IC-4 - 6th International Conference
Materials and Devices Technologies for Energy-efficient Neuromorphic and Unconventional Computing

ABSTRACTS

Session IC-4.A Advances in memory and memristive technologies for computing: materials, devices, advanced characterization and modelling

IC-4.A:IL01  A Novel ePCM Cell for Edge AI Applications
M. BALDO, STMicroelectronics, Technology R&D, Agrate Brianza, Italy

Memory data transfer is becoming the main limiting factor in the execution of AI loads at the edge. To overcome speed and power limitations imposed by the classical von Neumann architecture new computational paradigms are being proposed as possible alternatives. In-Memory-Computing (IMC) is one of the most promising emerging approaches thanks to the absence of data transfer between the memory and the computational units. In this work, a new Ge-GST embedded PCM cell architecture called Rheostatic, is presented. Such geometry was chosen to optimize the device performance for Neural Network acceleration through Analog-IMC means. Details on the integration flow and differences with standard Wall architecture are discussed. The cell dimensions are optimized to find an optimal fit for Analog-IMC applications. An extensive electrical characterization of conductance levels is performed to assess the cell proprieties and stability. Performance evaluation of different Neural Networks for image recognition are assessed through simulations based on the gathered data. The results demonstrate the superior performance of the Rheostatic cell when compared with standard Wall structure ePCM.

IC-4.A:IL02  Phase-Change Memory for In-Memory Computing
G.S. SYED, IBM Research – Europe, Zurich, Switzerland

In-memory computing (IMC) is arguably the most promising brain-inspired computational paradigm aimed at addressing the processor-memory divide in modern computing systems. This talk will elucidate how phase-change memory (PCM) is a leading technology for realizing analog IMC. The talk is organized to begin with a foundational overview of PCM-based IMC, followed by a review of the current state of the art in the design and fabrication of PCM devices and IMC chips, and finally, a survey of the broader application landscape that is being improved through PCM-based IMC. We will conclude by outlining promising, interdisciplinary research directions in this rapidly advancing field.

IC-4.A:IL03  Modelling of Phase Change Heterostructures for Neuromorphic Computing
R. MAZZARELLO, Department of Physics, Sapienza University of Rome, Rome, Italy; S. RITAROSSI, University of Roma Tre, Rome, Italy

Present-day computers based on the von Neumann architecture are becoming inadequate for data-centric applications, such as artificial intelligence and machine learning, due to the physical separation between processing and storage units. Neuromorphic hardware offers a novel approach inspired by the brain to overcome this limitation: its basic building blocks are artificial neurons, which can act both as computing and storage units. We have designed phase-change heterostructures (PCHs) for neuromorphic applications, consisting of alternately stacked nanolayers of phase-change materials (Sb, GeTe or Ge₂Sb₂Te₅) and confinement materials (TiTe₂). By applying suitable electrical pulses, phase-change materials undergo fast and reversible transitions between multiple states exhibiting resistivity contrast. However, variability and temporal resistance drift in conventional phase-change cells hinder their use in neuromorphic devices. We discuss how the introduction of interface layers in PCHs can mitigate both variability and drift, potentially enabling reliable iterative RESET and cumulative SET operations for high-performance neuro-inspired computing.

IC-4.A:L04  1D van der Waals Materials for Phase-Change Memory
YI SHUANG, YUJI SUTOU, Tohoku University, Sendai, Japan

We report the discovery and characterization of a novel phase transition in one-dimensional van der Waals (1D vdW) NbTe4. By slightly tuning Nb content via sputtering, NbTe4 thin films transform controllably from a monoclinic phase to a tetragonal phase. By observing the atomic images of the phase boundaries between monoclinic and tetragonal phases via TEM and HAADF-STEM, the lateral atomic chain shuffling and Te-Te bond reorganization within vdW gaps were confirmed. [1] Using this phase tunability, we fabricated phase-change memory (PCRAM) devices with NbTe4 films as active layer, demonstrating reversible switching between low-resistance tetragonal and high-resistance monoclinic phases under electrical pulses. [1] These results advance the understanding of polymorphic transitions in 1D vdW materials and establish NbTe4 as a promising candidate for PCRAM, enabling novel device architectures with enhanced functionality, energy efficiency, and scaling potential.
[1] Yi Shuang, Daisuke Ando, and Yuji Sutou. "Phase Engineering of a 1D van der Waals Thin Film." Advanced Functional Materials (2025): 2503094.

IC-4.A:IL05  Flexoelectric Modulation of Electroresistance in 2D van der Waals Semiconductors
A. GRUVERMAN, University of Nebraska-Lincoln, Lincoln, NE, USA

Strain engineering has been adopted as a powerful strategy to control the physical properties of materials. High strain tolerance of two-dimensional (2D) van der Waals (vdW) semiconductors makes them fertile ground for manipulation of their electronic properties via application of the elastic strain fields. Here, we investigate modulation of the electroresistance effect in the 2D vdW semiconductors MoS2 and In2Se3 under the mechanical stress induced by an atomic force microscope (AFM) tip. We show that the resistance of the 2D semiconductor junctions can be reversibly tuned by up to 4 orders of magnitude resulting from a combined effect of the tip-induced strain and strain gradient. While the strain effect results in a gradual reduction of the barrier height with the tip loading force increase, the induced strain gradient, which generates a flexoelectric field, modifies the barrier profile. In addition, it is shown that the tip-generated flexoelectric effect leads to significant enhancement of the photovoltaic effect in the 2D vdW semiconductors. A combination of the optical and mechanical stimuli facilitates reversible photomechanical tuning of resistance of the 2D vdW semiconductors and development of devices with an enhanced photovoltaic response.

IC-4.A:IL06  Magneto-ionic Synaptic Devices
L. HERRERA DIEZ, Centre de Nanosciences et de Nanotechnologies, CNRS, Université Paris-Saclay, Palaiseau, France

Magneto-ionics takes inspiration from memristor technologies and offers a path for controlling magnetic properties using ionics. Integrating ionic and spintronic technologies offers new degrees of freedom to design neuromorphic hardware with novel magnetic functionalities, alongside the established ionic analogue behavior. I will present a variety of material combinations and device designs allowing to generate multiple stable and electrically detectable magneto-ionic states. I will also demonstrate that magneto-ionic nanodevices can not only function as basic synaptic elements, using their capacity to encode multiple analogue states, but also enable new bioinspired functionalities. We show that in magneto-ionic synaptic elements, synaptic depression can be tuned using a magnetic field, allowing to dynamically control the linearity of the synaptic weight update. This functionality is reminiscent of neuromodulation, observed in biological systems, and neural network simulations reveal that a magnetically induced enhancement in weight-update linearity improves learning accuracy across a wide range of learning rates. These findings highlight the versatility and promise of magneto-ionic devices for developing multifunctional synaptic elements for neuromorphic hardware.

IC-4.A:L07  Spintronic Advantage of Molecular Spin-valves for Reinforcement Learning
C. BALDASSINI^1,2, R. LICATA^1,2, S. BOSE^1,2, M. PISTOIA^1,2, S. SANNA², I. BERGENTI¹, R. CECCHINI¹, V.A. DEDIU¹, L. GNOLI¹, P. GRAZIOSI¹, R. RAKSHIT¹, S. ROY¹, M. SINGH¹, A. RIMINUCCI¹, ¹Institute for the Study of Nanostructured Materials (CNR-ISMN), Bologna, Italy; ²Department of Physics and Astronomy, University of Bologna, Bologna, Italy

The development of neuromorphic devices is a pivotal step in the pursuit of low power artificial intelligence. In particular, molecular spin valves show great potential in neuromorphic applications. A synaptic analogue is one of the building blocks of this vision. We study the synaptic behaviour of molecular La_0.7Sr_0.3MnO₃/tris(8-hydroxyquinolinato)gallium/AlOx/Co spintronic devices. By arranging the devices in an N-crosspoint configuration and by emulating the synaptic weight with the total conductance, we can store multiple synaptic weights, and so multiple neural networks, on a single chip. Each device’s conductance can be tuned via voltage pulses and by switching the magnetization of the electrodes[1] [2]. In this way, distinct synaptic weights can be associated with different magnetic configurations [3]. To validate the synaptic architecture, we evaluated its accuracy in the simultaneous reproduction of arbitrary weights. The system demonstrated excellent reproduction fidelity up to a number N for an N-crosspoint synapse. Furthermore, we implemented a reinforcement learning task on a simple network. In this demonstration, an experimental dataset was used to emulate the physical device behavior. Each neural network was trained to solve a distinct benchmark control problem such as the cart-pole and mountain car tasks. In conclusion, our results demonstrate the strong potential of molecular resistive spin valves as fundamental building blocks for innovative neuromorphic architectures.
[1] Shumilin, A. et al. (2024). Glassy Synaptic Time Dynamics in Molecular La0.7Sr0.3MnO3/Gaq3/AlOx/Co Spintronic Crossbar Devices. Advanced Electronic Materials. https://doi.org/10.1002/aelm.202300887. [2] Riminucci, A. and Legenstein, R. (2019). Fast learning synapses with molecular spin valves via selective magnetic potentiation. https://doi.org/10.48550/arXiv.1903.08624. [3] Baldassini, C. (2025). Spintronic advantage in the training of a molecular cross-bar neural network.

IC-4.A:IL08  Strain Mediated Changes in Negative Differential Resistance in Manganite Thin Films
A. JAMAN, K.P. ROMPOTIS, I. BHADURI, T. BANERJEE, University of Groningen, Groningen, Netherlands

The Metal-Insulator transition (MIT) in ferromagnetic LSMO thin films is an example of a strongly coupled phase transition enabled by temperature. Control of this MIT transition by Joule heating and its manifestation as a negative differential resistance (NDR) is associated with the thermodynamics of competing phases. We study the impact on NDR due to changes in epitaxial strain by growing LSMO on different substrates. The energetics of different orbitals in such strained films, due to extrinsic and intrinsic entropic contributions is found to stabilize distinctive non-linear evolution of resistances when driven out-of-equilibrium with an electrical stimulus. I will discuss our findings and show how our work paves the way for utilizing NDR to understand the coexisting yet competing phase dynamics accelerating the design of non-CMOS computing primitives.

IC-4.A:IL09  Lead-Free Halide Perovskite Memristors for Neuromorphic Computing
C. MANDAR MHASKAR¹, S. ROY CHAUDHURI², A. ROY CHAUDHURI¹, ¹Materials Science Centre, Indian Institute of Technology Kharagpur, India; ²Department of Chemistry, Raja Narendralal Khan Women’s College, Midnapore, India

The slowdown of Moore’s law and the end of Dennard scaling are driving alternative computing paradigms, with neuromorphic systems offering energy-efficient, brain-inspired architectures. High-performance artificial synapses require materials combining stable resistive switching with rich plasticity. Halide perovskites are promising due to high ionic mobility and defect-tolerant lattices, but most research focuses on lead-based HPs, whose toxicity limits practical use. Lead-free alternatives are essential for sustainable devices. KCuCl₃, an all-inorganic, lead-free HP, exhibits multi-level and analog bipolar resistive switching via chloride vacancy migration. KCuCl₃ memristors achieve endurance >3100 cycles, retention >10⁴ s, and gradual, linear conductance modulation. Pulse experiments show diverse synaptic behaviors, including paired-pulse facilitation/depression, spike-rate and spike-timing-dependent plasticity, and associative learning. Neural network simulations based on experimental conductance updates reach ~98% (MNIST), ~90% (Fashion-MNIST), and ~80% (CIFAR-10). KCuCl₃ provides a sustainable, high-performance halide perovskite platform integrating multi-level switching, long-term reliability, and complex synaptic functionality enabling scalable, brain-inspired electronics.

IC-4.A:IL10  Understanding Effects of Bottom Electrode Materials on Ferroelectricity in (Hf,Zr)O2 thin films
MIN HYUK PARK, Department of Materials Science and Engineering & Inter-university Semiconductor Research Center, College of Engineering, Seoul National University, Seoul, Republic of Korea

HfO₂‑based films, especially Hf₀.₅Zr₀.₅O₂ (HZO), remain ferroelectric below 10 nm and are compatible with the complementary-metal-oxide-semiconductor. The polar o‑phase (Pca2₁) is stabilized by dopants, surface energy, annealing, and electrode confinement. TiN can cause wake‑up via oxygen vacancies resulting from scavenging effect; Mo/W/Ru supply oxygen during ALD/anneal, lowering vacancies and suppressing non‑ferroelectric phases. Electrode texture sets orientation and fields; α‑W and β‑W favor distinct o‑phase orientations. Ru also offers tunable oxygen (via atomic layer deposition temperature), high work function, and 3D compatibility. We compare Mo, W, and Ru. Mo increases o‑phase content, reduces vacancies, suppresses wake‑up, and boosts Pr. W: α‑W(110) raises remanent polarization (Pr~29.23 μC/cm²) and speed but more fatigue; β‑W(200) lowers Ec (~0.75 MV/cm) and improves low‑field; both endure ~10⁹ cycles. Ru: oxygen content tunes phase, polarization, leakage, and wake‑up; higher oxygen reduces vacancies/leakage but can increase monoclinic content and limit endurance via interfacial defects. Therefore, engineering electrode chemistry, texture, and oxygen activity tailors HZO phase stability, defects, and switching for optimized devices.

IC-4.A:L11  Deterministic Intermediate Polarization States for Ferroelectric Synaptic Transistors: A TiN/ Hf0.5Zr0.5O2/TiN Model Study
N. SIZYKH, M. SPIRIDONOV, A. KHANAS, A. ZENKEVICH, Moscow Institute of Physics and Technology, Dolgoprudny, Moscow region, Russia

Neuromorphic computing is emerging to overcome the von-Neumann memory bottleneck and the energy/latency costs of transferring data between memory and processors. Real-time, edge-scale AI further demands low-power, massively parallel hardware. These needs point to development of novel analog devices, especially with synaptic functionality. We treat TiN/Hf0.5Zr0.5O2/TiN capacitors as a model system for Metal-Ferroelectric-Metal-FET devices and demonstrate deterministic setting and stabilization of intermediate polarization states suitable for synaptic weights storage. We have implemented a searching procedure to find the voltage pulse amplitudes to write exact intermediate states. Thereby, we reproducibly set eight well-separated levels (2Pr=4–28 uC/cm2). For this, we introduce a feedback-assisted continuous cycling procedure to track state evolution under repeated operation. Crucially, it appears that to rewrite the preceding and subsequent polarization levels after storing information at a given level has to be explicitly verified. To probe this rewriting sensitivity, we modified the conventional retention testing protocol. Our results outline read/write pulses shaping strategies for MFMFET-based synapses, while highlighting open questions on multilevel state accessibility.

IC-4.A:L12  The Effect of Chargeable Defects at Interfaces on the Functional Properties of M/HZO/M (M=TiN, W) Ferroelectric Memory Capacitors
N. SIZYKH, M. SPIRIDONOV, A. KHANAS, A. ZENKEVICH, Moscow Institute of Physics and Technology (National Research University), Dolgoprudny, Moscow region, Russia

Increasing demands for mass data storage and emerging computing paradigms are driving the need for energy-efficient, fast, non-volatile memory devices. Ferroelectric Hf0.5Zr0.5O2 (HZO) is a leading candidate owing to its excellent CMOS compatibility. In this work, we correlate the switching-pulse-amplitude dependence of reliability data for TiN/HZO/TiN and W/HZO/W memory capacitors with the energy distribution of chargeable defects, presumably oxygen vacancies, in HZO band gap, measured via the Step-Recovery with Multi-Pulse Test (SRMPT) technique. We further utilize SRMPT-derived information on the energy landscape of defects to perform consistent calculations of the evolution of the defects spatial distribution as well as the associated internal electric fields across the ferroelectric HZO layer under applied voltage pulses. By aligning experimental data with time-dependent charged defects dynamics and local electric fields, we establish the link between the interface defects fingerprint and reliability outcome of the ferroelectric capacitors, which are the functional basis for non-volatile memory as well as synaptic devices for neuromorphic computing hardware.

IC-4.A:IL13  Yttrium Oxide: An Analog OxRAM Material
E. PIROS¹, P. SCHREYER¹, T. KIM¹, Y. LI¹, Y. DUAN¹, A. ARZUMANOV¹, L. MOLINA-LUNA², J. GEHRUNGER³, L. MAYRHOFER³, C. HOCHBERGER³, T. OSTER⁴, K. HOFMANN⁴, F. AGUIRRE^5,6, J. SUÑE⁵, E. MIRANDA⁵, L. ALFF¹, ¹ATFT Div., Inst. Mat. Sci., Technische Universität Darmstadt, Darmstadt, Germany; ²AEM Div., Inst. Mat. Sci., Technische Universität Darmstadt, Darmstadt, Germany; ³Computer Systems, Dep. El. Eng. and Inf. Techn., Technische Universität Darmstadt, Darmstadt, Germany; ⁴Integrated Electronic Systems Lab, Dep. El. Eng. and Inf. Techn., Technische Universität Darmstadt, Darmstadt, Germany; ⁵Departament d'Enginyeria Electrònica, Universitat Autònoma de Barcelona, Spain; ⁶Intrinsic Semiconductor Technologies, Ltd., Buckinghamshire, UK

In OxRAM devices, the properties of the switching medium strongly affect switching performance. Among transition metal oxides, yttrium oxide stands out for its stability and high intrinsic anion vacancy concentration arising from its cubic bixbyite structure. Yttria-based memristors thus show excellent retention under heat-accelerated aging [1], high stability at elevated temperatures [2], and reduced read noise variability [3]. Through oxygen engineering, we have isolated the cubic phase of yttrium oxide, achieving analog switching during both the set and reset processes [4]. The intrinsic vacancy network lowers energy barriers for filament formation, reducing switching voltage and power consumption. Increasing oxygen deficiency promotes controlled, field-driven switching, yielding stable intermediate states with nonlinear conductance quantization [5]. Our latest results demonstrate superior temporal stability over 1 trillion readouts, even at elevated temperatures, across all resistive states—an important step toward reliable device operation.
[1] Semicond. Sci. Technol. 34(7), 075008 (2019). [2] Appl. Phys. Lett. 117, 013504 (2020). [3] Phys. Rev. Appl. 14, 034029 (2020). [4] Adv. Electron. Mater. 6, 2000439 (2020). [5] Sci. Rep. 14, 1122 (2024).

IC-4.A:IL14  Resistance Switching in SiOx Layers: From Demonstration to Commercialisation
A. MEHONIC, Department of Electronic & Electrical Engineering, University College London, London, UK

The rapid growth of computing has created unsustainable energy demands, with projections suggesting that by 2035 computing could outpace realistic energy supply. Since memory dominates energy use in modern systems, breakthroughs are essential. Over the past 15 years, my group has advanced resistive RAM (ReRAM) from fundamental studies of resistance switching in amorphous oxides to a manufacturable technology addressing both non-volatile storage and energy-efficient computing. This talk highlights insights into switching physics, performance optimisation, and reliable device design, culminating in a roadmap for ReRAM’s role in future memory-centric systems and lessons from translating academic research into industrial innovation.

IC-4.A:IL15  Advances in Modelling of Memristive Devices
S. MENZEL, Forschungszentrum Jülich, Peter-Grünberg-Institut (PGI-7), Jülich, Germany

Memristive devices based on the filamentary valence-change mechanism have recently entered the market of embedded non-volatile memories, i.e., as so-called RRAM. Due to the filamentary nature, Joule heating plays a crucial role for the dynamics of the switching process. The heating efficiency depends on several parameters such as the size of the filament, the thermal conductivity of the materials, and the interfaces between the dissimilar materials. In a static picture all these parameters are reduced to one parameter, i.e., thermal resistance. To design RRAM cells with improved heating efficiency, however, a better physical understanding is required. In this presentation, first the influence of different parameters on the heating efficiency is discussed. One important – and often overlooked – aspect is the thermal boundary resistance at the interfaces. Including these effects lead to a better understanding of Joule heating and improved design guidelines. Moreover, it will be shown that the thermal resistance of the device may change during switching. Based on these insights an improved compact model, i.e., the JART VCM Rth model is presented. By including the state-dependence of thermal resistance, it can model accurately multilevel programming.

IC-4.A:L16  Towards atomistic understanding of digital and analog filamentary switching
L. ALFF, TU Darmstadt, Darmstadt, Germany

While digital applications of memristive devices are more and more taking off, the use of multi-state or analog devices is still in an emerging state. A key aspect is to understand which material modifications drive the transition from digital two states to multiple states with suffcient stabiliy. Here we discuss several material strategies to achieve multiple resistive states by oxygen engineering and doping of CMOS compatible oxides, in this case lanthanum substitution in hafnium oxide. We also address recent progress in observing memristive model structures operando during a transmission electron microscope experiment on a focused ion beam cut lamella.

IC-4.A:L17  Resistive Switching Characteristics of Single, Bi, and Tri-layered Oxides in Memristive Devices
K. DORAI SWAMY REDDY¹, E. PÉREZ^1,2, CH. WENGER^1,2, ¹IHP – Leibniz Institute for High Performance Microelectronics, Frankfurt Oder, Brandenburg, Germany; ²Brandenburgische Technische Universität (BTU) Cottbus-Senftenberg, Cottbus, Germany

Memristive devices are promising for neuromorphic computing due to their resistive switching properties. These properties are significantly influenced by the dielectric materials and their spatial arrangement in the device stack, determining the formation of various interfaces. Consequently, a systematic study of various dielectric layers and their configurations is crucial to understanding how such arrangements impact the switching characteristics. In this study, Metal-Insulator-Metal devices with an area of 600 x 600 nm2, fabricated in IHP’s 130 nm CMOS process line, are used to examine the switching characteristics of single-layer (HfO2 doped with Aluminum), bi-layer (Al2O3, HfO2), and tri-layer (Al2O3, HfO2, TiO2) dielectric stacks. Additionally, the tri-layer devices are analyzed with TiO2 positioned either at the bottom or at the top of the stack. Compared to single-layer devices, the bi-layer devices exhibit improved gradual switching during the reset process but suffer from a smaller memory window. Among the tri-layer configurations, devices incorporating TiO2 as the bottom dielectric layer exhibit the largest memory window, whereas those with TiO2 positioned at the top display the most pronounced gradual switching characteristics during both the Set and Reset processes.

IC-4.A:IL18  How to Control the Behaviour and Functionalities of Nanoionic-enhanced Memristive Devices
I. VALOV, Research Centre Juelich, Wilhelm-Johnen-Str., Juelich, Germany; Institute for Electrochemistry and Energy Systems, BAS, Sofia, Bulgaria

Memristive devices have significantly developed in the last decade, expanding their application horizon much beyond memory applications. Especially important is their functionalities as artificial neurons and synapses, making them promising building units for the next generation bio-inspired neuromorphic hardware. In this presentation, different ways to controll and tune memristive behaviour will be discussed as well as new aspects on materials design and its influence on the physicochemical processes and resulting functionalities of both ECM (CBRAM) and VCM (OxRAM) devices. The magnitude of the applied voltage, materials and thicknesses of the layers appear of special importance and as well the combination combination of thicknesses of all involved layers. The selection of different materials is changing the electrochemical nanoionics processes and as well the performance of the memrsitors. For reliable operation conditions and performance of the device stack should be considered as a whole, and materials used in the stack and their thicknesses should be coordinated and harmonised.

IC-4.A:IL19  Disentangling Electronic and Ionic Behaviour in Redox-based Memristive Devices
J. HELLWIG, C. WITTBERG, D. SPITHOURIS, C. FUNCK, R. DITTMANN, P. GRÜNBERG, Institute 7, Forschungszentrum Jülich GmbH, Jülich, Germany

The physical origin of resistive switching and the relaxation of the low resistive state of redox-based memristive devices is often debated. This work evaluates the contributions of electrons and ions to these measurements on the example of memristive SrTiO3 devices using a novel Deep Level Transient Spectroscopy -like approach which has not been presented for memristive devices before. The current response during and after a voltage pulse is investigated in a temperature range between 15K and 450 K. Current transients arising from electron capture and emission on oxygen vacancy related defects, were only found in negative polarity pronounced in distinct temperature ranges between 50 K and 120 K, and between 280 K and 400 K. These activation ranges match our calculations based on Shockley-Read-Hall dynamics using results of previous spectroscopic measurements to determine the energy positions of the traps. The current transients are attributed to a modulation of the tunneling current induced by an electron capture process as a result of the dominant quasi-Fermi level formation in reverse polarity. Our unique measurement approach demonstrates the strong interaction between the tunnelling current and the electronic structure of these devices for the first time. This methodology is a powerful method to separate electronic and ionic effects in memristive devices in particular in their transient behavior.

IC-4.A:IL20  Nano-Investigation of Energy Materials with Light
G. DI MARTINO, University of Oxford, Oxford, UK

The quest for efficient energy materials is crucial to meeting global demand for sustainable energy solutions. This talk explores the challenges and opportunities in studying these materials, particularly the dynamic behaviours that govern performance. Atomic-level understanding is vital but difficult due to limitations in observing real-time nanoscale processes. I will present how advanced optical characterization in my research addresses these challenges, offering insights into material behaviour under realistic conditions. Techniques like ultrafast spectroscopy, near-field microscopy, and single-particle imaging allow us to track defects, phase transitions, and structural changes in real time. I will highlight studies on oxides, 2D materials, ferroelectrics, and magnetic systems. Key examples include oxygen vacancy migration in memristors (Nature Electronics, 2020), optical probing of MoS₂ switching (Adv. Mater., 2022), and phase changes in HfZrO ferroelectrics (Adv. Funct. Mater., 2023).

IC-4.A:L21  N-Type Behavior from a P-Type Dopant? On Vacancies and Charge Compensation Mechanisms in HfO2
O. REHM¹, L. BAUMGARTEN², F. WUNDERWALD³, A. FUHRBERG¹, P.M. DUERING¹, A. GLOSKOVSKII⁴, C. SCHLUETER⁴, T. MIKOLAJICK^3,5, U. SCHROEDER³, M. MUELLER¹, ¹Fachbereich Physik, Universität Konstanz, Konstanz, Germany; ²Forschungszentrum Jülich GmbH, Peter Grünberg Institut (PGI-6), Jülich, Germany; ³NaMLab gGmbH, Dresden, Germany; ⁴Deutsches Elektronen-Synchrotron, Hamburg, Germany; ⁵Technische Universitat Dresden, Dresden, Germany

Contrary to the anticipated p-type doping behaviour of HfO₂ when doped with trivalent Y, our experimental results demonstrate an n-type doping behaviour, as evidenced by HAXPES Fermi level shifts. The analysis of all possible charge compensation mechanisms in oxide insulators yields a coherent explanation: The complex behaviour of the Fermi level arises, because trivalent Y doping of HfO2 leads to the creation of additional vacancies for the compensation of doping charges. This phenomenon is beneficial in compensating for the charge of residual vacancies, as created by lattice strain or oxygen deficiency in the sample growth, as long as the Fermi level remains within the allowed doping range. The analysis of the possible charge compensation paths also provides valuable insights into the quantification of oxygen vacancies: Despite its frequent utilisation, the Hf3+ signature in photoelectron spectroscopy spectra is not a reliable quantitative indicator of oxygen vacancies. In the case of heterovalent doping, the appearance of Hf3+ is related only to one of at least four possible charge compensation paths. This quantification is only possible in pure HfO₂ and for isovalent doping, for example by Zr (HZO).
[1] O. Rehm et al. accepted in Advanced Physics Research (2025).

IC-4.A:L22  Deconstructing the Interfacial Origins of a Multifunctional High-Mobility 2DEG
H. COX, D. RUBI, M. AHMADI, S. SANJAY, M. SAROTT, A.C. GARCÍA CASTRO, J. BAAS, K. CHERKAOUI, B. NOHEDA, University of Groningen, Groningen, Netherlands

The all-perovskite LaScO3/BaSnO3 interface hosts a high-mobility 2DEG ideal for transparent electronics. We engineer this system as a multifunctional platform whose conducting channel is tuneable over three orders of magnitude via oxygen vacancy control. Building on this tuneability, the system exhibits robust memristive switching driven by ionic motion and a persistent optical response with an unusual positive-to-negative photoconductivity crossover in the infrared. To understand the origins of these properties, we deconstruct the 2DEG's confinement mechanism at the atomic scale. Using iDPC-STEM to map local polarity, we directly visualise a strong structural polarisation stabilised in the BSO at the interface, revealing a uniform 'head-to-tail' profile. This provides direct evidence of the interface's polar nature but challenges polar catastrophe models. Our ongoing work, which will be presented, correlates these atomic-scale structural maps with oxygen vacancy density to disentangle the competing roles of strain and polar fields, providing a framework for designing advanced reconfigurable electronic and optoelectronic devices.

IC-4.A:L23  Resistive Switching Devices Using Novel 2D Materials
M. GRÁCIO, H. TEIXEIRA, C. DIAS, J. VENTURA, IFIMUP, Departamento de Fisica e Astronomia, Faculdade de Ciências, Universidade do Porto, Porto, Portugal

The integration of 2D materials on resistive switching (RS) devices has shown good neuromorphic capabilities and performance [1]. MXenes (Ti3C2TX) have excellent electrical/chemical properties, with RS ability [2]. Their integration into polymeric matrices, such as Polyvinylidene fluoride (PVDF), can enhance their electrochemical and thermal stability [3]. We fabricated structures based on MXenes composites and obtained bipolar RS by inserting 3 wt.% Ti3C2TX in the PVDF. The composite was synthesized by dispersion of the 2D flakes in the polymer, using dimethylformamide (DMF) as solvent, under magnetic stirring. The composite was deposited on the bottom electrode (BE) through spin coating. We used Ag (top) and W, ITO (BE) as electrode materials. RS is based in filament of Ag ions formation/rupture. The structures exhibit low switching voltage and ON/OFF resistance ratio around 10, and present endurance/retention. XRD and Raman analysis shows no oxidation of MXenes in the polymer up to 10 months after fabrication. These results provide a path to develop resilient and flexible memristors, towards non-volatile memories and neuromorphic systems.
[1] D. B. Strukov et al., Nature, 453, 80–83, 2008. [2] X. Yan et al, Small, 15(25), 2019. [3] M. Grácio et al., Polymers, 17(10), 1309, 20.

IC-4.A:L24  Leveraging Liquid-based Memristors for Neuromorphic Computing
A.V. SILVA¹, A.T.S.C. BRANDÃO², C.M. PEREIRA², J. VENTURA¹, C. DIAS¹, ¹IFIMUP, Departamento de Física e Astronomia, Faculdade de Ciências, Universidade do Porto, Portugal; ²CIQUP, Departamento de Química e Bioquímica, Faculdade de Ciências, Universidade do Porto, Portugal

As computing components continue to miniaturize, computing paradigms are evolving to biologically inspired information processing architectures. Neuromorphic systems replicate the behavior of neural networks with dynamic, adaptive components replicating synaptic function. To this end, resistive switching devices have been one promising route toward artificial synapses with high efficiency. Here, we explore the same behavior in liquid systems where ionic mobility governs conductivity and enables reconfigurable behavior. The liquid devices studied, based on copper salts, exhibit low-voltage operation, stable cycling, and key neuromorphic functionality, such as potentiation and depression, and temporal learning effects, such as PPF and STDP. By modulating solvent composition, the devices exhibit volatile and non-volatile state changes and maintain functionality across varied timescales. Their adaptability highlights liquid-state architectures as energy-efficient and multifunctional building blocks for future neuromorphic and bio-interfacing applications.

IC-4.A:IL25  Toward a Scalable Oscillatory Neural Network Architecture Based on VO₂ Neurons and Analog ReRAM Coupling Elements
V. BRAGAGLIA, IBM Research - Zurich, Rüschlikon, Switzerland

Oscillatory neural networks (ONNs), which encode information in the relative phase of coupled nonlinear oscillators, offer a neuromorphic alternative to von Neumann architectures by enabling parallel, energy-efficient computation. Vanadium dioxide (VO₂) oscillators serve as CMOS-compatible neuron elements through self-sustained oscillations driven by the insulator–metal transition. To achieve scalable and programmable ONN connectivity, we integrate multilevel resistive random-access memory (ReRAM) as tunable coupling elements. Bilayer TiN/CMO/HfOₓ/TiN ReRAM devices provide stable, low-power, and compact coupling suitable for adaptive network configurations. Design considerations such as voltage-induced state drift and nonlinear resistance response are addressed to ensure reliable operation. This work presents a ReRAM-coupled VO₂ ONN architecture and experimental validation of key building blocks toward energy-efficient, hardware-based neuromorphic computing.

IC-4.A:L26  Ultra High Frequency Oscillators with Nanoscale VO2 Memristors
Z. POLLNER, T.N. TÖRÖK, L. PÓSA, S.W. SCHMID, Z. BALOGH, A. HALBRITTER, Department of Physics, Budapest University of Technology and Economics, Hungary; M. CSONTOS, J. LEUTHOLD, Institute of Electromagnetic Fields, ETH Zurich, Switzerland; A. BÜKKFEJES, Emerson - Test and Measurement (NI), Hungary; H. KIM, A. PIQUÉ, Naval Research Laboratory, USA; J. VOLK, Institute of Technical Physics and Materials Science, Centre for Energy Research, Hungary

Oscillating neural networks (ONNs) are promising candidates for a new computational paradigm, where optimization problems are solved through the synchronization of coupled oscillators. Nanoscale VO2 Mott memristors [1] are promising building blocks for such ONNs. Until now, however, not only the maximum frequency of VO2 based ONNs, but also the maximum frequency of individual VO2 oscillators has been severely limited, which has restricted their efficient and energy-saving use. Here, we show how the oscillating frequency can be increased beyond the 100 MHz range by optimizing the sample layout and circuit layout, while maintaining an endurance of 10^13 cycles [2]. The limiting factors of the oscillation frequencies are studied with circuit-level and physics-based electrothermal simulations. Furthermore , we investigate how switching times slow down under oscillator conditions compared to the fastest switching achieved with single pulses [3]. These results pave the way towards the realization of ultra-fast and energy-efficient VO2-based oscillating neural networks.
[1] Pósa, L. et al. ACS Appl. Nano Mat. 6, 9137–9147 (2023). [2] Pollner, Zs. et al. Adv. Electron. Mater. https://doi.org/10.1002/aelm.202500433 (2025). [3] Schmid, S. et al., ACS Nano, 18, 21966–21974 (2024).

IC-4.A:L27  Toward GHz Operation of Oscillating Neural Networks
Z. POLLNER^1,2, T.N. TÖRÖK^1,3, L. PÓSA¹, Z. BALOGH^1,2, A. HALBRITTER^1,2, J. VOLK³, J. LEUTHOLD⁴, M. CSONTOS⁴, ¹Department of Physics, Budapest University of Technology and Economics, Budapest, Hungary; ²HUN-REN–BME Condensed Matter Research Group, Budapest University of Technology and Economics, Budapest, Hungary; ³Institute of Technical Physics and Materials Science, Centre for Energy Research, Budapest, Hungary; ⁴Institute of Electromagnetic Fields, ETH Zurich, Zürich, Switzerland

Phase-encoded oscillating neural networks (ONNs) realized with capacitively coupled simple circuits of Mott memristors have been demonstrated recently to solve complex optimization problems such as graph-coloring and max-cut [1]. In essence, the problem is encoded in the couplings of the oscillator network. The solution, worked out exclusively by the physics of the neural network hardware, is represented in the synchronized phases of the oscillators. To exploit the full potential in operation speed and energy-efficiency of such ONNs, in our recent work we have explored the pathways to optimize the device layout and circuit environment of the VO2 Mott memristors. As a result, the operating frequency was increased from the MHz range to 170 MHz [2], using devices whose active area was limited by the resolution of e-beam lithography. Here, we present an electromigration-based patterning technique that further reduces the active area to truly nanometer dimensions. This, combined with the on-chip integration of the oscillator circuits, facilitates stable oscillations up to 0.65 GHz, enabling energy efficient operation approaching the clock frequencies of conventional computing.
[1] O. Maher et al. Nat. Comms. 15, 3334 (2024). [2] Z. Pollner et al. Adv. Electron. Mater. e00433 (2025).

IC-4.A:IL28  Realizing Neuron-synapse Integration with Vanadium Oxide Based Memristors for Multiple Neuronal Spiking Functionalities
JEN-SUE CHEN, ZIH-SIAO LIAO, SHENG-JIE HONG, LI-CHUNG SHIH, SHUAI-MING CHEN, KAI-SHIN HSU, CHI-CHEIN CHEN, Department of Materials Science and Engineering, National Cheng Kung University Tainan, Taiwan

Threshold switching memristors (TSM) and resistive switching memristors (RSM) have been employed to mimic artificial neurons and synapses, respectively. However, the integration of both remains a significant challenge due to the distinct material systems required. In this work, we first introduce a hybrid memristor-based neuromorphic system comprising a volatile TSM as the spiking neuron and a non-volatile RSM as the synapse. The neuron circuit realizes a leaky integrate-and-fire (LIF) neuron model with tunable synaptic weights and temporal encoding functions. In addition, by connecting multiple RSMs in parallel, we demonstrate that the neuronal membrane potential dynamics is jointly governed by input strength and timing, thereby enabling precise modulation of spike generation. Secondly, we present a dual-mode TSM that achieves intrinsic neuronal plasticity (INP) through resistance-dependent threshold modulation, enabled by the coupling of conductive filament formation and Mott transition. Excitatory and inhibitory pulses independently modulate spiking frequency, time-to-first-spike, and LIF behavior without altering synaptic input. These findings offer a compact and scalable solution for the next-generation energy-efficient neuromorphic hardware implementation.

IC-4.A:L29  Cryogenic Endurance of HfO₂-Based 1T1R RRAM for Quantum-Compatible Memory
E. PEREZ-BOSCH QUESADA, A. MISTRONI, K.D.S. REDDY, F. REICHMANN, C. WENGER, E. PEREZ, IHP - Leibniz Institute for High Performance Microelectronics, Frankfurt (Oder), Germany; A. CANTUDO, J.B. ROLDAN, Department of Electronics and Computer Technology, University of Granada, Granada, Spain; R. JIA, Micro- and Nanosystems Technology, Technical University of Munich, Munich, Germany; H. CASTAN, S. DUEÑAS, Department of Electronics, University of Valladolid, Valladolid, Spain; C. WENGER, E. PEREZ, BTU Cottbus-Senftenberg, Cottbus, Germany

Cryogenic non-volatile memory units are essential for superconducting single-flux and quantum computing systems. However, the lack of compatible memory technologies operating below 4 K limits the development of scalable architectures. In previous work, we demonstrated resistive switching in HfO₂-based Resistive Random-Access Memory (RRAM) integrated in a 1-Transistor-1-Resistor (1T1R) structure at temperatures down to 1.5 K. Yet, no endurance tests have been reported at such cryogenic conditions. Building on these results, we present a comprehensive overview of the electrical characteristics of HfO₂-based 1T1R RRAM devices from room temperature down to 1.5 K. Special attention is given to how the forming operation temperature affects endurance behavior at cryogenic conditions. We present the first endurance study at 1.5 K, where devices formed at 300 K and 1.5 K were subjected to 1000 DC reset/set cycles using the Multilevel-Cell (MLC) approach at the minimum temperature. A multivariate coefficient of variation analysis revealed that cryogenic forming and switching reduce reset threshold variation while increasing variation during set transitions to the most conductive level. This study advances the understanding and development of next-generation cryogenic memory technologies.

IC-4.A:L30  Impact of NMOS Transfer Characteristics on the Electrical Behavior of 1T1R Structure
T. RIZZI, M. UHLMANN, K.D.S. REDDY, E. P-B. QUESADA, C. WENGER, E. PÉREZ, IHP - Leibniz Institute for High Performance Microelectronics, Frankfurt (Oder), Germany; T. ZANOTTI, F. M. PUGLISI, Dipartimento di Ingegneria “Enzo Ferrari” Università di Modena e Reggio Emilia, Modena, Italy; C. WENGER, E. PÉREZ, BTU Cottbus-Senftenberg, Cottbus, Germany

Resistive Random-Access Memory (RRAM) devices are commonly integrated in a 1-transistor–1-resistor (1T1R) structure, where an NMOS transistor is used to set a compliance current during SET operations to enable accurate multi-level programming. The transistor transfer characteristics influence the 1T1R electrical response and, consequently, the memory cell resistance. This work presents an experimental study on the impact of NMOS transfer behavior during programming and read operations of 1T1R structures. The tested devices combine TiN/Al:HfO₂/Ti/TiN RRAM with NMOS transistors of different sizes fabricated in a 130 nm CMOS technology. The intermediate node between RRAM and transistor is measured to capture the voltage drop dynamics. Moreover, a mismatch is observed between resistive states obtained from programming and reading operations, attributed to gate-voltage-dependent variations of compliance current and parasitic channel resistance. The findings are validated via circuit-level simulations using foundry-provided transistor model and the UNIMORE physics-based RRAM compact model. These results clarify the influence of transistor transfer characteristics on 1T1R operations and support the definition of optimal programming conditions.

IC-4.A:L31  Effect of Residual Ion Drift during Programming of CMOS-Integrated Nanoscale HfO2-based Memristive Devices
S. HOFFMANN-EIFERT, O. ARTNER, F. CUEPPERS, SEOKKI SON, XIAOHUA LIU, S. WIEFELS, S. MENZEL, Forschungszentrum Jülich GmbH, Peter Grünberg Institute for Electronic Materials (PGI 7), Jülich, Germany

Redox-based resistive random-access memory (ReRAM) are key enabling elements for neuromorphic applications. In combination with a transistor that acts as a selector and current limiter, an array of 1-transistor-1-ReRAM elements supports energy-efficient computing. High integration density requires small device size and multi-level programming. Both properties have been demonstrated by HfO2 ReRAM devices. In the ideal 1T1R element, the programmed low resistance state (LRS) should be defined by the set voltage and the gate current. However, it has been found that the read conductivities GLRS,read exceed the programmed values GLRS,prog. Here, we show that this correlates to a systematic intrinsic effect [1] that occurs in addition to the stochastic nature [2] of the set process. Investigating the transient switching behaviour of HfO2 devices integrated in a 1T1R configuration [3] it is found that the voltage on the ReRAM cell, which describes the internal state, shows a continuous decrease as a function of time under constant load. This behaviour, which is beyond the universal switching law, is discussed in view of the physical compact model [4].
[1] O. Artner et al., prep. [2] F. Cueppers et al., APL Mater, 7, 091105, 2019. [3] X. Liu et al, IEEE TED, 2024. [4] S. Seon et al, prep.

IC-4.A:L32  Comprehensive Noise Diagnostics of Memristive Systems
Z. BALOGH, B. SÁNTA, A. HALBRITTER, Department of Physics, Institute of Physics, Budapest University of Technology and Economics, Budapest, Hungary and MTA-BME Condensed Matter Research Group, Budapest, Hungary; T.N. TÖRÖK, Department of Physics, Institute of Physics, Budapest University of Technology and Economics, Budapest, Hungary and Institute of Technical Physics and Materials Science, HUN-REN Centre for Energy Research, Budapest, Hungary; S.W. SCHMID, Department of Physics, Institute of Physics, Budapest University of Technology and Economics, Budapest, Hungary and Experimental Physics V, Center for Electronic Correlations and Magnetism, University of Augsburg, Augsburg, Germany

In conventional electrical engineering, noise, i.e. the fluctuation of the measured current around its expected value, is regarded as a disturbance during device operation, since elevated noise levels can lead to erroneous or unstable outputs. Thus, understanding noise phenomena, revealing their physical origins, and scaling with the device parameters or developing effective suppression strategies are of high technological relevance. It has already been demonstrated that resistive switching memories exhibit pronounced 1/f-type noise, with properties depending on device operation principles and the structural characteristics or the resistance states of a given device. This variability underpins the importance of systematic and extensive noise characterization. In this presentation, I show the comprehensive noise diagnostics of various memristive systems (such as filamentary phase change or tunnel-like systems) focusing on the resistance and voltage scaling, the device-to-device and cycle-to-cycle reproducibility of their noise levels. These results not only contribute to the fundamental understanding of the physical processes governing memristive behavior but also help to find optimized devices for different neuromorphic and advanced computing applications.

Session IC-4.B1 Memristive devices for neuromorphic and unconventional computing: devices, modelling, applications

IC-4.B1:IL33  Design and Implementation of Memristive Locally Coupled Sensor-processor Systems: Recent Results
R. TETZLAFF¹, A. DEMIRKOL¹, V. NTINAS², C. YU¹, D. PROUSALIS¹, I. MESSARIS¹, A. ASCOLI³, LEON CHUA⁴, ¹Institute of Circuits and Systems, TU Dresden, Dresden, Germany; ²Department of Electronic Systems, Aalborg University, Denmark; ³Department of Electronics and Telecommunications of Politecnico di Torino, Turin, Italy; ⁴Department of Electrical Engineering and Computer Sciences, University of California Berkeley, Berkeley, CA, USA

Currently, there is a highly dynamic transformation towards AI-driven technologies, which are to be integrated into a rapidly increasing number of everyday applications, leading to an unprecedented demand for computing power. Not only are computer systems struggling to keep pace with this demand, but the collection and processing of large amounts of sensor data cannot be carried out in its current form and, in particular, is insufficient in terms of robustness and reliability in complex scenarios. Memristive analog computing arrays are increasingly coming to the foreground. Directly linked to sensors, these so-called autonomous, AI-based sensor-processor systems also enable complex data processing on the sensor, eliminating the need to transfer data to a central computer or cloud. Current work is focusing on locally coupled nonlinear dynamic systems, known as Cellular Neural Networks (CellNN), which have demonstrated significantly higher performance than classic Convolutional Neural Networks in recent studies. In particular, locally active memristive networks with suitable parameter selection can give rise to dynamics at the edge of chaos, which can be used to solve difficult combinatorial problems. Analog memristive CellNN [1] computing arrays as bio-inspired systems have enormous potential when implemented with sensors, i.e., as sensor processor systems that are mathematically universal in the internal processing of data from their own sensors, typically under real-time conditions. Due to their local coupling structure of elementary nonlinear dynamic systems, known as cells, they are capable of representing the dynamics of spatially discretized partial differential equations. Due to their inherent efficiency and low parameter requirements, they are suitable for replacing parts or even entire deep neural networks, especially in conjunction with distributed memory and stored programmability. This paper investigates the dynamic behavior of CellNN arrays, considering volatile and non-volatile memristors as cell elements, but also for the implementation of synaptic couplings. The results obtained for different memristor models are discussed in detail.
[1] R. Tetzlaff, A. Ascoli, I. Messaris and L.O. Chua, "Theoretical Foundations of Memristor Cellular Nonlinear Networks: Memcomputing with Bistable-Like Memristors," in IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 67, no. 2, pp. 502-515, Feb. 2020, doi: 10.1109/TCSI.2019.2940909.

IC-4.B1:IL34  Utilizing Emerging Memory Technologies to Solve Optimization Problems
D. STRUKOV, UC Santa Barbara, ECE Department, Santa Barbara, CA, USA

This talk presents my group’s recent advances in hardware accelerators for combinatorial optimization. The core innovations are twofold. First, we introduce a massively parallel gradient computation method for high-degree polynomials, ideally suited for efficient mixed-signal in-crossbar-memory computing circuits. Second, we propose a unified framework that leverages this method to design macro-blocks capable of solving higher-order problems, such as Boolean satisfiability (SAT) and polynomial unconstrained binary optimization, directly in their native encoding. Our gradient computation scheme accelerates gradient-descent-based heuristics, which underpin state-of-the-art local search SAT solvers and Ising machines. Compared with formulations that reduce problems to the Quadratic Unconstrained Binary Optimization form, our approach achieves lower area complexity and benefits from the smoother energy landscapes inherent to native formulations. These properties make it particularly effective for tackling high-order problems. Experimental results from ReRAM and SRAM in-memory computing circuits, benchmarked on small SAT instances, together with large-scale simulation studies, demonstrate substantial gains in both speed and energy efficiency over existing solutions.

IC-4.B1:L35  Data-Driven Flux-Controlled Memristor Model for Neuromorphic Applications
K. NIKIRUY, I. PETRENYOV, A. SHKURMANOV, M. ZIEGLER, Chair of Energy Materials and Devices, Department of Materials Science, Kiel University, Kiel, Germany; J. SCHNEEGAß, T. IVANOV, Department of Electrical Engineering and Information Technology, Micro- and Nanoelectronic Systems, TU Ilmenau, Ilmenau, Germany; D. ROSSETTI, F. CORINTO, A. ASCOLI, Department of Electronics and Telecommunications, Politecnico di Torino, Turin, Italy; A.S. DEMIRKOL, R. TETZLAFF, Faculty of Electrical and Computer Engineering, Institute of Circuits and Systems, TU Dresden, Dresden, Germany

This work presents a compact, physically interpretable modelling framework for resistive‑switching devices grounded in Chua’s theory of extended memristors. The central idea is to represent the device conductance as the parallel of two nonlinear dynamical circuits, each selectively active under opposite voltage polarities. This decomposition captures the device’s intrinsic conduction asymmetry and provides a modular structure that mirrors distinct microscopic mechanisms (for example, filamentary formation, ionic drift, or barrier modulation) that dominate under switching behaviour. Model calibration follows a hybrid machine‑learning approach aimed to provide physically meaningful parameter sets. The initial parameter space is obtained via Latin Hypercube Sampling to ensure diverse starting points; Bayesian optimization is then used to explore promising regions efficiently; finally, gradient‑based refinement refines the solution to maximize fit quality while respecting physically derived parameter bounds. In summary, the proposed flux-based extended memristor model combines physical interpretability, SPICE compatibility, and data‑driven parameterization to offer a unified framework for circuit‑level design and neuromorphic system simulation.

IC-4.B1:L36  Logistic Map Circuit for Chaotic Sequence Generation in DNA-based Image Encryption
R. CAVAZZANA, S. HALIUK, A. ASCOLI, D. VOVCHUK, T. SALGALS, V. BOBROVS, F. PARESCHI, F. CORINTO, J. SECCO, Department of Electronics and Telecommunications, Politecnico di Torino, Turin, Italy; Institute of Telecommunications, Riga Technical University, Riga, Latvia; Department of Radio Engineering and Information Security, Yuriy Fedkovych Chernivtsi National University, Chernivtsi, Ukraine

As sensitive data management expands at an exponential pace, securing data transmission and storage has become critical, especially in clinical settings. Conventional encryption standards are increasingly exposed as adversaries gain computational power, motivating exploration of robust, resource-efficient alternatives. Chaotic systems, unpredictable, highly sensitive, and aperiodic, offer rich entropy for next-generation security. We present a compact, low-cost circuit implementation of the logistic map that reliably generates pseudorandom sequences for an image-encryption pipeline. A novel post-processing technique yields balanced, mutually orthogonal binary sequences, strengthening keystream quality and reducing correlation leakage.The resulting sequences are then employed in the proposed image encryption algorithm, which combines permutation processes enhanced through DNA-inspired encoding and operations rules. We validate the system using the NIST Statistical Test Suite and additional security and robustness evaluations. All tests are successfully passed, demonstrating strong randomness and resistance to brute-force and statistical attacks, and confirming the practicality of the approach for secure transmission and storage of sensitive data.

IC-4.B1:L37  High Frequency Characterization and Modeling of CMOS Integrated RRAM Devices
M. UHLMANN, S. DILEK, M. INAC, E.P.-B. QUESADA, R. THILEEBAN, E. PÉREZ, F. KORNDÖRFER, P. OSTROVSKYY, C. CARTA, G. KAHMEN, C. WENGER, A. MALIGNAGGI, IHP – Leibniz Institute for High Performance Microelectronics, Frankfurt (Oder), Germany; E. PÉREZ, C. WENGER, G. KAHMEN, BTU Cottbus-Senftenberg, Cottbus, Germany; C. CARTA, TU Berlin, Berlin, Germany

Resistive Random Access Memory (RRAM) has emerged as a technology for "more-than-Moore" applications due to its simple two-terminal structure, high level of scalability, fast switching speed, and, most importantly, non- volatility and compatibility with standard CMOS fabrication processes. These characteristics make RRAM a promising candidate for embedded memory applications and in-memory computing (IMC) architectures. Although RRAM has been studied extensively for low-frequency mixed-signal systems, its potential in higher-frequency (HF) applications remains less investigated. This work, shows for the first time the design, characterization, and modeling of a CMOS-compatible RRAM device for HF applications. S-parameter measurements were performed over a frequency range from 20 MHz to 67 GHz. Based on these measurements, a lumped-element equivalent circuit model was derived for both the low-resistance state (LRS) and high-resistance state (HRS) of the RRAM. The resulting RF model shows that RRAM devices could be used in circuits for HF applications up to 67 GHz. Among these circuits are amplifier front-ends, millimeter-wave imaging- and satellite communication systems. All of such circuits and systems are able to be designed with the RRAM dynamic model resulting from this work.

IC-4.B1:L38  Neural Information Processing and Time-Series Prediction with Only Two Dynamical Memristors
D. MOLNÁR^1,2, T.N. TÖRÖK^1,3, J. VOLK Jr.¹, R. KÖVECS¹, L. PÓSA^1,3, P. BALÁZS¹, G. MOLNÁR³, N. JIMENEZ OLALLA⁴, Z. BALOGH^1,2, J. VOLK³, J. LEUTHOLD⁴, M. CSONTOS⁴, 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; ⁴Institute of Electromagnetic Fields, ETH Zurich, Zurich, Switzerland

Memristive devices are commonly benchmarked by the multi-level programmability of their resistance states. Neural networks utilizing memristor crossbar arrays as synaptic layers largely rely on this feature. However, the dynamical properties of memristors, such as the tailorable response times arising from the exponential voltage dependence of the resistive switching speed remain largely unexploited. Here, an information processing scheme which fundamentally relies on the latter is proposed [1]. Simple dynamical memristor circuits capable of solving complex temporal information processing tasks are realized. A scheme is presented in which a single non-volatile meristor and a series resistor can perform temporal pattern recognition tasks, such as the discrimination of sub-threshold and super-threshold voltage pulses, or the identification of neural spikes buried in high noise. Furthermore, a time series prediction circuit is implemented using a dynamic layer of only two memristors and a readout layer based on the linear combination of their output signals. This scheme can learn the operation of an external dynamical systems and predict their output with high accuracy.
[1] D. Molnar et al. Adv. Electron. Mater. e00353 (2025), https://doi.org/10.1002/aelm.202500353.

IC-4.B1:IL39  Receptron Hardware Platform for Real Time Learning and Classification
B. PAROLI, S. RADICE, F. BORGHI, M.A.C. POTENZA, P. MILANI, CIMAINA-Interdisciplinary Centre for Nanostructured Materials and Interfaces, Department of Physics “Aldo Pontremoli”, Università degli Studi di Milano, Milano, Italy

A novel threshold logic gate design, called Receptron, has been recently proposed: it is based on nonlinear weights thus widening the spectrum of Boolean computable functions while simplifying training thanks to a random search protocol. The hardware implementation of the Receptron model has been demonstrated by standard microcontrollers and CMOS components. The reconfigurability and functional completeness of the receptron allows faster, more efficient and possibly edge, data processing compared to traditional architectures used for artificial neural networks Here we report the fabrication of a receptron-based electronic board for edge data processing and classification working on analog inputs and capable to learn with an extremely reduced training.

IC-4.B1:IL40  Resistive Switching Devices at the Crossroad of RF Switching and Flexible Electronics
A. KIAZADEH, T. MINGATES, J. DEUERMEIER, M. PEREIRA, A.G. KELLY, E. CARLOS, E. FORTUNATO, CENIMAT|i3N, Faculty of Science and Technology, NOVA University Lisbon, Caparica, Portugal

Resistive switching devices, including RRAMs, emerge as versatile building blocks for next-generation electronics, bridging memory, computation, and reconfigurable circuitry. This talk explores the crossroads of RF switching and flexible electronics, highlighting how resistive switching devices can unlock novel functionalities beyond conventional technologies. We present the design, fabrication, and characterization of RRAM-based RF switches, demonstrating low insertion loss, high isolation, and fast switching suitable for high-frequency applications. Furthermore, we show how integrating resistive switching devices with oxide TFTs in flexible electronics platforms enables low-power, printable, and reconfigurable devices, realized through low-temperature and sustainable fabrication processes. Beyond RF applications, these platforms offer promising pathways toward neuromorphic computing architectures, where memory and processing functions are combined in a compact, energy-efficient, and multifunctional system. The talk emphasizes how resistive switching technology can bridge high-frequency performance, flexible electronics, and emerging computing paradigms in a sustainable and versatile manner.

IC-4.B1:IL41  Memristors with Organic Self-Assembled Monolayers
P. KIRSCH, TU Darmstadt, Darmstadt, Germany; and Merck Electronics KGaA, Darmstadt, Germany

Functionalized self-assembled monolayers (SAMs) can be used as active components in electronics and spintronics devices. In particular, SAMs composed of dipolar, conformationally flexible phosphonic acids enable electrically switchable tunnel junctions with memristive characteristics. The performance of such organic memristive devices is comparable to that of their state-of-the-art inorganic counterparts. The big advantage of organic systems is their intrinsic functional versatility: a rather complex property profile can be easily designed into the individual component molecules. A potential challenge of many SAM systems - thermal stability - was resolved by using thermally and chemically robust phosphonic acids as anchoring groups which are compatible with oxide and nitride substrate materials common in electronics industry. Another class of functional SAMs is composed of chiral compounds, such as binaphthol phosphates (BNP), which can act as electron spin filters. Phenomena around the interaction between electrons of different spin and chiral thin films are known as the chirality-induced spin selectivity (CISS) effect. The CISS effect of chiral SAMs can be utilized for simplified spintronics devices where the SAM replaces the pinned magnetic layer.

IC-4.B1:IL42  Molecular Neuromorphic Building Blocks for Artificial Intelligence
S. GOSWAMI, Centre for Nano Science and Engineering, Indian Institute of Science, Bangalore, India

Artificial Intelligence (AI) has long inspired both fascination and frustration, oscillating between extraordinary promises and recurring disillusionment. Recent milestones—such as AI surpassing human experts in complex tasks—herald a new era of computing, yet these advances come at a steep cost: immense energy demands and prohibitively intensive training. Despite this progress, even the most powerful machines remain far less efficient and compact than the human brain. The root of this gap lies in conventional circuits and architectures that cannot reproduce the brain’s rich, nonlinear dynamics operating near the edge of chaos. In this talk, I will present a new class of molecular circuit elements that capture reconfigurable, brain-like logic at the nanoscale. These devices function as analog, digital, or near-instability elements, offering unprecedented potential to emulate neural behaviour. We trace their development from molecular physics to integrated chip design, outlining a pathway toward energy-efficient AI systems that move beyond the constraints of Moore’s Law.

IC-4.B1:L43  From Nature to Neuromorphic: Harnessing Nigella Melanin as a Green Material for Artificial Synapses
M. AMBRICO¹, G. LUPIDI², Gb. LUPIDI⁴, P.F. AMBRICO¹, L. VALGIMIGLI³, S. MATTIELLO⁴, D. ACETO¹, A. GUZZINI⁴, A. GRIGORYEVA⁵, A. DE STRADIS⁶, R. GUNNELLA⁴, ¹CNR-ISTP - Bari Branch, Bari, Italy; ²School of Pharmaceutical Sciences and Health Products, University of Camerino, Camerino (Italy); ³Department of Chemistry “G. Ciamician”, University of Bologna, Bologna, Italy; ⁴International School Advanced Studies, University of Camerino, Camerino, Italy; ⁵Tecnopolo di Rimini, Rimini, Italy; ⁶CNR-IPSP, Bari Branch, Bari, Italy

The Von Neumann bottleneck, caused by the separation of memory and processing units, limits CPU performance in conventional computing. Artificial synapses offer a solution through in-memory computing and parallel processing, essential for AI and neuromorphic systems. While memristors effectively mimic synaptic behavior, their high power consumption is a drawback. Memcapacitors provide a more energy-efficient alternative due to their non-dissipative nature and minimal thermal effects. Melanin biopolymers, especially those derived from Nigella sativa (black cumin seeds), have emerged as promising materials for artificial synapses thanks to their biocompatibility, ionic conductivity, and organic semiconductor-like properties. Devices fabricated using Nigella melanin via drop-casting on Au and Pt interdigitated electrodes have undergone electrical and dielectric characterization, demonstrating memory switching, retention, and endurance—highlighting their potential in neuromorphic computing and bioelectronic applications.

Session IC-4.B2 Diffusive and volatile memristors

IC-4.B2:IL44  Diffusive Memristor Devices and Circuits for Neuromorphic Computing
QIANGFEI XIA, University of Massachusetts, Amherst, MA, USA

A diffusive memristor is a volatile resistance switch that returns to a high resistance state once the stimulus is removed. Owing to their structural resemblance to biological ion channels, diffusive memristors have been effectively employed to mimic typical synaptic and neuronal behavior. The stochastic dynamics, including the randomness in switch-on time, have been harnessed to develop true random number generators. In this presentation, we will discuss our advancements in diffusive memristors with consistent and controllable relaxation behavior, which have been utilized to construct large-scale circuits for event-based spatiotemporal pattern recognition. Additionally, we will present an oscillatory neuron created with diffusive memristors for neuromorphic computing.

IC-4.B2:IL45  Investigation and Engineering of Switching and Relaxation Dynamics of Ag-based Diffusive Memristors
S. BRIVIO, M. DUTTA, A. BELLINGERI, F. VACCARO, S. SPIGA, CNR - IMM, Unit of Agrate Brianza, Agrate Brianza, Italy

Computing has been acquiring an increasingly large share of the energy footprint in the last decades. Therefore, it is the right time now to lay the foundations for novel, sustainable computing paradigms, inspired by the brain’s efficiency. Such revolution must be accompanied by the development of components building up physical computing substrates. Volatile diffusive memristors can emulate the basic brain-inspired cognitive functions that encode information in the time variable. Diffusive memristors display voltage-driven switching from high to low resistance states and self-relaxation. This behaviour is due to ion formation and migration, like in our brain, and it can be realized over multiple timescales, from ns to, possibly, minutes, through appropriate programming protocols and material design. In this paper, we discuss our work on diffusive memristors based on a Ag/SiOx/Pt stack and draw a route from device characterization, to understanding of physics at the basis of the device operation,[1] and to the definition of programming strategies and engineering of materials for volatile diffusive memristors towards computing applications across multiple timescales.[2]
[1] Vaccaro et al Appl. Math. Model. 134 , 591 (2024). [2] Dutta et al Adv. Electron. Mater. 10, 2400221 (2024).

IC-4.B2:L46  Effect of Al2O3 Layer Thickness and of Programming Parameters on Cumulative and Intrinsic Retention in Ag/Al2O3/SiOx/Pt Volatile Memristors
A. BELLINGERI^1,2, S. BRIVIO¹, S. SPIGA¹, ¹CNR - IMM, Unit of Agrate Brianza, Italy; ²Università degli Studi di Milano, Department of Physics “Aldo Pontremoli”, Italy

Volatile memristors are promising candidates for implementing short-term plasticity and neuronal functions in brain inspired hardware neural networks. Despite various efforts in literature to engineer retention times toward a wide range of timescales, most of the volatile memristors exhibit retention times in the µs-s range. Our study focuses on stack engineering of Ag/Al2O3/SiOx/Pt volatile memristors, whose mechanism relies on the formation and dissolution of Ag filament, to expand retention times. We measure retention times in fresh devices (intrinsic retention) with and without Al2O3 thanks to DC measures at different compliance currents and staircase speeds. The results evidence that the introduction of the Al2O3 layer leads to increased retention times up to tens of seconds, depending on Al2O3 thickness and programming parameters. Furthermore, retention times increase up to hundreds of s with repetition of measurements, outlining the presence of cumulative effects, which are anyway reversible within a short time scale. We will discuss how intrinsic and cumulative retention effects can be exploited for applications.
This work has been partially funded by Ministero delle Imprese e del Made in Italy (MIMIT) under IPCEI Microelettronica 2, project MicroTech_for_Green.

IC-4.B2:L47  Self-aligned Single-Nanoparticle Ag Memristors
M. FISCHER-BUTESHEVA, D. EGLIN, M. LEWERENZ, G.-L. FRANCHINI, E. PASSERINI, N. JIMENEZ OLALLA, K. SRIKRISHNAPRABHU, M. STECHER, R. GISLER, Y. FEDORYSHYN, M. CSONTOS, J. LEUTHOLD, Institute of Electromagnetic Fields, ETH Zurich, Zurich, Switzerland

Filamentary memristors are especially attractive devices for next-generation computing applications like neuromorphic computing due to the inherently small filament sizes, theoretically allowing for footprints of a few atoms. However, reliably decreasing device footprints below 100 x 100 nm2 is increasingly challenging due to the stochastic nature of filament formation. The key to reliable, low-footprint switching is the confinement of the filament site with high precision. In this work, we demonstrate a novel bottom-up fabrication method for memristors featuring a self-aligned filament formation site. Ag nanoparticles (NP) were synthesized via an aqueous on-chip chemical process at a geometrically predefined location. Via controlling the nucleation process, single NPs could be precisely nucleated in Pt gaps. The resulting Ag NPs were electrically conditioned to yield memristive gaps. Compact memristors with a clear I-V hysteresis and a switching site footprint of about 50 x 50 nm2 were obtained. The new NP implementation technique circumvents cumbersome NP placement and manipulation techniques. Beyond a new fabrication route for small-footprint Ag memristors, this work highlights an unexplored route for single NP implementation in versatile functional devices.

IC-4.B2:L48  Investigation of the Transient Behaviour of Volatile Electrochemical Metallization Cells Operated in Integrate-and-fire Neuron Circuits
J. RASBACH, NINGYUAN MA, R. WALIED AHMAD, S. MENZEL, S. HOFFMANN-EIFERT, Forschungszentrum Jülich GmbH, Peter Grünberg Institut 7, Jülich, Germany

Neuromorphic computing (NC) concepts use the hardware-based emulation of the biological brain to store and process digital information with much lower power consumption than classical von Neumann computers. The integration of volatile electrochemical metallization (vECM) type devices in integrate-and-fire (IF) neuron circuits enables the neuronal functionality of small footprint circuits. Highly power efficient vECM devices receive growing attention because of their low threshold voltage and high off resistance [1]. In this study, the integration of a Pt/HfO2/Ag-based vECM device in an IF neuron circuit is successfully demonstrated. The switching behaviour of the vECM device in different oscillating modes is investigated as a function of the applied voltage, the synaptic weight, and the integrator. By combining the measurement data with circuit simulations and physical modeling [2], conclusions were drawn regarding the wake-up effect [3] and the filament growth and dissolution in the vECM device. This work provides insights for the design and programming of artificial neuron circuits paving the way to innovation in NC.
[1] Moon et al., Mat. Horizon 11, 4840 (2024). [2] Ahmad et al., Adv. Int.Sys, accepted (2025). [3] Dutta et al., Adv. Eletr. Mater. 10, 2400221 (2024).

IC-4.B2:L49  Volatile Amorphous-SrTiO3 Devices with Tunable Decay Time for Event-based Sensing
D. SPITHOURIS, J. HELLWIG, C. WITTBERG, R. DITTMANN, PGI-7, Forschungszentrum Jülich, NRW, Germany; H. GREATOREX, E. CHICCA, BICS Lab, Zernike Institute for Advanced Materials, University of Groningen, The Netherlands; CogniGron, University of Groningen, The Netherlands

Tunable gradual switching volatile devices can enable adaptation to diverse temporal coding tasks and richer information encoding making them ideal for neuromorphic edge applications. Most volatile memristive devices reported to date exhibit abrupt filamentary switching that suffers from reliability and variability issues. Here we present an area-dependent resistive switching Pt/amorphous-SrTiO3/TaOx/Ta device stack, that leverages its ionic-based volatility to realize tunable, reproducible, and gradual switching dynamics. The CMOS BEOL-compatible devices demonstrate very low variability, stable operation, forming-free, self-compliance and rectifying behavior. We provide experimental evidence over the ionic origin of the volatility and the conduction mechanism. By leveraging the rich ionic dynamics and kinetics, we demonstrate how the volatile behavior of the device can be tuned through applied stimulus and stack engineering, enabling decay times ranging from tens of milliseconds to tens of seconds. Furthermore, we conduct a comprehensive study on the reliability of the device and its capability for multilevel operation. Finally, we demonstrate the potential of our volatile amorphous-SrTiO3-based devices for event-based sensing, with a particular focus on event-based vision.

IC-4.B2:L50  Three-Terminal Memristor with Tunable Volatility and Set-Voltage
K. SRIKRISHNAPRABHU, M. LEWERENZ, M. FISCHER-BUTESHEVA, E. PASSERINI, A. SCHNEUWLY, N.J. OLALLA, R. GISLER, M.A. STECHER, Institute of Electromagnetic Fields, ETH Zurich, SWITZERLAND; A. EMBORAS, M. LUISIER, Integrated Systems Laboratory, ETH Zurich, SWITZERLAND; M. CSONTOS, J. LEUTHOLD, Institute of Electromagnetic Fields, ETH Zurich, Switzerland

Increasing functional complexity on hardware is key to realizing efficient neuromorphic computing on-chip. Tunable memristors that can be reconfigured to multiple functions are suitable candidates for increasing the functional density on hardware.However, a versatile memristor, where multiple memristor properties like set-voltage, on-resistance, volatility and decay times can be tuned in the same device is not reported yet. In this work, we report a tunable three-terminal memristor with a silver-tin alloyed source. The heterogenous electrochemical properties of silver and tin play an important role in the tunability of the device. In this device, the polarity of the gate voltage tunes the volatility of the device. Positive gate voltages result in non-volatile switching while negative gate voltages result in volatile switching. The magnitude of the gate voltage in the volatile and non-volatile regimes tunes the set-voltage and on-resistance. Thus, due to the volatility tuning, synaptic and neuronal behaviors can be realized on the same device.Also, these synaptic and neuronal behaviors can be tuned using the magnitude of the gate voltage.Thus, this tunable three-terminal memristor is a suitable candidate for functionally complex neuromorphic hardware.
1) NatRevMat 7 575–591,4 22.

Session IC-4.C Brain inspired hardware and computing

IC-4.C:IL51  Modeling the Brain as a Complex Adaptive System: Computational Approaches to Information Transfer in Neuronal Circuits
D. GANDOLFI, University of Modena and Reggio Emilia, Modena, Italy

The quest for bottom-up brain modeling—grounded in neuronal biophysics—echoes Richard Feynman's famous assertion: "What I cannot create, I do not understand." Today, large-scale, single-cell resolution models of brain microcircuits [Gandolfi et al., 2023]—are at the forefront of computational neuroscience. Relying on biologically realistic construction strategies that integrate morphological and anatomical data, these models not only advance the field, but also offer new insights into brain functions and information transfer within neuronal circuits. As a case study, a pattern completion task in a biologically realistic model of the hippocampus will be presented. At the same time, it is becoming increasingly evident that we must develop theories and strategies to integrate bottom-up models with top-down approaches—those based on ensemble representations of large-scale network activity. Within the theoretical framework of The Virtual Brain, it is now possible to simulate personalized brain activity, opening new perspectives for both clinical applications and basic research. These models offer powerful tools for individualized therapeutics interventions and for deepening our understanding of the neural basis of information transfer in physiological and pathological states.

IC-4.C:IL52  Self-organizing Neuromorphic Networks as Dynamical Systems for Computing
G. MILANO¹, C. RICCIARDI², E. MIRANDA³, ¹Advanced Materials Metrology and Life Science Division, INRiM (Istituto Nazionale di Ricerca Metrologica), Italy; ²Department of Applied Science and Technology, Politecnico di Torino, Italy; ³Departament d’Enginyeria Electrònica, Universitat Autònoma de Barcelona (UAB), Cerdanyola del Vallès, Spain

Besides crossbar arrays of memristive devices, self-organizing memristive networks based on nanomaterials have been demonstrated as promising substrates for in materia reservoir computing. In this context, the relationship between emergent dynamics and computational properties of these systems is still a challenge. Here, we report through an experimental and modeling approach on a new framework enabling the description of self-organizing nanowire networks as stochastic dynamical systems. We show that emergent dynamics can be holistically described in terms of an Ornstein-Uhlenbeck process where stimuli-dependent deterministic trajectories are coupled with noise and jump effects. Also, we show that this framework allows to investigate and quantify the role of deterministic and stochastic dynamics on information processing capabilities of the system, evaluating computing properties through benchmark tasks such as nonlinear autoregressive moving average (NARMA) and nonlinear transformation (NLT). These results represent a step ahead for the implementation of physical computing paradigms exploiting emergent dynamics of self-organizing systems.

IC-4.C:L53  Reversible Self-assembly of Neuromorphic Particle Networks
S. VAN KESTEREN¹, L. BÉGON-LOURS¹, S. SACANNA², ¹ETH Zurich, Dept. of Information Technology and Electrical Engineering, Zurich, Switzerland; ²New York University, Department of Chemistry, New York, NY, USA

Topological plasticity—the ability of a network to break and reform its connections—is central to learning in biological neural systems. Reproducing this adaptability in solid-state neuromorphic hardware remains a major challenge. Self-assembling nanoparticle networks offer a promising route, as their connections can be reversibly formed and dissolved under external stimuli. In this talk, we present recent experimental and numerical progress toward realizing such rewireable neuromorphic systems. Experimentally, we demonstrate light-controlled self-assembly of tunable random and crystalline nanoparticle structures, precisely tuning interaction strengths from -30 to 0 KbT, and characterize the electrical properties of single-particle junctions that govern their memristive-based processing. Complementarily, we develop a molecular dynamics model that reproduces the dynamic formation and purging of interparticle connections, providing a quantitative framework for understanding and designing reconfigurable neuromorphic networks.

IC-4.C:L54  Programmable Connectivity of Self-assembled Materials for Algebraic and Classification Tasks on Edge Systems
F. BORGHI, D. DECASTRI, F. PROFUMO, P. MILANI, Physics Department and CIMaINa, University of Milano, Milano, Italy

The exploitation of self-assembled systems characterized by nonlinear dynamics is actively investigated as an alternative strategy to develop energy-efficient data processing devices, relying on the emerging behavior of stochastic network activities. Here, the implementation of data processing devices based on nanostructured thin films is explored in a dual way. First, reconfigurable nonlinear threshold logic gates (TLGs) have been implemented using nanostructured devices. The control over the synchronous activity and connectivity of the micrometric active sites, which rule the emerging network dynamics, resulted essential for the performance and reliability of the nonlinear TLG devices. These have also been investigated through a phenomenological model, which includes the external stimulus and a state variable of the system. Secondly, we demonstrated the use of these systems to classify with high accuracy and in real-time neuronal traces recorded from in vivo systems. The classification is carried out by means of a linear classifier and is characterized by higher interpretability and accuracy compared to artificial neural networks. These results pave the way to the integration of such systems into edge devices, such as brain-machine interfaces.

IC-4.C:L55  Dense Ag/PVP-based Nanowire Networks for Brain-like Electronics
J. DIAZ SCHNEIDER, C. QUINTEROS, E. MARTÍNEZ, P. LEVY, Centro Atómico Bariloche, Comisión Nacional de Energía Atómica (CNEA), S.C. de Bariloche, Río Negro, Argentina; Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET), Argentina; Instituto de Nanociencia y Nanotecnología (CNEA + CONICET), S.C. de Bariloche, Río Negro, Argentina ICIFI, (UNSAM-CONICET), San Martín, Argentina

We study the emergence of resistive switching effect on dense self assembled nanowire networks, a commonly used material for transparent electrodes. The injection of high current into over-percolated samples produces a transition to a still percolated regime with enabled resistive switching between states of definite resistance. The resulting circuit is rationalized as a random network consisting resistors and junctions with memristive properties in which conductive filaments connect and disconnect nanowires. The dynamics of the formation and disruption of filaments unveils collective properties of the system. As nanowire networks can emulate synaptic processes, neuromorphic computation capabilities with the additional functionality of transparency are envisaged.

IC-4.C:L56  Steady-state and Non-steady-state Noise Dynamics of VO₂ Memristors revealed by Full Cycle Noise Spectroscopy Measurements
B. SANTA, S.W. SCHMIED, Z. SINOROS-SZABO, T.N. TÖRÖK, L. POSA, Z. BALOGH, A. HALBRITTER, Department of Physics, Budapest University of Technology and Economics, Budapest, Hungary

Memristive devices are promising candidates for neuromorphic computing applications, but their performance is fundamentally limited by non-idealities, such as cycle-to-cycle variations and the internal fluctuations in the active volume, i.e. device noise. Here, we investigate these non-idealities in VO2, a fundamentally important memristive material system that is a key building block for spiking artificial neuron circuits. We perform a comprehensive noise diagnostics [1], that goes beyond the readout noise analysis, and also tracks the nonlinear and non-stationary evolution of internal fluctuations at higher voltages. This knowledge not only reveals the source of the fluctuations, but the nonlinear noise characteristics may also precursor the switching process [2]. Comparing the noise data to resistor network simulations, we identify distinct regimes, where either active volume dependent intrinsic resistance fluctuations, or the so-called phase transition noise dominate. Furthermore, we investigate how the noise characteristics can be influenced by the optimization of the device fabrication, and how the noise properties of VO2 compare to those of other memristive systems.
[1] Z. Balogh et al., Nano Fut. 2021, 5, 042002; [2] A. Nyáry et al., ACS Appl. Mater. Int. 2025, 17, 25654

IC-4.C:L57  High Frequency Neural Information Processing with Dynamical Memristors
D. MOLNAR, T.N. TÖRÖK, J. VOLK Jr., A. HALBRITTER, Department of Physics, Institute of Physics, Budapest University of Technology and Economics, Budapest, Hungary

Energy efficiency and computational speed are among the most critical parameters in neuromorphic computing. Traditional approaches focus on translating neural networks into more efficient hardware implementations, where individual elements are simple but the interconnecting networks are highly complex. A less explored strategy replaces network complexity by dynamical complexity. This is not only supported by the availability ultra-short switching times of 10ps [1] but also the common exponential voltage dependence of the resistive switching speed. Here, we explore the dynamic properties of memristive devices approaching the GHz range. We use this vast dynamic range to show, how our previously demonstrated low-frequency dynamical information processing scheme [2] can be extended to the high-frequency regime, where signal processing is achieved faster with significantly less energy consumption. We also show that the actual dynamical properties may differ in the different material systems and frequency ranges. This can be utilized to tailor the information processing circuit for the actual task, i.e. to build more time-domain or more voltage-domain sensitive circuits.
[1] M. Csontos et al., Adv. Electron. Mater. 9, 2201104 (2023). [2] D. Molnar et al., Adv. Electron. Mater. e00353.

IC-4.C:IL58  Ionic Nanoarchitectonics for Neuromorphic Computing
KAZUYA TERABE, National Institute for Materials Science (NIMS), Tsukuba, Ibaraki, Japan

Advancements in next-generation AI technology are expected to be driven by edge computing and brain-inspired computing approaches. The development of neuromorphic devices and circuits is essential for applications that require compact, low-power information processing and real-time responsiveness—goals that are challenging to achieve using conventional methods. By mimicking the structure and function of biological neural networks, these devices possess diverse information processing capabilities and may enable AI to perform flexible, advanced intellectual activities similar to those of the human brain. This presentation will introduce ‘ionic nanoarchitectonics,’ a cutting-edge approach to creating such neuromorphic devices. This method enables multifunctional devices through precise control of localized ion transport and electrochemical phenomena within solids. Ionic nanoarchitectonics manipulates ion movement from the nanoscale to the atomic scale, controlling state changes within materials and facilitating the development of novel information processing functions and adaptive devices. We present neuromorphic devices that have been constructed using ionic nanoarchitectonics, and demonstrate their various performance characteristics, including their synaptic functions.

IC-4.C:IL59  Electrochemical Ionic Synapses for Energy-Efficient Brain-Inspired Computing
BILGE YILDIZ, Massachusetts Institute of Technology, Cambridge, MA, USA

In this talk, I will share our work on the ionic electrochemical synapses, whose electronic conductivity we control deterministically by electrochemical insertion/extraction of dopant ions. This work is motivated by the need to enable significant reductions in the energy consumption of computing, and is inspired by the ionic processes in the brain. Proton as the working ion in our research presents with very low energy consumption, on par with biological synapses in the brain. The target material is a mixed proton-electron conductor, whose electronic conductivity depends on the proton concentration through doping and phase change effects. The candidate materials are a spectrum of intercalation oxides as well as 2D van der Waals materials. In addition, the conductance change in these electrochemical devices depends non-linearly on the gate voltage, due to field-enhanced ion migration in the electrolyte, and charge transfer kinetics at the electrolyte-channel interface. We are leveraging these intrinsic nonlinearities to emulate bio-realistic learning rules deduced from neuroscience studies, such as spike timing dependence of plasticity and Hebbian learning rules. Our findings provide pathways towards brain-inspired hardware that is bio-plausible and energy-efficient.

IC-4.C:IL60  Electrochemical Random Access Memory - Self Heating Opens a New Frontier
E.J. FULLER, Sandia National Laboratories, Livermore, CA, USA

Analog information processing is intrinsically more efficient than digital computing and can circumvent its bottlenecks. Programmable resistors, for example, could enable efficient and reconfigurable voltage division, dynamic-gain amplification, and matrix-vector multiplication – but the required analog properties remain locked behind a missing circuit element: a programmable linear resistor. We introduce a resistive device based upon a self-heated, metal oxide electrochemical cell. The approach combines the reliability and linear response of trim resistors with the programmability and density of resistive memory – and with far greater precision and dynamic range than either approach. The key advance is an electrothermal gate that simultaneously spreads heat and vacancy reactions to enable nine-decades of tunable analog resistance. Devices are linear across the entire range which enables low harmonic distortion signal processing. The self-heating profoundly reduces noise, yielding 100x lower conductance errors than other resistive memory devices, and thousands of distinguishable states. Simulations indicate matrix multiplication efficiency could approach >1,000 TOPS/W.

IC-4.C:IL61  In-Materio Reservoir Computing Utilizing Spatiotemporal Dynamics of Ion, Electron, and Spin
T. TSUCHIYA, D. NISHIOKA, W. NAMIKI, R. IGUCHI, Y. SHINGAYA, K. TERABE, National Institute for Materials Science, Tsukuba, Ibaraki, Japan

In-materio computing, in which inherent properties of materials are harnessed to perform computing, has recently been attracting attention for overcoming the low energy efficiency of Artificial neural network (ANN) computing. In particular, in-materio reservoir computing is attractive because of its ability to significantly reduce the computational resources required to process time-series data by utilizing the nonlinear responses of a ‘reservoir’ (material or device, as a dynamical system) to input signals. Realizing nonlinear and diverse dynamics with nanomaterials and/or nanospace is thus a great challenge for the development of low-power consumption and highly integrated ANN-based computing devices. Recently, we have developed high-performance in-materio reservoir computing devices on the basis of spatiotemporal dynamics of information carriers in materials. One example is an ion-gating reservoir, which utilizes ion-electron coupled dynamics in the vicinity of electrochemical interfaces. Another example is a magnonic reservoir device utilizing chaotic or edge-of-chaotic dynamics of interfered spin waves in magnetic materials. In the presentation, the relationship between the computation performance and the dynamical behavior of the devices will be discussed.

IC-4.C:L62  Energy-Efficient Programming of First-Order Memristive Devices
V.A. SLIPKO¹, A. ASCOLI², F. CORINTO², Y.V. PERSHIN³, ¹Institute of Physics, Opole University, Opole, Poland; ²Department of Electronics and Telecommunications Politecnico di Torino, Turin, Italy; ³Department of Physics and Astronomy, University of South Carolina, Columbia, SC, USA

We introduce a theoretical approach for designing stimuli that accurately program first-order memristive devices while simultaneously reducing the energy costs associated with the resistance switching phase. By examining two first-order predictive models for resistance switching memories, our theoretical analysis shows how, based on the device's physical characteristics—reflected in the model equations and parameter settings—the stimulus protocol aimed at minimizing Joule losses during SET or RESET transitions can involve various fixed-polarity voltage stimuli. These may include a single voltage pulse of a specific height within an acceptable range for part of the programming time, or analog continuous waveforms, which can be approximated as sequences of square voltage pulses with varying heights throughout the entire programming period. Considering that maximizing energy efficiency during resistance programming operations—frequently performed in compute-in-memory engines—are primary objectives for memristor researchers worldwide, our innovative optimization strategy, rooted in control-theoretic principles, and promising substantial energy savings in memristive systems for AI applications, may inspire new approaches to update the information stored in resistive memory banks.

IC-4 - 6th International ConferenceMaterials and Devices Technologies for Energy-efficient Neuromorphic and Unconventional Computing

Cimtec 2026

IC-4 - 6th International Conference
Materials and Devices Technologies for Energy-efficient Neuromorphic and Unconventional Computing