorphic data processing based on a very simple hardware.

H-3:L05b The Simplest Ever-Reported Three-Circuit-Element Hodgkin-Huxley Neuristor
A. ASCOLI¹, A.S. DEMIRKOL², S. SLESAZECK³, F. CORINTO¹, M. GILLI¹, T. MIKOLAJICK^3,4, R. TETZLAFF^2, L.O. CHUA⁵, ¹Department of Electronics and Telecommunications, Politecnico di Torino, Turin, Italy; ²Institut für Grundlagen der Elektrotechnik und Elektronik, Technische Universität Dresden, Dresden, Germany; ³Institut für Halbleiter- und Mikrosystemtechnik, Technische Universität Dresden, Dresden, Germany;⁴NaMLab gGmbH, Dresden, Germany; ⁵Department of Electrical Engineering and Computer Sciences, University of California at Berkeley, Berkeley, CA, USA

Local Activity [1] is the property, which characterizes a physical system capable to act as a local source of energy around an operating point. Edge of Chaos [1] is a special region of the Local Activity domain, where the system is also stable besides being acting about an operating point. Despite sitting on an apparently quiet quiescent state, a physical system poised on the Edge of Chaos is endowed with a high degree of excitability. This explains why an infinitesimally-small perturbation, displacing a system of this kind away from its locally active and stable operating point, may trigger emergent phenomena across its physical medium. The Principle of Local Activity is grounded on solid theoretical foundations, which allow to determine whether there are parameter regions, where a physical system may operate on the Edge of Chaos about some operating point, and to define them quantitatively, in case they exist. Recurring to this theory, it was possible, very recently, to resolve longstanding open problems in science, including, especially, the origin for the appearance of symmetry-breaking effects across homogeneous reaction-diffusion systems from cellular biology [2]. Moreover, the current availability of solid-state memristive nanodevices [3], displaying Local Activity and Edge of Chaos, similarly as the biological potassium and sodium ion channels across neuronal axon membranes, allows to design bioplausible neuromorphic circuits, which may pave the way toward the development of brain-like computing systems in the years to come. In this oral presentation, we present the simplest ever-reported electronic neuron [4]-[5], neuristor for short, which, thanks to the Local Activity of a memristor from NaMLab, employs just two degrees of freedom to reproduce all the three fundamental bifurcation phenomena dictating the life cycle of an Action Potential in the fourth-order Hodgkin-Huxley model [6].
[1] L. O. Chua, “Local Activity is the Origin of Complexity,” Int. J. Bifurcation Chaos, vol. 15, no. 11, pp. 3435–3456, 2005
[2] A. Ascoli, A.S. Demirkol, R. Tetzlaff, and L.O. Chua, “Edge of Chaos is Sine Qua Non for Turing Instability,”IEEE Int. Trans. Circuits and Systems-I (TCAS-I): Regular Papers, vol. 69, no. 11, pp. 4596-4609, 2022
[3] A. Ascoli, A.S. Demirkol, R. Tetzlaff, S. Slesazeck, T. Mikolajick, and L.O. Chua, “On Local Activity and Edge of Chaos in a NaMLab Memristor,” Frontiers in Neuroscience, vol. 15, no. 651452, (30pp.), 2021
[4] A. Ascoli, A.S. Demirkol, I. Messaris, F. Corinto, M. Gilli, R. Tetzlaff, and L.O. Chua, “Edge of Chaos Theory Reveals the Simplest Ever Reported Hodgkin-Huxley Neuristor from NaMLab,” Nature Communications, under review
[5] A. Ascoli, A.S. Demirkol, I. Messaris, R. Tetzlaff, and L.O. Chua, “How to Leverage the Local Activity of a NbOx Threshold Switch to Induce the Fundamental Bifurcations of a Fourth-Order Neuron in a Second-Order Bio-Inspired Circuit,” Advanced Electronic Materials, under submission
[6] A.L. Hodgkin, and A.F. Huxley, “A quantitative description of membrane current and its application to conduction and excitation in nerve”, J. Physiol., vol. 117, no. 4, pp. 500-544, 1952

H-3:L06 Harnessing the Frequency Response of Silicon Oxide Memristors
H.R.J. COX¹, W.H. Ng¹, T. Benkohen¹, D. Das¹, A. Mehonic¹, C. Henderson¹, A. Xhameni², E. Zanganeh², A. Jaman³, R. Jackman¹, T. Banerjee³, A. Lombardo², A.J. Kenyon¹, ¹Dept. of Electronic and Electrical Engineering, University College London, London, UK; ²London Centre for Nanotechnology, University College London, London, UK; ³Faculty of Science and Eng., University of Groningen, Groningen AG, Netherlands

Neuromorphic systems have the potential to transform the fields of Artificial Intelligence and Machine Learning by emulating the continuous, parallel communication between neurons found in the human brain. A promising route to this is the use of oxide-based, intrinsic ReRAM memristors due to their continuous states and compatibility with present semiconductor fabrication processes. During operation, under an applied field, oxygen is driven across the metal-oxide-metal (MIM) stacks, forming conducting filaments which modulates the resistance between the electrodes. Typically, only the linear conductance states of the memristors are used to store information in such systems. However, their complex frequency response, which remains largely unstudied, offers radical new ways to extend the capabilities of memristive computation by exploiting the frequency and phase domains. It is shown here that memristors can be applied as highly tuneable components across radio and microwave frequencies. The devices exhibit rich dynamics, which hint at exciting new ways to encode and process information in the frequency domain. At RF frequencies, the memristors act as tuneable RC low-pass filters, in which both the dimensions and state of the device can be used to vary the cut off frequency. This reveals that it is not only the resistance of the device that is modified, but the permittivity of the silicon oxide dielectric and hence capacitance. Thus, instead of storing information in one dimension with conductance, it can be stored in two dimensions in the complex plane. This also opens the possibility of optoelectronic applications where an electrical signal could be applied to the devices and used to modulate the phase of light passing through the dielectric. At higher frequencies transmission losses make it challenging to measure conventional devices. To overcome this, a coplanar waveguide structure is simulated, designed, and fabricated with a memristor embedded into it. This enables the devices frequency response to be measured across 10 orders of magnitude, up to 20 GHz revealing a highly dynamic system. Embedding ferromagnetic Lanthanum Strontium Manganate, which exhibits negative differential resistance, into a coplanar waveguide has enabled the realisation of a high frequency phase-transition nano-oscillator. Coupling these oscillators with tuneable memristors could enable the creation of a tuneable Ising Hamiltonian solver. Beyond this, initial results demonstrate the memristors can have a dynamic response to the oscillations making them a potential candidate for realising the holy grail of machine learning — a physical Boltzmann Machine.

H-3:L07 Tunable Photoresponsivity Associated with Synaptic Functions in Zinc-Tin Oxide Phototransistor for In-Sensor and Neuromorphic Computing
LI-CHUNG SHIH, Chun-Tao Chen, Ya-Chi Huang, Shuai-Ming Chen, Yu-Chieh Chen, Jen-Sue Chen, Department of Materials Science and Engineering, National Cheng Kung University, Tainan, Taiwan

Drawing inspiration from the visual processing of mammals, machine vision technology empowers the real-time recognition and classification of objects using image sensors combined with processing units, all driven by artificial neural networks (ANNs). To faithfully mimic the photoreceptor cells and the biological neural network in the retina, photodetectors within the image sensor array must be directly coupled with synaptic devices to create an artificial neural circuit. In this study, we demonstrate a dual-function photosynaptic transistor using zinc-tin oxide (ZTO) with embedded Au nanoparticles (NPs) heterostructure. Utilized as a photodetector, the photoresponsivity of our device can be modulated by applying a gate voltage, suggesting prospective applications in in-sensor vector-matrix multiplication (VMM) operations for encoding image information. Furthermore, our device emulates synaptic characteristics by simultaneously applying voltage and optical spikes. The optoelectrical-induced conductance can serve as the weight in the neural network for image classification, achieving a recognition accuracy of 98.2%. Consequently, the ZTO/Au NP photosynaptic transistor exhibits the potential for seamless integration of photodetectors and synaptic transistors for advanced visual cognition.

H-3:IL08 Bayesian Inference Leveraging Nanoscale Device Stochasticity
B. RAJENDRAN, King's College London, London, UK

Bayesian neural networks can provide quantifiable uncertainty metrics associated with their decisions, which could play a crucial role in improving the reliability and trustworthiness of AI models. However, since model parameters of Bayesian networks need to be sampled from learnt stochastic distributions, their realisation in hardware is challenging, especially in CMOS platforms. Alternatively, novel hardware architectures can be devised to leverage or engineer naturally occurring nanoscale stochasticity to implement software-trained Bayesian networks. In this talk, I will discuss our work on designing ensemble architectures based on phase change memory noise as a resource for synaptic sampling in Bayesian spiking neural networks. As a benchmark, I will also compare this with Gibbs sampling-based architectures based on a reduced precision CMOS implementation.

H-3:IL09 Deep Neural Network Inference with a 64-core in-Memory Compute Chip based on Phase-change Memory
M. LE GALLO, IBM Research Europe, Rüschlikon, Switzerland

The need to repeatedly shuttle around synaptic weight values from memory to processing units has been a key source of energy inefficiency associated with hardware implementation of artificial neural networks. Analog in-memory computing (AIMC) with spatially instantiated synaptic weights holds high promise to overcome this challenge, by performing matrix-vector multiplications directly within the network weights stored on a chip to execute an inference workload. We designed and fabricated a multi-core AIMC chip in 14-nm complementary metal–oxide–semiconductor (CMOS) technology with backend-integrated phase-change memory (PCM). The fully-integrated chip features 64 256x256 AIMC cores interconnected via an on-chip communication network. In this talk, I will present our latest efforts in employing this chip for performing inference of deep neural networks. First, the PCM technology and computational unit-cell we use will be described. Next, experimental inference results on ResNet and LSTM networks will be presented, with all the computations associated with the weight layers and the activation functions implemented on-chip. Finally, I will present our open-source toolkit (https://aihw-composer.draco.res.ibm.com/) to simulate inference and training of neural networks with AIMC.

H-3:L11 Single-node Reservoir Computing through a Memristive Circuit with Complex Dynamics
S. BRIVIO, M. Escudero, S. Spiga, CNR – IMM, Unit of Agrate Brianza, Italy

The reservoir computing concept prescribes an ensemble of interacting dynamical objects to preprocess information, through their collective dynamics, in favor of the post-processing by a linear and easily trainable readout. The design, realization and optimization of physical reservoirs is a complex task that led to the proposal of single-node systems (mainly in the photonic realm) that, despite their compactness, show useful complex dynamics. In this work, we realize an electronic single-node reservoir made of an oscillator circuit with nonlinear and tunable properties enabled by a nonvolatile memristor. Pt/HfO2/TiN devices are programmed through several resistance states featuring different nonlinear current-voltage characteristics required to generate complex dynamics in a circuit inspired from the Murali-Lakshmanan-Chua one. We demonstrate the hardware system to be able to perform nonlinear classification tasks and extend the results through simulations toward spatial and temporal tasks with increased complexity. We also complement the reservoir system with the implementation of the readout layer through a memristor array performing analog vector-matrix multiplication.
The work is partially supported by the PRIN2017-MIUR project COSMO (Prot. 2017LSCR4K)

H-3:L12 Autonomous Neural Information Processing by a Dynamical Memristor Circuit
D. Molnar^{1, 2}, T.N. Török^{1, 3}, R. Kövecs¹, L. Pósa^{1, 3}, P. Balázs¹, Gy. Molnár³, N.J. Olalla⁴, J. Leuthold⁴, J. Volk³, M. Csontos⁴, A. Halbritter^{1, 2}, ¹Department of Physics, Institute of Physics, Budapest University of Technology and Economics, Budapest, Hungary; ²HUN-REN–BME Condensed Matter Physics Research Group, Budapest, Hungary; ³Institute of Technical Physics and Materials Science, Centre for Energy Research, Budapest, Hungary; ⁴Institute of Electromagnetic Fields, ETH Zurich, Zurich, Switzerland

Analog tunable memristors are widely utilized as artificial synapses in various neural network applications. However, exploiting the dynamical aspects of their conductance change to implement active neurons is still in its infancy, awaiting the realization of efficient neural signal recognition functionalities. Here we experimentally demonstrate an artificial neural information processing unit that can detect a temporal pattern in a very noisy environment, fire an output spike upon successful detection and reset itself in a fully unsupervised, autonomous manner [1]. This circuit relies on the dynamical operation of only two memristive blocks: a non-volatile Ta2O5 device and a volatile VO2 unit. A fading functionality with exponentially tunable memory time constant enables adaptive operation dynamics, which can be tailored for the targeted temporal pattern recognition task. In the trained circuit false input patterns only induce short-term variations. In contrast, the desired signal activates long-term memory operation of the non-volatile component, which triggers a firing output of the volatile block.
[1] D. Molnar et al. arXiv:2307.13320

H-3:L13 Nonlinear Dynamics and Local Activity in Bio-inspired Memristor Networks
A. ASCOLI¹, F. CORINTO¹, M. GILLI¹, R. TETZLAFF², ¹Dept of Electronics and Telecommunications, Politecnico di Torino, Turin, Italy; ²Institute of Circuits and Systems, Faculty of Electrical and Computer Engineering, TU Dresden, Dresden, Germany

Nonlinear oscillators are systems which, undergoing very rich dynamics, are employable as primitives in several applications. Networks of oscillators exhibit complex interactions, which have been exploited in various works to address problems without efficient solutions on von- Neumann machines, including pattern recognition [1] and combinatorial problem optimization [2]. Under the fundamental principle of Local Activity, including its Crown Jewel, referred to as the Edge of Chaos, arrays of bio-inspired nonlinear oscillators, built with volatile memristors, may even process data similarly as biological systems [3]-[4]. Additionally, non-volatile memristive devices display nonlinear characteristics, which are particularly suitable to design circuits experiencing a plethora of complex dynamic behaviours, among which periodic oscillations and chaos, and transitioning between them via well-defined bifurcations. In a recent work [5], we presented the theoretical framework as well as the physical implementation of a memristive Chua’s circuit, which leverages the programmable nonlinear conductance of a non-volatile ReRAM cell to switch between a number of different oscillatory operating modes through fundamental bifurcation phenomena. All in all, the application of rigorous nonlinear circuit- and system-theoretic techniques [6], including the universal principle of Local Activity, allows to design bio-inspired neural networks, including reservoir computing systems, which exploit the unique physical properties of volatile and non-volatile memristors to outperform state-of-the-art computing machines [7].
1. D. E. Nikonov et al., “Coupled-oscillator associative memory array operation for pattern recognition,” IEEE J. Explor. Solid-State Comput. Devices Circuits, vol. 1, pp. 85–93, 2015.
2. N. Mohseni, P. L. Mcmahon, and T. Byrnes, “Ising machines as hardware solvers of combinatorial optimization problems,” Nature Rev. Phys., vol. 4, no. 6, pp. 363–379, May 2022.
3. A. Ascoli, A.S. Demirkol, R. Tetzlaff, and L.O. Chua, “Edge of Chaos is Sine Qua Non for Turing Instability,” IEEE Int. Trans. Circuits and Systems-I (TCAS-I): Regular Papers, vol. 69, no. 11, pp. 4596-4609, 2022, DOI: 10.1109/TCSI.2022.3194465
4. A. Ascoli, A.S. Demirkol, R. Tetzlaff, and L.O. Chua, “Edge of Chaos Theory Resolves Smale Paradox,” IEEE Trans. Circuits and Systems-I (TCAS-I): Regular Papers, vol. 69, no. 3, pp. 1252--1265, March 2022, DOI: 10.1109/TCSI.2021.3133627
5. M. Escudero et al., “Chua’s Circuit With Tunable Nonlinearity Based on a Nonvolatile Memristor: Design and Realization,” IEEE Transactions on Circuits and Systems-I: Regular Papers, DOI: 10.1109/TCSI.2023.3323854.
6. F. Corinto, M. Forti, and L.O. Chua, “Nonlinear Circuits and Systems with Memristors - Analogue Computing via the Flux-Charge Analysis Method,” Springer, 2020, ISBN-13: 978- 3-030-55650-1
7. M.Weiher, M. Herzig, R. Tetzlaff, A. Ascoli, S. Slesazeck, T. Mikolaijck, “Improved Vertex Coloring With NbOx Memristor-Based Oscillatory Networks,” IEEE Trans. Circuits and Systems-I (TCAS-I): Regular Papers, 2021, DOI: 10.1109/TCSI. 2021.3061973

H-3:IL14 Edge of Chaos Theory for Unconventional Computing
R. TETZLAFF¹, A. Demirkol¹, A. Ascoli¹, L.O. Chua², ¹Institute of Circuits and Systems, TU Dresden, Dresden, Germany; ²Department of Electrical Engineering and Computer Sciences, University of California Berkeley, Berkeley, CA, USA

Numerous investigations have focused on the development of new computer architectures to overcome the limitations of the classical von Neumann architecture, which is characterized by physically separate computation and storage operations, resulting in limited throughput and not enabling real-time computation in smart sensing devices. Recently developed memristor-based computing architectures, where processing and data storage occur in the same physical location, range from memristive crosspoint arrays to memristive artificial and bio-inspired neural networks. Although recent results from different developments show that dot-product operations or correlation detection can be efficiently performed by such memristive memory crosspoint arrays, recent developments point to general memristive computational structures known as Memristor Cellular Neural Networks (M-CNN) [1], which provide extremely high computational performance and, when equipped with memristors, can also be used as pure memory systems within a program flow. Typically, computing with these structures is often based on complex behaviour that occurs in these arrays of nonlinear identical systems. Fundamental results were derived by Chua [2], who proved that the emergence of complexity in these structures relies on local activity and, in particular, on a parameter subset called the Edge of Chaos (EOC). The edge of chaos theory is presented and discussed in detail in this paper. In addition, M-CNN are proposed for the implementation of general-purpose programmable mem computing structures that enable unconventional computing in sensor-processor systems.
[1] R. Tetzlaff, A. Ascoli, I. Messaris, and L.O. Chua, “Theoretical Foundations of Memristor Cellular Nonlinear Networks: Memcomputing with Bistable-like Memristors,” IEEE Trans. on Circuits and Systems–I: Regular Papers, 2019, 10.1109/TCSI.2019.2940909
[2] L.O. Chua, “Local Activity is the Origin of Complexity,” International Journal of Bifurcation and Chaos, Vol.15, No.11, pp. 3435-3456, 2005

Session H-4 Theory, modelling and simulation of materials and devices for future computing

H-4:IL01 Latest Advances in Modelling of Valence Change and Electrochemical Resistive Switching Devices
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. Despite the advances made in the understanding of the fundamental mechanism, a comprehensive model explaining all aspects is still missing. In this talk, the latest advances in the modeling of ECM and VCM devices will be discussed. For ECM devices a variability-aware compact model will be presented that reproduces the variability of Cu/SiO2 based ECM devices. In addition, an extension of this model is presented that allows for the simulation of volatile switching (v-)ECM cells. For v-ECM cells, using a 2D electrochemical continuum model of the filament growth, we show that the applied voltage influences the size of the filament due to the change of the growth-limiting process. We recently showed that the electronic transport of VCM devices depends on the energetic level of the oxygen vacancy defect state leading to two major types of conduction mechanism. These two types of mechanism have been included in our JART VCM compact model improving the description of electronic transport.

H-4:IL02 Density Functional Simulations of Ag Migration in a Conductive Bridging Random Access Memory Cell
J. AKOLA, Department of Physics, Norwegian University of Science and Technology (NTNU), Trondheim, Norway

We have performed density functional/molecular dynamics (DF/MD) simulations to investigate the drift of Ag atoms in an amorphous GeS2 solid-state electrolyte between Ag and Pt electrodes in the presence of a finite electric field. The model structure of 1019 atoms represents a conductive bridging random access memory (CBRAM) device. Simulations under an external electrostatic potential show Ag migration and the formation of percolating single-atom Ag wires through the solid-state electrolyte. The electronic structure analysis of selected snapshots shows that dissolved Ag atoms become markedly cationic, which changes when Ag clusters form at the Pt electrode. Sulfur becomes more anionic during the migration as a result of Ag-S bonding, and the effect is most pronounced near the active (Ag) electrode. The formation of conductive filaments requires a percolating network of Ag clusters to grow from the Pt interface, and the weakest links of this network appear at the Ag electrode. We also presents result for our latest electronic structure analyses of the prototypical phase-change material Ge2Sb2Te5. Hybrid DF simulations have enabled us to pinpoint electron/hole localization effects with local geometrical motifs both in the amorphous and re-crystallized phases.

H-4:IL03 Multi-scale Modelling of Valence Change Memory Cells
m. luisier, M. Kaniselvan, M. Mladenovic, Integrated Systems Laboratory, ETH Zurich, Zurich, Switzerland

The modelling of modern nanoscale devices such as valence change memory (VCM) cells, a well-established type of memristors, requires the availability of advanced technology computer aided design (TCAD) tools. The latter should go beyond classical approximations and rely on atomistic quantum mechanical concepts to reproduce experimental data and predict the "current vs. voltage" characteristics of not-yet-fabricated components. In particular, to shed light on the behaviour of resistive switching random access memories, which consist of metal-insulator-metal stacks, molecular dynamics (MD), kinetic Monte Carlo (KMC), density functional theory (DFT), and quantum transport (QT) should be combined. By doing so, the growth of nano-filaments through realistic oxides embedded between two metallic electrodes can be accurately simulated and the electrical current flowing through them computed with an atomic resolution. A better understanding of these key features will help design next-generation devices with improved performance.

H-4:L05 A Machine-learning Interatomic Potential for GeSbTe Phase Change Alloys
O. Abou El Kheir, D. Baratella, Department of Materials Science, University of Milano-Bicocca, Milano, Italy; L. Bonati, M. Parrinello, Italian Institute of Technologies (IIT), Genova, Italy; M. Bernasconi, Department of Materials Science, University of Milano-Bicocca, Milano, Italy

Atomistic simulations based on density functional theory (DFT) have provided useful insights on the properties of phase change materials over the last years. However, several key issues for the operation of memory devices are beyond the reach of DFT simulations. A route to enlarge the scope of DFT methods is the exploitation of machine learning techniques to generate interatomic potentials for large scale molecular dynamics simulations trained on a DFT database [1]. In this talk, we report on the generation of an interatomic potential for the Ge2Sb2Te5 compound within the neural network framework implemented in the DeePMD-kit package [2] that allows simulating several tens of thousands of atoms for tens of ns at a modest computational cost [3]. The validation of the potential and its application to the study of the crystallization kinetics will be discussed. Extension of the potential to study crystallization with phase separation in Ge-rich alloys of interest for applications in embedded memories will be discussed as well.
[1] G. C. Sosso and M. Bernasconi, MRS Bulletin 44, 705 (2019).
[2] H. Wang, L. Zhang, J. Han, and W. E, Comp. Phys. Commun. 228, 178 (2018)
[3] O. Abou El Kheir, L. Bonati, M. Parrinello, M. Bernasconi, arXiv: 2304.03109 (2023).

H-4:L06 Are Machine Learning Interatomic Potentials Always Better for Modeling Amorphous Metal Oxides?
S. GRAMATTE^{1, 2, 3}, V. Turlo¹, O. Politano², ¹Lab. for Advanced Materials Processing, Empa - Swiss Federal Labs for Materials Science and Technology, Thun, Switzerland; ²Lab. Interdisciplinaire Carnot de Bourgogne, UMR 6303 CNRS-Université de Bourgogne, Dijon Cedex, France; ³Lab. for Joining Technologies and Corrosion, Empa - Swiss Federal Labs for Materials Science and Technology, Duebendorf, Switzerland

Amorphous metal oxides offer unique electronic and optical properties that make them suitable for a variety of computing applications, ranging from digital storage and transistors to neuromorphic and unconventional computing devices. In our comprehensive study, we methodically assessed the performance of multiple classes of interatomic potentials in comparison to the DFT benchmark database, using bulk amorphous alumina as a model system. For stoichiometric alumina, the Born-Mayer-Huggins (BMH) fixed-charge model provides the most effective balance of computational efficiency and accuracy. For non-stoichiometric alumina, variable charge ReaxFF potential displayed a strong agreement with DFT calculations. Furthermore, we tested the graph neural network-based NequIP framework to train machine learning potentials (MLPs) on a small portion of the ab initio database, with a training time of a few hours. The Pareto-optimal MLPs easily surpassed ReaxFF in terms of both precision and computational speed, though cannot achieve the same for BMH potential. Thus, we highlight that simple classical models are still the best for modeling bulk amorphous metal oxides in cases when computational performance is critical, while machine learning models should be used for more complex cases.

H-4:L07 Modelling of Stochastic Switching in Monolayer MoS2 RRAMs with Kinetic Monte Carlo
L. Peddaboina, G. Hegde, J.S.A. Nandan Karalapati, O. Badami, S. Bhattacharjee, IIT Hyderabad, Kandi, Sangareddy, India

Stochastic switching, inherent in resistive random access memory (RRAM) can be leveraged to create devices for security and probabilistic computing. In this regard, it is necessary to model the C2C stochasticity of RRAM devices to predict the system-level performance of circuits built using these devices. Monolayer 2D materials have demonstrated stochastic multistate operation, and present excellent opportunity for voltage and area scaling [1]. Although the stochastic switching behavior for oxide-based RRAM is well explored, a similar study for 2D materials is in a very nascent stage. In this work, we propose an efficient physics-based model to capture and study the C2C variability in monolayer MoS2 RRAMs. The microscopic origin of stochasticity due to the S atoms popping in and out of the moly plane [2] is modeled using the kinetic Monte Carlo method. Carefully designed rate equations that capture the physics of abrupt SET (threshold switching) and gradual RESET (Joule heating) processes provide an excellent fit to experimental data. This enables us to extract key material parameters for further system-level optimization of devices for stochastic computing.
[1] S. Bhattacharjee et.al ACS App Mat & Interf.12(5),6022-6029(2020)
[2] S. Mitra et.al npj 2D Mater Appl5,33(2021)

Session H-5 2D materials- and soft materials-based devices

H-5:L03 Visual Memory in a 2D Memitter Based on WS2
F. FERRARESE LUPI, G. Milano, A. Angelini, Advanced Materials Metrology and Life Science Division, INRiM (Istituto Nazionale di Ricerca Metrologica), Torino, Italy; M. Rosero Realpe, B. Torre, Department of Applied Science and Technology (DISAT), Politecnico di Torino, Torino, Italy; E. Kozma, CNR-SCITEC, Milano, Italy; C. Martella, C. Grazianetti, CNR-IMM, Unit of Agrate Brianza, Agrate Brianza Italy

In the era of Big Data and Artificial Intelligence, there is a growing expectation for neuromorphic systems to outperform the current limitations of computing technology by emulating the information processing capabilities of the human brain. Neuromorphic architectures, both electronic and optical, have been developed for implementing neural networks and brain-inspired computing paradigms. Adaptive materials are also proposed for in-sensor information processing, eliminating the need for data transfer between sensors and computing units. Among emerging neuromorphic technologies, materials able to adapt their response to external stimulation have been used to emulate synaptic plasticity and neuron functionalities, with 2D materials showing promise due to their scalability and integration capabilities. Here we introduce the novel concept of a "2D memitter" that uses WS2 monolayers to achieve all-optical neuromorphic data processing. We demonstrate the time-dependent, highly nonlinear photoluminescence (PL) response of WS2 flakes under optical stimulation, which exhibits fading memory characteristics and synaptic behavior. Through experimental and modeling approaches, we illustrate the potential for PL dynamics to replicate biological functions like the Visual Short-Term Memory.

H-5:L04 2D Van der Waals NbTe4 Phase Change Material: Enabling Ultralow Thermal Consumption
YI SHUANG¹, Q. Chen^{2, 3}, M. Kim⁴, Y. Wang⁴, Y. Saito⁵, S. Hatayama⁵, P. Fons⁶, D. Ando⁴, M. Kubo^{2, 3}, Y. Sutou^1, 4, ¹WPI Advanced Institute for Materials Research, Tohoku University, Aoba, Sendai, Japan; ²New Industry Creation Hatchery Center, Tohoku University, Aramaki, Aoba-ku, Sendai, Japan; 3Institute for Materials Research, Tohoku University, Aoba-ku, Sendai, Japan; ⁴Department of Materials Science, Graduate School of Engineering, Tohoku University, Sendai, Japan; ⁵Device Technology Research Institute, National Institute of Advanced Industrial Science and Technology (AIST), Tsukuba Central 2, Tsukuba, Japan; ⁶Department of Electronics and Electrical Engineering, Faculty of Science and Technology, Keio University, Yokohama, Kanagawa, Japan

2D vdW transition metal di-chalcogenides (TMDs) are gaining attention in nonvolatile memory due to their adaptable properties and scalability. However, their intricate switching and complex production methods hinder large-scale manufacturing. Sputtering offers potential for extensive TMD production, but their high melting point (usually > 1000°C) requires elevated temperatures for good crystallinity. This study explores low-melting 2D vdW transition metal tetra-chalcogenides (TM4X4) and identifies NbTe4 with an exceptionally low melting point at around 447°C. Initially, deposited NbTe4 is amorphous and can be crystallized through annealing at temperatures over 272°C.[1] This unique combination of low melting point and high crystallization temperature addresses key challenges in phase-change memory materials, such as high thermal consumption and poor thermal stability in the amorphous phase. Consequently, NbTe4 is a highly promising candidate to potentially resolve these issues.
[1] Y. Shuang, et al, “NbTe4 Phase-Change Material: Breaking the Phase-Change Temperature Balance in 2D van der Waals Transition-metal Binary Chalcogenide”, Advanced Materials. adma.20230346

H-5:L06 3D Printable High-performance Soft Material for Neural Interface Applications
TAO ZHOU, Dept of Engineering Science and Mechanics, Pennsylvania State University, PA, USA; Center for Neural Engineering, Pennsylvania State University, PA, USA; Huck Institutes of The Life Sciences, Pennsylvania State University, PA, USA; Materials Research Institute, Pennsylvania State University, PA, USA

In order to establish effective and enduring communication with delicate and vulnerable neural tissues, neural interfaces necessitate an intimate and prolonged functional interaction. Consequently, it is imperative for these interfaces to exhibit a high level of biocompatibility and elicit minimal foreign body responses. The field of neural interfaces has witnessed significant advancements in recent decades, characterized by notable progress in both design and material innovations. These advancements have aimed to achieve tissue-matching properties, thereby minimizing foreign body responses, while simultaneously enabling desired electrophysiological functionalities such as neural recordings and stimulations. Due to their excellent tissue-matching qualities, hydrogels have emerged as very promising materials for facilitating biocompatible, long-term interactions with biological tissues. In this study, we present a micro-structured neural interface composed of conductive hydrogels that are both stretchy and soft, fabricated using 3D printing technology. The unique assortment of hydrogel inks that can be printed in three dimensions allows for the simple and flexible additive manufacturing method for the production of neural interfaces in a variety of configurations.

H-5:L07 Resistive Switching Memory Behaviours in Bio-degradable Composites
v.s. Vallabhapurapu¹, Z.W. Dlamini², S. Vallabhapurapu³, ¹Department of Physics, University of South Africa, Johannesburg, South Africa; ²Central University of Technology, South Africa; ³School of Computing, University of South Africa, Johannesburg, South Africa

Environmental friendly, bio compatible and biodegradable green technologies are the need of the next generation sustainable life style. In this context electrical conduction and resistive switching properties of several biomaterials and biodegradable composites have been studied in detail. Raw bio materials like plant and animal milk, bio-nano composites of Gelatin, Chitosan etc with the incorporation of suitable nano particles are used as functional materials to fabricate ReRAM devices. All devices are well characterized and studied in detail for conducting and resistive switching properties. We report on the unexpected resistive switching memory behaviours in raw cow milk and other plant extracts. The ReRAM parameters are not comparable to the state of art inorganic ReRAMs in the literature, but one can see a grand opening into the bio compatible and biodegradable ReRAM developments. Our measurements show a definitive cycling dependence and calls for further optimization of these devices. Here we present our latest results and review in the light of current ReRAM technologies.

Session H-6 Nanomaterials and unconventional subtrates for computing

H-6:IL01 Reservoir Computing with Nanowire Networks
G. MILANO¹, C. Ricciardi², ¹Advanced Materials Metrology and Life Science Division, INRiM (Istituto Nazionale di Ricerca Metrologica), Italy; ²Department of Applied Science and Technology, Politecnico di Torino, Italy

Physical reservoir computing represents a computational framework that leverages physical laws to enable a reduction of the effective computational costs. Here, we report on in materia implementation of reservoir computing in self-organizing memristive nanowire networks. We show that these multiterminal systems, that have been recently proposed as alternatives to conventional memristive crossbar arrays, represents biologically plausible substrates for neuromorphic-type of data processing. By exploiting emergent behavior of the system, we show through a combined experimental and modeling approach that emergent dynamics can be exploited for efficient data processing in the framework of reservoir computing. In this context, we show that a fully memristive computing architecture realized by coupling self-organized nanowire networks with conventional memristive devices can be exploited to perform pattern recognition, speech recognition and time-series prediction. Furthermore, we report on tomographical evidence of memory engrams, i.e., physicochemical changes in biological neural substrates supposed to endow the representation of experience in our brain, showing that these networks can be exploited for both encoding and storage of information in the same physical location.

H-6:IL02 Materializing Cognition – Information Processing in Cognitive Matter
W.G. VAN DER WIEL, Center for Brain-Inspired Nano Systems (BRAINS), University of Twente, Enschede, The Netherlands and Institute of Physics, University of Münster, Münster, Germany

By harnessing the natural richness of materials and developing design principles for adaptation and autonomy, we aim to realize ‘cognitive matter’ and compute with it [1,2]. Here we discuss experiments in disordered nanomaterial systems, where use 'material learning' to realize functionality. We show that a ‘designless’ network of gold nanoparticles can be configured into Boolean logic gates using artificial evolution [3]. We demonstrate that this principle is generic and exploit the nonlinearity of a disordered nanoscale network of dopants in silicon and significantly facilitate handwritten digit classification [4]. An alternative material-learning approach is followed by first mapping our DNPU on a deep-neural-network model, which allows for applying standard machine-learning techniques in realizing functionality [5]. We show that our devices are not only suitable for solving static problems but can also be applied in highly efficient real-time processing of time-dependent signals.
[1] C. Kaspar et al., Nature 594, 345–355 (2021).
[2] H. Jaeger et al., Nat. Commun. 14, 4911 (2023).
[3] S.K. Bose, C.P. Lawrence et al., Nat. Nanotechnol. 10, 1048 (2015).
[4] T. Chen et al., Nature 577, 341 (2020).
[5] H.-C. Ruiz Euler et al., Nat. Nanotechnol. 15, 992 (2020).

H-6:IL03 Emergent Brain-like Dynamics from Memristive Networks
Z. KUNCIC, School of Physics, University of Sydney, NSW, Australia; F. Caravelli, Theoretical Division (T4), Los Alamos National Laboratory, Los Alamos, NM, USA

In response to electrical signals, heterogeneous networks of highly interconnected metallic nanowires exhibit memristive switching, arising from electro-chemical evolution of conductive nano-filaments at their cross-point junctions. Due to their self-organized nature, these memristive networks behave similar to disordered many-body physical systems, characterized by emergent nonlinear dynamics such as dynamical phase transitions, synchronization and spatio-temporal correlations. In this talk, we draw a comparison of these emergent dynamics with brain dynamics, arising from neuronal microcircuits established by synaptic connectivity patterns. Using analysis tools and concepts from complexity science and statistical physics, we find a critical threshold at which the emergent dynamics of memristive networks can be identified with symmetry breaking, similar to that exhibited by other complex physical systems, including the brain. Furthermore, we show that this state is optimal for machine learning using memristive networks as reservoir computers.

H-6:L04 Composite Nanogranular Networks: Brain-like Resistive Switching Patterns and In Situ Current Path Imaging
B. Adejube¹, O. Gronenberg², T. Hemke³, N. Carstens¹, R. Gupta¹, O.-H. Asnaz⁴, T. Strunskus^{1, 5}, F. Faupel^{1, 5}, T. Mussenbrock³, J. Benedikt^{4, 5}, L. Kienle^{2, 5}, A. Vahl^{1, 5}, ¹Dept of Materials Science - Chair for Multicomponent Materials, Faculty of Engineering, Kiel University, Kiel, Germany; ²Dept of Materials Science - Synthesis and Real Structure, Faculty of Engineering, Kiel University, Kiel, Germany; ³Chair of Applied Electrodynamics and Plasma Technology (AEPT), Ruhr University Bochum, Bochum, Germany; ⁴Experimental Plasma Physics, Institute of Experimental and Applied Physics, Kiel University, Kiel, Germany; ⁵Kiel Nano Surface and Interface Science KiNSIS, Kiel University, Kiel, Germany

Nanoparticle networks, self-organized and poised at the percolation threshold, are promising candidates as a platform for bio-inspired information processing, i.e. by harnessing high dimensionality, non-linear responses and brain-like avalanche dynamics caused by the spatio-temporal distribution of resistive switching events within the networks. In this contribution, we present a versatile approach to fabricate composite nanoparticle networks from tailored building units by means of nanoparticle beam deposition techniques and a controlled surface modification of Ag nanoparticles in an RF plasma. Criticality and avalanche dynamics in the current response are showcased by a Ag/CxOyHz/Ag composite network at the percolation threshold. Furthermore, a novel approach to image the evolving current paths in nanoparticle networks is demonstrated using resistive contrast imaging and active voltage contrast imaging complementarily. To gain a deeper understanding of the complex current response in nanogranular matter the experimental results are corroborated by simulations.
Acknowledgements Funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – Project-ID 434434223 – SFB 1461.

H-6:L05 In-materia Adaptive Computing Devices based on Random-assembled Clusters Network
F. BORGHI, G. Nadalini, S. Bressan, P. Milani, CIMAINA and Dipartimento di Fisica, Università di Milano, Italy

Unconventional in-materia computing devices, which exploit the complexity and collective phenomena originating from various classes of physical substrates, have been proposed as alternative strategy to artificial neuromorphic systems based on components obtained by top-down lithographic technologies. In this contest, the employment of materials composed by a large number of non-linear nanoscale junctions obtained by the assembling of metallic nanoparticles have shown interesting non-linear electrical properties and complex resistive switching phenomena. Here we demonstrate the development and the integration of memristive device based on nanostructured gold able to integrate and process voltage pulse trains autonomously into a polymer matrix of polydimethylsiloxane (PDMS), which is one of the most common flexible and stretchable silicon elastomers. The composite Au/PDMS device preserves the electrical conductivity even in stress conditions and adapts it in response to the different mechanical stimuli the system receives in a non-trivial manner. Even more interestingly, the memristive flexible and stretchable composite device processes and integrates adaptively train pulses according to the external mechanical stimuli applied to the device.

Session H-7 New developments in characterization methods for materials and devices

H-7:IL02 Photoelectron Spectroscopy of Functional Oxides for Novel Electronic Device Concepts
M. Müller, University of Konstanz, Germany Complex Materials Group, Konstanz, Germany

If everything is about ‘the nothing’, is nothingness everything? In oxide materials, the „nothing” is mighty: The key to shaping the physical functionalities lies in the tunability of their oxygen sublattice and, in particular, their defect structure. Initiating or suppressing oxygen migration or redox reactions across interfaces is an effective means to create targeted oxide phases and unique functionalities. In dielectric oxides, defects like metal dopants and oxygen vacancies have been identified to induce ferroelectricity [1]. A well-balanced presence of vacancies can stabilize a robust ferroelectric state over thousands of electric field switching cycles [2]. Here, I will give an overview on defect detection and interface chemistry, and the novel possibilities of hard X-ray photoelectron spectroscopy for device optimization and for in operando experiments [3]. The focus will be on results obtained with HfO2, HZO and AlScN ferroelectric materials integrated into real-world capacitors.
[1] Th. Szyjka et al, ACS Applied Electronic Materials 2, 10, 3152 (2020)
[2] L. Baumgarten et al, Adv. Funct. Mater. in print (10/2023)
[3] M. Müller et al, J. Vac. Sci. Technol. A 40, 013215 (2022)

H-7:IL03 Advanced Nanoscale Spectroscopic Investigation of Nanostructures for Single Photon Source
P. PRETE, IMM-CNR, Lecce, Italy

Quantum Science and Technologies promise an epochal turning point, defined as quantum revolution 3.0, in many technological sectors, including that of ICT (Quantum Computing and Quantum Communication). The development of nano-photonic devices that rely on the generation, manipulation and detection of single photons is crucial for the realization of future quantum computers, whose logic is based on quantum bits. These devices exploit quantum correlation states, which are realized in semiconductor quantum dot (QD) structures. Therefore, the study of new cutting-edge epitaxial nano-structures for the fabrication of single photon light sources is pivotal. One-dimensional hetero-structures (so-called QD-in-nanowires, QDNWs) through vapor-liquid-solid self-assembly or selective area epitaxy, are the ideal system to engineer custom-designed materials and optimize their properties and, consequently, the performance of the device. The talk report on the nano-spectroscopic/structural properties of MOVPE grown GaAs-AlGaAs core-shell and core-multishell NWs forming quantum well tubes (QWTs), along with GaAs-based QDNWs, investigated by using advanced nano-characterization tools, such as high spatial resolution and low-temperature cathodoluminescence spectroscopic imaging.

H-7:IL04 Progress on Tomographic Filaments Observation with Adaptive Scalpel Scanning Probe Microscopy
U. CELANO, School of Electrical, Computer & Energy Engineering, Arizona State University, Scottsdale, AZ, USA

Recent years have seen widespread development of filamentary-based resistive memories (RRAMs) in diverse fields such as data storage, neuromorphic computing, and edge computing. This has been accompanied by significant progress in our ability to observe and study the switching mechanisms of these devices, namely formation, rupture, and modulation of nanosized conductive filaments. Tomographic atomic force microscopy (T-AFM), also known as scalpel scanning probe microscopy (SPM), is a slice-and-view tomographic technique widely used to observe conductive filaments in RRAMs. Ten years since its introduction, scalpel SPM has made significant progress in the field of RRAM research. In this review, we will discuss the current state of the art, future developments, and a novel measurement scheme called adaptive scalpel SPM, which has the potential to enhance both the sensitivity and lateral resolution of three-dimensional tomographic imaging of confined volumes.

H-7:L05 Infrared Nanoimaging of Hydrogenated Perovskite Nickelate Memristive Devices
S. Gamage¹, S. Manna^{2, 3}, M. Zajac⁴, S. SLAC Hancock⁴, Q. Wang⁵, S. Singh¹, m. Ghafariasl¹, K. Yao⁴, T. Tiwald⁶, T.J. Park⁵, d. landau⁴, H. Wen², S. Sankaranarayanan^{2, 3}, P. Darancet^{2, 7}, S. Ramanathan^{5, 8}, Y. Abate¹, ¹University of Georgia, Department of Physics and Astronomy, Athens, GA, USA; ²Argonne National Laboratory; ³University of Illinois Chicago; ⁴University of Georgia; ⁵Purdue University; ⁶J A Woollam Co Inc; ⁷Northwestern Argonne Institute of Science and Engineering; ⁸Rutgers The State University of New Jersey, USA

Solid-state devices made from correlated oxides such as perovskite nickelates are promising for neuromorphic computing by mimicking biological synaptic function. However, comprehending dopant action at the nanoscale poses a formidable challenge to understanding the elementary mechanisms involved. Here, we perform operando infrared nanoimaging of hydrogen-doped correlated perovskite, neodymium nickel oxide (H-NdNiO3) devices and reveal how an applied field perturbs dopant distribution at the nanoscale. This perturbation leads to stripe phases of varying conductivity perpendicular to the applied field, which define the macroscale electrical characteristics of the devices. Hyperspectral nano-FTIR imaging in conjunction with density functional theory calculations unveil a real-space map of multiple vibrational states of H-NNO associated with OH stretching modes and their dependence on the dopant concentration. Moreover, the localization of excess charges induces an out-of-plane lattice expansion in NNO which was confirmed by in-situ - x-ray diffraction and creates a strain that acts as a barrier against further diffusion. Our results and the techniques presented here hold great potential to the rapidly growing field of memristors and neuromorphic devices wherein nanoscale ion motion

H-7:L06 Metrology of Ferroelectric HZO with STEM EBIC Imaging
B.C. REGAN, H.L. Chan, T. O'Neill, Y. Chen, UCLA, Los Angeles, CA, USA; S.S. Fields, J.F. Ihlefeld, University of Virginia, Charlottesville, VA, USA; W.A. Hubbard, NanoElectronic Imaging Inc., Los Angeles, CA, USA

Scanning transmission electron microscopy (STEM) is already a standard tool for characterizing materials properties, but the overwhelmingly dominant contribution to STEM contrast comes from intra-atomic (nuclear) electric fields. To visualize the much smaller inter-atomic fields, we employ STEM electron beam-induced current (EBIC) imaging. STEM EBIC imaging proves itself to be a powerful technique for measuring the key properties of ferroelectric thin films. At low magnification, it can map polarization across an entire device. At high magnification, it can quantitatively map both the coercive fields and the remanent fields as they vary from domain to domain. Applying STEM EBIC imaging to ferroelectric Hf0.5Zr0.5O2 (HZO) devices, we gain new insight into the causes underlying the width of the coercive field distribution. For instance, (a) misalignments between the crystal polarization axes and the applied field or (b) locally varying bias fields are both plausible mechanisms. STEM EBIC imaging rules out the former, identifies the latter as the root cause, and measures its magnitude. STEM EBIC imaging's ability to discriminate between specific, mechanistic models of property dispersion promises to accelerate ferroelectric materials development.

H-7:L07 Dead Samples Tell No Tales: STEM EBIC of PFIB-prepared Devices
W.A. HUBBARD, NanoElectronic Imaging, Los Angeles, CA, USA

The function of modern electronic components derives from the electronic properties of their constituent materials. Transmission electron microscopy (TEM) is the standard technique for imaging microelectronic devices, but preparing a sample for TEM imaging typically destroys a device’s electronic structure, and thus its function. Moreover, TEM is blind to this electronic structure. Here we present a sample-preparation-and-imaging methodology that addresses both of these challenges. First, with a plasma focused ion beam (PFIB) we prepare electron-transparent, electrically contacted cross section samples that preserve the original device function. The electrical contacts enable device biasing, as well as measurement of electron beam-induced current (EBIC), in the TEM. We then use scanning TEM (STEM) EBIC imaging to map conductivity and electric field in these devices, at high resolution, while they are operating. Transistors (GaN, AlGaAs HEMTs), optical sensors (HgCdTe, InGaAs IR detectors), and nonvolatile memory devices, including RRAM (HfO2, TaO2), PCRAM (GST) and ferroelectrics (HZO), can all be analyzed using this workflow. Here we will describe the sample preparation process and the electronic contrast mechanisms, with a special focus on nonvolatile memory materials.

Track HAvances in Materials and Devices Research for Digital, Neuromorphic and Unconventional Computing

Cimtec 2024

Track H
Avances in Materials and Devices Research for Digital, Neuromorphic and Unconventional Computing