ReRAM Devices Research Articles

Fully visible light controlled neuro-synaptic ReRAM device based on metal oxide heterojunction

Photoelectric synaptic devices as a combination of electronic synapse and photodetector are considered as emerging bio-inspired device technologies. These devices have immense potential to conquer the bottleneck of von Neumann architecture based traditional computing systems. In this Letter, we propose an all-oxide based photoelectric neuro-synaptic resistive random access memory device with the structure of ITO/Ga2O3/ZnO/ITO/Glass, in which the conductance states are reversibly tuned by two different wavelengths (405/522 nm) of visible light spectrum. The strength of light pulse is altered to investigate the learning and forgetting phases of the photoelectric response of the device. A basic biomimetic function “learning-forgetting-rehearsal” behavior is imitated up to 20 cycles. Moreover, emulation of some typical synaptic functions such as associative learning and switching between short and long term plasticities indicate the wavelength awareness of the device. Based on the pure optically induced potentiation/depression characteristics, convolutional neural network simulation achieves an overall test accuracy of 82.5% for the classification of Zalando's article images. The noise tolerance capability of neural network is also examined by applying “salt and pepper” noise in high proportion (75%) to corrupt the images. This work may provide a promising step toward the development of transparent electronics in optogenetics-inspired neuromorphic computing.

Bidirectional Photoelectric Response of Perovskite Oxide-Based Synaptic ReRAM Device

Bidirectional Photoelectric Response of Perovskite Oxide-Based Synaptic ReRAM Device

Analytical modelling of the transport in analog filamentary conductive-metal-oxide/HfOx ReRAM devices.

The recent co-optimization of memristive technologies and programming algorithms enabled neural networks training with in-memory computing systems. In this context, novel analog filamentary conductive-metal-oxide (CMO)/HfOx redox-based resistive switching memory (ReRAM) represents a key technology. Despite device performance enhancements reported in literature, the underlying mechanism behind resistive switching is not fully understood. This work presents the first physics-based analytical model of the current transport and of the resistive switching in these devices. As a case study, analog TaOx/HfOx ReRAM devices are considered. The current transport is explained by a trap-to-trap tunneling process, and the resistive switching by a modulation of the defect density within the sub-band of the TaOx that behaves as electric field and temperature confinement layer. The local temperature and electric field distributions are derived from the solution of the electric and heat transport equations in a 3D finite element ReRAM model. The intermediate resistive states are described as a gradual modulation of the TaOx defect density, which results in a variation of its electrical conductivity. The drift-dynamics of ions during the resistive switching is analytically described, allowing the estimation of defect migration energies in the TaOx layer. Moreover, the role of the electro-thermal properties of the CMO layer is unveiled. The proposed analytical model accurately describes the experimental switching characteristic of analog TaOx/HfOx ReRAM devices, increasing the physical understanding and providing the equations necessary for circuit simulations incorporating this technology.

Flexible SnO2–MoS2 based memristive device exhibiting stable and enhanced memory phenomenon

Flexible non-volatile memory devices have been gaining interest in expanding the digital data storage world. Due to the burgeoning advancement in the healthcare industry, the Internet of Things, and wearable electronics, the demand for ultra-thin, low-power, and flexible memory is increasing. Further, the advancement of synthesis procedures for two-dimensional nanomaterials having better optical, electrical, and mechanical strength with flexibility has fuelled the flexible memory device area, as commonly used flash memory is approaching its physical limit. In this context, the present work reports the flexible resistive switching memory device based on pure molybdenum disulfide (MoS2) and tin oxide (SnO2) based nanocomposite powder synthesized using the simple hydrothermal process. The nanocomposite formation was characterized using x-ray diffraction and Raman spectroscopic techniques. The memory device was fabricated by spin-coating the pure MoS2, pure SnO2, and MoS2–SnO2 nanocomposite film over the ITO-PET flexible substrate. The device was completed by thermally evaporating the thin Al layer through a shadow mask. It was found that MoS2–SnO2 based memory devices exhibited improved switching performance having a higher I ON/I OFF ratio, and lower switching voltage in comparison to pure MoS2 and pure SnO2-based devices (I ON/I OFF ratio ∼100, V = 0.5 V). Furthermore, to check the stability and cyclic performance of the fabricated device, the retention and endurance test was also performed, and the MoS2–SnO2 device retained the HRS and LRS states up to 2 × 103s and showed stable performance up to 100 switching cycles without much degradation, respectively. It should be mentioned that the presently proposed ReRAM device based on SnO2 and MoS2 with flexible and low-power features had excellent potential for use in the wearable device industry.

Radiation-Hardened ReRAM Devices for Flexible Electronics

Radiation-Hardened ReRAM Devices for Flexible Electronics

(Invited) Potential of Silicon Oxide Films for Low-Cost and High-Performance Resistive Switching Devices

Although resistance switching in transition-metal oxide (TMO) films has been widely studied [1], the physics and chemistry of resistance switching in non-transition-metal oxide (nTMO) films remains under-studied [2]. This must be corrected because inexpensive and chemically stable device configurations are required for the future Internet-of-Things society. It is easily anticipated that the switching electric field of silicon oxide films is apt to be higher than that of TMO films because the dielectric constant of silicon oxide films is smaller than those of the TMO films. However, the use of two-layer stacks like SiOx/hi-k oxide reduces the switching electric field [3]. Therefore, the study of silicon oxide films is still meaningful.Many scientists have recently investigated resistance switching in sputter-deposited silicon oxide films in detail [4,5] because this structure dispenses with the TMO (Fig. 1). Many papers addressed the role of silicon sub-oxide (SiOx) [6] because it is anticipated that the unstable bonds of non-stoichiometric silicon oxide can create degraded, but reversible, conductive paths inside the film. However, it is not yet clear how the SiOx region can trigger resistance switching, how important the SiOx region is, and whether the SiOx region is the only determiner of resistance switching [7] (Figs. 2, 3). The author demonstrated that hot-electron injection from the Si substrate had great potential in triggering resistance switching and lowering the switching voltage of sputter-deposited Si oxide films [5,8]; it was also mentioned that Si precipitates played an important role in realizing repeatable bipolar switching [9] (Fig. 2). Relating to this study, the author also proposed the physical and chemical structure of conductive filaments and their switching behavior based on an analysis of a possibly equivalent circuit model [10] (Fig. 4). However, it was not definitely elucidated why unipolar switching is not easily observed in sputter-deposited silicon oxide films, even though it is not a TMO.Recently, the author performed various Monte Carlo simulations to elucidate the physical and chemical parameters that rule the unipolar switching process in sputter-deposited silicon oxide films. Generations of simple bond breaking, oxygen vacancies, metallic Si sites, and E’’ centers were implemented in the simulation algorithm [8]. All-positive voltage stress mode for both the electroforming process and the reset process will not yield devices with stable, repeatable switching [11]. On the other hand, the all-negative stress mode results in stable, repeatable switching because the recovery of the internal degradation of the Si oxide film is not completed [11] (Fig. 5). This difference stems from the physical asymmetry of the electrode materials (Fig. 1).Though some may consider that silicon oxide films are not preferable to ReRAM devices from the chemical points of view, the theoretical analysis provided by the author in this paper suggests that silicon oxide films can be applied to the ReRAM device.[1] Y. Tokura, Physics Today, vol. 56, pp. 50-55, 2003.[2] T. Yanagida, et al., Sci. Rep., vol. 3, No.1657, pp.1-6, 2013.[3] P. Broqvist and A. Pasquarello, Appl. Phys. Lett., vol. 91, 192905, 2007.[4] J. Yao, et al., Appl. Phys. Lett, vol. 93, pp. 253101-1-253101-3, 2008.[5] R. Yamaguchi, S. Sato, and Y. Omura, Jpn. J. Appl. Phys., vol. 56, pp. 041301-1-041301-6, 2017.[6] A. Mehonic, et al., J. Appl. Phys., vol. 111, pp. 074507-1-074507-9, 2012.[7] Y. Wang, et al., Appl. Phys. Lett., vol.102, pp. 042103-1-042103-5, 2013.[8] Y. Omura, Ind. J. Electrical Eng. & Comput. Sci., vol. 24, pp. 1367-1378, 2021.[9] Y. Omura, R. Yamaguchi, and S. Sato, IEEE Trans. Device Reliab. and Mat. Vol. 17, pp. 561-567 (2017).[10] Y. Omura, ECS J. Solid State Sci. and Technol., vol. 10, pp. 124006-1-124006-10, 2021.[11] Y. Omura, Materials Today Proc., vol. 20, pp. 273-282, 2020; the advanced study will be published in the ECS J. Solid State Sci. and Technol. Figure 1

Effects of stacking sequence and top electrode configuration on switching behaviors in ZnO-HfO2 hybrid resistive memories

The resistive switching (RS) behavior of multilayered ReRAM devices is typically influenced by two aspects: the intrinsic properties of the dielectric multilayer, and other extrinsic contributions. In this study, we deposited and investigated bi-layered memories incorporating ZnO and HfO2 dielectrics. Different stacking sequences of the two oxides (ZnO/HfO2 and HfO2/ZnO) were employed to explore the intrinsic contribution of the bilayer microstructure, while different top electrode configurations (Pt and Ti) were utilized to demonstrate the extrinsic impact. All devices exhibit characteristic bipolar RS behavior, showcasing their functionality. But interestingly, the device with a Pt top electrode displays digital RS behavior, while the one with a Ti top electrode exhibits analog RS behavior. The completely different switching types observed during the reset process of the ZnO-HfO2 hybrid resistive memories are attributed to different interface charge migration process and structural characteristic. And as compared to the tailoring effect of stacking sequences (bilayer microstructure and oxide/oxide interface), the tailoring effect of top electrode configuration (electrode materials and oxide/electrode interface) is much stronger. Based on the latter, the switching ratio can be remarkably improved by at least two orders of magnitude, the reset voltage (energy consumption) can also be significantly reduced, and other discussed switching parameters can also show greater improvement. The comprehensive comparative analysis of intrinsic and extrinsic contributions provides valuable insights for exploring the RS mechanism and optimizing the design of bi-layered memory devices.

The influence of thermal cycling on the activation energy of conduction electrons and filament temperature in Pt/NiOx/Pt ReRAMs

We investigate the electrical and thermal conduction properties of low- (ON) and high-resistance (OFF) states in Pt/NiOx/Pt based unipolar ReRAM devices during cooling and warming cycles between 300 and 180 K. The conduction electron-trap activation energy was found to decrease upon warming. Although thermal cycling did not significantly affect the average resistance-temperature coefficient of the Pt diffused conductive filaments in the system, the ON-state resistance fluctuations increase at high temperatures, indicating that ambient temperature significantly affects the sizes of the formed filaments. The mechanism behind these thermally activated changes is discussed.

A ReRAM-Based Convolutional Neural Network Accelerator Using the Analog Layer Normalization Technique

This paper presents a ReRAM-based convolutional neural network (CNN) accelerator with a new analog layer normalization (ALN) technique. The proposed ALN can be used to effectively reduce the effect of the conductance variation is ReRAM devices by normalizing the outputs of the vector-matrix multiplication (VMM) in the charge domain. The ALN achieves high energy and hardware efficiencies because it directly processes normalization of the VMM outputs without storing their values in memory and is merged into the neuron circuit of the accelerator. To verify the effect of the ALN through experiments, a VMM accelerator that consists of two 25 × 25 sized ReRAM arrays and peripheral circuits with ALN is used for a convolution layer with digital signal processing (DSP) in a field programmable gate array (FPGA). The MNIST dataset is used to train and inference a CNN employing two VMM accelerators that work as convolution layers in a pipelined manner. Despite the conductance variation of the ReRAM devices, the ALN successfully stabilizes the output distribution of the convolution layer, which improves the classification accuracy of the network. A final classification accuracy for the MNIST and Fashion-MNIST datasets of 96.2% and 83.1% is achieved, respectively, with an energy efficiency of 9.94 tera-operations per second per Watt (TOPS/W).

Analysis of random telegraph noise in resistive memories: The case of unstable filaments

Through Random Telegraph Noise (RTN) analysis, valuable information can be provided about the role of defect traps in fine tuning and reading of the state of a nanoelectronic device. However, time domain analysis techniques exhibit their limitations in case where unstable RTN signals occur. These instabilities are a common issue in Multi-Level Cells (MLC) of resistive memories (ReRAM), when the tunning protocol fails to find a perfectly stable resistance state, which in turn brings fluctuations to the RTN signal especially in long time measurements and cause severe errors in the estimation of the distribution of time constants of the observed telegraphic events, i.e., capture/emission of carriers from traps. In this work, we analyze the case of the unstable filaments in silicon nitride-based ReRAM devices and propose an adaptive filter implementing a moving-average detrending method in order to flatten unstable RTN signals and increase sufficiently the accuracy of the conducted measurements. The τe and τc emission/capture time constants of the traps, respectively, are then calculated and a cross-validation through frequency domain analysis (Lorentzian fitting) was performed proving that the proposed method is accurate.

Exploiting device-level non-idealities for adversarial attacks on ReRAM-based neural networks

Resistive memory (ReRAM) or memristor devices offer the prospect of more efficient computing. While memristors have been used for a variety of computing systems, their usage has gained significant popularity in the domain of deep learning. Weight matrices in deep neural networks can be mapped to crossbar architectures with memristive junctions, generally resulting in superior performance and energy efficiency. However, the nascent nature of ReRAM technology is directly associated with the presence of inherent non-idealities in the ReRAM devices currently available. Deep neural networks have already been shown to be susceptible to adversarial attacks, often by targeting vulnerabilities in the networks’ internal representation of input data. In this paper, we explore the causal relationship between device-level non-idealities in ReRAM devices and the classification performance of memristor-based neural network accelerators. Specifically, our aim is to generate images which bypass adversarial defense mechanisms in software neural networks but trigger non-trivial performance discrepancies in ReRAM-based neural networks. To this end, we have proposed a framework to generate adversarial images in the hypervolume between the two decision boundaries, thereby leveraging non-ideal device behavior for performance detriment. We employ state-of-the-art tools in explainable artificial intelligence to characterize our adversarial image samples, and derive a new metric to quantify susceptibility to adversarial attacks at the pixel and device-levels.

Device Modeling Bias in ReRAM-Based Neural Network Simulations

Emerging technologies based on resistive switching (ReRAM) devices promise to improve the speed and energy efficiency of next generation machine learning accelerators, but further research is required for achieving commercial maturity. System-level prototyping with emerging devices is costly, and algorithmic investigations require hardware neural network modeling which often deviates from experimental reality. In this work, the concept of modeling bias is proposed as a way to quantify this deviation and support reliable evaluation of device populations in the context of neural network algorithms. While applicable to other device modeling techniques, modeling bias is investigated here using jump tables - a promising physics-less technique to model emerging memory devices for hardware networks. Questions about the fidelity of these tables in relation to stochastic device behavior are answered. Two methods of jump table modeling – binning and a novel Optuna-optimized binning - are explored using synthetic data with known distributions for benchmarking and experimental data obtained from TiOx ReRAM devices for practical testing. Novel device metrics are proposed, and it is shown that these metrics can present crucial insights on the device population prior to training the hardware network. Results on a multi-layer perceptron trained on MNIST show that device models based on binning deviate from target network accuracy at a low number of points and high switching noise in the device dataset. The proposed approach opens the possibility for device-algorithm co-design investigations into statistical device models with better performance, as well as experimentally verified modeling bias in different in-memory computing and neural network architectures.

Multilevel resistive switching with negative differential resistance in Al/NiO/ZnFe2O4/ITO ReRAM device

The Resistive Random Access Memory devices have emerged as an energy-efficient alternative to Von Neumann computers by enabling in-memory computing. Here we demonstrate bipolar resistive switching in thin films of Nickel oxide (NiO) and Zinc Ferrite (ZFO) using a simple Al/NiO/ZFO/ITO configuration, making them a possible candidate for the next generation memory devices. The fabricated device demonstrated excellent resistive switching behavior with high endurance for up to 1000 cycles, good retention for more than 104 s, and a good resistance ratio of HRS to LRS ∼102. Ohmic conduction was observed in the LRS, while in the HRS, along with ohmic conduction, space charge limited current (SCLC) and Schottky mechanisms were observed. Besides the LRS and HRS, a number of stable intermediate resistance states can also be obtained during the RESET process using different stop voltages, which makes the current device a multilevel resistive switching device.

Tailor-made synaptic dynamics based on memristive devices

The proliferation of machine learning algorithms in everyday applications such as image recognition or language translation has increased the pressure to adapt underlying computing architectures towards these algorithms. Application specific integrated circuits (ASICs) such as the Tensor Processing Units by Google, Hanguang by Alibaba or Inferentia by Amazon Web Services were designed specifically for machine learning algorithms and have been able to outperform CPU based solutions by great margins during training and inference. As newer generations of chips allow handling of and computation on more and more data, the size of neural networks has dramatically increased, while the challenges they are trying to solve have become more complex. Neuromorphic computing tries to take inspiration from biological information processing systems, aiming to further improve the efficiency with which these networks can be trained or the inference can be performed. Enhancing neuromorphic computing architectures with memristive devices as non-volatile storage elements could potentially allow for even higher energy efficiencies. Their ability to mimic synaptic plasticity dynamics brings neuromorphic architectures closer to the biological role models. So far, memristive devices are mainly investigated for the emulation of the weights of neural networks during training and inference as their non-volatility would enable both processes in the same location without data transfer. In this paper, we explore realisations of different synapses build from memristive ReRAM devices, based on the Valence Change Mechanism. These synapses are the 1R synapse, the NR synapse and the 1T1R synapse. For the 1R synapse, we propose three dynamical regimes and explore their performance through different synapse criteria. For the NR synapse, we discuss how the same dynamical regimes can be addressed in a more reliable way. We also show experimental results measured on ZrOx devices to support our simulation based claims. For the 1T1R synapse, we explore the trade offs between the connection direction of the ReRAM device and the transistor. For all three synapse concepts we discuss the impact of device-to-device and cycle-to-cycle variability. Additionally, the impact of the stimulation mode on the observed behavior is discussed.

Investigation of α-Fe2O3 and Cr dopedα-Fe2O3 Based Nano-Film as Resistive Switching Material for ReRAM Device Application

Investigation of α-Fe2O3 and Cr dopedα-Fe2O3 Based Nano-Film as Resistive Switching Material for ReRAM Device Application

Oxide Dissolution Mediated Formation of Conducting Filament in ReRAM Devices: A Phase Field Study

Oxide Dissolution Mediated Formation of Conducting Filament in ReRAM Devices: A Phase Field Study

A Novel Architecture Design for Output Significance Aligned Flow with Adaptive Control in ReRAM-based Neural Network Accelerator

Resistive-RAM-based (ReRAM-based) computing shows great potential on accelerating DNN inference by its highly parallel structure. Regrettably, computing accuracy in practical is much lower than expected due to the non-ideal ReRAM device. Conventional computing flow with fixed wordline activation scheme can effectively protect computing accuracy but at the cost of significant performance and energy savings reduction. For such embarrassment of accuracy, performance and energy, this article proposes a new Adaptive-Wordline-Activation control scheme ( AWA-control ) and combines it with a theoretical Output-Significance-Aligned computing flow ( OSA-flow ) to enable fine-grained control on output significance with distinct impact on final result. We demonstrate AWA-control -supported OSA-flow architecture with maximal compatibility to conventional crossbar by input retiming and weight remapping using shifting registers to enable the new flow. However, in contrast to the conventional computing architecture, the OSA-flow architecture shows the better capability to exploit data sparsity commonly seen in DNN models. So we also design a sparsity-aware OSA-flow architecture for further DNN speedup. Evaluation results show that OSA-flow architecture can provide significant performance improvement of 21.6×, and energy savings of 96.2% over conventional computing architecture with similar DNN accuracy.

Neural network learning using non-ideal resistive memory devices

We demonstrate a modified stochastic gradient (Tiki-Taka v2 or TTv2) algorithm for deep learning network training in a cross-bar array architecture based on ReRAM cells. There have been limited discussions on cross-bar arrays for training applications due to the challenges in the switching behavior of nonvolatile memory materials. TTv2 algorithm is known to overcome the device non-idealities for deep learning training. We demonstrate the feasibility of the algorithm for a linear regression task using 1R and 1T1R ReRAM devices. Using the measured device properties, we project the performance of a long short-term memory (LSTM) network with 78 K parameters. We show that TTv2 algorithm relaxes the criteria for symmetric device update response. In addition, further optimization of the algorithm increases noise robustness and significantly reduces the required number of states, thereby drastically improving the model accuracy even with non-ideal devices and achieving the test error close to that of the conventional learning algorithm with an ideal device.

Effect of electron conduction on the read noise characteristics in ReRAM devices

The read variability of redox based resistive random access memory is one of the key characteristics with regard to its application in both data storage and novel computation in memory or neuromorphic architectures. While intrinsic noise limits the number of distinguishable states, it may be beneficial for probabilistic computing or to prevent overfitting. Thus, application and material system need to be carefully matched according to their read noise characteristics. Preceding density functional theory simulations suggested dividing oxides used in valence change memory into two categories based on the dominating conduction mechanism. We provide a comprehensive experimental study, which confirms the simulations and demonstrates how the conduction mechanism affects the variability. We analyze the signal-to-noise ratio (SNR) of five different switching oxides, revealing that oxides with shallow defect states (type 1) show high SNR whereas oxides with deep defect states (type 2) exhibit pronounced ionic noise. Thus, our results provide valuable input toward tuning of read noise characteristics by material design.

Electrical conduction and resistive switching in cow milk-based devices prepared using the spin-coat method

Electronic devices made of organic materials are critical for sustainable technology and the reduction of electronic waste. In this paper, we report on the resistive switching (RS) behavior of an organic full cream cow milk-based film placed between two electrodes. The switching properties of the milk film were investigated for use in RS memory (ReRAM), a new type of memory device. Two devices were created: one with electrodes made of silver (Ag) and indium doped tin oxide (ITO) (Ag/cow milk/ITO) and the other with electrodes made of Ag and tungsten (W) (Ag/cow milk/W). The study is unique in two ways: (1) the use of the spin coat method, which allowed us to deposit 15 nm thin milk films, and (2) the use of ITO and W bottom electrodes, which allowed us to compare the results. Our findings show that both ITO- and W-based devices exhibit RS properties with “S-type” bipolar behavior. The ITO-based device exhibited switching at 0.77 V and an ION/IOFF ratio of ∼2. The memory behavior of this device lasted ∼7 write/erase cycles, indicating the possibility of nonvolatile memory application. In contrast, the switching in the W-based device was characterized by low (0.1 V) voltage, and it lasted for only one cycle. Increasing the compliance current to 0.5 A improved the ON/OFF ratio to 102, but the device could not repeat the same hysteresis behavior. The analysis of the conduction and switching mechanisms in both devices revealed a conductive bridge mechanism attributed to the metalization of metal ions naturally found in milk. Overall, our findings show that the thickness of the milk film influences both conduction and switching mechanisms and that the (milk film)/ITO and (milk film)/W interfaces influence the conductance of full cream cow milk-based ReRAM devices.