Filamentary-based organic memristors for wearable neuromorphic computing systems

A filamentary-based organic memristor is a promising synaptic component for the development of neuromorphic systems for wearable electronics. In the organic memristors, metallic conductive filaments (CF) are formed via electrochemical metallization under electric stimuli, and it results in the resistive switching characteristics. To realize the bio-inspired computing systems utilizing the organic memristors, it is essential to effectively engineer the CF growth for emulating the complete synaptic functions in the device. Here, the fundamental principles underlying the operation of organic memristors and parameters related to CF growth are discussed. Additionally, recent studies that focused on controlling CF growth to replicate synaptic functions, including reproducible resistive switching, continuous conductance levels, and synaptic plasticity, are reviewed. Finally, upcoming research directions in the field of organic memristors for wearable smart computing systems are suggested.


Introduction
Von Neumann computing systems have rapidly developed in the last decades, roughly following Moore's law.However, the demands for processing vast amounts of data in diverse technologies such as self-driving cars, image recognition, IoT, and artificial intelligence considerably increases the importance of computation speed [1][2][3][4][5][6].Consequently, the traditional von Neumann architecture, based on CMOS devices, has been deemed an undesirable option due to a bottleneck problem.Recently, neuromorphic systems capable of rapidly processing large amounts of data have garnered considerable attention as a new computing paradigm because of their ability to overcome the limitations of von Neumann devices [7][8][9][10].In neuromorphic systems, biological networks consisting of neurons and synapses are mimicked, enabling parallel computation to be performed in a manner similar to their biological counterparts [3,11,12].In order to implement bio-realistic computing systems, it is essential to completely mimic biological neural networks in terms of integration density and energy consumption [13][14][15][16][17]. Specifically, complex and practical computations close to those in the human brain require the neuromorphic systems to contain approximately 10 12 neurons and about 10 15 synapses, while consuming less than 20 W during computations.
A resistive switching memory, called as a memristor, is one of the promising candidates as an artificial synapse, owing to its two-terminal structure suitable for high integration [18,19].In typical memristors, a sandwich structure comprising metal/switching layer/metal is employed, and the resistance state of the switching layer is adjusted by external electric stimuli.To date, various switching layers, including oxide films [20][21][22], 2D materials [23,24], and organic films [25][26][27], have been utilized in memristors.Among these, organic material-based memristors have been demonstrated as viable synaptic components for smart wearable computing systems, owing to their tunable electrical characteristics, mechanical flexibility, and biocompatibility [19,28,29].For organic memristors, the resistive switching behaviors can be induced through several types of operating principles, such as charge trapping [30][31][32], ion migration [33][34][35], and filamentary mechanisms [18,19].In the devices governed by charge trapping and ion migration phenomena, current conduction is determined by the capture of charge carriers and ions at trapping sites within a switching layer, respectively.For such devices, scalability is inherently limited by the dependence of electrical performance on cell size.Additionally, the unstable characteristics of trapped charges or ions results in poor memory retention properties.In contrast, the filamentary based organic memristors possess the stable memory properties and the large scalability [19,36].In these memristors, nano-sized conductive filaments (CFs) connecting the electrodes are formed (or ruptured) in the switching layer in response to the electric stimuli, leading to the occurrence of stable resistive switching behaviors, regardless of the cell size.
Organic memristors demonstrate high potentials for realizing various applications for smart electronics including logic operators, image recognition, combinatorial optimization, and electronic skin [37][38][39][40].Considerable efforts have been invested in the realization of synaptic functions in organic memristors through the engineering of CF growth (see figure 1) [41].However, the fabrication of bio-realistic synaptic devices utilizing organic memristors remains challenging; the same is the case for the development of practical wearable neuromorphic systems approaching the complexity of the brain.To realize hardware neural networks resembling their biological counterparts, artificial synapses must completely replicate synaptic functions including synaptic plasticity, continuous weight states, and high integration density [30,31].While it is important to emulate complete synaptic functions in organic memristors to achieve smart wearable computing systems, the direction and objectives for research in this area have not been adequately discussed.
In this paper, we review recent studies on the engineering of CF growth in organic memristors to mimic synaptic functions; additionally, we suggest future approaches toward realizing wearable neuromorphic computing systems.First, we expound on the fundamental background for CF dynamics that govern the resistive switching of organic memristors.Second, considering that organic memristors must replicate biological synaptic functions, including reliability, uniformity, continuous conductance levels, and synaptic plasticity, we explore strategies for constructing biomimetic neural networks for wearable devices.Next, we offer applications for organic memristor-based wearable neuromorphic systems, ranging from single devices to system-level implementations.Finally, we discuss the prospects of the organic memristors for wearable neuromorphic computing systems.

Mechanism of CF formation
In filamentary-based organic memristors, the formation of CFs is achieved through the phenomenon of electrochemical metallization (ECM) involving thress successive steps, as shown in figure 2(a) [42].When positive electric stimuli are applied to the active electrode, such as Ag and Cu, of the devices, metal ions are generated by the process of electrochemical redox reactions (Step 1).Then, the metal ions drift toward the inert electrode at the electric field (Step 2).The ions reaching the inert electrode, such as Au and Pt, undergo cathodic deposition, leading to electrocrystallization (Step 3).As these steps are repeatedly performed under the electric stimulation, CFs composed of active metal atoms grow between the electrodes, ultimately resulting in a transition of the memristor from the high resistance state (HRS) to the low resistance state (LRS).In addition, the CFs can be ruptured using electric stimuli of an opposite polarity through electrochemical dissolution, thereby causing the transition from the LRS to the HRS.The electrochemical reactions responsible for the resistive switching behaviors of the devices are reversible.The reactions involved in the growth and rupture of the CFs are referred to as set and reset processes, respectively.For the initial resistive switching called as an electroforming process, more electric stimuli are required to form the CFs in the memristors, compared to the subsequent set process for re-forming the CFs [42,43].

Organic materials for CF growth
In organic materials, the ECM phenomenon for the CF growth is governed by various parameters of the polymer including chain length, molecular weight, side chain, and conjugated core [36,45].Therefore, it is difficult to speculate the CF growth at a specific organic medium.Until now, the insulating polymer based memristors have been widely used for flexible neuromorphic systems due to their low leakage current and high flexibility [18,19,38,39].Furthermore, addressing growing environmental apprehensions, various strategies have been investigated to integrate bio-materials into memristors.Eco-friendly memristors, employing non-toxic materials such as silk fibroin, keratin, and pectin, were developed using conventional coating processes [46][47][48].These devices exhibited the stable resistive switching behaviors with the CF formation, and potentials for green electronics.

Parameters affecting CF structures
The ECM phenomena involving redox reactions and ion drift are primarily governed by the material properties of the organic memristors and the parameters of the external electric stimuli [49].Consequently, the growth and structure of CFs can be tuned through the use of different material combinations and the engineering of electric signals for switching the device resistance.This section elaborates on the parameters affecting the CF structure in organic memristors.

Material properties
The dynamics and structure of the CFs are dependent on the electrochemical redox reactions and ion drift in the memristors [44,50].For the inorganic material systems, it was reported that the correlation between the redox rates and the ion mobility determines the CF structures in the devices (see figure 2(b)) [44].When both the redox rates and ion mobility are high and uniform, the metal ions are electrocrystallized at the inert electrode, leading to the CF growth with an inverted cone-shaped structure.On the contrary, the low and inhomogeneous redox rates and ion mobility induce the formation of the cluster-structured CF inside the switching film.In addition, the CFs can be engineered to fill the gap between the electrodes and grow in branch-like structures from the inert electrode, by controlling the ion mobility and redox rates, respectively.
In the filamentary-based organic memristors, some studies have been carried out to explore the effects of the material parameters on the CF growth [36,[51][52][53].As the free volume distribution in the organic switching layer governs the ion mobility, it was utilized as a factor for tuning the CF structures [36,52].As shown in figure 3(a), in the organic memristor consisting of Ag/liquid crystalline polymer/indium tin oxide (ITO), the polymer was aligned to define the free volume distribution, and the ion migration as well as the CF growth were achieved along the free volumes [52].The device with the aligned polymer layer exhibited a comparatively reliable growth of the CFs along confined paths, in contrast to the reference device lacking polymer alignment (see figure 3(b)).The void density of a polymer switching layer was also adjusted by varying the molecular weight (M w ) to facilitate the growth of the CFs in the organic memristor (see figure 3(c)) [36].For the device with the structure of Ag/poly(methyl methacrylate) (PMMA)/Ag, the switching voltages were found to be lower in the device with the higher polymer M w owing to a larger number of voids for ion migration (see figure 3(d)).Additionally, doping the switching layer with the metal ions contributes to an increase in the ion mobility of the organic memristors, leading to the more stable CFs, as shown in figure 3(e) [53].In the Ag/Li + -doped polyethylene oxide (PEO)/Pt-based memristor with high ion mobility, a stable CF for resistive switching was formed, despite the large gap (approximately 2 µm) between the electrodes.In contrast, the conventional device comprising the PEO film with low ion mobility exhibited no formation of the CF when the electrode gap exceeded 1 µm.For controlling the redox rate for the ECM in the organic memristors consisting of Cu/poly-4-vinylphenol/poly(melamine-coformaldehyde)/Pt, a Ti-buffer layer was inserted between the switching layer and the inert electrode (see figure 3(f)) [51].With the insertion of the buffer layer at the polymer/inert electrode interface, the CF growth was promoted because of the enhanced redox rate, resulting in a significant reduction in switching voltages (see figure 3(g)).

Parameters of electric stimuli
In addition to the material properties, the parameters of the electric stimuli for inducing the ECM phenomena are important in relation to the CF growth [41,54].Generally, the number, width, and the amplitude of an external electric pulse were controlled to tune the CF thickness in the organic memristors.As the number or width of electric pulses increases, the size of the CF can be enhanced through facilitated ECM processes, as shown in figure 4(a) [41,42,54].In the organic memristor with the structure of Ag/Poly(vinyl alcohol)/ITO, the thicker CFs and the lower resistance state at the LRS were achieved with an increasing number of set pulses applied to the device, as shown in figure 4(b) [41].Moreover, the voltage pulse amplitude was gradually varied to adjust the CF thickness in the device structured with Ag/Ag-Polyvinyl imidazole/Au (see figure 4(c)) [54].The resistance at the LRS of the device was effectively changed via the set pulse amplitude, as shown in figure 4(d).

Confinement of CF formation for reliable switching
For realizing the bio-inspired neuromorphic systems, it is essential to develop the crossbar-structured memristor arrays with high integration [55,56].However, in the filamentary-based memristors, the reproducibility of resistive switching is inherently poor because of the stochastic formation of the CF, thereby leading to highly restricted integration density [36,57].During the set process of the organic memristors, the injection and drift of metal ions occur across the entire area of the cell, leading to the random growth of the multiple filaments.To obtain the reproducible resistive switching behaviors in the organic memristors, several methods for confining the CF growth have been proposed [58][59][60][61][62][63][64].

Localization of ion injection
For confining the CF growth in the organic memristors, the interface between the active metal and the switching layer was engineered in order to restrict ion injection [60,61,63].In the Au/PMMA/Ag-triggering sites/ITO memristor, the metal ions were ionized only at the triggering sites during the set process, and it led to the localized CF formation (see figure 5(a)) [61].Such a device showed the considerably improved reproducibility for the CF growth, compared to the conventional device with the Ag/PMMA/ITO structure, as presented in figure 5(b).
The use of an ion blocking layer can also limit ion injection in the organic memristors [60].In the organic memristors based on Al/parylene/W, a nanoporous graphene film was inserted between the active electrode (Al) and the switching medium (parylene) as an ion-blocking layer to restrict the injection of metal ions (see figure 5(c)) [60].Compared to a typical memristor lacking an ion-blocking layer, the device with the graphene layer exhibited improved reliability and uniformity, as well as a lower level of leakage current, which was attributed to the formation of finer CFs during the set process.Recently, the 2D covalent organic framework (COF) material has been used as a switching layer of the organic memristors to limit the ion injection area of the device, as presented in figure 5(d) [63].Note that COF materials have predictable porous structures with long-range order, enabling their use for defining the ion injection area in organic memristors.In the device with a configuration of Ag/COF-5/ITO, the CFs were found to be grown along the optimized porous structures, and the reproducible resistive switching behaviors were confirmed, as shown in figure 5(e).

Definition of ion migration path
Another approach to form the CFs in the localized regions involved guiding the ion drift into confined paths within the polymer switching medium [58,59,62,64].A specific molecule, tetracyanoquinodimethane (TCNQ), known for its efficient coordination with Ag, was employed to define the Ag ion migration path and ensure reliable switching behaviors in the Ag/TCNQ-embedded polymer/ITO organic memristors (see figure 5(f)) [64].The device showed the reliable switching cycles and the low device-to-device variation.
As ion migration is promoted within the voids of the polymer medium, the distribution of free volumes in the polymer layer can be engineered to guide the ion drift along specific pathways.The use of copper phthalocyanine nanowires (N-CuMe 2 Pc NWs) with a uniform mesh structure as a switching layer induced the drift of metallic ions along the void inside the N-CuMe 2 Pc NWs film (see figure 5(g)) [59].In the device comprising Ag/N-CuMe 2 Pc NWs/ITO, the N-CuMe 2 Pc NWs film directly contributed to the localized CF growth, leading to enhanced reliability, as shown in figure 5(h).In addition, the photocrosslinkable polymer of poly vinyl(cinnamate) (PVCi) was utilized as a switching layer to predefine the free volumes for ion migration in the organic memristors [62].In the memristor consisting of Ag/PVCi/ITO, the PVCi polymer medium was photocrosslinked after the forming process, to establish the paths for ion migration (see figure 5(i)).At the photocrosslinking process of the device, the voids in the PVCi film were reduced, and the paths for ion migration were defined along the initially formed CFs.Throughout the repeated cycles of the device, the CFs were locally formed in accordance with the predefined regions, resulting in significantly enhanced reproducible performance in resistive switching compared to the reference memristor without the crosslinking process.
An electric field is a critical factor for affecting the ion drift velocity in a polymer medium.One approach involved defining the path of ion migration by locally enhancing the electric field in the organic memristors [58].For example, a cone-shaped Al electrode was introduced into the Al/poly(N-vinylcarbazole)/ITO memristor to amplify the electric field at a wedge of the electrode, as shown in figure 5(j) [58].The memristor with the cone-shaped electrode exhibited more reliable and uniform behaviors in operations, as the electrode structure for enhancing the electric field was more sharply fabricated (see figures 5(k) and (l)).In such a device, during the set process, ion migration was facilitated at the specific paths by the electrode wedge with an enhanced electric field, leading to the formation of the confined CFs.

Control of CF thickness for continuous conductance levels regulation
In biological neural networks, parallel computation is achieved through the combination of signals transmitted along neurons that are connected via numerous synapses [16,65].For biological systems, synapses possess continuous weight values indicating the strength of the relationship between neurons, and thus, it is possible to effectively solve complex problems [16,26].To achieve bio-inspired computing systems, it is crucial to mimic a continuous and analog synaptic weight in artificial synapses.In filamentary-based organic memristors, the device conductance functions as a synaptic weight, and this value is controlled by the thickness of the CFs [18].However, the CF growth in the organic memristor occurs arbitrarily and abruptly, leading to non-linear changes in the conductance value during the set process.There are researches being conducted to precisely control the CF thickness in order to attain continuous conductance levels in organic memristors [28,[66][67][68][69].
Control of the compliance current (CC) during the set process consisting of the voltage sweeps has been widely used to adjust the CF growth and the non-volatile conductance state of the memristors, as illustrated in figure 6(a) [66].However, with such a method, it is difficult to precisely adjust the device conductance due to the overshoot current problem that encourages the excessive growth of the CFs [70].To minimize the overshoot current during the set process, a gradual increase in CC was utilized in the Cu/poly(1,3,5-trivinyl-1,3,5-trimethyl cyclotrisiloxane) (pV3D3)/Al memristor [67].In this study, the CF thickness and the device conductance were adjusted by changing the reset voltage amplitude, following the formation of a thin CF through a set process involving varying CCs (figures 6(b) and (c)).
To regulate the CF thickness and resulting device conductance with minimal power consumption, the set and reset processes of the organic memristors are executed in pulse mode, rather than through voltage sweeps [18].In this case, the device conductance can be finely controlled through delicately engineered set and reset pulse conditions [71].For the Ag/reduced graphene oxide + chitosan/FTO memristors, the gradually changing pulses (from −1.2 to −1.8 V for the set process, and from 1.2 to 1.8 V for the reset process) were utilized to systematically adjust the conductance, as depicted in figure 6(d) [68].Instead of pulse engineering, an external component such as a driving transistor can also be connected to the organic memristor to adjust the CF thickness during the set process, involving constant voltage pulses [28].The Ag/poly(2,2,3,3,4,4,4-heptafluoro-butyl methacrylate)/Au memristor serially integrated with an organic transistor, was developed using the transfer printing method, as an artificial synapse with the continuous conductance levels (see figure 6(e)).In the developed synapse, the CF thickness and the conductance level of the memristor were precisely set by the channel conductance of the driving transistor, using simple voltage pulses (−2.5 V-pulse with 100 ms), as presented in figure 6(f).Recently, linearly controllable conductance levels were achieved in the Ag/poly (3,4-ethylenedioxythiophene):poly(styrenesulphonate) (PEDOT:PSS)/Au memristor by engineering the polymer conductivity [69].In this memristor, the cluster-type CF was formed according to the electric stimuli, leading to the gradual change of the conductance.(e), (f) Adapted from [28], with permission from Springer Nature.

Tuning of CF stability for synaptic plasticity
Biologically, a synaptic weight between neurons is altered through two types of synaptic plasticity: (1) short-term plasticity (STP), which takes several milliseconds to minutes, and (2) long-term plasticity (LTP), which spans from several hours to months [7].STP temporally alters the synaptic weight based on the timing between the input signals, and it leads to LTP for triggering permanent weight changes following the signal history.To achieve learning and computation processes with energy efficiency close to their biological counterparts, it is necessary to completely mimic synaptic plasticity in artificial synapses within hardware systems [16,26].For the filamentary-based organic memristors, the stability of the CF is closely linked to the memory retention characteristics of the device [36].Thus, the stability of the CF is leveraged as a crucial factor in replicating synaptic plasticity in organic memristors.
Given the dependence of CF stability on its thickness, the number and frequency of the pulses were regulated to selectively induce STP and LTP in the organic memristors during the set process based on a pulse mode [62].As the ECM phenomenon in the devices is induced by electric stimuli, the thickness of the CFs can be increased by prolonging the pulse duration or reducing the pulse interval [42,54].In the organic memristors with the structures of as Ag/PEDOT: PSS/Ta (refer to figure 7(a)) [72], STP and LTP were sequentially represented according to the number of pulses (see figure 7(b)).Furthermore, the Cu/CNT-doped honey/ITO memristor demonstrated STP or LTP characteristics in response to pulse intervals, as illustrated in figure 7(c) [73].As the memristors were exposed to increased electrical stimuli, the CF thickened and stabilized, leading to transition of the device memory characteristics from STP to LTP.Moreover, a shorter pulse interval results in a thicker and more stable CF, attributed to a reduced duration for the diffusion of metal atoms from the filament.However, in these devices, STP and LTP were induced dependently, making it challenging to emulate history-dependent learning processes, such as spike timing-dependent plasticity (STDP) and spike rate-dependent plasticity (SRDP), without employing complex pulse engineering processes.
To obtain independent STP and LTP, the multi-stacking structured memristors, comprising two distinct parts dedicated to STP and LTP, were proposed, as shown in figure 7(d) [39,74].In such memristors, unstable CFs are formed in the STP part in response to the electric stimulus, allowing the subsequent stimulus to be applied to the LTP part when it propagates along the STP part before the CFs rupture.For controlling the CF stability of the memristor, the M w of the polymer switching layer was adjusted as a key parameter for the CF diffusion (see figure 7(e)) [74].In the Ag/PMMA/ITO memristors, the CF diffusion was observed to increase with decreasing polymer M w , attributed to the rise in free volume density.The Ag/PMMA (with low M w )/Ag/PMMA (with high M w )/ITO memristor exhibited independent STP and LTP characteristics, allowing the implementation of pulse-based learning rules, including STDP and SRDP, under the simple electrical stimuli (see figure 7(f)).Recently, the ion injection density was manipulated as a diffusive parameter of the CF in organic memristors [39].In this study, the memristor comprising the STP and LTP parts was effectively developed by systematically engineering the ion injection amount and the resultant CF stability in the devices (refer to figure 7(g)).The developed memristor, configured as Ag/PVCi/Au/Ag nanoparticles/PVCi/Ag nanoparticles/ITO, demonstrated independent STP and LTP properties, showcasing bio-mimetic learning capabilities for SRDP and STDP within the device (see figures 7(h) and (i)).

Logic operators
Boolean logic operators are basic applications that utilize multiple memristors to showcase the system's capability for parallel computation [75,76].Flexible operators for 'AND' and 'OR' logics have been implemented using the two organic memristor cells [51].As these systems mostly relied on offline learning, only the LTP characteristics (excluding STP) were necessary in the memristors.Furthermore, only two or three conductance levels were required to establish stable systems, which were easily constructed using the typical organic memristor cells acting as non-volatile memory components.
In a system composed of the two Cu/crosslinked poly-4-vinylphenol/Pt memristors and a single sensing resistor, the conductance states of the memristors were set by the logic input voltages (5 V or 2 V for '1' and 0 V for '0' logic inputs), which determined an output logic voltage at the sensing resistor (2 V for '1' and 0 V for '0' logic outputs), as shown in figure 8(a) [51].Additionally, the flexible 2 × 1 array of the organic memristors composed of Ag/PVCi/PEDOT: PSS was used for parallel computation of 'AND' and 'OR' logics (see figure 8(b)) [76].In this parallel system, each memristor cell's conductance was trained to an optimized value, and the bit line current level was examined to determine an output logic value when logic input voltages (0.5 V for '1' and 0.05 V for '0') were applied to the word lines.Both the systems operated stably as logic operators.Notably, the flexible memristor array consumed approximately 6.82 fJ and 17.90 fJ for the computation of 'AND' and 'OR' logics, respectively, showcasing significant superiority compared to conventional CMOS technology [17].

Image recognition systems
An image recognition system is a key application that showcases the potential of bio-inspired neuromorphic systems [77].This system is constructed following the offline learning rules, and the LTP characteristics for the multilevel conductance levels are required in an artificial synapse [67].Generally, the synapse cells in the crossbar arrays can be set by the V dd /2 scheme [78].In this set process, the selected bit and word lines are biased with a set voltage (V set ), and other lines are connected to V set /2 for avoiding the write disturbance from the sneak currents.However, it is still challenging to develop highly integrated organic memristor arrays due to the poor reliability and uniformity of the device cell [58][59][60][61][62][63][64].As a result, only numerical simulation methods have been used to verify the potential of organic memristors [38,67].In such simulations, the software-based neural networks for image recognition were first constructed, and the ideal synaptic weight distributions were calculated after training to obtain high recognition accuracy.Subsequently, the synaptic weights for the ideal networks were quantized to the memristor conductance values.The recognition accuracy of the system, utilizing the distribution of device conductance, was evaluated by measuring the reading current value for each bit line.This assessment considered that the word and bit lines were connected to input voltage signals and ground, respectively.Given that the recognition accuracy of the hardware system is primarily governed by the number of conductance states in the memristors, it is important to attain the continuous conductance levels and LTP properties in the devices.The flexible hardware neural networks, consisting of Ag/PMMA/ITO and Cu/pV3D3/Al, demonstrated high potential with the accuracy of about 90% in recognizing handwritten digit and face images, respectively (see figures 8(c) and (d)) [38,67].

Combinatorial optimization systems
To develop complex neural networks capable of solving combinatorial optimization problems, not only independent STP and LTP but also the homeostatic plasticity of artificial synapses must be realized [39].Specifically, to construct stochastic.Hopfield neural networks, suitable for combinatorial optimization, the transient decaying noise should be represented in the synapse.To emulate the transient noise attributed to homeostatic plasticity in organic memristors, the multi-stacked structured device consisting of the LTP and STP parts (Ag/PVCi/Au/Ag nanoparticles/PVCi/Ag nanoparticles/ITO), referred to in section 3.3, was used [39].In this memristor, the STP part has lower conductance values at the HRS and LRS than those of the LTP part, and a transient noise current during the reading process can be induced by the volatile conductance of the STP part.The 6 × 6 crossbar arrays of the flexible memristors were constructed and trained to function as the hardware neural networks for solving the graph max-cut problem (see figure 8(e)).The system showed reliable performance in solving the problem consisting of six different nodes, as shown in figure 8(f).

Electronic skins
Artificial peripheral nervous systems represent crucial applications of artificial synapses in achieving bio-realistic robotic systems and smart prosthetics [3].For the construction of peripheral nerves, such as electronic skins, achieving mechanical stretchability and biocompatibility in the memristors acting as nociceptors is crucial.Several studies have been conducted to develop electronic skins using the filamentary-based organic memristors, which exhibit biocompatible and mechanically flexible characteristics [40,79].An organic memristor, composed of carboxymethyl ι-carrageenan (CιC), was used as a nociceptor in the electronic skin system [79].The Ag/CιC/ITO structured resistive switching device was fabricated on a flexible poly(ethylene terephthalate) substrate, as illustrated in figures 9(a) and (b).When the device was interfaced with a conventional pressure sensor, it effectively emulated the functions of a pressure nociceptor system in electronic skins, as depicted in figures 9(c) and (d).In addition, a flexible memristor with a structure of Ag/a copolymer of chlorotrifluoroethylene and vinylidene fluoride/Pt was incorporated into a system comprising a triboelectric generator and a light-emitting diode, demonstrating the development of an artificial healable skin, as shown in figure 9(e) [40].This system effectively demonstrated injury response behaviors under various scenarios, where parameters such as pain duration, frequency, and intensity were varied (see figure 9(f)).While the organic memristors in these artificial skins demonstrated mechanical flexibility, the intrinsic stretchable and flexible features, akin to their biological counterparts, were not attained due to the rigidity of the electrodes.

Conclusion
In this paper, the researches on developing flexible artificial synapses utilizing filamentary-based organic memristors have been reviewed, from single-device studies to practical system implementations.Fundamental background information regarding the operating principles of the devices, as well as the various parameters governing CF growth, was discussed.The material properties of the switching layer and the input signal conditions, which are related to the redox reactions and ion drift for the ECM phenomena of the organic memristors, were reported as critical parameters influencing the CF dynamics and structures.
To mimic the biological synaptic functions such as the highly reproducible characteristics for high integration, continuous weight levels, and synaptic plasticity in the organic memristors, considerable effort has been made to delicately engineer the CF growth.However, it is still difficult to achieve the filamentary-based organic memristors with complete synaptic functions for practical wearable applications, and only simple or simulated systems have been constructed to demonstrate the device's potential.To realize the bio-inspired computing systems, suitable for wearable smart electronics, the more promising and facile strategy for developing the reproducible CF growth in the memristors is first required.Additionally, the continuous conductance levels should be attained in a single memristor without any external components for the complex pulse engineering.Ultimately, both strategies for achieving reliable and uniform resistive switching and analog conductance states should be applied to the device concept for independent STP and LTP, such as the multi-stacked structured organic memristors, simultaneously.Moreover, it is imperative to acquire intrinsic flexible and stretchable characteristics in organic memristors to advance the development of bio-realistic nervous systems, such as artificial skins.

Figure 1 .
Figure 1.Replication of synaptic functions in the filamentary-based organic memristors for developing bio-inspired computing systems in wearable electronics.

Figure 2 .
Figure 2. (a) Current-voltage curves presenting an operating mechanism of filamentary-based organic memristors.Reproduced from [42].© IOP Publishing Ltd.All rights reserved.(b) Effects of the redox rate and the ion mobility on the conductive filament growth of the memristors.Adapted from [44], with permission from Springer Nature.

Figure 3 .
Figure 3. (a) A concept of metal ion migration according to the polymer alignment in the organic memristors consisting of a liquid crystalline polymer (LCP) switching layer.(b) The structures of the conductive filaments (CFs) after the set process in the LCP based memristors without and with polymer alignment.(a), (b) Reprinted from [52], Copyright (2020), with permission from Elsevier.(c) Schematics showing the effects of the polymer molecular weight (Mw) on the ion migration in the organic memristors.(d) Resistive switching behaviors of the organic memristors with different polymer Mws.(c), (d) [36] John Wiley & Sons.© 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim.(e) Current-voltage curves (I-V) and the CF growth at the active area of the organic memristor without and with Li + doping.Reprinted (adapted) with permission from [53].Copyright (2022) American Chemical Society.(f) Device structures consisting of Cu/poly-4-vinylphenol/poly(melamine-coformaldehyde)/Pt and Cu/Ti/poly-4-vinylphenol/poly(melamine-co-formaldehyde)/Pt.(g) I-V curves of the devices.(f), (g) Reprinted (adapted) with permission from [51].Copyright (2017) American Chemical Society.

Figure 4 .
Figure 4. (a) Schematics presenting the conductive filament (CF) growth according to the number of the electric stimuli in the organic memristor.(b) Changes of the device conductance according to the number of the voltage pulses.(a), (b) [41] John Wiley & Sons.© 2023 The Authors.Advanced Intelligent Systems published by Wiley-VCH GmbH.Influences of the pulse amplitude on (c) the CF growth and (d) the conductance increase in the organic memristor.(c), (d) [54] John Wiley & Sons.© 2022 Wiley-VCH GmbH.

Figure 5 .
Figure 5.Studies for confining the conductive filament (CF) formation in the organic memristors.(a) A concept for localizing the ion injection in the organic memristor utilizing the interfacial triggering sites (ITSs) of Ag.(b) The confined CF growth in the device with ITSs, observed using the conductive atomic force microscopy.(a), (b) Reprinted (adapted) with permission from [61].Copyright (2019) American Chemical Society.(c) Schematics presenting the confined ion injection in the organic memristors with a graphene barrier.[60] John Wiley & Sons.© 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim.(d) An organic memristor structure with the 2D covalent organic framework for the refining the ion injection.(e) Reliable resistive switching behaviors in the device.(d), (e) Adapted from [63] with permission from the Royal Society of Chemistry.(f) The reproducible resistive switching characteristics in the organic memristor composed of tetracyanoquinodimethane embedded polymer for confining the ion migration.[64] John Wiley & Sons.© 2023 Wiley-VCH GmbH.(g) A concept for defining the ion migration path in the organic memristor introducing the copper phthalocyanine nanowires.(h) Reproducible resistive switching characteristics during the repeated cycles.(g), (h) [59] John Wiley & Sons.© 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim.(i) Schematics showing a method of predefining the ion migration path in the organic memristors based on the switching layer of photocrosslinkable polymer.Adapted from [62] with permission from the Royal Society of Chemistry.(j) An organic memristor structure with the cone-shaped electrode for enhancing the local electric field.(k) The reliable resistive switching characteristics of the device under the pulse modes.(l) The cell-to-cell uniformity of the device.(j)-(l) [58] John Wiley & Sons.© 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim.

Figure 6 .
Figure 6.(a) A control of the compliance current (CC) levels at the set process consisting of voltage sweeps for obtaining the multilevel conductance states in the organic memristors.[66] John Wiley & Sons.© 2018 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim.(b) Schematics illustrating a strategy for controlling the conductive filament thickness of the organic memristors at the reset process.(c) A current-voltage curve for the organic memristor at the gradual reset process.(b), (c) Reprinted (adapted) with permission from [67].Copyright (2019) American Chemical Society.(d) Changes of the organic memristor conductance via the gradual varying set and reset pulses for linearly controlling the device conductance.[68] John Wiley & Sons.© 2021 Wiley-VCH GmbH.(e) The artificial synapse composed of one organic transistor and a single organic memristor (1 T-1 M).(f) A systematic control of the 1 T-1 M cell conductance via a gate voltage of the organic driving transistor.(e),(f) Adapted from[28], with permission from Springer Nature.

Figure 7 .
Figure 7. (a) The Ag/poly(3,4-ethylenedioxythiophene):poly(styrenesulphonate)/Ta based memristor for emulating synaptic plasticity.(b) The change of the conductance stability in the organic memristor according to the number of electric stimuli.(a), (b) Adapted from [72] with permission from the Royal Society of Chemistry.(c) The short-term plasticity (STP) to long-term plasticity (LTP) transition in the organic memristor comprising Cu/CNT doped honey/ITO.[73] John Wiley & Sons.© 2023 Wiley-VCH GmbH.(d) The multi-stacking structured memristors, comprising two distinct parts with different polymer molecular weights.(e) Spike rate-dependent plasticity (SRDP) characteristics implemented in the device.(d), (e) Reprinted (adapted) with permission from [74].Copyright (2020) American Chemical Society.(f) The flexible artificial synapse based on the organic memristor consisting of the computation and memory parts with the different ion injection density.(g) SRDP and (h) Spike timing-dependent plasticity (STDP) represented in the device.(f)-(h) [39] John Wiley & Sons.© 2023 The Authors.Advanced Science published by Wiley-VCH GmbH.

Figure 9 .
Figure 9. (a) Flexible and (b) resistive switching features of the memristor consisting of Ag/carboxymethyl ι-carrageenan/ITO.(c) A schematic presenting a pressure nociceptor system in electronic skins.(d) The characteristics of an artificial nociceptor.(a)-(d) Adapted from [79] with permission from the Royal Society of Chemistry.(e) A structure of an artificial healable skin involving the Ag/a copolymer of chlorotrifluoroethylene and vinylidene fluoride/Pt memristors.(f) Injury response behaviors of the artificial skin under various scenarios.(e), (f) [40] John Wiley & Sons.© 2022 The Authors.Advanced Science published by Wiley-VCH GmbH.