Artificial intelligence is pushing the limits of digital computing to such an extent that, if current trends continue, global energy consumption from computation alone could surpass all other forms of energy use within the next two decades. Electrochemical random-access memory (ECRAM) presents a promising approach to storing and processing information in analog form, increasing efficiency and sustainability with in-memory computing [1-3].

In this discussion, I will introduce a novel approach to ECRAM, where devices are self-heated to overcome the kinetic barriers associated with electrochemical reactions. We refer to these devices as electro-thermo-chemical random-access memory (ETCRAM) [4]. The key innovation is an electrothermal gate that simultaneously distributes heat and facilitates oxygen vacancy reactions, enabling large reversible modulations of composition with a tunable analog resistance of up to one-billion times (×10⁹).

We demonstrate that programming an ETCRAM cell based on TaO_x achieves current-voltage linearity across the entire operational range, a feat not possible with other memory technologies. This linearity enables a wide variety of signal processing tasks inaccessible to other devices and is attributed to a unique set of physical properties, including a non-electrostatic, volumetric programming mechanism and distinct read and write paths. Importantly, the self-heating mechanism significantly reduces noise, resulting in precision that is ×100 higher than other resistive memory technologies, with thousands of distinguishable states.

We showcase a variety of signal processing tasks, including dynamic-gain amplification (up to 1V) and voltage-encoded matrix-vector multiplication (MVM), both demonstrating low harmonic distortion. Furthermore, we present the integration of a 1,024-element ETCRAM array with CMOS technology. Due to its unique linearity and precision at high resistance, simulations indicate that MVM efficiency could approach >1,000 TOPS/W.

Electrochemical random-access memory (ECRAM) is widely considered a prime candidate for future memory technologies due to its exceptional scalability and compatibility with standard CMOS processing. In particular, three-terminal synaptic transistors (SynTs) employ ion-conductor gating that enables highly energy-efficient control of ionic doping through redox-mediated processes, thereby decoupling write and read operations and enhancing the linearity of state programming for neuromorphic computing. The inherently faradaic nature of these processes naturally prompts an investigation into the feasibility of exploiting ECRAM devices as energy-storage elements and thereby further improves the system-level energy efficiency of the overall solution .

To evaluate their storage capability, a comprehensive set of electrochemical characterizations is performed. SynTs with extrinsic pseudocapacitive TiO2 channel material have been considered for study and are found to exhibit behavior consistent with that of conventional electrochemical storage systems, with charge capacity scaling according to the applied potential window and active device area. The resulting areal storage capacity surpasses that of high-density capacitor technologies by several orders of magnitude, while maintaining excellent synaptic performance: a nonvolatile conductance modulation on the order of tens of nanoSiemens is enabled by reversible ion intercalation within the channel, and synaptic functions such as long-term potentiation and depression operate with switching energies on the order of femtojoules per square micrometer. Additionally, the devices demonstrate endurance on the order of several hundred thousand weight-update cycles.

Finally, design perspectives are presented illustrating how an ECRAM-based “in-memory energy” approach could offer a novel technological pathway capable of addressing the requirements of both emerging and established application domains.

Different classes of oxide-based memristors are being actively investigated as enabling technologies for neuromorphic systems and unconventional computing paradigms. These devices can replicate synaptic and neuronal functionalities directly in hardware, operate as computing units for in-memory processing within neural networks, and serve as fundamental elements in nonlinear dynamical circuits. In this context, oxide-based memristors exhibiting both volatile and non-volatile resistive switching are extensively studied to implement diverse computing functionalities [1].

The first part of the presentation will focus on our latest results concerning volatile electrochemical memristors based on Ag/SiOₓ/Pt structure [2], and how their relaxation dynamics can be tailored by introducing an ultrathin Al₂O₃ layer (1–2 nm) via atomic layer deposition at the SiOₓ/Pt interface. We will examine the relationship between switching times and relaxation effects, which governs memristor dynamics and allows for multiple switching modes to emulate essential synaptic and neuronal functions. In the second part, I will introduce our recent advances on HfO₂-based analog memristors and their application in a memristor-driven circuit inspired by the Murali–Lakshmanan–Chua (MLC) architecture [3]. This circuit leverages the programmable and nonlinear properties of Pt/HfO₂/TiN memristor devices and enables single-node reservoir computing for various nonlinear classification tasks and real-time information processing [4].

Magneto-ionics, which refers to voltage-driven modification of the magnetic properties of materials as a result of electric-field-induced ion transport, offers a compelling route toward low-power, analog magnetic memories and computing devices. Just as spintronics revolutionized electronics by integrating spin, merging magnetism with voltage-driven ion motion (iontronics) opens exciting prospects for the development of analog low-power magnetic data storage and their application in emerging research areas such as synaptic devices, data security and in-memory computing. While most magneto-ionic systems rely on oxygen, hydrogen, or lithium ions, this work introduces nitrogen magneto-ionics: the voltage-driven transport of nitrogen ions in transition metal nitride films (e.g., CoN, FeN, CoMnN, FeCoN) at room temperature using solid or liquid electrolytes [1-3]. These nitrides exhibit a tunable coexistence of ferromagnetic, paramagnetic, and antiferromagnetic phases, dictated by alloy composition and nitrogen content.

Remarkably, and in contrast to oxygen magneto-ionics, nitrogen transport tends to occur uniformly through a plane-wave-like migration front, an effect particularly interesting for the implementation of multilayer memory devices. We discuss methods to enhance switching speed and cyclability, and demonstrate neuromorphic traits such as potentiation, depression, spike-timing-dependent plasticity, and multilevel memory, amongst others. Notably, some effects can be induced wirelessly via bipolar electrochemistry. New phenomena in nanoscale patterned devices, such as FeCoN sub-micron disks, reveal a voltage-controlled transition between paramagnetic, coherent rotation, and magneto-ionic vortex states. These are driven by precise control of the planar N³⁻ ion front, which modulates local magnetic layer thickness. Miniaturization of nitrogen magneto-ionic systems offers future prospects for brain-inspired devices and data security, where multiple synapses can be interconnected to perform in-memory computing.

Recent predictions highlight that the surging demand for computational power could account for a major share of global energy consumption by 2050. To address this sustainability challenge, a paradigm shift toward new, energy-efficient computing architectures is required. The possibility of dynamically tuning material properties via voltage-driven ion migration represents a promising pathway to realize ultralow-power storage and logic devices. In perovskite oxides, the voltage-driven modulation of the oxygen vacancy concentration (δ) ^[1,2] acts as the primary lever to tune these properties. However, the efficiency and thermodynamic trajectory of this oxygen exchange are intrinsically governed by the cation composition. Consequently, understanding how B-site substitution influences the oxygen reduction pathway is critical for designing devices with tailored switching characteristics.

In this work, we present a comparative study between SrFeO_3-δ (SFO) and the B-site substituted SrFe_0.5Co_0.5O_3-δ (SFCO). Using a solid electrolyte gating geometry, we drive the reversible (de)intercalation of oxygen ions, allowing for precise control over the oxygen non-stoichiometry across the full crystallographic range: from the fully oxidized perovskite to the reduced infinite-layer phase. We demonstrate that the introduction of Cobalt significantly alters the reaction pathway during the voltage-driven reduction process. While the pure SFO system transitions through the expected sequence of Perovskite à Brownmillerite à Infinite Layer, the SFCO films reveal a more complex trajectory due to the easier reducibility of Cobalt. This results in the stabilization of distinct intermediate states, specifically revealing distinguishable "oxidized" and "reduced" perovskite phases prior to the Brownmillerite transformation.

We further correlate these structural evolutions with their functional properties. For the SFO reference system, the phase transition is accompanied by a conductivity modulation of up to 5 orders of magnitude. By employing a setup that allows for simultaneous structural, electrical, and optical characterization, we provide a comprehensive map of the stoichiometry-functionality relationship for both materials. These results highlight the interplay between B-site chemistry and oxygen dynamics, offering new guidelines for controlling the redox pathways in future ionotronic devices.

Spintronic and magnetoelectric systems continue to face intrinsic trade-offs among switching speed, energy consumption, and reliability. Recently, optical excitations of antiferromagnetic (AFM)/ferromagnetic (FM) bilayers have been used to modulate the interfacial interaction between the two layers, commonly known as “exchange bias”. While interesting fast magnetization dynamics have already been observed [1], so far, in most cases laser illumination has triggered thermal effects, which is detrimental in terms of energy efficiency. Inducing strain-mediated changes in AFM/FM exchange interactions through the photovoltaic effect (using a low-power laser and a ferroelectric, FE, substrate) is appealing to reduce power consumption. Moreover, the optical control of ferroelectric states and its magnetoelectric coupling with multi-state functionality [3] has the potential to overcome current performance bottlenecks while enabling new device paradigms for neuromorphic computing, reconfigurable logic architectures, and ultra-low-power wireless data storage.

Here, we present the first demonstration of significant optical control of exchange bias at room temperature in a FE/FM/AFM heterostructure via visible-light-induced photostriction [2]. This effect arises from abnormal strain generated via the photostrictive response of the ferroelectric substrate, which induces compressive stress along both in-plane ferroelectric polarization directions. This strain is transferred to FM layer interfaced with AFM layer, leading to a modulation of magnetic anisotropy and interfacial exchange bias coupling. Unlike conventional electric-field-driven or strain-mediated magnetoelectric coupling, this approach enables non-thermal, reversible, multilevel tuning of magnetic interactions, with programmable states achieved solely by adjusting light intensity as low as 0.1 W cm⁻². This light-controlled, multistate response provides a new degree of freedom for neuromorphic and opto-magnetic memory devices, offering energy-efficient, wireless, and multifunctional operation.

Being able to control how materials exchange and conduct heat could open new avenues of research, generating a novel class of heat-driven devices such as thermal switches and thermal transistors1,3. One of the most promising approaches to build these components is based on the variation of the thermal conductivity of an active layer pumping ions in/out the layer through  an electrolyte from/towards a counter electrode. Some examples of such transistors have been developed, usually employing liquid electrolytes2. While variations in thermal conductivity of efficient active layers have been reported, a short variation of conductivities and the inability to be integrated into a circuit have hindered the performance and the applicability of thermal transistors. Solid-state thermal transistors with systems formed by a solid electrolyte and an oxide active layer have recently been studied as a response to these limitations3.  

In this work, we study the effect of oxygen stoichiometry on the thermal properties of complex oxides that can change their crystalline structure by reduction/oxidation processes through a solid electrolyte. In particular, we focus on SrFeO3, in which we can produce phase changes from perovskite to brownmillerite in a fully reversible manner. For this purpose, we designed and fabricated a platform to modulate the oxygen stoichiometry of the active layer, producing a wide span of oxygen contents in a single sample. In the first step, a thin film of the active oxide was deposited by PLD on a substrate of yttrium-doped zirconia, which is an ion-conductive oxide, functioning as an electrolyte. Later, we designed and fabricated a gold pad stripe pattern on top of the thin film. Finally, by applying a voltage between this strip and a silver counter electrode, we modified the oxygen stoichiometry along the gold stripe, which was determined using a non-destructive optical-based procedure. Finally, by using the frequency domain thermoreflectance technique we were able to make precise measurements of the thermal conductivity at each point, thus correlating the oxygen content in the film with the thermal conductivity. The studied material shows that there is a lot of potential for this family of oxides in the field of thermal modulation. 

The ability to manipulate magnetic properties through ionic motion in ferromagnetic/oxide structures in a non-volatile way, rather than through volatile, purely electronic means, presents exciting opportunities for the development of functionalities like reconfigurable multistate memories and the implementation of cumulative gate effects in spintronics devices. Magneto-ionics takes inspiration from memristor technologies and offers one of the most advanced approaches today 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 different strategies to develop multistate magneto-ionic memory devices. A variety of material combinations and device designs allows to explore the control of nucleation/propagation of a spin-reorientation transition under gate voltage [1] and the exploration of an extended gate-induced oxidation-reduction spectrum [2] to generate multiple stable, 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 [3]. 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.

References

[1] G. Bernard, X. Lafosse, C. Tataru, M.-A. Syskaki, A. Durnez, F. Mahut, D. Ravelosona, J. Langer and L. Herrera Diez, ‘Magneto-ionic Control of a Propagating Spin Reorientation Transition’ Nano Letters 25, 32, 12241–12247 (2025).

[2] I. Benguettat-El Mokhtari, R. Pachat, V. Porée, A. Lamperti, Y. Roussigné, M.-A. Syskaki, J. Wrona, G. Bernard, A. Cataldo, A. Resta, A. Nicolaou, S. Ono, S. M. Ch´erif, J. Langer, D. Ravelosona, M. Belmeguenai, A. Solignac, L. Herrera Diez, ‘Exploring the full magneto-ionic oxidation spectrum in Pt/CoFeB/HfO2’, Appl. Phys. Lett. 126, 232402 (2025).

[3] G. Bernard, K. Cottart, M-A. Syskaki, V. Porée, A. Resta, A. Nicolaou, A. Durnez, S. Ono, A. Mora Hernandez, J. Langer, D. Querlioz and L. Herrera Diez, ‘Dynamic Control of Weight-Update Linearity in Magneto-Ionic Synapses’ Nano Letters 25, 4, 1443–1450 (2025).

Oxide heterostructures are common building blocks for memristive devices, where electron, ion, and thermal transport across the stacked layers critically define device performance for neuromorphic computing applications. In particular, the band alignment between oxides and the spatial distribution of trap states strongly influence current transport mechanisms, thereby shaping the resistance window, switching voltage, and read-out conditions. The relative position of traps governs the dominant transport pathways, while the resulting electric-field distribution sets the switching speed. Trap-induced transient charging further increases required read-out voltages and limits read-out speed.

In this contribution, we present a comparative study of non-filamentary, area-dependent memristive device stacks in which ionic motion and resistance modulation occur across the full device area. We focus on devices employing p-type Pr₀.₇Ca₀.₃MnO₃ (PCMO) and n-type InGaZnO (IGZO) stacked  with other metal oxides. We compare band-bending calculations with experimentally determined band alignments obtained via in-situ XPS studies and operando hard X-ray spectroscopy.  

We discuss how how the babd alignment and the trap positions influence switching polarity, control the transition between transport regimes, and impact switching voltage and read-out delay. Our findings highlight how engineered trap landscapes and interface band structures can be leveraged to optimize non-filamentary memristive device performance.

Halide perovskites (HPs) have gained significant attention in optoelectronics and photovoltaics due to their tunable bandgap, high color purity, and long charge diffusion lengths. However, the application of HPs in resistive random-access memory (RRAM) has been hindered by material and electrical instability in traditional thin-films resulting in subpar figures-of-merit such as retention, endurance, and switching speed [1, 2].

To address these limitations, a novel switching matrix has been developed that replaces the thin-film architecture with vertically aligned, three-dimensional, high-density monocrystalline HP nanowires and quantum wires. These nanostructures are embedded within a porous alumina membrane (PAM) sandwiched between metal contacts. The PAM provides excellent passivation, imparting crucial electrical and material stability to the environmentally sensitive HPs by drastically reducing surface diffusion pathways and thereby thwarting moisture-induced degradation.

This approach yields record-breaking performance for HP-based RRAMs, with extrapolated retention times of up to 28.3 years, measured endurance of 5 million cycles, and switching speeds as fast as 100 picoseconds. These results represent the best figures-of-merit reported for HP RRAMs and stand in stark contrast to HP thin-film devices, which typically demonstrate a maximum endurance of 10 thousand cycles, retention of 10⁵ seconds, and switching speeds of 10 nanoseconds [3, 4]. Furthermore, the scalability potential is demonstrated through the fabrication of a 14 nm lateral size HP quantum wire RRAM cell and the development of a cross-bar device architecture featuring a unique metal-semiconductor-insulator-metal (MSIM) sneaky path mitigation scheme.

Beyond data storage, this technology pioneers the development of HP-based physical neural networks utilizing nanowires in PAM for low-power, high-precision computing. Operation in two distinct modes: electrochemical metallization (ECM) with a silver top electrode and non-electrochemical metallization (non-ECM) with an indium tin oxide top electrode, enables the control of multi-level conductance states. These states are utilized as synaptic weights in artificial neural networks to perform image processing tasks such as outlining, sharpening, and embossing in the ECM mode [5]. In the non-ECM mode, the multi-level states are used to emulate the cognitive learning model of the Gestalt Closure Principle [6].

The use of HP nanowires embedded in a PAM provides superior material reliability, retention, and jitter of the conduction states compared to thin-film counterparts, enabling the construction of a physical artificial vision system. A physical processor is integrated with a physical pre-processor, resulting in reliable photo-synaptic behavior and stable, temporally robust conduction states. The devices exhibit retention greater than 10⁵ seconds and temporal jitter below 10%, and respond to changes in the intensity, duration, and frequency of light pulses, allowing for functionalities such as contrast enhancement. The system's capability was validated by successfully recognizing four different geometric shapes using a 6×6 array of nanowire devices [7].

In summary, these findings advance HP RRAMs to the state-of-the-art standard and showcase the potential of  perovskite nanowire devices as a compelling alternative technology for future data storage and neuromorphic computing modules.

New paradigms in computing architectures require the development of devices based on materials that extend beyond conventional semiconductor components. Memristors[1] have emerged as main building blocks for these next-generation technologies. Current research explores several systems—such as resistive random-access memories, spin-torque magnetic memories, phase-change memories, and other resistive-based elements—as potential memristors. However, all these approaches rely on current injection through the device, which leads to disadvantages primarily related to device reliability and power consumption [2]. Ferroelectric materials can overcome many of these challenges. As insulating materials capable of operating under open-circuit conditions, they avoid current flow during switching and offer inherently low-power operation.[2] Despite these advantages, their technological impact was limited by the difficulty of integrating ferroelectric compounds into highly dense memory architectures [3,4]. This situation changed with the discovery of ferroelectricity in doped HfO₂ [5]. Hafnium oxide—and its doped variants—is fully compatible with CMOS processes, making it an attractive material for neuromorphic devices where it can function as a memristive element [6]. In recent years, progress in understanding the memristive and neuromorphic characteristics of ferroelectric HfO₂ has been remarkable. The majority of studies have focused on polycrystalline films, which are directly compatible with industrial fabrication. However, their functional properties are difficult to interpret because polycrystalline hafnia typically contains mixed phases, multiple orientations, and diffuse interfaces. These structural complexities hinder a clear understanding of the intrinsic ferroelectric and neuromorphic behavior.

Epitaxial ferroelectric HfO₂ films offer a powerful alternative platform for fundamental studies [7]. These films exhibit atomically flat surfaces, well-defined interfaces, and single-phase structures, enabling more precise investigation of switching mechanisms. Epitaxial films also avoid the wake-up effect commonly observed in polycrystalline hafnia [8], and they show fast neuromorphic responses [9] together with large polarization and exceptional endurance [10] even in ultrathin layers [11]. These features position epitaxial ferroelectric HfO₂ as an excellent model system for exploring new phenomena relevant to multi-input/multi-output neuromorphic devices. In this work, we highlight several recent developments using epitaxial ferroelectric HfO₂ systems. First, we present the creation of a robust multiferroic heterostructure capable of achieving magnetoelectric neuromorphic-like responses [12]. This system demonstrates the potential of coupling ferroelectric and magnetic orders to achieve neuromorphic magnetoelectric response. Second, we summarize our recent investigations into the interaction between ferroelectric order and light. We show that photovoltaic conversion can be used to monitor polarization via short-circuit photocurrent signals [13]. This voltage-free readout enables non-destructive detection of multilevel ferroelectric states—an especially valuable feature for neuromorphic architectures requiring energy efficiency. In addition, light can be used to remotely manipulate the ferroelectric polarization state, enabling multi-state optoelectrical response. Taken together, these results underscore the broad potential of epitaxial ferroelectric HfO₂ films for advancing neuromorphic device technologies.

Filamentary oxide-based memristors, especially those based on HfO₂, have attracted significant interest due to their CMOS compatibility, well-defined resistive states, fast switching speed, and high endurance [1]. Their operation relies on the reversible formation and partial disruption of oxygen-vacancy conductive filaments, driven by redox reactions and field-assisted ion migration within the oxide layer [2]. A detailed understanding of forming, SET, and RESET processes reveals the strong dependence of switching characteristics on voltage amplitude, current compliance, and material stack engineering [3]. Moreover, multilevel resistance states can be achieved by carefully tuning the applied voltage during programming, enabling fine control of filament dimensions and conductance levels [3]. A major challenge for these devices is their intrinsic cycle-to-cycle variability. Experimental data show significant dispersion in SET/RESET voltages, arising from the stochastic nature of vacancy generation, filament regrowth, and gap evolution across cycles [4]. Reliability concerns also include random telegraph noise (RTN), caused by trapping/detrapping events near the filamentary path, as well as irreversible current fluctuations associated with structural modifications within or around the conductive filament [5]. In long-term operation, the memory window gradually narrows due to oxide degradation mechanisms that affect filament stability. Beyond switching performance, these devices exhibit synaptic-like behavior such as gradual potentiation and depression under pulsed stimuli, together with multilevel conductance states, and they can reproduce spike-timing-dependent plasticity (STDP), where the conductance change depends on the relative timing between pre- and post-synaptic spikes [6,7]. Although challenges remain regarding linearity, asymmetry, and energy efficiency, these characteristics highlight the versatility of filamentary oxide-based memristors and their potential for a broad range of emerging electronic applications.

Halide perovskite memristors have been reported to exhibit a wide variety of switching mechanisms depending on material composition, device architecture, and interfacial design, leading to diverse physical interpretations and device behaviors, while comprehensive material–device design rules and physically grounded modeling frameworks remain limited.¹

Here, we present a combined experimental and modeling study of low-voltage resistive switching in lead-free bismuth halide perovskite memristors, establishing a quantitative link between microscopic ion dynamics and macroscopic hysteresis behavior. Ag/Cs₃Bi₂I₉₋ₓBrₓ/ITO devices with I-rich (x = 3) and Br-rich (x = 6) compositions crystallize in a layered trigonal structure and form highly uniform thin films. Both devices exhibit reproducible bipolar switching with SET/RESET voltages below 0.3 V, ON/OFF ratios >10¹, and excellent cycling and retention stability.

Crucially, this work provides the first direct experimental validation of the conductance-activated quasi-linear memristor (CALM) framework in bismuth-based halide perovskite systems,^2,3 showing quantitative agreement between measured and simulated I–V hysteresis across multiple scan rates. Physical modeling reveals ion-migration-controlled filament dynamics, with I-rich devices forming more stable filaments due to reduced halide vacancy density, while Br-rich devices achieve ultralow-voltage operation at the expense of slightly broader switching distributions. These results demonstrate that compositional engineering in lead-free bismuth perovskites enables precise control over the trade-off between switching voltage and stability, further strengthened by dynamic physical modeling, positioning this materials platform for energy-efficient non-volatile memory and neuromorphic hardware.

Magneto-ionic materials, which enable non-volatile control of magnetism through voltage-driven ion migration, are emerging as promising candidates for neuromorphic computing [1, 2]. Unlike conventional memristors, these systems typically operate through dual-actuation protocols, involving both electric and magnetic fields, thus providing a broader range of functional capabilities. The utilization of voltage rather than electric currents significantly reduces Joule heating effects and enhances energy efficiency. Remarkably, voltage-triggered ion motion can induce the formation of ferromagnetic regions within films that are initially paramagnetic. However, the general need for external magnetic fields to control the orientation of the voltage-induced ferromagnetic phases remains a key limitation, undermining the full energy-saving potential of these systems. In this work, we present a magneto-ionic strategy in CoFeN that fully decouples the electric and magnetic field actuation requirements. First, we apply magnetic field to fix a pre-defined direction of the magnetization in the magneto-ionically generated ferromagnetic phase. Then, once the magnetic field is removed, we demonstrate the modulation of magnetic remanence solely with the applied voltage. Such tuning of the remanent magnetization state is enabled by the voltage-controlled propagation of a planar N³⁻ ion migration [3], along with the ferromagnetic exchange interactions between pre-existing and newly generated CoFe magnetic regions. The system exhibits behaviors reminiscent of neuromorphic-inspired functionalities, such as synaptic potentiation and depression [4], while also showing a cumulative, voltage-driven increase in magnetization in the absence of a magnetic field. Once the magnetic field is switched off, synaptic weight remains influenced by the sample’s magnetic and electric history. By eliminating the need for magnetic fields, our approach contributes to reduce energy consumption, decreasing the amount of energy spent typically using these systems by several orders of magnitude, thereby offering a more efficient pathway for brain-inspired magneto-ionic devices.

The exponential rise of Big Data is driving a dramatic increase in energy consumption by information technologies [1]. One reason is that most memory systems utilize electric currents to write data, which inherently dissipates energy through Joule heating [2]. Electric-field control of magnetic properties has emerged as a sustainable leading strategy to address this issue. Among the diverse voltage-driven mechanisms for the tuning of magnetism (e.g. electrostatic charging, strain-mediated multiferroic coupling or electrochemical reactions [3]), control of magnetism through electric-field-induced ion motion in magneto-ionics is rapidly gaining momentum. Magneto-ionics provides unprecedented non-volatile control of coercivity, anisotropy, exchange bias or magnetization, ultimately enabling conversion between magnetic and non-magnetic states [4,5]. However, despite the pressing need for strategies to control magnetic states at the nanoscale, magneto-ionics has largely been confined to continuous thin films, leaving nanoscale magnetic architectures underexplored.

In this talk, we will demonstrate in-situ probing of magnetic properties and associated ion dynamics in nanometer-scale structures, revealing dynamically evolving spin configurations that strongly depend on the gating duration. More precisely, we will introduce a so far unexplored nanoscale magnetic object: magnetic vortex controlled by electric-field-driven ion motion, termed the magneto-ionic vortex or, for simplicity, “vortion” [6]. Vortions are generated within initially paramagnetic FeCoN nanodots via voltage-driven gradual extraction of N^3– ions. What distinguishes vortions from conventional magnetic vortex states is that their key properties such as magnetization amplitude, nucleation and annihilation fields, coercivity, remanence and anisotropy, can be controlled and fine-tuned post-synthesis in an analog, reversible and energy-efficient manner [6]. This obviates the need for energy-demanding methods like laser pulses or spin-torque currents. Such tunability is made possible by taking advantage of a so far overlooked aspect of N^3– magneto-ionics, namely the occurrence of a planar ion migration front, which allows precise, post-synthesis control of the magnetic layer's thickness. Consequently, we demonstrate voltage-mediated transitions between paramagnetic, single-domain, and vortion states, unlocking a new paradigm for energy-efficient control of magnetism at the nanoscale.

This unprecedented level of control over magnetic properties at the nanoscale and at room temperature opens new horizons for the development of advanced magnetic devices with functionalities that can be tailored at the post-synthesis stage, therefore providing enhanced flexibility, needed to meet specific technological demands. Magneto-ionic states induced within patterned units enable transformative potential for neuromorphic devices [7], analog computing, multi-state data storage, and adaptive inference protocols implemented directly within magneto-ionic architectures [8]. Moreover, we find that the inherent physical nature of patterned magneto-ionic systems enables robust, self-protected hardware security primitives–crucial as Big Data growth outpaces the reliability of software-based defenses [8]. These findings pave the way for a new class of hardware security solutions rooted in emergent magnetic phenomena.

Our body is a self-regulating system, able to adapt by autonomously adjusting its internal processes in response to changes in the environment, to keep vital internal variables constant, such as body temperature, blood pressure, pH levels, etc. [1] To do so, the body uses sensors, a control center and effectors distributed over the entire organism, together with feedback loops which keep internal conditions stable. This can be considered a kind of biological intelligence through adaptive regulation mechanisms, in which internal feedback loops play a crucial role.^[2,3] Brain itself is a self-regulating system, whose ability to perform efficient and fault-tolerant data processing is strongly related to its peculiar interconnected adaptive architecture, based on redundant neural circuits interacting at different scales. By emulating these systems' processing and learning mechanisms, novel neuromorphic computing technologies can strive to achieve higher levels of energy efficiency and computational performance.^[4]

Here the implementation of neuromorphic data processing devices based on self-assembled nanostructured thin films^[5,6], able to adapt to electrical and environmental stimuli, is explored. In operando observation of the structural and morphological reorganization of the nanostructured films at the nano, meso and macroscale under applied bias and ambient conditions is first reported. Furthermore, an experimental strategy based on micro-thermography is exploited for the study of the spatial and temporal dynamic of the resistive switching (RS) activity of the reorganizing networks. In particular, we investigated the control over the synchronous activity and connectivity of the micrometric active sites which rule the emerging network dynamic, of paramount importance for the correct performance of the data processing devices.

The RS activity of these materials is also described in response to mechanical and environmental stimuli (for instance, temperature and humidity),^[7] allowing the interplay between the sensing and the embodied processing capabilities of these multifaceted materials.

The controlled programmability of ns-Au network’s connectivity, by means of mechanical and electrical inputs, demonstrates the capability of our devices to encode external stimuli into specific connectivity, which can last in time and be reprogrammed on demand. This enables the fabrication of data processing devices based on self-assembled systems, such as reconfigurable nonlinear threshold logic gates,^[4,7,8] featuring tailored and adaptable connectivity that allows the controlled emerging responses of the complex networks. We also demonstrate the potentiality of the use of these self-assembled reconfigurable devices to classify with high accuracy and in real-time neuronal traces corresponding to physiological and evoked spiking activity recorded from the barrel cortex of a rat.^[9]

The use of neuromorphic self-assembled materials with complex wiring and redundant morphological features is proposed for the implementation of energy-efficient reprogrammable edge computing devices.  

The rapid expansion of machine learning capabilities is driven by the exponentially increasing complexity of deep neural network (DNN) models, which demand hardware that is both energy‑efficient and chip‑area efficient to handle computationally intensive inference and training tasks. Electrochemical random‑access memories (ECRAMs) have emerged as a promising solution, specifically designed to enable efficient analog in‑memory computing for these data‑intensive workloads. In this talk, I will present a CMOS‑compatible ECRAM prototype fabricated with inorganic metal oxides. The device operates by shuttling protons within a symmetric gate stack composed of a zirconium oxide protonic electrolyte sandwiched between a hydrogenated tungsten oxide channel and gate. This architecture yields nearly perfectly symmetric programming characteristics with exceptionally low cycle‑to‑cycle variability (<1%) under voltage pulse operation. By optimizing zirconium oxide stoichiometry, the prototype achieves fast operation with latency down to 100 nanoseconds, endurance exceeding 10⁸ cycles, robust retention of analog memristive states. and ultra‑low energy consumption (<1 femtojoule per weight update). These ECRAMs can be monolithically integrated on top of silicon electronics to form pseudo‑crossbar arrays. The test chips function as in-memory computing processing elements to accelerate both inference and training of deep neural networks.