• <tr id="yyy80"></tr>
  • <sup id="yyy80"></sup>
  • <tfoot id="yyy80"><noscript id="yyy80"></noscript></tfoot>
  • 99热精品在线国产_美女午夜性视频免费_国产精品国产高清国产av_av欧美777_自拍偷自拍亚洲精品老妇_亚洲熟女精品中文字幕_www日本黄色视频网_国产精品野战在线观看 ?

    Flexible Memristive Devices Based on Graphene Quantum-Dot Nanocomposites

    2022-08-24 07:01:08SungWonHwangandDaeKiHong
    Computers Materials&Continua 2022年8期

    Sung Won Hwang and Dae-Ki Hong

    Department of System Semiconductor Engineering,SangMyung University,Cheonan-si,31066,Korea

    Abstract: Artificial neural networks (ANNs) are attracting attention for their high performance in various fields,because increasing the network size improves its functioning.Since large-scale neural networks are difficult to implement on custom hardware,a two-dimensional(2D)structure is applied to an ANN in the form of a crossbar.We demonstrate a synapse crossbar device from recent research by applying a memristive system to neuromorphic chips.The system is designed using two-dimensional structures,graphene quantum dots(GQDs)and graphene oxide(GO).Raman spectrum analysis results indicate a D-band of 1421 cm-1 that occurs in the disorder;band is expressed as an atomic characteristic of carbon in the sp2 hybridized structure.There is also a G-band of 1518 cm-1 that corresponds to the graphite structure.The G bands measured for RGO-GQDs present significant GQD edge-dependent shifts with position.To avoid an abruptly-formed conduction path,effect of barrier layer on graphene/ITO interface was investigated.We confirmed the variation in the nanostructure in the RGO-GQD layers by analyzing them using HR-TEM.After applying a negative bias to the electrode,a crystalline RGO-GQD region formed,which a conductive path.Especially,a synaptic array for a neuromorphic chip with GQDs applied was demonstrated using a crossbar array.

    Keywords: Memristive devices;neuromorphic chip;resistive RAM;quantum dot;graphene

    1 Introduction

    Recently,the von Neumann method of implementing the existing serial processing structure has been found to have performance limitations in memory and computing system processing due to the functionally limited memory access process and bottlenecks of processed information.To overcome these performance limitations,over the past few years,artificial neural networks(ANNs)have made significant progress in many fields,such as image processing [1-5]and object recognition [6-13].However,the performance of ANNs is significantly new heterogeneous parallel device structures that enable computational performance can be alternatives.This area of research involves the development of flexible devices,which enable many exciting applications,from big data processing capacity to image analysis;deep neural networks (DNNs) have also been tried in various structures recently[14,15].However,the complex circuit structure and large power consumption needed to increase the performance limits device use in various embedded-type applications[16,17].The use of a memristive device is widely regarded as a hardware-implemented system solution,using an internal memory circuit in a deep neural network based on a parallel architecture.

    As such,the implementation of a new heterogeneous structure-based device can completely avoid unnecessary noise and leakage current in the circuit[18-20].In particular,the advent of new memory technologies,such as resistive technology,which have emerged as real alternatives,make it possible to maximize the performance of next-generation random access memory(RRAM)structures[21,22].Until recently,breakthrough technologies were realized through innovative structures providing smallarea,high-speed,nonvolatile memory circuits applicable to Complementary Metal Oxide Semiconductor (CMOS) network systems [23].However,Parallel heterogeneous neuromorphic engineering applies CMOS-based digital-analog Very Large Scale Integration(VLSI)to realize high-performance and flexible ANNs[24].For resistive memory,various geometries of devices have been used to engineer memristive devices and integrate them into crossbar structures[25].Therefore,to take full advantage of the parallel structure,a continuous hardware configuration is essential to integrate crossbar array integrated memory devices with CMOS technology.

    In particular,memristors have advantages as next-generation storage devices due to their high power density and low power consumption,and their flexible structures[26].A passive two-terminal memristor is a primary electronic device with a built-in synaptic nonvolatile memory function in an array-structure crossbar.Various studies have reported on the design of a heterostructure of the active layer that controls the spatial resolution and resistive memory properties of crossbar ion channel electrodes [27].A memristor is a passive two-terminal structure element composed of a metal layer that supplies carriers and a dielectric layer where ion channels form.The physical mechanism is important in forming conductive filaments (CF),which serve as ion paths at the interface between the metal contact and the active layer[28].In the case of a formation of a CF between electrodes when oxygen vacancy occurs in the metal oxide layer by an externally applied voltage,a filament forms by penetration of the electrode metal material into the oxide film due to electrical stress [29].The CFs formed in this way repeat the generation and rupture of the filament path according to the magnitude or direction of the voltage,and the resistance value changes.

    Device has a memory characteristic of having high resistance (OFF state) at a low voltage and a low resistance (On state) at a specific voltage.Rapid changes in resistive memory occur depending on the structure used in the device and the way the conductive channels are implemented.After CF formation via voltage application the memristor unit can perform a reversible transition between the high-resistance state(HRS)and the low-resistance state(LRS).The reversible conversion operation from HRS to LRS is controlled through a set process.On the other hand,the reversible conversion operation from LRS to HRS is controlled through a reset process.The crossbar arraybased neuron structure implements the calculation function of the weight value,which is the main parameter of the synaptic device[30].Therefore,considering these fundamental physical performance limitations,because it is no longer possible to narrow the desired level gap compared to the latest technology via device expansion alone,a crossbar array synaptic device with a new structure can be an alternative[31].

    The physical properties,such as the van der Waals force of low-dimensional materials,are important in constructing a resistive memory with a new structure [32].Among low-dimensional structures,we focused on graphene quantum dots(GQDs)and reduced graphene oxide(RGO),which can control quantum effects and effective mass,and have advantages not only as novel active layers but also as flexible electrical device applications[33].GQDs are two-dimensional planar nanoflakes with thicknesses of a few atomic layer;they are defect-free and have dominant size and edge properties[34].GQDs can be applied as active layers in crossbar structures,enabling the implementation of next-generation RRAM.GQD-based CF formation characteristics are consistent,and reproducible performance is possible.

    The Strukov’s research team at the University of California Santa Barbara developed an ANNs using a 12×12 array of resistive memristors [35].As the iteration of the algorithm increased,the number of misclassified patterns tended to approach zero.Although ANNs using memristors have shown some performance in synaptic operation,problems with neuromorphic circuits are appearing due to high error rates.This is due to the unstable structure of the filament path and structural defects,and the low on/off conversion rate.Recently,artificial synaptic structures designed with inorganic and organic materials showed different results by applying atomic vacancies and interfacial trap principles based on metal and oxide layers.In this work,we demonstrate a parallel-structured memristive crossbar array based on GQDs and GO,without the need for additional passive devices.

    2 Experiment

    2.1 Preparation of Graphene Quantum Dots

    Graphene was synthesized on Cu foil in a thermal furnace at 950°C for 7 min under Ar:H2:CH4mixed-gas flow.Subsequently,a pre-prepared poly (methyl methacrylate) (PMMA,Sigma Aldrich)solution was dropped all over the graphene surface,spincoated at 5000 rpm for 45 s,and then heat treated on a hot plate at 160°C for 3 min.The PMMA solution was prepared by adding 0.5 g PMMA powder to 15 mL chlorobenzene(Sigma Aldrich)and stirring the mixture at 65°C for 8 h.The Cu was removed by dipping the PMMA/graphene/Cu foil stack in 1 M FeCl3solution(Sigma Aldrich).The PMMA/graphene stack was then transferred onto glass or PET substrate and heat-treated at 110°C for 3 h.Finally,the PMMA was dissolved by acetone.The following processes were used to fabricate GQDs.Graphene oxide(GO)was prepared from graphite powder using the Hummers method.GO was deoxidized in a horizontal furnace at 350-450°C for four hours under Ar to produce reduced graphene oxide powder.About 5.0-6.5 g of graphene oxide powder was oxidized by ultrasonication in concentrated 15 mL H2SO4and 40 mL HNO3for 20 h.In addition,350 mL deionized water was used to dilute the mixture.After cooling to room temperature,the resulting powder was redispersed in 35-mL DI water for two hours under ultrasonication.Then,by filtering the resulting suspension through a 200-nm nanoporous membrane,a brown solution was separated.Since the colloidal solution still contained some large graphene nanoparticles (<200 nm) emitting weak blue fluorescence,it was further filtered overnight in a dialysis bag with a cutoff of 3500 Da molecular weight,thereby producing strongly fluorescent GQDs.The GQDs were separated into different sizes using several dialysis bags of 1000-50,000 Da and a 20-nm nanoporous membrane[36].

    2.2 Device Fabrication

    To fabricate RRAM-based synapse devices,after the direct synthesis of the extracted GQD and RGO,graphene quantum dots of each shape were directly grown on patterned transparent electrodes on a Cu thin films through a modified CVD method.After washing the composite GQD layer in a deionized water cleaning process,heat treatment was performed.The solution used to establish the dielectric structure of RGO was prepared by sonication of 2.5-7.5 wt%RGO in 2-propanol for 60 min to prepare five samples with different RGO concentrations.The solution was spin coated at 5500 rpm for 35 s to obtain a ~55 nm thick RGO layer.A cross-bar type Au/RGO GQD/Au/graphene structure was deposited on an ITO substrate by thermal evaporation using a 150 μm wide overlapping hard mask.The coated substrate was heat-treated at 80°C for 10 min to finally obtain the RGO/GQDs layer,which was then cooled to room temperature.Finally,Au metal was thermally evaporated on the RGO/GQDs layer to form the contact.The active areas of all the devices were defined using a shadow mask of 15 mm2.

    2.3 Characterization

    The surface roughness of the graphene and RGO layers was measured using an atomic force microscope in non-contact mode (AFM,Park Systems,model XE-100).Plane-and cross-sectionalview images of the materials and devices were obtained by field emission scanning electron microscopy(FE-SEM,Carl Zeiss,model LEO SUPRA 55).Morphologies of GQDs were analyzed using an HRTEM(High-resolution Transmission Microscopy)(FEI Tecnai F30 S-Twin).To make the HRTEM specimens,the GQDs were dispersed in DI water,drops of which were then put on C-or SiO-coated Cu grids(Tedpella,Inc.,Redding,CA,USA)and mica substrate,respectively.The current-voltage(I-V)curve of the device was analyzed using an electrical and electronic workstation(Keithley 2400).Current density-voltage (J-V) characteristics were monitored by forward/reverse scans at 200 ms and 10 mV under standard conditions.Bending test of the flexible devices was conducted at a 0.5 Hz frequency.

    3 Results and Discussion

    3.1 RRAM-Based Synapse Devices for Neuromorphic Systems

    The characteristic of increasing the amount of current according to the frequency with which bias is applied to the memory resistor electrode is similar to that of the synapse that controls the amount of current according to the spike stimulus[37].From the background of the neuromorphic crossbar system,the computational performance of the synaptic capability is capable of parallel structure-based signal processing [38].Responses to increases in activity can be mimicked by the proposed resistive memory by reinforcement and depression of the conductance value by proper bias pulses [39].To show that the RGO-GQDs nonvolatile(NV)random-access devices provided consistent performance as an artificial neural system capable of implementing spike signal output improvement,a 20 ns pulse signal designed to increase in amplitude was used to modulate the resistance change of the RRAM,memristor with properties similar to those of synapses were incorporated into the integrated circuits.

    Since the resistance-variable memristor has a structure of two electrodes and an oxide film located between them,it has high resistance (OFF state),and has a memory characteristic having low resistance (On state) from a specific voltage.The rapid change in resistance is related to the microstructure used in the device,and typically appears in two ways.First,with an externally applied potential difference,oxygen vacancies form at the interlayer with a defect between the terminal and the oxide layer.A conductive filament forms through both electrodes[40].In the second case,the electrode metal material penetrates the oxide layer via electrical stress to form a filament [41].The channels formed in this way repeat the generation and rupture of the filament according to the magnitude or direction of the potential difference and the resistance changes.In principle,the memristor can persistently store an amount of current that has previously passed by controlling the amount of current and maintaining the charge transfer path between the two even if no voltage is applied in the middle[42].

    A 10 nm thick RGO-GQD layer was designed on a Gr/ITO/PI substrate by spray coating method;its structure,confirmed by the high-resolution transmission electron microscopy(HR-TEM),is shown in the image Fig.1.This layer consists of empty space between GQDs.Defects are negatively biased by oxygen ion diffusion and accumulation in the composite RGO structure,which is associated with the formation of defects.A CF of an active layer based on the resistive memory structure with Au electrode was applied to prevent the destruction of an electric current of 15 μA and prevent the permanent destruction of the electrode.The TE/RGO-GQDs/BE layers indicate stable resistive switching behavior without defect-related bubble formation.Therefore,by applying a negative voltage through the electrode,the electrical resistance increases rapidly at the positive minimum turn-on voltage.

    Figure 1:Schematic of synaptic transmission in neural system and illustration of RRAM-based bionic neural cell

    The structure of a convolutional neural network with excellent performance in image classification will be briefly reviewed[43].In addition,a neuromorphic architecture that mimics the human brain and a compression technique that can reduce the number of weight parameters of the neural network are examined.The convolution layer extracts a feature map by conducting convolution calculation of the input image.The convolution operation involves constant size and weights.The kernel with the previously determined parameters extracts features by multiplying and adding parameters corresponding to pixel values while moving the input image[44].

    After going through the convolution and weight layers,the stably connected process is performed.A fully connected layer is a structure in which each input node is associated with the complete nodes of the next function.The fully connected layer receives as input features extracted from the convolutional layer and classifies them into objects.After passing through the stably placed layer,the objects are finally output as probability values.A convolutional layer has several parameters but a large amount of computation,and a connection layer has several parameters,but a large amount of computation.

    Atomic-scale RGO-GQD layers were designed on a Gr/ITO/PI substrate;these are shown in a schematic of the synaptic transmission in the conductive path,in Fig.2.RGO-GQD layers structured with several oxygen-functional groups are formed between the top and bottom contacts.Au ions in the bottom electrode penetrate to the Gr of the crystal layer at an initial step of the set process,and these terminal ions create conductive Au/Gr interface defects.This forming of locally conductive paths on the RGO composite layer is a significant phenomenon.

    3.2 Memristive Devices Based on RGO Nano-sheet Nanocomposites

    Device was fabricated on a single-layer Gr on an ITO substrate through chemical vapor deposition(CVD);device formation is confirmed in the schematic of the synaptic crossbar in the conductive layers with Gr,as shown in Fig.3.To prepare the graphene layer for the RGO-GQD RRAM component,through the hard mask,a part of the RGO-GQD layer,and the substrate under the Au layer,were etched down to ~10 nm,completing the 3D device structure.Then,a GQD layer was formed on the RGO layer by spray coating,and 15 nm of Au was deposited by electron beam deposition to form an electrode for carrier supply.Subsequently,the configuration of the RRAM device with the RGO-GQD active layer was coated with several layers of GQD pieces,on the top and on the memory switching layers.Then,an additional array layer was completed by sputtering of a metal layer on top of the RGO-GQDs by e-beam deposition and lift-off process.Before evaluating the electrical properties,we improved the electrical contact properties.We removed the residual contaminant particles introduced during RGO-GQD coating by annealing the fabricated device at 110°C for 4 h in an H2carrier gas atmosphere.

    Figure 2:Pattern recognition with neural network using multilayer perceptron algorithm

    Figure 3:Fabrication process flow and schematic device architecture based on Au/RGOGQDs/Au/graphene/ITO/PI

    The stress generated during device fabrication was relieved through the annealing process,and contaminants and moisture were removed to improve the interlayer adhesion in the structure.To electrically connect a resistive memory device having a flexible structure to an electric circuit,a wire bonding process was performed on an electrode supplying a carrier to the active layer.

    Chemically induced GO has a unique set of properties resulting from oxygen functional groups introduced via chemical exfoliation of graphite.The thickness of GQDs is 0.34 nm,as measured by atomic force microscopy(AFM).GQDs with sizes of 5 to 10 nm consist of a single layer and exhibited a bilayer structure in localized areas.GQDs were also measured and found to have average height,as shown in Fig.4a;they were imaged using AFM in contact mode to determine the number of graphene layers.The GQD particles were measured and found to have a height distribution between 0.5 and 3.5 nm on the RGO substrate.The highest part of the GQDs was measured at 1.3 nm.This corresponds to a single graphene layer there were uniform properties inside the RGO layers synthesized by other methods.Raman spectrum analysis indicated a D-band of 1357 cm-1that occurs in areas of disorder,expressed according to the atomic characteristic of carbon in the sp2hybridized structure,and a Gband of 1596 cm-1that corresponds to the graphite structure.

    Figure 4:Characterization of chemical state in GQD layers in pristine Raman intensity ratio maps of 2D bands.(b)Raman profiles of black,red,green and blue boxes in(a)

    The centers of widely distributed bands at 2690 and 2760 cm-1correspond to the 2D and D+G bands,respectively,as shown in Fig.4b.The ratio of D-to-G band intensity changed according to the composition of the RGO-GQD region due to the diffusion of oxygen ions;analysis of the relative intensities of the D bands showed that hexagonal networks of defective carbon atoms were reduced because interatomic stresses were generated as carbon atoms recombine with oxygen-containing groups.It was confirmed that a strong electric field generated conduction channels locally in the composite RGO-GQD region.

    Fig.5 shows Raman spectra of GQDs to indicate chemical states in the GQD layers,the ON and the OFF states,and RGO-GQD/Au/Gr/ITO substrate.The Raman spectrum of the GQDs resolves into two distinctive D and G bands at ~1421 and ~1518 cm-1,respectively.The D and G bands hardly move even when the ON and OFF states,which are the devices electrical states,proving that the basic optical properties were not affected.In contrast,the G bands measured for the RGO-GQDs showed significant GQD edge-dependent shifts with position.According to this relationship,the average size of GQDs was 5.8~8.9 nm,which was consistent with the measurement results within the range of 6-9 nm obtained by HR-TEM.

    Fig.6a shows an HR-TEM image of the ON state after applying a negative bias to the upper electrode on the active layer.The non-uniform interface state was confirmed in the RGOGQD/Au/graphene/ITO structure.Although the GQD region in RGO has a steadier and more densely stacked path than that of the off state,oxygen vacancies in the metal atomic layer increased at the interface,which means that defects in the interfacial region in RGO-GQD/Au/graphene increased.As shown in the TEM image,in the RGO-GQD and graphene layers,locally coupled atomic layer regions were located at the heterogeneous interface,and it can be judged that the RGO-GQD layers affected the charge transport path formation in the ON state,as shown in the red color box.An irregular interfacial region appeared in another localized area,and the boundary between the GQD conduction channel crystallization region and the RGO region could be identified.

    Figure 5:Characterization of chemical state in GQD layers in pristine,ON and OFF states.Raman intensity ratio maps of D and G bands in (a) ON state and (c) OFF state.(b) and (d) are Raman profiles of blue and yellow boxes in(a),(c)

    Figure 6:TEM image of RGO-GQD/Au layer.Scale bar indicates 20 nm.Inset of right panel:highresolution TEM image showing GQDs.Scale bar,5 nm

    The composite RGO-GQD region has more regular and densely stacked GQDs than does the conventional GO thin film,which means that the crystallinity of the composite RGO-GQD region is increased.To avoid an abruptly formed conduction path,the effect of a barrier layer on the graphene/ITO interface was investigated.As can be seen in the HR-TEM images,the diameters of the GQDs at different locations in the RGO were uniformly distributed in the range of 3.6-7.5 nm,as shown in the green color box.In the ON state,it can be inferred that the GQD layer in the RGO is essential to forming a conductive charge transport layer.An irregular stack structure appeared in the localized region,and the boundary between the crystalline region and the randomly stacked region could be distinguished.Because typical thin GO layers have non-uniform thickness and roughness,creating a local electric field is not easy.

    Figs.7a and 7b is an HR-TEM image of the RGO-GQD/graphene/ITO layer in the ON and OFF states.The insets in the figures are HR-TEM images indicated in the red and blue rectangular regions of(c)and(d),respectively.The observation is that GQD is organized into an RGO layer as nanoparticles that uniformly incorporate the interface of graphene in the ON state.It can be inferred that the atomic crystallinity of the GQD layer within the RGO layer increased.In contrast,the GQDs are activated into the RGO layer as an aggregation that non-uniformly penetrates the ITO substrate interface in the OFF state.We investigated the variation in the nanostructure of the RGO-GQD layers by HR-TEM analysis.After applying a negative bias to the electrode,a crystalline RGO-GQD region was found,which is related to a conductive path.This decreased the number of hexagonal carbon-oxygen bonds in the RGO layer and caused a relative increase in the number of carbon-carbon orbital bonds in the GQD layer.

    Figure 7:Cross-sectional TEM images of(a)conductive channel region in ON state and(b)OFF state;(c,d)corresponding cross-sectional HR-TEM images obtained in red and blue rectangular regions of(a,b),respectively

    Furthermore,structural and chemical variation in the RGO-GQD layers can be proved by the dspacing of the GQD clusters,as seen in the yellow box;these clusters have a considerable distance between them,caused by the desorption of RGO oxygen ions,as shown in Fig.7c.However,the RGO layers in the yellow box are a relatively small distance apart due to the vast number of oxygenfunctional groups without defects,as shown in Fig.7d.The d-spacing results clearly prove the variation in RGO-GQD layers under an electric field.We found that conductive GQD pathways were created by movement of oxygen ions at interfaces in a strong electric field.The electric field is localized to a specific area of the GQDs because the thickness and roughness of the RGO layer are uniform,and the size and edge effects of the GQDs dominate.

    During the ON state,oxygen and graphene layers play important roles in forming conductive graphene pathways.By applying a negative bias to the top electrode,the reproducibility characteristics of the resistive memory were confirmed to be related to the nanostructure.As the resistive properties were maintained during LRS,the nanostructures in the RGO-GQD devices appeared to be uniform,suggesting a correlation with the irreversible resistive switching behavioral properties.Observing the interface between the electrode and the low-dimensional structure reveals the presence of a very thin layer associated with defects.In the active layer heterostructure to which GQD is applied,surface recombination and diffusion of Au ions occur due to rapid energy transfer at the interface.The GQD composite layer resolves the energy band mismatch between this heterostructure and the grains generated from the heterogeneous interface structure.Therefore,carrier movement occurs without clogging from Au adatom to GQD when the graphene layer is applied to the RGO active layer.This confirms the possibility of effective filament channels for charge recombination and transport in RGObased resistive memories.

    Fig.8a shows the DC switching results measured over 30 loops,used to analyze the resistive memory characteristics of the flexible RGO-GQD structure.As can be seen in the figure,the butterfly shape during DC switching hardly changed after cycling,indicating that the designed RGO-GQD structure-based device did not degrade during the fabrication process.The synaptic crossbar array device was set up with a single 15 ns wide pulse (2.3 V high),and the active stay resistance was reduced by a factor of 10 during DC switching.These electrical properties allowed the RGO-GQD structure-based devices to exist in a RESET range of<10 ns per bit for multi-bit implementation of the synaptic crossbar.The LRS and HRS show uniform distributions,and the HRS shows a slightly wider distribution than LRS at the selected bias.

    Figure 8:(a) I-V curves of Au/RGO-GQDs/Au/graphene/ITO/PI memory devices plotted on semilogarithmic scale.(b)I-V curves of Au/RGO-GQDs/Au/graphene/ITO/PI memory devices plotted on semi-logarithmic scale.The inset is an optical microscopy image of RRAM crossbar device(scale bar:200 μm)

    The strong electric field across the electrode and the RGO-GQD layer at a positive bias voltage induced the oxygen ions in the negatively charged RGO to drift towards the top electrode interface.As a result,the height of the Schottky barrier between the electrode and the heterostructure interface decreased,and carrier diffusion through the tunneling barrier increased due to the reduction of defects in the RGO-GQD layer.This mechanism provides advantages compared to crossbar synapse devices.Through the electric field,the conductive RGO-GQD layer enabled uniform resistive memory value control through the structural transport of ions and the switch characteristics observed in flexible RRAM devices with well-defined interfacial defects and GQDs.

    Fig.8b shows the SET/RESET I-V curves of 15 devices,in which ten have 5 nm size GQDs and the rest have 10 nm size GQDs.The dramatic change of the ON/OFF state current was confirmed by looking at the RGO switching layer applied with 10 nm particle size and comparing it to the RGO-GQD layer coated to 5 nm particle size.For the SET/RESET voltages,it was observed that the electrical properties of the RGO active layer RRAM synaptic crossbar to which GQDs were applied by size increased to 2.4/3.3 V for SET/RESET.The ON/OFF current ratio is reduced to 103 of that of the discrete RRAM crossbar with GQDs.This is because the mobile ions penetrating the RGOGQDs conductive path were transferred to the active area by the depression of the voltage,which had an amplitude of 2.9 V.Here,the various pulse frequencies and pulse widths were changed at 10 Hz and 20 ms,respectively.The thickness of the RGO-GQDs layer was set in a range of 10-25 nm by varying the concentration of GQDs,whereas the active region was fixed at 25×25 μm2.A compliance current of 5 mA was applied to protect the device.The initial switching occurred at Vf=2.31 V for 1.5 wt%GQSs,and Vfchanged from 2.12 to 2.85 V when increasing from 1.0 to 2.5 wt%of the GQDs.When the sweep voltage was changed to 5 V,there was no resistive switching characteristic at 5.0 wt%of GQDs.In the next switching stage,at 1.5 wt%of GQDs,the HRS was maintained when bias was applied below 1.2 V,when bias was applied at over 2.84 V,the current rapidly increased.As a result,the SET/RESET I-V loops characteristic curve under a depression pulse of 2.5 V exhibited the best symmetricity.Therefore,we optimized the SET/RESET I-V characteristics of the designed synapse array device by adjusting the carrier penetration behavior through variation of the active path thickness and region of the crossbar interface.

    Fig.9a shows the electrical characteristics measured under various temperature conditions in the range of 150~300 K.The High Resistance State (HRS) and Low Resistance State (LRS) show the change in resistance.The carriers penetrated the composite RGO-GQD layers under an electric field,and conductive metal filaments composed of GQD ions were produced.As the temperature increases,the resistance decreases in the HRS and LRS,and the conduction path of the composite RGO-GQDs improves.According to the temperature,the presence of GQDs,and the composite RGO high resistance stemming from linked oxygen functional groups that can be dissociated,the resistance change in the state is more significant than the resistance in the LRS.Based on these experimental results,the high-temperature region was found to be consistent with the Arrhenius model.Defects of the interface in the RGO-GQDs,and the ejected carriers,support the electrical conduction of the composite RGO structure.Furthermore,the flexible RGO-GQD device showed that SET/RESET cycling endurance could be greater than 200.To predict how fast the RRAM device can operate,SET and RESET operations were performed with uniform pulses with width ranging from 20 μs to 15 ns,which is the resolution of the pulse generator.LRS is caused by ion diffusion in the top electrode.It can be inferred that,related to the composite RGO region with metallic behavior,the formation of a conductive path(CP)occurs.

    Figure 9:(a) Temperature dependence of resistance at HRS and LRS.(b) Conductive AFM images of RGO-GQDs/Au/graphene/ITO/PI at LRS and HRS after removing Au top electrode;scale bar indicates 0.3 μm

    The structure of the CP was analyzed using conductive AFM,with results shown in Fig.9b.After removing the top Au electrode at a voltage of 0.5 V,conductivity of state in the measured HRS and LRS.The AFM image shows a relatively high current,indicating that a negative bias is applied to the upper electrode.It can be seen that a local conduction path is formed.For CMOS device implementation,because process flexibility of the array design is required,a system of resistive memory crossbars can incorporate electrically characterized memristors.We aimed to develop a monolithically integrated array capable of obtaining stable resistive memory performance in RGO-GQD flexible devices.

    To improve performance and portability,and to reduce size,electronic devices are evolving to become more flexible,such as including low-power devices,batteries,and sensors to improve performance and portability and reduce size.Flexible electronics bend when,as rolling and torsion occur,as such,mechanical reliability during cyclic deformation is an important characteristic.We investigated the bending stress of RGO-GQDs layers on flexible polyimide substrates with different thicknesses using single-layer and bi-layer bending models and monitored the electrical resistance of the devices with an RGO-GQDs during a bending fatigue test.Fig.10a shows the variation of the crossbar synapse array with an RGO-GQDs by under repeated bending of up to 500 times iterations for R=5 mm.Even after the bending test,more than 95% of the resistive switching value is was maintained.

    Figure 10:(a) Schematic representation and the insets describing the bending tests,and show a real image of a typical memristive sample.(b)Changes of normalized ΔR/R0(%)as functions of bending cycles at R=5 mm for devices under outer bending

    To clarify the cause of the device degradation,we also confirmed the evolution of normalizedΔR/R0 (%) changes as functions of bending cycles in the same way.After 100 cycles,the relativeΔR/R0(%)increased by 6%,as shown in Fig.10b,which the cracking of the interface can interpret shows this effect.Therefore,an exact design of the to determine applied mechanical strain applied during bending is necessary to model highly reliable flexible electronics.These results suggest that the defect-free RGO-GQDs are very promising as a flexible memristive devices.Compared to these devices,our synapse array with a GQDs showed a low cycle-to-cycle variation of less than 3%and a high recognition rate of 95.5%.From the perspective,in terms of cycle-to-cycle variation,we can say that the proposed device is acceptable for implementing an artificial neural network.

    4 Conclusion

    This work successfully implemented a reproducible flexible crossbar synaptic array based on a low-dimensional structure with GQD-RGO layers.This two-dimensional synapse array was achieved by adopting a crossbar synapse with a GQD structure.The mobile carriers in the RGO penetrated the interface of a CF channel located at the crossbar between the top and bottom electrode,which resulted in resistive memory structures with excellent conductance due to GQD ion pathways.The crossbar synaptic device showed reliable long-term plastic behavior because predictable currents did not return to their initial values.Raman spectrum analysis indicate a D-band of 1421 cm-1that occurs due to disorder,expressed as an atomic characteristic of carbon in the sp2hybridized structure,and a G-band of 1518 cm-1that corresponds to the graphite structure.The intensity ratio I(D)/I(G)of the D peak to the G peak is about 0.96,and is related to the size and edge structure of the GQDs coated on the RGO.In the ON state,it can be inferred that the GQD layer in the RGO is an essential variable in forming a conductive charge transport layer.We found that conductive GQD pathways were created by the movement of oxygen ions at the interface in a strong electric field.The electric field is localized to a specific area of the GQD because the thickness and roughness of the RGO layer are uniform,and the size and edge effects of the GQDs dominate.These results will impact emerging VLSI design and lead to the optimization of ANNs.

    Funding Statement:The authors received no specific funding for this study.

    Conflicts of Interest:The authors declare that they have no conflicts of interest to report regarding the present study.

    精品国产三级普通话版| 中文字幕av在线有码专区| 亚洲精品第二区| 亚洲va在线va天堂va国产| av在线播放精品| 国产精品三级大全| 亚洲伊人久久精品综合| 深夜a级毛片| 少妇高潮的动态图| 国产精品国产三级国产专区5o| 欧美激情在线99| 色综合站精品国产| 黑人高潮一二区| 国内少妇人妻偷人精品xxx网站| 成年免费大片在线观看| 亚洲国产最新在线播放| 亚洲av一区综合| 激情五月婷婷亚洲| 人人妻人人澡欧美一区二区| 精品久久久噜噜| 三级国产精品片| 亚洲精品久久午夜乱码| 男女边吃奶边做爰视频| 亚洲自偷自拍三级| 22中文网久久字幕| 国产精品综合久久久久久久免费| 岛国毛片在线播放| 精品久久久久久成人av| 久久99热6这里只有精品| 最近2019中文字幕mv第一页| av女优亚洲男人天堂| 国产成人a∨麻豆精品| 亚洲欧美成人综合另类久久久| 我的老师免费观看完整版| 不卡视频在线观看欧美| 国产伦精品一区二区三区四那| 极品教师在线视频| 国产久久久一区二区三区| 久久精品久久久久久久性| 日日摸夜夜添夜夜爱| 麻豆国产97在线/欧美| 久久这里只有精品中国| 国内精品美女久久久久久| 欧美激情国产日韩精品一区| av在线老鸭窝| 欧美成人精品欧美一级黄| 国产淫片久久久久久久久| 欧美性感艳星| 精品久久久久久电影网| 亚洲精品乱码久久久久久按摩| 亚洲在久久综合| 成人综合一区亚洲| 亚洲国产精品国产精品| 男人狂女人下面高潮的视频| 午夜老司机福利剧场| 在线观看美女被高潮喷水网站| 国产亚洲91精品色在线| 亚洲国产精品sss在线观看| 身体一侧抽搐| 国产高清不卡午夜福利| 不卡视频在线观看欧美| 亚洲欧美成人精品一区二区| 老女人水多毛片| 蜜桃久久精品国产亚洲av| 人妻系列 视频| 男人和女人高潮做爰伦理| 国产高清有码在线观看视频| 搡老乐熟女国产| 精品久久久久久电影网| 亚洲乱码一区二区免费版| 永久免费av网站大全| 天堂中文最新版在线下载 | 久久久久久久午夜电影| 精品一区二区免费观看| 2021天堂中文幕一二区在线观| 国产一级毛片七仙女欲春2| 国产精品一区二区三区四区免费观看| 免费看不卡的av| 亚洲在线观看片| 精品人妻一区二区三区麻豆| 一级av片app| 色视频www国产| 最近的中文字幕免费完整| 能在线免费看毛片的网站| 久久99热这里只频精品6学生| 丝袜美腿在线中文| 久久久精品欧美日韩精品| 一本—道久久a久久精品蜜桃钙片 精品乱码久久久久久99久播 | 成年av动漫网址| 亚洲va在线va天堂va国产| 一边亲一边摸免费视频| 久久久久久伊人网av| 男女边摸边吃奶| 99re6热这里在线精品视频| 成年av动漫网址| 精品人妻熟女av久视频| 最近手机中文字幕大全| 国产免费视频播放在线视频 | 熟妇人妻不卡中文字幕| 国产精品国产三级国产专区5o| 国产人妻一区二区三区在| 国产色婷婷99| 内地一区二区视频在线| 亚洲成人一二三区av| 久久99热6这里只有精品| 国产精品久久久久久精品电影小说 | 在线观看免费高清a一片| 亚洲av电影在线观看一区二区三区 | 色网站视频免费| 久久久久久久久久久免费av| 欧美+日韩+精品| 岛国毛片在线播放| 国产成人一区二区在线| 成人午夜精彩视频在线观看| 99久久精品一区二区三区| 日韩欧美国产在线观看| 成人亚洲精品av一区二区| 国产一级毛片在线| 成年免费大片在线观看| 韩国高清视频一区二区三区| 麻豆乱淫一区二区| 成人二区视频| 非洲黑人性xxxx精品又粗又长| 一级av片app| 日本黄色片子视频| 国产成人aa在线观看| 亚洲欧美清纯卡通| 日本一二三区视频观看| 性色avwww在线观看| 欧美xxxx黑人xx丫x性爽| 一二三四中文在线观看免费高清| 日韩成人av中文字幕在线观看| 超碰av人人做人人爽久久| 日本免费a在线| 乱人视频在线观看| 久久久精品94久久精品| 色哟哟·www| 一级a做视频免费观看| 97在线视频观看| 97在线视频观看| 噜噜噜噜噜久久久久久91| 男女国产视频网站| 激情五月婷婷亚洲| 噜噜噜噜噜久久久久久91| 丰满乱子伦码专区| 成人漫画全彩无遮挡| 成人特级av手机在线观看| 男女国产视频网站| 男女国产视频网站| 亚洲精品国产成人久久av| 亚洲av电影在线观看一区二区三区 | 免费无遮挡裸体视频| 尾随美女入室| 欧美日本视频| 91狼人影院| av在线亚洲专区| 国内精品一区二区在线观看| 天天一区二区日本电影三级| 国产毛片a区久久久久| 成人亚洲精品一区在线观看 | 91午夜精品亚洲一区二区三区| 亚洲精品乱码久久久久久按摩| av播播在线观看一区| 亚洲精品成人久久久久久| 国产伦精品一区二区三区视频9| 只有这里有精品99| 国产成年人精品一区二区| 日韩强制内射视频| 男女视频在线观看网站免费| 男插女下体视频免费在线播放| 韩国av在线不卡| 欧美极品一区二区三区四区| 日韩一区二区视频免费看| 伊人久久精品亚洲午夜| 国产精品一区二区在线观看99 | 精品国产三级普通话版| av免费在线看不卡| 久久久精品欧美日韩精品| 免费观看无遮挡的男女| 毛片女人毛片| 人妻少妇偷人精品九色| 干丝袜人妻中文字幕| 九九爱精品视频在线观看| 国产免费又黄又爽又色| 韩国av在线不卡| 如何舔出高潮| 亚洲在线自拍视频| 欧美日本视频| 少妇高潮的动态图| 永久网站在线| 蜜臀久久99精品久久宅男| 日韩中字成人| 777米奇影视久久| 日本av手机在线免费观看| 午夜福利在线观看吧| 禁无遮挡网站| 一级毛片我不卡| 国产人妻一区二区三区在| 91精品国产九色| 国产综合懂色| 亚洲性久久影院| 欧美区成人在线视频| 欧美xxⅹ黑人| 国产精品无大码| 成人国产麻豆网| 精品午夜福利在线看| 亚洲av一区综合| 亚洲aⅴ乱码一区二区在线播放| 久久午夜福利片| 好男人在线观看高清免费视频| 久久久久性生活片| 狠狠精品人妻久久久久久综合| 高清毛片免费看| 午夜精品一区二区三区免费看| av在线亚洲专区| 国产精品.久久久| 身体一侧抽搐| 97人妻精品一区二区三区麻豆| 免费av毛片视频| 99九九线精品视频在线观看视频| 在线观看免费高清a一片| 熟女人妻精品中文字幕| 欧美 日韩 精品 国产| 成人综合一区亚洲| 一个人看视频在线观看www免费| 亚洲精品日韩在线中文字幕| 日韩制服骚丝袜av| 国产高清有码在线观看视频| 亚洲四区av| 色视频www国产| 大又大粗又爽又黄少妇毛片口| kizo精华| 欧美日韩综合久久久久久| 亚洲欧美精品专区久久| 精品国产三级普通话版| 国国产精品蜜臀av免费| 精品人妻熟女av久视频| 久久亚洲国产成人精品v| 亚洲精品视频女| 日韩强制内射视频| 成人二区视频| 国产免费又黄又爽又色| 久久久久久久久久久免费av| 日韩一本色道免费dvd| 国产精品av视频在线免费观看| 国产一区二区亚洲精品在线观看| 嫩草影院新地址| 国产精品嫩草影院av在线观看| 嘟嘟电影网在线观看| 久久久a久久爽久久v久久| 久久久久久久久久成人| 插逼视频在线观看| 久久这里只有精品中国| 最近视频中文字幕2019在线8| 一本一本综合久久| 高清毛片免费看| 国产色爽女视频免费观看| 麻豆成人av视频| 一区二区三区高清视频在线| 精品国产三级普通话版| 日韩强制内射视频| 在现免费观看毛片| 日日啪夜夜爽| 国产伦在线观看视频一区| 人妻夜夜爽99麻豆av| 黄片wwwwww| 男女国产视频网站| 国产精品熟女久久久久浪| 亚洲精品一区蜜桃| 亚洲成人av在线免费| 日日撸夜夜添| 成人无遮挡网站| 国产久久久一区二区三区| 国产黄片美女视频| 亚洲欧洲国产日韩| 一个人观看的视频www高清免费观看| 亚洲国产成人一精品久久久| 九草在线视频观看| 少妇被粗大猛烈的视频| 女人十人毛片免费观看3o分钟| 欧美日韩亚洲高清精品| www.av在线官网国产| 欧美bdsm另类| 日日摸夜夜添夜夜爱| 成年av动漫网址| 国产探花极品一区二区| 亚洲激情五月婷婷啪啪| 亚洲熟妇中文字幕五十中出| 亚洲综合精品二区| 特大巨黑吊av在线直播| 日韩成人伦理影院| 日韩精品青青久久久久久| 高清av免费在线| 人妻少妇偷人精品九色| 久久久精品免费免费高清| 国产午夜福利久久久久久| 久久99热这里只有精品18| 男女视频在线观看网站免费| 狠狠精品人妻久久久久久综合| 又黄又爽又刺激的免费视频.| 80岁老熟妇乱子伦牲交| 国产高潮美女av| 亚洲18禁久久av| 国产一区亚洲一区在线观看| 日日干狠狠操夜夜爽| 男人和女人高潮做爰伦理| 啦啦啦啦在线视频资源| 国产精品人妻久久久久久| 欧美精品一区二区大全| 日韩欧美三级三区| 国语对白做爰xxxⅹ性视频网站| 日本三级黄在线观看| 伦精品一区二区三区| 2021天堂中文幕一二区在线观| 淫秽高清视频在线观看| 深爱激情五月婷婷| 国产淫片久久久久久久久| 十八禁国产超污无遮挡网站| 国产av码专区亚洲av| 国产精品一及| 国语对白做爰xxxⅹ性视频网站| 久久精品国产鲁丝片午夜精品| 亚洲欧美一区二区三区黑人 | 免费黄频网站在线观看国产| 最近中文字幕2019免费版| 亚洲欧洲国产日韩| 国产精品久久视频播放| 又爽又黄a免费视频| 色综合色国产| 少妇被粗大猛烈的视频| 女人十人毛片免费观看3o分钟| 免费看日本二区| 天天一区二区日本电影三级| 亚洲欧美精品专区久久| 欧美性猛交╳xxx乱大交人| 午夜激情欧美在线| a级一级毛片免费在线观看| 日韩欧美 国产精品| 天堂√8在线中文| 老司机影院成人| 天堂网av新在线| 亚洲第一区二区三区不卡| 日日撸夜夜添| 最后的刺客免费高清国语| 亚洲图色成人| 欧美最新免费一区二区三区| 亚洲熟妇中文字幕五十中出| 亚洲精品第二区| 国产精品.久久久| 亚洲av成人av| 亚洲四区av| 亚洲精华国产精华液的使用体验| 美女大奶头视频| 日本一二三区视频观看| 尾随美女入室| 亚洲一级一片aⅴ在线观看| 51国产日韩欧美| 色综合色国产| 国产成年人精品一区二区| 97精品久久久久久久久久精品| 免费大片黄手机在线观看| 亚洲av成人精品一二三区| 日日摸夜夜添夜夜爱| 高清在线视频一区二区三区| 国产黄片美女视频| 亚洲美女视频黄频| 一个人观看的视频www高清免费观看| 亚洲精品乱码久久久v下载方式| 精品不卡国产一区二区三区| 2018国产大陆天天弄谢| 麻豆精品久久久久久蜜桃| 女人久久www免费人成看片| 一级毛片 在线播放| 亚洲成人久久爱视频| 日本熟妇午夜| 免费不卡的大黄色大毛片视频在线观看 | 美女xxoo啪啪120秒动态图| 中文字幕亚洲精品专区| 精品国产一区二区三区久久久樱花 | 亚州av有码| 久久99热6这里只有精品| 免费无遮挡裸体视频| 亚洲精品乱久久久久久| 午夜激情福利司机影院| 国产成人freesex在线| 日日撸夜夜添| 国产黄色视频一区二区在线观看| 国产精品一区二区三区四区免费观看| 国内揄拍国产精品人妻在线| 亚洲av免费高清在线观看| 真实男女啪啪啪动态图| 1000部很黄的大片| a级毛片免费高清观看在线播放| 欧美 日韩 精品 国产| 天天一区二区日本电影三级| 婷婷色综合大香蕉| 最后的刺客免费高清国语| 国产成人a区在线观看| 亚洲精品乱码久久久久久按摩| 国产精品精品国产色婷婷| 国产精品蜜桃在线观看| 国产精品一区二区在线观看99 | 欧美性猛交╳xxx乱大交人| 啦啦啦啦在线视频资源| 黄色一级大片看看| 热99在线观看视频| 欧美成人一区二区免费高清观看| 亚洲图色成人| 国产免费福利视频在线观看| 听说在线观看完整版免费高清| 韩国av在线不卡| 人妻一区二区av| 亚洲真实伦在线观看| 国国产精品蜜臀av免费| 亚洲成人一二三区av| 国产高清不卡午夜福利| 五月玫瑰六月丁香| 非洲黑人性xxxx精品又粗又长| 高清视频免费观看一区二区 | 国产成人aa在线观看| 91狼人影院| 在线观看一区二区三区| 欧美日韩国产mv在线观看视频 | 国产成人91sexporn| 校园人妻丝袜中文字幕| 乱系列少妇在线播放| 日韩成人伦理影院| 麻豆久久精品国产亚洲av| 777米奇影视久久| 日韩一区二区三区影片| 九草在线视频观看| 日韩欧美三级三区| 熟女人妻精品中文字幕| 免费av观看视频| 日韩伦理黄色片| 一本久久精品| 啦啦啦韩国在线观看视频| 亚洲av中文av极速乱| 超碰av人人做人人爽久久| 熟妇人妻不卡中文字幕| 肉色欧美久久久久久久蜜桃 | 亚洲最大成人手机在线| 天堂影院成人在线观看| 亚洲av日韩在线播放| 青青草视频在线视频观看| 中文字幕免费在线视频6| 国产av码专区亚洲av| 一级毛片电影观看| 欧美 日韩 精品 国产| 夜夜看夜夜爽夜夜摸| 男女国产视频网站| 精品99又大又爽又粗少妇毛片| 人妻夜夜爽99麻豆av| 国精品久久久久久国模美| 国产亚洲av片在线观看秒播厂 | 国产一区亚洲一区在线观看| 国产成人aa在线观看| 国产精品一区二区三区四区免费观看| 日韩欧美三级三区| 国产乱来视频区| h日本视频在线播放| 有码 亚洲区| 中文字幕免费在线视频6| 国产成人精品久久久久久| 舔av片在线| 精品99又大又爽又粗少妇毛片| 精品一区二区三区人妻视频| 成人特级av手机在线观看| 蜜桃久久精品国产亚洲av| 国产精品综合久久久久久久免费| 免费观看性生交大片5| 毛片一级片免费看久久久久| 日韩欧美一区视频在线观看 | 最新中文字幕久久久久| 午夜爱爱视频在线播放| 岛国毛片在线播放| 色哟哟·www| 亚洲精品一区蜜桃| 久久99精品国语久久久| 春色校园在线视频观看| 日本熟妇午夜| 亚洲欧洲日产国产| 高清在线视频一区二区三区| 国产精品国产三级专区第一集| 亚洲va在线va天堂va国产| 亚洲无线观看免费| 亚洲国产欧美在线一区| 亚洲av不卡在线观看| 亚洲精品aⅴ在线观看| 亚洲aⅴ乱码一区二区在线播放| 联通29元200g的流量卡| 成人国产麻豆网| 久久久久性生活片| 欧美三级亚洲精品| 两个人视频免费观看高清| 国产伦精品一区二区三区四那| 亚洲国产av新网站| 美女高潮的动态| 纵有疾风起免费观看全集完整版 | 亚洲一区高清亚洲精品| 国内精品美女久久久久久| 丝袜美腿在线中文| 中文资源天堂在线| 亚洲18禁久久av| 国产91av在线免费观看| 神马国产精品三级电影在线观看| 最近手机中文字幕大全| 人妻制服诱惑在线中文字幕| 欧美另类一区| 真实男女啪啪啪动态图| 看非洲黑人一级黄片| 韩国av在线不卡| ponron亚洲| 婷婷色综合www| 高清午夜精品一区二区三区| 欧美性猛交╳xxx乱大交人| 久久精品熟女亚洲av麻豆精品 | 一本久久精品| 欧美最新免费一区二区三区| 日本av手机在线免费观看| 国产在线男女| 国产国拍精品亚洲av在线观看| 老师上课跳d突然被开到最大视频| 亚洲成人中文字幕在线播放| 欧美xxxx性猛交bbbb| 纵有疾风起免费观看全集完整版 | 晚上一个人看的免费电影| 亚洲激情五月婷婷啪啪| 99久国产av精品国产电影| .国产精品久久| 欧美日韩视频高清一区二区三区二| 免费人成在线观看视频色| 美女国产视频在线观看| 狠狠精品人妻久久久久久综合| 国产av在哪里看| 国产精品一区www在线观看| 国产av码专区亚洲av| 最近中文字幕高清免费大全6| 久久久欧美国产精品| 又粗又硬又长又爽又黄的视频| 极品少妇高潮喷水抽搐| 日韩伦理黄色片| 国产精品久久久久久久久免| 国产av在哪里看| 能在线免费观看的黄片| 你懂的网址亚洲精品在线观看| 91精品伊人久久大香线蕉| 人妻制服诱惑在线中文字幕| 乱人视频在线观看| 中文字幕久久专区| 一本—道久久a久久精品蜜桃钙片 精品乱码久久久久久99久播 | 最近最新中文字幕免费大全7| 中文字幕亚洲精品专区| av播播在线观看一区| 少妇人妻精品综合一区二区| 成人毛片60女人毛片免费| 亚洲最大成人手机在线| 你懂的网址亚洲精品在线观看| 色播亚洲综合网| 日韩,欧美,国产一区二区三区| 欧美性猛交╳xxx乱大交人| 国产欧美日韩精品一区二区| 国产精品人妻久久久影院| 国产亚洲av片在线观看秒播厂 | 青青草视频在线视频观看| 午夜免费男女啪啪视频观看| 国产精品三级大全| 中文字幕久久专区| 亚洲国产欧美在线一区| 亚洲av成人精品一区久久| 黑人高潮一二区| 又黄又爽又刺激的免费视频.| 亚洲国产精品专区欧美| 熟妇人妻不卡中文字幕| 少妇被粗大猛烈的视频| 亚洲精品456在线播放app| av在线观看视频网站免费| 最近的中文字幕免费完整| ponron亚洲| 久99久视频精品免费| 国内少妇人妻偷人精品xxx网站| av女优亚洲男人天堂| 国产免费一级a男人的天堂| 亚洲国产色片| 亚洲乱码一区二区免费版| 亚洲精品视频女| 男人爽女人下面视频在线观看| 国产精品一区二区性色av| 午夜精品在线福利| 国产爱豆传媒在线观看| 欧美丝袜亚洲另类| 欧美日韩精品成人综合77777| 老女人水多毛片| 91久久精品国产一区二区三区| 精品酒店卫生间| 国产片特级美女逼逼视频| 国产一区二区三区av在线| 精品亚洲乱码少妇综合久久| 麻豆国产97在线/欧美| 一个人看视频在线观看www免费| 一个人免费在线观看电影| 99re6热这里在线精品视频| 久久99热这里只频精品6学生| 人妻制服诱惑在线中文字幕| 免费观看无遮挡的男女| 看十八女毛片水多多多| av卡一久久| 夜夜看夜夜爽夜夜摸| 精品欧美国产一区二区三| 三级国产精品欧美在线观看| 在线 av 中文字幕| 成人毛片a级毛片在线播放| 简卡轻食公司| 91午夜精品亚洲一区二区三区| 午夜激情福利司机影院|