True random number generators (TRNGs) are essential for hardware security in edge AI systems, yet conventional designs often incur large analog overhead and limited throughput imposed by standalone macros. We present a Compute-in-Memory (CiM)-compatible TRNG that directly exploits the intrinsic stochasticity of TaOx-based 1T1R RRAM arrays, enabling entropy extraction within the memory fabric. Bit-rate is scaled by parallel column readout circuits consisting of a transimpedance amplifier (TIA) and an ADC. Our design achieves up to similar to 270 Mbps throughput with TIA + 16-bit ADC per column. Furthermore, a lightweight shift-XOR post-processing stage permits reduction to 8-bit ADC resolution, lowering energy consumption to similar to 51 pJ bit-1 without degrading randomness quality. Fully compatible with standard CiM read paths, the architecture introduces minimal hardware overhead and provides a scalable and energy-efficient foundation for secure random number generation in edge-AI applications.

Epitaxial heterostructures of iron oxide thin films on oxide substrates are promising for spintronic applications. Scaling down such heterostructures into ordered nanostructures enables integration into functional devices. However, fabricating well-ordered nanostructures while retaining their structural and functional integrity remains challenging. A bottom-up approach is used to grow epitaxial Fe3O4 nanodot arrays on Nb-doped SrTiO3(Nb:STO) substrates integrating anodic aluminum oxide (AAO) templates combined with pulsed laser deposition. Following this method, the lateral confinement of Fe3O4 into 30 and 70 nm nanodots with 3D long-range ordering is then confirmed by grazing-incidence small-angle X-ray scattering (GISAXS) and scanning electron microscopy (SEM). Building on this structural evidence, the Verwey transition is found to be retained, as observed in the continuous film. To explore its applicability in nanoelectronics, conductive atomic force microscopy (c-AFM) is used to probe local electrical behavior, and reveals bipolar resistive switching at room temperature in individual nanodots, consistent with behavior observed in thin films. Together, these results establish the first demonstration of ordered epitaxial Fe3O4/Nb:STO nanodots with preserved structural, magnetic, and electrical functionalities, providing a generalizable route for nanoscale integration of complex oxides.

Engineering complex oxide heterostructures and their interfaces has revealed a plethora of emergent electronic and magnetic properties. To achieve these functionalities beyond conventional epitaxy, building heterogeneous integrated electronic architectures have been possible from the advances in free-standing oxides and their integration with semiconductors. However, to harness the physical phenomena of oxides in such co-integrated environments, it is necessary to achieve atomically defined membrane-based oxide heterostructures and subsequent control of charge-transfer. Here, we report on the direct growth-control of the surface termination of SrTiO3 membranes, eliminating any B-HF requirements, and subsequent transfer of TiO2-terminated SrTiO3 membranes onto silicon (Si) via a sacrificial layer route. This approach yields atomically defined step-terraced membrane-based substrates on silicon support. By systematic growth control of LaAlO3 on these templates, we demonstrate distinct signatures of oxygen-vacancy-induced (ionic) and intrinsic (electronic) charge transfer mechanisms. A systematic crossover between these processes is observed, based on near-ambient pressure XPS, probing reversible and irreversible contributions of interfacial charge transfer during redox-cycling. These results indicates that TiO2-termination in the oxide membrane may have been achieved, which is a prerequisite for tailoring and fine-tuning membrane-based oxide heterointerfaces for electronic and ionic phenomena in confined systems beyond conventional epitaxy.

Voltage-controlled ion insertion provides a powerful strategy for the analog tuning of material properties, enabling adaptive devices such as neuromorphic transistors and smart displays. Among tunable materials, mixed ionic-electronic conducting oxides undergoing topotactic phase transitions are particularly compelling due to their dramatic property changes between fully oxidized and fully reduced states. However, intermediate oxidation states remain largely underexplored because of significant control limitations. In this work, we investigate the topotactic phase transition in strontium ferrite (SrFeO3-δ) thin films by progressively and precisely modulating and quantifying oxygen non-stoichiometry via solid-state electrochemical pumping. This fine-tuning approach unveils the co-existence of multiple stable phases in equilibrium configurations across a broad range of oxidation states. A crystallographic mixing model that captures the structural-electronic coupling underlying this phenomenon is proposed, complemented by a defect chemistry framework that quantitatively describes the oxidation mechanism under applied voltage. These findings highlight the critical role of intermediate states in governing functional properties and open new pathways for designing advanced ionotronic oxygen-responsive devices.

Emerging Kolmogorov–Arnold networks (KANs) replace the linear weights of neural networks with trainable nonlinear functions. This modification is particularly attractive for scientific computing, where KANs can match the accuracy of conventional multilayer perceptrons (MLPs) while reducing model size by up to 100×. However, this efficiency comes at the cost of computationally expensive nonlinear evaluations, unlike conventional MLPs dominated by linear matrix multiplications. We present a flexible and energy‐efficient compute‐in‐memory accelerator tailored for KANs, developed through cross‐layer optimization across algorithm, architecture, circuit, and device levels. The accelerator computes arbitrary nonlinear functions using a single‐read scheme and read‐optimized memory arrays with nonvolatile memristive devices. Our system achieves a lowest energy of 8.69 pJ per KAN function. In terms of energy‐delay product, it provides 1996× improvement over CPUs, 208× over standard MLP‐oriented compute‐in‐memory accelerators, and up to 71× over prior KAN accelerators. These results establish energy‐efficient hardware primitives for implementing advanced nonlinear networks in scientific computing.

Filamentary valence change mechanism (VCM)-type memristive devices based on transition metal oxides offer great potential for the realization of energy-efficient analog hardware accelerators used in machine learning and neuromorphic computing. To fully exploit this potential, integration of nanostructured memristive devices with complementary metal oxide semiconductor (CMOS) circuits and multi-level programming are essential prerequisites. In crossbar arrays for in-memory computing, a transistor in series with the VCM device acts as a selector and limits the current in the SET process, which allows programming of different low-resistance states (LRS). However, a discrepancy between the programmed LRS value and the measured conductivity value, G LRS, is often observed, even for VCM devices with linear current-voltage characteristics. In this study, we analyze the physical origin of this effect. Therefore, 100 nm & times; 100 nm-sized HfO2-based VCM devices were integrated on foundry-built 180 nm CMOS wafers. Through transient analysis of the device response during the SET, we show that the conductivity of the VCM cell continuously increases for the duration of the SET pulse. This finding could be understood from physical simulation through thermally assisted ion migration in the filament region. This insight can support the development of devices with improved accuracy under multi-state programming.

In-memory computing (IMC) with memristive crossbar arrays offers a promising solution to the energy and latency limitations of conventional von Neumann architectures. However, most existing IMC systems rely on perceptron-based neurons, limiting computational flexibility and requiring large, dense networks to achieve competitive performance. Inspired by the nonlinear processing of biological dendrites, this work introduces a hardware-efficient IMC architecture, “DARWIN”, that integrates nonvolatile memristive devices as synaptic weights arranged as a tree with volatile memristive devices functioning as integration filters. To further optimize hardware utilization, we incorporate learnable sparsity, enabling the network to automatically discover a compact synaptic pattern within a dendritic tree tailored to the underlying device characteristics. Experimental results demonstrate that the proposed architecture achieves more than two orders of magnitude lower memory footprint and at least an order of magnitude lower power consumption than other in-memory computing architectures. These findings highlight the potential of combining dendritic computation, heterogeneous memristive technologies, and sparsity-aware learning to advance scalable and bio-inspired in-memory computing hardware.

Volatile memristive devices with controllable temporal dynamics enable adaptation to diverse temporal coding tasks, making them attractive for low-power neuromorphic edge applications that process asynchronous sensory streams. However, most reported volatile memristors rely on abrupt filamentary switching, suffering from high variability, unreliable operation, and the need for current compliance and forming steps. In this work, an area-dependent volatile memristive device based on a Pt/ α - SrTiO 3 / TaO x /Ta stack is presented, exhibiting ionic-based volatility and a CMOS BEOL-compatible process flow. The device combines low variability with forming-free, self-compliant, rectifying operation and gradual, analog-like switching, achieving an ON/OFF ratio of ∼ 10 3 . An in-depth experimental and physical analysis of the mechanisms governing current transport and volatile behavior is provided. By exploiting the device's rich ionic dynamics, its volatility can be systematically tuned through applied stimulus and stack engineering, yielding decay time constants from tens of milliseconds to several seconds. A reliability study, including endurance and multilevel operation, demonstrates reproducible access to distinct volatile states. Finally, it is shown how a complete system, combining the device with spiking neuron models and CMOS circuitry, could exploit its programmable temporal dynamics for event-based vision, with the wide range of decay constants supporting multiple sensory modalities.

Logic-in-memory (LiM) has emerged as a promising paradigm to address the von Neumann bottleneck by integrating data storage and in-situ computation. Resistive random-access memory (RRAM) is a strong candidate for LiM owing to its non-volatility, fast switching characteristics, and compatibility with CMOS integration. However, intrinsic cycle-to-cycle and device-to-device variability fundamentally limits logic reliability in RRAM-based LiM architectures, necessitating well-defined device specifications and variability margins to ensure correct operation. In this work, we experimentally demonstrate a CMOS-integrated TaO _x -based 1T1R RRAM computing fabric that supports reconfigurable LiM operations, including a functionally complete Boolean set and in-memory arithmetic primitives. Through extensive statistical measurements and variability-aware statistical modeling, we systematically evaluate the impact of key variability sources, SET voltage ( V _SET ), low resistance state, and high resistance state, on logic correctness and identify the dominant contributors to computational failure. Finally, we derive the device specifications and requirements to achieve error-free stateful LiM operations in the 1T1R RRAM arrays.

Heavily doped perovskite oxide thin films have emerged as versatile catalysts based on exsolution processes, which yield functional nanoparticles, but also introduce dopant inhomogeneities and internal phase separation. Here, we resolve the initial-stage of thin film growth in heavily Ni-doped and (Nb,Ni)-co-doped SrTiO3-δ utilizing scanning tunneling microscopy and spectroscopy (STM/S). While the growth of co-doped SrTiO3-δ follows statistically distributed island nucleation, Ni-doped SrTiO3-δ (STNi) exhibits a distinct pentagonal or hexagonal pattern of monolayer islands. A similar pattern is observed in thicker films by transmission electron microscopy (TEM). STS reveals Ni-dopant clusters forming from the onset of the thin film growth and serving as nucleation points for embedded nanostructures at increasing film thickness. Quantitative analysis reveals that in the as-grown monolayers, approximately 20% of the Ni ions segregate into nanoclusters, while the remaining ∼80% are dissolved in the perovskite lattice. Upon reduction (600°C, UHV), the entire Ni content of the sub-monolayer STNi is found to nucleate as metallic nanoparticles at the surface, revealing two exsolution pathways from the perovskite lattice as well as from pre-defined clusters. These findings provide deeper insights into the nucleation processes in heavily doped oxide and further reveal the role of dopant inhomogeneities for metal exsolution reactions.

Metal exsolution is explored as a means of tuning catalytic activity in perovskite oxides for the oxygen evolution reaction in aqueous medium. Controlling nanoparticle size and distribution through external reaction parameters is inherently coupled to the simultaneous evolution of the functional properties of the exsolved nanoparticle population and the host oxide under reducing synthesis conditions. Our analysis reveals that insufficient control over electronic properties of both the exsolved nanoparticles and the oxide support can fundamentally decouple the performance of exsolution catalysts from morphological descriptors such as nanoparticle density. Building on these findings, we demonstrate that metal exsolution can be employed to achieve catalytic performance comparable to that of established high-performance perovskite oxide electrocatalysts which rank close to the top of the volcano plot, while reducing the required amount of critical transition metals by a factor of 20, as evaluated through activity and stability testing.

ABSTRACT Heavily doped perovskite oxide thin films have emerged as versatile catalysts based on exsolution processes, which yield functional nanoparticles, but also introduce dopant inhomogeneities and internal phase separation. Here, we resolve the initial‐stage of thin film growth in heavily Ni‐doped and (Nb,Ni)‐co‐doped SrTiO 3‐δ utilizing scanning tunneling microscopy and spectroscopy (STM/S). While the growth of co‐doped SrTiO 3‐δ follows statistically distributed island nucleation, Ni‐doped SrTiO 3‐δ (STNi) exhibits a distinct pentagonal or hexagonal pattern of monolayer islands. A similar pattern is observed in thicker films by transmission electron microscopy (TEM). STS reveals Ni‐dopant clusters forming from the onset of the thin film growth and serving as nucleation points for embedded nanostructures at increasing film thickness. Quantitative analysis reveals that in the as‐grown monolayers, approximately 20% of the Ni ions segregate into nanoclusters, while the remaining ∼80% are dissolved in the perovskite lattice. Upon reduction (600°C, UHV), the entire Ni content of the sub‐monolayer STNi is found to nucleate as metallic nanoparticles at the surface, revealing two exsolution pathways from the perovskite lattice as well as from pre‐defined clusters. These findings provide deeper insights into the nucleation processes in heavily doped oxide and further reveal the role of dopant inhomogeneities for metal exsolution reactions.

ABSTRACT Memristive Pr 0.7 Ca 0.3 MnO 3 (PCMO) heterostructures exhibit area‐dependent resistive switching via a valence change mechanism, making them promising for neuromorphic architectures. A major challenge in PCMO‐based memory is higher‐dimensional lattice defects that affect oxygen‐vacancy migration and concentration. This study mitigates these defects using highly conductive amorphous PCMO fabricated via a CMOS back‐end‐of‐line‐compatible process and compares it with low‐conductive amorphous and polycrystalline PCMO. The resistance differences are attributed to changes in electronic mobility, based on the analysis of short‐ and long‐range order, Mn–O hybridization, and Mn valence state. AlO x /qa‐PCMO devices showed the highest ON/OFF ratio compared to low‐conductive amorphous and polycrystalline PCMO, because the field‐accelerated oxygen vacancy movement switches the mechanism from Poole–Frenkel emission in the LRS to trap‐assisted tunneling in the HRS. The mechanism change was identified by systematically analyzing the I–V asymmetry, device band diagrams for different PCMO types, and shape changes in the I–V curve fits. The band diagrams were calculated from the measured bandgaps and work functions of the different PCMO types. Analysis of the electric field distribution in the devices showed a clear correlation between the pre‐switching field strength in AlO x and the resulting ON/OFF ratio.

Raman spectroscopy was employed to investigate the topotactic phase transition in SrFeOx thin films for memristive device applications. This study provides direct spectroscopic evidence of local structural changes associated with the redox-driven phase transition. In support of our previous reports, which primarily focused on X-ray absorption spectroscopy, our current operando Raman analysis reveals that the formation of conductive filaments involves not only variations in oxygen content but also a distinct structural transformation within the perovskite lattice. Notably, a much smaller change in oxygen content is sufficient to trigger significant alterations in the electronic properties, leading to the formation of conducting pathways in these memory devices. The spatially resolved Raman intensity mapping further confirms the localized nature of this structural phase transition within the active region of the device.

This work presents OTTER, a 28 nm CMOS platform co-integrated with TaOx-based valence-change mechanism (VCM) RRAM, demonstrating a two-transistor-one-memristive-device (2T1R) architecture for reliable in-memory computing. The 2T1R cell combines a low-drive-current (LD) transistor and a high-drive-current (HD) transistor in parallel, providing dedicated bias paths for SET programming and RESET operation, respectively. Through systematic experimental and simulated comparison of various transistor-pairing configurations using the physical compact model JART VCM Rth, design guidelines for transistor sizing are derived, establishing the minimum RESET transistor W/L required for complete RESET as a function of the SET current compliance. The 2T1R cell is further characterized under pulse-based programming, demonstrating multilevel analog conductance tuning with narrow, well separated conductance states across six programmable levels. An analog content-addressable memory (aCAM) design based on the same 2T1R cell is additionally analyzed at the circuit level, evaluating trade-offs between top- and bottom-connected RRAM comparator configurations. A hardware implementation of compute-in-memory (CIM) multiply-and-accumulate (MAC) operations is further demonstrated on a 15 x 15 2T1R crossbar array.

Area-dependent memristive devices based on the valence change mechanism are promising candidates for emerging analog and neuromorphic computing due to their intrinsic analog switching behavior and reduced variability. Among these, IGZO-based devices are particularly attractive owing to their potential for multifunctional applications, including optoelectronics and flexible electronics. However, systematic design strategies that directly link the material and interface properties to their device performance remain unexplored. In this work, we demonstrate that the switching polarity and electrical characteristics of IGZO-based memristive devices can be systematically tuned by modifying the top electrode. This control is shown to originate from changes in the band alignment and the resulting spatial electric field distribution across the device. To identify the governing interface and underlying transport mechanisms, we combine energy band diagram simulations with XPS-based band alignment measurements. Using this experimentally validated framework, the measured I–V characteristics are quantitatively reproduced within the Tsu–Esaki tunneling model. These results establish a consistent physical understanding of switching in IGZO-based devices and demonstrate that band engineering provides a powerful route to control both transport and switching behavior, enabling targeted optimization for large-scale analog and neuromorphic systems.

The effect of X-ray nano-beam irradiation has been investigated, along with a subsequent electric discharge between the metallic electrodes of memristive-like devices fabricated on top of rutile single crystals. X-ray irradiation with a photon flux of about 6 x 10(12) photons/second impinging on an area of 74 x 60 nm(2), and with irradiation times of the order of tens of seconds per point, has been able to pattern a conducting path between the electrodes, which has guided the electric discharge process. The results from Ti K-edge XANES mapping with submicron spatial resolution show small changes in the configuration of the pre-edge peaks, suggesting the presence of anatase phase in the affected region, which is also confirmed by micro-Raman mapping. This implies the occurrence of very intense Joule heating during the electric discharge (T > 2140 K), followed by a very quick cooling process. This conversion of rutile to anatase represents a novel electrical method for the formation of localized anatase nanoparticles from rutile single crystals, which may be of help for specific applications.