Recent progress in proton involvement and coupling for bio-realistic synaptic devices

Abstract

Neuromorphic systems aim to emulate the energy efficiency and adaptive learning of the human brain. Recent studies have suggested that ions such as Na⁺, K⁺, and H⁺ play key roles in biological decision-making, emphasizing ion-specific signaling in neural processes. Among emerging hardware approaches, proton-based synaptic devices have gained attention for their ability to replicate ion-mediated signaling in synapses. Owing to their high mobility arising from their low atomic mass, protons enable fast, low-power, analog switching, which is crucial for mimicking biological efficiency. This review classifies the switching mechanisms of protonic neuromorphic devices into two types: proton involvement and proton coupling. By analyzing two- and three-terminal architectures, we present a framework showing how each mechanism modulates resistance in different materials. Proton involvement typically involves field- or environment-driven ionic motion, whereas proton coupling refers to mechanisms in which protons interact with other ions to regulate redox activity and transport. In this context, we emphasize the importance of studying ion-mediated processes, particularly those involving protons, as windows into biological adaptability and intelligence. Ultimately, this review highlights that understanding proton-based switching mechanisms is crucial for realizing neuromorphic hardware that emulates the energy efficiency and adaptability of the brain.

---

1. Introduction

The rapid expansion of artificial intelligence (AI) research has brought neuromorphic computing into the spotlight.¹ This biologically inspired approach seeks to emulate the structural and functional principles of the human brain by implementing hardware architectures that replicate neural circuits to enable efficient, brain-like information processing.² Neuromorphic computing applications are gaining traction across diverse fields, including autonomous vehicles, robotics, bio-signal processing, and space computing.³ The appeal of neuromorphic computing lies in its ability to process information in a parallel, event-driven manner, which enables low power consumption and real-time responsiveness—features that are highly desirable for next-generation computing.⁴ At the heart of this technology is the development of artificial synaptic devices that integrate memory and computation within a single unit.⁵ To enhance the performance of these devices, a wide range of materials—including metal oxides,^5–11 chalcogenides,^12–16 and organic compounds^17–20 —have been explored.²¹ Additionally, various device types, such as resistive random-access memory (RRAM),^22,23 phase-change random-access memory (PRAM),²⁴ and others,^25,26 have been proposed for integration into neuromorphic systems. To emulate complex biological signaling, these devices are often coupled with control strategies that regulate the pulse width, frequency, and amplitude to mimic diverse forms of synaptic plasticity.^27,28 Despite the promising potential of various engineering strategies, achieving the energy efficiency of biological synapses remains challenging.²⁹ While some artificial devices can approach biological performance at the device level, scaling up to full neuromorphic systems introduces parasitic losses, leakage, and reliability issues that prevent realization of comparable efficiency.^30,31

Biological synapses feature an approximately 20 nm-wide synaptic cleft through which neurotransmitters are released from the presynaptic neuron and bind to receptors on the postsynaptic neuron, where ion and proton dynamics coordinate to ensure efficient signal transmission, as visualized in Fig. 1.³² This architecture ensures directional and selective signal transmission, while consuming only 10–100 fJ per synaptic event.³³ Moreover, synapses exhibit a range of time-dependent behaviors such as short-term plasticity (STP), long-term potentiation (LTP), and spike-timing-dependent plasticity (STDP), which regulate the synaptic strength in response to neural activity.^34–36 While these mechanisms have been widely emulated in artificial devices, they represent only fragmented aspects of biological learning.³⁷ To emulate the remarkable energy efficiency of the brain, it is essential to replicate its adaptive learning capabilities, such as metaplasticity, which allow synapses to adjust their plasticity based on past activity and changing stimuli.^29,38 Metaplasticity refers to the brain's ability to adjust the sensitivity of a synapse to plasticity based on its history of activity, i.e., it modulates how readily a synapse responds to future stimuli, thus enabling context-aware learning and adaptation.³⁹

Fig. 1 (a) Schematic of a biological synapse showing neurotransmitter release and potential local proton (H⁺) accumulation in the ∼20 nm cleft following Ca²⁺ influx. These signals bind to postsynaptic receptors, opening ion channels. (b) Enlarged view illustrating how a proton gradient can induce an ion gradient by modulating receptor activation: greater H⁺ concentrations lead to greater ion influx. (c) Conceptual comparison between single-ion and multi-ion synaptic signaling, showing how H⁺ contributes to sensitive stimulus-response coupling that supports adaptive learning behavior.

Adaptive learning in biological synapses is fundamentally mediated by ion-based signal transduction, as ions modulate synaptic behavior through their dynamic and stimulus-responsive migration across membranes.^40,41 The migration of ions such as Na⁺, Ca²⁺, and K⁺ across neuronal membranes following electrochemical gradients is crucial for efficient information storage and processing.^42,43 These ionic movements drive essential synaptic activities, including excitatory postsynaptic potentials (EPSPs) through Na⁺ influx,⁴⁴ neurotransmitter release through Ca²⁺ influx,⁴⁵ and membrane potential restoration via K⁺ efflux.⁴⁰ In biological synapses, protons (H⁺) play a significant signaling role, particularly through their influence on pH-sensitive receptors such as acid-sensing ion channels (ASICs)^46,47 and N-methyl-D-aspartate (NMDA) receptors.⁴⁸ Subtle shifts in pH can alter the activation thresholds of these receptors, highlighting how local ionic conditions affect synaptic responsiveness.^39,49 These receptors are activated by extracellular acidification, allowing protons to modulate synaptic excitability and contribute to sensory perception and neural plasticity. More recently, protons have gained attention because of their distinctive advantages, such as an inherently high mobility owing to their low atomic mass, which enables fast, low-power, and analog switching behaviors that are crucial for mimicking biological efficiency.^7,10 These properties are key determinants of the faster response times, lower power consumption, and enhanced dynamic modulation capabilities observed in proton-based devices, making them particularly suitable for replicating biological learning processes. Consequently, mimicking the modulatory role of protons at biological synapses could be the key to developing artificial synapses with unprecedented efficiency, scalability, and adaptability. Therefore, proton-based devices are a promising foundation for next-generation neuromorphic engineering.

Given these advantages, recent research has increasingly focused on the development of proton-based memristors and transistors, in which gate-controlled proton migration enables the replication of synaptic plasticity and dynamic signal modulation.^10,11 Protonic artificial synapses thus offer a promising platform for implementing low-power, high-speed, and high-density neuromorphic hardware. By accurately mimicking ion-based signaling in biological synapses, these systems provide an energy-efficient foundation for next-generation neuromorphic device design.⁵⁰ This review aims to analyze the switching mechanisms of proton-based artificial synaptic devices by categorizing them into two distinct frameworks: proton involvement and proton coupling, as illustrated in Fig. 2. First, we examine the physical principles and switching characteristics of each mechanism in the context of two-terminal devices. Representative examples of proton involvement include conductivity modulation in organic–inorganic hybrid perovskite⁵¹ and oxide-based⁵² materials, whereas proton coupling is discussed in terms of continuous analog switching driven by proton-coupled electron transfer (PCET) reactions in bio-polymers,^53,54 with a focus on their potential to replicate biological synaptic functionalities. We then extend the discussion to three-terminal devices, focusing on bio-polymer-based systems that utilize a third terminal to inject protons for enhanced linearity and dynamic modulation,⁵⁵ as well as Complementary Metal–Oxide–Semiconductor (CMOS)-compatible devices employing inorganic electrolytes that exhibit rapid and reproducible switching behavior through proton involvement.^56,57 By examining the switching mechanisms of proton-based artificial synapses, this review aims to elucidate their potential to emulate the adaptive learning and energy efficiency of the brain.

Fig. 2 Conceptual illustration connecting biological and artificial synapses through proton-based mechanisms. Inspired by ion-driven signaling in the brain, artificial devices employ proton involvement and proton–electron coupling to control proton migration via voltage and humidity. These mechanisms allow for precise conductance adjustment and more linear weight updates, which are key features for mimicking adaptive learning and realizing energy-efficient neuromorphic computing.

2. Exploring proton-driven Switching in two-terminal architectures

2.1. Proton involvement: moisture and local electric field-induced ionic control

The proton involvement mechanism refers to a mode of switching behavior in which protons that are introduced into a device by external stimuli, such as an electric field or humidity, modulate the local conductivity and thereby influence resistive switching (RS) characteristics. These protons may originate from environmental sources such as moisture or hydrogen, or they may be externally injected. Within a device, they induce changes in the local electronic structure or bonding network, often acting as transient dopants or structural modifiers. These modulations can alter conduction pathways and trigger changes in the resistance state. Protons play a central role in this mechanism by modulating the electronic and structural properties through humidity-induced lattice distortion or field-driven ionic motion. This has been demonstrated in various material systems,⁵⁸ including perovskite-based compounds,^59,60 oxide-based materials,^52,61–63 and polymer-based platforms.⁶⁴

Zhang et al.⁶⁰ demonstrated that the RS behavior of CH₃NH₃PbI₃ perovskite memristors can be indirectly modulated by protonic effects in moisture-sensitive environments. As illustrated in Fig. 3a, the two-terminal device consists of a CH₃NH₃PbI₃ active layer positioned between a gold (Au) top electrode and a fluorine-doped tin oxide (FTO) bottom electrode. The Au/CH₃NH₃PbI₃/FTO memristor exhibits pronounced variations in RS behavior as a function of the relative humidity (RH), highlighting the critical role of moisture in the device performance. As shown in Fig. 3b, the current in the low-resistance state (LRS) increases gradually with increasing RH; this trend is similar to that observed in the endurance behavior shown in Fig. 3c. This enhancement is interpreted as the result of moisture reducing the migration barrier for iodide ions (I⁻), which facilitates the formation of broader and more stable conductive channels. In terms of endurance, the switching behavior remains stable within the RH range of 5%–75%, whereas at 90% RH, the device begins to degrade, with a reduction in the high-resistance state (HRS) occurring within just 20 switching cycles. To further elucidate the mechanism, high-resolution X-ray diffraction (XRD) was performed on CH₃NH₃PbI₃ films exposed to different humidity levels (Fig. 3d). At 75% RH, a shift in the (220)/(004) diffraction peak, typically centered around 28.3°, toward lower angles indicated lattice expansion, which is consistent with water molecule intercalation. This structural distortion facilitates ion transport by weakening the Pb–I bonds and lowering the activation energy for vacancy migration. Notably, these changes were reversible upon returning to a low RH, suggesting that humidity can act as a dynamic and non-destructive modulator of the switching behavior. This reversible structural tuning under moderate humidity appears to be the main contributor to the RS enhancement, whereas irreversible degradation and reduced endurance occurred primarily under excessively high RH conditions (e.g., 90% RH, Fig. 3c). The dominant switching mechanism is iodide ion migration facilitated by moisture-induced lattice softening. In addition, under high electric fields, water molecules undergo electrolysis to generate protons (H⁺) and hydroxide ions (OH⁻), which also contribute to conductivity modulation by supplying additional mobile carriers.

Fig. 3 Humidity-dependent resistive switching behavior of CH₃NH₃PbI₃-based memristors. (a) Schematic diagram of the CH₃NH₃PbI₃-based memristor composed of an Au/CH₃NH₃PbI₃/FTO structure. (b) I–V curves measured at different relative humidity (RH) levels, showing a progressive reduction in the low-resistance state (LRS) and narrowing of the ON/OFF window with increasing RH. (c) Endurance characteristics under RH ≤ 75% and RH = 90%, highlighting the degradation of switching behavior in high-humidity conditions. (d) High-resolution X-ray diffraction (XRD) profiles of CH₃NH₃PbI₃ thin films exposed to RH levels of 5%, 75%, and returned to 5%, showing reversible peak shifts indicative of lattice expansion. Reproduced with permission from ref. 60. Copyright 2021, American Chemical Society.

Another study demonstrated that the RS behavior in Au/Nb : SrTiO₃ interface-type memristors could be effectively modulated by controlling the RH.⁶² Kunwar et al. showed that protons from moisture are absorbed at the metal–semiconductor interface and modulate the Schottky barrier via proton-assisted electron trapping and detrapping. Under dry conditions, the HRS and I–V hysteresis disappear, while re-exposure to humid environments restores both, highlighting the critical role of protons in enabling switching. This forming-free device architecture, based solely on a Schottky junction without an oxide switching layer, exhibited reliable performance with an on/off ratio over 10⁴ and endurance beyond 10⁴ cycles.

Zhang et al.⁵² demonstrated that localized electric fields can directly induce proton-based switching in a WO₃ memristor using a platinum-coated atomic force microscopy (AFM) tip. Upon positive bias, protons generated from H₂ dissociation diffuse into the WO₃ lattice, leading to lattice expansion and electron doping, which together trigger a hydrogenation-induced insulator-to-metal transition (IMT), hereafter termed proton-driven IMT. Unlike conventional abrupt IMTs, this proton-driven IMT in WO₃ is structurally stable and enables gradual multistate resistance modulation with endurance over 500 cycles (Fig. 4b and c). Secondary ion mass spectrometry (SIMS) analysis (Fig. 4d) confirmed surface-localized proton accumulation, indicating a field-driven insertion mechanism. These protons modulate the conductivity without invoking bulk redox reactions. Fig. 4e shows that protons cause vertical lattice expansion (∼2.3%) without disrupting the in-plane structure, thus supporting reversible, structurally confined switching behavior. Electron energy loss spectroscopy (EELS) results (Fig. 4f) reveal increased electron occupancy in the W 5d orbitals, confirming electron doping via proton intercalation. These results confirm that protons contribute both structurally and electronically by modulating the local carrier density. These results validate that applying a localized electric field via a Pt-coated AFM tip inserts protons into WO₃, thus inducing electron doping and vertical lattice expansion. This results in a reversible proton-driven IMT and stable multi-level RS, representing a novel switching mechanism driven by proton-induced electronic and structural modulation rather than by conventional oxygen vacancy dynamics.

Fig. 4 Protonic switching behavior in WO₃ memristors driven by localized electric fields. (a) Schematic illustration of a WO₃-based memristor probed by a Pt-coated conductive AFM tip to induce localized electric fields. (b) I–V characteristics showing a stable proton-driven insulator-to-metal transition (IMT) over 200 cycles. (c) Multi-level resistance states (HRS, IRS, LRS) obtained by varying the voltage amplitude (±2 V, ±4 V); the LRS conductance increases with voltage. (d) SIMS profile indicating increased H⁺ concentration near the hydrogenated surface. (e) c-Axis lattice expansion verified by XRD, with in-plane epitaxy maintained. (f) EELS spectra showing suppression of the O K-edge peak A, confirming electron injection into W 5d orbitals. Reproduced from ref. 52 © 2023, F. Zhang et al., licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/).

As a cost-effective alternative to expensive Pt-based local proton insertion, Deng et al.⁶¹ employed spatially selective proton doping using Pd electrodes in VO₂-based Mott devices to integrate volatile neuron and non-volatile synapse functionalities within a single structure. Protonation of VO₂ induces lattice expansion and stabilizes a metallic phase (H_xVO₂), enabling nonvolatile conductance switching. This behavior is attributed to a proton-driven electronic insulator–metal transition (E-IMT), distinct from conventional thermally or field-induced IMT. In the absence of H⁺, the device exhibits only volatile threshold switching, whereas H⁺ insertion induces stable electronic phase modulation. The system demonstrated strong neuromorphic performance, achieving 92.3% and 73.2% classification accuracies on the Modified National Institute of Standards and Technology (MNIST) dataset (a widely used handwritten digit recognition benchmark) and Fashion-MNIST dataset, respectively. These results, obtained through spiking neural network (SNN) simulations, reflect the importance of achieving linear and symmetric synaptic weight updates, which are key factors for improving the classification accuracy and learning efficiency in neuromorphic systems.

In addition to oxide-based examples, Josberger et al.⁶⁴ reported a two-terminal protonic device that utilized PdH_x electrodes in a polymeric environment. In this system, protons (H⁺) functioned as both charge carriers and state variables, exemplifying proton involvement in solid-state systems. The device employed palladium hydride (PdH_x) electrodes and a proton-conducting Nafion polymer membrane as the channel. Under an applied bias, protons are injected from the PdH_x source into the Nafion layer, generating a transient H⁺ current (ISD) that decays as hydrogen is depleted, thus effectively mimicking synaptic short-term depression. This proton-mediated switching is reversible, with performance metrics including a switching speed of ∼25 ms, energy consumption of ∼30 nJ per event, and clear I–V hysteresis, underscoring its potential for low-power, neuromorphic memory applications.

Overall, proton involvement provides a versatile route for dynamically modulating resistive states through humidity or local electric fields. Nevertheless, most reported systems rely on external proton sources or localized fields, which limits quantitative control and device reproducibility. Future efforts should aim to achieve stable and scalable proton transport— for example, by incorporating solid-state or hybrid interfaces—while maintaining the unique advantage of moisture-assisted tunability.

2.2. Proton coupling: redox-reactive biomaterials for H⁺-mediated switching

Proton coupling refers to a RS mechanism closely related to electrochemical metallization (ECM), in which protons facilitate the redox reaction of metal cations (e.g., Ag⁺) and the formation of conductive filaments. In the ECM mechanism, Ag⁺ ions migrate through the switching medium under an electric field and are reduced to metallic Ag⁰, thus forming conductive filaments that create an LRS. These filaments can be ruptured or dissolved to return to a high-resistance (OFF) state, thus enabling memory switching. Song et al.⁵³ proposed a distinct RS mechanism in peptide-based memristors, in which this process is further assisted by protons. Rather than merely interacting with other carriers, protons actively participate in the redox reaction and lower the energy barrier for Ag⁺ reduction and migration. This coupling occurs via a process known as PCET, in which the transfer of protons is chemically linked to the movement of electrons. In tyrosine-rich peptides, protons dissociate from phenol groups under an electric field and assist in reducing Ag⁺ to Ag⁰, thereby supporting filament formation. Consequently, the PCET enables redox-driven ECM switching with improved tunability, analog control, and energy efficiency. Importantly, the proton conduction itself does not simply enhance mobility, but rather modulates the electrochemical environment in a manner that facilitates ionic conductivity and redox kinetics for Ag⁺.

Building on this concept, Song et al.⁵³ designed a peptide-based memristor and proposed a new PCET-driven switching mechanism using a tyrosine-rich peptide sequence (YYACAYY, or Y7C). Fig. 5a shows a schematic illustration of the device configuration consisting of a Pt bottom electrode, Y7C peptide film, and Ag top electrode. When a positive voltage is applied, Ag is oxidized to Ag⁺, which diffuses into the peptide layer. The phenol groups in tyrosine donate electrons to reduce Ag⁺ to Ag⁰ and simultaneously release protons that promote filament formation. This sequence exhibited significantly lower set voltages than a phenylalanine-based peptide (FFACAFF), owing to the proton-donating ability of the phenol side chains in tyrosine.⁶⁵ To validate the proton-conducting nature of the Y7C film, electrochemical impedance spectroscopy (EIS) was performed using symmetric Au electrodes. Fig. 5b shows that as the RH increases, the impedance of the peptide film decreases significantly, indicating enhanced proton conductivity. Fig. 5c and e demonstrate the RH-dependent switching behavior: as RH increases from 15% to 90%, the set voltage decreases significantly from 4.6 V to 0.4 V, indicating that increased humidity enhances proton conduction and facilitates Ag⁺ reduction. To further clarify the role of protons, H₂O was replaced with D₂O vapor, and the set voltage at 90% RH increased to 2.1 V. This unusually large increase—known as a kinetic isotope effect (KIE) due to the slower motion of heavier deuterium—confirms that proton transfer plays a central role in the switching process and is chemically coupled to the redox reaction, consistent with a PCET mechanism.

Fig. 5 Summary of proton-coupled switching in Y7C peptide-based memristors. (a) Schematic of the device structure and PCET mechanism for Ag filament growth. (b) EIS measurements of proton conductivity under different RH levels. (c and d) I–V curves under increasing RH in H₂O vapor (c) and D₂O vapor (d); higher RH shown in darker curves, all measured after 2 h of equilibration. (e) Box plots of RH-dependent set voltages for both H₂O and D₂O, showing the interquartile range, mean, median, and outliers. (f) Alternating voltage and RH stimuli demonstrate dual-mode resistive switching. Reproduced from ref. 53 © 2020, M. K. Song et al., licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/).

At low RH, limited proton availability impedes the uniform reduction of Ag⁺ ions and the stable formation of conductive filaments. Consequently, the devices exhibited larger variability and lower ON/OFF ratios, reflecting incomplete or unstable switching behavior. In contrast, under high RH conditions, enhanced proton mobility supports more uniform Ag⁺ reduction and filament growth, resulting in stable, symmetric, and analog switching characteristics.⁵⁴Fig. 5f highlights the dual-mode operation of the device: the device was written by either voltage pulses or increasing RH, and erased by voltage pulses, demonstrating multimodal switching behavior. This responsiveness reflects a degree of functional mimicry of biological synapses, where multiple types of stimuli modulate signal transmission, providing neuromorphic systems with high flexibility and functionality. In a subsequent study, Song et al.⁵⁴ expanded the system from memory switching to analog switching suitable for neuromorphic computing. By leveraging improved proton conduction under high humidity, more linear and repeatable switching behavior was achieved through solid-state PCET. The enhanced device exhibited an image classification accuracy of 82.5% in a Fashion-MNIST simulation at 90% RH, demonstrating its viability in practical neuromorphic applications.

In summary, proton coupling provides a promising mechanism for fine-tuned, energy-efficient analog switching through tightly linked proton–electron dynamics. This mechanism supports key requirements in analog computing, such as gradual and symmetric conductance modulation, low-power tunability, and multilevel memory states, thereby offering concrete design pathways for next-generation neuromorphic hardware. The main challenge is to translate this bio-inspired proton–electron coupling into stable and scalable device platforms. This will require materials that can sustain rapid and repeatable proton exchange without structural degradation, and device architectures that confine redox reactions in a well-controlled nanoscale region to minimize variability. System-level strategies that integrate soft protonic layers with robust inorganic matrices may further bridge biological functionality with process compatibility. Moreover, extending PCET-based concepts toward cooperative interactions among multiple ions could open new directions for dynamic and adaptive synaptic behavior.

3. Exploring proton-driven switching in three-terminal architectures

3.1. Three-terminal proton coupling for analog switching

This section explores an advanced approach to proton-coupled switching that leverages a three-terminal configuration. By decoupling the pathways for conduction and ion injection, this architecture allows for more precise conductance modulation and improved synaptic functionality. The integration of a separate control terminal for proton modulation leads to enhanced linearity, switching fidelity, and learning performance. Yoon et al.⁵⁵ proposed a proton-coupled peptide-based memristive system incorporating a pseudo-three-terminal layout, in which a Pd electrode was integrated laterally to facilitate proton injection into the switching medium.

The core device structure comprises a Pt bottom electrode, self-assembled tyrosine-rich peptide (Y7C) layer, and Ag top electrode that is responsible for switching. Fig. 6a shows a schematic illustration of this three-terminal architecture. A Pd electrode, positioned laterally, serves as the control terminal, which modulates the local proton concentration within the peptide matrix without directly contributing to conduction. The Pd terminal plays a crucial role by absorbing H₂ gas and dissociating it into protons, subsequently injecting protons into the Y7C film through the formation of PdH_x. These injected protons migrate toward the switching interface, where they participate in PCET reactions, thus facilitating the reduction of Ag⁺ ions and the nucleation of metallic filaments. This configuration separates the pathways of proton modulation and RS, allowing for more precise and independent control of the synaptic behavior. Comparative transient current measurements (Fig. 6b(i–iii)) verify the effectiveness of Pd as a proton injector under vacuum, hydrogen-rich, and humid air conditions, with Pd–Pd devices showing significantly higher charge transfer than their Au–Au counterparts (Fig. 6c). This contrast arises from intrinsic material properties: Pd efficiently injects protons via PdH_x formation, whereas Au blocks proton transport. Fig. 6b and c highlight this difference, confirming that Pd enables a much greater current response under identical conditions. These experiments provide direct evidence that the Pd terminal actively injects protons into the switching layer, thereby validating the proposed proton coupling mechanism.

Fig. 6 Summary of proton-coupled three-terminal switching in peptide-based memristive devices. (a) Schematic of the Pd-integrated three-terminal structure and PCET-based switching mechanism. (b) Transient current responses of Pd–Pd and Au–Au configurations under (i) vacuum, (ii) dry H₂, and (iii) humid air. (c) Integrated charge comparison across environmental conditions, validating the proton injection efficiency. (d and e) I–V curves under various V_Pd (d: Pd terminal; e: Au terminal). (f) Average set voltage versus applied terminal voltage, showing linear tuning in Pd-based devices. (g) Synaptic weight modulation under pulse stimuli with varying V_Pd. (h) Neural network architecture used for MNIST classification simulation. (i) MNIST test accuracy for each V_Pd condition after 100 training epochs. Reproduced from ref. 55 © 2025, J. H. Yoon et al., licensed under CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/).

To assess the impact of proton injection on the switching behavior, the voltage applied to the Pd control terminal (V_Pd) was systematically varied. At V_Pd = +10 V, the set voltage (V_set) dropped to ∼0.5 V, whereas at V_Pd = −10 V, V_set rose to ∼4.5 V, indicating that greater proton availability facilitated easier filament formation (Fig. 6d). In contrast, devices using a non-reactive Au terminal showed negligible changes in V_set across all voltages (Fig. 6e), confirming the unique role of Pd in proton injection. These results were further validated by plotting the average V_set against the control voltage (Fig. 6f). In Pd-based devices, the graph reveals a nearly linear decreasing trend, where increasing V_Pd leads to a systematic reduction in V_set. This linear relationship confirms that proton availability, modulated via the Pd terminal, directly affects the energy barrier for filament formation and thus governs the switching threshold. The neuromorphic capabilities of this system were demonstrated by evaluating the synaptic weight modulation under different V_Pd values. Linear and gradual conductance changes were observed at +10 V, whereas lower or negative voltages resulted in abrupt and unstable transitions (Fig. 6g). These behaviors were incorporated into a software-based simulation of a four-layer artificial neural network trained on the MNIST dataset (Fig. 6h), where recognition accuracy reached 91.1% at +10 V but dropped to 43.3% at −10 V (Fig. 6i). These results underscore the importance of proton regulation via a dedicated terminal for achieving reliable and efficient neuromorphic learning.

Together, these results underscore the advantages of integrating proton-coupled switching into three-terminal devices, where an independent control terminal can regulate the ionic concentration and switching characteristics. This decoupled architecture enables precise analog modulation and robust learning capabilities, making peptide-based systems promising candidates for bio-inspired neuromorphic computing. However, despite the advantage of independent proton control, the current pseudo-three-terminal configuration—relying on a lateral Pd electrode and hydrogen exposure—remains complex and unsuitable for large-scale or CMOS-compatible integration. Simplifying this architecture into a vertically stacked or fully solid-state format would greatly improve scalability and enable compact device layouts. At the same time, reliability must be further enhanced by stabilizing the Pd–electrolyte interface and mitigating PdH_x-related degradation during repeated hydrogen absorption and release cycles. Replacing precious Pd with more cost-effective proton-storage materials could also reduce fabrication costs while maintaining effective proton delivery. Future efforts should therefore focus on developing structurally simplified, reliable, and economically viable three-terminal systems based on solid-state or hybrid protonic platforms that can combine precise ionic control with robust long-term performance.

3.2. Proton involvement in three-terminal electrochemical RAM

Various three-terminal ion-involved synaptic transistors (electrochemical RAM, ECRAM)⁶⁶ have been developed using a range of mobile ions, such as Li⁺,^67–70 Na⁺,^71,72 and O²⁻,^73–75 to achieve controllable and analog conductance switching. Van de Burgt et al.⁷⁶ first proposed a proton-based ECRAM using a PEDOT:PSS(poly(3,4-ethylenedioxythiophene):polystyrene sulfonate)/Nafion system, demonstrating its potential as a fast and accurate synaptic device. Building on this concept, Onen et al.⁵⁶ introduced a CMOS-compatible three-terminal structure that integrated a PdH_x gate, phosphosilicate glass (PSG) solid electrolyte, and WO₃ channel. This configuration enabled non-volatile, symmetric conductance modulation, high energy efficiency, and compatibility with semiconductor manufacturing, making it a strong candidate for next-generation analog computing devices.

Fig. 7a depicts the vertically stacked device architecture composed of a PdH_x gate, PSG solid electrolyte, and WO₃ channel. This layout allows ion-based conductance modulation via electrochemical proton insertion. Fig. 7b shows the scanning electron microscopy (SEM) image of a 5 μm × 25 μm channel, confirming the scalability and precise patterning of the device. Fig. 7c and d highlight the key features of the proposed CMOS-compatible protonic programmable resistor. Fig. 7c quantitatively demonstrates the ability of the device to modulate conductance in a linear and symmetric fashion. In this experiment, ±3 V gate voltage pulses, each lasting 1 s, were applied 100 times in both directions while maintaining V_DS = 0.1 V. The measured source current showed a gradual increase with positive pulses and a corresponding decrease with negative pulses, confirming symmetric and reversible conductance modulation. Notably, the conductance remained stable after pulse removal, indicating non-volatility due to the electron-blocking nature of the PSG layer. A distinct symmetry point, where incremental and decremental conductance changes became equal, was also observed, highlighting the ideal weight update characteristics for neuromorphic learning. Fig. 7d shows the dynamic conductance response under repeated gate pulses obtained by monitoring the source current in real time. When a pulse is applied, the current exhibits an immediate shift, followed by a gradual change while the pulse is active; a second abrupt shift then occurs once the pulse is removed, stabilizing at a new conductance level. These two-stage transitions originate from distinct physical mechanisms: a volatile field-effect modulation (ΔG_FE) caused by proton-induced interfacial charge accumulation during the pulse, and a non-volatile modulation (ΔG_{intercalation}) resulting from actual proton insertion or extraction from the WO₃ lattice. This coexistence of volatile and non-volatile behaviors demonstrates the potential of the device to support both short- and long-term memory functions in neuromorphic hardware. Additionally, the measured gate current remains in the sub-picoampere range throughout, indicating highly energy-efficient operation. Together, Fig. 7c and d confirm that the PdH_x/PSG/WO₃ device combines fast response, non-volatility, low power operation, and symmetric weight tuning, all of which are essential characteristics for energy-efficient analog learning in neuromorphic systems.

Fig. 7 Structure and proton-driven switching behavior of the three-terminal PdH_x/PSG/WO₃ device (a). (b) SEM image of the top view showing the source, drain, and gate terminals with a 5 μm × 25 μm channel. (c) Reversible and symmetric conductance modulation under ±3 V pulses, with stable non-volatile states. (d) Real-time I_S response showing transient field-effect and non-volatile proton intercalation. Reproduced with permission from ref. 56 Copyright 2021, American Chemical Society.

In a subsequent study, the authors demonstrated nanosecond-scale conductance switching by further refining the device structure and operating conditions.⁵⁷ The device dimensions were drastically reduced to 50 × 150 nm, and advanced fabrication techniques such as electron-beam lithography and self-aligned patterning were introduced to improve the alignment precision and integration density. The WO₃ channel was crystallized to enhance its proton sensitivity, and the PSG electrolyte was thinned to less than 10 nm to enable high-field operation under ±10 V gate pulses, resulting in electric fields exceeding 1 V nm⁻¹. Under these extreme conditions, proton transport within the PSG was significantly accelerated, allowing conductance modulation with gate pulses as short as 5 ns. The conductance change remained stable for hundreds of seconds after the pulse, confirming the excellent non-volatility. Remarkably, the energy consumed per switching event was as low as ∼15 aJ, making the system several hundred times more energy efficient than biological synapses. Therefore, this follow-up study established a significant milestone in neuromorphic engineering, proving that proton-based devices can simultaneously achieve ultrafast speed, low energy consumption, long-term retention, and high scalability, which are key requirements for practical analog deep learning applications.

Overall, these studies establish proton-based ECRAM as a major milestone in neuromorphic hardware, demonstrating that solid-state protonic devices can operate at nanosecond speeds—comparable to conventional electronic memories—while maintaining symmetric, non-volatile, and ultralow-energy switching. Nevertheless, such remarkable performance does not yet overcome the intrinsic limitation of proton mobility or the material-level constraints associated with ionic conductivity and interfacial stability. Achieving consistently high-speed and durable operation will require further advances in material design to engineer faster proton transport pathways and chemically robust solid electrolytes. Continued progress in scalable fabrication and the exploration of cost-effective proton-injection materials will also be crucial for translating these high-performance devices into large-scale, low-power analog computing systems.

4. Conclusions

In this review, we categorized the switching mechanisms of protonic synaptic devices into two conceptual modes: proton involvement and proton coupling. Based on this classification, we systematically analyzed relevant materials, device architectures, and switching principles. To provide a clearer overview, we have additionally summarized representative device structures, switching mechanisms, and performance metrics in a comparison table (Table 1). In two-terminal devices, proton involvement refers to resistance switching mediated by ion migration induced by environmental stimuli such as humidity or electric fields, as demonstrated in CH₃NH₃PbI₃ and WO₃-based devices. In contrast, proton coupling involves proton-coupled electron transfer (PCET), in which the release of protons triggers redox reactions and filament formation, as observed in tyrosine-based peptide devices. In three-terminal systems, proton coupling enables enhanced linearity and learning accuracy through the independent control of proton injection using lateral Pd electrodes. Notably, a higher proton concentration in the switching layer improves conductance linearity, underscoring the role of proton content in device optimization. While this review separates proton involvement and proton coupling for clarity, they are not mutually exclusive. In peptide-based memristors with PdH_x electrodes, field-driven proton migration and proton coupling effects can occur together, showing that both mechanisms can synergistically coexist within a single device.⁷⁷ For proton involvement, electrochemical random-access memory (ECRAM) devices utilizing protons have demonstrated non-volatile analog switching with highly linear and symmetric conductance modulation across multiple states in response to incremental programming pulses; they also exhibit CMOS compatibility. This comparative framework clarifies how proton presence and transport affect the RS behavior depending on the material context, offering design insights for future neuromorphic systems. Notably, proton-based mechanisms feature high ionic mobility, redox tunability, and structural adaptability, making them well-suited for replicating the key features of biological synapses, such as activity-dependent plasticity, analog signal modulation, and long-term memory retention.

Table 1 Overview of representative proton-based synaptic devices

Terminal type	Mechanism	Materials	Device performance metrics				Analog computing metrics			Advantages	Ref.
			On/off ratio	Programming voltage	Retention	Endurance	Linearity	Neural network accuracy	Synaptic functionality
a SNDP – Spike-Number Dependent Plasticity. b SFDP – Spike-Frequency Dependent Plasticity.
2-Terminal	Proton involvement: moisture-induced lattice expansion	Au/CH3NH3PbI3/FTO	10^3(5% RH)–10^4(75% RH)	V set: 0.5 V, Vreset: −0.6 V (28% RH)	>10^4 s (at RH ≤ 75%)	>10^4 cycles (at RH ≤ 75%)	—	—	—	Humidity-sensitive RS, low-voltage switching	60
	Proton involvement: moisture-driven protons modulate Schottky barrier	Au/Nb : SrTiO3/Au	>10^4 (at 0.6 V)	Set Vth: 1.2V	>10 000 s	>10^4 cycles	—	—	—	Humidity-dependent RS, forming-free	62
				Reset V < −5V
	Proton involvement: proton-driven IMT	Pt/WO3/Nb : SrTiO3	—	±4 V (writing/erasing voltage)	—	>500 cycles	Symmetric potentiation/depression	—	—	Multilevel conductance (analog tuning), nanoscale patterning	52
	Proton involvement: proton-driven E-IMT	Pd/HxVO2/VO2/TiAu	—	±10 V μm^−1 (potentiation/depression)	>10^4 s	>10^3 cycles	Nonlinear factor: ∼0.8	MNIST: 92.1%, fashion-MNIST: 73.22%	—	Multilevel conductance (analog tuning), SNN simulation, Co-integration with CMOS	61
	Proton Involvement: Proton transport through Nafion	PdHx/Nafion/PdHx	∼100	V set: +1.25 V, Vreset: −1.25 V	—	22 cycles	—	—	STD	Low-power (30 nJ per bit), synaptic-like behavior	64
		Ag/Y7C/Pt	>10^6 (at 45% RH)	V set: 0.4 V(at 90% RH)/2.6 V (at 45% RH)/4.6 V (at 15% RH)	>10^4 s	>100 cycles	—	—	PPF, STP, LTP, SNDP^a, SFDP^b	Humidity-controlled switching, multistate, low-voltage switching, RH-only write	53
	Proton coupling: proton-coupled electron transfer (PCET)	Ag/Y7C/Pt	—	V set: 1.15 V (F7C) vs. 0.5 V (Y7C) (at 80% RH)	>10^4 s (Y7C, at 80% RH)	>100 cycles (Y7C, 80% RH)	—	—	—	Low-voltage switching, RH-only write	65
		Ag/Y7C peptide/Pt	∼10^6 (60% RH)	±0.3 V (potentiation/depression) (at 90% RH)	>10^4 s	—	Asymmetric nonlinearity factor: 0.597 (at 90% RH)	Fashion-MNIST: 82.50% (at 90% RH), 10.00% (at 75% RH)	—	Analog switching, low-voltage switching, humidity-tunable, metaplasticity	54
			∼10^4 (90% RH)
3-Terminal	Proton coupling: external Pd gate injects protons	Ag/Y7C/Pt, PdHx (side contact)	—	V set : 0.56 V (at VPd = +10 V)/3.84 V (at VPd = −10 V)	—	—	Improvement with Vpd	MNIST: 91.1% (at VPd = +10 V)	LTP, LTD	Analog tuning, low-voltage operation, biorealistic operation	55
	Proton involvement: proton transfer between PEDOT:PSS and Nafion	PEDOT:PSS/Nafion/PEDOT:PSS + PEI(S/D)	—	Low switching voltage (∼0.5 mV)	>25 h	>300 cycles	High linearity	MNIST: 93–97%	STP, LTP, PPF, STDP	Analog tuning(>500 states), ultra-low voltage switching, flexibility, low energy (10 pJ)	76
		Pd (gate)/PSG/WO3 (channel)/Au (S/D)	—	V Pd: ±3 V (potentiation/depression), (at VDS = 0.1 V)	—	> 5 × 10^4 pulses	Symmetric & gradual	—	—	CMOS-compatible, low-energy (1 fJ per step), multistate, Symmetric modulation, Scalable	56
	Proton involvement: gated proton (H+) intercalation	Pd (gate)/PSG/WO3 (channel)/Au (S/D)	—	V Pd: 10 V/−8.5 V (potentiation/depression) (at VDS = 0.1 V)	≥100 s	>10^5	Nearly linear	—	—	Ultrafast (5 ns), low-energy (∼15 aJ), multistate, CMOS-compatible	57

---

Proton-based synaptic devices offer a fundamentally different route toward adaptive learning and energy-efficient computation, inspired by biological systems. Rather than relying solely on fast electron transport, these devices utilize proton-mediated mechanisms that inherently support gradual and analog conductance changes that are crucial for mimicking synaptic plasticity. Although the intrinsic transport speed of protons can, in theory, approach that of electrons, practical limitations currently constrain their performance.^10,78–80 Nevertheless, this inherent slowness can also be viewed as an opportunity. The gradual proton dynamics may be advantageous for slow electronics^63,81 and edge computing⁸² that require stable, low-power, and time-dependent processes, thereby opening alternative pathways for neuromorphic applications.⁸³ Recently, however, notable progress has been achieved in accelerating protonic responses. For instance, proton–conducting PSG-based devices⁵⁷ have demonstrated switching speeds on the order of ∼5 ns, representing a substantial improvement over earlier protonic systems and suggesting the feasibility of practical computing applications. Despite this advancement, further improvement in proton mobility and interfacial control remains crucial to bridge the gap between conventional high-speed electronic devices and emerging ionic-based computing paradigms. Thus, ongoing research should focus on overcoming these material bottlenecks to fully leverage proton dynamics for low-power, brain-inspired systems. By embracing ionic, rather than purely electronic, signal transduction, protonic systems can unlock new possibilities for implementing energy-efficient, adaptive, and analog computation systems that more closely parallel biological intelligence.

One of the critical limitations in solid-state protonic systems is their relatively low proton mobility compared to that in aqueous biological environments. This limited mobility slows the device response and restricts the achievable energy efficiency.^57,84 To overcome this fundamental bottleneck, a multi-pronged roadmap focusing on materials, device architecture, and system integration is essential. From a materials perspective, it is essential to enhance intrinsic proton conductivity through the design of highly conductive oxide frameworks and the incorporation of proton storage media such as Pd or Pt, which play a crucial role in optimizing device performance.^7,85,86 The oxide framework is prioritized due to its inherent CMOS compatibility,⁵⁶ superior thermal/chemical stability,⁷⁹ and demonstrated capability for stable multi-level conductance modulation, which often surpass those of organic/polymeric or perovskite alternatives.⁵⁷ Achieving these material advancements should lead to devices with key figures of merit suitable for synaptic applications, including target switching energy below 10 aJ (approaching the biological limit), switching speed in the nanosecond range.⁵⁷

Furthermore, materials must be engineered to offer faster proton conduction pathways, e.g., through nanostructured oxide lattices⁸⁷ or interfacial hydrogen-bond networks,⁸⁸ and the rational design of inorganic solid-state electrolytes that support selective proton transport while minimizing electronic leakage.⁸⁰ From a device perspective, both memristor-like and transistor-type architectures should be optimized to enable efficient proton involvement and/or coupling either internally or at interfaces.^80,89,90 This includes refining fabrication techniques to drastically reduce device dimensions and optimize device architectures to reduce transport distances and retain protons effectively.⁵⁶ At the system level, solid-state protonic platforms should be capable of rapid signal processing and data retention with low power consumption, while remaining compatible with standard semiconductor processes.⁹¹ To achieve CMOS circuit compatibility, innovation is required to address the thermal limitations of peptide-based biopolymer organic materials. Moreover, while noble-metal hydrogen reservoirs (Pd, Pt) provide excellent stability and hydrogen uptake kinetics, low-cost alternatives (Al, Mg) still have room for improvement in stability and uniformity,⁹² leaving the cost-performance balance as a major challenge for large-scale implementation. These persistent issues highlight the importance of continuous material innovation toward reliable and scalable protonic neuromorphic systems.

The ultimate goal of biomimetic intelligent synaptic devices is to emulate the energy efficiency and computational performance of the human brain. More broadly, from an academic standpoint, understanding the mechanisms underlying high-level human cognition and communication will require the development of multimodal neuromorphic systems capable of replicating the complexity of biological ion signaling. Such systems must go beyond proton-only architectures to incorporate and interpret the roles of other biologically essential ions, including Na⁺, K⁺, and Ca²⁺. Exploring how these ions interact and modulate synaptic behavior may offer foundational insights into the nature of biological intelligence.

Author contributions

Yubeen Park: writing – original draft, investigation, visualization. Jung-El Ryu: data curation, visualization. Seok Daniel Namgung: conceptualization, supervision, writing – review & editing, project administration. Min-Kyu Song: conceptualization, supervision, writing – review & editing, project administration.

Conflicts of interest

There are no conflicts to declare.

Data availability

This review does not include any primary research data, software, or code, nor were any new data generated or analyzed in the course of this work.

Acknowledgements

This work was supported by Korea Institute for Advancement of Technology (KIAT) grant funded by the Korea Government (MOTIE) (P0023718, Inorganic Light-emitting Display Expert Training Program for Display Technology Transition). This research was supported by the 2025 Chung-Ang University Research Scholarship Grant. This work is supported by a Korea University Grant.

References

1. D. Ivanov, A. Chezhegov, M. Kiselev, A. Grunin and D. Larionov, Front. Neurosci., 2022, 16, 959626.
2. G. Indiveri, B. Linares-Barranco, T. J. Hamilton, A. V. Schaik, R. Etienne-Cummings, T. Delbruck, S.-C. Liu, P. Dudek, P. Häfliger and S. Renaud, Front. Neurosci., 2011, 5, 73.
3. D. Kudithipudi, C. Schuman, C. M. Vineyard, T. Pandit, C. Merkel, R. Kubendran, J. B. Aimone, G. Orchard, C. Mayr and R. Benosman, Nature, 2025, 637, 801–812.
4. M. Yao, O. Richter, G. Zhao, N. Qiao, Y. Xing, D. Wang, T. Hu, W. Fang, T. Demirci and M. De Marchi, Nat. Commun., 2024, 15, 4464.
5. Z. Wang, S. Joshi, S. E. Savel'ev, H. Jiang, R. Midya, P. Lin, M. Hu, N. Ge, J. P. Strachan and Z. Li, Nat. Mater., 2017, 16, 101–108.
6. Q. Tian, X. Shan, J. Bian, Y. Cheng, J. Zheng, Z. Wang, X. Zhao, H. Xu and Y. Liu, J. Mater. Chem. C, 2025, 13, 12046–12065.
7. X. Yao, K. Klyukin, W. Lu, M. Onen, S. Ryu, D. Kim, N. Emond, I. Waluyo, A. Hunt and J. A. Del Alamo, Nat. Commun., 2020, 11, 3134.
8. M. Lanza, H. S. P. Wong, E. Pop, D. Ielmini, D. Strukov, B. C. Regan, L. Larcher, M. A. Villena, J. J. Yang and L. Goux, Adv. Electron. Mater., 2019, 5, 1800143.
9. S. P. Pati and T. Yajima, Jpn. J. Appl. Phys., 2024, 63, 030801.
10. Y. Yuan, R. K. Patel, S. Banik, T. B. Reta, R. S. Bisht, D. D. Fong, S. K. Sankaranarayanan and S. Ramanathan, Chem. Rev., 2024, 124, 9733–9784.
11. L. Liu, P. A. Dananjaya, E. K. Koh, F. Tan, Z. Chen, G. J. Lim, C. X. X. Lee, J. L. Yang and W. S. Lew, Small Methods, 2025, 2500445.
12. S. G. Sarwat, T. Moraitis, C. D. Wright and H. Bhaskaran, Nat. Commun., 2022, 13, 2247.
13. Y. Li, Y. Zhong, L. Xu, J. Zhang, X. Xu, H. Sun and X. Miao, Sci. Rep., 2013, 3, 1619.
14. S. R. Bauers, M. B. Tellekamp, D. M. Roberts, B. Hammett, S. Lany, A. J. Ferguson, A. Zakutayev and S. U. Nanayakkara, Nanotechnology, 2021, 32, 372001.
15. Z. Yang, Y. Li and X. Miao, Photo-Electroactive Nonvolatile Memories for Data Storage and Neuromorphic Computing, Elsevier, 2020, pp. 293–315.
16. L. Q. Zhu, C. J. Wan, L. Q. Guo, Y. Shi and Q. Wan, Nat. Commun., 2014, 5, 3158.
17. T.-G. Hwang, H. Park and W.-J. Cho, Biomimetics, 2024, 9, 157.
18. M. M. H. Tanim, Z. Templin and F. Zhao, Micromachines, 2023, 14, 235.
19. S. Liu, J. Zeng, Z. Wu, H. Hu, A. Xu, X. Huang, W. Chen, Q. Chen, Z. Yu and Y. Zhao, Nat. Commun., 2023, 14, 7655.
20. J. Zhang, D. Liu, Q. Shi, B. Yang, P. Guo, L. Fang, S. Dai, L. Xiong and J. Huang, npj Flexible Electron., 2022, 6, 30.
21. A. Pal, Z. Chai, J. Jiang, W. Cao, M. Davies, V. De and K. Banerjee, Nat. Commun., 2024, 15, 3392.
22. J. Park, A. Kumar, Y. Zhou, S. Oh, J.-H. Kim, Y. Shi, S. Jain, G. Hota, E. Qiu and A. L. Nagle, Nat. Commun., 2024, 15, 3492.
23. J. H. Yoon, Y.-W. Song, W. Ham, J.-M. Park and J.-Y. Kwon, APL Mater., 2023, 11, 090701.
24. S. Nandakumar, M. Le Gallo, I. Boybat, B. Rajendran, A. Sebastian and E. Eleftheriou, J. Appl. Phys., 2018, 124, 152135.
25. F. Bashir, F. Zahoor, A. Alzahrani and H. Abbas, Sci. Rep., 2025, 15, 12638.
26. R. Oshio, T. Kuwahara, T. Aoki, M. Kimura and Y. Nakashima, Sci. Rep., 2025, 15, 3954.
27. L. Hammarström and S. Styring, Energy Environ. Sci., 2011, 4, 2379–2388.
28. Y. Sun, H. Wang and D. Xie, Nano-Micro Lett., 2024, 16, 211.
29. D. Kireev, S. Liu, H. Jin, T. P. Xiao, C. H. Bennett, D. Akinwande and J. A. C. Incorvia, Nat. Commun., 2022, 13, 4386.
30. C. Liu, L. Zhu, J. K. Weng, Y. C. Li, Z. Chen, J. Lei, T. Q. Zi, D. Wu, W. Fa and S. Chen, Adv. Electron. Mater., 2023, 9, 2200821.
31. W. Xu, S.-Y. Min, H. Hwang and T.-W. Lee, Sci. Adv., 2016, 2, e1501326.
32. B. A. Vogt, L. Vogt, N. B. Farber and G. Bush, J. Comp. Neurol., 2005, 485, 218–239.
33. Y. Liang, H. Li, H. Tang, C. Zhang, D. Men and D. Mayer, Nano-Micro Lett., 2025, 17, 198.
34. N. Caporale and Y. Dan, Annu. Rev. Neurosci., 2008, 31, 25–46.
35. G. Neves, S. F. Cooke and T. V. Bliss, Nat. Rev. Neurosci., 2008, 9, 65–75.
36. R. S. Zucker and W. G. Regehr, Annu. Rev. Physiol., 2002, 64, 355–405.
37. F. Pulvermüller, R. Tomasello, M. R. Henningsen-Schomers and T. Wennekers, Nat. Rev. Neurosci., 2021, 22, 488–502.
38. S. Farashahi, C. H. Donahue, P. Khorsand, H. Seo, D. Lee and A. Soltani, Neuron, 2017, 94, 401–414.
39. W. C. Abraham, Nat. Rev. Neurosci., 2008, 9, 387–387.
40. G. Voglis and N. Tavernarakis, EMBO Rep., 2006, 7, 1104–1110.
41. Y. Hu, F. Zeng, C. Chang, W. Dong, X. Li, F. Pan and G. Li, ACS Omega, 2017, 2, 746–754.
42. J. Kim and D. A. Hoffman, Neuroscientist, 2008, 14, 276–286.
43. S. Huang, W. Dong, X. Lin and J. Bian, Neural Regener. Res., 2024, 19, 2684–2697.
44. H. Takagi, Neurosci. Res., 2000, 37, 167–171.
45. T. C. Südhof, Cold Spring Harbor Perspect. Biol., 2012, 4, a011353.
46. Y. Huang, N. Jiang, J. Li, Y.-H. Ji, Z.-G. Xiong and X.-M. Zha, Neuropharmacology, 2015, 94, 42–48.
47. W.-Z. Zeng, D.-S. Liu, L. Liu, L. She, L.-J. Wu and T.-L. Xu, Sci. Rep., 2015, 5, 14125.
48. X.-M. Zha, J. A. Wemmie, S. H. Green and M. J. Welsh, Proc. Natl. Acad. Sci. U. S. A., 2006, 103, 16556–16561.
49. D. Mango and R. Nisticò, Front. Pharmacol., 2019, 10, 555.
50. C. Weilenmann, A. N. Ziogas, T. Zellweger, K. Portner, M. Mladenović, M. Kaniselvan, T. Moraitis, M. Luisier and A. Emboras, Nat. Commun., 2024, 15, 6898.
51. Y. Li, Y. Pan, C. Zhang, Z. Shi, C. Ma, S. Ling, M. Teng, Q. Zhang, Y. Jiang and R. Zhao, ACS Appl. Mater. Interfaces, 2022, 14, 44676–44684.
52. F. Zhang, Y. Zhang, L. Li, X. Mou, H. Peng, S. Shen, M. Wang, K. Xiao, S.-H. Ji and D. Yi, Nat. Commun., 2023, 14, 3950.
53. M.-K. Song, S. D. Namgung, D. Choi, H. Kim, H. Seo, M. Ju, Y. H. Lee, T. Sung, Y.-S. Lee and K. T. Nam, Nat. Commun., 2020, 11, 5896.
54. M.-K. Song, S. D. Namgung, H. Lee, J. H. Yoon, Y.-W. Song, K. H. Cho, Y.-S. Lee, J.-S. Lee, K. T. Nam and J.-Y. Kwon, Nano Res., 2023, 16, 858–864.
55. J. H. Yoon, M. K. Song, W. Ham, K. J. Park, J. M. Park, S. D. Namgung and J. Y. Kwon, Adv. Funct. Mater., 2025, 35, 2415222.
56. M. Onen, N. Emond, J. Li, B. Yildiz and J. A. Del Alamo, Nano Lett., 2021, 21, 6111–6116.
57. M. Onen, N. Emond, B. Wang, D. Zhang, F. M. Ross, J. Li, B. Yildiz and J. A. Del Alamo, Science, 2022, 377, 539–543.
58. Y. Wang, T. Seki, P. Gkoupidenis, Y. Chen, Y. Nagata and M. Bonn, Proc. Natl. Acad. Sci. U. S. A., 2024, 121, e2314347121.
59. T. Dey, X. Lai, S. Manna, K. Patel, R. K. Patel, R. S. Bisht, Y. Zhou, S. Shah, E. Y. Andrei and S. K. Sankaranarayanan, ACS Nano, 2025, 19, 6815–6825.
60. X. Zhang, X. Zhao, X. Shan, Q. Tian, Z. Wang, Y. Lin, H. Xu and Y. Liu, ACS Appl. Mater. Interfaces, 2021, 13, 28555–28563.
61. S. Deng, H. Yu, T. J. Park, A. N. Islam, S. Manna, A. Pofelski, Q. Wang, Y. Zhu, S. K. Sankaranarayanan and A. Sengupta, Sci. Adv., 2023, 9, eade4838.
62. S. Kunwar, C. B. Somodi, R. A. Lalk, B. X. Rutherford, Z. Corey, P. Roy, D. Zhang, M. Hellenbrand, M. Xiao and J. L. MacManus-Driscoll, Adv. Electron. Mater., 2023, 9, 2200816.
63. S. P. Pati, Y. Geng, S. Hamasuna, K. Fujiwara, T. Iizuka, H. Inoue, I. Inoue and T. Yajima, Commun. Mater., 2024, 5, 177.
64. E. E. Josberger, Y. Deng, W. Sun, R. Kautz and M. Rolandi, Adv. Mater., 2014, 26, 4986–4990.
65. M.-K. Song, S. D. Namgung, J. M. Suh, J. H. Yoon, K. T. Nam and J.-Y. Kwon, ACS Mater. Lett., 2023, 6, 275–280.
66. A. A. Talin, Y. Li, D. A. Robinson, E. J. Fuller and S. Kumar, Adv. Mater., 2023, 35, 2204771.
67. R. D. Nikam, M. Kwak, J. Lee, K. G. Rajput, W. Banerjee and H. Hwang, Sci. Rep., 2019, 9, 18883.
68. S. Thakoor, A. Moopenn, T. Daud and A. Thakoor, J. Appl. Phys., 1990, 67, 3132–3135.
69. J. Tang, D. Bishop, S. Kim, M. Copel, T. Gokmen, T. Todorov, S. Shin, K.-T. Lee, P. Solomon and K. Chan, Proc. IEEE Int. Electron Devices Meeting (IEDM), 2018, 13.1.1–13.1.4.
70. J. Lee, R. D. Nikam, S. Lim, M. Kwak and H. Hwang, Nanotechnology, 2020, 31, 235203.
71. K. Lee, J. Lee, R. D. Nikam, S. Heo and H. Hwang, Nanotechnology, 2020, 31, 455204.
72. J. Lin, S. Lan, L. Yang, Q. Lin, Y. Chen and W. Xu, J. Mater. Chem. C, 2025, 13, 13246.
73. P. Langner, F. Chiabrera, N. Alayo, P. Nizet, L. Morrone, C. Bozal-Ginesta, A. Morata and A. Tarancón, Adv. Mater., 2025, 37, 2415743.
74. Y. Li, E. J. Fuller, J. D. Sugar, S. Yoo, D. S. Ashby, C. H. Bennett, R. D. Horton, M. S. Bartsch, M. J. Marinella and W. D. Lu, Adv. Mater., 2020, 32, 2003984.
75. S. Kim, T. Todorov, M. Onen, T. Gokmen, D. Bishop, P. Solomon, K.-T. Lee, M. Copel, D. B. Farmer, J. A. Ott, T. Ando, H. Miyazoe, V. Narayanan and J. Rozen, Proc. IEEE Int. Electron Devices Meeting (IEDM) 2019, 35.7.1–35.7.4.
76. Y. Van De Burgt, E. Lubberman, E. J. Fuller, S. T. Keene, G. C. Faria, S. Agarwal, M. J. Marinella, A. Alec Talin and A. Salleo, Nat. Mater., 2017, 16, 414–418.
77. J. H. Yoon, W. Ham, K. J. Park, S. D. Namgung, M. K. Song and J. Y. Kwon, Small Methods, 2025, e01472.
78. A. A. Talin and B. Yildiz, Chem. Rev., 2025, 125, 4765–4767.
79. X. Deng, Y.-X. Liu, Z.-Z. Yang, Y.-F. Zhao, Y.-T. Xu, M.-Y. Fu, Y. Shen, K. Qu, Z. Guan and W.-Y. Tong, Sci. Adv., 2024, 10, eadk9928.
80. R. Li, L. Cheng, W. Liu, C. Jin, X. Shi, P. Xie, Q. Sun, M. Long, J. Yang and J. C. Ho, Appl. Phys. Rev., 2025, 12, 021318.
81. H. Inoue, H. Tamura, A. Kitoh, X. Chen, Z. Byambadorj, T. Yajima, Y. Hotta, T. Iizuka, G. Tanaka and I. H. Inoue, Adv. Mater., 2025, 37, 2407326.
82. O. Krestinskaya, A. P. James and L. O. Chua, IEEE Trans. Neural Netw. Learn. Syst., 2019, 31, 4–23.
83. M. Mukhtar, M. J. Sultan, U. Ali and S. Y. Lee, J. Mater. Chem. C, 2025, 13, 17458–17482.
84. N. Ma, S. Kosasang, A. Yoshida and S. Horike, Chem. Sci., 2021, 12, 5818–5824.
85. S. Fop, J. A. Dawson, D. N. Tawse, M. G. Skellern, J. M. Skakle and A. C. Mclaughlin, Chem. Mater., 2022, 34, 8190–8197.
86. M. Tayyab, S. Rauf, A. Z. Khan, Z. Tayyab, K. Khan, I. Hussain, M. B. Hussain, M. Waseem, A. N. Alodhayb and X.-Z. Fu, Ceram. Int., 2024, 50, 40526–40552.
87. Z. Chen, Y. Peng, F. Liu, Z. Le, J. Zhu, G. Shen, D. Zhang, M. Wen, S. Xiao and C.-P. Liu, Nano Lett., 2015, 15, 6802–6808.
88. R. Guo, Y. Zhou, W. Wang, Y. Zhai, X. Liu, W. He, W. Ou, R. Ding, H.-L. Zhang and M. Wu, Sci. Adv., 2025, 11, eadr5374.
89. W. Sun, B. Gao, M. Chi, Q. Xia, J. J. Yang, H. Qian and H. Wu, Nat. Commun., 2019, 10, 3453.
90. K. Gou, Y. Li, H. Song, R. Lu and J. Jiang, iScience, 2024, 27, 111327.
91. H. Kwak, N. Kim, S. Jeon, S. Kim and J. Woo, Nano Converg., 2024, 11, 9.
92. L. Schlapbach and A. Züttel, Nature, 2001, 414, 353–358.

---