Stretchable synaptic transistors based on the field effect for flexible neuromorphic electronics

Structural designs for stretchable synaptic FETs

Stretchable synaptic transistors play a core role in flexible neuromorphic electronics and usually employ both organic and inorganic materials to fabricate functional layers in devices. Recently, many materials have been used in neuromorphic electronics, of which emerging two-dimensional (2D) materials have drawn great attention owing to their unique properties; synaptic devices based on 2D materials (such as graphene, graphene oxide, hexagonal boron nitride, transition metal dichalcogenides, transition metal oxides, 2D perovskites, and black phosphorus) have been developed^[67-70]. Organic materials are also considered as ideal candidates for neuromorphic electronics because of their excellent optoelectronic and mechanical properties. For instance, Chen et al. developed different types of synaptic devices based on organic materials, such as synaptic transistors with multi-sensing-memory-computing^[71], stretchable synaptic transistors with tunable behavior^[23], and artificial multisensory integration nervous systems based on synaptic transistors^[72]. Notably, certain organic and inorganic functional layer materials usually have high electrical qualities; however, they cannot withstand large mechanical deformations owing to their natural rigidity and brittleness, limiting their application in stretchable electronics.

In recent years, many efforts have been made to develop structural designs for synaptic devices that can improve their stretchability and electrical performance^[13,16,19]. Providing these devices with proper structural designs can enable them to dissipate mechanical deformation and become stretchable, decreasing the mechanical damage they experience and finally improving their mechanical and electrical performances. Stretchable synaptic transistors have been fabricated using structural designs for flexible neuromorphic electronics. Nguyen et al. fabricated a stretchable and stable synaptic electrochemical transistor on a 3D mogul-patterned polydimethylsiloxane (PDMS) substrate [Figure 2]^[13]. These transistors can successfully mimic synaptic behaviors even when stretched up to 30% strain, owing to the excellent stress absorption capability of the patterned PDMS substrate. Li et al. fabricated buckled monolayer MoS₂ FETs on a PDMS substrate via strain engineering [Figure 3A]^[16]. The buckled MoS₂ devices exhibit a stable transistor performance with mobility of ~30 cm²·V^-1·s^-1, an on/off ratio of ~10⁸, and a subthreshold swing (SS) of ~180 mV·dec^-1, even after many stretching cycles under more than 10% strain [Figure 3B]. Moreover, the devices can mimic optoelectronic synapse functions, such as STM/LTM and long-term potentiation/depression (LTP/LTD) characteristics, and the image recognition accuracy can reach ~95% even when the strain is as high as 6.5% [Figure 3C]. Lee and co-workers fabricated an organic optoelectronic sensorimotor synapse based on the s-ONWST^[19]. The s-ONWST employs a single wavy nanowire as the semiconductor layer, and the wavy nanowire retains its configuration after repeated stretching to 100% strain. During stretching tests, the s-ONWST, which is fabricated on an ion gel-gated platform, successfully emulates biological synapse functions, such as EPSC, PPF, spike voltage-dependent plasticity (SVDP), and spike number-dependent plasticity (SNDP), even after 50 cycles at 100% strain. These studies demonstrated that stretchable structural designs were promising approaches for endowing non-stretchable synaptic devices with high stretchability.

Figure 2. (A) Structure of the OECT with a membrane on a 3D mogul-patterned PDMS substrate and FESEM images of the 3D mogul-patterned substrate before (top) and after (bottom) stretching for 1,000 cycles at 30% strain, respectively; (B) Optical image of the OECT without membrane in the direction of the channel length at 0% (left) and 30% (right) strains; (C) Transfer characteristics and current change (I_DS/I_DS,0) of the OECT obtained under uniaxial static stretching from 0% to 30%; (D) I_DS value after 10 s upon removal of gate pulsing and SW values generated in the OECT with the membrane after different gate pulses number before and after cyclic stretching of 1,000 cycles at 30% strain. Reprinted with permission from Ref.^[13]. Copyright 2021, John Wiley and Sons. OECT: organic electrochemical transistor; PDMS: polydimethylsiloxane.

Figure 3. (A) Schematic diagram of the buckled MoS₂ FET on a stretchable PDMS substrate; (B) Transfer curves, mobility, and SS value of the device under biaxial strain with different strains, V_ds = 3 V; (C) Optical potentiation (1 Hz, 0.5 s) and electric depression (V_g = 12 V, 4.5 s) of the stretchable MoS₂ device under 0%, 3.2%, and 6.5% strain and the simulation results of image recognition accuracy under different strains. Reprinted with permission from Ref.^[16]. Copyright 2022, John Wiley and Sons. FET: field-effect transistor; PDMS: polydimethylsiloxane.

Intrinsically stretchable synaptic FETs

Stretchable synaptic transistors require that all components be stretchable and exhibit excellent electrical performance. Therefore, in addition to the development of stretchable synaptic transistors using structural designs, intrinsically stretchable synaptic transistors have been explored. Chen et al. developed ultra-stretchable organic electrochemical transistors (OECTs) based on a honeycomb-like porous microstructured organic semiconductor film with a biaxially pre-stretched platform [Figure 4A]^[12]. The resulting semiconductor film stabilized the effective electronic and ion transport pathways under deformation, enabling OECTs with stable output characteristics upon deformation and a maximum transconductance (g_m) degradation of 8% at 20% strain after 10,000 stretching cycles [Figure 4B]. Meanwhile, the stretchable OECT showed stable current output characteristics, even with strains of up to 30%-140%. The stretchable OECT also showed stable synaptic characteristics, such as PPF and EPSC, with minor variations in postsynaptic currents under different strains [Figure 4C]. Molina-Lopez and co-workers fabricated stretchable transistor arrays using intrinsically stretchable active materials via the inkjet printing process^[15]. Poly(3,4-ethylenedioxythiophene)-poly(styrenesulfonate) (PEDOT: PSS) was the source/drain/gate electrode; interconnect, a network of semiconducting single-walled carbon nanotubes (SCSWCNTs), was the semiconducting channel; and solid-state ionic poly(vinylidene fluoride-hexafluoropropylene) (PVDF-HFP) was the gate dielectric. The resulting devices exhibited excellent electrical performances, with an average mobility of 27 ± 5 cm²·V^-1·s^-1, I_on/I_off > 10⁴, and a maximum transconductance of 47 ± 9 µS. During the stretching tests, these parameters remained stable even for independent strains of up to 20%, owing to the tube-to-tube charge transport in the SWCNT networks without strain dependence. These stretchable synaptic transistors, which emulated the characteristics representative of the synaptic behavior of double-layer capacitive dielectrics, such as EPSC, inhibitory postsynaptic currents (IPSC), and short-term plasticity (STP), exhibited outstanding mechanical stability during stretching tests. Shim et al. developed a fully rubbery synaptic transistor using a composite of poly(3-hexylthiophene) (P3HT) nanofibrils percolated in polydimethylsiloxane (P3HT-NFs/PDMS) as the stretching semiconductor and an ion gel as the gate insulator [Figure 5A]^[17]. The synaptic transistors showed unique synaptic characteristics, including EPSC, PPF, and memory [Figure 5B]. These synaptic characteristics were retained even after the transistors were stretched by 30%. Finally, a neurological electronic skin based on a fully rubbery synaptic transistor and two pressure-sensitive mechanoreceptors, which could successfully emulate the Morse Code and facilitate communication with a robotic hand, was fabricated. These studies demonstrated the feasibility and importance of developing intrinsically stretchable synaptic transistors for flexible neuromorphic electronics.

Figure 4. (A) Illustrations of the pre-stretched OECT fabrication process; (B) the g_m and I_on dependence on stretching cycles for psh-DPP-g2T (ε_ps = 100%) OECTs under 0% and 30% elongation strains and the g_m/I_on data for psh-DPP-g2T (ε_ps = 150%) OECTs under various strains (from 0% to 150%); (C) EPSCs triggered by multiple spikes (each spike, -0.7 V, 100 ms; interval, 100 ms) under 0% and 60% elongation strains in the ε_⊥or ε_∥direction. Reprinted with permission from Ref.^[12]. Copyright 2022, The Author(s), under exclusive license to Springer Nature. EPSC: excitatory postsynaptic current; OECT: organic electrochemical transistors.

Figure 5. (A) Schematic view of an elastic neurological electronic skin and a fully rubbery, all-organic synaptic transistor, and the optical images of the fully rubbery synaptic transistor under mechanical strains of 0%, 10%, 20%, 30%, and 0% along the channel length direction; (B) Schematic illustration of the synapse and the organic synaptic transistor with an ion-gel gate dielectric, and biological synaptic functions of all-organic synaptic transistors, mainly including EPSC and PPF. Reprinted with permission from Ref.^[17]. Copyright 2021, Tsinghua University Press and Springer-Verlag GmbH Germany, part of Springer Nature. EPSC: excitatory postsynaptic current; PPF: paired-pulse facilitation.

Light sensory systems

In the human brain, which receives more than 80% of its external environmental information from the human visual system^[73,74], light can control the release of neurotransmitters and thereby adjust brain activity in vivo; this process is mediated by light sensory systems^[75]. Some scientists have attempted to design light sensory systems composed of synaptic devices and optical sensors to mimic the processes occurring in this system^[76-78]. Chen et al. designed an artificial flexible visual memory system by integrating memristors with ultraviolet (UV) image sensor arrays on flexible polyimide substrates^[76]. The visual memory system can detect and memorize image information for at least 1 week [Figure 6A]. Wang et al. developed a light-triggered organic neuromorphic device (LOND) by using a ferroelectric/electrochemical artificial synapse and a light-sensitive electronic component to mimic the visual perception system^[77]. This device can transduce incident light frequency, intensity, and wavelength into synaptic signals in both volatile and nonvolatile forms. The LOND can also recognize colors and shows varying degrees of synaptic signal volatility under 550 nm and 850 nm light [Figure 6B]. Additionally, a sensory synaptic device that combines synaptic and sensing capabilities into a single device has been used to emulate light sensory systems. Huang et al. fabricated organic/inorganic hybrid heterojunction short-wave infrared (SIR) synaptic phototransistors (OHSPTs) that successfully simulate typical synaptic behavior under SIR light irradiation^[78]. The OHSPT array can successfully realize the image recognition of SIR light on dark nights under artificial white-light conditions, exhibiting good self-adaptability and strong anti-interference ability.

Figure 6. (A) Schematic of the artificial visual memory unit integrated by the image sensor and resistive switching memory device, photos of the integrated devices array on flexible polyimide substrates, and information storage behaviors and reusability of the flexible device arrays for the applied patterned light. Reprinted with permission from Ref.^[76]. Copyright 2018, WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim; (B) A LOND cross-sectional and the P(VDF-TrFE)/P(VP-EDMAEMAES) bilayer SEM image cross-sectional, and signals in the LOND array recorded after NIR and green light exposure (10.80 mW cm^-2, 64 Hz for 20 s). Reprinted with permission from Ref.^[77]. Copyright 2018, WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim. LOND: light-triggered organic neuromorphic device.

Artificial light sensory systems based on stretchable synaptic FETs have also been developed. Lee et al. developed a stretchable optoelectronic sensorimotor synapse system based on a stretchable organic nanowire synaptic transistor^[19]. Regarding s-ONWSTs, they use a wavy organic nanowire and a high-capacitance ion-gel electrolyte as a stretchable semiconductor and gate insulator, respectively; and they show stable current-voltage characteristics and various typical postsynaptic behaviors at both 0 and 100% strains. Notably, s-ONWSTs can generate EPSCs by exploiting the output voltage of a photodetector, thereby converting light signals to postsynaptic currents. Therefore, by inducing an EPSC amplitude response, a stretchable optoelectronic sensorimotor synapse system comprising a photodetector and s-ONWSTs can achieve optical wireless communications. Every letter of the English alphabet can be expressed through the control of light signals; this expression can induce a distinct EPSC amplitude response in s-ONWSTs. Using the stretchable optoelectronic sensorimotor synapse system, the International Morse code (such as “SOS,” “HELLO” and “UNIVERSE”) has been successfully emulated; the outputted International Morse code did not show any notable changes at both 0 and 100% strains. When an s-ONWST was connected to a polymer actuator through a transimpedance circuit, forming an artificial neuromuscular junction, the polymer actuator could be activated by voltage signals converted from the EPSCs of the organic optoelectronic synapse. The biological muscle tension responses during contraction were successfully emulated using an artificial neuromuscular electronic system. This study provided a promising strategy for developing next-generation biomimetic soft electronics, soft robotics, neurorobotics, and electronic prostheses with wireless communication, and it demonstrated the role of stretchable synaptic transistors in artificial sensory systems.

Tactile sensory systems

Tactile perception enables animals to perceive and respond to external tactile stimuli, facilitating their communication with the external environment. The realization of an artificial tactile sensory system to achieve bionic touch perception is essential for applications in prosthetics, flexible robotics, and human-machine interfaces. Artificial tactile sensory systems based on synaptic devices and tactile sensors have been developed to emulate the biological processing of tactile signals^[5,29,79-81]. Shan et al. proposed a novel dual-output artificial tactile sensing (DOATS) system capable of realizing the parallel outputting of photoelectric signals^[80]. The DOATS system simulated human tactile perception and recognized 16 types of fabrics with an accuracy rate of 94.1%. Multitask operations that can simultaneously drive the manipulator and implement image recognition have also been successfully achieved [Figure 7]. Wan et al. developed a flexible neurological electronic skin with a ferroelectric nanogenerator (FENG) and an artificial synaptic transistor, which successfully processed tactile information containing the magnitude and activation history of the force stimuli [Figure 8]^[81].

Figure 7. (A) Parallel output signals simulate slow adaptative/fast adaptative, and the schematic of photoelectric signal DOATS system and the photoelectric hybrid artificial neural network; (B) Optical images of 16 different fabric textures and texture total classification accuracy; (C) Photographs of the manipulator from resting state to moving state, and the DOATS system realizes the image learning process through electrical signals. Reprinted with permission from Ref.^[80]. Copyright 2022, American Chemical Society. DOATS: dual-output artificial tactile sensing.

Figure 8. (A) Schematic illustration of artificial electronic sensory skin; (B) Photograph of a 1.2 × 1.2 cm² pristine PPFE film and complete ferroelectric nanogenerator device, respectively, and schematic of the regulating circuit based on ferroelectric nanogenerator; (C) Normalized PSC change of a synaptic transistor under different loading forces, and synaptic weight curves (peak and raw values of EPSC and IPSC) measured with periodical loading forces of 41.3 or 14.2 N at a frequency of 0.67 Hz. All synaptic behavior is measured at |V_DS| = 1.0 V. Reprinted with permission from Ref.^[81]. Copyright 2020, American Chemical Society. EPSC: excitatory postsynaptic current; ferroelectric nanogenerator ; IPSC: inhibitory postsynaptic currents; SWCNT: single-walled carbon nanotubes.

Studies have explored ways to improve the compatibility of stretchable artificial tactile sensory systems, which are based on stretchable synaptic transistors and tactile sensors, with animals. Shim and co-workers developed a fully rubbery, neurologically integrated, tactile sensory skin consisting of an array of rubber pressure-sensitive mechanoreceptors and fully rubbery synaptic transistors^[5]. Mechanoreceptors respond to physical touches and generate presynaptic pulses, which are transmitted to synaptic transistors and excite postsynaptic potentials. Noticeably, the excitatory postsynaptic potential (EPSP) mapping from the sensory skin was achieved; for instance, the “U” and “H” patterns have been successfully marked by tapping the sensory skin with objects and recording the EPSP mapping. The EPSPs remained stable even after a mechanical strain of 50% was applied. Moreover, a soft adaptive neurologically integrated robot, namely a soft pneumatic robot covered with fully elastic skins of triboelectric nanogenerators (TENGs) and synaptic transistors, was fabricated. This robot could adaptively sense physical stimuli such as taps, touches, and locomotion, which were attributed to synaptic memory-encoded signals. When the TENG was physically tapped, presynaptic pulses were generated and transmitted to the synaptic transistors. Subsequently, the EPSC and short-term weight change w_n/w₁ values were obtained, using which the locomotion of the programmable robot could be controlled. Notably, the skin on the left and right sides guided the robot to turn right and left, respectively; and the top skin controlled the time at which the robot turned on or off. Controlling the tapping numbers from one to four times on the side skin could guide the robot to turn 10, 20, 30, and 40°. For instance, tapping the right skin of the soft robot four times could turn it 40° to the right. The soft neurorobot could perform adaptive locomotion along a complex path, turning right by 30° and then continuing to locomote straight for 10 s. Stretchable rubbery synaptic transistors based on elastomeric electronic materials such as rubbery conductors, semiconductors, and ion gels play an important role in the processing of tactile signals. Stretchable synaptic transistors have exhibited a set of synaptic characteristics including SM, STM, LTM, and filter characteristics, retaining them even after stretching to 50% strain. This study demonstrated the possibility of enabling neurological functions in soft machines and other applications.

Multisensory artificial nervous systems

Biological sensory systems can receive external stimuli and transfer different forms of physical stimuli into information coding for brain processing, enabling precise perception and appropriate reactions to complex real-world scenarios^[82,83]. In biological systems, external stimuli are often processed synchronously and integrated into a comprehensive awareness; thus, humans are more sensitive to complex stimuli than they are to single stimuli^[84]. Therefore, the incorporation of multisensory learning capabilities into electronic devices for the development of artificial sensory systems is increasing. Previous studies have attempted to emulate biological multisensory functionalities by integrating various types of sensors with artificial neuromorphic devices^{[27,72,85,86]}. Wu et al. proposed an artificial multisensory integrated nervous system that integrates a flexible triboelectric nanogenerator (TENG) with an organic photosynaptic transistor and can emulate both haptic and iconic perception behaviors^[72]. The multisensory integration nervous system successfully emulated typical multisensory integration behaviors, including the inverse effectiveness effect and temporal congruency, and the pattern recognition accuracy of the multisensory integration system was higher than that of a single-sense system. Liu et al. fabricated a multisensory artificial nervous system with a multisensory-memory-computing function and multitask emotion recognition in a self-powered vertical tribotransistor (VTT), which integrated a triboelectric nanogenerator (TENG) and transistor^[71]. The multisensory artificial nervous system could simulate tactile, auditory, and visual perceptions, and a multi-model emotion recognition system based on the VTT was achieved with an emotion recognition accuracy reaching 94.05%.

Multisensory nervous systems based on stretchable synaptic FETs have been introduced into flexible neuromorphic electronics. Liu and co-workers developed a nanowire-channel intrinsically stretchable neuromorphic transistor (NISNT) that integrated the functions of a synapse and tactile sensor and could perceive both tactile and visual information^[21]. This transistor employed P3HT/PEO nanowires and ion gel as the stretchable semiconductor and gate insulator, respectively, and it successfully mimicked synaptic characteristics, including EPSC, PPF, SFDP, SNDP, LTP, and LTD. These characteristics did not degrade even after 1,000 stretching cycles, demonstrating the high tolerance of the NISNT to stretching. When the stretchable NISNT was used as a tactile afferent nerve, it adhered to the five figures of the hand and achieved gesture recognition. For example, the 270 samples for the hand gestures “Good,” “Yeah,” and “OK” could be distinguished even under stretching deformation, and the accuracy was approximately 92.6%. Notably, NISNT could also process visual information owing to the light-sensitive feature of the P3HT/PEO channel, and through the introduction of light information, the accuracy of gesture recognition could be further improved, reaching 96.3%. Wang and co-workers developed a stretchable temperature-responsive multimodal neuromorphic electronic skin (STRM-NES) based on an array of intrinsically stretchable carbon nanotube synaptic transistors with mechanoreceptors and thermoreceptors [Figure 9]^[20]. Temperature sensing and modulation, deformation sensing and modulation, perception and memory, and healing from burning could be simultaneously realized in STRM-NES. Integrated synaptic carbon nanotube transistors (IS-CNT-STs) can emulate essential synaptic characteristics such as EPSC, IPSC, PPF, LTP, LTD, and the Pavlovian conditional reflex. The synaptic characteristics of these devices showed no obvious changes, even after 1,000 stretching cycles at 30% strain. These synaptic characteristics could be modulated by different stretching and deforming states; for instance, the PPF index declined by 120% when the strain was up to 50%. Temperature variations can also modulate the synaptic characteristics of a device, and the postsynaptic current can increase or decrease with the accumulation of thermal temperature. The synaptic characteristics of the STRM-NES could be tuned by both temperature and stretch deformation; the postsynaptic currents of the skin were weaker after a 30% strain with an increasing temperature than when it was unstretched. Additionally, the STRM-NES successfully emulated the self-healing abilities and temperature-location resolution of the skin. These studies demonstrated that stretchable synaptic transistors had great potential for the development of multisensory artificial nervous systems.

Figure 9. (A) Structure diagram of the neuromorphic device and its application; (B) Multiple negative presynaptic pulse induced postsynaptic currents at different stretch cycles and under different levels of mechanical strain, respectively; (C) Temperature-responsive LTP/LTD after 30% strain. Reprinted with permission from Ref.^[20]. Copyright 2022, American Chemical Society. LTP/LTD: long-term potentiation/depression.

Different sensory systems based on stretchable synaptic FETs with a wide range of application prospects in bioinspired soft robots and neural prostheses have been developed^[5,17,20,87]. For instance, Shim et al. proposed an artificial neuromorphic cognitive skin based on an array of biaxially stretchable elastomeric synaptic transistors^[87]. Each biaxially stretchable synaptic transistor could emulate a full set of synaptic behaviors, such as the EPSC, PPF index, and memory characteristics, under various mechanical strains. The stretchy neuromorphic imaging sensory skin device based on stretchable elastomeric synaptic transistors exhibited a stable pattern reinforcement function under non-uniform deformation [Figure 10]. These studies demonstrated that stretchable synaptic transistors had the potential for application in next-generation wearable electronics, soft robotics, and neuroprosthetics, which are potential future development directions for flexible neuromorphic electronics.

Figure 10. (A) Schematic of the stretchy neuromorphic imaging sensory skin device; (B) Optical photographs of the neuromorphic imaging sensory skin device at initial and biaxially stretched states before and after light illumination (“7”-shaped); (C) Mapped photoresponsive EPSC of the stretchy neuromorphic imaging sensory skin device with a different number of light pulses, and optical photographs of the stretchy neuromorphic imaging sensory skin device adhered on human skin at initial and fisted state before (Top) and after (Bottom) the light illumination. Reprinted with permission from Ref.^[87]. Copyright 2022, the Author(s). Distributed under Creative Commons Attribution-NonCommercial-NoDerivatives License 4.0 (CC BY-NC-ND). EPSC: excitatory postsynaptic current.