In situ Parallel Training of Analog Neural Network Using Electrochemical Random-Access Memory

CMOS-Compatible Protonic Three-Terminal Memristor for Analog Synapse in Neuromorphic Computing

All-solid-state inorganic hydrogen-based three-terminal memristors (H-3TMs) suffer from poor retention, susceptibility to humidity and temperature, and the reliance on wet chemistry during fabrication, hindering their manufacturability within existing foundry processes. To address these, this study presents a CMOS-compatible H-3TM based on reversible intercalation and extraction of protons between the SiNx electrolyte and WOx channel. The protons are introduced via a straightforward hydrogen plasma treatment, promoting a compatible fabrication process with back-end-of-line integration. Experimental and simulation results indicate that the low proton transport tendency across the electrolyte/channel interface without an external electric field contributes to high retention performance. Furthermore, the device demonstrates linear potentiation and depression, 512 conductance states with a dynamic range of ≈40, low energy operation (≈73 fJ per write), and excellent overall device-to-device variation. Its analog properties are evaluated under the training and inference framework of MNIST and Fashion-MNIST datasets. The device achieved training and inference accuracies only 0.4% and 0.3% below the ideal benchmark on the F-MNIST dataset. This work offers a rational approach for future artificial synaptic device design and fabrication.

Unveiling ECRAM switching mechanisms using variable temperature Hall measurements for accelerated AI computation

Electrochemical random-access memory devices are promising for analog cross-point array-based artificial intelligence accelerators due to their high stability and programmability. However, understanding their switching mechanism is challenging due to complex multilayer structures and the high resistivity of oxide materials. Here, we fabricate multi-terminal Hall-bar devices and conduct alternating current magnetic parallel dipole line Hall measurements to extract transport parameters. Through variable-temperature Hall measurements, we determine the oxygen donor level at approximately 0.1 eV in tungsten oxide and reveal that conductance potentiation even at low temperatures results from increased mobility and carrier density. This behavior is linked to reversible electronic and atomic structure changes, supported by density functional theory calculations. Our findings enhance the understanding of electrochemical random-access memory switching mechanisms and provide insights for improving high-performance, energy-efficient artificial intelligence computation in analog hardware.

Organic Artificial Nerves: Neuromorphic Robotics and Bioelectronics

Neuromorphic electronics are inspired by the human brain's compact, energy-efficient nature and its parallel-processing capabilities. Beyond the brain, the entire human nervous system, with its hierarchical structure, efficiently preprocesses complex sensory information to support high-level neural functions such as perception and memory. Emulating these biological processes, artificial nerve electronics have been developed to replicate the energy-efficient preprocessing observed in human nerves. These systems integrate sensors, artificial neurons, artificial synapses, and actuators to mimic sensory and motor functions, surpassing conventional circuits in sensor-integrated electronics. Organic synaptic transistors (OSTs) are key components in constructing artificial nerves, offering tunable synaptic plasticity for complex sensory processing and the mechanical flexibility required for applications in soft robotics and bioelectronics. Compared to traditional sensor-integrated electronics, early implementations of organic artificial nerves (OANs) incorporating OSTs have demonstrated a higher signal-to-noise ratio, lower power consumption, and simpler circuit designs along with on-device processing capabilities and precise control of actuators and biological limbs, driving progress in neuromorphic robotics and bioelectronics. This paper reviews the materials, device engineering, and system integration of the OAN design, highlights recent advancements in neuromorphic robotics and bioelectronics utilizing the OANs, and discusses current challenges and future research directions.

Bio-plausible reconfigurable spiking neuron for neuromorphic computing

Biological neurons use diverse temporal expressions of spikes to achieve efficient communication and modulation of neural activities. Nonetheless, existing neuromorphic computing systems mainly use simplified neuron models with limited spiking behaviors due to high cost of emulating these biological spike patterns. Here, we propose a compact reconfigurable neuron design using the intrinsic dynamics of a NbO2-based spiking unit and excellent tunability in an electrochemical memory (ECRAM) to emulate the fast-slow dynamics in a bio-plausible neuron. The resistance of the ECRAM was effective in tuning the temporal dynamics of the membrane potential, contributing to flexible reconfiguration of various bio-plausible firing modes, such as phasic and burst spiking, and exhibiting adaptive spiking behaviors in changing environment. We used the bio-plausible neuron model to build spiking neural networks with bursting neurons and demonstrated improved classification accuracies over simplified models, showing great promises for use in more bio-plausible neuromorphic computing systems.

Trivalent Ionic Molecular Bridges as Efficient Charge-Trapping Method for All-Solid-State Organic Synaptic Transistors toward Neuromorphic Signal Processing Applications

Achieving high retention of memory state is crucial in artificial synapse devices for neuromorphic computing systems. Of various memorizing methods, a charge-trapping method provides fast response times when it comes to the smallest size of electrons. Here, for the first time, it is demonstrated that trivalent molecular bridges with three ionic bond sites in the polymeric films can efficiently trap electrons in the organic synaptic transistors (OSTRs). A water-soluble polymer with sulfonic acid groups, poly(2-acrylamido-2-methyl-1-propanesulfonic acid) (PAMPSA), is reacted with melamine (ML) to make trivalent molecular bridges with three ionic bond sites for the application of charge-trapping and gate-insulating layer in all-solid-state OSTRs. The OSTRs with the PAMPSA:ML layers are operated at low voltages (≤5 V) with pronounced hysteresis and high memory retention characteristics (ML = 25 mol%) and delivered excellent potentiation/depression performances under modulation of gate pulse frequency. The optimized OSTRs could successfully process analog (Morse/Braile) signals to synaptic current datasets for recognition/prediction logics with an accuracy of >95%, supporting strong potential as all-solid-state synaptic devices for neuromorphic systems in artificial intelligence applications.

Vector Symbolic Finite State Machines in Attractor Neural Networks

Hopfield attractor networks are robust distributed models of human memory, but they lack a general mechanism for effecting state-dependent attractor transitions in response to input. We propose construction rules such that an attractor network may implement an arbitrary finite state machine (FSM), where states and stimuli are represented by high-dimensional random vectors and all state transitions are enacted by the attractor network's dynamics. Numerical simulations show the capacity of the model, in terms of the maximum size of implementable FSM, to be linear in the size of the attractor network for dense bipolar state vectors and approximately quadratic for sparse binary state vectors. We show that the model is robust to imprecise and noisy weights, and so a prime candidate for implementation with high-density but unreliable devices. By endowing attractor networks with the ability to emulate arbitrary FSMs, we propose a plausible path by which FSMs could exist as a distributed computational primitive in biological neural networks.

Electrochemical random-access memory: recent advances in materials, devices, and systems towards neuromorphic computing

Artificial neural networks (ANNs), inspired by the human brain's network of neurons and synapses, enable computing machines and systems to execute cognitive tasks, thus embodying artificial intelligence (AI). Since the performance of ANNs generally improves with the expansion of the network size, and also most of the computation time is spent for matrix operations, AI computation have been performed not only using the general-purpose central processing unit (CPU) but also architectures that facilitate parallel computation, such as graphic processing units (GPUs) and custom-designed application-specific integrated circuits (ASICs). Nevertheless, the substantial energy consumption stemming from frequent data transfers between processing units and memory has remained a persistent challenge. In response, a novel approach has emerged: an in-memory computing architecture harnessing analog memory elements. This innovation promises a notable advancement in energy efficiency. The core of this analog AI hardware accelerator lies in expansive arrays of non-volatile memory devices, known as resistive processing units (RPUs). These RPUs facilitate massively parallel matrix operations, leading to significant enhancements in both performance and energy efficiency. Electrochemical random-access memory (ECRAM), leveraging ion dynamics in secondary-ion battery materials, has emerged as a promising candidate for RPUs. ECRAM achieves over 1000 memory states through precise ion movement control, prompting early-stage research into material stacks such as mobile ion species and electrolyte materials. Crucially, the analog states in ECRAMs update symmetrically with pulse number (or voltage polarity), contributing to high network performance. Recent strides in device engineering in planar and three-dimensional structures and the understanding of ECRAM operation physics have marked significant progress in a short research period. This paper aims to review ECRAM material advancements through literature surveys, offering a systematic discussion on engineering assessments for ion control and a physical understanding of array-level demonstrations. Finally, the review outlines future directions for improvements, co-optimization, and multidisciplinary collaboration in circuits, algorithms, and applications to develop energy-efficient, next-generation AI hardware systems.

WOx channel engineering of Cu-ion-driven synaptic transistor array for low-power neuromorphic computing

The multilevel current states of synaptic devices in artificial neural networks enable next-generation computing to perform cognitive functions in an energy-efficient manner. Moreover, considering large-scale synaptic arrays, multiple states programmed in a low-current regime may be required to achieve low energy consumption, as demonstrated by simple numerical calculations. Thus, we propose a three-terminal Cu-ion-actuated CuOx/HfOx/WO3 synaptic transistor array that exhibits analogously modulated channel current states in the range of tens of nanoamperes, enabled by WO3 channel engineering. The introduction of an amorphous stoichiometric WO3 channel formed by reactive sputtering with O gas significantly lowered the channel current but left it almost unchanged with respect to consecutive gate voltage pulses. An additional annealing process at 450 °C crystallized the WO3, allowing analog switching in the range of tens of nanoamperes. The incorporation of N gas during annealing induced a highly conductive channel, making the channel current modulation negligible as a function of the gate pulse. Using this optimized gate stack, Poole-Frenkel conduction was identified as a major transport characteristic in a temperature-dependent study. In addition, we found that the channel current modulation is a function of the gate current response, which is related to the degree of progressive movement of the Cu ions. Finally, the synaptic characteristics were updated using fully parallel programming and demonstrated in a 7 × 7 array. Using the CuOx/HfOx/WO3 synaptic transistors as weight elements in multilayer neural networks, we achieved a 90% recognition accuracy on the Fashion-MNIST dataset.

Exploring the Cutting-Edge Frontiers of Electrochemical Random Access Memories (ECRAMs) for Neuromorphic Computing: Revolutionary Advances in Material-to-Device Engineering

Advanced materials and device engineering has played a crucial role in improving the performance of electrochemical random access memory (ECRAM) devices. ECRAM technology has been identified as a promising candidate for implementing artificial synapses in neuromorphic computing systems due to its ability to store analog values and its ease of programmability. ECRAM devices consist of an electrolyte and a channel material sandwiched between two electrodes, and the performance of these devices depends on the properties of the materials used. This review provides a comprehensive overview of material engineering strategies to optimize the electrolyte and channel materials' ionic conductivity, stability, and ionic diffusivity to improve the performance and reliability of ECRAM devices. Device engineering and scaling strategies are further discussed to enhance ECRAM performance. Last, perspectives on the current challenges and future directions in developing ECRAM-based artificial synapses in neuromorphic computing systems are provided.

ECRAM Materials, Devices, Circuits and Architectures: A Perspective

Non-von-Neumann computing using neuromorphic systems based on two-terminal resistive nonvolatile memory elements has emerged as a promising approach, but its full potential has not been realized due to the lack of materials and devices with the appropriate attributes. Unlike memristors, which require large write currents to drive phase transformations or filament growth, electrochemical random access memory (ECRAM) decouples the "write" and "read" operations using a "gate" electrode to tune the conductance state through charge-transfer reactions, and every electron transferred through the external circuit in ECRAM corresponds to the migration of ≈1 ion used to store analogue information. Like static dopants in traditional semiconductors, electrochemically inserted ions modulate the conductivity by locally perturbing a host's electronic structure; however, ECRAM does so in a dynamic and reversible manner. The resulting change in conductance can span orders of magnitude, from gradual increments needed for analog elements, to large, abrupt changes for dynamically reconfigurable adaptive architectures. In this in-depth perspective, the history of ECRAM, the recent progress in devices spanning organic, inorganic, and 2D materials, circuits, architectures, the rich portfolio of challenging, fundamental questions, and how ECRAM can be harnessed to realize a new paradigm for low-power neuromorphic computing are discussed.

Tunable Intervalence Charge Transfer in Ruthenium Prussian Blue Analog Enables Stable and Efficient Biocompatible Artificial Synapses

Emerging concepts for neuromorphic computing, bioelectronics, and brain-computer interfacing inspire new research avenues aimed at understanding the relationship between oxidation state and conductivity in unexplored materials. This report expands the materials playground for neuromorphic devices to include a mixed valence inorganic 3D coordination framework, a ruthenium Prussian blue analog (RuPBA), for flexible and biocompatible artificial synapses that reversibly switch conductance by more than four orders of magnitude based on electrochemically tunable oxidation state. The electrochemically tunable degree of mixed valency and electronic coupling between N-coordinated Ru sites controls the carrier concentration and mobility, as supported by density functional theory computations and application of electron transfer theory to in situ spectroscopy of intervalence charge transfer. Retention of programmed states is improved by nearly two orders of magnitude compared to extensively studied organic polymers, thus reducing the frequency, complexity, and energy costs associated with error correction schemes. This report demonstrates dopamine-mediated plasticity of RuPBA synapses and biocompatibility of RuPBA with neuronal cells, evoking prospective application for brain-computer interfacing.

Bayesian neural networks using magnetic tunnel junction-based probabilistic in-memory computing

Bayesian neural networks (BNNs) combine the generalizability of deep neural networks (DNNs) with a rigorous quantification of predictive uncertainty, which mitigates overfitting and makes them valuable for high-reliability or safety-critical applications. However, the probabilistic nature of BNNs makes them more computationally intensive on digital hardware and so far, less directly amenable to acceleration by analog in-memory computing as compared to DNNs. This work exploits a novel spintronic bit cell that efficiently and compactly implements Gaussian-distributed BNN values. Specifically, the bit cell combines a tunable stochastic magnetic tunnel junction (MTJ) encoding the trained standard deviation and a multi-bit domain-wall MTJ device independently encoding the trained mean. The two devices can be integrated within the same array, enabling highly efficient, fully analog, probabilistic matrix-vector multiplications. We use micromagnetics simulations as the basis of a system-level model of the spintronic BNN accelerator, demonstrating that our design yields accurate, well-calibrated uncertainty estimates for both classification and regression problems and matches software BNN performance. This result paves the way to spintronic in-memory computing systems implementing trusted neural networks at a modest energy budget.

Metaplastic and energy-efficient biocompatible graphene artificial synaptic transistors for enhanced accuracy neuromorphic computing

CMOS-based computing systems that employ the von Neumann architecture are relatively limited when it comes to parallel data storage and processing. In contrast, the human brain is a living computational signal processing unit that operates with extreme parallelism and energy efficiency. Although numerous neuromorphic electronic devices have emerged in the last decade, most of them are rigid or contain materials that are toxic to biological systems. In this work, we report on biocompatible bilayer graphene-based artificial synaptic transistors (BLAST) capable of mimicking synaptic behavior. The BLAST devices leverage a dry ion-selective membrane, enabling long-term potentiation, with ~50 aJ/µm2 switching energy efficiency, at least an order of magnitude lower than previous reports on two-dimensional material-based artificial synapses. The devices show unique metaplasticity, a useful feature for generalizable deep neural networks, and we demonstrate that metaplastic BLASTs outperform ideal linear synapses in classic image classification tasks. With switching energy well below the 1 fJ energy estimated per biological synapse, the proposed devices are powerful candidates for bio-interfaced online learning, bridging the gap between artificial and biological neural networks.

Hafnium Oxide (HfO2 ) - A Multifunctional Oxide: A Review on the Prospect and Challenges of Hafnium Oxide in Resistive Switching and Ferroelectric Memories

Hafnium oxide (HfO₂ ) is one of the mature high-k dielectrics that has been standing strong in the memory arena over the last two decades. Its dielectric properties have been researched rigorously for the development of flash memory devices. In this review, the application of HfO₂ in two main emerging nonvolatile memory technologies is surveyed, namely resistive random access memory and ferroelectric memory. How the properties of HfO₂ equip the former to achieve superlative performance with high-speed reliable switching, excellent endurance, and retention is discussed. The parameters to control HfO₂ domains are further discussed, which can unleash the ferroelectric properties in memory applications. Finally, the prospect of HfO₂ materials in emerging applications, such as high-density memory and neuromorphic devices are examined, and the various challenges of HfO₂ -based resistive random access memory and ferroelectric memory devices are addressed with a future outlook.

Strategies to Improve the Synaptic Characteristics of Oxygen-Based Electrochemical Random-Access Memory Based on Material Parameters Optimization

Oxygen-based electrochemical random-access memories (O-ECRAMs) are promising synaptic devices for neuromorphic applications because of their near-ideal synaptic characteristics and compatibility with complementary metal-oxide-semiconductor processes. However, the correlation between material parameters and synaptic properties of O-ECRAM devices has not yet been elucidated. Here, we propose the critical design parameters to fabricate an ideal ECRAM device. Based on the experimental data and simulation results, it is revealed that consistent ion supply from the electrolyte and rapid ion diffusion in the channel are critical factors for ideal synaptic characteristics. To optimize these parameters, crystalline WO_2.7 exhibiting fast ion diffusivity and ZrO_1.7 exhibiting an appropriate ion conduction energy barrier (1.1 eV) are used as a channel and an electrolyte, respectively. As a result, synaptic characteristics with near-ideal weight-update linearity in the nanosiemens conductance range are achieved. Finally, a selector-less O-ECRAM device is integrated into a 2 × 2 array, and high recognition accuracy (94.83%) of the Modified National Institute of Standards and Technology pattern is evaluated.