Recent advances in fabrication and functions of neuromorphic system based on organic field effect transistor

The development of various artificial electronics and machines would explosively increase the amount of information and data, which need to be processed via in-situ remediation. Bioinspired synapse devices can store and process signals in a parallel way, thus improving fault tolerance and decreasing the power consumption of artificial systems. The organic field effect transistor (OFET) is a promising component for bioinspired neuromorphic systems because it is suitable for large-scale integrated circuits and flexible devices. In this review, the organic semiconductor materials, structures and fabrication, and different artificial sensory perception systems functions based on neuromorphic OFET devices are summarized. Subsequently, a summary and challenges of neuromorphic OFET devices are provided. This review presents a detailed introduction to the recent progress of neuromorphic OFET devices from semiconductor materials to perception systems, which would serve as a reference for the development of neuromorphic systems in future bioinspired electronics.

Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence.Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.

Introduction
The development of artificial intelligence (AI) and internet of things technologies has driven substantial progress in the fields of human-computer interaction interfaces, artificial perception systems, machines, and prosthetics, which has also led to an explosive increase in the information and data that need to be processed in-situ remediation.The traditional digital system based on Von Neumann architecture faces great challenges due to its reliance on centralized processing and sequential operations for information processing, which will inevitably lead to a series of problems of slow processing speed and high-power consumption in the current system [1][2][3].It is worth noting that the biological synapse, as the basis of interaction between organisms and the external environment, can store and learn rapidly while processing external signals in parallel [4,5].Therefore, the simulation of biological synapses and preparation of neuromorphic systems can simulate the biological data processing process at the synapse unit level, and then parallel processing distributed information, improve fault tolerance and efficiency, and decrease the power consumption of the system [6][7][8].Recently, the devices mimic biological synapses composed of two main architectures, such as two-terminal memristors and three-terminal transistors [9][10][11][12][13][14][15][16][17][18][19].Compared with the two-terminal synaptic devices, the three-terminal synaptic transistors have the advantages of independent control of pre-synaptic electrodes and post-synaptic electrodes, excellent stability, and materials diversity [11-13, 15, 16].Meanwhile, with appropriate materials selection and structural design, transistors can convert external stimuli (light, pressure, temperature, etc) into electrical signals, providing the possibility for artificial synapses to directly perceive external environments [20][21][22][23][24].In addition, the collaborative control of devices can be easily achieved in transistor synaptic-based circuits, which indicates that stable neural networks can be developed using neural units [25,26].More importantly, signal transmission and self-learning can be performed simultaneously in transistor-based three-terminal/multi-terminal artificial contact [13,27].Therefore, transistors are better suited than other types of devices to simulate synaptic functions and neuromorphic systems, especially to simulate parallel learning and dendrite integration that requires multi-terminal operations.
Noticeably, organic semiconductor materials have great advantages due to their easy processing, pollution-free, and suitability for large-area preparation, making them suitable for neuromorphic field effect transistors and applications [28][29][30][31][32][33].Many natural organic materials, with the advantages of renewability, sustainability, biodegradability, and environmentally friendliness, are investigated [34][35][36][37].Moreover, organic materials can adjust their optoelectronic properties through the design or modification of organic molecules, which are suitable for sensing and responding to light.Furthermore, the organic field effect transistor (OFET) can be used for flexible electronic devices by solution-processable fabrication techniques, such as roll-to-roll coating, screen printing, and inkjet printing [38][39][40].Therefore, the OFET device is a promising component for future neuromorphic systems.
In this review, the recent progress in neuromorphic systems based on OFET devices is summarized.Section 2 introduces the different organic materials in neuromorphic OFET devices, such as single p-type, single n-type, and heterojunction materials.Section 3 presents the different structures and working mechanisms based on neuromorphic OFET devices.The basic working mechanisms such as the electric double layer (EDL) effect, ferroelectric deflection effect, and floating gate trapping effect are summarized.Then, section 4 briefly reviewed the different artificial perception-based neuromorphic OFET devices, as shown in figure 1, which include tactile, visual, auditory, and other perception systems.Finally, the summary and the challenges of neuromorphic OFET devices are provided.

Organic semiconductor materials for neuromorphic OFET devices
Neuromorphic devices are functional devices that mimic the synapses in a biological brain, and the behavior of neuromorphic devices can be adjusted to achieve the function of learning, memory, and adaptation.Noticeably, the conductance of organic semiconductor materials can be modulated by charge transfer or redox, and thus, the neuromorphic OFET devices using organic semiconductor materials as channel layers have been reported.And the output performance of neuromorphic device is determined by the organic semiconductor materials.Meanwhile, the memory and computing performance are always influenced by the semiconductor layer.Thus, it is important to investigate different semiconductor materials for neuromorphic OFET devices.In this section, we will briefly introduce semiconductor materials in neuromorphic OFET devices, such as single semiconductor materials (P-type and N-type) and heterojunction semiconductor materials (bulk heterojunction and layer heterojunction), and then the semiconductor materials, synaptic functions, flexibility and stimuli of different neuromorphic OFET devices are finally summarized in table 1.
N-type organic semiconductor materials have also been investigated for neuromorphic OFET devices, and n-type OFET devices are essential for the all-organic complementary circuits [59][60][61].Xie et al reported an n-type neuromorphic OFET device with the all-solid-state vertical structure, while the PEO/LiClO4 material functioned as the gate dielectric layer and naphthylene-1,4,5,8-tetracarboxylic-diimidethiophene-vinyl-thiophene (NDI-gTVT) as the organic semiconductor layer, whose schematic structure is illustrated as figure 2(b) [60].The n-type material in this work was a donoracceptor conjugated material, which possesses the advantages of highly operated stability in electrolyte media, low operating voltage, and high conductivity.Noticeably, the channel conductance in this neuromorphic OFET device was increased or decreased by the coordinating of Li + and NDI-gTVT, because of the electronegativity differences between the side-chain of ethylene glycol (EG) and the main-chain of NDI-gTVT.The PEO can separate Li + and be induced to NDI-gTVT when a positive voltage arrives at the gate electrode, and then the conductance of the channel layer can be increased with the redox effect.Finally, the functions of a synapse can be well mimicked, such as excitatory post-synaptic current (EPSC), paired-pulse facilitation (PPF), short-term plasticity (STP), long-term plasticity (LTP) and basic functions, and the minimum power consumption of this device can be decreased to 6.16 pJ.Moreover, the recognition accuracy based on this novel n-type device achieved 94%, and the high accuracy lies in the different electronegativity between EG and NDI-gTVT and the reversibly Li+ doping mechanism.

Heterojunction semiconductor materials in OFET
Recently, the neuromorphic OFET device with heterostructure semiconductor materials has emerged, while single semiconductor material neuromorphic OFET devices always need a charge-storage layer or other functional layers between the insulating layer and channel layer.Noticeably, the heterostructure semiconductor materials always possess a heterointerface that promotes the separation of excitons [62][63][64][65][66][67][68][69][70][71][72][73][74].Gao et al reported a P-type and N-type bulk heterostructure as the channel layer of the OFET device (figure 3(a)), which showed the different functions with different mixed ratios of N-type semiconductor to p-type materials [62].Interestingly, with the ratio increased, the OFET device functioned from nonvolatile memory to synapse, as shown in figure 3(c).The low ratio of n-type materials cannot form a continuous network in p-type materials, thus, the discontinuous N materials can be considered as the quantum well to capture carries and make the device work as a memory.Then, with the ratio of N-type semiconductor materials increased, the continuous network would offer a charge transmission path way and resulting in the current leakage.Hence, the P/N layer heterostructure device can transform its function from memory to synaptic.Meanwhile, Gao et al also reported another bulk heterostructures neuromorphic OFET device, in which the channel layers are MXene-TiO 2 complexes and organic semiconductors [63].The power consumption and time were reduced by 96% and 95% in retraining this heterostructures device.Furthermore, Wang et al investigated a simple and effective method that fabricated an inorganic perovskite quantum dots (QDs) and organic semiconductor bulk heterostructure light-stimulated neuromorphic OFET device [67].This device can respond to light sensitively and can realize several synaptic functions, which also can be easily applied to large-scale applications.
Additionally, several workers reported the neuromorphic OFET device with layer heterostructure semiconductors, for example, Liu et al depicted a synaptic transistor with the semiconductor material of 4H-SiC and organic semiconductors for neuromorphic ultraviolet (UV) vision [68].In this device, the n-type 4H-SiC semiconductor material was selected as the substrate and light-sensitive layer of OFET, while the p-type P3HT organic material was used as the channel layer.With the design of layer heterostructure and the effect of 4H-SiC photogate, the device achieved non-volatility synaptic functions.Meanwhile, Wang et al first reported a novel two-dimensional (2D) organic material perylene-3,4,9,10tetracarboxylic dianhydride (PTCDA) and inorganic material molybdenum disulfide (MoS 2 ) hybrid layer heterostructure OFET device with the neuromorphic functions of photoelectric dual modulation [71].Notably, this dual-modulation neuromorphic OFET device working in electric mode needs a four-terminal configuration, as illustrated in figure 3(d).The top Au gate electrode acted as the control electrode to drive the separation and capture or gradual release of charges at the heterostructure interface, and then led to the electric synaptic functions (EPSC/PPF, figure 3(e)).This device works in the optical mode and just needs a three-terminal configuration (figure 3(f)), which does not need another top gate electrode as the control electrode.In this mode, a laser pulse was selected as the control terminal, and the optical signal as input signal to simulate different synaptic functions.Moreover, the synaptic weight changes and PPF ratios are all superior to previously reported works, which also depicted that this device opens up a new path for future artificial neuromorphic system design.

The structures of neuromorphic OFET devices
Notably, neuromorphic device requires the function of memory and learning, thus, the dielectric layers with their intrinsic property are crucial for OFET devices to mimic brain-liked synaptic function.Dielectric layers can affect the functions of OFETs and their neuromorphic functions due to their intrinsic ability, which can determine the operation voltage of devices and lead to significant energy consumption.Thus, different dielectric layers of neuromorphic OFETs are reviewed in this section, and we mainly reviewed three main types of neuromorphic OFET devices, such as electrolyte gate devices, ferroelectric devices and floating gate devices.Finally, the structures, gate materials, preparation methods, device sizes, synaptic functions, and flexibility of different structures of neuromorphic OFET devices are summarized in table 2. Compared with traditional OFET devices, these devices can mimic synaptic function with different working mechanisms.

Electrolyte gate neuromorphic OFET device
The OFET devices with ionic electrolytes have attracted great attention because of their large specific capacitance, low operating voltage, and sensitive interfacial properties, which are suitable for bioinspired neuromorphic computing devices [52,53,57,[75][76][77][78][79].Electrolyte gate OFET devices always show two different working mechanisms, such as the EDL effect and electrochemical reaction, and the difference between these two models is whether the ions from electrolyte penetrate into the semiconductor layers.
For the EDL neuromorphic OFET devices, when a positive voltage was applied to the gate electrode, the cations and anions in the electrolyte moved to the electrolyte/channel and gate electrode/electrolyte interfaces, respectively, and then formed the EDL effect.Gkoupidenis et al presented the simple device with a PEDOT: PSS film as a channel layer and a KCl as an electrolyte, which is confined in a polydimethylsiloxane (PDMS) elastomer well, as depicted in figure 4(a) [52].Noticeably, a glass substrate with a photolithographically patterned Au electrode was coated with a PEDOT: PSS film to form the device channel, and then resulting in an active area with dimensions of 2.4 mm × 0.25 mm, which is different from the typical EDL device structure.This device can realize the basic neuromorphic functions such as shortterm synaptic plasticity functions in a simple single device level for the first time.Similar to the traditional synaptic transistor, the pre-synaptic pulse in this device was gate voltage, and the post-synaptic current was PEDOT: PSS channel current.Meanwhile, the function of adaptation when responding to the external signal stimuli can be measured by continued 300 pulses, while this phenomenon was achieved with the incompletion recovery of ions into the electrolyte and the ion migration-relaxation dynamics.Moreover, Gkoupidenis et al also reported another electrolyte multi-gate neuromorphic OFET device, which achieved the function of orientation selectivity and is similar to the visual system in a single unit [53].The multi-gate device was different from existing systems that did not need extra optical signals, while the orientation function was realized through mapping the grids or arrays of this multi-gate device.The different input signals can be presented by the superimposing voltage which is applied to more than one gate electrode.Furthermore, Qian et al also investigated a multi-gate electrolyte gate neuromorphic OFET device for spatiotemporal information processing, and temporal correlation of output signal is dependent on the distance between the gate electrode and the semiconductor layer [76].
In addition, Dai et al recently reported a device with the wood-derived cellulose nanopaper as an ionic conductivity dielectric material [57].The cellulose's specific characteristic of the main chains is that they entangled with each other so that the anionic cannot be movable optionally.Thus, only the sodium ions and protons can be moved with the gate voltage and then produce EPSC.Meanwhile, the spatial summation and high-pass filter function can be successfully simulated in this device, which has great potential for future neuromorphic computing.Noticeably, Xu et al demonstrated an electrolyte neuromorphic device with a core-sheath nanowire as a channel layer [79].The device's configuration was composed of a conducting line probe, ion-gel, organic nanowire, and metal contacts while the conducting line probe acted as the gate electrode.The organic nanowire was composed of PEO as the sheath and P3HT as the inner core.When a pre-synaptic pulse was applied to the gate electrode, the ions were migrated to the PEO sheath or even to the P3HT core, and then the STP and LTP could be induced with the ion migration-relaxation dynamics.Besides, benefiting from the large surface to volume ratio of organic nanowires and short channel length, the power consumption decreased to 1.23 fJ per spike for this device, which was lower than reported synaptic devices.Notably, the low consumption is similar to the biological synapse, which was suitable for the development of bioinspired electronic devices.

Ferroelectric neuromorphic OFET device
Ferroelectric thin film materialis also a promising candidate dielectric layer for the neuromorphic OFET device because of its distinctive ferroelectric associated resistive switching effect and nonvolatile polarization [80][81][82][83][84][85].Meanwhile, the unique polarization plasticity of ferroelectric material is similar to the brain-like synaptic function, and Poly(vinylidene fluoride) (PVDF) is the most used ferroelectric material in OFET devices, which possess excellent ferroelectricity and thermal stability and is suitable for neuromorphic OFET device [86].In addition, other ferroelectric materials also attracted reporters' attention, such as Pb(Zr, Ti)O 3 (PZT) [87][88][89], BaTiO3 (BTO) [90,91], HfxZr1-xO2 [92][93][94], and BiFeO3 [95] ferroelectric materials.Li et al illustrated a neuromorphic OFET device with the PVDF-TrFE dielectric layer, and this OFET device was demonstrated with a vertical configuration and a nanoscale channel (figure 5(a)) [96].In this device, PDVT-10 organic material was selected as the semiconductor layer, and the synaptic function was achieved from the property of spontaneous and delicate ferroelectric polarization control.Meanwhile, different from traditional planner OFET devices, the source electrode, semiconductor layer, and drain electrode of this ferroelectric OFET device were vertically stacked.The working mechanism of this vertical ferroelectric OFET device is shown in figure 5(b).When a negative voltage was applied to the gate electrode, the polarization of PVDF-TrFE could be tuned, and then the Schottky barrier between the networksource electrode and channel layer was decreased, simultaneously.As a result, the carriers in the channel can be controlled by the polarization direction and the number of gate pulses.In contrast, when a positive voltage was applied to the gate electrode, the carriers in the channel and the conductivity decreased because of the changed polarization and higher Schottky barrier.Noticeably, the OFET device with a ferroelectric thin film device showed a linear conductance variation, which was suitable for pattern recognition.The recognition accuracy reached 91.38%, which was the record-high recognition accuracy of OFET synaptic at the time.Meanwhile, the OFET devices with 2D semiconductors and ferroelectric thin film have reported that the carrier concentration in 2D materials can be effectively modulated by ferroelectric.Tian et al investigated an OFET synaptic device with 2D MoS 2 as the channel layer, the channel conductance can be adjusted by altering the gate voltage [97].The basic synaptic functions can be well demonstrated with this 2D OFET synaptic device, such as LTP/LDP, and Hebbian spike-timing dependent plasticity (STDP) learning.
Furthermore, the flexible neuromorphic OFET device is considered important for the future E-skin and humanmachine interaction.Then Jang et al reported an ultrathin freestanding substrate PVDF-TrFE based OFET synaptic device for wearable applications, as illustrated in figure 6(a) [98].The freestanding OFET device was achieved by mechanically peeling off the prepared device from the rigid Si substrate, and the device was attached to the PET flexible substrate sequentially.Similarly, the charge carried in the pentacene layer was regulated by the polarization changing of the PVDF-TrFE, and the polarization was also changed by the voltage applied to the gate electrode.The output current of this OFET can not only mimic the STDP biological functions well, but also work at extremely small bending states with excellent performance.Figure 6(b) shows the device was transferred onto the brain-shaped PDMS mold and folded into a small bending radius of 50 µm, which was aimed at investigating the adaptive ability with complex surroundings and bioinspired environment.This device can maintain the performance under these conditions, as depicted in figure 6(c), the folded OFET device shows stable output properties over 6000 gate voltage pulses.This free-standing ferroelectric OFET device had great potential for new-generation wearable bioinspired device applications.

Floating gate neuromorphic OFET device
Floating gate neuromorphic OFET devices always have a control gate electrode embedded in the insulating layer to capture the charges from the semiconductor layer.Noticeably, the charges were trapped in a floating gate which cannot be easily escaped and led to excellent retention characteristics.Meanwhile, the channel conductivity and trapped charges can be regulated by the gate voltage [56,[99][100][101][102][103][104][105][106][107].For example, in 2018, Ren et al used C60 molecular nanomaterials as a floating-gate neuromorphic OFET device, and the morphology controlled C60 was dispersed in the PMMA insulating layers, enabling the device to show bidirectional threshold voltage shift and learning abilities like synapses [56].The schematic illustration of this bottom-gate top-contact OFET device is depicted in figure 7(a), which is composed of a pentacene organic semiconductor layer, and this device was integrated on a flexible PET substrate.Noticeably, the amount of charge carried in the floating gate was determined by the  amplitude and the duration time of applied gate voltage pulses and the program temperatures.Meanwhile, as the number of pulses increased, more electrons were trapped into the floating gate and similar to the action potential of synapse, which also demonstrated that this OFET device processes significant charge trapping ability.Moreover, QDs also have great potential for floating gate devices owing to their high charge carrier mobility and nanoparticle dimension.Li et al investigated a strain neuromorphic OFET device and inorganic CsPbBr3 QDs functioned as the floating gate [101].The channel carrier density was determined by the trapped ability of CsPbBr3 QDs and gate voltage, meanwhile, the CsPbBr3 QDs also had excellent absorption properties, and thus this device also showed significant synaptic function with different UV radiations.This device also obtained higher recognition accuracy under UV radiations than under dark states, which is owing to the different mechanisms with two different states.Moreover, a neuromorphic OFET device was also demonstrated with silver nanowires (AgNWs) as floating gate and exhibited a high transparency rather than 81%, which was suitable for working with transparent electronic devices [103].Meanwhile, the device can transform from a memory device into a synaptic device by changing the concentration of AgNWs, which is also enabled for neuromorphic OFET device applications.
Zheng et al designed an innovative configuration device that combined floating gate and EDL effect, and investigated a planar floating gate electric-double-layer transistor (EDLT) device [105].The schematic illustration of this OFET geometry is shown in figure 7(d).Similar to conventional EDLT devices, chitosan was stacked with the gate electrode, while the floating gate layer was different from the traditional structure, which is demonstrated between the gate electrode and blocking layer.Thus, the ions transport of ion-gel caused by the EDL effect and gate voltage would cause the tunneling effect of the floating gate, in general, the channel layer would induce floating gate programming in this device.Meanwhile, under the influence of EDL and floating gate, the effect of gate voltage on electrolyte would decrease and extend the saturation of conductance modulation and further improve the weight update characteristics.Additionally, the flexible device was fabricated based on a Kapton substrate, and the LTP/LDP curve of this flexible floating-gate EDLT device under a 75 • bending angle also depicted excellent performance.Finally, this device exhibited a high recognition accuracy of 87.8% after 100 training epochs.

Perception systems based on neuromorphic OFET devices
OFET-device-based perception systems can open up new opportunities for future intelligent artificial applications and human-machine interaction interface development.It is notable that the efficient construction of intelligent electronics and mimic biological functions need integrated sensing and signal processing functions, and perception such as tactile, visual, auditory, and other sensing functions are highly desired to simulate [108][109][110].In this section, the different perception systems based on neuromorphic OFET devices will be summarized.

Tactile perception system based on neuromorphic OFET devices
Tactile perception, as the basic biological perception system, always consists of tactile receptors to sense external signals and synapses to process transferred signals.Correspondingly, artificial tactile perception systems based on neuromorphic OFET devices need to integrate tactile sensors and OFET synaptic devices [110][111][112][113]. Zang et al reported a neuromorphic OFET-based tactile perception system, and the system was composed of a dual-organic-transistor-based tactileperception element (DOT-TPE) [110].The schematic illustration of this DOT-TPE system is shown in figure 8(a), and the equivalent circuit of this tactile perception system is also illustrated in figure 8(a).This DOT-TPE was composed of an OFET-based tactile perception sensor and a neuromorphic OFET device, which enabled to simulate human beings.When different pressure was applied to the OFET sensor, the space between the gate and the insulating layer was deformed and then the capacitance of dielectric layer was changed, simultaneously.Correspondingly, the delivered pre-synaptic spikes from pressure sensor were changed, leading to the channel conductance change of the neuromorphic OFET device.Moreover, based on this DOT-TPE system, the function of high-pass temporal filtering function can be well mimicked.With the touch speed increased, the position resolving and code ability can be achieved, which has great potential for future human-machine interaction.Meanwhile, with the tactile perception time increased from 1 s to 5 s, a sustained-timedependent sensing capability and a short responding time of this system can be recorded.Furthermore, Kim et al investigated an artificial afferent nerve to emulate the biologically slowly adapting type I sensory neurons [113].The pressure sensors transfer external pressure changes into voltage changes, and then the voltage is transferred into voltage pulses through an organic ring oscillator, and at last, the postsynaptic currents can be recorded by the synaptic transistor.It is noteworthy that the synaptic transistor can process multiple signals from pressure sensors and ring oscillators because of the specific functions of the ion-gels dielectric layer.Several functions such as sum operation, directions distinction, and braille characters identification can be achieved, which also had great significance for information encoding.Finally, a hybrid reflex arc was demonstrated on a discoid cockroach to emulate a biological reflex arc.Actuation of the tibial extensor muscle can be recorded with multiple external pressures applied to the hybrid reflex arc, and the activated muscle fibers would be affected by the amplitude and frequency of pressure signals.
Recently, triboelectric nanogenerators (TENGs) have attracted great attention, which can convert mechanical energy into electronic signals, and TENGs are also suitable for sensing units in tactile neuromorphic OFET devices [42-44, 112, 114-119].Liu et al reported a high-sensitivity self-powered tactile perception system with a TENG as a tactile receptor and a neuromorphic OFET synaptic device as a signal processing unit, as shown in figure 8(b) [42].To improve the sensitivity of this tactile system, the micro-patterned post-processed double-network friction layer and micro-structured electrode were integrated as the tactile receptor.This tactile receptor showed a short response time of just 6 ms, which is suitable for artificial tactile perception systems.Meanwhile, the hierarchical memorial mechanism can be completely simulated by this self-powered tactile system (such as from sensory memory to long-term memory), and the different memory state was achieved with the different working mechanism of the synaptic transistor.Finally, by integrating a 28 × 28 TENG matrix, a real-time digit hand written recognition system was demonstrated, which also achieved high hand written digit recognition accuracy.Notably, the tactile perception system would provide broad application in future neuromorphic systems and bioinspired electric devices.

Visual perception system based on neuromorphic OFET devices
Visual is the most important perception system for organisms to acquire and transmit external information, and they also have great significance for biological activities.Different from artificial tactile perception, artificial visual perception systems need not integrate extra receptors to obtain light signals.In visual neuromorphic OFET devices, light is regarded as the input or pre-synaptic spike [41,44,[120][121][122][123][124].For instance, Wang et al illustrated a visual perception system based on an OFET device, whose structure is a floating gate device (figure 9(a)) [120].CsPbBr 3 QDs with excellent absorption properties enabled this system to respond to multiple wavelengths, such as 365 nm, 450 nm, 520 nm, and 660 nm.During the on state, photogenerated carriers from QDs can easily escape to the pentacene channel layer, and the charges would be accumulated and retained for a long time.Whereas, a negative voltage applied to the gate electrode would induce the carriers to return to QDs and form an off state.This visual perception system exhibits excellent light-induced charge trapping and electrical-mediated charge release characteristics, which also allow optical programming operation and electrical erasing operation in one device.Notably, basic synaptic functions can be modulated in an optical way, which is suitable for mimicking the visual perception system.In addition, Chiang et al demonstrated a ptype and n-type channel layer visual perception OFET device with rod-coil as a floating gate [41].The charge-trapping layer of this device was bis((2E, 6E, 10E, 14E, 18E, 22E, 26E, 30E)- 3, 7, 11, 15, 19, 23, 27, 31, 35-nonamethylhexatriaconta-2, 6, 10, 14, 18, 22, 26, 30, 34- [3, 2-b] thiophene (C10-DNTT), and then followed by the channel layer of 1, 3, 8, 10(2H, 9H)-Tetraone, 2, 9-bis(2-phenylethyl)anthra[2,  1, 9-def:6, 5, 10-d ′ e ′ f ′ ]diisoquinoline (BPE-PDI) or DNTT.By integrating the function of external voltage and light illumination, this visual perception neuromorphic OFET device can stabilize switching in bistable states and exhibit excellent photonic synaptic functions.Meanwhile, Lee et al reported a stretchable visual perception system based on an OFET device, and the channel layer was an organic nanowire, which was suitable for a stretchable device [121].A neuromuscular electronic system is fabricated with a photodetector, a stretchable OFET device, and an artificial muscle.When the photodetector is exposed to a light signal, the output voltage changes and is transferred to the OFET device for output current, and finally, the artificial muscle responds to different light intensity or duration time.Different strain states also induce different variations of artificial muscle.Moreover, Deng et al recently reported an artificial visual perception system with the organic molecular crystal, which achieved the photo-synaptic function in a single device with its unique photo-induced charge transfer effect (figure 9(b)) [122].Notably, in this study, 5,11-bis(triethylsilylethynyl) anthradithiophene (Dif-TES-ADT) crystal arrays were used as photoactive layers due to their broad light absorption spectrum (300-650 nm), excellent air stability, and high carrier mobility (∼6 cm 2 V −1 s −1 ).The organic molecular crystalbased visual perception system exhibited photoresponsivity up to 1650 A W −1 at a low gate voltage of 5 V, and the photosynaptic functions were achieved by the oxygen-induced deep traps storage of photogenerated photon-induced charge from the organic molecular crystal.In addition, basic functions such as STP, LTP, and spike-timing-dependent plasticity (STDP) can be realized, and a proof-of-concept artificial image-perception sensor is constructed on a flexible substrate, possessing the capability to recognize and remember optical images.
Noticeably, in a real biological system, we not only need to transfer one perception information, but also need to integrate different sensory information into a comprehensive processing, which is called multisensory integration.Thus, many reporters have investigated multisensory integration perception systems, such as visual and tactile systems.In 2021, Yu et al showed a graphene/MoS 2 heterostructure OFET device that was capable of multisensory learning ability [123].In this multisensory system, the tactile perception, the visual perception, and the sensing information processing were simulated by a distance-changed TENG, a light-sensitive channel layer, and an OFET device, respectively.When the tactile stimuli applied to TENG can be converted into triboelectric potential and then transferred to the synaptic transistor, the Fermi level of graphene can be shifted and downward, then the carriers can be doping into the graphene channel layer.When the visual stimuli are applied to the channel layer, the effect of triboelectric and illuminated would infuse the photogenerated carriers from the MoS 2 layer into the graphene layer.In addition, the distance between two friction layers determines holes or electrons that can be doped into a channel layer.Moreover, the multisensory perception system achieved image recognition with the synergistic effect of tactile and visual perception.In the same year, Wu et al showed another multisensory perception system based on TENG and lightsensitive OFET devices [44].Meanwhile,TENG also acted as a tactile receptor, and the channel layer and QDs floating gate in this device acted as visual receptors (figure 10).This multisensory perception system not only successfully mimicked the behavior of the multisensory integration, but also stimulated the inverse effectiveness effect and temporal congruency, which were the primary principles of multisensory integration in biological systems.In addition, the digit recognition accuracy with multisensory perception is higher than single perception, which showed superior performance when meeting extreme or different environments.Then, a 3 × 3 pixels array actuated by TENG is fabricated to simulate the function of environment adaptable behaviors and showed high recognition accuracy with complex environments.In addition, neuromorphic OFET devices not only need to work for small-scale AI computing but also need integrated neuromorphic circuits to explore the external environment and provide corresponding consequences.Recently, Krauhausen et al demonstrated robotics that can integrate and learn sensorimotor [124].The autonomous robot can perceive touch and optical signals from sensors, and the real-time sensing signal from the robot can be transferred to the organic neuromorphic circuit for real-time feedback and learning, and then the processed signal can be transferred to actuators.The robot can achieve path selection in the two-dimensional maze through this self-feedbacked multisensory perception neuromorphic system.Notably, through the real-time learning of this system, a relationship between sensory receptors in the robot and actuators was established, and then the maze showed two different paths before and after training.The multisensory perception systems based on tactile and visual showcase more applications in complex and energy-restricted environments.

Other perception systems based on neuromorphic OFET devices
As one of the five senses, auditory perception system can perceive sound signals through ear and transfer them to the neural center for action potential generation [43,[125][126][127][128][129][130][131][132].This system is efficient and important, He et al in 2019 achieved the function of sound azimuth detection with two neuromorphic OFET devices [129].Two pre-synaptic electrode presents two ears and with the position of the sound source changed, the output currents can be changed.Thus, the orientation of the sound source can be judged through the ratio of two output currents, and the devices can act as the spatiotemporal signal processing units.Meanwhile, Liu et al investigated another auditory perception system, which integrated frequency-sensitive TENG and OFET devices as sound receptors and synaptic units, respectively, as shown in figure 11(a) [43].Notably, the fs-laser processed friction layer not only increased surface charge density but also increased the sensitivity of the  TENG receptor, which also improved the recognition accuracy of different sound instructions.In addition, to further investigate the self-adaptation performance of this auditory perception system, a noise-adjustable neuromorphic circuit based on a TENG, an OFET, and a synaptic transistor is demonstrated (figure 11(b)).The pre-synaptic current of V G can be adjusted by the output voltage of TENG and another load transistor, thus, the post-synaptic current can be regulated during noisy and soft environments.With this circuit, the specific instructions can be distinguished with recognizable postsynaptic current, which is suitable for working in complex environments.Moreover, Liu et al also reported a multisensory perception system composed of tactile, visual, and auditory perception in a one-structured vertical tribo-transistor device, as depicted in figure 11(c) [130].The distance between gate electrode and insulating layer would change the distribution of ions and channel conductance, and simulate the perception of tactile and auditory.Meanwhile, the light-sensitive MXenes network source electrode played an important role in visual perception detection.Furthermore, this multisensory perception system also achieved the function of multisensing-memory-computing with one device and increased the emotion distinguish accuracy.These functions would decrease the data exchange between multi-sensors, memory units, and computing units, and increase the working and transfer speed.
Despite the basic tactile, visual, and auditory perceptions, several other artificial perceptions are also investigated, such as kinesthetic perception [131][132][133].Shan et al showed a selfpowered kinesthetic perception system based on a TENG and a synaptic device (figure 12(a)) [131].Noticeably, the flexible single-electrode tribotronic sensor consisted of a PDMS protective layer, AgNWs electrode, and PDMS/MXenes friction layer, and the sensor can fit skin that can be pasted on the human body, and then the movements signal could be collected and transferred to the synaptic device.Based on this perception system, the bending angle and bending times of fingers, and the kinesthetic orientation angle of the arm can be recorded and distinguished.Moreover, this system was not only to recognize the movement of body but also to identify the dangerous driving and sign language.Authors designed a system that can integrate and evaluate the fatigue driving degree, which also can detect and analyze the state of the driver so that it can reduce traffic accidents.Furthermore, Chen et al recently reported a motion sensory system, which integrated the perception of visual and rotation signals (figure 12(b)) [133].This system was close to a biological reality system with multi-input signals and achieved temporal congruency.As schematically illustrated, the visual signal was simulated by a synaptic transistor array, and the rotation signal was mimicked by TENG.The TENG provided here can sense rotation and axial motion and then serve as the power supply of synaptic devices, simultaneously.Due to the function of sense motion and vision, authors demonstrated a spiking correlated neural network to distinguish different numbers and angles (0 • , 90 • , 180 • and 270 • ) simultaneously, which achieved the function of recognition of multimodal information.In addition, this system can judge the body state more quickly and accurately.A mechanical arm was connected with this multisensory perception system, and then the multisignals were transmitted to system and transferred to the mechanical arm for further reaction.The shorter time of the mechanical arm to grip a boll depicted that the multisensory perception inputs show superiority for motion perception and self-protection.

Summary and outlook
In summary, this study systematically reviewed the recent progress of different materials, structures, and sensory systems based on neuromorphic OFET devices.Significant efforts have been made to demonstrate bioinspired synaptic transistors, and several basic synaptic functions have been emulated by neuromorphic p/n/heterojunction OFET devices, such as EPSC, PPF, STDP, and other functions.Meanwhile, the implementation of neuromorphic OFET devices in bioinspired electronics, and the working mechanism with different structures have also been discussed.The studies also introduced different sensory perception systems with tactile, visual, auditory, and multisensory integration, which demonstrated that neuromorphic OFET devices have a wide range of applications.
Although many considerable advances have been achieved in neuromorphic OFET devices, many challenges urgently need to be resolved.First of all, considering the high operating voltage and output current of traditional OFET devices, which not only lead to significant power consumption but are not suitable for integrating circuits.Hence, it is critical to develop low-operation voltage OFETs for brain-inspired neuromorphic devices.Recently, Ren et al [134] reviewed the recent advances in the development of low-voltage OFETs and highlighted some key challenges for low-voltage OFETs for applications.Meanwhile, the understanding of brain function is relatively simple, accordingly, further studies should be conducted to investigate and explore the devices and mechanisms with different structural device to mimic biological function more accurately.Moreover, the organic semiconductor materials are unstable in atmosphere and not compatible with bioinspired electronics.Thus, pursuing high-stability and compatible materials should be further investigated.Furthermore, multisensory integration plays an important role in biological systems, while the inherent mechanism between different perceptions is not distinct.The mechanism with multiperception simultaneously impacted needs to be explored.Finally, the dimensions and integration level of neuromorphic OFET devices are also important for the application of neuromorphic computing and large-scale integration circuits.The fabricated devices need to be compatible with the current standard CMOS process, and downscaling for high-density integration and low-power consumption devices should be considered.

Figure 2 .
Figure 2. Single semiconductor materials in OFET.(a) Schematic illustration of single p-type semiconductor layer OFET and typical working mechanism.Reprinted with permission from [45].Copyright (2020) American Chemical Society.(b) All-solid-state vertical structured n-type device and (c) schematic structure of n-type materials.(d) Basic synaptic functions and the power consumption of this device.[60] John Wiley & Sons.© 2022 Wiley-VCH GmbH.

Figure 3 .
Figure 3. Heterojunction semiconductor materials in OFET.(a) Diagram of the structure of PDVT-10:N2200 bulk heterojunction device.(b) Typical transfer curve of this device with optical programming and electrical erasing.(c) The behavior of the device under different doping concentrations.Reprinted with permission from [62].Copyright (2021) American Chemical Society.(d) Schematic of MoS2/PTCDA layer heterojunction modulated by electrical or optical spike.(e) Heterojunction device works at electric mode; (f) heterojunction device works at optical mode.[71] John Wiley & Sons.© 2018 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim.

Figure 4 .
Figure 4. Electrolyte gate neuromorphic OFET device.(a) Schematic illustration of typically EDL device structure and writing diagram, post-synaptic current as a function of 300 pre-synaptic pulses.[52] John Wiley & Sons.© 2015 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim.(b) The working state and architecture of the multi-gate OECT device, the simplified visual system, and the definition of input spatial pulse patterns with different orientations and corresponding polar diagrams.Reproduced from [53].CC BY 4.0.

Figure 5 .
Figure 5. Ferroelectric neuromorphic OFET device with PVDF-TrFE dielectric layer.(a) Schematic illustration of neuromorphic OFET device with vertical configuration.(b) The working mechanism of vertical ferroelectric OFET device.(c) The pattern recognition of ferroelectric OFET device utilizing multilayer perception neural network.Reprinted from [96], © 2021 Elsevier Ltd.All rights reserved.

Figure 6 .
Figure 6.Flexible neuromorphic OFET device.(a) Diagram of ultrathin freestanding substrate ferroelectric OFET device.(b) Images of the conformable device on the brain-shaped PDMS mold.(c) Repetitive LTP and LTD operations of the folded device during 6000 spikes.Reprinted with permission from[98].Copyright (2019) American Chemical Society.

Figure 7 .
Figure 7. Floating gate neuromorphic OFET device.(a) The 3D device configuration of the neuromorphic device with C60/PMMA as a floating gate and corresponding cross-sectional side view SEM.(b) The energy band diagram of C60/PMMA device at hole trapping and electron trapping state.(c) EPSC of the device under different program duration times and temperatures.[56] John Wiley & Sons.© 2018 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim.(d) Schematic illustration of the floating-gate EDLT device.(e) LTP/LTD under different blending angles.Reprinted with permission from [105].Copyright (2022) American Chemical Society.

Figure 8 .
Figure 8. Tactile perception system based on neuromorphic OFET devices.(a) Schematic illustration of this DOT-TPE system, the equivalent circuit of the DOT-TPE system, and corresponding functions.[110] John Wiley & Sons.© 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim.(b) The biological tactile perception system, TENG and OFET device as tactile receptor and synapse, respectively, and 28 × 28 TENG receptor array for real-time digit recognition and its recognition accuracy with different digit numbers.Reprinted from [42], © 2020 Elsevier Ltd.All rights reserved.

Figure 9 .
Figure 9. Visual perception system based on neuromorphic OFET devices.(a) Device structure of the visual perception system based on OFET device, energy diagram of the device during a light programming operation and electrical erasing operation, and basic synaptic functions mimicked by artificial visual perception system.[120] John Wiley & Sons.© 2018 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim.(b) Device structure, working mechanism, and STDP of organic molecular crystal-based device.Reproduced from [122].CC BY 4.0.

Figure 10 .
Figure 10.Multisensory perception system based on neuromorphic OFET devices.(a) Schematic illustration of multi-sensory perception system based on TENG and neuromorphic OFET device.(b) Multi-sensory integration and the behavior of the device under light and touch pulses with the same pulse and interval time.Reprinted from [44], © 2021 Elsevier Ltd.All rights reserved.

Figure 11 .
Figure 11.Auditory and multisensory perception systems based on neuromorphic OFET devices.(a) An auditory perception system integrated with a TENG as receptor and OFET as synapse.(b) The circuit diagram, schematic configuration, and current level of the noise-adjustable neuromorphic circuit.Reprinted from [43], © 2020 Elsevier Ltd.All rights reserved.(c) The configuration of a multisensory perception system composed of tactile, visual, and auditory perception in a one-structured vertical tribo-transistor device.Reproduced from [130].CC BY 4.0.

Figure 12 .
Figure 12.Other perception systems based on neuromorphic OFET devices.(a) A self-powered kinesthetic perception system based on a TENG and a synaptic device.Reprinted from [131], © 2021 Elsevier Ltd.All rights reserved.(b) and (c) A motion sensory system integrates the perception of visual and rotation signals.Reprinted with permission from [133].Copyright (2022) American Chemical Society.

Table 1 .
Summary of semiconductor materials, synaptic functions, flexibility and stimuli of OFETs.

Table 2 .
Summary of structure, gate materials, preparation method, device size, synaptic functions, and flexibility of different neuromorphic OFET devices.