1. Introduction
The nervous system of all organisms, including brain structures, is composed of two types of synapses: electrical and chemical. In electrical synapses, the cytoplasm of adjacent neuros is directly connected by gap junctions, allowing the bidirectional passage of electrical currents. In contrast, chemical synapses transmit signals by releasing neurotransmitters from the presynaptic neuron, which diffuse across the synaptic cleft and then bind to receptors on the postsynaptic neuron[1, 2]. Both chemical and electrical signals are essential for the functions of living organisms. In recent years, artificial synapses using electronic devices have garnered significant attentions due to their ability to mimic biological synapses. Although enormous efforts have been devoted to developing artificial electrical synapses, chemical synapses have received comparatively less attention[3]. Neurotransmitters can be classified based on their chemical structures and functional characteristics. Major neurotransmitters include amino acids (e.g., glutamic acid and tyrosine), serotonin, choline (e.g., acetylcholine), catecholamines (e.g., epinephrine and dopamine), peptides (e.g., endorphins) and soluble gases (e.g., nitric oxide)[4, 5]. These neurotransmitters are produced by chemical reactions and transmitted through processes such as endocytosis and exocytosis[6, 7]. Many diseases originate from abnormal levels of neurotransmitters, including physical, psychotic, and neurodegenerative conditions such as Alzheimer, dementia, addiction, Parkinson, depression, and schizophrenia[4, 8−11]. Therefore, understanding neuromodulatory processes in the human brain or animals through monitoring target neurotransmitters can help to shed light on complex behaviors in contemporary neuroscience. The artificial synapses with neurotransmitter-mediated functions are capable of utilizing neurotransmitters to build spiking neural networks (SNN) and artificial neural networks (ANN) that mimic the function of biological neural networks. These devices hold great potentials to overcome the limitations of traditional von Neumann computing architectures. Incorporating neurotransmitter functionality into these devices enables the effective modelling of the complex synaptic interactions in the human brain. This, in turn, enhances the capabilities of brain-computer interfaces (BCIs) and facilitates connection seamlessly and intelligently with biological systems[12].
Achieving synaptic modulation based on biochemical signaling detection is the first step towards a biologically integrated neuromorphic system[12]. Nevertheless, prior to neurotransmitter regulation, it is crucial to detect low concentrations of neurotransmitters for artificial synapses. Given that neurotransmitter concentrations in biological samples are typically in the nanomolar range[11], it is necessary to develop highly sensitive, selective, and reliable biosensors for neurotransmitter detection. Once accurate detection is achieved, the next step is to implement the simulation of neurotransmitter-regulated synaptic behaviors, including synaptic plasticity, action potentials, neurotransmitter release, etc. This simulation enables the construction of ANN and SNN to enhance brain-computer interface capabilities. Subsequently, a complete circuit for biologically integrated neuromorphic systems is required, where the operation process can be continuously corrected through feedback within a closed-loop structure[13, 14].
Organic electrochemical transistors (OECTs) are electronic devices that rely on organic mixed ionic–electronic conductors (OMIECs) and can operate in aqueous environments[15]. Additional advantages of OECTs include low operating voltage, good biocompatibility, and high transconductance[16, 17], making them ideal platforms for interfacing with biological systems for biochemical and electrophysiology sensing. These attributes are particularly beneficial in biochemical sensors, point-of-care diagnostics, and implantable and wearable technologies[18−24]. Note that OECTs possess a similar structure to biological synapses as shown in Fig. 1, with channels forming the synaptic gap of artificial chemical synapses. The gate electrode corresponds to the presynaptic membrane, which receives the stimulus, and the drain electrode resembles the postsynaptic membrane, collecting the postsynaptic current (PSC) for further signal processing[25]. Therefore, OECTs are an ideal platform for biomimic neurotransmitter-synapse interactions.

While several good review articles related to artificial synapses[18, 21, 26−29] and bioelectronic interfaces[19, 23, 30, 31] have been published recently, there is a notable lack of comprehensive summaries specifically on artificial synapses mediated by neurotransmitters and resultant neuromorphic systems, which are key for biomimic and bioinspired neuromorphic systems. This review article aims to fill this gap. In the first part of this contribution, the device structure and operational mechanism of OECTs are introduced. The second part focuses on the advances on the detection of a representative neurotransmitter, dopamine (DA), which can modulate synaptic transmission between neuros and thereby influence synaptic conditioning and plasticity. The detection principles and methods to improve sensor performance as well as in vivo detection are also summarized. The third part discusses the neurotransmitter-mediated artificial synapses, including synaptic plasticity, and examples related to simulation of biological synapses and biologically integrated neuromorphic systems. Finally, the possible prospects and future research directions for future biologically integrated neuromorphic systems through neurotransmitter-mediated artificial synapses are proposed.
2. Structure and operational mechanism of OECTs
OECTs are three-terminal devices, with a drain, source and gate electrodes. Same with metal−oxide semiconductor field-effect transistors (MOSFETs) and organic field-effect transistors (OFETs), OECTs can be considered as amplifiers or switches, where the input signal (VG) amplifies and controls the output signal (ID). Distinct from OFETs, the gate electrode of OECTs is isolated from the channel through the use of an electrolyte. The electrolyte can be a liquid, solid, or gel[15, 32, 33]. The channel layer in OECTs is an OMIEC, which is both electronically and ionically conductive, and the electrolyte with ionic conductivity can promote the movement of ions under the action of an electric field. OECTs can be divided into two types according to different working states: 1) The channel material conducts electricity and typically operates in depletion mode (ON to OFF), and the input (VG) causes electrochemical dedoping, thereby reducing the electrical conductivity of the channel (Fig. 2(a)). 2) The channel material is poorly conductive and operates in the accumulation mode (OFF to ON), and its doping requires the application of a gate voltage (VG), thereby increasing the conductivity of the channel (Fig. 2(b)). The most popular semiconductor material in OECTs is poly(3,4-ethylenedioxythiophene):poly(styrene sulfonate) (PEDOT:PSS), which operates in the depletion mode[22].
According to Bernards' model[34], OECT can be divided into two circuits (Fig. 2(c)). The first is the electronic circuit, modeled as a resistor (RCH), which considers the movement of charge carriers within the source−channel−drain structure. The second is the ionic circuit in the gate-electrolyte-channel configuration. A resistor (RE), gate capacitor (CG) and channel capacitor (CCH) were involved. As a main figure of merit, the transconductance (gm) of OECTs is defined as the ratio of the change in ID to the change in VG, and can be expressed as (Eq. (1)):
gm=∂ID∂VG=WdLμC∗(VT−VG), |
(1) |
where d is the channel thickness, L is the channel length, W is the channel width, C* is the volumetric capacitance, VT is the threshold voltage, and μ is the charge carrier mobility. Increasing W and decreasing L can effectively enhance gm, but increasing d is a more practical approach for the gm improvement due to the challenge of device miniaturization[35, 36]. Note that a thicker channel layer typically leads to a slower response time for OECTs because of the longer pathways for ion penetration. Another method for gm improvement is to design and synthesize new channel materials with higher μC*[37, 38]. Besides, switching times, on/off current ratio (Ion/Ioff), and stability also need to be considered to comprehensively evaluate the performance of OECTs[39, 40].
3. Detection of DA
3.1 Working principle of OECTs for DA detection
The working principle of the DA sensor based on OECTs is shown in Fig. 3(a). Within the electrolyte, DA molecules can undergo electro-oxidation in the form of o-dopaminequinone at the gate electrode, and therefore a faradic current is generated[41]. The advantages of high transconductance, low operating voltage and good aqueous compatibility allow OECTs amplifying the weak signal of DA in solution. This process decreases the potential drop at the electrolyte/gate interface and increases the effective gate voltage (Vg-eff) (Fig. 3(b)), as detailed by[42]:

ΔVg-eff={2.30(1+γ)kT2qlog[C]higher concentrationA×[C]βlow concentration, |
(2) |
where [C] is the DA concentration, T is the absolute temperature, k is the Boltzmann constant, q is the charge of an electron, A and β are constants determined by fitting experimental data, γ is the ratio between the volumetric capacitance of the channel (CCH) and the capacitance of the gate-electrolyte interface (CG). In the higher concentration range, the potential ΔVg-eff will increase linearly with the increase in the logarithm of [C]. However, in the low concentration range (below nM), the nonlinear relation was often observed[41]. Therefore, Zhang et al. proposed an empirical relationship in which ΔVg-eff is power-exponential to [C][43]. In practice, the sensitivity of the device at low concentrations needs to be calibrated.
The real-time ID changes of an OECT device during adding different concentrations of DA into phosphate buffer saline is shown in Fig. 3(c), with the fitted calibration curve (Fig. 3(d)). The sensor performance can be evaluated by two key parameters that are extracted from the calibration curve: the sensitivity and the limit of detection (LOD). The sensitivity defines the minimum input required to generate a detectable output response from the sensor. The most common unit of sensitivity is mV/dec, while some reports also used current change to describe the sensitivity with a unit of mA/dec or the gm change with a unit of SM−1[44, 45], while the LOD indicates the lowest analyte concentration that can be reliably detected compared to a blank measurement. In this detection principle, selectivity is defined as the ratio of the slope of the calibration line for the analyte of interest to the slope of the specific interferent. However, in practical measurements, less attention is paid to this parameter[46]. Another parameter that needs to be considered is the linear range of DA detection, where the output is proportional to the input. Within this range, the sensitivity is theoretically maintained at a constant level, thereby enabling the sensor to ensure a certain level of measurement accuracy as the range increases. The greater the linear range of the sensor, the larger its detection range, which is essential for ensuring measurement accuracy[47].
From Eq. (3), ΔVg-eff is proportional to log[C]. The slope of the fitted curve can be modulated by controlling the capacitance ratio γ[41]:
γ=CCHCG=cchAchcgAg, |
(3) |
where AG and ACH are the areas of gate and channel electrode and cg and cch are the gate and channel capacitances per unit area, respectively. The capacitance ratio is proportional to ACH/AG. Therefore, optimizing the ratio of the channel area to the gate area in OECTs can enhance the performance of DA detection. Additionally, it has been reported that alternating current (AC) methods offer greater sensitivity and more stable data in sensing applications compared to traditional direct current (DC) methods[48].
3.2 DA sensors based on OECTs
DA is an electrically active molecule that can be oxidized without the enzymes, and therefore it can be quantified through the electrochemistry technique. In particular, the oxidation peak of DA is 150 mV (compared to Ag/AgCl)[46]. However, a few challenges still exist. On one hand, most reported electrochemical DA sensors exhibited linear detection ranges from 1 to 100 mM with the detection limits around 1 mM[49−51], which are orders of magnitude higher than the DA concentrations in clinic samples (nM for plasma, μM for urine and fM for single adrenal chromaffin cell, respectively[52, 53]). On the other hand, DA has many analogues, and a number of potential interferences like ascorbic acid (AA) and uric acid (UA) also complicate the specific detection[46, 54]. Interestingly, the high transconductance of OECTs allows sufficient signal amplification and consequent detection limit of DA down to the nM, suggesting the detection capability in real samples like urine and plasma[41]. OECTs-based DA sensors have a faster response time compared to conventional methods that typically require several minutes of response time, often within seconds of detection[55, 56]. Besides, the specificity for DA sensing can be improved by modifying various active materials on the gate of OECTs. Table 1 summarizes the characteristics and structures of recent DA detectors based on OECTs.
Gate materials | Gate area (μm2) | Channel size (W/L (μm/μm)) | LOD (M) | Linear range (M) | Sensitivity (mV/dec) | Ref. |
Aptamer-enhanced Au | / | 30/22 | 5 × 10−16 | 5 × 10−12−10−9 | / | [57] |
9 × 106 | / | 1 × 10−14 | 10−11−10−8 | 6.9 | [58] | |
Pt | 9 × 106 | 22/5 | 1 × 10−9 (AC) 1 × 10−8 (DC) | 10−7−10−5 | 45.2 | [48] |
4.8 × 105 | 40/10 | 3 × 10−8 | 3 × 10−8−1 × 10−7 | / | [42] | |
/ | 6000/100 | <5 × 10−9 | 5 × 10−8 −3 × 10−6 | 174 | [41] | |
Vertical Pt wire | 2.8 × 105 | 700/1900 | 4.4 × 10−6 | / | 410 | [59] |
Nafion and graphene flakes co-modified Pt | / | 6000/200 | 5 × 10−9 | 5 × 10−9−1 × 10−6 | 242 | [60] |
Overoxidized MIP (o-MIP/Pt) | / | 1000/20 | 3.4 × 10−8 | 4 × 10−7−1 × 10−5 | / | [61] |
Ag/AgCl pseudo reference | / | 10/10 | / | 1 × 10−6−3 × 10−4 | 279 | [62] |
/ | 5000/1550 | 4.3 × 10−8 | 1 × 10−9−1 × 10−4 | 0.312 mA/dec | [44] | |
/ | 1500/5000 | 3.7 × 10−8 | 5 × 10−8−1 × 10−4 | / | [63] | |
Nitrogen/oxygen-codoped carbon cloths | / | / | 1 × 10−9 | 1 × 10−6−3 × 10−4 | 151 | [39] |
Carbon/PEDOT: PSS | 3.9 × 10−2 | 0.45 × 0.45 × π (needle-typed) | 1 × 10−12 | 1 × 10−9−1.6 × 10−7 2 × 10−9−7 × 10−6 | 190 ± 20 pA/dec 210 ± 40 pA/dec | [64] |
CNT/ (Pt NPs) fiber | / | / | 5 × 10−9 | 5 × 10−9−5 × 10−7 | / | [65] |
rGO/CNT/PEDOT:PSS | / | 240/77 | 6 × 10−6 | 1 × 10−5−1 × 10−4 | / | [66] |
Polypyrrole nanofiber | / | / | 1 × 10−9 | 1 × 10−9− 1 × 10−6 | / | [67] |
PEDOT:PSS | 9 × 106 | 3000/3000 | 6 × 10−6 | 5 × 10−6−1 × 10−4 | / | [68] |
Copper phthalocyanine | / | 166/20 | 3.04 × 10−6 | 5 × 10−5−5 × 10−4 | / | [16] |
By employing various gate electrodes (gold, graphite, and platinum (Pt)), Tang et al.[41] realized DA detection using OECTs. It was found that the OECT sensitivity were dependent on the gate electrode's characteristics and the operational voltage. Due to the excellent catalytic properties of Pt, the device with a Pt electrode at a gate voltage of 0.6 V had the highest sensitivity. Moreover, a LOD of less than 5 nM for DA was achieved, indicating a significant improvement over conventional electrochemical methods with the same electrode. But this sensor had poor selectivity and sensitivity. Several research groups have also used Ag/AgCl pseudo reference electrode[44, 62, 63] as gate electrode. Although they could achieve good performance due to unpolarizable property[38], metal electrodes were more advantageous in terms of integration. Thereafter, a wide variety of DA sensors based on OECTs were developed through gate modification and architecture design.
3.2.1 Gate modification
Modifying the gate electrode with various materials in OECTs can effectively enhance both the selectivity and sensitivity for DA detection. These modifications help reduce interference and get a lower LOD. Liao et al.[60] coated the Pt gate surface with a biocompatible polymer Nafion or chitosan, by which the interference induced by UA and AA was effectively eliminated, particularly after the modification with Nafion. Additionally, the incorporation of graphene flakes on the gate electrodes further increased the sensitivity, achieving a DA detection limit of 5 nM. This value was significantly lower than that of conventional electrochemical methods. Besides, widely explored nanomaterials such as gold NPs[63], platinum NPs[65], or various composite materials[67] were also sufficient to enhance the performance of OECTs-based DA sensors as the modification layer.
3.2.2 Architecture design
Different from commonly used planar device structure, Mariani et al.[64] designed a needle-type OECT for spatially resolved detection of DA. Due to its unique structure, the sensor featured a high gate-to-channel area ratio. Besides, this study combined the advantages of carbon nanoelectrodes (CNEs) with OECTs to realize nanoscale pioneer OECTs, where both single- and double-tube structures were used with a conductive polymer film on the CNEs. Intriguingly, this needle-type OECT had a size similar with a single cell, and therefore high spatial resolution detection of neurotransmitters can be realized. By applying appropriate drain and gate voltages, this OECT sensor achieved a LOD on the order of pM and a linear DA detection range from 2 nM to 7 μM. Other OECT architecture such as self-curling[16] and fiber-shaped structures[59, 65, 67] were also proposed with good detection performance.
3.2.3 MIP-based electrochemical sensors
MIPs[69] are useful materials for selective binding, where a template molecule (target) is dissolved in a suitable solvent with one or more monomers and a cross-linking agent to form a highly cross-linked polymer. Following polymerization, the template molecules are removed, and consequently nanocavities are formed that correspond to the size, shape, chemical interactions and orientation of the template molecules[70]. Therefore, the gate modification using MIPs was found to be an effective approach for high-performance OECT DA sensor[61]. The flow of electrochemical synthesis of overoxidized MIP (o-MIP) on OECT gate is shown in Fig. 4(a).

The OECT sensor with an o-MIP/Pt gate showed a LOD of 34 nM, which is almost identical to the one with bare Pt. However, the selectivity was remarkably improved, and a DA/AA signal ratio over 5 ranging from ~0.4 to ~10 μM was obtained.
3.2.4 Aptamer-based electrochemical biosensors
Aptamers are a superior option for DA sensors, irrespective of cost considerations. Aptamers, which are made up of single-stranded RNA or DNA molecules, have the ability to bind detection molecules with outstanding affinity and specificity[71]. In contrast to large antibodies, aptamers typically have a lower molecular weight. In addition, aptamers can be used to develop biosensors with specificity and fast response due to their conformational changes upon binding. A considerable number of naturally occurring aptamers have been identified in the untranslated regions of mRNA. These aptamers have been demonstrated to bind small molecule metabolites[71, 72]. As a result, many efforts have been made to extract DA aptamers and have been used in sensors[46, 57, 73, 74].
Soliman et al.[58] developed aptamer-enhanced OECTs for ultra-sensitive DA detection. Fig. 4(b) illustrates the operational principle of aptamer immobilization in conjunction with an integrated OECT with a functionalized gate. The sensor demonstrated real-time detection with a linear response from 10 fM to 10 nM and a detection limit of 10 fM. In addition, Liang et al. reported similar aptamer sensor, where a fM level DA detection was also obtained with high sensitivity[57]. This provides an efficient platform for future disease diagnosis and clinical applications. However, more efforts are needed to ensure reusability and versatile integration with other devices.
3.3 In vivo detection
Real-time in vivo detection of DA is necessary to build integrated bio-logical neuromorphic systems. However, the complex environments that usually accompany in-vivo situations in living organisms require the fabrication of more well-designed structure with higher sensitivity and stability[11]. This can be addressed by using OECT devices. Li et al.[75] demonstrated a fast-scanning potential (FSP)-gated OECT, where a sensitivity of 0.899 SM−1 and a LOD of 5 nM were reported. Importantly, the real-time dopamine monitoring was realized using this device in the living rat brain. Unlike the previous individual devices, Xie et al.[42] fabricated an OECT array for monitoring DA in rat brains (Fig. 5(c)). The OECT array consists of multiple OECTs that can detect transient DA releases in real time with nanomolar scale LOD and high temporal resolution. Notably, continuous in vivo operation for several hours was reported at an operating voltage under 1 V without significant degradation. Additionally, these OECT arrays were capable of simultaneously detecting evoked DA release in multiple striatal brain regions.

4. DA-mediated artificial synapses
4.1 Synaptic plasticity
Synaptic plasticity, a biological process, is defined as the alteration of synaptic weight under certain conditions. It is regarded as a fundamental mechanism that underlies the ability of the brain to adapt and learn[79]. Simulating synaptic plasticity through OECTs is a prerequisite for constructing ANNs and SNNs. In addition, neurotransmitters are important for synaptic plasticity. In response to the activation of receptors by a variety of neurotransmitters, chemical synapses amplify and transform presynaptic signals, which in turn have a complex effect: excitatory postsynaptic potential (EPSP) or inhibitory postsynaptic potential (IPSP) on the postsynaptic neuron[28].
Based on the duration of synaptic weight changes, synaptic plasticity can be categorized into two distinct types: short-term (STP) and long-term plasticity (LTP). Taking the neuronal excitation process as an example, a short single presynaptic voltage spike triggers a sharp increase in excitatory postsynaptic current (EPSC) to a level, A1, which then quickly returns to its baseline. This behavior is called STP (Fig. 5(a))[76]. If a subsequent voltage pulse is applied before the EPSC has fully decayed, the current A2 is higher than A1. This behavior is defined as paired-pulse facilitation (PPF), and its strength is quantified as A2/A1 (Fig. 5(b))[76]. LTP occurs in response to a high frequency of presynaptic action potentials. With increasing the number of presynaptic spikes applied to the organic synapse, a transition of synaptic response from STP to LTP is possible (Fig. 5(c))[77, 80]. During LTP, the synaptic weight increases, and then decays slowly to a certain state that maintains for a long time.
4.2 Artificial synapses with DA-mediated plasticity
Achieving synaptic modulation based on biochemical signaling detection is the first step towards a biologically integrated neuromorphic system[12]. Significant contributions to devices that simulate DA recognition and release have been made, bridging the connection between biological neural networks and artificial intelligence systems[81, 82]. The neuromorphic response induced by the redox reaction of DA is one of the most efficient ways to realize this process. Table 2 summarizes the DA-mediated artificial synapse based OECTs.
Type | Gate materials | Key properties | Details | Ref. |
Biohybrid synapse | Au/PEDOT: PSS | DA recycling/LTP/LTD | OECT/microfluidic channel | [12] |
PEDOT: PSS fiber | DA recycling | Fiber OECT | [78] | |
Artificial spiking neuron | Ag/AgCl | DA/ions sensitive and neuron spiking properties | OECT | [85] |
Au | DA/5-HT sensitive and neuron spiking properties/ biological functions of a retinal pathway | OECT/microfluidic channel | [86] | |
Artificial synapses | Ag ink | STP/LTP/flexible | Fully printed dual-gate OECT | [83] |
ITO | Mimicking biological neurotransmitter corelease | Electrical and optical modulation OECT/microfluidic channel | [84] | |
Closed-loop neuromorphic system | PEDOT: PSS | Closed-loop control, actuation and reinforcement learning | OECT/microfluidic channel/robotic hand | [14] |
Keene et al.[12] demonstrated a functional biohybrid synapse, in which a dopaminergic presynaptic domain of PC-12 cells was coupled to an organic neuromorphic device serving as the postsynaptic domain. This design allowed OECTs as artificial synapses direct contact with living tissue and adaption through biofeedback. Furthermore, the electrochemical simulation of long-term potentiation and long-term depression of synaptic weights induced by DA were also achieved. Additionally, polydimethylsiloxane (PDMS) microfluidic channels allowed for precise control of fluid flow within the device, which mimicked DA endocytic recycling of in biological synapses. Subsequently, several successive studies[78, 83, 84] have investigated artificial synapses with DA-mediated plasticity based on OECTs. DA-mediated plasticity is markedly distinct from the short-term synaptic behavior observed in electrolytes devoid of DA. The rationale for this phenomenon can be elucidated by the chemical reaction where DA is oxidized to DA o-quinone in the presynaptic compartment, resulting in the simultaneous production of 2e− and 2H+. These generated H+ ions then inserted into the channel, where they combined with PSS− to form stable sulfonic acid groups, thereby facilitating LTP[83].
The device structure is important in neuromorphic devices. Fibers are lightweight, flexible, breathable and comfortable compared to thin-film structures[87, 88]. These characteristics confer a significant advantage in the field of wearable technology and BCIs. Alarcon-Espejo et al.[78] developed the fiber-electrochemical neuromorphic organic devices (Fiber-ENODes), and demonstrated that these high hole mobility fibers hold great potential as foundational elements for future bio-hybrid technologies. Fig. 5(d) showed the structure of artificial bio-hybrid synapse based on double PEDOT: PSS fiber. It was revealed that the optimization of electrode/organic interface and fiber alignment resulted in stable contacts, high μC* products, and high hole mobility. The device respond to DA was found to affect STP and LTP for Fiber-ENODes. Moreover, a shape factor very similar to neuronal axon terminals was also observed, as shown in Fig. 5(e).
4.3 Biologically integrated neuromorphic systems with DA-mediated artificial synapses
BCI technology is a transformative human-computer interaction technology[31, 89]. Its mechanism of action is to bypass peripheral nerves and muscles and directly establish a brand new communication and control channel between the brain and external devices. Artificial synapses based on OECTs are expected to accelerate the development of BCI technology as a bridge between machines and living organisms, thereby enabling biologically integrated neuromorphic systems. Currently the detection of electrical signals in organisms has been successfully realized by OECTs[20, 90, 91]. In addition to electrical signals, chemical signals are also a crucial part of living organisms[92]. Therefore, artificial synapses with neurotransmitter-mediated is of great importance in future applications of BCIs.
Sarkar et al.[85] fabricated an organic artificial neuron (OAN) using a compact nonlinear electrochemical element. This device with inherent biosensing capabilities could operate in liquid environments, as shown in Figs. 6(a) and 6(b), and more importantly, mimics the spiking behavior of neurons. Its sensitivity to ion concentration and dopamine concentration was confirmed, facilitating the modulation of spiking excitatory behavior (Fig. 6(c)). Matrone et al. realized a more complex system[86]. This system used OANs to mimic afferent neurons that received light stimulation and employed synapses modulated by DA and Serotonin (5-HT) to simulate the neuromodulatory activity of intermediate neurons. As shown in the Fig. 6(d), an OECT synapse can modulate the spike pulses from the preceding neuron based on DA concentration. As concentration increased, the synaptic output significantly enhanced, leading to an increase in the spike pulse frequency of the subsequent neuron. The degree of modulation depended on the concentration of the neurotransmitter. This process effectively simulated the spike-rate coding dependent on neurotransmitters in intermediate neurons and the transmission of neural signals between multiple neurons.

In order to control the machine by the organisms, Bruno et al. designed an OECT platform through neurotransmitter closed-loop control, actuation and reinforcement learning[14]. The overall neuromorphic closed-loop architecture was shown in Fig. 6(e), where the standard silicon technologies were utilized. This design enabled brain-inspired computing through adaptive synaptic potentiation and depression within a closed-loop system. Specifically, it utilized DA release and detection to facilitate closed-loop learning and training processes akin to those in the human brain. Notably, a robotic hand could be interfaced and controlled by this microfabricated platform, as shown in Fig. 6(f), and autonomous learning to grasp objects of different sizes was also realized. Many efforts have been made by researchers to realize bio-integrated neuronal networks. However, the field is still in its infancy.
5. Conclusion and outlook
In recent years, the development of OECT technology and the trend of multidisciplinary intersection have facilitated the rapid development of OECT-based DA sensors. In this review, an overview of the structure and operating mechanism of OECTs is first presented. Then a comprehensive review of the advances in DA sensors and artificial synapses based on OECT technology is discussed. By combining DA detection and modulation, artificial synapses enable chemical communication with neurotransmitter interactions. The platform has significant implications for neuroscience, disease diagnosis, neuroprosthetic control, bioelectronic medicine, and the integration of brain and machine intelligence. However, many challenges remain to be addressed. The first problem faced is how to reduce power consumption and improve long-term stability. As more devices are integrated, reducing power consumption to levels comparable to that of the biological brain is becoming increasingly critical. Furthermore, the working life on a yearly basis is required for the implantable devices, and therefore the long-term stability of current OECTs should be significantly improved. The second is to increase the diversity of sensing. While considerable progress has been made in DA detection, synaptic events are usually the result of multiple neurotransmitters acting together[93, 94]. Therefore, there is a high demand to develop sensors that can detect multiple neurotransmitters simultaneously. The third is to achieve a closed loop of OECTs and neurotransmitter signaling. OECTs can simulate synaptic plasticity, spike modulation and adaptive learning, but integration with biological systems requires the bidirectional communication[95]. This requires the use of BCI technology to create a closed loop between the artificial synapse and the organism.
Acknowledgments
This work is supported by the National Natural Science Foundation of China (Grant No. 62074163) and Beijing Natural Science Foundation (Grant No. JQ24030).