Initial Resistance State Research Articles

Volatile Memristor in Leaky Integrate-and-Fire Neurons: Circuit Simulation and Experimental Study

In this paper, circuit implementation of a leaky integrate-and-fire neuron model with a volatile memristor was proposed and simulated in the SPICE simulation environment. We demonstrate that simple leaky integrate-and-fire (LIF) neuron models composed of: volatile memristor, membrane capacitance and neuron resistance can mimic spatial and temporal integration, firing function and signal decay. The existing leaky term originates from the recovery of the initial resistive state in the memristor in the spontaneous reset cycle, which is essential for emulating the forgetting process in all-memristive neural networks (MNNs). Furthermore, a diffusive perovskite memristor was used to validate the model where intrinsic memristors’ capacitance acts as neuron membrane capacitance. Good agreement with experimental and simulation results was observed. Volatility, as an inherent property of specific memristors, eliminates the need for usage of an additional peripheral circuit which will reinitialize device state, thus allowing the development of energy-efficient, large scale complex memristive neural networks. The presented circuit level model of LIF neurons can facilitate the design of MNNs.

Yttria-stabilized zirconia cross-point memristive devices for neuromorphic applications

An array of cross-point memristive devices has been implemented on the basis of yttria-stabilized zirconia thin film for applications in prototypes of spiking neural networks. The resistive switching phenomena and the plasticity nature of such memristive devices are studied. Reproducible bipolar resistive switching and precise tuning of resistive state are demonstrated and used to implement the plasticity rule according to STDP (spike-timing-dependent plasticity) mechanism. STDP learning is found to be dependent on the memristor's initial resistive state value and discussed in terms of the finite conductance change in studied structures. Obtained results provide the foundation for autonomous neuromorphic circuits with unsupervised learning development.

A Data-Driven Verilog-A ReRAM Model

The translation of emerging application concepts that exploit resistive random access memory (ReRAM) into large-scale practical systems requires realistic yet computationally efficient device models. Here, we present a ReRAM model, where device current–voltage characteristics and resistive switching rate are expressed as a function of: 1) bias voltage and 2) initial resistive state (RS). The model versatility is validated on detailed characterization data, for both filamentary valence change memory and nonfilamentary ReRAM technologies, where device resistance is swept across its operating range using multiple input voltage levels. Furthermore, the proposed model embodies a window function which features a simple mathematical form analytically describing RS response under constant bias voltage as extracted from physical device response data. Its Verilog-A implementation captures the ReRAM memory effect without requiring integration of the model state variable, making it suitable for fast and/or large-scale simulations and overall interoperable with current design tools.

A simple test for ideal memristors

An ideal memristor is defined as a resistor with memory that, when subjected to a time-dependent current, , its resistance RM(q) depends only on the charge q that has flowed through it, so that its voltage response is . It has been argued that a clear fingerprint of these ideal memristors is a pinched hysteresis loop in their I– curves. However, a pinched I– hysteresis loop is not a definitive test of whether a resistor with memory is truly an ideal memristor because such a property is shared also by other resistors whose memory depends on additional internal state variables, other than the charge. Here, we introduce a very simple and unambiguous test that can be utilized to check experimentally if a resistor with memory is indeed an ideal memristor. Our test is based on the duality property of a capacitor-memristor circuit whereby, for any initial resistance states of the memristor and any form of the applied voltage, the final state of an ideal memristor must be identical to its initial state, if the capacitor charge finally returns to its initial value. In actual experiments, a sufficiently wide range of voltage amplitudes and initial states are enough to perform the test. The proposed test can help resolve some long-standing controversies still existing in the literature about whether an ideal memristor does actually exist or it is a purely mathematical concept.

Diverse spike-timing-dependent plasticity based on multilevel HfO x memristor for neuromorphic computing

Memristors have emerged as promising candidates for artificial synaptic devices, serving as the building block of brain-inspired neuromorphic computing. In this letter, we developed a Pt/HfO x /Ti memristor with nonvolatile multilevel resistive switching behaviors due to the evolution of the conductive filaments and the variation in the Schottky barrier. Diverse state-dependent spike-timing-dependent-plasticity (STDP) functions were implemented with different initial resistance states. The measured STDP forms were adopted as the learning rule for a three-layer spiking neural network which achieves a 75.74% recognition accuracy for MNIST handwritten digit dataset. This work has shown the capability of memristive synapse in spiking neural networks for pattern recognition application.

Error free physically unclonable function with programmed resistive random access memory using reliable resistance states by specific identification-generation method

A high performance physically unclonable function (PUF) implemented with WO3 resistive random access memory (ReRAM) is presented in this paper. This robust ReRAM-PUF can eliminated bit flipping problem at very high temperature (up to 250 °C) due to plentiful read margin by using initial resistance state and set resistance state. It is also promised 10 years retention at the temperature range of 210 °C. These two stable resistance states enable stable operation at automotive environments from −40 to 125 °C without need of temperature compensation circuit. The high uniqueness of PUF can be achieved by implementing a proposed identification (ID)-generation method. Optimized forming condition can move 50% of the cells to low resistance state and the remaining 50% remain at initial high resistance state. The inter- and intra-PUF evaluations with unlimited separation of hamming distance (HD) are successfully demonstrated even under the corner condition. The number of reproduction was measured to exceed 107 times with 0% bit error rate (BER) at read voltage from 0.4 to 0.7 V.

(Invited) Material and Device Design for Practical Application of Highly-Reliable and High-Density ReRAM

Electric-pulse-induced resistance change phenomena in PrCaMnO3 were reported in 2000 [1]. From this finding, Resistive Random Access Memory (ReRAM) using this phenomena attracted both academic and industrial researchers as a promising candidate of next-generation nonvolatile memory. In 2013, we started the world’s first mass-production of ReRAM-embedded micro-controller unit. Among issues to be solved for its practical application, the most fundamental problem was that the mechanism of resistive switching (RS) remained unclear, causing the lack of guide for the design of materials and devices for ReRAMs. Therefore, we investigated the mechanism of RS and considered the appropriate materials and device structures for ReRAM from the viewpoint of device reliability. Based on the clarified RS mechanism, we successfully developed highly-reliable Ta oxide ReRAM. ReRAM cells with top electrode/Ta oxide/bottom electrode was fabricated. Ta oxide was deposited by reactive sputtering. Device structures, RS characteristics and reliability of ReRAMs were evaluated by electrical measurements and physical analyses.In order to elucidate the mechanism of RS, chemical states in ReRAM cells for initial resistance state, high-resistance state (HRS) and Low-resistance state (LRS) were directly observed by hard x-ray photoemission spectroscopy. The ratio of Ta 4d photoemission peaks for Ta sub-oxide (TaOx) to Ta2O5 increases with the decrease in the resistance of ReRAM cells (from Initial to HRS to LRS). This indicates that the mechanism of RS is redox reactions in Ta oxide layer. Based on the RS model, the guidelines for the material selection of oxide and electrode were presented. Oxide material should be bi-stable: The Gibbs free energy for the reaction between oxides with different valence states such as TaO2/Ta2O5should be small. Electrode material should not be easy to oxidize: Standard electrode potential of the electrode metal should be sufficiently higher than that of material constituting the RS oxide [2]. We next investigated the detailed structure of Ta oxide ReRAM and found that the RS locally occurred in small conductive area (Conductive Filament: CF) consisting of oxygen-deficient Ta oxide [3]. Therefore, we evaluated the retention characteristics of Ta oxide ReRAMs focusing on the CF. Experimental results indicate that the retention degradation of LRS/HRS is caused by the oxygen diffusion into/out of CF. From detailed analyses, we clarified that the control of CF characteristics including filament size and the density of oxygen vacancy was the key for the improvement of the reliability [4]. Utilizing these insights, we established the reliability model of Ta oxide ReRAM and successfully started the mass-production of ReRAM. The delivery of higher-density ReRAM is essential for further spread of ReRAM in the market. The following two the issues are inevitable for the realization of higher-density ReRAM; (1) scaling down of ReRAM cells and (2) low-current operation, which may degrade the reliability. Therefore, we addressed these challenges and successfully demonstrated the good array-level reliability with scaled-down cells and low-current operation by controlling the CF from the standpoint of fabrication process and device structure, respectively [5,6].

Improving Detection Accuracy of Memristor-Based Bio-Signal Sensing Platform.

Recently a novel neuronal activity sensor exploiting the intrinsic thresholded integrator capabilities of memristor devices has been proposed. Extracellular potentials captured by a standard bio-signal acquisition platform are fed into a memristive device which reacts to the input by changing its resistive state (RS) only when the signal ampitude exceeds a threshold. Thus, significant peaks in the neural signal can be stored as non-volatile changes in memristor resistive state whilst noise is effectively suppressed. However, as a memristor is subjected to increasing numbers of supra-threshold stimuli during practical operation, it accumulates (RS) changes and eventually saturates. This leads to severely redsignaluced neural activity detection capabilities. In this work we explore different signal processing and memristor operating procedure strategies in order to improve the detection rate of significant neuronal activity events. We analyse the data obtained from a single-memristive device biased with a reference neural recording and observe that performance can be improved markedly by a) increasing the frequency at which the memristor is reset to an initial resistive state where it is known to be highly responsive, b) appropriately preconditioning the input waveform through application of gain and offset in order to optimally exploit the intrinsic device behaviour. All results are validated by benchmarking obtained spike detection performance against a state-of-the-art template matching system utilising computationally-heavy, multi-dimensional, principal component analysis.

(Invited) Mechanism Study of Reversible Resistivity Change in Oxide Thin Film

The field of applied physics is driven by the need to understand materials structure and electronic properties with the ultimate goal of harnessing them in functional devices such as nonvolatile memories. One needs to understand the mechanism, which dictates the behavior of the charged defects and electrons under external stimuli such as electrostatic potential, thermal gradient and photons, to obtain a more microscopic understanding of the system. Then, one arrives at a predictive description of designing functional devices. In the first part of the talk, we report the x-ray-induced reversible resistance change in 40-nm-thick TiO2 films sandwiched by Pt top and bottom electrodes, and propose the physical mechanism behind the emergent phenomenon. Our findings indicate that there exists a photovoltaic-like effect, which modulates the resistance reversibly by a few orders of magnitude, depending on the intensity of impinging x-rays. We found that this effect, combined with the x-ray irradiation induced phase transition confirmed by transmission electron microscopy, triggers a non-volatile reversible resistance change. Understanding x-ray-controlled reversible resistance changes can provide possibilities to control initial resistance states of functional materials, which could be useful for future nonvolatile memories. In the second part, we cover the investigation of Ta2O5 powder and oxygen deficient thin films using synchrotron x-ray studies at the Advanced Photon Source, combining x-ray diffraction, extended x-ray absorption fine structure (EXAFS) and resonant x-ray scattering spectroscopy (RIXS) and Ab initio band structure calculations. Ta-based resistive switching devices have been extensively investigated due to their fast switching and reliable endurance among other materials. Despite extensive recent interests, however, there is still lack of fundamental understanding of electronic structure and local structure of the Ta-based device. Oxygen vacancies play essential roles in the electric functionality of resistive switching devices, which is based on the formation and rupture of conducting paths within the insulating matrix. We found that there are strong correlations among oxygen vacancy number and positions and energy gaps. Ab initio band structure calculations successfully explain the evolution of the electronic excitation spectrum as a function of oxygen vacancy number and positions and importantly provide a predictive description of the oxygen deficient Ta oxide that may improve the desired performance based on atomic level design rather than the traditional trial-error methods.

X-ray Irradiation Induced Reversible Resistance Change in Pt/TiO2/Pt Cells

The interaction between X-rays and matter is an intriguing topic for both fundamental science and possible applications. In particular, synchrotron-based brilliant X-ray beams have been used as a powerful diagnostic tool to unveil nanoscale phenomena in functional materials. However, it has not been widely investigated how functional materials respond to the brilliant X-rays. Here, we report the X-ray-induced reversible resistance change in 40-nm-thick TiO2 films sandwiched by Pt top and bottom electrodes, and propose the physical mechanism behind the emergent phenomenon. Our findings indicate that there exists a photovoltaic-like effect, which modulates the resistance reversibly by a few orders of magnitude, depending on the intensity of impinging X-rays. We found that this effect, combined with the X-ray irradiation induced phase transition confirmed by transmission electron microscopy, triggers a nonvolatile reversible resistance change. Understanding X-ray-controlled reversible resistance changes can provide possibilities to control initial resistance states of functional materials, which could be useful for future information and energy storage devices.

Improved endurance in ultrathin Al2O3 film with a reactive Ti layer based resistive memory

A 3-nm-thick Al2O3 based resistive memory with a Ti layer was prepared in this work. The Ti/HfOx devices with the same thickness were also fabricated for comparison. The oxygen gettering of Ti form Al2O3 is lower than that from HfOx. The Al2O3 devices with strong dielectric strength exhibit tight distribution of initial resistance state and high resistance state. The low resistance state, SET and RESET voltage of the devices seems insensitive to the dielectric film. Without an ideal current limiter, high forming voltage (VF) leads to the Al2O3 device with poor endurance (<100 cycles). Reduction of current overshoot by an external resistor of 800Ω during forming process, the endurance of the memory device can be improved and increase up to 104 cycles. The current overshoot and VF in the Al2O3 device with high operational temperature during forming step can be also suppressed.

Resistive Switching Characteristics of Cu/SiO&lt;sub&gt;2&lt;/sub&gt;/Pt Structure

SiO2thin films were fabricated as resistive layers of Cu/SiO2/Pt devices to investigate resistive switching properties. A thermal annealing was performed to allow for the diffusion of Cu ions into the SiO2thin films, leading to the formation of Cu-doped SiO2layers. Occurrence probabilities of the resistive switching and initial resistance-states of the devices were influenced by SiO2thickness, which was dependent on the Cu diffusion status within the SiO2layer. The resistive switching behaviors were characterized by the voltage sweeping mode and the current sweeping mode. The current sweeping mode provided a desired compliance current to well control the resistive switching from the high resistance-state to the low resistance-state (SET). Therefore, the large RESET (from the low resistance-state to the high resistance-state) current was not inherent in the device, due to poor control of the compliance current by the voltage sweeping mode. The current sweeping mode is a simple method to characterize the RESET current.

Impact of insertion of ultrathin TaO x layer at the Pt/TiO 2 interface on resistive switching characteristics

A TaO x layer as thin as ∼1 nm was inserted at the interface between top Pt electrode and TiO 2 as a restive switching dielectric to find a clue for a control of the resistive switching characteristics. By inserting an ultrathin TaO x layer, a marked change in initial resistance state and an improvement of endurance property were observed. These results can be interpreted in terms that such an ultrathin TaO x layer works as an oxygen reservoir.