2Memristor-1Capacitor Integrated Temporal Kernel for High-Dimensional Data Mapping.

Compact but precise feature-extracting ability is core to processing complex computational tasks in neuromorphic hardware. Physical reservoir computing (RC) offers a robust framework to map temporal data into a high-dimensional space using the time dynamics of a material system, such as a volatile memristor. However, conventional physical RC systems have limited dynamics for the given material properties, restricting the methods to increase their dimensionality. This study proposes an integrated temporal kernel composed of a 2-memristor and 1-capacitor (2M1C) using a W/HfO2/TiN memristor and TiN/ZrO2/Al2O3/ZrO2/TiN capacitor to achieve higher dimensionality and tunable dynamics. The kernel elements are carefully designed and fabricated into an integrated array, of which performances are evaluated under diverse conditions. By optimizing the time dynamics of the 2M1C kernel, each memristor simultaneously extracts complementary information from input signals. The MNIST benchmark digit classification task achieves a high accuracy of 94.3% with a (196×10) single-layer network. Analog input mapping ability is tested with a Mackey-Glass time series prediction, and the system records a normalized root mean square error of 0.04 with a 20×1 readout network, the smallest readout network ever used for Mackey-Glass prediction in RC. These performances demonstrate its high potential for efficient temporal data analysis.

Ferroelectric domain wall memory with embedded selector realized in LiNbO3 single crystals integrated on Si wafers.

Interfacial 'dead' layers between metals and ferroelectric thin films generally induce detrimental effects in nanocapacitors, yet their peculiar properties can prove advantageous in other electronic devices. Here, we show that dead layers with low Li concentration located at the surface of LiNbO3 ferroelectric materials can function as unipolar selectors. LiNbO3 mesa cells were etched from a single-crystal LiNbO3 substrate, and Pt metal contacts were deposited on their sides. Poling induced non-volatile switching of ferroelectric domains in the cell, and volatile switching in the domains in the interfacial (dead) layers, with the domain walls created within the substrate being electrically conductive. These features were also confirmed using single-crystal LiNbO3 thin films bonded to SiO2/Si wafers. The fabricated nanoscale mesa-structured memory cell with an embedded interfacial-layer selector shows a high on-to-off ratio (>106) and high switching endurance (~1010 cycles), showing potential for the fabrication of crossbar arrays of ferroelectric domain wall memories.

Fluorite-structure antiferroelectrics.

Ferroelectricity in fluorite-structure oxides like hafnia and zirconia have attracted increasing interest since 2011. Two spontaneous polarizations of the fluorite-structure ferroelectrics are considered highly promising for nonvolatile memory applications, with their superior scalability and Si compatibility compared to the conventional perovskite-structure ferroelectrics. Besides, antiferroelectricity originating from a field-induced phase transition between the paraelectric and ferroelectric phases in fluorite-structure oxides is another highly interesting matter. It was suggested that the field-induced phase transition could be utilized for energy conversions between thermal and electrical energy, as well as for energy storage. The important energy-related applications of antiferroelectric fluorite-structure oxides, however, have not been systematically reviewed to date. Thus, in this work, the fluorite-structure antiferroelectrics are reviewed from their fundamentals to their applications based on pyroelectricity as well as antiferroelectricity. Another important application field of the fluorite-structure antiferroelectrics is the semiconductor memory devices. The fluorite-structure antiferroelectrics can be utilized for antiferroelectric random-access-memories, negative capacitance field-effect-transistors, and flash memories. Moreover, the recently reported morphotropic phase boundary (MPB) between the ferroelectric and antiferroelectric phases in this material system marks another significant progress in this material system, and thus, the fundamentals and applications of the MPB phase are also reviewed.

Temporary formation of highly conducting domain walls for non-destructive read-out of ferroelectric domain-wall resistance switching memories.

Erasable conductive domain walls in insulating ferroelectric thin films can be used for non-destructive electrical read-out of the polarization states in ferroelectric memories. Still, the domain-wall currents extracted by these devices have not yet reached the intensity and stability required to drive read-out circuits operating at high speeds. This study demonstrated non-destructive read-out of digital data stored using specific domain-wall configurations in epitaxial BiFeO3 thin films formed in mesa-geometry structures. Partially switched domains, which enable the formation of conductive walls during the read operation, spontaneously retract when the read voltage is removed, reducing the accumulation of mobile defects at the domain walls and potentially improving the device stability. Three-terminal memory devices produced 14 nA read currents at an operating voltage of 5 V, and operated up to T = 85 °C. The gap length can also be smaller than the film thickness, allowing the realization of ferroelectric memories with device dimensions far below 100 nm.

In2Ga2ZnO7 oxide semiconductor based charge trap device for NAND flash memory.

The programming characteristics of charge trap flash memory device adopting amorphous In2Ga2ZnO7 (a-IGZO) oxide semiconductors as channel layer were evaluated. Metal-organic chemical vapor deposition (MOCVD) and RF-sputtering processes were used to grow a 45 nm thick a-IGZO layer on a 20 nm thick SiO2 (blocking oxide)/p++-Si (control gate) substrate, where 3 nm thick atomic layer deposited Al2O3 (tunneling oxide) and 5 nm thick low-pressure CVD Si3N4 (charge trap) layers were intervened between the a-IGZO and substrate. Despite the identical stoichiometry and other physicochemical properties of the MOCVD and sputtered a-IGZO, a much faster programming speed of MOCVD a-IGZO was observed. A comparable amount of oxygen vacancies was found in both MOCVD and sputtered a-IGZO, confirmed by x-ray photoelectron spectroscopy and bias-illumination-instability test measurements. Ultraviolet photoelectron spectroscopy analysis revealed a higher Fermi level (E F) of the MOCVD a-IGZO (∼0.3 eV) film than that of the sputtered a-IGZO, which could be ascribed to the higher hydrogen concentration in the MOCVD a-IGZO film. Since the programming in a flash memory device is governed by the tunneling of electrons from the channel to charge trapping layer, the faster programming performance could be the result of a higher E F of MOCVD a-IGZO.

Atomic layer deposition of GeSe films using HGeCl3 and [(CH3)3Si]2Se with the discrete feeding method for the ovonic threshold switch.

The ovonic threshold switch (OTS) based on the voltage snapback of amorphous chalcogenides possesses several desirable characteristics: bidirectional switching, a controllable threshold voltage (V th) and processability for three-dimensional stackable devices. Among the materials that can be used as OTS, GeSe has a strong glass-forming ability (∼350 °C crystallization temperature), with a simple binary composition. Described herein is a new method of depositing GeSe films through atomic layer deposition (ALD), using HGeCl3 and [(CH3)3Si]2Se as Ge and Se precursors, respectively. The stoichiometric GeSe thin films were formed through a ligand exchange reaction between the two precursor molecules, without the adoption of an additional reaction gas, at low substrate temperatures ranging from 70 °C-150 °C. The pseudo-saturation behavior required a long time of Ge precursor injection to achieve the saturation growth rate. This was due to the adverse influence of the physisorbed precursor and byproduct molecules on the efficient chemical adsorption reaction between the precursors and reaction sites. To overcome the slow saturation and excessive use of the Ge precursor, the discrete feeding method (DFM), where HGeCl3 is supplied multiple times consecutively with subdivided pulse times, was adopted. DFM led to the saturation of the GeSe growth rate at a much shorter total injection time of the Ge precursor, and improved the film density and oxidation resistance properties. The GeSe film grown via DFM exhibited a short OTS time of ∼40 ns, a ∼107 ON/OFF current ratio, and ∼104 selectivity. The OTS behavior was consistent with the modified Poole-Frenkel mechanism in the OFF state. In contrast, the similar GeSe film grown through the conventional ALD showed a low density and high vulnerability to oxidation, which prevented the OTS performance. The ALD method of GeSe films introduced here will contribute to the fabrication of a three-dimensionally integrated memory as a selector device for preventing sneak current.

Voltage Drop in a Ferroelectric Single Layer Capacitor by Retarded Domain Nucleation.

Ferroelectric (FE) capacitor is a critical electric component in microelectronic devices. Among many of its intriguing properties, the recent finding of voltage drop (V-drop) across the FE capacitor while the positive charges flow in is especially eye-catching. This finding was claimed to be direct evidence that the FE capacitor is in negative capacitance (NC) state, which must be useful for (infinitely) high capacitance and ultralow voltage operation of field-effect transistors. Nonetheless, the NC state corresponds to the maximum energy state of the FE material, so it has been widely accepted in the community that the material alleviates that state by forming ferroelectric domains. This work reports a similar V-drop effect from the 150 nm thick epitaxial BaTiO3 ferroelectric thin film, but the interpretation was completely disparate; the V-drop can be precisely simulated by the reverse domain nucleation and propagation of which charge effect cannot be fully compensated for by the supplied charge from the external charge source. The disappearance of the V-drop effect was also observed by repeated FE switching only up to 10 cycles, which can hardly be explained by the involvement of the NC effect. The retained reverse domain nuclei even after the subsequent poling can explain such behavior.

Four-Bits-Per-Cell Operation in an HfO2 -Based Resistive Switching Device.

The quadruple-level cell technology is demonstrated in an Au/Al2 O3 /HfO2 /TiN resistance switching memory device using the industry-standard incremental step pulse programming (ISPP) and error checking/correction (ECC) methods. With the highly optimistic properties of the tested device, such as self-compliance and gradual set-switching behaviors, the device shows 6σ reliability up to 16 states with a state current gap value of 400 nA for the total allowable programmed current range from 2 to 11 µA. It is demonstrated that the conventional ISPP/ECC can be applied to such resistance switching memory, which may greatly contribute to the commercialization of the device, especially competitively with NAND flash. A relatively minor improvement in the material and circuitry may enable even a five-bits-per-cell technology, which can hardly be imagined in NAND flash, whose state-of-the-art multiple-cell technology is only at three-level (eight states) to this day.

Preparation and characterization of ferroelectric Hf0.5Zr0.5O2 thin films grown by reactive sputtering.

HfO2-ZrO2 solid-solution films were prepared by radio frequency sputtering, and the subsequent annealing process was optimized to render enhanced ferroelectric behavior. The target power, working pressure and O2 partial pressure ratios were varied, along with the annealing gas, time and temperature. Then, the film's structural and electrical properties were carefully scrutinized. Oxygen-deficient conditions were necessary during the sputter deposition to suppress grain growth, while annealing by O2 gas was critical to avoid defects and leakage problems. It is expected that the grain size difference under various deposition conditions combined with the degree of TiN top and bottom electrode oxidation by O2 gas will result in different ferroelectric behaviors. As a result, Hf0.5Zr0.5O2 prepared by radio frequency sputtering showed optimized ferroelectricity at 0% of O2 reactive gas, with a doubled remnant polarization value of ∼20 µC cm-2 at a thickness of 11 nm. Film growth conditions with a high growth rate (4-5 nm min-1) were favorable for achieving the ferroelectric phase film, which feasibly suppressed both the grain growth and accompanying monoclinic phase formation.

Regulation of transport properties by polytypism: a computational study on bilayer MoS2.

Being a member of the van der Waals class of solids, bilayer MoS2 exhibits polytypism due to different possible stacking arrangements, namely, 2Hc, 2Ha and 3R-polytypes. Unlike monolayer MoS2, these bilayers exhibit indirect band gaps. Band extrema states originate from a linear combination of Mo-(d) and S-(p) orbitals which are sensitive to the interlayer interactions. We have studied the impact of stacking pattern on the electronic structure and electron/hole transport properties of these polytypes. Based on first-principles computations coupled with the Boltzmann transport formalism, we found that a strong electron-hole anisotropy can be realised in the 2Ha-MoS2 polytype unlike in a monolayer which is isotropic in nature. A staggered arrangement between two layers results in a higher relaxation time for electrons compared to holes leading to anisotropy which is of importance in device engineering.

Low-Power, Self-Rectifying, and Forming-Free Memristor with an Asymmetric Programing Voltage for a High-Density Crossbar Application.

A Pt/NbOx/TiOy/NbOx/TiN stack integrated on a 30 nm contact via shows a programming current as low as 10 nA and 1 pA for the set and reset switching, respectively, and a self-rectifying ratio as high as ∼105, which are suitable characteristics for low-power memristor applications. It also shows a forming-free characteristic. A charge-trap-associated switching model is proposed to account for this self-rectifying memrisive behavior. In addition, an asymmetric voltage scheme (AVS) to decrease the write power consumption by utilizing this self-rectifying memristor is also described. When the device is used in a 1000 × 1000 crossbar array with the AVS, the programming power can be decreased to 8.0% of the power consumption of a conventional biasing scheme. If the AVS is combined with a nonlinear selector, a power consumption reduction to 0.31% of the reference value is possible.

Time-Dependent Negative Capacitance Effects in Al2O3/BaTiO3 Bilayers.

The negative capacitance (NC) effects in ferroelectric materials have emerged as the possible solution to low-power transistor devices and high-charge-density capacitors. Although the steep switching characteristic (subthreshold swing < sub-60 mV/dec) has been demonstrated in various devices combining the conventional transistors with ferroelectric gates, the actual applications of the NC effects are still some way off owing to the inherent hysteresis problem. This work reinterpreted the hysteretic properties of the NC effects within the time domain and demonstrated that capacitance (charge) boosting could be achieved without the hysteresis from the Al2O3/BaTiO3 bilayer capacitors through short-pulse charging. This work revealed that the hysteresis phenomenon in NC devices originated from the dielectric leakage of the dielectric layer. The suppression of charge injection via the dielectric leakage, which usually takes time, inhibits complete ferroelectric polarization switching during a short pulse time. It was demonstrated that a nonhysteretic NC effect can be achieved only within certain limited time and voltage ranges, but that these are sufficient for critical device applications.

A crossbar resistance switching memory readout scheme with sneak current cancellation based on a two-port current-mode sensing.

This paper describes a novel readout scheme that enables the complete cancellation of sneak currents in resistive switching random-access memory (RRAM) crossbar array. The current-mode readout is employed in the proposed readout, and a few critical advantages of the current-mode readout for crossbar RRAM are elucidated in this paper. The proposed scheme is based on a floating readout scheme for low power consumption, and one more sensing port is introduced using an additional reference word line. From the additional port, information on the sneak current amount is collected, and simple current-mode arithmetic operations are implemented to cancel out the sneak current from the sensing current. In addition, a simple method of handling the overestimated-sneak-current issue is described. The proposed scheme is verified using HSPICE simulation. Moreover, an example of a current-mode sense amplifier realizing the proposed cancelling technique is presented. The proposed sense amplifier can be implemented with less hardware overhead compared to the previous works.

Introduction to memristors and neuromorphic systems.

Recent generative artificial intelligence (AI) has exerted a profound and far-reaching global impact across diverse fields and society. However, it comes at the cost of substantial energy and computational resource consumption. Neuromorphic computing endeavors to create highly efficient computing hardware that emulates biological neural networks and even mimics some human brain functions, and it is expected to play an essential role in the next-generation computing hardware. Memristors open up novel opportunities for neuromorphic computing due to their feasible ability to mimic neural functions. Innovation in memristors may lead to novel algorithms and contribute to conventionally challenging tasks like nondeterministic polynomial time (NP)-hard problem. To this end, we present a themed collection in Materials Horizons and Nanoscale Horizons, in which we publish the latest developments in memristive materials, device fabrication, characterization, and circuit design for neuromorphic systems.

Tunable stochastic memristors for energy-efficient encryption and computing.

Information security and computing, two critical technological challenges for post-digital computation, pose opposing requirements - security (encryption) requires a source of unpredictability, while computing generally requires predictability. Each of these contrasting requirements presently necessitates distinct conventional Si-based hardware units with power-hungry overheads. This work demonstrates Cu0.3Te0.7/HfO2 ('CuTeHO') ion-migration-driven memristors that satisfy the contrasting requirements. Under specific operating biases, CuTeHO memristors generate truly random and physically unclonable functions, while under other biases, they perform universal Boolean logic. Using these computing primitives, this work experimentally demonstrates a single system that performs cryptographic key generation, universal Boolean logic operations, and encryption/decryption. Circuit-based calculations reveal the energy and latency advantages of the CuTeHO memristors in these operations. This work illustrates the functional flexibility of memristors in implementing operations with varying component-level requirements.

Spatiotemporal Data Processing with Memristor Crossbar-Array-Based Graph Reservoir.

Memristor-based physical reservoir computing (RC) is a robust framework for processing complex spatiotemporal data parallelly. However, conventional memristor-based reservoirs cannot capture the spatial relationship between the time-varying inputs due to the specific mapping scheme assigning one input signal to one memristor conductance. Here, a physical "graph reservoir" is introduced using a metal cell at the diagonal-crossbar array (mCBA) with dynamic self-rectifying memristors. Input and inverted input signals are applied to the word and bit lines of the mCBA, respectively, storing the correlation information between input signals in the memristors. In this way, the mCBA graph reservoirs can map the spatiotemporal correlation of the input data in a high-dimensional feature space. The high-dimensional mapping characteristics of the graph reservoir achieve notable results, including a normalized root-mean-square error of 0.09 in Mackey-Glass time series prediction, a 97.21% accuracy in MNIST recognition, and an 80.0% diagnostic accuracy in human connectome classification.

A high-dimensional in-sensor reservoir computing system with optoelectronic memristors for high-performance neuromorphic machine vision.

In-sensor reservoir computing (RC) is a promising technology to reduce power consumption and training costs of machine vision systems by processing optical signals temporally. This study demonstrates a high-dimensional in-sensor RC system with optoelectronic memristors to enhance the performance of the in-sensor RC system. Because optoelectronic memristors can respond to both optical and electrical stimuli, optical and electrical masks are proposed to improve the dimensionality and performance of the in-sensor RC system. An optical mask is employed to regulate the wavelength of light, while an electrical mask is used to control the initial conductance of zinc oxide optoelectronic memristors. The distinct characteristics of these two masks contribute to the representation of various distinguishable reservoir states, making it possible to implement diverse reservoir configurations with minimal correlation and to increase the dimensionality of the in-sensor RC system. Using the high-dimensional in-sensor RC system, handwritten digits are successfully classified with an accuracy of 94.1%. Furthermore, human action pattern recognition is achieved with a high accuracy of 99.4%. These high accuracies are achieved with the use of a single-layer readout network, which can significantly reduce the network size and training costs.

Heterogeneous reservoir computing in second-order Ta2O5/HfO2 memristors.

Multiple switching modes in a Ta2O5/HfO2 memristor are studied experimentally and numerically through a reservoir computing (RC) simulation to reveal the importance of nonlinearity and heterogeneity in the RC framework. Unlike most studies, where homogeneous reservoirs are used, heterogeneity is introduced by combining different behaviors of the memristor units. The chosen memristor for the reservoir units is based on a Ta2O5/HfO2 bilayer, in which the conductances of the Ta2O5 and HfO2 layers are controlled by the oxygen vacancies and deep/shallow traps, respectively, providing both volatile and non-volatile resistive switching modes. These several control parameters make the second-order Ta2O5/HfO2 memristor system present different behaviors in agreement with its history-dependent conductance and allow the fine-tuning of the behavior of each reservoir unit. The heterogeneity in the reservoir units improves the pattern recognition performance in the heterogeneous memristor RC system with a similar physical structure.

Thresholding Computing with Heterogeneous Integration of Memristive Kernel with Metal-Oxide-Semiconductor Capacitor for Temporal Data Analysis.

Precise event detection within time-series data is increasingly critical, particularly in noisy environments. Reservoir computing, a robust computing method widely utilized with memristive devices, is efficient in processing temporal signals. However, it typically lacks intrinsic thresholding mechanisms essential for precise event detection. This study introduces a new approach by integrating two Pt/HfO2/TiN (PHT) memristors and one Ni/HfO2/n-Si (NHS) metal-oxide-semiconductor capacitor (2M1MOS) to implement a tunable thresholding function. The current-voltage nonlinearity of memristors combined with the capacitance-voltage nonlinearity of the capacitor forms the basis of the 2M1MOS kernel system. The proposed kernel hardware effectively records feature-specified information of the input signal onto the memristors through capacitive thresholding. In electrocardiogram analysis, the memristive response exhibited a more than ten-fold difference between arrhythmia and normal beats. In isolated spoken digit classification, the kernel achieved an error rate of only 0.7% by tuning thresholds for various time-specific conditions. The kernel is also applied to biometric authentication by extracting personal features using various threshold times, presenting more complex and multifaceted uses of heartbeats and voice data as bio-indicators. These demonstrations highlight the potential of thresholding computing in a memristive framework with heterogeneous integration.

Memristive Crossbar Array-Based Probabilistic Graph Modeling.

Modern graph datasets with structural complexity and uncertainties due to incomplete information or data variability require advanced modeling techniques beyond conventional graph models. This study introduces a memristive crossbar array (CBA)-based probabilistic graph model (C-PGM) utilizing Cu0.3Te0.7/HfO2/Pt memristors, which exhibit probabilistic switching, self-rectifying, and memory characteristics. C-PGM addresses the complexities and uncertainties inherent in structural graph data across various domains, leveraging the probabilistic nature of memristors. C-PGM relies on the device-to-device variation across multiple memristive CBAs, overcoming the limitations of previous approaches that rely on sequential operations, which are slower and have a reliability concern due to repeated switching. This new approach enables the fast processing and massive implementation of probabilistic units at the expense of chip area. In this study, the hardware-based C-PGM feasibly expresses small-scale probabilistic graphs and shows minimal error in aggregate probability calculations. The probability calculation capabilities of C-PGM are applied to steady-state estimation and the PageRank algorithm, which is implemented on a simulated large-scale C-PGM. The C-PGM-based steady-state estimation and PageRank algorithm demonstrate comparable accuracy to conventional methods while significantly reducing computational costs.