A bioinspired neuromuscular system enabled by flexible electro-optical N2200 nanowire synaptic transistor

N-type poly{[N,N'-bis(2-octydodecyl)naphthalene-1,4,5,8-bis(dicarboximide)-2,6-diyl]-alt-5,5'-(2,2'-bithiophene)} (N2200, Mw > 50 000; Derthon) and poly(ethylene oxide) (PEO, Mw ∼ 400 000; Sigma-Aldrich) were dissolved in chlorobenzene at a 7:3 weight ratio, then stirred for 12 h at 60 °C to create a uniform printing ink. N2200 nanowire synaptic transistors were prepared using e-NWP. The process is controlled by an intelligent digital platform, which enables fast and highly reproducible printing of NW arrays. For the printing process, a polyimide (PI) substrate with deposited gold source and drain electrodes (70 nm) was placed on the printing platform, and a syringe was filled with the prepared ink. The syringe had a metal nozzle that had 80 μm inner diameter and was mounted with its tip 3.5 mm above the substrate. A voltage of ∼2 kV was applied to the nozzle, and the blended ink was extruded at 20 nl min⁻¹ with the nozzle moving at 1 m s⁻¹. The N2200/PEO ink formed a fine stream under the electric field. During the printing process, the solvent fully evaporated, so a continuous solid NW forms on the PI substrate.

The FNST can be stimulated by light pulses. To realize electrical spikes modulation, an ion gel composed of polymer (poly(vinylidene fluoride-co-hexafluoropropylene)) (PVDF-HFP) and ionic liquid (1-ethyl-3-methylimiazolium bis-(trifluoromethyl sulfonyl) imide) ([EMIM-TFSI]) was transferred onto the surface of N2200 NWs. The ion gel was prepared by mixing PVDF-HFP, [EMIM-TFSI], and acetone in a weight ratio of 1:4:7 at room temperature, then drying the mixture in a vacuum oven at 70 °C for 0.5 h. A metal probe was used as a presynaptic input terminal.

To form the electrolyte layer, chitosan (∼0.6 g) was added into 2 wt% acetic acid aqueous solution (20 ml). The mixture was stirred for 30 min at 70 °C, then sonicated for 15 min to remove remnant air bubbles. Finally, the mixture was placed in a vacuum oven for 4 h at 75 °C to remove the solvent.

To form the electrode layer, chitosan (∼0.2 g) was added to the acetic acid solution and stirred for 30 min at 70 °C. Then multiwalled carbon nanotubes (10.0%; Boyu Gaoke) and reduced graphene oxide (0.5 wt%; Boyu Gaoke) were added to the solution and stirred at 70 °C for 10 min. Polyaniline (∼0.08 g) was then added and the mixture was stirred for 40 min, then sonicated for 30 min. Finally, the mixture was poured into a mold then placed in a vacuum oven for 4 h at 75 °C.

To fabricate artificial muscles, the electrolyte layer was sandwiched between two electrode layers, which were then cut to 20 mm × 5 mm and assembled using hot-pressing technology.

To construct the bioinspired neuromuscular system, the synaptic device was integrated with the amplifier circuit and artificial muscle fibers. To control the efferent muscular actions, an operational amplifier was used to transform the output from the synaptic device. By connecting the drain electrode of the synaptic device to the conversion amplification module, the postsynaptic current is converted to an output voltage, which can drive the artificial muscle fibers.

The surface morphology of N2200 nanowires was investigated using a scanning electron microscope (SEM) (TESCAN MIRA LMS) and an atomic force microscope (AFM) (Bruker). X-ray diffraction (XRD) patterns were obtained using a Rigaku D diffractometer. X-ray photoelectron spectra (XPS) were obtained using a Thermofisher Nexsa. Absorption spectra were measured using a Hitachi U4150 spectrometer. All electrical characterizations were performed using a probe station in a nitrogen-filled glovebox, and the data were gathered using a Keithley 4200A semiconductor parameter analyzer. A continuous-wave semiconductor laser with a wavelength of 405 nm was used as the light source.

Motor neurons connect with skeletal muscle fibers via chemical synapses within living organisms [37]. Upon receiving excitatory signals, the presynaptic area of a motor neuron releases the neurotransmitter acetylcholine (ACh) [38], which diffuses across the synaptic cleft, then interacts with ACh receptors in the postsynaptic membrane. This interaction causes ion channels to open and thereby generate a postsynaptic potential [39]. This chain of events ultimately leads to the contraction of muscle fibers.

To construct a bioinspired neuromuscular system, the FNST was used to emulate the functions of biological synapses. Polymer-ion-gel fibers were fabricated to serve as artificial muscles, simulating biological muscles. Electrical spikes with different frequencies can modulate the postsynaptic currents generated by the FNST. After amplification through a circuit, the postsynaptic currents drive the artificial muscles to contract, and thus mimic the excitatory behavior of motor neurons (figure 1(a)).

Zoom In Zoom Out Reset image size

The electrodynamically-printed NWs were long and continuous. Parallel NWs were ∼60 μm apart (figure 1(b)), and each NW had a diameter of ∼650 nm (figure 1(c)). An AFM image of a single N2200 NW showed a height of ∼200 nm (figure 1(d)). The XRD pattern of N2200 nanowires showed a peak at 2θ = 3.92°, which represents the (100) orientation of N2200 (figure 1(e)) [40]. XPS spectra showed peaks at binding energies E_B = 163.88 eV for S 2p_3/2 and E_B = 165.08 eV for S 2p_1/2; these peaks confirm the existence of sulfur in N2200 NWs (figure 1(f)) [41]. The XPS spectra also confirmed the presence of oxygen and nitrogen (figure S2). Absorption spectra of the N2200 nanowires had a broad absorption peak between 500 and 800 nm (figure 1(g)).

In a human retina, when photoreceptor cells receive light stimulation of a specific intensity or wavelength, they undergo chemical and electrophysiological changes, which generate electrical signals (figure 2(a)) [42, 43]. They are passed through synapses to connected bipolar cells, which relay the signals through synapses to retinal ganglion cells [44]. N2200 as an n-type organic semiconductor exhibits visible light responsiveness (figure 1(g)). Additionally, its excellent flexibility makes it suitable for use in flexible electro-optical synaptic transistors to emulate the photoreceptor neurons in the human eye.

Zoom In Zoom Out Reset image size

The FNST can act as a light-sensitive synapse modulated by light pulses (figure 2(b)). The FNST shows a higher current under illumination than in darkness; this difference confirms the production of photo-generated carriers (figure 2(c)). Under drain voltage V_d = 1 V, the application of a light pulse (405 nm, 50 ms) induced generation of photo-carriers in the N2200 NWs, and I_ds was considered as the excitatory postsynaptic current (EPSC). The photo-carriers gradually dissipate once the pulse stimulus ceases, and the EPSC returns to its initial state (figure 2(d)).

Paired-pulse facilitation (PPF) represents short-term synaptic plasticity [45, 46]. PPF was simulated by applying two consecutive light pulses under V_d = 1 V (figures 2(e) and S3). Experiments about synaptic plasticity were repeated more than three times, and corresponding statistics were shown in mean value and standard deviation. The height of the second EPSC peak (A2) was higher than that of the first EPSC peak (A1) (figure 2(e)), because the photo-induced carriers generated by the first light pulse had not entirely recombined by the time the second pulse arrived, so the carriers accumulated, and thereby increased the photocurrent. The increase was quantified as PPF index = A2/A1 × 100%. When two light pulses separated by intervals 50 ⩽ Δt ⩽ 450 ms were applied to the FNST, the PPF index gradually decreased as Δt increased (figures 2(f) and S4).

Spike-duration dependent plasticity (SDDP) refers to the correlation between EPSC and the duration of synaptic stimuli. To measure the SDDP, the FNST was stimulated using light pulses with durations 50 ⩽ D ⩽ 500 ms in increments of 50 ms, under V_d = 1 V. The EPSC increased as D increased (figure 3(a)). The SDDP index was defined as the ratio of EPSC triggered by the nth light pulse (An) and that triggered by the first light pulse (A1). As D increased, the SDDP Index increased continuously (figure 3(b)). Increase in D causes increase in the number of photo-generated carriers in the semiconductor, so the photocurrent increases.

Zoom In Zoom Out Reset image size

Spike-rate dependent plasticity (SRDP) is a form of synaptic plasticity and is among the most important mechanisms for synaptic learning and memory in neural cognitive processes [47, 48]. In the context of this mechanism, high-frequency neuronal activities generally result in the reinforcement of synaptic weights. The mechanism facilitates self-organization and adaptability within neural networks, and thereby achieves high flexibility in information encoding. The FNST emulated SRDP. When consecutive light pulses with pulse frequencies 3.33 Hz ⩽ ƒ ⩽ 10 Hz were applied to the FNST under V_d = 1 V, its EPSC gradually increased as f increased (figure 3(c)). The SRDP index was defined as the ratio between the EPSC triggered by the 6th or 10th (A6 or A10) light pulse to the EPSC triggered by the first light pulse (A1). The SRDP index increased as f increased (figure 3(d)).

Spike-number dependent plasticity (SNDP) is a form of synaptic plasticity in which the EPSC increases when the number of stimuli increases. At a constant V_d = 1 V, the EPSC increased as the number N of light pulses was increased from 1 to 10 (figure 3(e)). This phenomenon occurs because increase in N causes increase in the number of photo-generated carriers in the semiconductor, with consequent increase in photocurrent. The SNDP index was defined as the ratio between the EPSC triggered by the light pulses with the pulse number of N (An) and the EPSC triggered by a single light pulse (A1). As N increased, the SNDP index increased (figure 3(f)).

The responses of the FNST can also be modulated by electrical spikes (figure 4(a)). Cations are initially distributed randomly in the ion gel. When an electrical spike (2 V, 50 ms) is applied under V_d = 0.4 V, they move to and accumulate at the interface between the ion gel and the N2200 nanowires to form an electric double layer. In this process, electrostatic coupling between cations and electrons leads to an accumulation of electrons in the n-type channel, and thereby increases the drain current or EPSC. Once the pulse stimulus ceases, the ions within the ion gel start to diffuse back to their original distribution. Consequently, the carriers cease accumulating, and the drain current returns to a steady state after a certain relaxation time (figure 4(b)).

Zoom In Zoom Out Reset image size

PPF was simulated by applying two consecutive electrical spikes (2 V, 50 ms) under V_d = 0.4 V. Δt between the electrical spikes was less than the relaxation time of the cations, so the cations in the ion gel could not fully return to equilibrium before the second stimulus was applied. The residual cations in the interface summed with additional cations that were driven to the interface by the second spike, so the current increased; as a result, the EPSC triggered by the second electrical spike (B2) was higher than that triggered by the first electrical spike (B1) (figure 4(c)). The PPF index B2/B1 × 100% was gradually decreased as Δt increased from 100 to 1400 ms (figures 4(d) and S5).

Spike-voltage dependent plasticity (SVDP) was quantified under V_d = 0.4 V at gate voltages 1 ⩽ V_g ⩽ 5 (figures 4(e) and S6). EPSC gradually increased as V_g increased. Therefore, the synaptic plasticity of the FNST can be effectively modulated by the amplitude of gate voltage.

SNDP was mimicked by applying electrical spikes. Under V_d = 0.4 V, as the number N of electrical spikes (2 V, 50 ms) increased from 1 to 10, the accumulation of ions in the interface or channel was increased, so EPSC increased (figure 4(f)). SNDP index = Bn/B1 × 100%, where Bn and B1 are the EPSC triggered by electrical spikes with N = n and 1, respectively, increased monotonically due to the increase of EPSC with an increase in the number of spikes (figure S7).

SDDP was mimicked by applying electrical spikes with different D under V_d = 0.4 V. SDDP index was defined as Bn/B1 × 100%, where Bn and B1 are the EPSC triggered by the n^th and the first electrical spikes, respectively. As D increased from 50 to 500 ms, both the EPSC and SDDP index increased monotonically (figures 4(g) and S8). This phenomenon occurs because increase in D causes increase in the number of cations that move from their equilibrium positions to the interface between ion gel and N2200, so the conductance of the N2200 channel increases.

The SDDP feature can be exploited to generate international Morse code by using electrical spikes to modulate the FNST. International Morse code consists of 'dot' (·) and 'dash' (-) [49, 50]. Morse code was encoded by adjusting the duration of electrical spikes. The threshold was set as 1.2 μA. EPSC triggered by electrical spikes with D = 50 ms were smaller than this threshold and correspond to '·'; whereas spikes with D = 150 ms were larger than the threshold and correspond to '-'. Signals 'NKU' were successfully generated by the FNST (figure 4(h)).

Neuromorphic devices offer a more efficient emulation of brain-like mechanisms for information processing than von Neumann computational architectures do [51, 52]. Neuromorphic devices can perform fundamental neural tasks like spike propagation and synaptic plasticity, and can support intricate dynamic operations such as multi-layer accumulation processing [53]. A bioinspired neuromuscular system was constructed by integrating the FNST, an amplifier circuit, and artificial muscle fibers (figure 5(a)). As applying currents, the ions inside the artificial muscle fibers rapidly diffuse and accumulate, resulting in irregular bending of their surface. The efferent muscular actions can be controlled by conveying the output signals from the synaptic devices to the artificial muscle fibers. The applications of this bioinspired neuromuscular system are mainly geared towards neuromorphic prosthetic limbs and neuro-controlled robots [7, 54].

Zoom In Zoom Out Reset image size

Controllable efferent muscular actions of the bioinspired neuromuscular system were realized by exploiting the characteristics of SRDP. When ten consecutive electrical spikes (2 V, 50 ms) were applied under V_d = 0.4 V, EPSC increased monotonically as f increased from 1.82 to 10 Hz (figure 5(b)). SRDP index was defined as the ratio of the EPSCs triggered by the 10th spike (B10) and the first spike (B1); this index also increased monotonically as f increased (figure 5(c)). The initial state of the artificial muscle fiber was unflexed (figure 5(e)). When electrical spikes were applied at f = 3.33 Hz to the FNST, the flexion angle of the artificial muscle fiber was ∼10° (figure 5(f)); as f increased to 10 Hz, the flexion angle of the artificial muscle fiber reached ∼60° (figure 5(g)). Figure S9 shows the statistical curve of signal output for each part of the bioinspired neuromuscular system. These results confirm that the efferent muscular actions of the bioinspired neuromuscular system can be effectively controlled by the FNST by exploiting the SRDP coding strategy.

Furthermore, the FNST showed excellent flexibility. EPSC was collected after different cycles to a bending radius of 1 cm. Under V_d = 0.4 V, the EPSC peak triggered by a single electrical spike (2 V, 50 ms) was slightly degraded after 1, 10, 100, 1000, and 5000 bending cycles (figure S10). Here, the retention rate R was defined as the ratio of the EPSC peak with bending to that without bending. R was greater than ∼95% after 1000 bending cycles and greater than ∼85% after 5000 bending cycles (figure 5(d)). These results demonstrate that this FNST has potential application in wearable electronics.