Synapses and neurons are the building blocks of the biological neural network, allowing for the neuromorphic transmission and processing of information to support learning and memories^[1−3]. Inspired by neurons and their functions, many efforts have been devoted to developing advanced artificial neurons for efficient and energy-saving neuromorphic computing^[4−8], as well as the construction of biomimetic neural system^[9−15]. One of the critical goals for the energy-efficient artificial neurons is to achieve bioinspired spike-based dynamic spatiotemporal signal processing^[16]. This spatiotemporal encoding through spike frequency and pattern recently has recently led to remarkable breakthroughs in neuromorphic robotics, prosthesis, and human-machine interfaces^{[11, 17, 18]}.

To achieve energy-efficient artificial neurons with spike coding, a variety of devices have been exploited, including resistive switching, phase-change, and ferroelectric memory^{[6, 19]}. Among them, the resistive switching memristor is deemed as an attractive device for the construction of artificial neurons because its two-terminal structure and stimuli-activated resistive switching characteristic are analogous to biological synapses^[20−22]. Volatile resistive switching memristors with threshold switching behaviors have been widely employed as the stimuli-spike coding unit for artificial spiking neurons^{[23, 24]}. However, most existing artificial spiking neurons rely on inorganic threshold switching memristors, such as TaO_x^{[25, 26]}, HfO_x^[27], NbO_x^{[16, 28]}, etc.^{[29, 30]}. By contrast, organic memristive materials are rarely used for the construction of artificial spiking neurons. Two main challenges still remain towards the development of spiking neurons made of organic materials: (ⅰ) the homogeneous preparation process^[31], (ⅱ) relatively large threshold voltages (V_th) caused by their thick layers^[32−34]. Moreover, regardless of organic or inorganic materials, a large intrinsic parasitic capacitance (C_p) is necessary for the highly integrated artificial spiking neurons without external parallel capacitors.

As a neurotransmitter, dopamine in biological synapses is vital for multiple neurophysiological activities (Fig. 1(a)). Its polymerized form, polydopamine (PDA), is an interesting biomaterial, whose functions include but are not limited to strong adhesion and broadband monotonic absorption^{[35, 36]}. Because of its distinctive capacity for spontaneous oxidation and polymerization into adherent films on numerous substrates, self-assembled PDA with the homogeneous thin film forming process has been exploited as an organic polymer memristive material^[37−40]. However, these reported PDA-based memristors still have a large V_th and even can only be written once, which are inappropriate for the construction of artificial spiking neurons.

In this study, in order to improve the memristive performance of PDA, we introduce a rapidly and scalably prepared PDA memristive film based on the electropolymerization technique for the first time. Compared with the existing PDA-based memristor, the as-fabricated electropolymerized PDA-based memristor here shows remarkable and uniform memristive performance with a V_th less than 0.3 V, a thin thickness of 16 nm, and a large C_p about 0.1 μF∙mm⁻². These device characteristics are suitable for the construction of artificial spiking neurons, which addresses the aforementioned challenges effectively. Leveraging its threshold switching behavior, the prepared sandwich PDA-based artificial neurons (Ag/PDA/Au) on the memristor array (Figs. 1(b) and 1(c)) could output the spiking voltage with an oscillation frequency of 4−75 Hz under the current stimulation (0.5–10 nA), and had a low power consumption of 10 pJ per spike. Such current-activated spiking neuron successfully mimicks the action potential (spikes) activity of chemical synaptic neurons (Fig. 1(a)).

Dopamine hydrochloride and tris-buffered saline (TBS) were purchased from Sigma-Aldrich. High-purity silver wires (5N), gold, and chromium targets were obtained from ZhongNuo Advanced Material (Beijing) Technology Co., Ltd. All chemicals and materials were used directly without further purification.

The fabrication process is shown in Fig. 2(a). In detail, first, the substrates (Si, polyethylene terephthalate (PET), etc.) were ultrasonically cleaned in acetone, ethanol, and deionized water for 10 min each in sequence, followed by drying in a nitrogen environment. Second, the bottom electrode (Cr/Au of 10 nm/90 nm) was deposited on the substrate by magnetron sputtering. Third, the electropolymerization of PDA was described as follow^{[41, 42]}: (ⅰ) 30 mL TBS containing 1 mg∙mL⁻¹ of dopamine hydrochloride was prepared as the electropolymerization solution; (ⅱ) such solution was bubbled with high-purity N₂ to remove dissolved oxygen prior to electropolymerization; (ⅲ) the Au-coated substrate, a standard calomel electrode (SCE), and a Pt sheet electrode were used as the working electrode, the reference electrode, and the counter electrode, respectively; (ⅳ) the dopamine electropolymerization was conducted by cyclic voltammetry (CV) from −0.5 to 0.5 V versus SCE for 15 cycles at a scan rate of 0.02 V∙s⁻¹; (ⅴ) after electropolymerization, the prepared PDA-based memrisitve film on the Au bottom electrode was washed with deionized water and used in the further experiments. Finally, the top active electrodes (Ag of 50 nm) with a size of 75 μm × 75 μm were uniformly deposited on the PDA-based memrisitve film through photolithography and thermal evaporation.

The properties and composition of electropolymerized PDA were characterized by AFM (atomic force microscope), XRD (X-ray Diffraction), Raman spectrum, FTIR−ATR (Fourier transform infrared spectroscopy−attenuated total reflectance), XPS (X-ray photoelectron spectroscopy), and ultraviolet−visible (UV−vis) spectroscopy. The electrical and spiking properties of the PDA-based memristor were measured by Keysight B1500A Semiconductor Device Parameter Analyzer linked with a probe station at the atmospheric environment (50%−60% RH and 23−27 °C).

The threshold switching (TS) behavior of the PDA-based memristor under positive voltage sweeps (0−1 V) is shown in Fig. 1(d). In the first forming step, the carrier transport behavior in its high resistance state (HRS) consists of three parts: the Ohmic conduction (linear part), the Child’s law (quadratic part), and the steep current increase part, which are consistent with the typical space-charge-limited conduction (SCLC) effect and the formation of filament. Moreover, in its low resistance state (LRS), Ohmic conduction is dominant. These phenomena may indicate the formation and rupture of Ag filaments between top and bottom electrodes in the PDA memrisitve layer^{[37, 38, 40, 43]}. After forming, the PDA-based memristor can self-reset to HRS without a negative voltage sweep, and subsequently set to LRS at a lower positive V_th of 0.3 V. Such TS behavior may be related to the reformation and self-dissolution of the residual Ag filaments. Fig. 1(e) schematically illustrates the TS mechanism according to the above analysis. Notably, the active electrode, Ag, can be oxidized to Ag ions being injected into the PDA layer under a positive electrical field, and the ion carrier transport mainly relies on the defects in the PDA before forming^{[31, 44]}.

The whole fabrication process of the PDA-based memristor is rapid and mass-produced, including magnetron sputtering, thermal evaporation, and electropolymerization (Fig. 2(a)). Importantly, compared to other memristors based on the self-polymerized PDA, our electropolymerized PDA can reduce the preparation time (25 min here rather than hours or even days in other reports)^[37−40]. Fig. 2(b) exhibits typical CV curves recorded during dopamine electropolymerization. As electropolymerization progresses, the current on the working electrode diminishes, indicating that the thickening PDA layer impedes the subsequent electrochemical reaction. As shown in Figs. 2(b) and 2(c), the obtained voltammetric peaks can be attributed to two redox couples of dopamine/dopaminequinone (ⅰ and ⅰ*) and leucodopaminechrome/dopaminechrome (ⅱ and ⅱ*)^{[41, 45, 46]}. The electropolymerized ultrathin PDA layer (~16 nm) was confirmed by SEM (Fig. 2(d)) and AFM (Fig. 2(e)).

The origin of the superior memrisitve performance of electropolymerized PDA layers has been investigated through some key characterization techniques. Fig. 2(f) shows the XRD pattern, and the wide diffraction peak at 2θ ≈ 24° suggests that PDA is an amorphous phase^[38]. As shown in Fig. 2(g), the D (~1393 cm⁻¹) and G (~1607 cm⁻¹) peaks in the Raman spectra attribute to sp² carbonaceous materials in the PDA layer^[47], which correspond to the stretching and deformation of catechol, respectively^[48]. Notably, the large D peak is related to abundant structural defects in the hexagonal carbon lattice of the PDA layer. As a result, both the ultrathin and defect-rich characteristics lead to the low V_th of our PDA-based memristors, where silver ions are easily transported through the defects after oxidation under an electric field.

Other characterizations, such as FTIR−ATR and XPS were utilized for the functional group analysis of PDA. The FTIR bands in Fig. 1(h) and the XPS peaks in Figs. 1(i) and 1(j) correspond to specific functional groups of PDA, which is in agreement with former detailed studies^{[49, 50]}. Interestingly, the electropolymerized PDA may be a disordered organic semiconductor due to its UV−vis spectra with an onset absorption edge of 610 nm (Fig. 1(k)).

For the electrical characterization of the PDA-based memristor, the measurements were conducted by electrically grounding the Au bottom electrode and applying direct current sweeps to the Ag top electrode. The current compliance (I_cc) was set at 100 nA to avoid the strong Ag filament formation under a strong current, which ensures the stability of the threshold switching behavior^[51]. Fig. 3(a) shows the forming step and typical threshold switching behavior. After forming, the PDA-based memristor can be easily switched from HRS to LRS when the sweeping voltage reaches V_th ~ 0.3 V, and self-reset from LRS to HRS when the sweeping voltage is lower than a hold voltage (V_hold ~ 0.01 V). Fig. 3(a) also shows the consecutive 300 sweeping cycles, demonstrating the stability of volatile threshold switching behavior. In detail, there was little variation in V_th, V_hold, and HRS in these 300 sweeping cycles, as demonstrated in Figs. 3(b)−3(d), respectively.

Additionally, the volatile PDA-based memristor shows a steep slope of 2.1 mV/dec in the steep current increase part during threshold switching (Fig. 3(e)). The PDA-based memristor is a unipolar resistive switching device due to the asymmetric electrode structure. As shown in Fig. 3(f), no resistive switching occurred under a negative voltage sweep, but a distinct parasitic capacitance effect appeared (from ⅳ to ⅴ). Fig. 3(g) illustrates the non-volatile resistive switching behavior of the PDA-memristor at a large I_cc of 1 mA.

Owing to the large C_p of the PDA-based memristor, the artificial spiking neuron can be constructed in a capacitor-free configuration. Fig. 4(a) illustrates the equivalent circuit of the artificial current-activated spiking neuron. The C_p here was obtained by the impedance analyzer to be about 0.1–1 μF∙mm⁻², which is a considerable value comparable to that of other inorganic materials^{[8, 25, 28]}. By applying a constant current bias (I_in), the artificial spiking neuron can output the spiking voltage (V_out), whose oscillation frequency increases from 4 to 75 Hz as I_in increases from 0.5 to 10 nA (Fig. 4(c)). As shown in Fig. 4(d), the spiking voltage waveform exhibits the charge-discharge process corresponding to spontaneous and synchronous resistive switching^[25]. The power consumption per spike does not exceed 10 pJ under different current stimulations according to the calculation.

In summary, we have reported an electropolymerized PDA-based memristor with improved memrisitve performance, including a low V_th of 0.3 V, a small thickness of 16 nm, and a high C_p of about 1 μF∙mm⁻². Combining these device properties with its stable threshold switching behavior, we have successfully constructed an artificial current-activated spiking neuron. This artificial neuron could generate oscillation voltages with the input-current-dependent spiking frequency and a low power consumption of 10 pJ per spike. These results show the promising potential of the PDA-based artificial neuron for neuromorphic applications, such as robotics, prosthesis, and human-machine interfaces. In the future, the memristor’s performance can be further improved by co-electropolymerizing dopamine with other organics.