Retention-aware zero-shifting technique for Tiki-Taka algorithm-based analog deep learning accelerator.

We present the fabrication of 4 K-scale electrochemical random-access memory (ECRAM) cross-point arrays for analog neural network training accelerator and an electrical characteristic of an 8 × 8 ECRAM array with a 100% yield, showing excellent switching characteristics, low cycle-to-cycle, and device-to-device variations. Leveraging the advances of the ECRAM array, we showcase its efficacy in neural network training using the Tiki-Taka version 2 algorithm (TTv2) tailored for non-ideal analog memory devices. Through an experimental study using ECRAM devices, we investigate the influence of retention characteristics on the training performance of TTv2, revealing that the relative location of the retention convergence point critically determines the available weight range and, consequently, affects the training accuracy. We propose a retention-aware zero-shifting technique designed to optimize neural network training performance, particularly in scenarios involving cross-point devices with limited retention times. This technique ensures robust and efficient analog neural network training despite the practical constraints posed by analog cross-point devices.

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.

Strategy to improve synaptic behavior of ion-actuated synaptic transistors-the use of ion blocking layer to improve state retention.

Synaptic transistors (STs) with a gate/electrolyte/channel stack, where mobile ions are electrically driven across the solid electrolyte, have been considered as analog weight elements for neuromorphic computing. The current (ID) between the source and drain in the ST is analogously updated by gate voltage (VG) pulses, enabling high pattern recognition accuracy in neuromorphic systems; however, the governing physical mechanisms of the ST are not fully understood yet. Our previous physics-based simulation study showed that ion movement in the electrolyte, rather than the electrochemical reactions that occur in the channel, plays an important role in switching. In this study, we experimentally explore the properties of the HfOx electrolyte and show that by tuning the density of oxygen vacancies, it can assume the dual role of electrolyte and channel. We demonstrate analog synaptic behavior using a novel ST with a two-layer stack of CuOx/HfOx, where the CuOx is the gate and Cu ion reservoir, and the HfOx is the electrolyte and channel. To improve state retention and linearity, we introduce a Cu ion transport barrier in the form of a dense and stoichiometric Al2O3 layer. The CuOx/Al2O3/HfOx exhibits excellent state retention and improved potentiation and depression response. Energy dispersive spectroscopy mapping following potentiation confirms the role of the Al2O3 layer in confining the Cu ions in the HfOx layer. We also show that a two-step programming scheme can further enhance synaptic response and demonstrate high recognition accuracy on the Fashion-MNIST dataset in simulation.

Internal Resistor Effect of Multilayer-Structured Synaptic Device for Low-Power Operation.

A synaptic device with a multilayer structure is proposed to reduce the operating power of neuromorphic computing systems while maintaining a high-density integration. A simple metal-insulator-metal (MIM)-structured multilayer synaptic device is developed using an 8-inch wafer-based and complementary metal-oxide-semiconductor (CMOS) fabrication process. The three types of MIM-structured synaptic devices are compared to assess their effects on reducing the operating power. The obtained results exhibited low-power operation owing to the inserted layers acting as an internal resistor. The modulated operational conductance level and simple MIM structure demonstrate the feasibility of implementing both low-power operation and high-density integration in multilayer synaptic devices.

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.

Definition of a Localized Conducting Path via Suppressed Charge Injection in Oxide Memristors for Stable Practical Hardware Neural Networks.

Oxide-based memristors have been demonstrated as suitable options for memory components in neuromorphic systems. In such devices, the resistive switching characteristics are caused by the formation of conductive filaments (CFs) comprising oxygen vacancies. Thus, the electrical performance is primarily governed by the CF structure. Despite various approaches for regulating the oxygen vacancy distributions in oxide memristors, controlling the CF structure without modifying the device configuration related to material compatibility is still a challenge. This study demonstrates an effective strategy for localizing CF distributions in memristors by suppressing charge injection during the formation of conducting paths. As the injected charge quantity is reduced in the electroforming process of the oxide memristor, the CF distributions become narrower, leading to more reproducible and stable resistive switching characteristics in the device. Based on these findings, a reliable hardware neural network comprising oxide memristors is constructed to recognize complex images. The developed memristor has been employed as a synaptic memory component in systems without degradation for a long time. This promising concept of oxide memristors acting as stable synaptic components holds great potential for developing practical neuromorphic systems and their expansion into artificial intelligent systems.

Three-dimensional vertical structural electrochemical random access memory for high-density integrated synapse device.

Three-terminal (3T) structured electrochemical random access memory (ECRAM) has been proposed as a synaptic device based on improved synaptic characteristics. However, the proposed 3T ECRAM has a larger area requirement than 2T synaptic devices; thereby limiting integration density. To overcome this limitation, this study presents the development of a high-density vertical structure for the 3T ECRAM. In addition, complementary metal-oxide semiconductor (CMOS)-compatible materials and 8-inch wafer-based CMOS fabrication processes were utilized to verify the feasibility of mass production. The achievements of this work demonstrate the potential for high-density integration and mass production of 3T ECRAM devices.

Linear Frequency Modulation of NbO2-Based Nanoscale Oscillator With Li-Based Electrochemical Random Access Memory for Compact Coupled Oscillatory Neural Network.

Oscillatory neural network (ONN)-based classification of clustered data relies on frequency synchronization to injected signals representing input data, showing a more efficient structure than a conventional deep neural network. A frequency tunable oscillator is a core component of the network, requiring energy-efficient, and area-scalable characteristics for large-scale hardware implementation. From a hardware viewpoint, insulator-metal transition (IMT) device-based oscillators are attractive owing to their simple structure and low power consumption. Furthermore, by introducing non-volatile analog memory, non-volatile frequency programmability can be obtained. However, the required device characteristics of the oscillator for high performance of coupled oscillator have not been identified. In this article, we investigated the effect of device parameters of IMT oscillator with non-volatile analog memory on coupled oscillators network for classification of clustered data. We confirmed that linear conductance response with identical pulses is crucial to accurate training. In addition, considering dispersed clustered inputs, a wide synchronization window achieved by controlling the hold voltage of the IMT shows resilient classification. As an oscillator that satisfies the requirements, we evaluated the NbO2-based IMT oscillator with non-volatile Li-based electrochemical random access memory (Li-ECRAM). Finally, we demonstrated a coupled oscillator network for classifying spoken vowels, achieving an accuracy of 85%, higher than that of a ring oscillator-based system. Our results show that an NbO2-based oscillator with Li-ECRAM has the potential for an area-scalable and energy-efficient network with high performance.

Ion-Driven Electrochemical Random-Access Memory-Based Synaptic Devices for Neuromorphic Computing Systems: A Mini-Review.

To enhance the computing efficiency in a neuromorphic architecture, it is important to develop suitable memory devices that can emulate the role of biological synapses. More specifically, not only are multiple conductance states needed to be achieved in the memory but each state is also analogously adjusted by consecutive identical pulses. Recently, electrochemical random-access memory (ECRAM) has been dedicatedly designed to realize the desired synaptic characteristics. Electric-field-driven ion motion through various electrolytes enables the conductance of the ECRAM to be analogously modulated, resulting in a linear and symmetric response. Therefore, the aim of this study is to review recent advances in ECRAM technology from the material and device engineering perspectives. Since controllable mobile ions play an important role in achieving synaptic behavior, the prospect and challenges of ECRAM devices classified according to mobile ion species are discussed.

Ferroelectric Switching in Trilayer Al2O3/HfZrOx/Al2O3 Structure.

Since ferroelectricity has been observed in simple binary oxide material systems, it has attracted great interest in semiconductor research fields such as advanced logic transistors, non-volatile memories, and neuromorphic devices. The location in which the ferroelectric devices are implemented depends on the specific application, so the process constraints required for device fabrication may be different. In this study, we investigate the ferroelectric characteristics of Zr doped HfO2 layers treated at high temperatures. A single HfZrOx layer deposited by sputtering exhibits polarization switching after annealing at a temperature of 850 °C. However, the achieved ferroelectric properties are vulnerable to voltage stress and higher annealing temperature, resulting in switching instability. Therefore, we introduce an ultrathin 1-nm-thick Al2O3 layer at both interfaces of the HfZrOx. The trilayer Al2O3/HfZrOx/Al2O3 structure allows switching parameters such as remnant and saturation polarizations to be immune to sweeping voltage and pulse cycling. Our results reveal that the trilayer not only makes the ferroelectric phase involved in the switching free from pinning, but also preserves the phase even at high annealing temperature. Simultaneously, the ferroelectric switching can be improved by preventing leakage charge.

Exploiting defective RRAM array as synapses of HTM spatial pooler with boost-factor adjustment scheme for defect-tolerant neuromorphic systems.

A crossbar array architecture employing resistive switching memory (RRAM) as a synaptic element accelerates vector-matrix multiplication in a parallel fashion, enabling energy-efficient pattern recognition. To implement the function of the synapse in the RRAM, multilevel resistance states are required. More importantly, a large on/off ratio of the RRAM should be preferentially obtained to ensure a reasonable margin between each state taking into account the inevitable variability caused by the inherent switching mechanism. The on/off ratio is basically adjusted in two ways by modulating measurement conditions such as compliance current or voltage pulses modulation. The latter technique is not only more suitable for practical systems, but also can achieve multiple states in low current range. However, at the expense of applying a high negative voltage aimed at enlarging the on/off ratio, a breakdown of the RRAM occurs unexpectedly. This stuck-at-short fault of the RRAM adversely affects the recognition process based on reading and judging each column current changed by the multiplication of the input voltage and resistance of the RRAM in the array, degrading the accuracy. To address this challenge, we introduce a boost-factor adjustment technique as a fault-tolerant scheme based on simple circuitry that eliminates the additional process to identify specific locations of the failed RRAMs in the array. Spectre circuit simulation is performed to verify the effect of the scheme on Modified National Institute of Standards and Technology dataset using convolutional neural networks in non-ideal crossbar arrays, where experimentally observed imperfective RRAMs are configured. Our results show that the recognition accuracy can be maintained similar to the ideal case because the interruption of the failure is suppressed by the scheme.

Artificial Perspiration Membrane by Programmed Deformation of Thermoresponsive Hydrogels.

Thermal management is essential for living organisms and electronic devices to survive and maintain their own functions. However, developing flexible cooling devices for flexible electronics or biological systems is challenging because conventional coolers are bulky and require rigid batteries. In nature, skins help to maintain a constant body temperature by dissipating heat through perspiration. Inspired by nature, an artificial perspiration membrane that automatically regulates evaporation depending on temperature using the programmed deformation of thermoresponsive hydrogels is presented. The thermoresponsive hydrogel is patterned into pinwheel shapes and supported by a polymeric rigid frame with stable adhesion using copolymerization. Both shape of the valve and mechanical constraint of the frame allow six times larger evaporation area in the open state compared to the closed state, and the transition occurs at a fast rate (≈1 s). A stretchable membrane is selectively coated to prevent unintended evaporation through the hydrogel while allowing swelling or shrinking of the hydrogel by securing path of water. Consequently, a 30% reduction in evaporation is observed at lower temperature, resulting in regulation of the skin temperature at the thermal model of human skins. This simple, small, and flexible cooler will be useful for maintaining temperature of flexible devices.

Microstructural engineering in interface-type synapse device for enhancing linear and symmetric conductance changes.

The origins of the nonlinear and asymmetric synaptic characteristics of TiO x -based synapse devices were investigated. Based on the origins, a microstructural electrode was utilized to improve the synaptic characteristics. Under an identical pulse bias, a TiO x -based synapse device exhibited saturated conductance changes, which led to nonlinear and asymmetric synaptic characteristics. The formation of an interfacial layer between the electrode and TiO x layer, which can limit consecutive oxygen migration and chemical reactions, was considered as the main origin of the conductance saturation behavior. To achieve consecutive oxygen migration and chemical reactions, structural engineering was utilized. The resultant microstructural electrode noticeably improved the synaptic characteristics, including the unsaturated, linear, and symmetric conductance changes. These synaptic characteristics resulted in the recognition accuracy significantly increasing from 38% to 90% in a neural network-based pattern recognition simulation.

Controllable quantized conductance for multilevel data storage applications using conductive bridge random access memory.

In this paper, we investigate the quantized conduction behavior of conductive bridge random access memory (CBRAM) with varied materials and ramping rates. We report stable and reproducible quantized conductance states with integer multiples of fundamental conductance obtained by optimizing the voltage ramping rate and the Ti-diffusion barrier (DB) at the Cu/HfO2 interface. Owing to controlled diffusion of Cu ions by the Ti-DB and the optimized ramping rate, through which it was possible to control the time delay of Cu ion reduction, more than seven levels of discrete conductance states were clearly observed. Analytical modeling was performed to determine the rate-limiting step in filament growth based on an electrochemical redox reaction. Our understanding of the fundamental mechanisms of quantized conductance behaviors provide a promising future for the multi-bit CBRAM device.

Effects of High-Pressure Hydrogen Annealing (HPHA) on Reliability Characteristics of RRAM.

Reliability characteristics (retention and endurance) of RRAM are critical for its practical realization and need to be improved. In this work, we confirmed the trade-off between retention and endurance by using various top electrode thickness conditions. The trade-off between retention and endurance characteristics was mainly due to the different amount of oxygen in scavenging layer (Ta) and the amount of oxygen vacancy in switching layer (HfO2). The amount of the oxygen in scavenging layer (Ta) and the amount of the oxygen vacancy in switching layer (HfO2) will be increased with the increase of Ta thickness. Therefore, the thicker Ta cells have worse retention because the large amount of oxygen in scavenging layer (Ta) can diffuse back into switching layer (HfO2) and recombine with oxygen vacancies in the filament. However, it has longer endurance because the large amount of oxygen vacancy in switching layer (HfO2) can be a source of the filament. Hence, there exists a trade-off relation between retention and endurance under the various Ta thickness conditions. To improve both retention and endurance characteristics, we proposed a new method by using high-pressure hydrogen annealing (HPHA). The thin Ta cells have longer retention and worse endurance because it has small amount of both oxygen in scavenging layer (Ta) and oxygen vacancy in switching layer (HfO2). Therefore, to generate more oxygen vacancies in switching layer (HfO2) maintaining small amount of oxygen in scavenging layer (Ta), we treated the samples by HPHA before Ta deposition. Finally, we obtained both improved retention and endurance characteristics in HfO2 based RRAM devices after high-pressure hydrogen annealing treatment.

Structurally engineered stackable and scalable 3D titanium-oxide switching devices for high-density nanoscale memory.

A 3D high-density switching device is realized utilizing titanium oxide, which is the most optimum material, but which is not practically demonstrated yet. The 1S1R (one ReRAM with the developed switching device) exhibits memory characteristics with a significantly suppressed sneak current, which can be used to realize high-density ReRAM applications.

Hardware implementation of associative memory characteristics with analogue-type resistive-switching device.

We have investigated the analogue memory characteristics of an oxide-based resistive-switching device under an electrical pulse to mimic biological spike-timing-dependent plasticity synapse characteristics. As a synaptic device, a TiN/Pr0.7Ca0.3MnO3-based resistive-switching device exhibiting excellent analogue memory characteristics was used to control the synaptic weight by applying various pulse amplitudes and cycles. Furthermore, potentiation and depression characteristics with the same spikes can be achieved by applying negative and positive pulses, respectively. By adopting complementary metal-oxide-semiconductor devices as neurons and TiN/PCMO devices as synapses, we implemented neuromorphic hardware that mimics associative memory characteristics in real time for the first time. Owing to their excellent scalability, resistive-switching devices, shows promise for future high-density neuromorphic applications.

Internal resistor of multi-functional tunnel barrier for selectivity and switching uniformity in resistive random access memory.

In this research, we analyzed the multi-functional role of a tunnel barrier that can be integrated in devices. This tunnel barrier, acting as an internal resistor, changes its resistance with applied bias. Therefore, the current flow in the devices can be controlled by a tunneling mechanism that modifies the tunnel barrier thickness for non-linearity and switching uniformity of devices. When a device is in a low-resistance state, the tunnel barrier controls the current behavior of the device because most of the bias is applied to the tunnel barrier owing to its higher resistance. Furthermore, the tunnel barrier induces uniform filament formation during set operation with the tunnel barrier controlling the current flow.

High current density and nonlinearity combination of selection device based on TaO(x)/TiO2/TaO(x) structure for one selector-one resistor arrays.

We demonstrate a high-performance selection device by utilizing the concept of crested oxide barrier to suppress the sneak current in bipolar resistive memory arrays. Using a TaO(x)/TiO(2)/TaO(x) structure, high current density over 10(7) A cm(-2) and excellent nonlinear characteristics up to 10(4) were successfully demonstrated. On the basis of the defect chemistry and SIMS depth profile result, we found that some Ta atoms gradually diffused into TiO(2) film, and consequently, the energy band of the TiO(2) film was symmetrically bent at the top and bottom TaO(x)/TiO(2) interfaces and modified as a crested oxide barrier. Furthermore, the one selector-one resistor device exhibited significant suppression of the leakage current, indicating excellent selector characteristics.

Defect engineering: reduction effect of hydrogen atom impurities in HfO2-based resistive-switching memory devices.

In this study, we propose a new and effective methodology for improving the resistive-switching performance of memory devices by high-pressure hydrogen annealing under ambient conditions. The reduction effect results in the uniform creation of oxygen vacancies that in turn enable forming-free operation and afford uniform switching characteristics. In addition, H(+) and mobile hydroxyl (OH(-)) ions are generated, and these induce fast switching operation due to the higher mobility compared to oxygen ions. Defect engineering, specifically, the introduction of hydrogen atom impurities, improves the device performance for metal-oxide-based resistive-switching random access memory devices.