Electrical Properties and Biological Synaptic Simulation of Ag/Mxene/Sio2/Pt RRAM Devices

electronics

Article Electrical Properties and Biological Synaptic Simulation of Ag/MXene/SiO2/Pt RRAM Devices

Xiaojuan Lian 1,2 , Xinyi Shen 1, Jinke Fu 1, Zhixuan Gao 1, Xiang Wan 1, Xiaoyan Liu 1, Ertao Hu 1 , Jianguang Xu 3 and Yi Tong 1,*

1 The Department of Microelectronics, Nanjing University of Posts and Telecommunications, Nanjing 210023, China; [email protected] (X.L.); [email protected] (X.S.); [email protected] (J.F.); [email protected] (Z.G.); [email protected] (X.W.); [email protected] (X.L.); [email protected] (E.H.) 2 National and Local Joint Engineering Laboratory of RF Integration and Micro-Assembly Technology, Nanjing 210023, China 3 The School of Materials Science and Engineering, Yancheng Institute of Technology, 211 East Jianjun Road, Yancheng 224051, China; [email protected] * Correspondence: [email protected]

Received: 9 November 2020; Accepted: 7 December 2020; Published: 9 December 2020

Abstract: Utilizing electronic devices to emulate biological synapses for the construction of artificial neural networks has provided a feasible research approach for the future development of artificial intelligence systems. Until now, different kinds of electronic devices have been proposed in the realization of biological synapse functions. However, the device stability and the power consumption are major challenges for future industrialization applications. Herein, an electronic synapse of MXene/SiO2 structure-based resistive random-access memory (RRAM) devices has been designed and fabricated by taking advantage of the desirable properties of SiO2 and 2D MXene material. The proposed RRAM devices, Ag/MXene/SiO2/Pt, exhibit the resistance switching characteristics where both the volatile and nonvolatile behaviors coexist in a single device. These intriguing features of the Ag/MXene/SiO2/Pt devices make them more applicable for emulating biological synaptic plasticity. Additionally, the conductive mechanisms of the Ag/MXene/SiO2/Pt RRAM devices have been discussed on the basis of our experimental results.

Keywords: RRAM devices; 2D MXene; resistance switching; volatile; nonvolatile; synaptic plasticity

1. Introduction Brain-inspired computing systems have been extensively investigated for their capability to break through the bottlenecks of dominant von Neumann computer architecture [1–8]. Different kinds of electronic synapses, the key components of brain-inspired systems, have been proposed to imitate biological synaptic functions, including transistors [9,10], phase change memory [11,12], ferroelectric devices [13,14], and resistive random-access memory (RRAM) devices [6–8], among others. Specifically, RRAM devices are considered one of the most competitive candidates as electronic synapses due to their low power consumption, high speed switching, multiple resistance states and so on [15–25]. However, the device reliability of the RRAM devices hinders the commercialization in future machine learning and neuromorphic computing and the power consumption requires further reduction. To break through these challenges, new materials and innovative structure is demanded. Two-dimensional (2D) material MXene, a family of transition metal carbides or nitrides, has attracted considerable attention in different research fields for their layered structure, high stacking density, ultra-high conductivity ( 6000–8000 S/cm), fast charge response and other ∼

Electronics 2020, 9, 2098; doi:10.3390/electronics9122098 www.mdpi.com/journal/electronics Electronics 2020, 9, 2098 2 of 10

outstanding performances [26–35]. Generally, MXene is in the form of Mn+1Xn where M is early translation metal and X is carbon or nitrogen. MXene can be synthesized by wet etching A from MAX compounds (A is Al or Si) due to the stronger connection of M-X bond than M-A bond [29,30]. Moreover, MXene can become narrow-band gap semiconductors or metal due to surface functionalities such as -O,Electronics -F, or 2020 -OH, 9, x FOR [31 PEER]. It REVIEW has been reported that some 2D materials such as Graphene,2 of 9 MoS2,

BN, and WSdensity,2 [17 ultra-high,36–39] areconduc abletivity to ( 6000–8000 ameliorate S/cm), the fast performances charge response of and RRAM other devices.outstanding Therefore, by combiningperformances the compatibility [26–35]. Generally, of SiOMXene2-based is in the RRAM form of M devicesn+1Xn where with M is the early dominant translation metal silicon CMOS fabricationand technology X is carbon or [ 40nitrogen.–42] andMXene those can be of synthesized a 2D MXene by wet etching material, A from we MAX designed compounds and (A fabricated is Al or Si) due to the stronger connection of M-X bond than M-A bond [29,30]. Moreover, MXene can Ag/MXene/SiO2/Pt structure-based RRAM devices. The introduction of 2D material MXene greatly become narrow-band gap semiconductors or metal due to surface functionalities such as -O, -F, or - lowers the operation voltage of silicon dioxide RRAM devices by controlling the growth of conductive OH [31]. It has been reported that some 2D materials such as Graphene, MoS2, BN, and WS2 [17,36– filaments (CFs).39] are Inable addition, to ameliorate the studiedthe performances devices exhibitof RRAM the devices. resistance Therefore, switching by combining (RS) characteristics the where bothcompatibility the volatile of SiO and2-based nonvolatile RRAM devices behaviors with the coexist dominant in silicon a single CMOS device. fabrication The technology coexistence of the volatile and[40–42] nonvolatile and thoseswitching, of a 2D MXene capable material of, we implementing designed and fabricated short-term Ag/MXene/SiO plasticity2/Pt (STP) structure- and long-term based RRAM devices. The introduction of 2D material MXene greatly lowers the operation voltage plasticity (LTP)of silicon rules, dioxide respectively, RRAM devices can by e controllingffectively the reduce growth the of conductive complexity filaments of brain-inspired (CFs). In addition, systems [43]. Finally, thethe working studied devices mechanisms exhibit the of resistance the Ag/MXene switching/SiO (RS)2/Pt characteristics devices have where been both discussed the volatile on and the basis of our experimentalnonvolatile results. behaviors This coexist work in maya single provide device. a The forward-looking coexistence of the solution volatile forand reducingnonvolatile the power consumptionswitching, of traditional capable of transitionimplementing metal short-term oxide-based plasticity RRAM(STP) and devices long-term as plasticity well as (LTP) the development rules, of respectively, can effectively reduce the complexity of brain-inspired systems [43]. Finally, the artificial intelligence systems on the hardware level. working mechanisms of the Ag/MXene/SiO2/Pt devices have been discussed on the basis of our experimental results. This work may provide a forward-looking solution for reducing the power 2. Materialsconsumption and Methods of traditional transition metal oxide-based RRAM devices as well as the development of artificial intelligence systems on the hardware level. The studied Ag/MXene/SiO2/Pt RRAM devices were successfully fabricated on Si wafer. Figure1a shows the2. scanning Materials and electron Methods microscope (SEM) image of the crossbar structure of devices that are more likely toThe be integratedstudied Ag/MXene/SiO and extended2/Pt RRAM with devices the conventional were successfully silicon fabricated CMOS on Si fabrication wafer. Figure technology. The vertical1a shows lines the of thescanning array electron are bottom microscope electrodes (SEM) imag ofe 90-nm-thickof the crossbar structure Pt and of the devices parallel that are lines are top electrodesmore of TiN likely/Ag to (80 be nm integrated/100 nm). and Pads extended (300 µwithm the300 conventionalµm) are illustrated silicon CMOS by whitefabrication dashed lines. × The red dashedtechnology. line framesThe vertical the lines core of intersection the array are region bottom (100electrodesµm of100 90-nm-thickµm) of topPt and and the bottom parallel electrodes, lines are top electrodes of TiN/Ag (80 nm/100 nm). Pads (300 μm× × 300 μm) are illustrated by white where the RSdashed occurs. lines. TheThe red 80 nmdashed RS line layer frames of SiO the2 corefilm intersection was sputtered region using(100 μm the × 100 physical μm) of vaportop and deposition (PVD) method.bottom The electrodes, MXene where (Ti3 Cthe2) RS layer occu onrs. SiOThe 280, preparednm RS layer by of etching SiO2 film Ti was3AlC sputtered2 with hydrogenusing the fluoride (HF), wasphysical deposited vapor by deposition spin-coating (PVD) at method. 500 rpm The for MXene 60 s. (Ti The3C2) cross-sectional layer on SiO2, prepared SEM image by etching indicates that the thicknessTi3AlC of2 thewith MXenehydrogen film fluoride is about (HF), was 50 nm,deposited as shown by spin-coating in Figure at 500 S1. rpm The for surface 60 s. The morphologycross- of sectional SEM image indicates that the thickness of the MXene film is about 50 nm, as shown in Figure Ti3C2 powderS1. The was surface characterized morphology by of Ti SEM3C2 powder (Figure was1b) characterized showing the by SEM chip-like (Figure multi-layered 1b) showing the nanostructure.chip- The X-raylike di ffmulti-layeredraction (XRD) nanostructure. figure indicates The X-ray that diffraction the main (XRD) component figure indicates of the that MXene the main used in our component of the MXene used in our experiment is Ti3C2 [44–46], as shown in Figure 1c. All the experiment is Ti3C2 [44–46], as shown in Figure1c. All the electrical characteristic measurements were performedelectrical by Keithley characteristic 4200A meas SCSurements semiconductor were performed parameter by Keithley analyzer. 4200A SCS semiconductor parameter analyzer.

Figure 1. (a) Scanning electron microscope (SEM) image of the TiN/Ag/MXene/SiO2/Pt crossbar structure. (b) The surface morphology of Ti3C2 powder was characterized by SEM, showing the classical multi-layered nanostructure. (c) The X-ray diﬀraction (XRD) shows that the MXene used in

this experiment is mainly composed by Ti3C2. Electronics 2020, 9, 2098 3 of 10

Electronics 2020, 9, x FOR PEER REVIEW 3 of 9 3. Results Figure 1. (a) Scanning electron microscope (SEM) image of the TiN/Ag/MXene/SiO2/Pt crossbar 3.1. Electrical Characteristicsstructure. (b) The surface morphology of Ti3C2 powder was characterized by SEM, showing the classical multi-layered nanostructure. (c) The X-ray diffraction (XRD) shows that the MXene used in By controllingthis experiment a compliance is mainly composed current by Ti3 limit,C2. we can obtain the electrical characteristics in Ag/MXene3./SiO Results2/Pt RRAM devices where both volatile and nonvolatile behaviors coexist in a single device. The volatile switching behavior would appear when the compliance current limit is relatively low3.1. Electrical (i.e., approximately Characteristics from 1 to 500 nA), whereas the nonvolatile switching behavior occurs with a relativelyBy controlling high compliance a compliance current current limit limit, (i.e., we approximatelycan obtain the electrical from 10 characteristicsµA to 5 mA). in Cycling Ag/MXene/SiO2/Pt RRAM devices where both volatile and nonvolatile behaviors coexist in a single experiments were performed on the Ag/MXene/SiO2/Pt RRAM devices under different compliance device. The volatile switching behavior would appear when the compliance current limit is relatively current limits,low (i.e., each approximately of which from consists 1 to 500 ofnA), 100 whereas cycles. the nonvolatile Figure2 switchinga,b shows behavior the typicaloccurs with volatile a I-V characteristicrelatively curves high under compliance the compliance current limit current (i.e., approximately limits of 1 from and 10010 μ nA,A to respectively.5 mA). Cycling They have similar thresholdexperiments switching were performed (TS) behaviors on the Ag/MXene/SiO [47], wherein2/Pt theRRAM forward devices scanningunder different (from compliance 0 to 0.2 V) excites them intocurrent the low limits, resistance each of which state consists (LRS) andof 100 the cycles. reverse Figure scanning 2a,b shows (from the typical 0.2 to volatile 0 V) makesI-V them characteristic curves under the compliance current limits of 1 and 100 nA, respectively. They have completelysimilar return threshold to the highswitching resistance (TS) behaviors state (HRS).[47], wherein Without the forward the Reset scanning operation, (from 0 theto 0.2 device V) would automaticallyexcites return them into to thethe low HRS. resistance Furthermore, state (LRS) we and statistically the reverse scanning analyzed (from the 0.2 to Set 0 V) voltages makes them of 100 cycles under twocompletely different return compliance to the high current resistance limits, state as (H shownRS). Without in Figure the Reset2c,d. operation, Most values the device of Set would voltages are around 0.06automatically V for a compliance return to the HRS. current Furthermore, limit of we 1 nA,statistically while analyzed the values the Set of voltages Set voltages of 100 cycles are observed under two different compliance current limits, as shown in Figure 2c,d. Most values of Set voltages around 0.18are Varound for a0.06 higher V for a compliancecompliance current current limit of limit 1 nA, of while 100 the nA. values Irrespective of Set voltages of are this, observed they are both ultra-low Setaround voltages 0.18 V applicablefor a higher compliance to low-power current artificial limit of 100 neural nA. Irrespective microcircuits. of this, Itthey should are both be ultra- noted that Set voltages decreaselow Set voltages with cycles applicable increasing to low-power under artificial a relatively neural lowmicrocircuits. compliance It should current be noted limit that (approximately Set from 1 nAvoltages to 500 nA),decrease which with may cycles be attributedincreasing under to few a residualrelatively metalliclow compliance Ag nanoparticles current limit during the (approximately from 1 nA to 500 nA), which may be attributed to few residual metallic Ag former phasenanoparticles [48] and during make the the former device phase easier [48] toand turn make on the at device lower easier operation to turn on voltages. at lower operation More supporting materials forvoltages. TS behaviors More supporting can be materials found for in TS supplementary behaviors can be Figuresfound in supplementary S2–S4. Figures S2–S4.

Figure 2. TypicalFigure 2. thresholdTypical threshold switching switching (TS) (TS)I-V I-Vcurves curves under under the the compliance compliance current current limits of limits (a) 1 and of (a) 1 and (b) 100 nA.(b The) 100 statistics nA. The statistics and distribution and distribution of Set of Set voltages voltages under under the compliance compliance current current limits limits of (c) 1 of (c) 1 nA, most of values are around 0.06 V, and (d) 100 nA, the values of Set voltages are observed around 0.18 V.

nA, most of values are around 0.06 V, and (d) 100 nA, the values of Set voltages are observed around 0.18 V.

When we continuously increased the value of the compliance current limit, non-volatile behavior was achieved in the same Ag/MXene/SiO2/Pt device. Figure 3a,b shows the typical RS I-V characteristics curves of 100 cycles under the compliance current limits of 500 μA and 1 mA, respectively. They have similar RS behaviors, wherein the Set process puts them into the LRS under Electronics 2020 9 positive voltage, , 2098 sweep and the Reset process restores them back to the HRS under the negative4 of 10 voltage sweep. This process is completely different from volatile behavior that can automatically isreturn completely to the HRS diﬀerent without from the volatile Reset behavioroperation. that In addition, can automatically the resistances return of to ON the and HRS OFF without states the at Reset0.01 V operation. were extracted In addition, for both the compliance resistances current of ON limits and OFF of 500 states μA and at 0.01 1 mA, V were as shown extracted in Figure for both 3c, complianced. It can be currentobserved limits that ofthe 500 resistanceµA and 1rati mA,o of as the shown OFF-State in Figure and3 c,ON-State d. It can is be almost observed 103. thatBesides, the resistancethe data retention ratio of thein both OFF-State ON and and OFF ON-State states wa is almosts measured 103. Besides,under the the compliance data retention current in both limits ON of and100 μ OFFA (shown states wasin Figure measured S5), under500 μA the and compliance 1 mA, respectively. current limits From of Figure 100 µA 3e,f, (shown we can in Figure see that S5), a 4 500relativelyµA and high 1 mA, and respectively. stable retention From (more Figure than3e,f, 1 we 10 can s)see can that be a obtained relatively under high anda higher stable compliance retention (morecurrent than limit1 of10 14 mA,s) can which be obtained means underthat the a higherstronger compliance the electrical current stimulation, limit of1 the mA, longer which the means data × thatretention. the stronger This phenomenon the electrical is stimulation, consistent thewith longer the LTP the datacharacteristic retention. of This biol phenomenonogical synapses, is consistent which is withregarded the LTPas the characteristic basis of learning of biological and memory synapses, [20]. However, which is regardedthe Set and as Reset the basis voltages of learning in these andtwo memoryRS processes [20]. are However, still very the low, Set andaround Reset 0.2 voltages V and − in0.2 these V, respectively, two RS processes benefiting are still the very development low, around of 0.2low-power V and 0.2brain-inspired V, respectively, computing beneﬁting systems. the development The low ofoperating low-power voltages brain-inspired in both TS computing and RS − systems.situations The are more low operating likely to be voltages associated in both with TS the and introduction RS situations of 2D are materials more likely MXene, to bewhich associated makes withthe CFs the introductiontend to grow of along 2D materials the location MXene, of MXen whiche makesnanostructures. the CFs tend Our torecent grow research along the work location has 3 2 2 2 ofdemonstrated MXene nanostructures. that the Ti C Our appears recent to research be bonded work to has the demonstrated amorphous SiO that to the form Ti3C an2 appears MXene/SiO to be 2 bondedcompound to the due amorphous to the surface SiO activity2 to form of anthe MXene dangling/SiO bonds2 compound in the cleaved due to amorphous the surface activitySiO substrate of the dangling[33]. bonds in the cleaved amorphous SiO2 substrate [33].

Figure 3. Typical resistance switching (RS) I-V curves of 100 consecutive cycles (gray curves) under the compliance current limits of (a) 500 µA and (b) 1 mA, respectively. ON and OFF resistance states during 100 cycles under the compliance current limit of (c) 500 µA and (d) 1 mA, respectively, extracted at 0.01 V. The data retention under the compliance current limits of (e) 500 µA and (f) 1 mA for the same device. Electronics 2020, 9, 2098 5 of 10

3.2. Biological Synaptic Simulation Electronics 2020, 9, x FOR PEER REVIEW 5 of 9 In biological synapses, once the action potential (AP) arrives at the presynaptic nerve terminal, excitatory postsynapticFigure 3. Typical current resistance (EPSC) switching would (RS) I-V curves generate of 100 withconsecutive the cycles open (gray of curves) ion channels under of Na+ in the compliance current limits of (a) 500 μA and (b) 1 mA, respectively. ON and OFF resistance states postsynaptic membranes,during 100 cycles which under causesthe compliance the change current oflimit synaptic of (c) 500 weight μA and between(d) 1 mA, respectively, two neurons, as shown in Figure4a [ 4,extracted5]. As aat pair0.01 V. of The identical data retention pulses under is the given compliance to the current presynaptic limits of (e nerve) 500 μA terminal, and (f) 1 mA EPSC-2 caused by the second pulsefor the same is significantly device. enhanced. This phenomenon is called paired-pulse facilitation (PPF) of synapses,3.2. anBiological important Synaptic characteristic Simulation of STP [49,50]. Hereby, we simulated the PPF characteristic in Ag/MXene/SiO /Pt RRAM devices. A pair of spikes with the pulse amplitude of 0.2 V and pulse In biological2 synapses, once the action potential (AP) arrives at the presynaptic nerve terminal, width of 90excitatory ms was postsynaptic applied to current the Ag (EPSC)/MXene would/SiO generate2/Pt device, with the and open the of corresponding ion channels of responsesNa+ in of the PPF characteristicpostsynaptic are membranes, obtained, which as shown causes in the Figure change4b. of It synaptic can be weight observed between that two EPSC-2 neurons, ismuch as larger than EPSC-1shown caused in Figure by the 4a [4,5]. first As pulse, a pair which of identical is consistent pulses is withgiven theto the phenomenon presynaptic nerve of biological terminal, synaptic EPSC-2 caused by the second pulse is significantly enhanced. This phenomenon is called paired-pulse plasticity [51,52]. The PPF characteristic can be better elucidated by considering the change between facilitation (PPF) of synapses, an important characteristic of STP [49,50]. Hereby, we simulated the two peakPPF currents characteristic of EPSC-2 in Ag/MXene/SiO and EPSC-12/Pt RRAM induced devices. by A a pair of of spikes identical with the pulses pulse amplitude versus theof interval time (∆t).0.2 As V shown and pulse in width Figure of4 90c, ms the was index applied of toPPF the =Ag/MXene/SiO(I2 I1)/I12/Pt 100%device, exponentiallyand the corresponding decreases with responses of the PPF characteristic are obtained, as shown− in Figure ×4b. It can be observed that EPSC- ∆t increasing, I1 and I2 are the values of the peak currents after the first and second pulses, respectively. 2 is much larger than EPSC-1 caused by the first pulse, which is consistent with the phenomenon of In this experiment, the index of PPF can be fitted with the following equation: biological synaptic plasticity [51,52]. The PPF characteristic can be better elucidated by considering the change between two peak currents of EPSC-2 and EPSC-1 induced by a pair of identical pulses PPF∆𝑡= C exp( ∆t/τ ) + C exp( ∆t/τ𝑃𝑃𝐹) 𝐼 𝐼 /𝐼 100% versus the interval time ( ). As1 shown− in Figure1 4c,2 the index− of 2 exponentially decreases with ∆𝑡 increasing, 𝐼 and 𝐼 are the values of the peak currents after the where twofirst characteristic and second relaxationpulses, respectively. times τ 1In= this5.83 experiment,ms and τthe2 = index73.45 of msPPFwere can be obtained, fitted with corresponding the following equation: to fast and slow decay terms, respectively [20,53]. The constants of C1 and C2 are equal to 1.21 and 694.98, 𝑃𝑃𝐹 𝐶 exp∆𝑡/𝜏 𝐶 exp∆𝑡/𝜏 respectively. The PPF characteristic in the Reset process was also obtained and the index of PPF where two characteristic relaxation times 𝜏 5.83 ms and 𝜏 73.45 ms were obtained, exponentially decreases with ∆t increasing as well (Figure S6). Furthermore, the LTP characteristic corresponding to fast and slow decay terms, respectively [20,53]. The constants of 𝐶 and 𝐶 are has been mimickedequal to 1.21 in and our 694.98, Ag/ MXenerespectively./SiO The2/Pt PPF RRAM characteristic devices, in the as Reset shown process in Figure was also4d. obtained After applying 100 consecutiveand the positiveindex of PPF spikes exponentially (pulse amplitudedecreases with is ∆𝑡 0.2 increasing V, pulse as width well (Figure is 25 S6). ms), Furthermore, the conductance of devices beginsthe LTP to characteristic graduallyincrease has been frommimicked the in initial our Ag/MXene/SiO state to a saturation2/Pt RRAM resistance devices, as state.shown Thisin process Figure 4d. After applying 100 consecutive positive spikes (pulse amplitude is 0.2 V, pulse width is 25 lasts for ams), long the period conductance of time of anddevices is termedbegins to thegradually long-term increase potentiation from the initial (red state scatters). to a saturation Subsequently, 100 consecutiveresistance negative state. This spikes process last (pulses for a amplitude long period of is time 0.2 and V, pulseis termed width the long-term is 25 ms) potentiation were applied to the same device.(red scatters). The Subsequently, conductance 100 gradually consecutive decreases negative spikes to a (pulse certain amplitude resistance is 0.2 V, state, pulse being width known as the long-termis 25 ms) depression were applied (blue to the scatters).same device. Commonly, The conductance the gradually LTP is decreases regarded to a ascertain the resistance basis of learning state, being known as the long-term depression (blue scatters). Commonly, the LTP is regarded as and memorythe basis [20 ,of54 learning–56] that and canmemory be applied[20,54–56] that into can the be synapticapplied into weights the synaptic training weights intraining artificial in neural networks [artificial4]. neural networks [4].

Figure 4. Figure(a) The 4. (a structural) The structural schematic schematic of the the signal signal transmission transmission between between synapses. (b synapses.) The blue curve (b) The blue is a pair of identical spikes applied to the top electrode with the pulse amplitude of 0.2 V and pulse curve is a pair of identical spikes applied to the top electrode with the pulse amplitude of 0.2 V and pulse width of 90 ms. The red curve is the corresponding responses of the paired-pulse facilitation (PPF) characteristic, and the excitatory postsynaptic current (EPSC)-2 is signiﬁcantly enhanced. (c) The relationship between the index of PPF and pulse intervals ∆t, displaying an exponential decrease. (d) The long-term potentiation (red scatters) and long-term depression (blue scatters)

characteristics of the Ag/MXene/SiO2/Pt RRAM devices. Electronics 2020, 9, 2098 6 of 10

4. Discussion

The Ag/MXene/SiO2/Pt RRAM devices have been designed and fabricated for the application of electronic synapses. By controlling a compliance current limit, the RS characteristics can be obtained in a single Ag/MXene/SiO2/Pt RRAM device where the volatile and nonvolatile switching behaviors coexist. The volatile switching behavior appears when the compliance current limit is relatively low, whereas the nonvolatile switching behavior occurs with a relatively high compliance current limit. Furthermore, the synaptic plasticity functions have been mimicked in the same Ag/MXene/SiO2/Pt RRAM device, such as the PPF and LTP characteristic. After combining the electrical characteristics with the biological synaptic simulation, we speculated the RS mechanism. When the positive voltage is applied to the Ag/MXene/SiO2/Pt RRAM device, the Ag ions generate by oxidation reaction at the top electrode and migrate to the bottom electrode. The metallic CFs of Ag would form with the accumulated Ag atoms, and this is the Set process. When a relatively low compliance current limit was applied to the Ag/MXene/SiO2/Pt device, a narrow metallic CF may be composed of discrete Ag nanoparticles [48,57,58] and forms in the RS layer. The CFs in this situation are unstable and tend to rupture spontaneously. The above process is the TS: the device would automatically return to the HRS without the Reset operation, similar to the STP characteristic of biological synapses. When we apply a relatively high compliance current limit, a thicker CF would be formed in the Ag/MXene/SiO2/Pt RRAM device. In this situation, the ON-State can be maintained after the Set operation and the Reset process is needed to restore them back to the OFF-State again. This is the typical nonvolatile RS process that is consistent with the LTP characteristic of biological synapses. Moreover, it is noted that the Ti3C2 appears to be bonded to the amorphous SiO2 to form an MXene/SiO2 compound due to the surface activity of the dangling bonds in the cleaved amorphous SiO2 substrate [33], which makes the CFs more likely to grow along the location of MXene nanostructures, accordingly reducing the power consumption of Ag/MXene/SiO2/Pt RRAM devices. Finally, the electrical characteristics of the Cu/MXene/SiO2/W RRAM device under the compliance current limits of 500 µA and 1 mA (shown in Figure S7) further demonstrate our proposed RS mechanism where the RS processes are determined by oxidation-reduction reactions of Cu or Ag ions in the SiO2 layer, which rule out the phase transition characteristic of the MXene material.

Supplementary Materials: The following are available online at http://www.mdpi.com/2079-9292/9/12/2098/s1, Figure S1: a single layer or few-layered MXene on glass slide and the cross-sectional SEM image of the MXene film, Figure S2: the forming processes under different compliance current limits, Figure S3: the TS behaviors in both positive and negative bias directions under the compliance current limit of 100 nA, Figure S4: the statistics of TS voltages versus the initial HRS resistances under the compliance current limits of 1 nA and 100 nA, respectively, Figure S5: the retention measurement under the compliance current limit of 100 µA, Figure S6: the PPF characteristic in the Reset process, Figure S7: The typical I-V curves of Cu/MXene/SiO2/W RRAM device under the compliance current limits of 500 uA and 1 mA, respectively. Author Contributions: Conceptualization, X.L. (Xiaojuan Lian) and Y.T.; methodology, X.L. (Xiaojuan Lian), X.S., E.H.; investigation, X.L. (Xiaojuan Lian), X.S. and X.W.; data curation, X.S., J.F. and Z.G.; writing—original draft preparation, X.L. (Xiaojuan Lian) and X.S.; writing—review and editing, X.L. (Xiaojuan Lian), X.S., J.F., Z.G., X.W., X.L.(Xiaoyan Liu), E.H., J.X. and Y.T.; supervision, X.L. (Xiaojuan Lian); project administration, X.L. (Xiaojuan Lian) and Y.T.; funding acquisition, X.L. (Xiaojuan Lian), J.X. and Y.T. All authors have read and agreed to the published version of the manuscript. Funding: This research was funded in part by the National Natural Science Foundation of China (61804079, 21671167), the Science Foundation of Jiangsu Province (CZ1060619001, SZDG2018007), the Science Foundation of National and Local Joint Engineering Laboratory of RF Integration and Micro-Assembly Technology (KFJJ20200102), the Science Research Funds for Nanjing University of Posts and Telecommunications (NY218110, BK20191202). Conflicts of Interest: The authors declare no conflict of interest. Electronics 2020, 9, 2098 7 of 10

References

1. Xia, Q.; Yang, J.J. Memristive crossbar arrays for brain-inspired computing. Nat. Mater. 2019, 18, 309–323. [CrossRef][PubMed] 2. Tang, J.; Yuan, F.; Shen, X.; Wang, Z.; Rao, M.; He, Y.; Sun, Y.; Li, X.; Zhang, W.; Li, Y.; et al. Bridging Biological and Artificial Neural Networks with Emerging Neuromorphic Devices: Fundamentals, Progress, and Challenges. Adv. Mater. 2019, 31, 1902761. [CrossRef][PubMed] 3. Sun, Z.; Ambrosi, E.; Bricalli, A.; Ielmini, D. Logic Computing with Stateful Neural Networks of Resistive Switches. Adv. Mater. 2018, 30, 1802554. [CrossRef][PubMed] 4. Zhang, T.; Yang, K.; Xu, X.; Cai, Y.; Yang, Y.; Huang, R. Memristive Devices and Networks for Brain-Inspired Computing. Phys. Status Solidi RRL 2019, 13, 1900029. [CrossRef] 5. Kuzum, D.; Yu, S.; Philip Wong, H.-S. Synaptic electronics: Materials, devices and applications. Nanotechnology 2013, 24, 382001. [CrossRef][PubMed] 6. Wu, Q.; Wang, H.; Luo, Q.; Banerjee, W.; Cao, J.; Zhang, X.; Wu, F.; Liu, Q.; Li, L.; Liu, M. Fully imitation synaptic metaplasticity based on memristor device. Nanoscale 2018, 10, 5875–5881. [CrossRef] 7. Ambrogio, S.; Balatti, S.; Milo, V.; Carboni, R.; Wang, Z.-Q.; Calderoni, A.; Ramaswamy, N.; Ielmini, D. Neuromorphic Learning and Recognition with One-Transistor-One-Resistor Synapses and Bistable Metal Oxide RRAM. IEEE Trans. Electron Devices 2016, 63, 1508–1515. [CrossRef] 8. Ielmini, D. Brain-inspired computing with resistive switching memory (RRAM): Devices, synapses and neural networks. Microelectron. Eng. 2018, 190, 44–53. [CrossRef] 9. Dai, S.; Zhao, Y.; Wang, Y.; Zhang, J.; Fang, L.; Jin, S.; Shao, Y.; Huang, J. Recent Advances in Transistor-Based Artificial Synapses. Adv. Funct. Mater. 2019, 29, 1903700. [CrossRef] 10. Hur, J.; Jang, B.C.; Park, J.; Moon, D.-I.; Bae, H.; Park, J.-Y.; Kim, G.-H.; Jeon, S.-B.; Seo, M.; Kim, S.; et al. A Recoverable Synapse Device Using a Three-Dimensional Silicon Transistor. Adv. Funct. Mater. 2018, 28, 1804844. [CrossRef] 11. Kuzum, D.; Jeyasingh, R.G.D.; Lee, B.; Wong, H.-S.P. Nanoelectronic Programmable Synapses Based on Phase Change Materials for Brain-Inspired Computing. Nano Lett. 2012, 12, 2179–2186. [CrossRef] 12. Burr, G.W.; Shelby, R.M.; Sidler, S.; di Nolfo, C.; Jang, J.; Boybat, I.; Shenoy, R.S.; Narayanan, P.; Virwani, K.; Giacometti, E.U.; et al. Experimental Demonstration and Tolerancing of a Large-Scale Neural Network (165000 Synapses) Using Phase-Change Memory as the Synaptic Weight Element. IEEE Trans. Electron Devices 2015, 62, 3498–3507. [CrossRef] 13. Yang, N.; Ren, Z.-Q.; Hu, C.-Z.; Guan, Z.; Tian, B.-B.; Zhong, N.; Xiang, P.-H.; Duan, C.-G.; Chu, J.-H. Ultra-wide temperature electronic synapses based on self-rectifying ferroelectric memristors. Nanotechnology 2019, 30, 464001. [CrossRef] 14. Boyn, S.; Grollier, J.; Lecerf, G.; Xu, B.; Locatelli, N.; Fusil, S.; Girod, S.; Carrétéro, C.; Garcia, K.; Xavier, S.; et al. Learning through ferroelectric domain dynamics in solid-state synapses. Nat. Commun. 2017, 8, 14736. [CrossRef][PubMed] 15. Wang, Z.; Joshi, S.; Savel’ev, S.E.; Jiang, H.; Midya, R.; Lin, P.; Hu, M.; Ge, N.; Strachan, J.P.; Li, Z.; et al. Memristors with diffusive dynamics as synaptic emulators for neuromorphic computing. Nat. Mater. 2017, 16, 101–108. [CrossRef][PubMed] 16. van de Burgt, Y.; Lubberman, E.; Fuller, E.J.; Keene, S.T.; Faria, G.C.; Agarwal, S.; Marinella, M.J.; Alec Talin, A.; Salleo, A. A non-volatile organic electrochemical device as a low-voltage artificial synapse for neuromorphic computing. Nat. Mater. 2017, 16, 414–418. [CrossRef] 17. Yan, X.; Zhao, Q.; Chen, A.P.; Zhao, J.; Zhou, Z.; Wang, J.; Wang, H.; Zhang, L.; Li, X.; Xiao, Z.; et al.

Vacancy-Induced Synaptic Behavior in 2D WS2 Nanosheet–Based Memristor for Low-Power Neuromorphic Computing. Small 2019, 15, 1901423. [CrossRef] 18. Valentian, A.; Rummens, F.; Vianello, E.; Mesquida, T.; Boissac, L.-M.B.; Bichler, O.; Reita, C. Fully Integrated Spiking Neural Network with Analog Neurons and RRAM Synapses. In Proceedings of the 2019 IEEE International Electron Devices Meeting (IEDM 2019), San Francisco, CA, USA, 7–11 December 2019; pp. 314–317. Electronics 2020, 9, 2098 8 of 10

19. Yu, S.; Gao, B.; Fang, Z.; Yu, H.; Kang, J.; Wong, H.-S.P. A Low Energy Oxide-Based Electronic Synaptic Device for Neuromorphic Visual Systems with Tolerance to Device Variation. Adv. Mater. 2013, 25, 1774–1779. [CrossRef] 20. Whitlock, J.R. Learning Induces Long-Term Potentiation in the Hippocampus. Science 2006, 313, 1093–1097. [CrossRef] 21. Milo, V.; Zambelli, C.; Olivo, P.; Pérez, E.; Mahadevaiah, M.K.; Ossorio, O.G.; Wenger, C.; Ielmini, D.

Multilevel HfO2-based RRAM devices for low-power neuromorphic networks. APL Mater. 2019, 7, 081120. 22. Lin, C.-Y.; Chen, J.; Chen, P.-H.; Chan, T.-C.; Wu, Y.; Eshraghian, J.K.; Moon, J.; Yoo, S.; Wang, Y.-H.; Chen, W.-C.; et al. Adaptive Synaptic Memory via Lithium Ion Modulation in RRAM Devices. Small 2020, 16, 2003964. [CrossRef] 23. Prezioso, M.; Merrikh-Bayat, F.; Hoskins, B.D.; Adam, G.C.; Likharev, K.K.; Strukov, D.B. Training and operation of an integrated neuromorphic network based on metal-oxide memristors. Nature 2015, 521, 61–64. [CrossRef][PubMed] 24. Shen, Z.; Zhao, C.; Qi, Y.; Xu, W.; Liu, Y.; Mitrovic, I.Z.; Yang, L.; Zhao, C. Advances of RRAM Devices: Resistive Switching Mechanisms, Materials and Bionic Synaptic Application. Nanomaterials 2020, 10, 1437. [CrossRef][PubMed] 25. Park, J.; Kwak, M.; Moon, K.; Woo, J.; Lee, D.; Hwang, H. TiOx-Based RRAM Synapse With 64-Levels of Conductance and Symmetric Conductance Change by Adopting a Hybrid Pulse Scheme for Neuromorphic Computing. IEEE Electron Device Lett. 2016, 37, 1559–1562. [CrossRef] 26. Naguib, M.; Kurtoglu, M.; Presser, V.; Lu, J.; Niu, J.; Heon, M.; Hultman, L.; Gogotsi, Y.; Barsoum, M.W.

Two-Dimensional Nanocrystals Produced by Exfoliation of Ti3AlC2. Adv. Mater. 2011, 23, 4248–4253. [CrossRef][PubMed] 27. Naguib, M.; Mashtalir, O.; Carle, J.; Presser, V.; Lu, J.; Hultman, L.; Gogotsi, Y.; Barsoum, M.W. Two-Dimensional Transition Metal Carbides. ACS Nano 2012, 6, 1322–1331. [CrossRef][PubMed]

28. Soundiraraju, B.; George, B.K. Two-Dimensional Titanium Nitride (Ti2N) MXene: Synthesis, Characterization, and Potential Application as Surface-Enhanced Raman Scattering Substrate. ACS Nano 2017, 11, 8892–8900. [CrossRef] 29. Alhabeb, M.; Maleski, K.; Anasori, B.; Lelyukh, P.; Clark, L.; Sin, S.; Gogotsi, Y. Guidelines for

Synthesis and Processing of Two-Dimensional Titanium Carbide (Ti3C2Tx MXene). Chem. Mater. 2017, 29, 7633–7644. [CrossRef] 30. Naguib, M.; Gogotsi, Y. Synthesis of two-dimensional materials by selective extraction. Acc. Chem. Res. 2015, 48, 128–135. [CrossRef] 31. Shen, L.; Jeon, J.; Jang, S.K.; Xu, J.; Choi, Y.J.; Park, J.-H.; Hwang, E.; Lee, S. Surface group modiﬁcation and

carrier transport property of layered transition metal carbides (Ti2CTx, T: -OH, -F and –O). Nanoscale 2015, 7, 19390–19396. 32. Naguib, M.; Come, J.; Dyatkin, B.; Presser, V.; Taberna, P.-L.; Simon, P.; Barsoum, M.W.; Gogotsi, Y. MXene: A promising transition metal carbide anode for lithium-ion batteries. Electrochem. Commun. 2012, 16, 61–64. [CrossRef] 33. Lian, X.; Shen, X.; Zhang, M.; Xu, J.; Gao, F.; Wan, X.; Hu, E.; Guo, Y.; Zhao, J.; Tong, Y. Resistance

switching characteristics and mechanisms of MXene/SiO2 structure-based memristor. Appl. Phys. Lett. 2019, 115, 063501. [CrossRef] 34. Chen, Y.; Wang, Y.; Luo, Y.; Liu, X.; Wang, Y.; Gao, F.; Xu, J.; Hu, E.; Samanta, S.; Wan, X.; et al. Realization of Artiﬁcial Neuron Using MXene Bi-Directional Threshold Switching Memristors. IEEE Electron Device Lett. 2019, 40, 1686–1689. [CrossRef] 35. Gogotsi, Y.; Anasori, B. The Rise of MXenes. ACS Nano 2019, 13, 8491–8494. [CrossRef][PubMed] 36. Lee, J.; Du, C.; Sun, K.; Kioupakis, E.; Lu, W.D. Tuning Ionic Transport in Memristive Devices by Graphene with Engineered Nanopores. ACS Nano 2016, 10, 3571–3579. [CrossRef][PubMed] Electronics 2020, 9, 2098 9 of 10

37. Li, D.; Wu, B.; Zhu, X.; Wang, J.; Ryu, B.; Lu, W.D.; Lu, W.; Liang, X. MoS2 Memristors Exhibiting Variable Switching Characteristics toward Biorealistic Synaptic Emulation. ACS Nano 2018, 12, 9240–9252. [CrossRef][PubMed] 38. Wang, M.; Cai, S.; Pan, C.; Wang, C.; Lian, X.; Zhuo, Y.; Xu, K.; Cao, T.; Pan, X.; Wang, B.; et al. Robust memristors based on layered two-dimensional materials. Nat. Electron 2018, 1, 130–136. [CrossRef] 39. Hui, F.; Villena, M.A.; Fang, W.; Lu, A.-Y.; Kong, J.; Shi, Y.; Jing, X.; Zhu, K.; Lanza, M. Synthesis of large-area multilayer hexagonal boron nitride sheets on iron substrates and its use in resistive switching devices. 2D Mater. 2018, 5, 031011. [CrossRef] 40. Lian, X.; Shen, X.; Lu, L.; He, N.; Wan, X.; Samanta, S.; Tong, Y. Resistance Switching Statistics and Mechanisms of Pt Dispersed Silicon Oxide-Based Memristors. Micromachines 2019, 10, 369. [CrossRef] 41. Bricalli, A.; Ambrosi, E.; Laudato, M.; Maestro, M.; Rodriguez, R.; Ielmini, D. Resistive Switching Device Technology Based on Silicon Oxide for Improved ON–OFF Ratio—Part II: Select Devices. IEEE Trans. Electron Devices 2018, 65, 122–128. [CrossRef] 42. Mehonic, A.; Shluger, A.L.; Gao, D.; Valov, I.; Miranda, E.; Ielmini, D.; Bricalli, A.; Ambrosi, E.; Li, C.; Yang, J.J.; et al. Silicon Oxide (SiOx): A Promising Material for Resistance Switching? Adv. Mater. 2018, 30, 1801187. [CrossRef][PubMed] 43. Shi, Y.; Pan, C.; Chen, V.; Raghavan, N.; Pey, K.L.; Puglisi, F.M.; Pop, E.; Wong, H.-S.P.; Lanza, M. Coexistence of volatile and non-volatile resistive switching in 2D h-BN based electronic synapses. In Proceedings of the 2017 IEEE International Electron Devices Meeting (IEDM 2017), San Francisco, CA, USA, 2–6 December 2017; pp. 5.4.1–5.4.4. 44. Zhang, M.; Wang, Y.; Gao, F.; Wang, Y.; Shen, X.; He, N.; Zhu, J.; Chen, Y.; Wan, X.; Lian, X.; et al. Formation of new MXene ﬁlm using spinning coating method with DMSO solution and its application in advanced memristive device. Ceram. Int. 2019, 45, 19467–19472. [CrossRef] 45. Wang, K.; Zhou, Y.; Xu, W.; Huang, D.; Wang, Z.; Hong, M. Fabrication and thermal stability of

two-dimensional carbide Ti3C2 nanosheets. Ceram. Int. 2016, 42, 8419–8424. [CrossRef] 46. Feng, A.; Yu, Y.; Wang, Y.; Jiang, F.; Yu, Y.; Mi, L.; Song, L. Two-dimensional MXene Ti3C2 produced by exfoliation of Ti3AlC2. Mater. Des. 2017, 114, 161–166. [CrossRef] 47. Wang, Z.; Rao, M.; Midya, R.; Joshi, S.; Jiang, H.; Lin, P.; Song, W.; Asapu, S.; Zhuo, Y.; Li, C.; et al. Threshold Switching of Ag or Cu in Dielectrics: Materials, Mechanism, and Applications. Adv. Funct. Mater. 2018, 28, 1704862. [CrossRef] 48. Sun, H.; Liu, Q.; Li, C.; Long, S.; Lv, H.; Bi, C.; Huo, Z.; Li, L.; Liu, M. Direct Observation of Conversion between Threshold Switching and Memory Switching Induced by Conductive Filament Morphology. Adv. Funct. Mater. 2014, 24, 5679–5686. [CrossRef] 49. Thomson, A.M. Facilitation, augmentation and potentiation at central synapses. Trends Neurosci. 2000, 23, 305–312. [CrossRef] 50. Kim, M.-K.; Lee, J.-S. Short-Term Plasticity and Long-Term Potentiation in Artiﬁcial Biosynapses with Diﬀusive Dynamics. ACS Nano 2018, 12, 1680–1687. [CrossRef] 51. Regehr, W.G. Short-Term Presynaptic Plasticity. Cold Spring Harb. Perspect. Biol. 2012, 4, a005702. [CrossRef] 52. Zucker, R.S.; Regehr, W.G. Short-Term Synaptic Plasticity. Annu. Rev. Physiol. 2002, 64, 355–405. [CrossRef] 53. Hu, S.G.; Liu, Y.; Chen, T.P.; Liu, Z.; Yu, Q.; Deng, L.J.; Yin, Y.; Hosaka, S. Emulating the paired-pulse facilitation of a biological synapse with a NiOx-based memristor. Appl. Phys. Lett. 2013, 102, 183510. [CrossRef] 54. Toyoda, H.; Zhao, M.-G.; Xu, H.; Wu, L.-J.; Ren, M.; Zhuo, M. Requirement of Extracellular Signal-Regulated Kinase/Mitogen-Activated Protein Kinase for Long-Term Potentiation in Adult Mouse Anterior Cingulate Cortex. Mol. Pain 2007, 3, 1744–8069. [CrossRef][PubMed] 55. Hirano, T. Regulation and Interaction of Multiple Types of Synaptic Plasticity in a Purkinje Neuron and Their Contribution to Motor Learning. Cerebellum 2018, 17, 756–765. [CrossRef][PubMed] 56. Citri, A.; Malenka, R.C. Synaptic Plasticity: Multiple Forms, Functions, and Mechanisms. Neuropsychopharmacology 2008, 33, 18–41. [CrossRef] Electronics 2020, 9, 2098 10 of 10

57. Hsiung, C.-P.; Liao, H.-W.; Gan, J.-Y.; Wu, T.-B.; Hwang, J.-C.; Chen, F.; Tsai, M.-J. Formation and Instability of Silver Nanofilament in Ag-Based Programmable Metallization Cells. ACS Nano 2010, 4, 5414–5420. [CrossRef] 58. Wang, W.; Wang, M.; Ambrosi, E.; Bricalli, A.; Laudato, M.; Sun, Z.; Chen, X.; Ielmini, D. Surface diﬀusion-limited lifetime of silver and copper nanoﬁlaments in resistive switching devices. Nat. Commun. 2019, 10, 81. [CrossRef]

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional aﬃliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).