Introduction

With the rapid advancement of artificial intelligence technologies, there has been growing interest in neuromorphic hardware that mimics the brain’s highly interconnected and parallel-processing neural networks1. Among the various candidates for such systems, organic electrochemical transistors (OECTs) have emerged as a promising option due to their unique ion-intermixing interface effects, low operating voltage, flexibility, and biocompatibility2,3,4. Leveraging electrochemical modulation mechanisms, OECTs have successfully mimicked various synaptic behaviors, such as excitatory postsynaptic currents (EPSC), paired-pulse facilitation (PPF), long-term plasticity (LTP)5,6. Recent studies have demonstrated that OECT-based neuromorphic devices can achieve high accuracy in neural network simulations7,8,9. The biological nervous system, however, comprises a diversity of neuronal components with distinct dynamics. For example, one category exhibits volatile characteristics, ensuring its current state is shaped by recent inputs while remaining independent of distant past events. These neurons are typically employed to pre-process information. In contrast, another category relies on non-volatile properties to project low-dimensional signals into a high-dimensional state space. These neurons are generally responsible for integrating and making inferences based on the pre-processed signals10,11,12. Therefore, the development of future neuromorphic systems hinges on the ability to emulate and integrate a diverse set of neural behaviors13.

Recent studies have focused on tuning device functionality via material design and operational parameters14,15,16. For example, varying the ion types of the electrolyte enables control over the device’s volatility and non-volatility5,17. Through the optimization of anion species, Wang and co-authors achieved a remarkable enhancement in state retention, realizing a duration of ~1000 s while maintaining 64% of the peak current17. In addition, Liu et al. designed two n-type Organic mixed ionic–electronic conductors (OMIECs), featuring hydrophilic ethylene glycol and hydrophobic butyl side chains, respectively. The oxygen free side chains of OMIEC induce dense molecular aggregation and ordered π-stacking, yielding a larger hysteresis window18. Ji et al. converted the OECT operation from volatile to non-volatile by incorporating polytetrahydrofuran as a dopant into the polymerized poly(3,4-ethylenedioxythiophene): tosylate (PEDOT:Tos)19. Additionally, the OECTs can be programmed and reconfigured between volatile and non-volatile states by coordinating the gate voltage and its duration20,21. Wang et al. demonstrate a vertical OECT with a crystalline–amorphous channel that is selectively doped by ions. While low-voltage doping of the amorphous region results in volatile behavior with fast recovery, high-voltage ion trapping in the crystalline region leads to non-volatile operation22. However, the heterogeneous integration of OECTs, fabricated with different materials, into a neuromorphic system presents a key challenge of process incompatibility, which in turn escalates manufacturing complexity and difficulty. The use of voltage-regulating OECT circuits comes with the requirement for a multi-level bias architecture, adding complexity to the circuit. This leads to a question: can the functional diversity of OECTs be achieved through the regionally controlled structural design, such as the electrolyte, rather than changes in material composition or operational parameters?

The development of high-resolution electrolytes serves as the foundation for enabling regionally controlled strategy. Drop-casting, unfortunately, restricts the fabrication of small-sized devices, and the uniform nature of spin coating makes it incompatible with strategies that demand localized or patterned regulation23,24. The high resolution of photolithography is offset by the degradation that its chemical developers cause to electrolytes. This incompatibility has limited the technique’s application in patterning electrolytes25. Additive manufacturing has emerged as a flexible fabrication approach capable of high-resolution patterning, including screen printing26,27, inkjet printing28,29,30, aerosol jet printing31, and nanoimprint32. Nevertheless, it should be noted that while fully printed OECTs have been demonstrated, the simultaneous achievement of high-resolution (<100 µm) and high-capacitance electrolyte printing remains challenging due to the high ion transport requirements of the electrolytes2,33,34. Additionally, the influence of electrolyte film thickness on device performance has not been systematically investigated. Thus, the key challenge lies in fabricating high-resolution electrolyte patterns to modulate OECT functionality, which is an essential step toward highly integrated and multifunctional OECT circuits.

Here, we introduce a regionally controlled ion-doping strategy that enables functional modulation of OECTs from high-performance computing to non-volatile memory modes, using the same materials and operating parameter. The foundation of this advancement is the precise deposition of electrolytes with a controllable thickness gradient, spanning from 100 nm to several tens of micrometers, within OECT channel region. Through regionally controlled printing, we fabricated two OECTs: an ion-rich OECT with multilayers electrolyte and an ion-deficient OECT with a single layer. The ion-rich OECT features rapid ion doping, yielding sub-1ms response dynamics and a 36.38 ± 1.35 mS transconductance for computational roles. In contrast, the ion-deficient OECT possesses a slow doping rate, which grants it a charge retention capability of over 300 s and non-volatile state retention, catering to neuromorphic memory. Furthermore, we demonstrate integration of the fully-printed OECTs for neuromorphic systems that emulate human brain learning and memory, enabling processing and storage of Morse code signals. This study demonstrates a straightforward strategy to circumvent material compatibility constraints and complex circuitry requirements, facilitating the application of OECTs for constructing high-integrated neuromorphic systems.

Results

Electrochemical performance of OECTs

We implemented regional control of electrochemical doping by varying the printed thickness of the electrolyte, thus tuning OECT characteristics through structural means rather than chemical composition35. The manufacturing process of fully printed OECTs is illustrated in Fig. 1a. The device, comprising the electrodes, channel, and electrolyte, was successively fabricated by inkjet printing onto a flexible polyethylene naphthalate (PEN) substrate (Supplementary Fig. S1a–c, Supporting Information). A 3×5 OECTs array was fabricated within an area of 0.5 cm² (Supplementary Fig. S1d), showcasing the viability of this approach for creating functional circuits. Figure 1b presents a schematic representation and micrograph of a fully additive-manufactured OECT, which features a regionally controlled electrolyte layer deposited via inkjet printing over the PEDOT:PSS channel, forming a confined 3D ion transport pathway that interfaces with the gate electrode. Figure 1c–e shows the optical microscope images of the printed AgNP, PEDOT:PSS, and hydrogel, respectively, with a magnified view provided in the inset. The principle of double network hydrogel electrolyte is shown in Supplementary Fig. S2. Under UV irradiation, the photoinitiator promotes the polymerization of monomer AM and crosslinking agent MBA to form polyacrylamide (PAAm). PAAm and PVA form PVA/PAAm double network hydrogel. EG and Gly are used as solvents to enhance the water retention of the hydrogel, while sodium ions are used as free moving ions in the hydrogel network36,37,38.

Fig. 1: Fabrication and characterization of the all-printed OECT.
Fig. 1: Fabrication and characterization of the all-printed OECT.
Full size image

a Schematic of the ultrasonic vibrated dispenser in the direct printer; Magnified view of the schematic of the all-printed OECT. b Schematic diagram and optical microscope image of the all-printed OECT. c–e Optical microscope images of AgNP, PEDOT:PSS, and hydrogel, respectively; the insets show the magnified views. f Schematic of OECTs that mimic the neural computing and memory functions of the human brain. g, h Transfer curve and pulse response of ion-rich OECT (10-layer electrolyte). i, j Transfer curve and pulse response of ion-deficient OECT (1-layer electrolyte).

As the full-printing technology endows the design and fabrication of the OECT with flexibility and scalability, it enables the monolithic integration of two distinct OECTs (with 10- and 1-layer electrolytes) into a neuromorphic circuit via direct printing, to emulate the information processing and memory activities of the human brain (Fig. 1f). The design is based on that the human nervous system consists of diverse and highly specialized neurons, such as those in the cerebral cortex focused on information processing and those in the hippocampus essential for memory function39. Although both types of OECTs utilize the same electrolyte ink (\(c\) = 1.28 mmol/ml), they differ in structure and function. The device with a multilayer electrolyte, owing to its greater thickness, possesses a higher ion areal concentration and is thus designated the ion-rich OECT. It is employed as a computing neuron. In contrast, the single-layer electrolyte OECT, with a lower ion areal concentration, is termed the ion-deficient OECT and is designed to mimic a hippocampal neuron.

As shown in Fig. 1g, i and Supplementary Fig. S3a–d, the transfer characteristics reveal a critical dependence on electrolyte thickness. When the electrolyte is sufficiently thick (10 or 20 layers), an adequate ion reservoir is established. The IDS of the ion-rich OECT decreases sharply with VGS, with minimal hysteresis. In contrast, reducing the electrolyte thickness results in a rightward shift of the transfer curve, requiring a higher gate voltage to turn the transistor off and exhibiting significant hysteresis. Particularly, the ion-deficient OECT (with a single layer) cannot be completely turned off even at 2.5 V and exhibits a wide hysteresis of 0.38 V, due to the critically insufficient ion supply. Furthermore, when driven by the same pulse stimulus (Fig. 1h, j and Supplementary Fig. S3e–h), the OECTs with 10- or 20- electrolyte layers exhibit both a large ΔIPSC (ΔIPSC = IDS-IDS0) and fast recovery. Significantly, the ion-rich OECT is capable of responding to a pulse spike of 1 ms, demonstrating high stimulus-resolution capability for computational operations (Supplementary Fig. S4). This combination of high response amplitude and speed, arising from the efficient injection and removal of ions, is a hallmark of ion-rich OECTs that makes them ideal for processing units. Conversely, this performance sharply declines with thinner electrolytes. The 5-layer device recovers slowly (taking ~40 s), whereas the 1-layer OECT retains about 40% of its memory window for 300 s. This pronounced persistent current effect makes the ion-deficient device promising as a memory unit. Overall, these results demonstrate that regionally controlled ion doping, governed by electrolyte thickness, allows for the modulation of transistor operating modes. To highlight this advantage, a concise comparison between our method and existing OECT regulation strategies is presented in Supplementary Table S1.

Notably, the ΔIPSC continues to rise even after the gate pulse is removed. We attribute this behavior to the maintained VDS of –0.6 V, which preserves a residual electric field between the gate and the channel. This field drives further gradual migration of cations into the channel, resulting in a slow increase in ΔIPSC. In contrast, under a positive bias (0.6 V), the current recovers fully (Supplementary Fig. S5). A positive VDS reduces the effective gate–channel potential, leading to an elevated off-state current. Therefore, to maximize the on/off ratio and dynamic range in subsequent experiments, we maintained VDS at –0.6 V as the optimal bias condition. Additionally, the retention time achieved in our ion-deficient OECT still falls short when compared to state-of-the-art neuromorphic devices like memristors, which can exhibit retention times on the order of hours40. We suggest that the retention time can be further extended through strategies such as spatially confining the electrolyte to trap ions or tailoring the ionic liquid composition to inhibit dedoping17. Looking ahead, we foresee that the use of high-resolution Super Inkjet (SIJ) printing for electrolyte patterning can extend this regional doping strategy to the vertical OECT architectures.

Capacitance analysis

Figure 2a illustrates schematic representations of ion-rich (top) and ion-deficient (bottom) OECT architectures, highlighting the distinct electrolyte configurations enabled by direct-write printing technology. The thickness of the PEDOT:PSS films in the two OECTs was ~300 nm (Supplementary Fig. S6). Notably, this method achieves precise modulation of the electrolyte thickness, which directly governs the local ion doping level in the OECT channel. As shown in Fig. 2b, the ion-rich and ion-deficient OECTs are characterized by electrolyte thicknesses of 7.61 μm and 0.14 μm, respectively. The flatness measurements of the electrolyte layers are shown in Supplementary Fig. S7. Moreover, the FTIR spectra presented in Supplementary Fig. S8 reveal that the absorption peaks in the ion-rich OECT are notably more intense compared to those in the ion-deficient OECT, indicating an augmentation in film thickness38,41. Since the ion areal concentration \({c}^{* }\) can be calculated using the formula \({c}^{* }={dc}\), where \(d\) denotes the electrolyte thickness and \(c\) represents the volumetric ion concentration (1.28 mmol/mL), the ion-rich OECT exhibits a high \({c}^{* }\) of 10 mmol/m², whereas the ion-deficient OECT shows a relatively lower \({c}^{* }\) of 0.18 mmol/m².

Fig. 2: Comparison of ion-rich and ion-deficient OECTs.
Fig. 2: Comparison of ion-rich and ion-deficient OECTs.
Full size image

a Device schematic of ion-rich OECT (top) and ion-deficient OECT (bottom). b Step profiler measurements of the two types of OECTs. c Capacitance-gate voltage curves of ion-rich OECT. d Capacitance-gate voltage curves of ion-deficient OECT. e Experimental and calculated transfer curves of ion-rich OECT. f Experimental and calculated transfer curves of ion-deficient OECT.

Then, the capacitance of the OECT electrolytes is investigated by electrochemical impedance spectroscopy (EIS) measurements. According to the theory of double-layer effect, the capacitance of the electrolyte (C) is calculated as:

$$\frac{1}{C}=\frac{1}{{C}_{H}}+\frac{1}{{C}_{D}}$$
(1)

where \({C}_{H}\) is the Helmholtz layer capacitance and \({C}_{D}\) is the diffusion layer capacitance. It dictates that the electrolyte capacitance is constrained by the minimum value of the compact layer and diffuse layer capacitances. The compact layer, dominant at high concentrations, is roughly concentration-independent, whereas the diffuse layer governs at low concentrations, making capacitance highly dependent on it42,43,44,45. Therefore, the regionally controlled ion-doping strategy allows for capacitance modulation by controlling the electrolyte areal concentration, which is achieved by fabricating electrolyte with sub-micron thickness. As shown in Supplementary Fig. S9, the capacitance of the ion-rich OECT measured at 1 Hz (0.46 μF) exceeds that of the ion-deficient OECT (0.29 μF). Additionally, Fig. 2c, d is the capacitance-gate voltage curves of the ion-rich OECT and ion-deficient OECT, respectively. Both the maximum capacitance and the capacitance-voltage slope of the ion-rich OECT are much higher than that of the ion-deficient OECT. According to the formulas,

$${Q}_{i}=C^{\prime} {V}_{{GS}}$$
(2)
$${p}_{i}={Q}_{i}/e{t}_{E}+{n}_{i}$$
(3)
$${p}_{i}\propto \partial {C}^{{\prime} }/\partial {V}_{{GS}}$$
(4)

where \({Q}_{i}\), \({p}_{i}\), \(e\), and \({t}_{E}\) are total charges, ion density at the semiconductor/electrolyte interface, elementary charge, and the thickness of the electric double layer46. This indicates that, compared to the ion-deficient OECT, a higher gate voltage in the ion-rich OECT promotes greater ion migration from the electrolyte toward the channel interface, thereby enhancing the electrochemical doping of the PEDOT:PSS channel layer.

Figure 2e, f presents the experimental and calculated transfer curves of the ion-rich and ion-deficient OECTs, respectively. And the detailed modeling framework is provided in Supplementary Note 1. As the ion charges \({Q}_{i}\) induce a shift in the threshold voltage \({V}_{{th}}\) relative to its initial value \({V}_{{th}0}\), that is, \({V}_{{th}}={V}_{{th}0}-\frac{{Q}_{i}}{C}\)46. The ion-rich OECT demonstrates a significantly reduced threshold voltage (\({V}_{{th}}\,\)= −0.1 V) compared to the ion-deficient OECT (\({V}_{{th}}\) = 0.4 V), thereby enabling fast response and computational capability. Compared to computing units, non-volatile storage devices need a relatively wide hysteresis window to ensure reliable retention. The main factor leading to the hysteresis of I-V characteristics is the asymmetry of the dynamic response time scale between the drift-dominated (forward scan) and diffusion-dominated (reverse scan) ion transport regimes. Meanwhile, in ion-deficient OECTs, the total number of ions is inherently limited, leading to a formed concentration gradient (\(\nabla C\)) that is approximately two orders of magnitude lower than that in ion-rich OECTs. According to Fick’s law, the ionic diffusion rate (\({v}_{{ion}}\)) is directly proportional to the concentration gradient (\(\nabla C\)), that is, \({v}_{{ion}}\propto \nabla C\). The diffusion rate of ions in ion-deficient OECT is reduced, leading to a smoother reverse scan curve for the OECT. Eventually, the ion-deficient OECT shows a hysteresis window of ΔV = 0.38 V, which makes it a suitable candidate for storage device applications.

Ion-rich OECT for computing unit

The electrical characteristics of the all-printed ion-rich OECT are shown in Fig. 3a, b and Supplementary Fig. S10. The ion-rich OECT is characterized by a high on/off ratio (>103) and a transconductance as high as 36.38 ± 1.35 mS. The conductivity of the OECT decreases with increasing VGS, demonstrating a characteristic output curve of a depletion transistor. Then, the device was subjected to 1000 bending cycles at a radius of 4 mm. Supplementary Fig. S11 presents the transfer characteristics of the OECT after the bending cycles. Although a slight rightward shift in the transfer curves was observed, the device maintained a high on/off ratio of ~103 and a transconductance of ~27.7 mS after 1000 cycles, demonstrating that our devices possess remarkable flexibility, making them suitable for flexible electronics applications. Additionally, after applying 1000 pulses (Fig. 3c), the ion-rich OECT demonstrated a stable cycling test performance, indicating that the device can be employed in practical applications.

Fig. 3: Electrical performance and application of the ion-rich OECT.
Fig. 3: Electrical performance and application of the ion-rich OECT.
Full size image

a Transfer and transconductance curves of the ion-rich OECT. b Output curves of the ion-rich OECT. c Stability of the ion-rich OECT over 1000 pulse cycles (VGS = ± 1.5 V, tp = 150 ms, Δt = 150 ms). The current response of the ion-rich OECT under different d pulse voltage amplitude (tp = 10 ms), e pulse widths (VGS = 1.5 V), and f pulse frequencies (VGS = 1.5 V). g Schematic diagrams and testing results of NAND gate. h Schematic diagrams and testing results of NOR gate.

To emulate the process of information transmission between neurons, we characterize the synaptic plasticity behavior of the ion-rich OECT under various pulse conditions (Fig. 3d–f) with a constant drain bias of VDS = −0.6 V. An increase in pulse voltage amplitude, pulse width, or pulse frequency results in larger inhibition of IDS, due to a greater quantity of Na+ ions permeating and reacting with the PEDOT:PSS film. These data demonstrate the sensitive response of the ion-rich OECT to the pulse stimuli, providing good controllability for processing units. Building on previous reports of OECT-based logic circuits47,48,49,50,51, we now demonstrate the successful operation of our devices in these functional circuits. The gate voltages exerted on the OECTs are denoted as A and B. Specifically, VIN (comprising VIN-A and VIN-B) is set up in such a way that a voltage of 0 V corresponded to the logic state ‘0’, while a voltage of 1.5 V signified the logic state ‘1’. When both VIN-A and VIN-B are concurrently at a high voltage level (logic 1), the output is low; in all other cases, the output is high (logic 1). The outcome aligns with the logic of the NAND gate (Fig. 3g). Additionally, Fig. 3h verifies that the NOR gate invariably exhibited ideal logic functions across all possible VIN logic combinations, confirming the capability of the computing tasks.

Ion-deficient OECT for memory unit

Compared with the ion-rich OECT, the transfer curve of the ion-deficient OECT shifts to the right and attains a maximum gm of 17.35 ± 1.41 mS (Fig. 4a and Supplementary Fig. S12). The on/off ratio of the ion-deficient OECT is ~102, which is lower than that of ion-rich devices. We attribute this reduction to the insufficient amount of Na+ in the electrolyte, which restricts channel doping and leads to an elevated off-state current. To address this limitation, a primary strategy in our future research will be to optimize the device performance through precise control of the electrolyte-to-semiconductor thickness ratio. Then, Supplementary Fig. S13 shows the current response of the ion-deficient OECT under various pulse voltage amplitude and pulse widths. The current modulation in ion-deficient transistors under pulsed voltage is lower than in ion-rich devices (Supplementary Tables S2, S3), a consequence of their increased threshold voltage and slowed ion dynamics. This inherent limitation, however, corresponds to an extended duration of current change. Figure 4b plots the magnitude of the conductance change (|ΔG|) against the pulse width. |ΔG| is defined as |(I – I0)/VDS|, where I is the transient current during the pulse and I0 is the initial current before the pulse. This change in conductance grows progressively with longer pulse widths, underscoring the potent modulation capability achieved through temporal control of the gate stimulus7. Remarkably, this elevated conductance level is sustained even after the removal of the VGS (Fig. 4c), demonstrating a retention capability for pulses. Furthermore, Supplementary Fig. S14 provides the complete temporal evolution and confirms the long-term retention capability of the ion-deficient OECT, demonstrating a retention state that can be maintained for up to 1000 s.

Fig. 4: Electrical performance and application of the ion-deficient OECT.
Fig. 4: Electrical performance and application of the ion-deficient OECT.
Full size image

a Transfer and transconductance curves of the ion-deficient OECT. b Response characteristics and memory window of the ion-deficient OECT under different pulse widths (VGS = 1.5 V). c Retention test for five simulated states. d Paired-pulse facilitation (PPF) behavior. e ΔIPSC of ion-deficient OECT to different pulse numbers (VGS = 1.5 V, tp = 10 ms, Δt = 10 ms). f Schematic illustration of learning, forgetting, and consolidate processes in the human brain. g Simulation of Brain-like Learning-Forgetting in an ion-deficient OECT. h Magnified view of (g).

Then, the ion-deficient OECT is subjected to paired-pulse stimulation, as shown in Fig. 4d and Supplementary Fig. S15. As a vital feature of short-term plasticity, paired-pulse facilitation (PPF, defined as A₂/A₁) reflects the ability of synapses to respond to two consecutive stimuli, where the ΔIPSC amplitude (A₂) triggered by the second pulse is significantly larger than that (A₁) triggered by the first52,53. The relatively slow ion motion in the ion-deficient OECT underlies this phenomenon, resulting in a significant PPF index of 1.9 (tp = 10 ms). Subsequently, as shown in Fig. 4e, upon further increasing the number of pulses, the ΔIPSC shows a proportional growth and stabilizes at distinct levels, this behavior showcases enhanced synaptic responses and effective signal retention, aligning well with the cumulative impacts commonly witnessed in neuromorphic systems. These results strongly suggest that the ion-deficient OECT holds great promise as a candidate for memory and signal storage applications.

Moreover, Fig. 4f, g illustrates how the ion-deficient OECT mimics the learning behavior of the human brain and its ability to acquire new information. A magnified view is shown in Fig. 4h. When the pulse voltage is applied for the first time, simulating the human brain’s initial reception of new information and first learning experience, the ion-deficient OECT exhibits a short-term memory (STM) effect and the 1st learning process takes 0.5 s to achieve the ΔIPSC of 200 μA. Subsequently, once applying a negative pulse, the current recovers quickly, simulating the forgetting process of the human brain. However, applying multiple consecutive positive and negative voltage pulses simulates the learning process, during which the current is progressively enhanced. So, the 5th learning process takes only 0.3 s to achieve the ΔIPSC of 200 μA. Subsequently, Supplementary Fig. S16 demonstrates the memory consolidation capability of the ion-deficient OECT-simulated brain after several learning cycles. The retained memory current progressively strengthens with an increasing number of learning cycles, with the sustained change persisting even after learning has ceased. These experimental results demonstrate the competence of the ion-deficient OECTs in mimicking the information consolidation process of the human brain.

Neuromorphic system

A neuromorphic circuit combining the ion-rich OECT with ion-deficient OECT is designed to mimic the learning and memory process within the human brain. The schematic of the OECT circuit is provided in Supplementary Fig. S17, and its corresponding application for emulating human brain learning and memory functions is demonstrated in Fig. 5a. We simulate the learning process by programming external stimuli (such as the Morse code of the letter ‘A’) as gate voltage input (VIN). The ion-rich OECT mimics the brain’s learning and processing functions with the manifestation being the output voltage(VB), while the signal storage function is emulated by employing the variation in drain current (ΔIPSC) of the ion-deficient OECT. The drain current was selected as the output signal for three key reasons. First, this approach functionally emulates classical telegraphy, where information is encoded by interrupting a continuous current. In our system, the gate voltage modulates the channel conductivity to generate analogous current pulses. Furthermore, as a current-output device, the OECT provides a high on/off ratio and wide dynamic range under gate modulation, ensuring a high signal-to-noise ratio for the encoded signals. Finally, directly reading the current eliminates the need for load resistors, simplifying the system architecture and ensuring higher signal fidelity54,55,56.

Fig. 5: Neuromorphic circuit for simulating human brain study and memory functionalities.
Fig. 5: Neuromorphic circuit for simulating human brain study and memory functionalities.
Full size image

a Schematic diagram of on-chip integrated OECTs for simulating human brain study and memory functionalities. b Simulating information writing, storage, and erasing with the OECTs circuit. c Ion-rich OECT for 26-letter Morse code learning. d VIN, VB, and ΔIPSC response of the letters ‘H’, ‘E’, ‘L’, ‘L’, and ‘O’. e On-chip integrated OECTs for Morse-coded word ‘HELLO’ learning, memory, and discrimination. f ΔIPSC of letters over time after removing VIN.

As demonstrated in Fig. 5b and Supplementary Fig. S18, the proposed system successfully performs the functions of writing, storing, and erasing signals. The writing time of the devices is 0.40 s, and the erasing time is 1.16 s. It can be observed that upon applying a positive VGS, the ΔIPSC rapidly increases, signifying the writing of information. Upon removal of the gate voltage, the ΔIPSC does not immediately reset but decays gradually, retaining over 60% of its value for more than 60 s. This sustained retention is a direct manifestation of the memory function to the ion-deficient OECT in the post-synaptic stage. Finally, the application of a reverse VGS erases the information by restoring the drain current of ion-deficient OECT to its original value. Hence, the circuit based on all-printed OECTs can integrate the information processing and memory functionalities analogous to those of the human brain. Additionally, Fig. 5c demonstrates that the ion-rich OECT simulates the encoding, learning, and transmission of Morse code. A voltage pulse of 0.2 s mimics dots, while a longer pulse of 2 s represents dashes. The amplitude variation of VB successfully manifested the Morse code corresponding to 26 letters.

Subsequently, Fig. 5d demonstrates the encoding and memory capabilities of the neuromorphic circuit. Applying gate voltage pulses (VIN) corresponding to the Morse code of the letters ‘H, E, L, L, and O’, the amplitude of VB varies correspondingly, demonstrating a fast-learning response of ion-rich OECT to the Morse-coded letters. Then, owing to the memory retention characteristics of the ion-deficient OECT, the distinct Morse-encoded letter, ‘HELLO’, can be differentiated and remembered through the ΔIPSC of the ion-deficient OECT. Notably, removing the gate voltage, the Morse-encoded letters remain discernible through the neuromorphic circuit (Fig. 5e, f). It should be noted that the integrated circuit exhibits a shorter memory retention time than isolated devices and differential fading rates among characters. This discrepancy primarily may be attributed to non-ideal effects introduced during system-level integration: First, the shared conductive line within the integrated circuit creates leakage paths and a current division effect, which accelerates charge dissipation in the ion-deficient OECT memory unit. Second, voltage fluctuations on adjacent signal lines couple into the ion-deficient device via parasitic capacitance, leading to an accelerated decay of the stored state. Additionally, the voltage division in the circuit results in an insufficient gate voltage for programming, which limits ion injection and consequently shortens the retention time. To address these challenges, our future work will focus on the collaborative optimization of materials, fabrication processes, device architecture, and circuit design. Overall, analogous to compartmentalized ion dynamics in biological neurons, our devices spatially separate ionic reservoirs and depletion regions, reproducing distinct time constants for fast signal processing and long-term memory consolidation.

Discussion

We have successfully fabricated two types of OECT by direct printing technique and achieved integrated neuromorphic circuits, which incorporate computing and memory functionalities. Remarkably, this is accomplished using the same material, structure, and operating voltage. Through the application of direct printing technology, the printed electrolyte can be flexibly regulated, achieving the modulation of the areal ionic concentration of the OECT. The computing unit utilizes an ion-rich OECT, distinguished by its fast response, high transconductance, and large on/off ratio. Conversely, the memory unit employs an ion-deficient OECT, characterized by its stable and wide memory window. The combination of these two devices effectively mimics the learning and memory processes of the human brain, showcasing their potential in information encoding, transmission, and memory. The presented research serves as a valuable inspiration for the advancement of next-generation neuromorphic circuits, particularly those incorporating sense-computation-memory integrated systems. This work further establishes regionally controlled ion-doping as a structural tuning route for OECT functionality. Unlike previous approaches that rely on bias modulation or material composition, this geometry-based strategy provides a unified and scalable way to co-integrate computing and memory behaviors within the same device platform.

Methods

Materials

PEDOT:PSS (Clevios PH1000) was purchased from Heraeus Electronic Material. Dimethyl sulfoxide (DMSO, average Mw = 78.13), Glycerol (Gly, >99%), and 2-Hydroxy-4’-(2-Hydroxyethoxy)-2-Methylphenylacetone (photo initiator, average Mw = 224.25) were provided by Aladdin. While Acrylamide (AM), Ethylene Glycol (EG, >99%), (N,N)-Methylenebis-acrylamide (MBA), and sodium chloride (NaCl) were obtained from Sigma-Aldrich. Polyvinyl alcohol (PVA, 87-89% hydrolyzed, high molecular weight) was purchased from Alfa Aesar. The silver nanoparticle (AgNP) ink was purchased from Xianfeng Technology.

PEDOT:PSS ink preparation

10 wt% DMSO was added to PEDOT:PSS solution and stirred for 4 h. The addition of DMSO enhanced the film formation temperature, preventing rapid solvent evaporation at the needle tip during room-temperature printing and thus avoiding needle clogging. The thoroughly mixed solution was filtered using a 0.22 μm nylon syringe filter to achieve a uniform PEDOT:PSS ink.

Electrolyte ink preparation

0.05 g PVA, 0.25 g AM, 2 mg MBA, 0.22 g NaCl was dispersed in 2 ml H2O/EG/Gly (vol% 50/25/25), followed by stirring at 50 °C for 1 h. Then, cooling the mixture to room temperature, 4 mg of photoinitiator was added. Finally, the solution was placed in a vacuum chamber for 30 min to remove all the air bubbles.

Fabrication of OECTs

The fabrication of the all-printed OECTs used SonoPlot’s Nanomaterial Deposition Inkjet Printing System (Microplotter System). First, a flexible substrate (PEN) was subjected to plasma treatment (30 W, 10 s) to achieve a moderately hydrophilic surface. Then, the gate, source, and drain electrodes were printed on the substrate using the AgNP ink with the substrate heated to 50 °C. The printing speed was set to 600 μm s−1 along both the X and Y-axes. After printing, the electrodes were annealed at 150 °C for 30 min. Once the substrate returned to room temperature, PEDOT:PSS ink was printed in the channel region between the source and drain electrode, and the layer was then annealed at 90 °C for 30 min. Subsequently, the hydrogel electrolyte ink was printed onto the gate and channel regions. The number of printed layers was adjustable, with ten layers of hydrogel electrolyte forming the ion-rich OECT, and a single layer forming the ion-deficient OECT. Finally, the electrolyte layer was cured under UV light for one minute, completing the fabrication of the OECT.

OECTs for logic circuits

The NAND gate consists of a pull-up network and a pull-down load connected in series between the power supply (VD) and ground. The pull-up network is formed by two ion-rich OECTs connected in parallel acting as the drive transistors. The pull-down load is a resistor (103 Ω). The NOR gate also features a series-connected pull-up network and a pull-down load between VD and ground, but its pull-up network is formed by two ion-rich OECTs connected in series, serving as the driving transistors, with the pull-down load similarly being a 103 Ω resistor.

OECTs for the on-chip learning and memory circuit

As a front-end circuit, the ion-rich OECT was configured with its source terminal grounded, its drain connected to a voltage source VD1 through a pull-up resistor (500 Ω), and its gate receiving the input voltage VIN. The output node (VB) of this voltage divider network was fed to the gate of the subsequent ion-deficient OECT, thereby modulating its drain current. Finally, the ion-deficient OECT stored information by translating the Morse code into a specific, non-volatile drain current signal.

Characterization

The electrochemical properties of the electrolytes were characterized via an electrochemical workstation (CHI660E, Shanghai Chenhua). For the capacitance-frequency test, the source and drain were connected to the reference and counter electrodes, while the gate was connected to the working electrode. An AC voltage of 1 V was applied, with the scanning frequency swept from 105 Hz down to 1 Hz. The transfers, output, and pulsed characteristics of the OECTs were measured using a semiconductor device analyzer (PDA, FS-Pro, PRIMARIUS.)