Search

With a great success of the generative model via deep neural networks, monocular depth estimation has been actively studied by exploiting various encoder-decoder architectures. However, the decoding process in most previous methods, which repeats simple up-sampling operations, probably fails to fully utilize underlying properties of well-encoded features for monocular depth estimation. To resolve this problem, we propose a simple but effective scheme by incorporating the Laplacian pyramid into the decoder architecture. Specifically, encoded features are fed into different streams for decoding depth residuals, which are defined by decomposition of the Laplacian pyramid, and corresponding outputs are progressively combined to reconstruct the final depth map from coarse to fine scales. This is fairly desirable to precisely estimate the depth boundary as well as the global layout. We also propose to apply weight standardization to pre-activation convolution blocks of the decoder architecture, which gives a great help to improve the flow of gradients and thus makes optimization easier. Experimental results on benchmark datasets constructed under various indoor and outdoor environments demonstrate that the proposed method is effective for monocular depth estimation compared to state-of-the-art models. The code and model are publicly available at: | https://github.com/tjqansthd/LapDepth-release |.

Various images captured in complicated lighting conditions often suffer from deterioration of the image quality. Such poor quality not only dissatisfies the user expectation but also may lead to a significant performance drop in many applications. In this paper, anovel method for low-light image enhancement is proposed by leveraging useful propertiesof the Laplacian pyramid both in image and feature spaces. Specifically, the proposed method, so-called a deep stacked Laplacian restorer (DSLR), is capable of separately recovering the global illumination and local details from the original input, and progressively combining them in the image space. Moreover, the Laplacian pyramid defined in the feature space makes such recovering processes more efficient based on abundant connectionsof higher-order residuals in a multiscale structure. This decomposition-based scheme is fairly desirable for learning the highly nonlinear relation between degraded images and their enhanced results. Experimental results on various datasets demonstrate that the proposed DSLR outperforms state-of-the-art methods. The code and model are publicly available at: https://github.com/SeokjaeLIM/DSLR-release .

We demonstrate a hybrid memory device with both memory and selector characteristics by using Al2O3 (memory) and TiO2 (selector) layers. The proposed device exhibits tunable threshold voltage characteristics based on the length of the residual conductive filament (CF) in the Al2O3 layer. To confirm the switching characteristics, the stability of the metallic CFs in each of the Al2O3 and TiO2 layers is investigated via electrical measurements. Our findings reveal that the CF in the TiO2 layer, which acts as a selector device, spontaneously ruptures. In contrast, a stable residual CF is observed in the Al2O3 layer, which acts as a memory device. Furthermore, we confirm that the tunable threshold voltage characteristics relate to the different tunneling gap distance from the residual CF in Al2O3. These results enable the hybrid memory device to achieve a high on–off ratio, high selectivity, and low leakage current without integrating an additional selector device, thereby enabling implementation of a high-density three-dimensional cross-point array.

Inspired by the great success of the deep neural networks in various fields of computer vision, studies for image deblurring have begun to become more active in recent days. However, most previous approaches often fail to accurately remove the blur artifacts, e.g., ghosting effects at the object boundaries and degradation of local details, in restored results. In this paper, we propose a deep spectral-spatial network (DSSN) for resolving the problem of single image deblurring. Specifically, the proposed method is able to efficiently recover scene characteristics in a global manner by minimizing differences of the frequency magnitude between the blurred input and corresponding sharp image via the spectral restorer, and the spatial restorer fine-tunes local details of the intermediate result, which is estimated by the spectral one, based on the intensity similarity. This cascaded scheme of deblurring processes is fairly desirable for clearly restoring edge-like structures as well as the textural information in a coarse-to-fine manner. Experimental results on benchmark datasets demonstrate that the proposed DSSN outperforms state-of-the-art methods. The code and model are publicly available at: https://github.com/SeokjaeLIM/DSSN_release .

Hardware neural networks (HNNs) which use synapse device (SD) arrays show promise as an approach to energy efficient parallel computation of massive vector-matrix multiplication. To maximize the inference accuracy of application-specific HNNs, we propose a highly reliable 2-terminal SD with fixed resistance based on WOx films. First, we investigate the device requirements of an array-based HNN through MATLAB and SPICE simulations taking into account the parasitic resistance effects in the array. On top of that, to fabricate the SD we utilize the intrinsic properties of the WOx film, which exhibits substantial changes in conductivity from 10−8 to 104 Ω−1cm−1 by varying the oxygen vacancy concentration. After the process optimization of oxide stoichiometry and electrode materials, we can form nanoscale WO×-based SDs with excellent conductance uniformity and I-V linearity. Our results show that inference accuracy is significantly improved by using WOx-based SD arrays even in advanced node scaling. Through experimental hardware implementation, 16 × 16 pixel images are correctly classified and we show the potential of WOx-based SD for future large-scale HNN applications.

In this work, we explore the electrical properties and data retention of Te-based conductive-bridge random-access memory (CBRAM) of ZrxTe1−x/Al2O3/Pt cells. The virgin resistance and forming voltage are decreased with increasing Zr composition (0.5 ≦ x ≦ 0.7) and decreasing electrolyte thickness. The resistance of the conductive filament (CF) formed in the Te-CBRAM shows semiconducting behavior that is decreased with increasing temperature, whereas a Cu-based CBRAM shows metallic behavior. Furthermore, the conductance change of Te-based CBRAM, during the filament dissolution step, occurs with lower conductance units than those of Cu/Ag-based CBRAM. The most differentiable characteristics of both devices are the data retention. Te-based CBRAM shows better data stability at high temperature (150 °C) than Cu-based device. Accelerated tests (250, 270, and 300 °C) were performed to understand the data retention of the Te-CBRAM, yielding excellent retention characteristics (10 years at 177 °C) despite its relatively low activation energy (Ea, 1.07 eV) than Cu/Ag- based devices. We believe that the excellent retention properties of Te-based devices are more influenced by the wide effective CF size than by Ea.

In this study, we introduce a lithium (Li) ion-based three-terminal (3-T) synapse device using WO

Emerging nonvolatile memory devices, also named memristive devices, have drawn great attention due to their potential for ultra-high density memory application. To achieve such high-density memristive device, a crossbar array structure is introduced. However, implementing such crossbar array requires addressing issues such as unwanted current flow in the unselected cell, so-called the “sneak-path” current. Therefore an access device called a selector device is required to suppress such undesired current and to operate the target memory cell both selectively and nondestructively in the crossbar array. Although different kinds of selector devices have been studied, none of them met all requirements, such as simple two-terminal structure, highly nonlinear I–V characteristics, fast operation speed, high endurance, excellent uniformity, and good thermal stability. The insulator–metal transition (IMT), the ovonic threshold switching (OTS), and the conductive bridging RAM (CBRAM)-type device are promising candidates as a selector device due to their scalability. IMT- and OTS-based selectors exhibit excellent performance, such as high on-current density and extremely fast switching speed, but the off-current, especially for IMT-based selectors, still needs further improvement for the effective reduction of leakage current in the crossbar array. CBRAM-type selectors have ultralow leakage current, suitable for high selectivity and low power device applications. However, the poor on-current density and reliability issues hinder their applicability. There has been great effort to improve the performance of these three types of selector devices, and this chapter reviews recent progress and future directions on their implementation for high-density crossbar array.

In this paper, we demonstrate the linear conductance-change characteristics of a conductive-bridging RAM (CBRAM) to be employed as an artificial synapse device in neuromorphic systems. The CBRAM with a bilayer electrolyte structure ( ${\mathrm {Cu/Cu}}_{{2}-{x}}\text{S}$ / $\mathrm {WO}_{{3}-{x}}$ /W) exhibits analog switching behavior during the depression process due to the well-controlled dissolution of the conductive filament. To analyze the origin of this motion, we investigate the effective voltage applied to $\mathrm {Cu}_{{2}-{x}}\text{S}$ and $\mathrm {WO}_{{3}-{x}}$ . Our findings reveal that $\mathrm {Cu}_{{2}-{x}}\text{S}$ , acting as a voltage divider, helps in suppressing the large voltage drop in $\mathrm {WO}_{{3}-{x}}$ , where the formation/dissolution of filament occurs. Furthermore, due to the diode-like characteristics of $\mathrm {Cu}_{{2}-{x}}\text{S}$ and the division of voltage drop between $\mathrm {WO}_{{3}-{x}}$ and $\mathrm {Cu}_{{2}-{x}}\text{S}$ , an optimum programming energy is applied to $\mathrm {WO}_{{3}-{x}}$ during the depression process. This leads to linear conductance-change characteristics under identical pulses. However, abrupt conductance-change characteristics are observed during the potentiation process. Thus, we use only the device characteristics of the depression part for the neuromorphic system. An excellent classification accuracy is achieved due to the linear conductance-change characteristics and optimized pulse conditions.

In this letter, we demonstrate the conductive-bridging RAM (CBRAM) with excellent multi-level cell (MLC) and linear conductance characteristics for an artificial synaptic device of neuromorphic systems. Our findings show that inherent characteristics of CBRAM can achieve the linear conductance and MLC characteristics as a product of an integer unit of the conductance. However, uncontrolled metal-ion injection into the switching layer results in a significant degradation of device uniformity, leading to degradation in the classification accuracy. Thus, we introduce a multi-layer CBRAM configuration (Cu/HfO2/Ta/Cu2S/W) to control the ionic motion in electrolytes. As a result of device engineering, highly improved classification accuracy is achieved using CIFAR-10 data set.

The effects of liner thickness on the reliability of AgTe/TiO2-based threshold switching (TS) devices were investigated. The off-state current of an AgTe/TiO2/Pt TS device was found to be significantly increased by in-diffusion of Ag into the TiO2 layer during the annealing process. Therefore, 3-, 5- and 7-nm TiN liners were introduced and compared to prevent the in-diffusion of Ag. While the 3-nm TiN liner was shown to be incapable of blocking Ag in-diffusion into the TiO2 layer, the 5- and 7-nm liners effectively suppressed in-diffusion and maintained high off-state resistance. However, the TS device with the 7-nm TiN liner exhibited wide threshold voltage distribution and poor endurance characteristics owing to a lack of Ag sources. The TS device with a 5-nm TiN liner, by contrast, was found to have an adequate amount of Ag sources and to demonstrate thermally stable and electrically reliable characteristics. The effects of TiN liner on Ag diffusion were also directly confirmed using energy dispersive spectrometry line profiles, transmission electron microscopy imaging, and mapping analyses.

In this letter, we demonstrate a self-limited conductive-bridging random access memory (CBRAM) that removes the necessity for external current compliance in a one selector–one resistor (1S1R) architecture. The standard Ge2Sb2Te5 (GST) is used as a CBRAM switching layer. In addition, Te-rich GST is also considered. Their performance is then compared. Both samples exhibit self-limited on-current characteristics, and the on-currents of the standard GST and Te-rich GST are ~300 and $\sim 20~\mu \text{A}$ , respectively. The observed self-limited characteristics are caused by the Te in the GST layer because in the presence of Te, Cu tends to form a more stable CuTe phase that restrict Cu filament growth. Furthermore, to confirm the feasibility of crossbar array applications, the 1S1R device is evaluated using a Ag/TiO2-based threshold selector device reported in our previous work. Hence, we confirm leakage current reduction, a uniform resistance distribution, and stable retention characteristics in the 1S1R configuration with no external current compliance.

In this paper, we propose a one transistor-two resistive RAM (RRAM) (1T2R) device to overcome the non-ideal switching behavior of artificial synapse devices, such as the unidirectional and abrupt change in the conductance. Our findings reveal that the 1T2R device can exhibit bidirectional conductance changes using unidirectional switching RRAMs. Thus, we introduce a unidirectional but analog switching Cu-based RRAM device (Cu/Cu2-X S/WO3-X /W) having an internal voltage suppressor (Cu2-X S) to realize a bidirectional and analog 1T2R synapse device. The synaptic behaviors of the 1T2R device are calculated using the subthreshold region of an NMOSFET. In addition, we improve the on/off conductance ratio and conductance change linearity owing to the nonlinear current transition characteristics of the subthreshold region. Finally, we demonstrate that an ideal synaptic behavior can be achieved through the 1T2R device even when non-ideal switching RRAM elements are used.

We demonstrate ultra-thin ALD-processed dual-oxide (Al 2 O 3 /TiO 2 ) hybrid device with memory and selector characteristics by engineering the stability of metal filament in Al 2 O 3 and TiO 2 layer. The optimized hybrid memory device shows outstanding performances such as low off current $( , low reset current $( , and high on/off ratio $(> 10^{4})$ . Inserting a Ti buffer layer which has a low electrode potential value, we observed excellent uniformity and retention property. Finally, an outstanding read/write margins and ultra-low power consumption are confirmed through array simulations of the proposed hybrid memory device.

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.

We perform experiments and device simulations to investigate the origin of current-voltage (I-V) linearity of TaOX-based resistive switching memory (RRAM) devices for their possible application as electronic synapses. By using electrical characterization and simulations, we link the electrical characteristics (linear or nonlinear I-V) to the microscopic properties of the conductive filament (CF). Our findings indicate that the shape and the thermal properties of the CF region are crucial to achieve linear I-V characteristics. These results allow optimizing the I-V curve linearity of TaOX-based RRAM devices, explaining the wide range of linear I-V characteristics experimentally observed on RRAM device obtained. When weight sum operation using SPICE simulations is performed, the read current is improved under the condition of linear I-V characteristics due to current loss minimization.

Emerging storage devices based on new physical switching mechanisms (noncharge based) have received great interest to implement ultrahigh-density crossbar arrays. However, in addition to the introduction of these new memory devices, implementing crossbar arrays requires addressing issues such as undesired current flow and data disturbance in nonselected cells. Therefore, an access device called a selector device is required to selectively operate and read the target device in the crossbar array. Though many selector devices have been researched, the ideal selector device to meet the requirements for a crossbar array architecture such as two-terminal structure, highly nonlinear I-V characteristics, fast switching speed, and excellent device uniformity has not been reported yet. Therefore, this chapter reviews the previously reported selector devices and discusses the development direction of the selector devices.

In this study, we introduce a lithium (Li) ion-based three-terminal (3-T) synapse device using WO x as a channel. Our study reveals a key stoichiometry of WO2.7 for excellent synaptic characteristics that is related to Li-ion diffusivity. The open-lattice structure formed by oxygen deficiency promoted Li-ion injection and diffusion. The optimized stoichiometry and improved Li-ion diffusivity were confirmed by x-ray photoelectron spectroscopy analysis and cyclic voltammetry, respectively. Furthermore, the transient conductance change that inevitably occurs in ion-based synaptic transistors was resolved by applying a two-step voltage pulse scheme. As a result, we achieved a symmetric and linear weight-update characteristic with reduced program/erase operation time.

To improve the classification accuracy of an image data set (CIFAR-10) by using analog input voltage, synapse devices with excellent conductance linearity (CL) and multi-level cell (MLC) characteristics are required. We analyze the CL and MLC characteristics of TaO

The switching uniformity and reliability of the TaOx based resistive random access memory (RRAM) device were investigated with varying nitrogen doping concentration. The nitrogen doped samples shows excellent electrical and reliability characteristics such as small switching variability for 3-bit multilevel per cell (MLC), low power operation and good retention properties. Compared with control sample, improved device characteristics of nitrogen doped device can be explained by nitrogen inducedfilament confinement. © The Author(s) 2015. Published by ECS. This is an open access article distributed under the terms of the Creative Commons Attribution 4.0 License (CC BY, http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse of the work in any

AgTe/TiN/TiO 2 /TiN threshold switching (TS) device was monolithically integrated with silicon MOSFET to demonstrate steep subthreshold slope field-effect transistors. The TS device with AgTe top electrode showed the high on-current, since the Te allows an extraction of the Ag out of the filament. The TiN liner was also inserted at the AgTe/TiO 2 interface to prevent in-diffusion of Ag into the TiO2 layer during back-end-of-line process. Finally, the transistor with TS device has a sub-5-mV/dec subthreshold slope (SS) and a high on/off current ratio (Ion/I off ) of >108 with a low drain voltage (0.5 V) even after the 400°C annealing process.

To realize a steep-slope-FET with low leakage current and low operating bias, we engineered two types of atom-switch devices and integrated them with a silicon MOSFET. The integrated atom-switch-FET exhibits extremely low leakage current (10−7 μA/μm), high I on /I off ratio (> 107), low operating bias ( off and Vth, as for optimal performances of atom-switch-FET were investigated at various operating bias conditions.

In this study, we investigate a proton-based three-terminal (3-T) synapse device to realize linear weight-update and I-V linearity characteristics for neuromorphic systems. The conductance states of the 3-T synapse device can be controlled by modulating the proton concentration in the WOx channel. Therefore, we estimate the dynamic change of proton concentration in the channel region, which directly affects synaptic behaviors. Our findings indicate that the supply of an excess number of protons from the SiO2-H electrolyte and low proton diffusivity in the WOx channel result in asymmetric and non-linear weight-update characteristics. In addition, though the linear I-V characteristics can be obtained using non-stoichiometric WOx, we observe that significant oxygen deficiency in the channel region increases the operating current levels. Thus, based on this information, we introduce optimized conditions of each component in the 3-T synapse device and shape of the gate voltage pulses. As a result, an excellent classification accuracy is achieved using linear weight-update and I-V linearity characteristics under optimized device and pulse conditions.

The effect of the interfacial layer on the reliability and off-current of a C-Te-based Ovonic threshold switching (OTS) device is investigated. By introducing an amorphous Ge (a-Ge) layer between both the top and bottom electrodes and the OTS film, a nearly 2-order of off-current reduction is confirmed. This could be attributed to the formation of a virtual electrode in the a-Ge layer that can scale down the effective area and localize electric field in a specific region. Area independent current-voltage characteristics were observed in the a-Ge introduced device while the without the barrier shows area dependent behavior, thus supporting the presence of a relatively small conductive filament that can act as a virtual electrode in the a-Ge inserted device. In addition, variabilities of the threshold voltage, hold voltage, and off-current are considerably improved owing to the confinement of the electric field. Transmission electron microscopy and energy dispersive spectroscopy analyses were employed to reveal the role of Ge at the interface. The findings are further supported by simulation using the thermally assisted hopping model.

We investigate the effect of Cu concentration On-state resistance retention characteristics of W/Cu/Ti/HfO2/Pt memory cell. The development of RRAM device for application depends on the understanding of the failure mechanism and the key parameters for device optimization. In this study, we develop analytical expression for cations (Cu+) diffusion model using Gaussian distribution for detailed analysis of data retention time at high temperature. It is found that the improvement of data retention time depends not only on the conductive filament (CF) size but also on Cu atoms concentration density in the CF. Based on the simulation result, better data retention time is observed for electron wave function associated with Cu+ overlap and an extended state formation. This can be verified by analytical calculation of Cu atom defects inside the filament, based on Cu+ diffusion model. The importance of Cu diffusion for the device reliability and the corresponding local temperature of the filament were analyzed by COMSOL Multiphysics simulation. (C) 2016 Author(s).

We investigated the reset breakdown phenomenon of HfOx-based resistive memory for reliable switching operation in a fully CMOS compatible stack. Through the understanding on the effect of electrode materials and device area, our findings show that observed failure is attributed to additional oxygen vacancies close to the electrode interface, where switching is occurred. Therefore, RuOx serving as an oxygen diffusion barrier was introduced to suppress the generation of unwanted oxygen vacancies by preventing out-diffusion of oxygen through the electrode. As a result, significantly enhanced breakdown strength in HfOx/RuOx stack is achieved and resulting in improved cycle endurance with larger on/off ratio. (C) 2016 Author( s).

In this paper, we demonstrate a volatile atomic switch that can be utilized for obtaining steep subthreshold swing (SS) ( 60 mV/dec). The result shows an improvement in the SS, which results from the transition of the atomic switch between the ON and OFF states, which is caused by the formation and rupture of a conductive filament. As a result, excellent switching characteristics are obtained for the FETs, such as low IOFF (∼10−5 μA/μm), high ION/IOFF ratio (∼105), low VDD (∼0.25 V), and steep SS ( 60 mV/dec). The result shows an improvement in the SS, which results from the transition of the atomic switch between the ON and OFF states, which is caused by the formation and rupture of a conductive filament. As a result, excellent switching characteristics are obtained for the FETs, such as low IOFF (∼10−5 μA/μm), high ION/IOFF ratio (∼105), low VDD (∼0.25 V), and steep SS (

To improve the classification accuracy of an image data set (CIFAR-10) by using analog input voltage, synapse devices with excellent conductance linearity (CL) and multi-level cell (MLC) characteristics are required. We analyze the CL and MLC characteristics of TaOx-based filamentary resistive random access memory (RRAM) to implement the synapse device in neural network hardware. Our findings show that the number of oxygen vacancies in the filament constriction region of the RRAM directly controls the CL and MLC characteristics. By adopting a Ta electrode (instead of Ti) and the hot-forming step, we could form a dense conductive filament. As a result, a wide range of conductance levels with CL is achieved and significantly improved image classification accuracy is confirmed.

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.

In this paper, a bipolar resistive random access memory (RRAM) device with a fab-friendly materials stack (TiN/Ti/HfO 2 /TiN) and process in a via-hole substrate with 200 nm cell size was successfully demonstrated on an 8-inch wafer scale. Furthermore, the robust device characteristics with reliable switching uniformity and stability were experimentally confirmed at the wafer level. Finally, from the standpoint of array architecture, the fabricated memory cell was evaluated with various selector devices such as a conventional silicon-based transistor and a newly developed tunneling-based diode device.

We demonstrate the device characteristics of W/Cu/N-GST/Al 2 O 3 /Pt conductive-bridging RAM, focusing on the nitrogen-doped Ge 2 Sb 2 Te 5 buffer layer to realize non-volatile memory applications. The on/off ratio of typical Cu/Al 2 O 3 -based CBRAM was improved from 102 to 105 with the N-GST buffer layer. The switching uniformity also improved compared to that of a non-buffer layer device. The improved properties that were realized are attributed to the effects of buffer layer such as controlled Cu-ion injection, internal resistor, and Joule heating confinement during the reset process. Furthermore, to verify the effect of nitrogen on the switching properties, we compared the GST and N-GST buffer layers. We believe that doped nitrogen helped to control Cu-ion injection into the resistive switching layer, and to confine the joule heating during the reset process, enabling a high on/off ratio and improved switching uniformity.

In this paper, the effect of the titanium oxide-based tunnel barrier on the non-linearity and switching uniformity of resistive random access memory has been investigated with the object of achieving excellent device non-linearity and reliability for cross-point array applications. To form the tunnel barrier of titanium oxide, its thickness was engineered using the deposition time. The tunnel barrier effectively controls the current flow in the devices with a tunneling mechanism that modifies the tunnel barrier thickness for non-linearity and switching reliability of devices. The tunnel barrier controls the current behavior of the device because most of the bias is applied to the tunnel barrier owing to its dominant resistance state. In addition, the tunnel barrier can exhibit uniform resistive switching during the set operation with the controlled current flow.