Neuromorphic devices for electronic skin applications

Abstract

Neuromorphic devices represent an important advancement in technology, drawing inspiration from the intricate and efficient mechanisms of the human brain. This review paper elucidates the diverse landscape of neuromorphic electronic skin (e-skin) technologies while highlighting their numerous applications. Here, neuromorphic devices for e-skin are classified as two types of direct neuromorphic e-skins combining both neuromorphic devices and sensors, and indirect e-skins separating neuromorphic devices and sensors. In direct neuromorphic e-skins, there are developing devices like memristor-based neuromorphic devices with sensors and transistor-based neuromorphic devices with sensors. On the other hand, indirect types are demonstrated as separated neuromorphic and sensor parts systems through the various interfacing structures. It also describes recent neuromorphic developments in artificial neural networks (ANNs), deep neural networks (DNNs), and convolutional neural networks (CNNs), for the real-time interpretation of sensory data. Moreover, it introduces multimodal sensory feedback, soft and flexible e-skins, and more intuitive human–machine interfaces. This review examines various applications, including smart textiles for the development of next-generation wearable bioelectronics, brain-sensing interfaces that enhance tactile perception, and the integration of human-machine interfaces aimed at replicating the biological sensorimotor loop, which can improve health monitoring and biomedical applications. Additionally, the review also highlights the potential of neuromorphic e-skin in human–robot interaction, particularly in the context of continuous prosthetic control and robotics. Through this analysis, the paper provides insights into current advancements, identifies key challenges, and suggests future research directions for optimizing neuromorphic e-skin devices and expanding their practical implementation.

---

Wider impact

Recently, many researchers have published brain-mimicked neuromorphic devices as conceptual memory computing interfacing five sensors, hearing, smell, taste, touch, and vision. Here, neuromorphic electronic skin (e-skin) represents a significant step forward in intelligent systems, merging bio-inspired neuromorphic computing with flexible, sensor-rich interfaces. This innovation enhances human–machine interactions by allowing machines to closely replicate the human sense of touch, achieving remarkable responsiveness, adaptability, and efficiency. This review paper highlights both direct and indirect neuromorphic e-skin systems, emphasizing their capacity to process sensory data in real-time using various approaches, including memristor, transistor, CMOS-based, and neural network-driven techniques. The potential applications of neuromorphic e-skin are vast and impactful. In smart textiles, emerging technologies promise to develop intelligent, adaptive devices that seamlessly integrate with complex environments. Meanwhile, brain-sensing interfaces enhance tactile perception, offering new possibilities for sensory augmentation. Inspired by biological systems, human–machine interfaces aim to replicate the sensorimotor loop, offering transformative opportunities in health monitoring and biomedical applications. Additionally, neuromorphic e-skin plays a crucial role in human–robot interaction, particularly in continuous prosthetic rehabilitation and advanced robotics, enabling more natural and responsive control. The creation of adaptable systems links theory to practice, enhancing human capabilities and quality of life through bio-inspired technologies.

1. Introduction

Neuromorphic engineering represents a paradigm shift in technology, drawing inspiration from the complex and efficient workings of the human brain. By emulating the structure and function of biological neural systems, neuromorphic devices aim to create intelligent systems capable of processing information in ways that mirror human cognition and sensory perception.^1–4 Central to this field is the development of neuromorphic electronic skin (e-skin) with flexible, sensor-rich interfaces designed to replicate the sense of touch, enabling machines to interact with their environment in more sophisticated and human-like ways.^5–8 These technologies not only enhance robotic capabilities but also pave the way for groundbreaking applications in rehabilitation, health monitoring, and environmental sensing.^9–12 The foundation of neuromorphic devices is rooted in the desire to replicate the efficiency and adaptability of biological systems. Unlike traditional computing architectures, neuromorphic systems leverage parallel processing and event-driven computation like a human, which aligns more closely with the way biological neurons operate.^13,14 Because of this paradigm change, neuromorphic systems for artificial neural networks (ANN), deep neural networks (DNN), and convolutional neural networks (CNN) that interpret sensory data in real-time have been developed, allowing machines to react to changing environments at previously unheard-of speed and accuracy.^15–17

Skin is a vital interface between the body and its environment, providing sensory feedback and flexibility that enables complex tasks.¹⁸ For individuals with skin damage or amputations, disrupted perception–action loops hinder even basic actions like object grasping. While prosthetic limbs restore some motor function, they lack sensory feedback and suffer from issues like phantom limb pain and poor dexterity.^19–21 To address these challenges, researchers have focused on creating soft, flexible e-skin that integrates multimodal sensory feedback and neuromorphic signal processing to enable more natural human–machine interfaces.^22–25 In rehabilitation, neuromorphic engineering is revolutionizing the design of brain–machine interfaces (BMIs) and prosthetic devices. BMIs utilize neural signals to control artificial limbs, offering means for individuals with motor impairments to regain mobility.^26,27 Recent advancements have shown that integrating neuromorphic principles can enhance the functionality of these systems, enabling them to adapt to the user's intentions more effectively.^28–30

One of the most promising applications of neuromorphic e-skin is in the realm of robotics, particularly in enhancing human–robot interaction. For instance, by incorporating tactile sensors that mimic the human sense of touch, robots can achieve a nuanced understanding of their surroundings. Recent studies have demonstrated that robots equipped with neuromorphic e-skin can perform delicate tasks, such as recognition of unknown objects and grip adjustment, by interpreting tactile feedback in real-time.^31–33 This capability is crucial for developing robots that can work alongside humans in collaborative environments, where precision and adaptability are paramount. Furthermore, wearable bioelectronics, another critical area of neuromorphic applications, are being developed to continuously monitor physiological signals, providing valuable insights into patients’ health status. By employing neuromorphic algorithms, these devices can process data from the various sensors to facilitate timely medical interventions.^5,7 As the field continues to advance, the integration of neuromorphic devices into environmental monitoring systems holds significant promise. These systems can interpret complex environmental cues, enabling the development of smart devices that interact seamlessly with their surroundings.^34,35 The incorporation of sensory feedback mechanisms in consumer electronics is also on the rise, intending to create more interactive and user-friendly products that enhance everyday experiences.

Neuromorphic e-skin represents a significant advancement in the field of artificial sensory systems, offering capabilities that closely emulate the sensing, processing, and adaptive qualities of human skin. This technology has the potential to address the limitations of conventional tactile systems, which often depend on centralized processing and may not provide the efficiency, responsiveness, and adaptability necessary for real-time interactions in complex environments.^36,37 By integrating bio-inspired neuromorphic computing with flexible, high-density sensor arrays, e-skin is poised to unlock transformative opportunities across various domains, including robotics, wearable bioelectronics, and environmental monitoring.^5,38–40 In addition to its applications in robotics, the development of wearable bioelectronics powered by neuromorphic algorithms holds promise for continuous monitoring of physiological signals. This innovation can offer valuable insights into health conditions, thereby enabling proactive medical interventions. Neuromorphic e-skin also contributes to advancements in smart textiles, human–machine interfaces, and environmental monitoring systems, leveraging the capability of neuromorphic systems to interpret complex environmental cues and foster more seamless interactions between devices and their surroundings.^34,41,42

To meet the diverse requirements of these applications, two distinct approaches to neuromorphic e-skin development have emerged. Direct systems, which embed neuromorphic hardware at the sensor level, provide localized, spike-based signal processing for immediate feedback and decision-making. This is particularly essential for applications such as prosthetics and robotics, which rely on real-time sensory–motor integration.^43,44 On the other hand, indirect systems utilize external neuromorphic processors to analyze data from distributed sensor networks, excelling in scenarios like wearable health monitoring and environmental sensing, where efficient processing of large-scale data is crucial for deriving meaningful insights.^45,46

This review aims to explore the transformative potential of neuromorphic e-skin technologies by bridging the gap between theoretical research and practical applications. It highlights the integration of BMIs with sensors and emphasizes their important roles in robotics, healthcare, and environmental systems. By drawing inspiration from the human nervous system, this work presents a thoughtful pathway toward the creation of artificial sensory systems that closely mimic human skin. These developments have the potential to reshape our interactions with technology, paving the way for smarter, more adaptive systems and encouraging innovative advances in human–machine interaction.

2. Neuromorphic devices for e-skin

2.1. Neuromorphic e-skin

Electronic skin (e-skin) represents a transformative development in the realm of flexible electronics, designed to replicate the diverse functionalities of human skin. This innovation involves a material that is flexible and stretchable and often also self-healing, embedded with a network of sensors capable of detecting various stimuli, including pressure, temperature, humidity, and chemical changes. The promise of e-skin spans multiple applications, such as robotics, prosthetics, health monitoring, interactive devices, etc., where it can provide both machines and humans with advanced sensory capabilities, enabling more nuanced and sophisticated interactions with their surroundings.^47–49

However, harnessing the full potential of e-skin is complex, primarily due to the enormous volume of data generated by its embedded sensors. These data need to be processed and interpreted effectively for the e-skin to function as intended. Neuromorphic devices, which are engineered to mimic the neuronal structures and processing mechanisms of the human brain, present a compelling solution to this challenge. The integration of neuromorphic devices with e-skin can facilitate real-time sensory data processing, low energy consumption, and adaptive learning, key attributes that are essential for the effective deployment of e-skin in various real-world scenarios (Fig. 1).^50,51 For instance, research conducted by Kim et al.⁵² presents an E-skin system integrated with a neuromorphic processor that could process tactile data in real-time. The neuromorphic processor emulates the brain's approach to sensory information processing, allowing the e-skin to respond to external stimuli with the speed and intelligence akin to biological skin. This capability is especially crucial in robotics, where the ability to quickly and accurately process tactile information can significantly improve the robot's dexterity and responsiveness.^50,52

Fig. 1 Integration of neuromorphic devices with e-skin.

Another vital advantage of incorporating neuromorphic devices into e-skin is their potential for energy efficiency.^46,53 Traditional processors tend to consume substantial amounts of power, especially when handling continuous and complex data streams from e-skin. Neuromorphic devices, on the other hand, replicate the energy-efficient nature of brain function, significantly reducing power consumption. It is proved that the neuromorphic e-skin system operated on extremely low power, making it particularly suitable for wearable and implantable applications. Such energy efficiency is critical for extending the operational lifespan of these devices and minimizing the frequency of recharging.^54,55

Furthermore, neuromorphic systems possess adaptive learning capabilities, enabling e-skin to enhance its sensitivity and responsiveness over time based on usage patterns and environmental factors.^45,56 This adaptability is particularly advantageous in the field of prosthetics, where e-skin must adjust to a variety of tactile inputs. For example, Harshil Patel and his coworkers developed an e-skin system integrated with a neuromorphic device capable of learning and adapting to different touch patterns, thereby improving its performance in practical applications.⁵⁷ This level of adaptability could lead to more intuitive and responsive prosthetic devices, significantly improving the quality of life for individuals with amputations. Overall, integrating neuromorphic devices with e-skin marks a critical advancement in the development of intelligent systems that can interact with their environment in a more human-like manner. By enabling real-time sensory processing, enhancing energy efficiency, and incorporating adaptive learning, neuromorphic e-skin systems hold the potential to revolutionize fields such as robotics, prosthetics, and beyond. This integration will make machines more perceptive, responsive, and capable of engaging with the world around them in increasingly sophisticated ways.^23,58,59

2.2. Neuromorphic e-skin devices: materials and integration considerations

Integrating neuromorphic devices with e-skin is a crucial advancement in improving its functionality, allowing it to closely mimic human skin's complex sensory processing abilities.⁶⁰ These neuromorphic devices, inspired by the neural architecture of the brain, enable real-time data processing, adaptive learning, and energy-efficient operations in e-skin systems.^36,61 There have been two highly effective and major types of neuromorphic devices, viz., memristors and transistors.^40,62,63 A memristor is a two-terminal non-volatile electronic component whose resistance depends on past current flow, enabling memory functionality. It typically consists of a thin switching layer between two metal electrodes.^64–66 In contrast, a transistor is a three-terminal semiconductor device used for signal amplification and switching. It regulates the current flow between two terminals based on the input to the third, making it essential in modern electronics.^67,68 There are various types of transistors, including bipolar junction transistors (BJTs), field-effect transistors (FETs), junction-field-effect transistors (JFETs), metal-oxide-semiconductor FETs (MOSFETs), complementary metal-oxide semiconductors (CMOS), insulated-gate bipolar transistors (IGBTs), etc.^67,69–72 They offer different operational principles and advantages according to their need. Both are important in modern neuromorphic electronic devices with distinct applications; however, it has been recognized that the memristor is among the most promising neuromorphic devices for e-skin applications up to now. The resistive switching characteristics of memristors with multi-state not only store information but also process it simultaneously, closely mirroring the synaptic functions found in biological neurons.^73,74 Memristors can form networks of artificial synapses, enabling intricate sensory data processing to occur directly within the e-skin, reducing the reliance on external computational resources.^75,76 Several studies explored the use of memristor networks within flexible electronics, demonstrating their potential for real-time data processing in wearable technologies. When applied to e-skin, memristors could greatly enhance the system's capability to process multiple sensory inputs simultaneously, leading to more sophisticated and responsive wearable devices.^7,50,77

Before assessing memristive devices for their potential application in neuromorphic computing and electronic skin, their characterization is mandatory. Electrical characterization includes measuring the current–voltage (I–V) characteristics, which give information on the switching mechanism, threshold voltage, and hysteresis. The set/reset voltage and ON/OFF ratio define the switching reliability between high and low resistive states which is an important characteristic of the memristive device. According to technology demand, many devices recently showed a set/reset voltage of less than ±0.5 V, and memory window 106 is significantly suitable for low-power-consumption electronic skin.^78,79 However, current compliance is needed to avoid the breakdown of the device. Synaptic performance of memristive devices can be assessed in terms of long-term and short-term plasticity (LTP/STP), spike-timing-dependent plasticity (STDP), as well as retention behavior, which together characterize the device's capability to emulate biological synapses. Many devices have large G_max/G_min ratios or dynamic ratios (suitable for multilevel conductance and for improvement of the identification accuracy of the resistance state), high linearity (ensures changes in conductance response to applied pulses, enabling precise synaptic tuning in memristive devices), and high I–V symmetricity, with low power consumption, meeting the needs of high-performance computing.⁷⁹ Endurance and reliability testing examines cycling stability which is reported to be more than 10⁶ cycles for memristive devices.⁸⁰ Although retention time is more than 10 years, environmental robustness verifies that the resistive states are stable and work smoothly in various environments.^32,81 Switching speed and energy efficiency are also paramount: the response times need to be rapid, and the power consumption needs to be low, especially for neuromorphic circuits operating in the real-time regime. Most memristive devices reported switching speeds within 100 ns and tens of fJ energy respectively.^82,83 Integrating this characteristic into a flexible substrate enhances the suitability of memristive devices for electronic skin, as studies suggest that fabricating these devices on flexible substrates enables consistent performance under stretching and bending conditions.^7,84

The conceptual working principle of neuromorphic systems based on multi-state memristors underscores their suitability for e-skin applications. Neuromorphic computing emulates biological neural systems, achieving efficient and adaptive processing through innovative architectures distinct from traditional computing. Multi-state memristors, a cornerstone of this paradigm, function as passive two-terminal devices that exhibit a nonlinear relationship between voltage and current. Their resistance, which changes based on the applied voltage's history, enables them to store information in a manner akin to biological synapses that adjust their strength based on neural activity.⁸⁵ Unlike conventional memory elements limited to binary states (0 or 1), multi-state memristors can represent several resistance levels simultaneously. This multi-state functionality enhances energy efficiency and processing speed by allowing multiple states to be accessed in a single operation.^86,87 Such capability aligns with synaptic behaviors like long-term potentiation and depression, where biological synapses modulate their strength in response to activity. This dynamic resistance modulation supports the design of artificial neural networks that can learn and adapt over time. Furthermore, the gradual resistance changes in these devices echo the nuanced strengthening and weakening of biological synapses, making them well-suited for implementing complex neuromorphic computing systems that process diverse sensory inputs.⁸⁸ The concept of the multi-memristive synapses is illustrated schematically in Fig. 2a.⁸⁹

Fig. 2 (a) Schematic of the concept of the multi-memristive synapses.⁸⁹ Reproduced from ref. 89 with permission from Springer Nature, copyright 2018. (b) The CMOS schematic for LIF neurons is divided into three functional blocks: the integration block (M₁–M₂–C_mem), the reset block (M₃–M₄–Cres), and the spike generation block (M₅–M₈).⁹⁰ Reproduced from ref. 90 with permission from arXiv preprint, copyright 2024. (c) The overall structure and main contributions of the CMOS-based ReRAM chip.⁹¹ Reproduced from ref. 91 with permission from Springer Nature, copyright 2022.

The operation of multi-state memristors relies on analog switching characteristics, which permit fine-tuning of resistance levels based on input signals. Such devices have demonstrated functionalities including potentiation and depression, mimicking synaptic plasticity found in biological systems. For instance, in the learning process, the memristor's resistance can be adjusted according to the intensity and frequency of input pulses, effectively storing information semantically as weights in artificial neural networks.⁸⁶ In neuromorphic architecture, multi-state memristors are often employed in crossbar configurations, where they function as synaptic weights between interconnected artificial neurons. This architecture allows for efficient vector-matrix multiplications, crucial for implementing neural network algorithms.⁸⁵ The inherent parallelism of memristive systems reduces the von Neumann bottleneck by enabling memory and processing to occur in the same location, akin to computations performed in the brain.⁷⁸

The deployment of multi-state memristors in neuromorphic computing is rapidly evolving, with research focusing on enhancing their stability, speed, and energy efficiency. Advances in materials science, such as the use of novel compounds and structuring techniques, aim to achieve greater reliability and dynamic range in memristive devices.^92,93 This paves the way for integration into real-world applications like robotics, artificial intelligence, and data processing systems that require high adaptability and low energy consumption. Neuromorphic computing based on multi-state memristors distinctly embodies the principles of biological computation, leveraging synaptic-like behavior for enhanced data processing capabilities. The ongoing research and development signify its potential impact on the future of intelligent systems, mimicking the brain's efficacy while overcoming the limitations posed by traditional computing architectures.

In general, the neuromorphic approach suitable for human e-skin is based on spiking neural networks (SNNs). SNNs process information using discrete spikes, similar to the action potentials in biological neurons, allowing for efficient handling of sensory data with temporal dynamics.⁹⁴ This timing-based processing is essential for replicating the nuanced temporal responses of human skin. For instance, research by Lee et al.⁹⁵ demonstrated the application of SNNs in flexible neuromorphic devices, highlighting their ability to effectively manage the complex, time-dependent data typical of tactile inputs. By integrating SNNs into e-skin, it becomes possible to mimic the temporal dynamics of human touch, enabling more lifelike and responsive sensory systems.⁹⁵ Besides oscillator-based neuromorphic devices provide a different approach, utilizing coupled oscillators to process information in a way that replicates the dynamic behaviors of neural circuits. These devices are particularly adept at detecting and processing rhythmic or periodic stimuli, such as vibrations. The integration of oscillator-based devices into e-skin can significantly improve its capacity to perceive and respond to dynamic environmental changes. For example, research conducted by Kim et al. involved the use of such devices to develop flexible sensors with high sensitivity to mechanical vibrations, showcasing their potential to enhance the e-skin ability to interact with its surroundings in real-time.⁵²

In recent years, neuromorphic computing has leveraged principles from neuroscience to develop computing systems that emulate the brain's information-processing capabilities. A significant research focus has been the integration of transistor-based neuromorphic e-skin, a bio-inspired, flexible platform designed to mimic the sensory functions of human skin. These transistor-based neuromorphic systems offer several advantages, including seamless integration with existing electronic infrastructure, high scalability, and low-power operation, making them highly suitable for next-generation artificial intelligence and human–machine interfaces.^8,96 Among all types of transistors, CMOS transistors serve as the foundation of modern electronics and neuromorphic computing architectures due to their unique combination of advantages that make them highly suitable for a wide range of applications nowadays. CMOS transistors comprise both p-type and n-type MOSFETs, which can efficiently switch between on and off states. Additionally, their dominance stems from several key features such as low power consumption, high noise immunity, scalability, and high density. This allows the formation of complex digital logic gates and analog circuits, essential for neuromorphic processing.^97,98 In neuromorphic e-skin applications, CMOS transistors are instrumental in emulating biological neurons and synapses, effectively translating neural activity into electronic signals. This capability enables real-time sensory feedback, adaptive learning, and energy-efficient signal processing, making CMOS-based neuromorphic e-skin a promising innovation in artificial intelligence, robotics, and wearable bioelectronics.⁹⁹Fig. 2b illustrates a schematic of the LIF neuron circuit in CMOS, which is divided into three functional blocks: the integration block (M₁–M₂–C_mem), the reset block (M₃–M₄–C_res), and the spike generation block (M₅–M₈).⁹⁰ Each block corresponds to a specific aspect of the LIF model and its underlying mathematical representation. The neuron integrates electrical signals over time, with a capacitor representing the accumulation of input. This capacitor charges as the neuron processes input signals until its voltage reaches a defined threshold. At this point, the circuit generates an output spike, signifying neural firing. The capacitors and resistive elements in the circuit emulate the membrane potential dynamics and leakage behavior characteristic of biological neurons. Specifically, input voltages from external signals charge the capacitor in the integration block. When the threshold is crossed, the spike generation block, driven by a network of transistors, produces a digital output signal. This implementation effectively mirrors the behavior of a spiking neuron. CMOS can emulate synaptic functions and plasticity, similar to biological synapses in neural signaling. CMOS–memristor hybrid circuits are a promising approach for implementing bio-plausible learning rules and various plasticity mechanisms. Memristors, as analog non-volatile memory devices, enhance energy efficiency and scalability while maintaining stable states. Their integration with CMOS allows for the implementation of spike-timing-dependent plasticity (STDP) curves, addressing the limitations of purely CMOS designs.¹⁰⁰

Neuromorphic computing architectures using CMOS feature large arrays of interconnected neurons and synapses, mimicking the brain's parallel processing. The designed connectivity reflects biological structures, with adaptable weights for learning from input data. CMOS enhances signal routing, maximizing speed and minimizing power consumption. For instance, spiking neural networks use crossbar-configured neuron arrays where memristors act as synapses with adjustable weights, allowing simultaneous signal processing. A key advantage of CMOS in neuromorphic computing is energy efficiency. Unlike traditional processors hampered by the von Neumann bottleneck, neuromorphic systems leverage in-memory computing, where data storage and processing occur together, significantly reducing energy overhead.^98,101 For instance, CMOS-based resistive RAM (RRAM) integrates storage with logic operations by utilizing the same architecture for both.¹⁰²Fig. 2c highlights the design, overall structure, and main contributions of the CMOS-based ReRAM chip.⁹¹ This structure minimizes energy dissipation typically associated with moving data across distinct memory and processing units, making it ideal for edge computing applications where power constraints are critical. Despite its advantages, CMOS-based neuromorphic systems encounter critical challenges as well, including higher power consumption relative to biological brains and limited capability in replicating complex neuronal dynamics. To overcome these constraints, researchers are investigating hybrid architectures that integrate CMOS with emerging technologies such as memristors. This approach aims to enhance energy efficiency and improve the fidelity of neuromorphic systems in mimicking biological neural processes, paving the way for more advanced and biologically realistic computing models.

To develop effective neuromorphic e-skin systems, it is crucial to carefully consider various key factors, such as the selection of materials, the design of device architecture, and the implementation of functional capabilities.^5,39,103 The choice of materials plays a fundamental role in determining the performance and durability of e-skin organic electronics. This is essential for creating the flexible and stretchable circuits required for e-skin applications. Until now various materials have been reported for neuromorphic e-skin applications such as biomaterials, carbonaceous materials, polymeric materials, and 2D materials. Table 1 summarizes neuromorphic devices based on various materials with potential applications in e-skin. Meanwhile, 2D materials like graphene and molybdenum disulfide (MoS₂) are gaining attention due to their exceptional electrical conductivity and flexibility, making them ideal for fabricating highly sensitive sensors within the e-skin.^104,105 These materials enable precise detection of stimuli such as pressure and temperature, which are essential for replicating the sensory functions of human skin. Additionally, biocompatible materials are indispensable for wearable or implantable e-skin, particularly in medical applications, where non-toxic and stable materials are critical for long-term safety and integration with the human body.¹⁰⁶ The architectural design of neuromorphic e-skin is equally important. Employing flexible and stretchable architectures is essential to ensure that neuromorphic devices retain their functionality while adapting to the contours of the human body.¹⁰⁷ Furthermore, the adoption of hierarchical structures that mimic the multilayered organization of natural skin significantly enhances the efficiency of e-skin systems, which work with different layers tailored for specific functions such as sensory detection, signal processing, and mechanical protection, thereby optimizing the overall performance. Functional capabilities within neuromorphic e-skin systems offer significant advantages in replicating the adaptive and responsive behaviors of biological skin. Synaptic plasticity, multimodal sensing and integration, and self-healing capabilities are essential functional aspects that contribute to creating more adaptive, resilient, and sensitive artificial skin. These capabilities enable e-skin to learn, adapt, and recover functionality after sustaining damage, thereby extending its operational lifespan and ensuring reliable performance over time for various applications.^108–110

Table 1 List of materials that have been used for neuromorphic e-skin applications to date

Material family	Name of the specific materials	Key features	Performance highlights	E-skin applications	Ref.
Biomaterials	• All cellulose-based	✓ Biocompatible	■ Responsive to stimuli	■ Pressure detection	5,8,31,111–113
	• Silk fibroin	✓ Biodegradable	■ Bio-mimic	■ Human motion monitoring
	• Gelatin	✓ Flexible	■ Response	■ Prosthetics
	• Sodium alginate	✓ Mechanical properties	■ Reversible contraction	■ Health monitoring
	• Collagen	✓ Renewable	■ Relaxation upon activation
		✓ Biodegradable resource	■ Precise detection of external pressure and temperature signals
Carbonaceous materials	• Activated carbon	✓ High electrical conductivity	■ High switching speed	■ Health monitoring	114–120
	• Amorphous carbon	✓ Large surface area	■ Energy efficiency	■ Tactile sensors
	• Carbon nanotubes	✓ Flexible	■ Fast response	■ Prosthetics
	• Graphene	✓ High thermal stability	■ Low power consumption	■ Robotics
	• Graphene oxide		■ Excellent memory characteristics
	• Carbon quantum dots		■ Significant hysteresis window
Polymeric materials	• All semiconducting polymers	✓ Good electrical conductivity	■ Short-term and long-term synaptic plasticity	■ Soft robotics	31,72,121–125
	• Polyvinylidene fluoride	✓ Lightweight	■ Perception of force, thermal, and light stimuli	■ Pressure sensor
	• Polydimethylsiloxane	✓ Flexible	■ Good sensitivity (10.89 ± 0.5 mV kPa^−1 in 80–230 kPa range)	■ Health monitoring
	• Ionic gels	✓ Piezoelectric behavior		■ Prosthetics
		✓ Stretchable
		✓ Multifunctional perception
2D materials	• Graphene	✓ High dielectric constant	• Linear synaptic weight updates	• Human electrocardiogram recognition	47,126–131
	• Ti3C2-MXene	✓ Controlled mechanical strain	• STDP behavior	• Wearable strain sensors
	• MoS2	✓ High responsivity	• High accessibility (<1% change between cycles)	• Pressure sensor
	• WSe2	✓ Specific detectivity	• Potentiation/inhibition	• Robotics
	• h-BN	✓ High electrical conductivity	• Spike-time-dependent plasticity	• Prosthetics
	• α-In2Se3	✓ Large surface area	• Pair-pulse facilitation	• Health monitoring
		✓ Flexible and stretchable	• Long retention
		✓ High thermal stability
		✓ Multifunctional perception

---

For neuromorphic e-skin to become a practical reality, it is crucial to focus on scalability. This means ensuring that these systems can be expanded to cover larger areas, such as full-body prosthetics or extensive robotic surfaces. Advanced manufacturing techniques based on printed electronics are essential for creating large-area e-skins.⁵⁵ These methods allow for the efficient production of extensive sensor arrays and integrated circuits over large surfaces. Another essential factor for the success of e-skin is power efficiency, especially for portable or wearable applications. Neuromorphic circuits need to operate with minimal power consumption, ideally harnessing energy from ambient sources or the user's body. This extends battery life and reduces reliance on external power supplies, making e-skin more suitable for continuous, long-term use. For instance, the development of ultra-low-power neuromorphic devices capable of energy harvesting, as highlighted by Durgesh Kumar et al.,¹³² significantly enhances the practicality of e-skin in wearable applications. This innovation presents substantial advancements in the development of energy-efficient systems for neuromorphic computing by optimizing the control of domain walls, as illustrated in Fig. 3a.¹³² This approach has enabled a reduction in the current density required to drive domain walls by a factor of 10⁴ compared to previously reported values for heavy metal-based spin Hall layers (Fig. 3b).¹³² Moreover, the seamless integration of neuromorphic e-skin with existing technological frameworks is vital for widespread adoption. This entails compatibility with current electronics, communication systems, and IoT networks to ensure that neuromorphic e-skin can be seamlessly incorporated into broader systems, such as healthcare monitoring platforms or smart robotic systems. This integration enables the creation of comprehensive, interconnected solutions that enhance the functionality and usability of e-skin in real-world settings. The study by Fen Miao et al. effectively demonstrated how neuromorphic e-skin could be seamlessly linked with IoT networks, providing real-time health monitoring and data transmission capabilities, marking a significant step toward practical implementation. Fig. 3c shows the architecture of the proposed system.¹³³

Fig. 3 (a) Schematic representation of the device structure designed for energy-efficient neuromorphic computing and (b) its corresponding motion behavior investigation under varying applied current densities.¹³² Reproduced from ref. 132 with permission from American Chemical Society, copyright 2023. (c) Conceptual architecture of the proposed neuromorphic electronic skin (e-skin) integrated with an internet of things (IoT) network for real-time health monitoring applications.¹³³ Reproduced from ref. 133 with permission from MDPI, copyright 2015.

Overall, memristor-based and transistor-based neuromorphic devices exhibit distinct performance characteristics. Memristors offer superior energy efficiency, especially at lower frequencies, with energy consumption in the range of tens of pJ per spike.¹⁷ They also demonstrate excellent scalability, particularly below 10 nm, and higher area efficiency compared to transistor devices. Memristors provide inherent non-volatility and can perform in-memory computing, which reduces power consumption and improves processing capabilities for neuromorphic applications.⁹⁸ Transistor-based devices benefit from established manufacturing processes, ensuring high reliability, precision, and scalability. However, their reliance on the von Neumann architecture limits energy efficiency and scalability for neuromorphic applications. In contrast, memristors effectively mimic biological neural functions, providing in-memory computation and synapse-like behavior ideal for neuromorphic computing. Despite their potential, memristors face challenges such as stability, repeatability, and integration into existing systems. Hybrid approaches combining memristors and transistors offer a promising solution, leveraging the speed of transistors for deterministic tasks while using memristors for adaptive, energy-efficient processing. This synergy is valuable for advanced neuromorphic systems, emphasizing parallel processing and learning capabilities.^98,134,135

The development of neuromorphic devices for e-skin presents a fascinating multidisciplinary opportunity that brings together advancements in materials science, electronics, and neuroscience. Recent research has identified the essential requirements for creating these neuromorphic devices, highlighting their features and characteristics tailored for specific applications.^98,136–139 Nowadays, neuromorphic systems have prominently relied on particularly memristor-based and CMOS transistor-based advances. In our analysis, the key performance metrics can generally understand differences in two technologies of memristors and transistors, as shown in Fig. 4. This study aims to provide valuable insights for the next generation of researchers and developers, guiding them toward the creation of more efficient and effective neuromorphic systems.

Fig. 4 Schematic illustration of comparative analysis between memristor-based neuromorphic systems and CMOS-based neuromorphic systems concerning performance metrics.

3. Types of neuromorphic e-skin

Neuromorphic e-skin is an advanced field that combines flexible electronics, neuromorphic computing, and sensory technologies to replicate the human skin's ability to sense and process environmental stimuli. E-skin uses flexible sensors and neuromorphic components to emulate the natural interactions between biological skin and the nervous system, enabling real-time detection and interpretation of signals like pressure, temperature, and texture. This capability has the potential to significantly advance areas such as prosthetics, robotics, and human–machine interfaces.^5,31,140 There are two main approaches for developing neuromorphic e-skin, “direct and indirect,” as shown in Fig. 5.

Fig. 5 Schematic of the types of neuromorphic e-skins.^141,142 Reproduced from ref. 141 with permission from Springer Nature, copyright 2014. Reproduced from ref. 142 with permission from Science, copyright 2018.

Recent advancements in neuromorphic e-skin devices have paved the way for two main systems: direct-type and indirect-type. Gaining a comprehensive understanding of the performance differences between these systems is crucial for designing effective and practical neuromorphic e-skin applications. Direct-type neuromorphic e-skin mimics skin-to-skin contact, offering several advantages that could enhance various applications. One major benefit is its rapid response time, ranging from ∼1 to 10 ms, which is essential for real-time applications in settings requiring immediate responses, such as prosthetics and interactive robotics. In contrast, indirect-type neuromorphic e-skin tends to have slower response times, ∼10–50 ms, due to added signal transmission layers or dependence on external computational units. This complexity can impact the system's efficiency, particularly in real-time-sensitive applications.^98,138,141

Another positive aspect of direct-type systems is their integration of neuromorphic computing. This allows for event-driven data processing, significantly reducing energy consumption. Research indicates that neuromorphic hardware within direct-type e-skins operates under extremely low energy requirements (<10 pJ per event), making them highly energy-efficient. Conversely, indirect-type systems may leverage traditional computing architectures that are less optimized for processing sparse, event-driven sensory data, resulting in higher energy consumption (∼100 pJ per event) due to continuous polling and less effective signal processing. This increased power demand could pose challenges for applications that need sustained operation in energy-constrained environments, such as wearable devices and autonomous robots.^98,138,143 Sensitivity is another area where direct-type e-skin shows promising potential. Studies have reported that these systems induce stronger neural activation in critical somatosensory areas, enhancing sensitivity to lower pressure levels (<10 kPa) for stimuli like stroking and tapping. This heightened sensitivity is advantageous for detecting subtle tactile cues and performing precision tasks. On the other hand, indirect-type e-skins tend to have reduced sensitivity (∼50 kPa), largely due to signal attenuation from additional layers.^6,98,138,144

In terms of multimodal sensing capabilities, direct-type e-skin excels through the seamless integration of neuromorphic components, allowing for efficient and simultaneous detection of various stimuli, including pressure, temperature, and texture. This capability is vital for applications requiring complex and adaptive sensory processing, such as advanced prosthetics, human–machine interfaces, and soft robotics. Although indirect-type systems can support multimodal sensing, they may experience challenges like a lower signal-to-noise ratio and slower integration times, which could affect their ability to offer high-fidelity sensory data. Additionally, direct-type systems possess greater ecological validity by closely emulating natural touch patterns and interpersonal interactions. Notable innovations in this area include utilizing the hypodermis layer of the skin as an electrolyte to improve electrical signal transmission in semiconducting films applied directly to the skin.^100,145 This method enhances signal transmission and improves device conformability to the skin, facilitating more realistic tactile sensing. In contrast, indirect-type systems, due to their layered structure, may not fully replicate the intricacies of direct touch, making them potentially less effective in contexts requiring intuitive and lifelike interactions.

Overall, the direct-type neuromorphic e-skin shows greater suitability for advanced applications compared to the indirect-type neuromorphic e-skin, which requires real-time feedback, high energy efficiency, superior sensitivity, and realistic tactile interactions from an application-oriented perspective. These applications include advanced prosthetics, precision robotics, healthcare monitoring, and immersive human–machine interfaces. While indirect-type systems may find relevance in less demanding contexts, addressing their limitations in response speed, energy consumption, and sensitivity could enhance their effectiveness for various technologies. It is crucial to recognize that both direct and indirect neuromorphic e-skin systems are continuously evolving. To enhance their utility across various applications, ongoing research should focus on improving multimodal sensing capabilities, optimizing energy efficiency, and minimizing latency. By understanding and addressing the performance differences highlighted in Fig. 6, researchers can make informed decisions in designing next-generation neuromorphic e-skin systems that are tailored to meet specific functional requirements and application domains.

Fig. 6 Schematic illustration of performance key characteristics between the direct type of neuromorphic e-skin and the indirect type of neuromorphic e-skin.

3.1. Direct neuromorphic e-skin

A direct neuromorphic e-skin integrates neuromorphic computing elements directly within the e-skin layer itself as shown in Fig. 7a.¹⁴¹ This approach allows the e-skin to not only detect stimuli but also process this information on-site, emulating the biological functions of human skin and the peripheral nervous system. The primary advantage of this method lies in its ability to perform real-time processing, as the sensory data are processed almost instantaneously within the e-skin. This capability is critical in applications where immediate feedback is necessary, such as in advanced prosthetics, where timely response to sensory inputs is essential for natural movement. Additionally, direct neuromorphic e-skin systems are typically more compact and lightweight because they combine sensing and processing functionalities into a single layer. This integration is particularly advantageous in wearable devices where space and weight constraints are significant concerns. Furthermore, the localized processing inherent to this approach results in lower power consumption, as it reduces the need to transmit large amounts of data to external processing units, making these systems more energy efficient. Majorly there are two types of direct neuromorphic e-skin, such as all memristor-based and all transistor-based direct types of neuromorphic e-skin. Both types offer unique capabilities regarding how it processes and responds to sensory inputs.

Fig. 7 (a) Schematic of an example of direct-type neuromorphic e-skin.¹⁴¹ Reproduced from ref. 141 with permission from Springer Nature, copyright 2014. (b) Schematic of the state-of-the-art for the memristive e-skin sensory system.³⁷ Reproduced from ref. 37 with permission from AIP Publishing, copyright 2024. (c) Illustration of a self-healing neuromorphic system mechanism.¹⁴⁶ Reproduced from ref. 146 with permission from Springer Nature, copyright 2020. (d) Schematic of working mechanisms of transistor-based neuromorphic e-skin.¹⁴⁷ Reproduced from ref. 147 with permission from Frontiers, copyright 2021. (e) Schematic image of the multigate flexible neuromorphic transistor for e-skin capable of learning through neural connections.³² Reproduced from ref. 32 with permission from Wiley, copyright 2023. (f) The functional plot of synaptic efficacy adjustments produced by the proposed TI-STDP and (g) its synaptic updates by our process closely align with the experimentally derived STDP.¹⁴⁸ Reproduced from ref. 148 with permission from arXiv preprint, copyright 2024. (h) Graphene-based synapse structure and (i) its long-term STDP potentiation and depression time scale vs. input spike duration.¹⁴⁹ Reproduced from ref. 149 with permission from IEEE, copyright 2020. (j) A neuromorphic tactile system combining sensing and mathematical computational coding functions.¹⁵⁰ Reproduced from ref. 150 with permission from Science, copyright 2024.

3.1.1. Memristor-based direct-type neuromorphic e-skin. Memristive e-skin represents a significant advancement in smart sensor technology by integrating memristors, which are memory-enabled electronic components, into a skin-like sensing architecture. This integration enhances functionality and efficiency, particularly in robotics, wearable electronics, and prosthetics where energy efficiency, adaptability, and responsiveness are crucial. The memristive e-skin offers two key advantages. Firstly, it has low power consumption as it can retain its last resistance state without continuous power, making it suitable for long-term use in wearables and prosthetic devices.¹⁵¹ Secondly, it demonstrates adaptive responses to external stimuli, mimicking the behavior of biological skin, which allows it to become more context-aware and adaptable over time.⁸ Additionally, its memory retention allows for neuromorphic computing, enabling the e-skin to recognize patterns and improve decision-making capabilities, particularly in advanced robotics and prosthetic limbs.¹⁵² From a scientific perspective, the integration of memristors into e-skin systems enhances energy efficiency and adaptability, which introduces advanced computational capabilities in flexible, wearable formats. By embedding memory and learning directly into the skin, technology reduces the need for external processing resources, paving the way for more autonomous, self-regulating devices.^31,135 These innovations push the boundaries of traditional electronic skin technologies, making memristive e-skin a promising platform for future intelligent systems capable of real-time learning, sensory perception, and adaptive decision-making. Significant research has propelled memristive e-skin forward, with numerous breakthroughs in its development.

In 2019, Kumar et al. proposed memristive e-skin sensors that are seamlessly integrated into a neuromorphic system designed for tactile sensing.³⁷ Using memristive e-skin allows for brain-like computing capabilities, enabling the e-skin to process sensory inputs similarly to biological systems. The e-skin has demonstrated impressive brain-inspired learning capabilities, enabling it to recognize and differentiate between various textures and pressure patterns. This advancement showcases the evolution of state-of-the-art bioinspired memristive sensory systems (Fig. 7b),³⁷ extending beyond replicating the physical properties of human skin to integrating artificial synapses and neural networks for advanced information recognition. The successful implementation of these memristive e-skin sensors contributes to the development of intelligent robotics and prosthetics capable of responding more intuitively to environmental stimuli. These advancements promise to enhance the interaction between robotic systems and their surroundings, improving their usability and effectiveness in real-world applications.⁵⁰

Further, the development of flexible and stretchable memristive e-skin with self-healing capabilities represents a significant advancement in electronic skin technology.²³ To this aim, satellite threshold-adjusting receptors (STARs) and satellite weight-adjusting resistive memories (SWARMs) are designed with self-healable ionic gels/dielectrics that heal themselves when subjected to damage (Fig. 7c).¹⁴⁶ This innovative system incorporates ionic liquid-based polymers and nanomaterials, such as metallic nanoparticles, to enhance both electrical conductivity and mechanical flexibility. These attributes allow the e-skin to serve as an efficient memristive device, capable of sensing environmental stimuli and retaining information through resistance-based memory.^60,153 A notable feature of this memristive e-skin is its self-healing capability, which enables it to autonomously repair itself following damage. This repair process is driven by molecular mechanisms within the ionic polymer matrix, allowing the material to restore its structure and functionality without external intervention. The ability to self-repair ensures continuous operation, even after physical damage, making technology particularly well-suited for long-term applications where reliability and durability are crucial. Additionally, the flexibility and stretchability of the memristive e-skin make it highly adaptable for integration into wearable technologies, including health monitoring systems and biometric sensors. Its ability to conform to the human body supports real-time physiological monitoring, advanced prosthetic systems with tactile feedback, and interactive devices capable of responding to touch and pressure. This combination of self-repair, memory retention, and sensory adaptability positions memristive e-skin as a promising platform for future wearable electronics, robotics, and prosthetics, where robust and intelligent sensory interfaces are required.¹⁵⁴

The development of array-based memristors has emerged as a pivotal advancement in the evolution of memristive e-skin technology. These memristors serve a dual role as both sensors and memory devices, allowing e-skin to mimic biological systems by providing sensory feedback while also retaining a long-term memory of inputs.^114,155 One key aspect of this research is the e-skin's ability to maintain memory trace of stimuli, particularly pressure, by altering its resistance. The ability of e-skin to “remember” and adapt to tactile stimuli opens doors to numerous practical uses, such as enhanced prosthetics and advanced robotics.^156–158 For instance, researchers at RMIT University have successfully created an artificial electronic skin capable of responding to pain stimuli like human skin.¹⁵⁹ This breakthrough could lead to improvements in prosthetic devices, smarter robotic systems, and non-invasive alternatives to traditional skin grafts. The implications of this technology are far-reaching, with potential benefits in human–machine interfaces, healthcare monitoring, and assistive technologies.¹⁶⁰ The enhanced ability of e-skin to provide both sensory feedback and memory retention not only improves user interaction but also paves the way for future advancements in fields requiring complex and adaptive human–machine interactions.

Although memristive e-skin technology shows immense potential, several critical challenges must be overcome to fully harness its capabilities. A key obstacle lies in the development of materials that can simultaneously provide flexibility, durability, and efficient memristive properties. Striking balance between these factors is essential for maintaining the long-term performance of e-skin, especially in dynamic environments that involve continuous bending, stretching, and deformation. While memristors are known for their energy efficiency compared to traditional electronic components, further optimization is needed to minimize power consumption during complex, multi-functional operations. Creating ultra-low-power systems that can handle sensing, data processing, and memory storage without excessive energy use is vital for practical applications in wearables, prosthetics, and robotics, where energy constraints are often critical. Another major challenge is the seamless integration of memristive e-skin with external electronic systems, such as wireless communication modules, neuromorphic processors, or AI-driven platforms. For memristive e-skin to fully realize its potential in applications like health monitoring, prosthetics, and robotic systems, efficient interaction between the e-skin and these systems is necessary, allowing real-time data transmission, processing, and decision-making.^7,161,162

Memristive e-skin represents a cutting-edge fusion of nanotechnology, advanced materials science, and neuroscience-inspired computing. Current research is focused on improving key functionalities such as adaptability, self-healing, and memory retention using memristive technologies. As advancements in materials science and fabrication techniques continue, memristive e-skin has the potential to transform industries such as healthcare, robotics, and prosthetics, making these systems more intuitive, intelligent, and responsive to real-world conditions. Looking ahead, memristive e-skin holds the promise of enabling next-generation technologies that can interact more naturally with both users and environments, offering new possibilities for smart, adaptive sensory systems and human–machine interfaces. Below, we have additionally summarized some of the previously reported all-memristor-based direct-type neuromorphic e-skin devices based on those previously reported in Table 2.

Table 2 Examples of all-memristor-based neuromorphic devices used for direct-type neuromorphic e-skin applications

Working mechanism	Materials	Neuromorphic functionality	Performance metrics	E-skin applications	Ref.
Electrochemical metallization (ECM)	Ag/GeSx	Mimics nociceptors, spike-timing-dependent plasticity (STDP)	High sensitivity to hazardous stimuli, fast adaptation (∼1 ms), amplification (>170%)	Robotic tactile sensing	17
Ion migration	TiO2, WO3	Multilevel conductance modulation, synaptic plasticity (PPF, EPSC)	Low operating voltage, high retention, multistate switching	Tactile sensing	163
Phase change mechanism (PCM)	Sb2Te3/GeTe	Gradual resistance changes for analog signal processing	High endurance, low power consumption, stable switching	Wearable electronics	7
Other emerging mechanisms (e.g., Mott transition, photonic-induced switching)	MoS2, WS2	Mimics neural dynamics, ion vacancy modulation for resistance switching	High scalability, low power operation (∼100 pA), fast response time	Flexible e-skin systems	96

---

3.1.2. Transistor-based direct-type neuromorphic e-skin. A transistor-based direct-type neuromorphic e-skin represents a significant advancement in bio-inspired electronics, designed to mimic the sensory and computational capabilities of human skin. These e-skins use neuromorphic architectures that emulate brain-like processing to detect and respond to stimuli like pressure, temperature, and touch in real-time. This technology is poised to transform fields like robotics, prosthetics, and healthcare, where adaptability and immediate sensory feedback are essential. The transistors in this e-skin perform both signal transduction and local computation. They amplify sensory input (e.g., pressure or strain) and mimic synaptic behaviors found in biological neurons, enabling real-time signal processing and pattern recognition. By integrating computational tasks directly into the transistors, these e-skins exhibit learning and adaptive responses, similar to biological skin. Numerous studies have been published on stretchable synaptic transistors, demonstrating progress in achieving various synaptic functions. These transistors operate based on several distinct mechanisms, as depicted in Fig. 7d, which compares biological synapses with the working principles of stretchable synaptic devices.¹⁴⁷ The key mechanisms include ion migration, electrochemical reactions, oxygen-induced persistent photoconductivity, and charge trapping. Each of these processes contributes uniquely to the transistor's ability to mimic synaptic behavior, making them essential for the development of flexible, neuromorphic systems. This feature enhances their ability to differentiate between various stimuli, making them highly autonomous and efficient. Direct-type neuromorphic e-skins integrate sensing and computation in the same layer, allowing for real-time responsiveness and energy efficiency by eliminating the need for external processors. They are typically made from oxide semiconductors or organic materials, which provide flexibility and stretchability, ensuring their functionality on irregular surfaces, such as in prosthetic applications, where they need to conform to the human body.

A key feature of these systems is their ability to mimic synaptic plasticity, enabling them to learn and adapt based on prior stimuli. Synaptic transistors emulate short-term and long-term plasticity, essential for learning processes. In prosthetics, this allows the E-skin to adjust sensitivity to repeated tactile inputs, improving user interaction with the environment. Additionally, spike-timing-dependent plasticity (STDP) enables neuromorphic E-skins to refine their responses over time by adjusting synaptic strength based on signal timing, similar to neuronal learning in the brain. Briefly, transistor-based direct-type neuromorphic E-skins combine advanced sensing, learning, and computational abilities into a single, flexible platform. By mimicking both the sensory and computational aspects of biological skin, these systems offer the transformative potential for robotic tactile feedback, prosthetics, and healthcare monitoring, pushing forward the development of intelligent, adaptive systems that function autonomously in real-time.

The development of transistor-based neuromorphic e-skin has rapidly progressed through a series of innovative research breakthroughs, particularly in the areas of synaptic transistors, flexible materials, and real-time processing. Fig. 7e effectively highlights the biological synapses and explores the diverse mechanisms that govern stretchable synaptic transistors. These mechanisms include ion migration, electrochemical reactions, oxygen-induced persistent photoconductivity, and charge trapping, each playing a crucial role in the functionality of these innovative devices.³² Each step in this evolution has contributed to the realization of intelligent artificial skin systems that mimic human skin's sensory and computational capabilities, with significant implications for prosthetics, robotics, and healthcare technologies. A critical advancement came in 2016 when Wan et al. developed a flexible synaptic transistor using oxide semiconductors that emulated the behavior of biological synapses. This important work has significantly advanced our understanding by demonstrating, for the first time, that neuromorphic plasticity encompassing both short-term and long-term memory responses can be effectively realized in flexible devices. The tunable plasticity of these artificial synapses allowed the e-skin to adapt its sensitivity based on previous stimuli, much like biological skin learns and adapts through neural connections. This research marked a significant leap toward creating an adaptive e-skin capable of learning from its environment, setting the stage for further developments in neuromorphic electronics.^9,164

Gebhardt et al. built on this foundation and introduced spike-timing-dependent plasticity (STDP) to neuromorphic e-skin. Fig. 7f and g visualize the functional shape of the synaptic adjustments produced by the core dynamics of TI-SDP compared to the classical curve yielded by (canonical) STDP, notably as it has been empirically derived and corroborated by various experiments conducted in history.¹⁴⁸ STDP is a learning mechanism observed in biological neurons, where the timing of stimuli directly influences synaptic strength. Gebhardt's work showed that integrating STDP into neuromorphic transistors enabled the e-skin to “learn” from repeated stimuli and adjust its response patterns accordingly. This breakthrough was particularly important for applications such as prosthetics and robotics, where the e-skin needed to refine its responses over time to improve tactile perception and dexterity.

Developing energy-efficient, compact, and bio-compatible artificial synapses is crucial for advancing large-scale neuromorphic systems. These artificial synapses need to transmit signals between neurons while behaving like biological synapses efficiently. Graphene, with its exceptional electronic properties, has emerged as a promising material due to its low-energy characteristics demonstrated in various electronic applications, including logic gates. Wang et al.'s study incorporates graphene to replicate crucial synaptic plasticity mechanisms such as spike-timing-dependent plasticity (STDP) and long-term plasticity, which are fundamental to learning and memory processes in the brain.¹⁴⁹ A key feature of graphene-based synapses is their programmability through the application of a back-gate bias voltage. This allows the artificial synapses to exhibit both excitatory and inhibitory behaviors, enabling tunable modulation of synaptic strength and timing. Such precise control over synaptic behavior is essential for the effective operation of neural networks, enabling dynamic adaptation of synaptic weights in response to input signals. Moreover, the proposed graphene-based synapses are designed (Fig. 7h) to meet stringent size and energy requirements, with a maximum area of 30 nm² and operating voltages as low as 200 mV². Additionally, the synapse can achieve a maximum synaptic weight change of 30%, which is influenced by variations in input spike durations. This flexibility enables the device to support a range of potentiation and depression time scales, from as narrow as [−1.5 ms, 1.1 ms] to as wide as [−32.2 ms, 24.1 ms] (Fig. 7i). These tunable temporal dynamics are critical for emulating the wide range of plasticity observed in biological systems. Overall, the graphene-based synapse demonstrates significant potential for application in large-scale, brain-inspired computational systems. Its combination of energy efficiency, compact design, and flexible plasticity behavior positions it as a strong candidate for future neuromorphic technologies, which demand scalable and adaptable components for the emulation of complex neural processes.

In progress, the fully integrated neuromorphic e-skin developed by Chen et al. represents a groundbreaking achievement in artificial tactile systems by effectively combining sensing and computational functions within a single layer (Fig. 7j).¹⁵⁰ This innovative design contrasts with traditional systems that separate these functions into multiple layers, which often results in increased complexity and latency. Toward this, several works have reported various sensors that provide similar functionalities as slow adapting (SA) and fast adapting (FA) receptors and also suggest their stacking to mimic the mechanoreceptor arrangement in the skin (Fig. 8a).¹⁶⁵ The direct integration of sensing and computational capabilities allows the neuromorphic e-skin to operate autonomously. By eliminating the need for external processing, this system significantly reduces latency, enhancing the speed and reliability of responses to tactile stimuli. As a result, it opens up new possibilities for real-time interactions in various applications, including robotics and prosthetics. With the ability to process sensory information immediately on the skin, the neuromorphic e-skin mimics the effectiveness of human touch for object recognition. This capability not only improves the accuracy of sensory information but also allows for more nuanced interactions with various surfaces and objects in the environment. This mimicking of human touch could lead to advancements in fields such as smart materials and human–machine interfaces.

Fig. 8 (a) The building blocks of neural pathways for tactile data processing in human skin including functions such as sensors, neurons, and synapses.¹⁶⁵ Reproduced from ref. 165 with permission from Science, copyright 2022. (b) Comparative concept of centralized and decentralized intelligence in robotics.¹⁴⁶ Reproduced from ref. 146 with permission from Springer Nature, copyright 2020. (c) The demonstration of the artificial tactile neural pathway capable of “in-skin” learning.¹⁶⁵ Reproduced from ref. 165 with permission from Science, copyright 2022.

Further, self-healing synaptic transistors have been developed to enhance the durability of neuromorphic electronic skin (e-skin). These advanced materials are designed to recover from mechanical stress and damage, addressing a critical challenge in wearable electronics and prosthetics. This breakthrough represents a significant step toward creating electronic skin with self-repairing capabilities, improving durability and functionality across various applications. The self-healing property is particularly important for devices exposed to mechanical stress in real-world environments, ensuring long-term viability and reliability. Fig. 8b further illustrates the comparative concept of centralized and decentralized intelligence in robotics.¹⁴⁶ In traditional centralized systems, signal transduction is separate from computation, with all learning processes centralized in a powerful processor. However, the proposed decentralized approach integrates learning directly into the sensor nodes, minimizing wiring complexity while simultaneously improving latency and fault tolerance. This decentralized architecture is particularly relevant for applications requiring real-time sensory processing and distributed intelligence, such as neuromorphic e-skin in robotics. This work, particularly highlighting self-healable neuromorphic mem-transistor elements for decentralized sensory signal processing, represents a significant advancement in robotics, enabling more resilient and adaptive systems.

Finally, the development of autonomous learning capabilities in e-skin systems was achieved by Liu et al., who created a neuromorphic e-skin capable of adjusting its sensitivity and responses based on environmental changes. The system demonstrates excellent bio-like synaptic behavior and shows great potential for in-hardware learning. This is demonstrated through a prototype computational e-skin, comprising event-driven sensors, synaptic transistors, and spiking neurons that bestow biological skin-like haptic sensations to a robotic hand. With associative learning, the presented computational e-skin could gradually acquire a human body-like pain reflex. Learning behavior could be strengthened through practice. Such a peripheral nervous system-like localized learning could substantially reduce the data latency and decrease the cognitive load on the robotic platform. These advancements collectively illustrate the rapid progress in the field of transistor-based neuromorphic e-skin, with each breakthrough contributing new functionalities such as adaptive learning, self-healing, and real-time signal processing (Fig. 8c).¹⁶⁵ These systems are increasingly capable of mimicking biological skin's complex sensory and computational functions, offering promising applications in prosthetics, robotics, and wearable technologies. Below, we have additionally summarized some of the previously reported all-transistor-based direct-type neuromorphic e-skin devices based on those previously reported in Table 3.

Table 3 Examples of all-transistor-based neuromorphic devices used for direct-type neuromorphic e-skin applications

Working mechanism	Materials	Neuromorphic functionality	Performance metrics	Applications	Ref.
Floating-gate transistor (FGT)	Carbon nitride (C3N4)	UV-responsive synaptic behavior	Low energy consumption (∼18.06 fJ per synaptic event)	Smart UV-detecting and blocking	125
Electric-double-layer transistor (EDLT)	IZO-based thin-film transistors	Long-term synaptic plasticity	High capacitance (>1.0 μF cm^−2), low operating voltage	Shape recognition	166
Electrochemical transistor (ECT)	Organic/inorganic hybrid	Pain perception	Sub-10 nm vertical structure	Nociceptor emulation	167
Ferroelectric field-effect transistor (FeFET)	—	Non-volatile synaptic plasticity	Large switching ratio, low power consumption	Smart textile	47
Optoelectronic synaptic transistor	InSe	Short-term and long-term plasticity (LTP)	Highly gate-modulated photo response	Gait perception	79

---

3.2. Indirect neuromorphic e-skin

Indirect neuromorphic e-skin involves a system where the e-skin functions primarily as a sensory interface, while neuromorphic processing occurs externally as shown in Fig. 9a.¹⁴² This separation between sensing and computation offers several key advantages in system optimization. Specifically, e-skin can be engineered for heightened sensitivity, flexibility, and durability, while the external neuromorphic processor can be designed for high-efficiency, large-scale data analysis and pattern recognition. This decoupling allows for independent advancement of both components, making the system well-suited for complex tasks requiring detailed sensory data processing, such as robotics, prosthetics, and human–machine interfaces. One significant benefit of this modular design is the ease of upgrading the neuromorphic processor without requiring changes to the e-skin itself. As neuromorphic hardware continues to evolve, the external processing unit can be replaced or upgraded independently, ensuring that the overall system can keep pace with advances in computational power, even in fast-moving fields.

Fig. 9 (a) Prototype example of an indirect neuromorphic e-skin where the e-skin functions primarily act as a sensory interface, while neuromorphic processing occurs externally.¹⁴² Reproduced from ref. 142 with permission from Science, copyright 2018. (b) Schematic demonstration of the flow system and e-skin components for implementing the artificial loop.¹⁰⁰ Reproduced from ref. 100 with permission from Springer Nature, copyright 2024. (c) Photos showing the excellent skin conformability of a bio-integrated e-skin system consisting of a temperature sensor, a pressure sensor, and two sets of RO-ED integrated circuits and (d) schematic showing the overall flow and e-skin components for the implementation of the artificial sensorimotor loop.⁸ Reproduced from ref. 8 with permission from Science, copyright 2023. (e) Overall schematic of the wirelessly powered tactile sensory system embedded in artificial skin (WTSA) and the converted tactile signal transfer process to stimulate the sciatic nerve.¹⁰⁰ Reproduced from ref. 100 with permission from Springer Nature, copyright 2024.

However, this separation between sensing and artificial intelligence (AI) processing does introduce challenges. One of the primary drawbacks is potential latency caused by the need to transmit data from the e-skin to the external processor. This can become a limiting factor in applications requiring real-time feedback, such as motion control in robotics or prosthetics. Additionally, wireless data transmission can increase power consumption, especially compared to systems where computation is performed locally within the e-skin. Higher power requirements pose challenges for wearable and portable devices, where battery life and energy efficiency are critical design considerations. Moreover, including an external processor may lead to bulkier systems, reducing the portability and comfort of wearable technologies, which can limit their practical use in certain applications.

Here, wireless neuromorphic e-skin systems are designed to mimic human skin's sensory functions, such as touch, temperature, and pressure. Fig. 9b shows an example of a flow system and e-skin components for implementing the artificial loop.¹⁰⁰ These systems convert external stimuli into electrical signals and are useful in healthcare, robotics, and prosthetics. The indirect type of wireless neuromorphic e-skin features arrays of sensors that detect pressure, temperature, humidity, or strain using piezoelectric, piezoresistive, or capacitive materials. Instead of processing these sensory data locally, the electrical signals are wirelessly transmitted to a nearby processing device using communication technologies like Bluetooth, near-field communication (NFC), or radio frequency identification (RFID). Each technology offers unique strengths and limitations, which dictate its applicability in different contexts. Bluetooth technology is widely used for its high data transfer rates and extensive compatibility with various electronic devices. It can transmit sensory data in real-time over distances of up to 10 meters, making it ideal for applications like prosthetics and robotics, where continuous monitoring and dynamic interaction are essential. For instance, in prosthetics, Bluetooth enables spiking neural network (SNN)-enabled e-skin systems to process complex sensory data, such as variable pressure levels or tactile patterns, in real-time.¹⁶⁸ However, the primary limitation of Bluetooth is its relatively high power consumption, which can constrain its use in ultra-low-power devices. Since energy efficiency is a critical parameter for wearable and implantable devices, developers must balance the benefits of real-time performance with the challenges of power optimization. Efforts to mitigate this issue include integrating advanced power management circuits or optimizing data transmission protocols to reduce energy demand. NFC is another widely used communication technology in wireless neuromorphic e-skin systems. It operates efficiently over short distances of less than 10 cm and is known for its extremely low consumption. These features make NFC particularly well-suited for secure and intermittent data transfer, such as in health monitoring and diagnostic applications. In health diagnostics, NFC-based e-skin can transfer localized data, such as heart rate, blood oxygen levels, or skin temperature, to an external device like a smartphone or a medical monitoring system. The energy efficiency of NFC is a major advantage for lightweight, portable, and power-constrained systems. This technology is especially beneficial for individuals who require frequent but short bursts of monitoring, such as patients with chronic conditions. Despite these advantages, NFC's limited range and lower data transfer rates make it unsuitable for applications requiring real-time or remote sensing capabilities. Its use is generally restricted to applications where secure, localized data exchange is more important than high-speed communication. This includes scenarios like wearable health patches, where small bursts of data transmission are sufficient for monitoring purposes. RFID technology, especially in its passive mode, offers a unique advantage in e-skin systems: it does not require a dedicated power source. This makes it highly energy-efficient and scalable for large sensor arrays.³⁶

Additionally, RFID is a cost-effective solution, making it attractive for applications in industrial and healthcare settings where extensive monitoring of environmental or physiological parameters is needed. RFID-based e-skin can be employed in tracking environmental data, such as humidity, pressure, and temperature, over extended periods. Its passive nature and energy efficiency allow for the deployment of large-scale, low-cost sensing networks without the need for frequent maintenance or battery replacement. However, like NFC, RFID has significant limitations, including a shorter range and limited bandwidth. These constraints make RFID less suitable for high-speed or real-time tasks, where the rapid processing and transmission of large amounts of data are necessary. Its primary use case lies in periodic or static monitoring, where scalability and cost-effectiveness are prioritized over speed and immediacy.¹⁶⁹

When comparing the three communication technologies, it becomes evident that each has its distinct advantages tailored to specific applications: bluetooth is the most viable choice for dynamic applications like prosthetics and robotics, where real-time processing and responsiveness are critical. Its higher data transfer rates and extended range provide a reliable solution for continuous monitoring and complex sensory data processing. However, the trade-off is its higher power consumption, which necessitates advanced energy management strategies for prolonged usage. NFC is best suited for low-power, short-range interactions, such as wearable health diagnostics. Its energy efficiency and secure data transfer capabilities make it ideal for localized applications, though its limited range and slower data rates restrict its use in more dynamic or remote sensing scenarios. RFID excels in scalable and periodic sensing tasks, particularly in industrial or healthcare contexts where passive operation and cost-effectiveness are priorities. While it is not suited for high-speed or real-time processing, its passive nature makes it a practical choice for large sensor networks or applications requiring extended monitoring durations.

This wireless approach eliminates the need for physical connections, making the system lightweight, flexible, and scalable, suitable for soft and stretchable electronics. Once the sensory data reach the remote device, neuromorphic algorithms, modeled on the brain's neural networks, are applied to interpret the signals. Neuromorphic computing mimics the spiking behavior of biological neurons and synapses, allowing the system to analyze patterns in the sensory data and convert them into meaningful information, such as identifying the presence of an object or gauging the intensity of a touch. This indirect processing approach mirrors the nervous system, where sensory signals are transmitted to the brain or spinal cord for interpretation. The result is a system that efficiently processes sensory information with high responsiveness, flexibility, and potential for diverse applications.

The development of flexible, pressure-sensitive e-skin using piezoresistive materials and wireless transmission modules represents a significant advancement in wireless neuromorphic systems.¹²⁶ This technology enhances interaction with environmental stimuli by mimicking the sensitivity of human skin, improving user experiences in applications such as health monitoring and robotics. The integration of piezoresistive materials allows the e-skin to accurately detect pressure stimuli, a crucial feature for these fields.^170,171 The system transmits the pressure signals to a remote processor, which utilizes spiking neural networks (SNNs) to interpret the data. By mimicking the behavior of biological neurons, SNNs enable the differentiation of various pressure levels and patterns, which can be complex and nuanced.¹⁷² The adoption of SNNs is particularly significant due to their computational efficiency and real-time processing capabilities, which are essential for responsive human–machine interactions. This innovative approach minimizes the energy consumption typically required for local data processing. By offloading data analysis to a remote processor, the design improves the e-skin applicability in compact and lightweight devices, maximizing portability and practicality.^154,173

In a study by Wang et al., researchers developed an innovative wireless e-skin that closely mimics the sensory feedback and mechanical properties of natural skin.⁸ This groundbreaking technology offers a promising solution for next-generation prosthetic devices by enhancing the user experience significantly. The e-skin design features flexible, thermosensitive, and capacitive sensors that can detect a range of stimuli such as temperature and touch. This capability is depicted in Fig. 9c.¹⁷⁴ One of the standout features of this system is its ability to wirelessly transmit sensory data to external processors, which emulate spike-based neuronal communication (Fig. 9d).⁸ This approach mirrors how sensory information is naturally processed in humans, enabling prosthetic users to receive real-time feedback expressed through spiking patterns. By allowing users to feel and respond to sensory information instantly, this wireless e-skin development has the potential to transform the overall experience for individuals with prosthetic devices. It paves the way for more natural interactions with the environment, ultimately enhancing the quality of life for those who rely on prosthetics. With this technology, users can more fully engage in their surroundings, bringing them closer to the sensory experiences that many take for granted.

Kyowon Kang and his colleagues recently made significant strides in wireless neuromorphic electronic skin (e-skin) technology by developing a fully implantable, wirelessly powered tactile sensory system embedded in artificial skin (WTSA). This groundbreaking research aims to restore tactile functions lost due to severe skin damage and enhance wound healing, offering an innovative solution in the field of regenerative medicine. The WTSA is designed for stable operation as a sensory system, addressing urgent needs in tactile sensory replacement and wound treatment. The system provides three main advantages: (i) it replaces severely damaged tactile sensory functions with a device that has broad biocompatibility, (ii) it supports wound healing and skin regeneration by incorporating collagen and fibrin-based artificial skin (CFAS), and (iii) it minimizes foreign body reactions through a hydrogel coating on the neural interface electrodes. The WTSA's stable performance was validated by quantitative analysis, demonstrating its ability to detect leg movement angles and electromyogram (EMG) signals in response to varying levels of applied pressure. The fabrication of the WTSA consists of five key components: an artificial skin layer, a crack-based tactile sensor, a wirelessly powered power management circuit (WPPFM), hydrogel-coated neural interface electrodes, and multilayered encapsulation that ensures durability for long-term implantation. Fig. 9e in the study illustrates the structure and step-by-step fabrication process, highlighting the integration of these elements to create a reliable, long-lasting sensory system.¹⁰⁰ This innovation represents a significant advancement in developing implantable e-skin technologies, aimed at restoring tactile sensation and facilitating the healing of severely damaged skin.¹⁰⁰

4. Applications of neuromorphic e-skin

4.1. Smart textiles applications

Neuromorphic devices, inspired by the neural structure and functioning of the human brain, have shown immense potential in smart textiles and wearable electronics. These devices offer efficient signal processing, adaptive learning, and low power consumption, making them suitable for e-skin applications that require continuous monitoring and real-time response. This section provides a detailed scientific explanation of neuromorphic devices in smart textiles and wearable electronics and outlines stepwise progress in this field. In 2019, Bae et al. introduced a new approach to creating neuromorphic textiles using pre-cleaned and catalyst-treated cotton yarns to make polydopamine-intercalated fabric memory (PiFAM).¹⁷⁵ The PiFAM was developed directly on a cotton substrate through a simple room-temperature and atmospheric-pressure fabrication process, making it perfect for wearable electronics. The fabrication method, shown in Fig. 10a, involved a two-step solution dip-coating process that resulted in a highly uniform deposition of polydopamine (PDA) layers onto the textile fibers. The PiFAM's memory function is achieved through resistive switching, where the formation and rupture of conductive filaments between adjoining PDA layers act as the resistive switching layer (RSL). This mechanism contributed to the excellent electrical performance of the device, including a wide sensing window, great data retention, and strong endurance over multiple switching cycles, which are crucial for high-performance wearable memory devices. The work paves the way for intelligent textile systems capable of advanced human–machine interaction and integration into IoT applications.¹⁷⁵

Fig. 10 (a) Schematics of Al and PDA coating procedures on fabric substrates using a two-step solution dip-coating method.¹⁷⁵ Reproduced from ref. 175 with permission from Wiley, copyright 2019. (b) Recognition simulation results for the Modified National Institute of Standards and Technology (MNIST) and electrocardiogram (ECG) pattern recognition tasks.¹⁷⁶ Reproduced from ref. 176 with permission from Science, copyright 2020. (c)–(e) Schematic image of the textile memristor network, including the top-layer device with synaptic plasticity and the bottom-layer device with neural functions. Schematic of the intelligent fiber heating system. The key fiber units are artificial synapses, artificial neurons, resistors, and the operation mechanism of the intelligent warm fiber.⁶⁰ Reproduced from ref. 60 with permission from Springer Nature, copyright 2022. (f) Schematic of neuromorphic e-skin-system integrated triboelectric nanogenerators (TENGs) for tactile sensing and organic electrochemical transistors (OECTs) for information processing and their (g) and (h) output performance with applied pressure (left image), and frequency (right image) and stability and durability tests of stretchable TENGs undergoing continuous 1000 cycles within 600 s, respectively.⁸⁸ Reproduced from ref. 88 with permission from Wiley, copyright 2023.

Ham and colleagues have achieved significant progress in e-textiles by successfully creating one-dimensional (1D) fiber-shaped artificial multi-synapse-based ferroelectric organic transistors on a thin silver (Ag) wire. These 1D synaptic devices exhibited a variety of distinct and well-defined synaptic behaviors, including short-term plasticity (STP), long-term potentiation (LTP), long-term depression (LTD), spike-rate-dependent plasticity (SRDP), and spike-timing-dependent plasticity (STDP). The transitions between LTP and LTD were stable under repeated stimulation cycles, maintaining consistent performance even after 100 mechanical bending cycles at fixed bending radii (R = ∞, 5, and 2.5 mm). The durability and flexibility of these 1D multi-synapse devices make them promising components for constructing neuromorphic networks within electronic textiles (e-textiles). The 1D design allows for easy integration into NOR-type synaptic arrays, which can be expanded to create large-scale textile-based neuromorphic systems. To showcase their potential, various synaptic arrays were produced, including 2 × 2, 3 × 2, and 10 × 12 configurations. These arrays were tested as proof-of-concept e-textile neural networks, successfully enabling the propagation of output signals through each synaptic cell without interference from undesired neural signals, confirming the functionality and reliability of the system. Notably, the 1D multi-synaptic arrays achieved recognition accuracies of approximately 90% for the Modified National Institute of Standards and Technology (MNIST) dataset and about 70% for electrocardiogram (ECG) pattern recognition tasks. Additionally, the initial recognition accuracy for the ECG patterns remained consistently within a margin of error of ∼2%, irrespective of mechanical bending stress, as illustrated in Fig. 10b.¹⁷⁶ This resilience under deformation underscores the durability of the synaptic devices for wearable applications.

In progress, Wang et al. developed a reconfigurable textile memristor network capable of functioning as both artificial synapses and neurons, with very low energy consumption.⁶⁰ The design involved a three-dimensional textile neural network with an interwoven structure, where the top and bottom layers acted as artificial synapses and neurons, respectively. The architecture of the network is illustrated in Fig. 10c,⁶⁰ showing interconnected textile memristors, with a photograph of the fabricated device provided in the inset. The system exhibited a remarkably low energy consumption of only 1.9 fJ per spike, making it approximately three orders of magnitude more energy-efficient than both biological neurons and previously reported artificial neurons. Additionally, Wang et al. integrated the artificial synapse, neuron, and functional memristors into a textile-based heating system, enabling intelligent temperature regulation. The neuromorphic textile autonomously adjusted the temperature in response to external stimuli by modulating the electrical current through the textile network, providing a responsive thermal management solution. Fig. 10d and e illustrate the integration of the neuromorphic e-skin into the textile heating system and outline the operational mechanism, demonstrating how the system combines neuromorphic computing with smart textile functionality.⁶⁰ The ultralow-power textile memristor network not only demonstrates the potential for reconfigurable brain-inspired computing but also opens new possibilities for wearable neuromorphic electronics within the framework of intelligent Internet of Things (IoT) applications.⁶⁰

Recently, M. Li and colleagues introduced a novel fibrous organic electrochemical transistor (FOECT), fabricated using functional boron nitride (FBN)-mediated polypyrrole (PPy) neurofibers combined with an ion-gel dielectric. This innovative design demonstrated exceptional device characteristics, including a high switching ratio (>10⁴), strong switching stability, and high transconductance (24.6 mS) at a low operating voltage (<1 V). The FOECT exhibited diverse synaptic behaviors, such as excitatory post-synaptic current (EPSC), inhibitory post-synaptic current (IPSC), and short-term plasticity (STP), showcasing its capability to mimic biological synapses. Benefitting from the robust reticular PPy nanonetwork, the FOECT maintained stable synaptic performance over 4000 cycles under different stimulation amplitudes, indicating its long-term operational reliability. Notably, the device operated efficiently at a low reading voltage of 1 mV and exhibited an ultralow energy consumption of 0.85 pJ/spike at a presynaptic stimulus of 30 mV. The energy requirements further decreased as the spike voltage increased, making it an energy-efficient option for neuromorphic computing. In addition to neuromorphic functionalities, the FOECT also demonstrated sensing capabilities, effectively monitoring C-reactive protein (CRP) levels across a linear range from 10 pg mL⁻¹ to 0.2 mg mL⁻¹ with a correlation coefficient (R²) of 0.966. The device exhibited consistent and reliable responses to various CRP concentrations, with stable current outputs measured at 97.33 ± 3.86 μA, 103 ± 2.45 μA, and 146 ± 5.35 μA for 10 pg mL⁻¹, 1 ng mL⁻¹, and 0.1 mg mL⁻¹, respectively. Overall, the development of the fibrous OECT combining neuromorphic computing and biosensing capabilities marks an important step toward future wearable, human–machine interactive devices. The ability to integrate both computing and sensing functions into a flexible, low-power platform opens new opportunities for advanced applications in wearable technology, including adaptive biosensors and intelligent electronic textiles.⁸

Expanding wearable technologies to include artificial tactile perception is crucial for the advancement of intelligent human–machine interfaces. Neuromorphic sensing devices, known for their low energy consumption and efficient operation, are promising candidates for achieving this goal. To enable wearable artificial tactile perception, the devices must exhibit skin-compatible and conformable properties that allow for seamless integration with the body. In this context, Wu and colleagues have reported a novel intrinsically stretchable, skin-integrated neuromorphic system that combines triboelectric nanogenerators (TENGs) for tactile sensing and organic electrochemical transistors (OECTs) for information processing (Fig. 10f).⁸⁸ The system is designed with a hierarchical structure, integrating TENGs with neuromorphic transistors to mimic the organization of biological tactile sensors and neurons. This configuration endows the system with essential properties for wearable applications, including high sensitivity to low pressures (∼0.04 kPa⁻¹), as demonstrated in Fig. 10g, and the ability to emulate both short- and long-term synaptic plasticity. Additionally, the device exhibits exceptional mechanical durability, withstanding over 10 000 switching cycles without performance degradation (as shown in Fig. 10h).⁸⁸ The system also supports symmetric weight updates, which is beneficial for learning algorithms, and can be stretched up to 100% strain, making it highly suitable for conformal wear on the skin. With neural encoding capabilities, the neuromorphic system can effectively recognize, extract, and encode tactile information, allowing it to mimic the functionality of biological skin in detecting and processing tactile stimuli. These features are sensitive to low pressure, robust synaptic plasticity, and high mechanical stretchability positioning this integrated neuromorphic system as a promising solution for the next generation of wearable devices aimed at providing artificial tactile perception. The capabilities demonstrated make the system an attractive platform for developing advanced wearable technologies for human–machine interfaces. It also enables the delivery of intelligent sensory feedback and adaptive responses for real-world applications.⁸⁸ The development of neuromorphic e-skin devices for smart textile applications has been widely reported. However, several scientific challenges need to be addressed. These include achieving energy efficiency comparable to biological systems and integrating rigid electronic components into flexible, durable, and washable textiles. To advance neuromorphic e-skin technology, future directions include developing reconfigurable memristor networks, improving in-sensor computing capabilities, integrating energy harvesting technologies, implementing more advanced neuromorphic architectures, and creating adaptive and learning e-skin systems. Progress in these areas has the potential to revolutionize applications in healthcare, human–machine interfaces, and smart clothing.

4.2. Tactile sensing applications

Seeing, touching, tasting, smelling, and hearing are the senses that people use to interact with and comprehend their environment. Our ability to use one or more of these senses allows us to move through unfamiliar and uncertain settings with ease. For example, we often take for granted how easily we can handle something as simple as an egg. Our brain receives instantaneous information about an egg's shape, size, color, texture, and temperature from our sense receptors when we hold it. We instinctively know that if we squeeze the egg too hard, it will shatter; if we squeeze it too weak, it will escape our grip. The secret is to apply precisely the proper amount of pressure, allowing our senses to continuously provide feedback to keep the egg intact. Additionally, our brain incorporates past information regarding the egg's fragility and weight, which makes the operation nearly effortless. But if a robot were given the same task, it would have to rely on human-like sensory inputs that would give it the feedback required for manipulating objects successfully. Since a robotic manipulator cannot learn about an object like human beings can, accurate sensory feedback becomes even more important to allow it to discover and communicate with its environment.¹⁷⁷

One sort of sensory technology called tactile sensing enables a system or device to collect data about an object's physical properties by making direct physical touch with it. Tactile sensors identify and interpret an object's texture, shape, softness, and temperature, much like the human sense of touch does. Guifen Sun et al. developed an adaptable and breathable tactile sensor with unique personal thermal management characteristics, as shown in Fig. 11a. A breathable, multipurpose tactile sensor with unique personal temperature management features has been developed.¹⁷⁸ Because of this, tactile sensing is crucial for jobs requiring a thorough comprehension of the physical world, such as robotics, where machines must precisely and carefully interact with objects. A wide range of sensors are categorized as tactile sensors if they are designed to collect tactile information by direct physical contact or touch with the object under study. Temperature, vibration, softness, surface texture, form, and the forces acting on the object – both shear forces, which operate perpendicularly to the surface, and normal forces, which act parallel to the surface – consisting of weight or pressure can all be detected by these types of sensors. One of these characteristics may be measured by a single touch sensor, or in certain situations, a combination of several. Specific features, such as pressure and torque sensing, are frequently excluded from the broad definition of tactile sense in specific talks. Nonetheless, torque and pressure are also regarded as crucial tactile factors because they can both be felt with the human touch. Torque describes the rotational forces that impact an object, while pressure indicates the amount of force applied to the surface. These characteristics are frequently included in the larger field of tactile sensing research since they are crucial for applications that call for precise object handling.

Fig. 11 The scope of tactile sensing technologies. (a) A flexible and breathable tactile sensor with unique personal thermal management ability is developed.¹⁷⁸ Reproduced from ref. 178 with permission from Elsevier, copyright 2023. (b) Schematic illustration of a remote tactile sensing system with magnetic synapse inspired by human tactile sensing and synaptic transmission.¹⁷⁹ Reproduced from ref. 179 with permission from Springer Nature, copyright 2017. (c) Schematic illustration of the tiny signal perception based on the HSP-MP&HM structured tactile sensor.¹⁸⁰ Reproduced from ref. 180 with permission from Elsevier, copyright 2022. (d) Schematic of the skin-inspired piezoelectric tactile sensor array.¹⁶⁷ Reproduced from ref. 167 with permission from Wiley, copyright 2021.

This review paper examines artificial tactile sensor design, showing how scientists have created a range of technologies that aim to replicate the human tactile experience. By broadening the opportunities for automation in dynamic and unstructured contexts, these artificial sensors are made to interpret physical interaction in a way that allows machines and robots to operate objects with higher sensitivity and precision. Fig. 11b depicts the workings of a distant sensory system modeled after human touch perception and synaptic transmission.¹⁷⁹ When exterior sensations that are tactile enter the body, the human fingertip's mechanoreceptors generate bioelectrical impulses in response. The design displays a remote tactile sensing system with a magnetic synapse, which is modeled after human tactile sensing and synaptic transmission. Mechanoreceptors in the fingertip sense external tactile inputs and convert them into electrical impulses, which are subsequently transported through the central nervous system via synaptic transmitters and then interpreted by the brain.¹⁷⁹

Tactile sensing has developed into sophisticated technology over the last thirty years. For a long time, a lot of people thought that touch sensors would completely change factory floors through revolutionary industrial robots and automation. However, despite these optimistic expectations, industrial settings hardly ever use these sensors. It begs the question: Did other factors influence the development of tactile sensing, or did it not live up to the hype in the field of robotics? We investigate the situation of tactile sensing presently in this paper, and we find that its direction has changed significantly. We review early hopes and forecasts, consider the findings of current studies, and show how the field has unexpectedly changed. Through an examination of ongoing initiatives and developing patterns, we pinpoint exciting new fields where tactile sensing has the potential to have a big influence. Interestingly, it appears that this technology is now poised to be extremely important in increasingly dynamic and unpredictable contexts, particularly in the areas of healthcare, medicine, surgery, service robotics, and handling natural products.

The 1980s marked a significant period for tactile sensing research and development. During this decade, there was growing recognition of the importance of tactile sensor technology. These pioneering investigations sparked a wider interest in the topic by laying the foundation for understanding how tactile sensing may be used in robotics and mechatronics. As evidenced by a plethora of literature, other researchers have since examined and developed tactile sensing technology in a variety of applications, notably in robotics and mechatronics.^181–184 Lee presented a succinct but comprehensive analysis of tactile sensing technology in 2000. It looked at how technology had evolved over time and critically analyzed the reasons behind its sluggish uptake in the consumer and industrial markets. The technological difficulties and commercial obstacles that have impeded the extensive adoption of touch sensors in many industries were brought to light in this review.¹⁸⁵

Hongsen Niu et al.¹⁸⁰ demonstrated an ultrasensitive capacitive touch sensor in their study. The sensor incorporates two Ag-PDMS electrode layers with a hierarchical microcone (HM) structure and a P(VDF-TrFE)-TiO₂ dielectric layer with a hierarchical sea-urchin-like TiO₂ particle-in-micropore (HSP-MP) structure. This was accomplished with a simple and economical method. To accurately record pressure mapping information such as the mass, shape, and location of different objects, a high-resolution flexible perception array was also created as presented in Fig. 11c.¹⁸⁰ Moreover, a piezoelectric tactile sensor array modeled after human skin was developed in a work by Weikang Lin et al.¹⁸⁶ Human skin, which is a remarkable sensor, can concurrently sense the intensity of multiple stimuli, including pressing, tapping, slipping, and bending. The array affixed to the human skin is depicted in the schematic depiction of the real-time tactile sensor array technology (Fig. 11d),¹⁶⁷ where it detects various external stimuli and relays the information to the signal processor and peripheral devices. Furthermore, Eltaib and Hewit presented a thorough examination of tactile sensing devices made especially for minimally invasive surgery three years later, in 2003.¹⁸⁷ Their research highlighted the critical role that tactile sense could play in improving surgical safety and accuracy. The argument put up was that tactile feedback technology holds particular significance in the medical industry, since it may significantly enhance the precision and control of surgical tools. This is especially useful for delicate treatments carried out in minimally invasive environments. According to their research, tactile sensing is becoming more and more relevant in domains other than standard industrial uses. These disciplines include healthcare and other specialized fields.¹⁸⁸ The current state-of-the-art in the field, trends in tactile sensor research, challenges that need not be met to be addressed, operational principles, and benefits and drawbacks of various tactile sensor designs are all covered in this paper. This expands on earlier reviews of tactile sensing technology. In addition to the previously investigated areas, we suggest new uses for this technology in the areas of leisure sports, aerospace engineering, automobile manufacturing, and rehabilitation medicine.

4.3. Healthcare and biomedical applications

The development of neuromorphic e-skin devices represents a significant leap forward in the fields of healthcare and biomedical applications. The field has seen remarkable growth, particularly in the last few years, with breakthroughs in materials science, sensor integration, and neuromorphic computing. Researchers have developed e-skin systems that not only sense various stimuli but also process this information in a manner analogous to biological neural networks. This innovative technology, emerging from the fusion of two fields of science and e-skin systems, aims to bridge the gap between artificial sensory systems and the complex, multifaceted capabilities of human skin. By mimicking the biological sensorimotor loop and integrating advanced materials with neuromorphic computing principles, these devices are poised to revolutionize prosthetics, health monitoring, and human–machine interfaces. This section will explore the recent advancements in neuromorphic e-skin devices, focusing on their stepwise progress, key innovations, and potential impact on healthcare and biomedical applications.

Researchers are making remarkable strides in the development of electronic skin (e-skin) devices, significantly enhancing their sensing capabilities. A central area of focus has been tactile sensing, which equips e-skins to detect various physical stimuli, including pressure, strain, slip, force vectors, and temperature. This feature is vital for applications in health monitoring and robotics, allowing e-skins to effectively mimic the sensory functions of human skin. In addition to tactile sensing, there is exciting progress in integrating chemical and electrophysiological sensing functionalities. By combining these capabilities, skin-attachable devices can monitor chemical changes and electrophysiological signals, crucial for evaluating users' health states. Such innovations pave the way for continuous health monitoring, providing invaluable data to both users and healthcare providers. Material advancements are playing an essential role in the evolution of e-skin technology.³¹ Researchers are concentrating on developing stretchable and self-healing materials that can endure prolonged stress and adapt to irregularly shaped surfaces. Emphasizing biocompatibility ensures these materials do not adversely affect the human body during extended use, addressing safety concerns as e-skin devices closely interact with human skin.⁵⁹Fig. 12a shows examples of seal-healable materials that can be used for electronic skin.¹⁸⁹ Furthermore, the self-healing capabilities of e-skin devices significantly enhance their durability and usability. As these devices may encounter mechanical damage over time, the ability to autonomously repair themselves is incredibly beneficial. Ideally, this self-healing process should take place at room temperature without requiring external stimuli, a feature known as autonomous self-healing. This advancement ensures that e-skin devices maintain long-term robustness and reliability, making them increasingly suitable for continuous use in real-world applications. In summary, the progress in e-skin technology, particularly in tactile, chemical, and electrophysiological sensing capabilities, complemented by innovations in stretchable and self-healing materials holds tremendous promise. These developments not only enhance the functionality of e-skin devices but also significantly improve their ability to replicate the sensory functions of human skin, thereby broadening their applications in healthcare and other vital domains.

Fig. 12 (a) Examples of seal-healable and biocompatible materials that can be used for electronic skin.¹⁸⁹ Reproduced from ref. 189 with permission from MDPI, copyright 2022. (b) The design and architecture of a self-sensing and haptic-reproducing e-skin, and its conformally real-time use on the arm, also their corresponding results under bending and double-side folding.¹⁹⁰ Reproduced from ref. 190 with permission from Science, copyright 2022. (c) An overview of the reported and proposed methodology for developing an MLA-based remote monitoring system. The new components are in bold. Stages in white are on a remote server while stages in gray are on a mobile device.¹⁹¹ Reproduced from ref. 191 with permission from Springer Nature, copyright 2018. (d) Prototype design of universal-sensor interface with self-X properties (USIX) chips.¹⁹² Reproduced from ref. 192 with permission from MDPI, copyright 2023. (e) A schematic overview of the emotion recognition process using the PSiFI system, from material fabrication to final classification.¹⁹³ Reproduced from ref. 193 with permission from Springer Nature, copyright 2024.

Further, advancements have significantly enhanced the integration of multifunctional capabilities in neuromorphic skin devices, particularly for healthcare and biomedical applications. As virtual and augmented reality (VR/AR) technologies rapidly evolve, conventional reliance on visual and auditory information is insufficient for creating truly immersive experiences. This drives the need for innovations in wireless touch perception, potentially revolutionizing how we interact with digital and physical environments. Addressing this challenge, Li et al. introduced an advanced approach for developing intelligent self-sensing and haptic-reproducing electronic skin (e-skin) designed for constructing a wireless touch-based Internet of Things (IoT).¹⁹⁰ Their work focuses on novel materials, device architecture, integration techniques, and communication protocols to achieve this goal. Fig. 12b illustrates the overall design and architecture of the proposed e-skin, demonstrating its real-time, conformal application on the arm, and showcasing performance under various deformation conditions, such as bending and double-sided folding.¹⁹⁰ E-skin employs innovative materials and structures to create flexible electromagnetic devices for tactile sensing and haptic feedback, acting as self-sensing actuators. Tactile sensing occurs through an electrical current generated by positional changes between a magnet and a coil during mechanical deformation. An alternating current applied to the coil causes the magnet to oscillate due to the Lorentz force, stimulating the skin. This system features multiple actuators organized in a scalable matrix design, enabling wireless touch sensing and feedback. It supports bidirectional, networked touch transmission for interactive communication among users. As a prototype for a touch-enabled Internet of Things (IoT), this technology could lead to significant advancements in touch communication and immersive digital experiences.¹⁶⁵

In progress, neuromorphic e-skin devices have made significant strides in biomedical applications over the past two years. These advancements leverage the unique capabilities of e-skin to mimic physiological sensing, enabling a variety of applications in health monitoring, emotion recognition, elderly care, and prosthetics. Neuromorphic e-skin technology has made significant advancements in healthcare, enabling personalized health monitoring across various applications. Flexible, wearable sensors can track multiple vital signs, such as heart rate, blood pressure, body temperature, and blood oxygen levels, simultaneously. These sensors integrate with mobile health platforms, allowing real-time data transmission to smartphones and cloud systems for advanced analytics. Machine learning algorithms applied to these data can facilitate early detection of health issues and provide personalized insights for proactive health management. Omar Boursalie et al. have overcome the challenges associated with the machine learning-based patient monitoring system on a mobile device and developed a new machine learning mobile platform that incorporates metrics beyond traditional classifier performance, as shown in Fig. 12c.¹⁹¹

Recent advancements in neuromorphic sensors have significantly improved the detection of early signs of circulatory conditions, such as peripheral artery disease (PAD). These sensors are designed to emulate the function of biological neurons, allowing them to identify subtle changes in physiological signals. This biomimetic approach enables the sensors to detect minute variations in blood flow patterns that may indicate the early onset of PAD. Additionally, these sensors can monitor multiple physiological parameters, including temperature, pressure, and the electrical conductivity of the skin, providing a comprehensive assessment of vascular health. However, contemporary devices still face challenges in managing time-domain data and implementing digital design techniques. Traditional technologies often fall short when compared to biological systems, which demonstrate highly efficient mechanisms capable of outperforming conventional approaches. To address these challenges, Hamam Abd and the team developed a neuromorphic spiking sensory system known as the universal-sensor interface with a self-X properties (USIX) Chip.¹⁹² This prototype, depicted in Fig. 12d, incorporates essential elements of an adaptive neuromorphic spiking sensory network, such as neurons, synapses, adaptive coincidence detection (ACD), and self-adaptive spike-to-rank coding (SA-SRC). The system was fabricated using XFAB CMOS 0.35 μm technology through the EUROPRACTICE program. The primary focus of their research was to evaluate the performance of the SA-SRC on-chip, specifically assessing the effectiveness of its adaptation scheme and its ability to generate spike sequences that reflect the temporal differences between incoming spikes. The SA-SRC is crucial to the adaptive neuromorphic spiking sensory system, as it is responsible for translating these temporal differences into meaningful spike orders. Measurement results from the fabricated chip confirmed the simulation outcomes of previous studies, demonstrating the system's capability to function effectively and adaptively. These findings suggest that the adaptive neuromorphic spiking sensory system holds promise for enhancing the efficiency and accuracy of neuromorphic sensors in detecting early vascular health issues, potentially paving the way for more advanced and biologically inspired diagnostic tools.¹⁹²

Neuromorphic e-skin technology has made significant strides in emotion recognition, offering promising implications for mental health monitoring. By integrating multimodal sensors, this e-skin can detect various physiological signals, such as skin conductance, temperature changes, and electrical activity, all of which correlate with different emotional states. Leveraging deep learning models trained in these physiological inputs, the system accurately identifies emotions like happiness, sadness, anger, and fear. This capability is particularly valuable for developing emotion-aware healthcare solutions that could facilitate early intervention strategies and enhance mental health outcomes. Recent research highlights the remarkable potential of wearable technologies for real-time health monitoring, focusing on the significance of ubiquitous sensing in the continuous collection of physiological and behavioral data within our daily lives. A notable innovation in this area is the creation of an emotion recognition framework designed for mental health monitoring, which utilizes galvanic skin response (GSR) signals to facilitate ongoing and objective assessment of emotional states. GSR serves as a well-established physiological marker that reflects fluctuations in skin conductance linked to autonomic nervous system activity, making it a dependable indicator of emotional arousal. By combining GSR sensing with cutting-edge wearable technology, this framework offers a more precise and non-invasive means of assessing emotions in real-time compared to traditional methods. One groundbreaking achievement in this field comes from Professor Jiyun Kim and his team,¹⁹³ who developed the world's first real-time wearable human emotion recognition system. This innovative system synthesizes multimodal data, merging verbal and non-verbal cues, which enhances the accuracy and depth of emotional analysis. A cornerstone of this technology is the personalized skin-integrated facial interface (PSiFI), a flexible, self-powered, and transparent wearable platform. With a bidirectional triboelectric strain and vibration sensor, the system dynamically detects micro-expressions and physiological changes, while wireless data transfer enables real-time processing and classification. A schematic overview of the emotion recognition process using the PSiFI system, from material fabrication to final classification, is presented in Fig. 12e. By employing advanced sensing mechanisms alongside machine learning algorithms, this framework introduces a novel approach to continuous mental health assessment. The capacity to accurately monitor emotional states with minimal intrusiveness holds great promise for early detection, intervention, and tailored treatment strategies in mental health care. In summary, this research represents a significant advancement in wearable sensing technology, offering a scalable and effective solution for monitoring emotional well-being in everyday life. The integration of GSR-based sensing with real-time data processing opens new avenues for future developments in digital health, enhancing our understanding and management of mental health conditions.¹⁹³

Meanwhile, the design and development of neuromorphic electronic skin (e-skin) devices require precise attention to biocompatibility, sensor accuracy, and long-term stability to ensure their effectiveness, safety, and applicability in healthcare and biomedical fields.¹⁹⁴ Biocompatibility ensures seamless interaction between e-skin devices and human tissue, preventing adverse effects such as irritation, immune responses, or cytotoxicity. This becomes particularly critical for devices intended for prolonged use or integration with neural tissues.¹⁹⁵ Recent studies have focused on the development of bio-inspired materials, such as polydopamine – a polymer with adhesive properties and structural similarity to melanin. Polydopamine has demonstrated excellent compatibility for skin contact applications.^16,175,196 Similarly, two-dimensional (2D) materials and their composites have shown low cytotoxicity and exceptional biocompatibility, making them suitable for advanced biomedical applications.^197–199 Natural polymers like silk fibroin and chitosan have also gained attention for their inherent biocompatibility, biodegradability, and suitability for e-skin systems.²⁰⁰ These materials serve as ideal substrates for integrating synthetic elements, enhancing the overall performance of neuromorphic e-skin systems.

Despite these advancements, challenges remain in scaling up the production of biocompatible materials while maintaining uniform quality and performance. Dynamic conditions such as sweating, repetitive motion, and extended skin contact also pose significant hurdles to maintaining biocompatibility. To address these issues, researchers are exploring hybrid materials that combine natural polymers' biocompatibility with synthetic components' enhanced functionality. Additionally, surface modification techniques, such as plasma-assisted treatments, have improved skin adherence, wettability, and compatibility. For example, Chen et al. successfully demonstrated polydopamine-modified super-elastic fibers with superior conductivity, achieving excellent biocompatibility and durability under dynamic conditions.²⁰¹

At the same time, sensor accuracy is another critical factor in developing wearable and implantable devices, as it directly impacts data reliability across applications. Advances in sensor technologies have significantly enhanced the capabilities of these devices, particularly in electrophysiological monitoring and chemical analysis. Flexible and wearable sensors have been developed to monitor vital signs such as EEG, ECG, and EMG, employing innovative materials like silicone for improved noise immunity and wearer comfort.²⁰² In chemical sensing, ion-sensitive field-effect transistors (ISFETs) have emerged as versatile tools, capable of converting changes in chemical solution composition into electrical signals. Recent advancements, such as integrating machine learning algorithms into ISFET systems, have addressed challenges like sensor variability and improved performance.²⁰³ Despite these developments, persistent issues such as noise interference, environmental calibration consistency, and power limitations continue to affect real-time accuracy. To overcome these limitations, neuromorphic computing, particularly spiking neural networks (SNNs), has shown promise in noise filtering and pattern recognition. A notable example is a resistive neuromorphic tactile system comprising a carbon nanotube-based microstructured pressure sensor, an ionic wire, and an electrolyte-gated tungsten oxide transistor.²⁰⁴ This system demonstrated superior performance in reducing errors and enhancing data interpretation, showcasing the potential of neuromorphic approaches in sensor systems. The future of sensor technology lies in the convergence of these advanced sensing modalities with neuromorphic computing and energy harvesting techniques. Biomolecular sensors are expanding beyond glucose detection to a broader range of physiological markers, promising more comprehensive health monitoring capabilities.¹¹¹ Additionally, the integration of neuromorphic technologies in edge AI and IoT applications is expected to enable more efficient and intelligent processing of sensor data at the edge, reducing reliance on cloud connectivity and improving privacy and security.^205,206

Long-term stability is crucial for reliable sensor performance in continuous health monitoring.²⁰⁷ Innovations such as self-healing materials and stretchable conductors have significantly improved device durability. For instance, silicone-based composites with autonomous self-healing properties restore functionality after damage, while stretchable conductors, like gold-coated titanium dioxide nanowires embedded in silicone, provide robust mechanical flexibility.^208,209 Hydrogels have also emerged as effective interfaces for maintaining flexibility and robustness in neural interfaces.^126,210 However, challenges like material degradation from mechanical stress, UV exposure, and sweat persist. To address these, researchers are developing autonomous self-healing systems that operate at room temperature and advanced encapsulation techniques to protect components from environmental factors. Hybrid e-skin systems using self-healing hydrogels and ion-gels have demonstrated sustained performance and stability under dynamic conditions, as reported in several recent studies.^5,211 These integrated strategies collectively drive the advancement of durable, accurate, and biocompatible e-skin technologies for next-generation healthcare and biomedical applications. The future of sensor technology lies in integrating these advanced sensing modalities with neuromorphic computing and energy harvesting. Such advancements are paving the way for the development of robust, biocompatible e-skin technologies capable of withstanding next-generation healthcare and biomedical applications.

4.4. Prosthetic applications

Neuromorphic e-skin devices represent a transformative approach in the development of advanced prosthetics, aiming to replicate the sensory and computational functions of human skin and the nervous system. These devices integrate flexible sensors and artificial synapses to mimic the mechanoreceptive, proprioceptive, and nociceptive capabilities of natural skin while leveraging neuromorphic computing for local processing and adaptive response. By incorporating features such as tactile sensing, temperature detection, and multi-modal integration, neuromorphic e-skin not only enhances the realism of prosthetic limbs but also addresses key challenges in real-time data processing and energy efficiency. This section reviews the evolution of neuromorphic e-skin technologies, highlighting recent advancements in materials, device architectures, and computational strategies that bring prosthetic systems closer to achieving lifelike sensory feedback and dynamic adaptability.

The development of graphene-based transparent neural microelectrode arrays by Park et al. represents a significant breakthrough in the field of neural interfaces, offering new possibilities for enhancing prosthetic technology. These arrays incorporate graphene-based carbon-layered electrode array (CLEAR) technology, which provides a unique combination of high electrical conductivity, optical transparency, and biocompatibility, making it an ideal candidate for advanced neural interfacing. The CLEAR device has been shown to enable simultaneous neural imaging and optogenetic stimulation, offering a multifaceted approach to studying and manipulating neural activity in vivo. The study by Park et al. further demonstrated the utility of the CLEAR device for in vivo imaging of the cortical vasculature. This was achieved through fluorescence microscopy, allowing researchers to visualize blood flow and vascular structures, and 3D OCT, which provided depth-resolved imaging of the brain tissue as shown in Fig. 13a.²¹² This type of device structure allows for simultaneous imaging, high-resolution neural recording, and optogenetic stimulation, which offers a versatile platform for advancing the capabilities of prosthetic devices. This multifunctional approach to neural interfacing has the potential to improve the functionality of prosthetics, offering users a more natural and lifelike experience.

Fig. 13 (a) Schematic demonstration of the utility of the CLEAR device for in vivo imaging of the cortical vasculature.²¹² Reproduced from ref. 212 with permission from Springer Nature, copyright 2014. (b) The long-term durability and performance of neural interfaces used in neural recording prosthetic applications and (c) their 3-month durability performance.²¹³ Reproduced from ref. 213 with permission from Wiley, copyright 2018. (d) The closed-loop prosthetic system for tactile sensing with a neuromorphic signal through the prosthesis controller.¹⁵⁶ Reproduced from ref. 156 with permission from Science, copyright 2018. (e) A real-time example of the sensorimotor loop with a monolithic soft e-skin.⁸ Reproduced from ref. 8 with permission from Science, copyright 2023. (f) Multifunctional conductive hydrogel interface for bioelectronic recording and stimulation. Conductive hydrogel interfaces establish a bridge between soft biological tissues (skin, heart, nerve, brain, etc.) and hard electronics to facilitate high-quality bidirectional bioelectronic stimulation and recording for implantable bioelectronics.¹²⁶ Reproduced from ref. 126 with permission from Wiley, copyright 2024.

Further, electrical interfacing with neural tissue is key to advancing diagnosis and therapies for neurological disorders, as well as providing detailed information about neural signals. A challenge for creating long-term stable interfaces between electronics and neural tissue is the huge mechanical mismatch between the systems. So far, materials and fabrication processes have restricted the development of soft electrode grids able to combine high performance, long-term stability, and high electrode density, aspects all essential for neural interfacing. To this end, Tybrandt et al. developed stretchable electrode grids using a composite material of gold-coated titanium dioxide nanowires in a silicone matrix, enabling long-term neural recording. This innovative approach facilitated long-term neural recording, thereby enhancing the durability and performance of neural interfaces used in prosthetic applications (Fig. 13b). More importantly, the developed grid can resolve high spatiotemporal neural signals from the surface of the cortex in freely moving rats with stable neural recording quality and preserved electrode signal coherence during 3 months of implantation as shown in Fig. 13c. The proposed material and device technology presented herein has the potential for a wide range of emerging biomedical applications.²¹³

The development of neuromorphic sensory feedback systems represents a significant breakthrough in advancing prosthetic technology. These systems are designed to replicate the biological behavior of tactile receptors, providing meaningful tactile feedback to prosthesis users, a feature often missing in conventional prosthetic designs. While modern advancements in prosthetic limbs and control mechanisms have enhanced the ability of amputees to regain lost function, they typically lack realistic tactile perception, limiting the user's ability to interact with their environment naturally. Osborn and colleagues made a key contribution in this area by utilizing transcutaneous electrical nerve stimulation (TENS) to provide sensory feedback to an amputee. They carefully calibrated the TENS parameters to elicit two types of tactile perceptions in the phantom hand: innocuous (non-painful) and noxious (painful) sensations. By implementing a closed-loop feedback system, they enabled a trans-humeral amputee to experience tactile sensations corresponding to specific areas of activation on the prosthetic limb. The closed-loop system could evoke either innocuous or painful sensations in response to different stimuli, enhancing the user's sensory experience with the prosthesis (Fig. 13d). To further advance the neuromorphic approach, the team developed a multilayered electronic dermis (e-dermis) inspired by the functional characteristics of human mechanoreceptors and nociceptors. The multilayered structure of the e-dermis enabled it to simulate the layered organization of human skin, where different layers contribute to distinct aspects of tactile perception. In an innovative pain detection task (PDT), the system demonstrated the capability to distinguish between safe (innocuous) and harmful (noxious) tactile stimuli during grasping. The prosthesis was programmed to react to noxious stimuli by initiating a protective reflex, mimicking a polysynaptic withdrawal reflex to avoid potential damage. The team also explored the differentiation of textures through neuromorphic sensory feedback. By employing stimulation paradigms that simulated slowly adapting (SA) receptor-like dynamics, the system allowed amputees to distinguish between different textures via nerve stimulation. By integrating neuromorphic techniques with sensory feedback mechanisms, the approach significantly improved the grip functionality and tactile realism of prostheses. The ability to convey both innocuous and painful sensations, along with the ability to differentiate textures, provided users with a more comprehensive sensory experience. This advancement represents a pivotal step towards developing prosthetic devices that offer not only functional movement but also meaningful sensory feedback, ultimately improving the quality of life for amputees.¹⁵⁶

Recent innovations in neuromorphic e-skin technologies focus on achieving low power consumption while maintaining efficient operation. This is crucial for prosthetic applications, where energy efficiency directly impacts the usability and longevity of the device. For instance, Wang et al. developed a soft e-skin using organic semiconductor transistors that replicate the mechanical properties and sensory feedback of natural skin. This system utilizes a trilayer, high-permittivity elastomeric dielectric, which allows for a low subthreshold swing comparable to polycrystalline silicon transistors, resulting in low operation voltage and power consumption.⁴⁶ The ability to generate neuromorphic pulse-train signals is a significant feature of modern e-skins. These signals mimic natural neural activity, enabling prosthetic devices to provide more realistic sensory feedback. The monolithic soft e-skin developed by Wang et al. can sense temperature and pressure and convert these stimuli into electrical pulses that induce neuronal firing, similar to natural skin. Fig. 13e illustrates the real-time demonstration of the sensorimotor loop with a monolithic soft e-skin.⁸ This biomimetic functionality allows for multimodal perception, essential for creating responsive interfaces in prosthetics. The integration of neuromorphic hardware into e-skin systems allows for real-time processing of tactile information. Such hardware can mimic the intricate neural encoding mechanisms of the somatosensory system, enabling artificial limbs to provide natural and informative feedback.^36,46 This real-time capability is crucial for seamless human–machine interaction, as it ensures that users can perceive and react to environmental stimuli without noticeable delay.

The latest advancements in biomimetic systems for prosthetics have focused on creating technologies that seamlessly integrate with the human body. These innovations aim to enhance the functionality and user experience of prosthetic devices by mimicking natural biological systems. Researchers are developing soft and elastic hydrogel-based microelectronics designed for localized low-voltage neuromodulation. These materials are particularly promising because they can conform to the body's natural movements and provide a comfortable interface between electronic devices and biological tissues. The use of hydrogels allows these systems to be both flexible and biocompatible, making them ideal for long-term integration with the human body. Fig. 13f comprehensively highlights the applications of bio-integrated hydrogel interfaces in bioelectronic recording and stimulation, especially in the context of implantable and integrated bioelectronic systems.¹²⁶ Another focus area is the creation of stretchable, self-healable, and breathable biomimetic iontronics. These materials are engineered to mimic the properties of natural tissues, allowing them to stretch and heal themselves after damage. This capability is crucial for maintaining the functionality of prosthetic devices over time, especially in dynamic environments where wear and tear are common. Efforts are also being made to develop wireless intelligent systems, such as three-lead electrocardiograph monitors with feedback functions. These systems leverage advanced sensor technologies and wireless communication to provide real-time health monitoring and feedback. Such capabilities are essential for integrating prosthetic devices with broader health management systems, allowing users to receive timely alerts and adjustments based on their physiological data.²¹⁴ Neuromorphic hardware plays a critical role in these advancements by providing real-time processing of sensory information. This technology mimics the neural encoding mechanisms of the somatosensory system, enabling prosthetic limbs to provide users with natural and informative feedback. By using analog/digital mixed-signal CMOS-based hardware, these systems can perform complex computations efficiently, supporting in-memory computing and neural computational primitives.⁴⁶ These technological advancements represent significant progress in creating biomimetic prosthetics that closely emulate natural limb function. By integrating advanced materials and neuromorphic hardware, researchers aim to develop prosthetic devices that offer enhanced sensory feedback, improved comfort, and seamless interaction with the human body. Such innovations hold great promise for improving the quality of life for individuals with limb loss or impairment by providing more naturalistic movement and interaction capabilities.

For the time being, the advancement of neuromorphic e-skin for prosthetics necessitates the seamless integration of essential technical parameters: biocompatibility, sensor accuracy, and long-term stability. Striking a balance among these factors is crucial for developing systems that can effectively interact with human physiology while providing superior performance. Biocompatibility ensures that prosthetic systems can interface with biological tissues without eliciting adverse reactions such as inflammation, immune rejection, or cytotoxicity.^215,216 Materials like natural polymers are widely used for their excellent bio-integration properties.^217,218 Furthermore, innovations such as graphene-based microelectrode arrays and hydrogel interfaces enable long-term adaptability and skin-like compliance. However, scaling up the production of biocompatible materials while maintaining uniform quality and performance remains a significant challenge.^219,220 Another critical issue is the mechanical mismatch between soft biological tissues and the typically rigid electronic components used in prosthetic systems. This mismatch can lead to discomfort and degradation of the interface. To address this, researchers have developed flexible, stretchable, and bioresorbable materials that closely mimic the mechanical properties of human skin. Hydrogels have gained attention as interface materials for bioelectronics due to their high water content, low modulus, and good biocompatibility. These properties make hydrogels ideal for bridging the mechanical and chemical differences between rigid electronic devices and soft biological tissues, potentially improving the stability and reliability of implantable electronic interfaces.²¹⁰ Despite significant progress, achieving reliable, long-term biocompatibility under dynamic conditions such as sweating, movement, and prolonged wear remains a key area for improvement.

The accuracy of sensors plays a crucial role in replicating the tactile and sensory feedback associated with natural skin. Neuromorphic e-skin systems effectively integrate tactile, temperature, and electrophysiological sensors to emulate the functions of mechanoreception, proprioception, and nociception.⁴⁰ These sensors are often paired with SNNs and neuromorphic hardware, which allows for real-time processing, improved noise filtering, and energy-efficient operations.^221,222 Recent innovations, such as organic semiconductor transistors⁸⁸ and carbon-based CLEAR technology,¹⁰⁸ provide promising low-power solutions for delivering real-time sensory feedback. However, achieving consistent sensor accuracy amidst environmental variations including humidity, temperature changes, and extended usage remains a challenge of ongoing importance. Researchers continue to work on adaptive calibration techniques and robust sensing architecture aimed at sustaining accuracy over time.

Long-term stability is vital for the reliable operation of prosthetic devices over extended periods. Researchers have facilitated the creation of highly durable components, such as gold-coated titanium dioxide nanowires embedded in silicone matrices and self-healing hydrogels, which demonstrate remarkable mechanical flexibility and resilience with a biocompatible, stretchable, and ultra-durable 3-month implantation period in freely moving rats.^213,223,224 Furthermore, the development of stretchable electrode grids allows for stable, high-resolution neural recordings, promoting precise and sustained interactions with biological systems. Nevertheless, challenges persist due to degradation resulting from mechanical stress, UV exposure, and chemical interactions (e.g., sweat). To combat these issues, researchers are implementing advanced encapsulation techniques to protect sensitive components from environmental influences.²²⁵ Additionally, self-healing mechanisms, including room-temperature autonomous self-repair, are being incorporated to restore material functionality following damage. Despite these significant advancements, maintaining the electrical and mechanical properties of materials after repeated use in harsh conditions remains an important area for future exploration and research. To address these challenges, future research should focus on hybrid materials, integration of computational techniques, and manufacturing processes. The field can overcome current limitations to deliver prosthetic e-skin systems that provide lifelike functionality, long-term stability, and widespread accessibility. These advancements hold great promise for improving the quality of life for individuals with limb loss or impairment.

4.5. Robotic applications

Touch is a complex and multifaceted sensory modality, facilitated by a vast network of mechanoreceptors, thermoreceptors, and nociceptors distributed across human skin. These receptors gather and encode tactile data, enabling humans to perceive and interact with the environment with remarkable efficiency.⁵ In contrast, the tactile sensing capabilities of modern robots remain limited, highlighting the need for neural-like hardware in e-skin.¹⁶⁵ This can be achieved through innovative distributed electronics or by leveraging neuromorphic circuits initially developed for other sensory modalities. This section reviews recent advancements in neuromorphic e-skin, focusing on material development, neural-like sensing, and data processing architecture in robotics applications.²²⁶

The field of electronic skin (e-skin) applications has made significant strides, particularly in developing materials with intrinsic stretchability and self-healing properties. These characteristics are vital for ensuring that the e-skin can endure prolonged mechanical stresses and conform to irregular surfaces, making it ideal for robotic applications. A remarkable advancement has been made by Jin Young Oh and his team,²²⁷ who successfully engineered a strain-sensitive, stretchable, and autonomously self-healing semiconductor film. This innovative material boasts a fracture strain exceeding 1300%, enabling it to withstand substantial deformation. Furthermore, it has the unique ability to recover its mechanical and electrical properties at room temperature without requiring external intervention. With an outstanding gauge factor of 5.75 × 10⁵ at 100% strain, it demonstrates excellent precision in deformation sensing. The team achieved these pioneering properties by combining a polymer semiconductor with a self-healing elastomer, employing dynamic cross-linking through metal coordination bonds. Fig. 14a shows a schematic illustration of Fe(III)-PDCA ligand bonding of PDMS-PDCA and DPP-TVT-PDCA in the blend film.²²⁷ This inventive fusion resulted in a material that elegantly integrates stretchability, self-healing, and sensitivity, key elements for the future of advanced e-skin systems. This groundbreaking work signifies a major step forward, as it harmonizes multiple desirable traits within a single material. The strain-sensitive, stretchable, and self-healing capabilities of this semiconductor film could greatly transform the design and functionality of e-skin, opening new avenues for applications in robotics and various other fields.²²⁷

Fig. 14 (a) Design and proposed mechanism of a strain-sensitive, stretchable, and self-healable thin film composed of diketopyrrolopyrrole (DPP) and polydimethylsiloxane (PDMS) dynamically cross-linked through Fe(III)-pyridine-2,6-dicarboxamide (PDCA) complexation, highlighting its mechanical and functional properties.²²⁷ Reproduced from ref. 227 with permission from Science, copyright 2019. (b) Comparison between the biological somatosensory system (i) and the artificial somatosensory system (ii), showcasing the integration of sensory receptors, neural pathways, and feedback mechanisms for tactile perception and response.¹¹⁵ Reproduced from ref. 115 with permission from Springer Nature, copyright 2022. (c) Neural network architecture for computational e-skin, consisting of sensory neurons and cuneate neurons interconnected via synapses to mimic the spatial sensitivity and functionality of biological tactile systems.¹⁶⁵ Reproduced from ref. 165 with permission from Science, copyright 2022. (d) Examples of spike-timing-dependent plasticity (STDP), a plasticity-based learning rule enabling unsupervised learning in spiking neural networks, demonstrating weight adjustments based on temporal correlations of pre- and post-synaptic neuron spikes.²²⁸ Reproduced from ref. 228 with permission from Frontiers, copyright 2010. (e) and (f) Prototype examples of flexible printed circuit board (PCB)-based e-skin and integrated neuromorphic chips, illustrating scalable fabrication techniques and dense neural network structures for computational e-skin applications.^229,230 Reproduced from ref. 229 with permission from Wiley, copyright 2017. Reproduced from ref. 230 with permission from Science, copyright 2014.

Building on the previous works, researchers made significant strides in improving the functionality and integration of e-skin systems. Fuqin Sun and colleagues have made significant advancements in flexible and neuromorphic electronics, paving the way for artificial perception systems that emulate biological functions.¹¹⁵ These systems hold great promise for intelligent robotics and human–machine interactions. However, despite notable progress, developing artificial systems that can mimic somatosensory feedback functions has remained a challenge. To address this gap, the researchers created an artificial somatosensory system that integrates tactile perception with instant feedback capabilities, drawing inspiration from human somatosensory feedback pathways. They achieved spatio-temporal information processing by integrating multiple sensors with multi-gate synaptic transistors, thereby mimicking the ability of biological systems to encode and interpret tactile data over time and space. The instant feedback mechanism was implemented using a comparator to analyze the processed signals and activate the artificial muscle when a critical stimulus intensity threshold was reached. The system directly draws inspiration from biological somatosensory feedback pathways, as illustrated in Fig. 14b, which compares the artificial system in Fig. 14b(i) with its biological counterpart in Fig. 14b(ii).¹¹⁵ In this artificial system, tactile stimuli generate signals similar to action potentials, which are transmitted to the artificial muscle to prevent further damage. This replicates the rapid response mechanism observed in living organisms. This research represents a paradigm shift in bionic tactile perception systems, demonstrating how artificial systems integrate sensory perception and feedback functions. By combining sensory, computational, and actuation components, this artificial somatosensory system provides a robust platform for developing next-generation intelligent robotics and advanced human–machine interfaces.

As research continues, the implementation of neuromorphic e-skin technology is expected to advance significantly. Future innovations may focus on improving the functionality and durability of e-skin devices, as well as integrating advanced sensors that can detect a wider range of stimuli. Recent developments in e-skin technologies have concentrated on enabling their application on complex three-dimensional surfaces, which is essential for enhancing robotic capabilities.^165,231 These advancements aim to improve the flexibility, stretchability, and conformability of e-skin systems while ensuring robust sensing abilities. The integration of novel fabrication techniques, advanced materials, and biomimetic designs has led to the creation of more versatile and high-performing robotic systems. Notably, spray coating technologies have emerged as a leading method for creating highly sensitive and durable e-skin for three-dimensional surfaces.^36,165 These improvements in applying e-skin to complex 3D surfaces represent a significant step towards more capable and versatile robotic systems. By enhancing the conformability, stretchability, and sensing capabilities of e-skin, researchers are paving the way for robots with improved tactile perception and interaction abilities.^165,231

Further advancements in electronic skin (e-skin) technology have focused on the development of computational architectures inspired by neural system structures and their corresponding training algorithms. A proposed model incorporates a neural network comprising one layer of sensory neurons and one layer of cuneate neurons connected via synapses (Fig. 14c)¹⁶⁵ This design draws inspiration from the human tactile system, where mechanoreceptors possess varying receptive field sizes that overlap and interconnect, forming the foundation for the spatial sensitivity of natural skin. Similarly, the proposed neural network leverages overlapping receptive fields to mimic the spatial sensitivity of biological tactile systems. A key consideration in implementing this neural network is the realization of its hardware components. While multiple learning strategies, including supervised, unsupervised, and reinforcement learning, are available, a fully hardware-based supervised learning system is often impractical due to the high cost and complexity of the required devices or circuits. Consequently, such systems are not typically considered for e-skin applications.

Instead, spike-timing-dependent plasticity (STDP), a fundamental learning rule in spiking neural networks (SNNs), provides a promising alternative for unsupervised learning (Fig. 14d).²²⁸ STDP adjusts synaptic weights based solely on the temporal correlation between spikes generated by pre- and post-synaptic neurons. Various pairing schemes can be employed, including “nearest neighbor takes all,” “nearest neighbor takes more,” or “all spike pairs count equally” approaches, each influencing the synaptic weight dynamics differently. This plasticity-based learning rule allows e-skin systems to autonomously adapt and refine their tactile sensing capabilities over time without requiring external supervision. Such an approach aligns well with the principles of neuromorphic computing, offering an efficient and biologically inspired framework for implementing computational e-skin systems. By integrating these neural-inspired architectures and learning rules, researchers aim to enhance the sensory and processing capabilities of e-skin, paving the way for more intelligent and adaptive robotic systems.^232,233

To replicate the remarkable mechanical properties of biological skin, electronic skin (e-skin) must be crafted on soft and flexible substrates. Recent advancements highlight the exciting integration of off-the-shelf sensors and electronic components onto flexible printed circuit boards (PCBs), paving the way for applications that span from localized hand-based manipulation to full-body coverage. This innovative approach extends to computational e-skin by connecting these sensors with neuromorphic chips (Fig. 14e and f)^229,230 forming a rich network of artificial neurons capable of performing a wide array of computational tasks essential for robotic systems.²³⁴ Yet, traditional chip bonding methods present significant challenges. Conventional techniques risk introducing cracks into the chips, jeopardizing their structural integrity and performance. The emergence of innovative bonding techniques, like printed bonding, offers a promising solution, ensuring a reliable and damage-free method for attaching chips to flexible substrates.^235–237 Moreover, ensuring the reliability of electrical interconnection within the flexible e-skin poses another critical challenge. This can be overcome using mechanically flexible conductive materials, such as liquid metals, which maintain electrical conductivity while gracefully accommodating the inherent mechanical deformations of soft substrates. Despite these obstacles, the combination of ultra-thin chips (UTCs) with flexible PCBs heralds an inspiring pathway toward the realization of computational e-skin. This approach effortlessly blends efficiency with scalability, fostering the development of e-skin systems with enhanced computational capabilities and bridging the gap between the wonders of biological and artificial skin functionalities in the realm of robotics.

5. Challenges and future direction

5.1. Challenges

Neuromorphic e-skin presents exciting opportunities across various applications, such as smart textiles, healthcare, biomedical devices, prosthetics, robotics, and advanced sensing platforms. While there has been considerable progress, the path to widespread adoption can be enhanced by addressing certain technical challenges, particularly in materials science.^55,238 Recent advancements in stretchable and self-healing materials have shown promise, but ensuring long-term durability remains a key focus area. The effects of mechanical stress like bending, stretching, and compression along with exposure to harsh environmental conditions can impact the electrical properties of these materials.^88,155,238 A significant challenge arises from the contrast between the soft, flexible nature of neuromorphic e-skin and the rigid components found in traditional electronic systems. This difference can lead to instability at the interfaces, potentially resulting in delamination, interfacial degradation, and reduced reliability over time.^155,166,239 Moreover, the integration of flexibility, durability, and advanced neuromorphic functionalities such as synaptic emulation, real-time processing, and low power consumption within a single system is an exciting area for further research. Understanding the root causes of these material challenges requires a multidisciplinary approach that brings together materials science, electrical engineering, and neuroscience.^40,155 Key strategies moving forward could include the development of innovative materials like flexible ionic conductors, self-healing polymers, stretchable piezoelectric and triboelectric materials, and hybrid organic–inorganic nanocomposites. Additionally, innovative device architectures designed to endure both mechanical and environmental stress while maintaining functionality will be crucial. By achieving a harmonious balance between mechanical adaptability and neuromorphic performance, we can unlock the full potential of durable, high-performance e-skin systems that mimic the intricate sensory and cognitive functions of human skin.

Neuromorphic systems often rely on external batteries, which limits their autonomy and scalability. Achieving high energy conversion efficiency in a compact form factor remains a significant challenge in the development of neuromorphic skin.¹¹² These e-skin systems must withstand continuous mechanical deformations, such as stretching, bending, and twisting, which can lead to material fatigue, cracks, and delamination. Such mechanical stresses can greatly reduce both the performance and lifespan of the e-skin.^127,240 For instance, in implanted e-skin, irregular deformations during use can cause cracks and delamination of battery electrodes, leading to decreased performance. Portable applications like prosthetics, robotics, and textiles may operate continuously, which further impacts their performance and longevity. Challenges in this area include ensuring electrical compatibility, minimizing interference, and maintaining overall system flexibility without sacrificing performance.^241–243 Collaborative, interdisciplinary efforts are essential for overcoming these barriers and unlocking the full potential of neuromorphic e-skin systems. The practicality and self-sufficiency of these devices could be greatly improved for real-world applications by developing low-power neuromorphic processors and integrating energy-harvesting technologies such as triboelectric and piezoelectric nanogenerators. The interplay between energy harvesting and neuromorphic e-skin technologies offers immense potential for creating autonomous, scalable systems for prosthetics, robotics, and wearable applications.

Neuromorphic e-skin systems face significant challenges in achieving real-time performance for complex tasks, particularly in terms of latency and processing capabilities. These challenges stem from the need to balance localized computation with the demands of millisecond-level responses required for applications like robotics. Recent advancements in neuromorphic computing and simulation techniques are addressing these issues, paving the way for more efficient and responsive e-skin systems.^244,245 The performance of neuromorphic e-skin in complex tasks is indeed limited by latency and real-time processing capabilities. This limitation is particularly evident in two scenarios, viz., localized computation and external processing. Direct neural processing within the e-skin offers faster response times due to reduced data transmission requirements. However, current setups often fall short of achieving the millisecond-level responses necessary for rapid action adjustments, such as those required in robotic motion control. On the other hand, when e-skin sensors transmit data to external processors, the transmission process introduces additional latency, further impeding real-time functionality.²⁴⁶ These issues should be mitigated by developing simulators with more efficient communication protocols and new neuromorphic hardware that can operate at timescales like those of real-time interactions. To further improve the real-time capabilities of neuromorphic e-skin systems, several areas of development are crucial.

The development of neuromorphic e-skin systems indeed faces several significant challenges, particularly in the areas of biological integration, mimicking natural skin perception, and scalable manufacturing. These challenges are critical to address for the successful implementation and widespread adoption of neuromorphic e-skin technologies. Stable and durable interfacing with the nervous system or neural interfaces presents two major hurdles, including immune rejection and tissue damage. The introduction of foreign materials into the body can trigger immune responses, potentially leading to device failure or tissue damage. Long-term implantation of neural interfaces can cause mechanical stress and inflammation, resulting in tissue damage and reduced device efficacy.²²⁶ Creating neuromorphic systems that accurately replicate the sophisticated perception of natural skin, including pain thresholds and proprioception, remains a significant challenge. This involves creating hardware and software systems that work seamlessly together to meet the user's nervous system needs. Furthermore, it is essential to ensure that the e-skin delivers consistent, repeatable, and natural-feeling sensations for the user.²⁴⁷ The wide adoption of neuromorphic e-skin is hindered by key challenges such as scalability and manufacturing costs. Creating flexible, multifunctional e-skins with embedded neuromorphic devices for large areas while maintaining high resolution is technically challenging. Current manufacturing processes for materials like stretchable electronics and soft neuromorphic transistors are limited in terms of large-scale production, cost, and consistency.²³⁹ By addressing these challenges, neuromorphic e-skin technologies can move closer to practical applications in prosthetics, robotics, and wearable health monitoring devices, potentially revolutionizing human–machine interfaces and personalized healthcare.

5.2. Future directions

The future development of neuromorphic e-skin technologies is poised for significant advancements, driven by innovative materials and fabrication techniques. These developments aim to overcome current limitations and enhance the performance of e-skin in real-time applications. One promising direction lies in the advancement of materials, with research efforts focusing on the development of novel nanomaterials and composites that exhibit superior mechanical and electrical properties. Materials with high flexibility, stretchability, and electrical conductivity, particularly those that integrate self-healing capabilities with enhanced sensory precision, hold immense potential to improve the durability and lifespan of neuromorphic e-skins.^121,248 Additionally, innovations in biomimetic materials that replicate the multilayered structure of human skin can significantly enhance the realism and accuracy of sensory feedback.^204,249 Hybrid materials that integrate organic and inorganic components present a compelling approach for achieving the dual objectives of flexibility and robustness.²⁵⁰ These materials combine the mechanical adaptability of organic polymers with the structural and functional stability of inorganic compounds, paving the way for highly efficient and durable e-skin systems. A noteworthy example in this domain is the research led by Tae Min Jang and his team, which has made significant strides in the development of eco-friendly electronic systems.¹⁹³ They introduced stretchable, biodegradable, and self-healing conductive materials featuring bilayer architecture. This architecture consists of (i) a synthetic self-healing elastomer and (ii) a self-healing conductive composite augmented with carefully selected additives. These materials exhibit remarkable self-repair capabilities, allowing e-skins to restore functionality after sustaining mechanical damage. This advancement not only extends the operational lifespan of e-skins but also enhances their reliability and suitability for various applications, including healthcare monitoring, robotics, and human–machine interfaces. Future research should aim to further optimize these materials by improving their mechanical resilience, electrical stability, and environmental sustainability. Additionally, integrating such advanced materials into scalable fabrication techniques will be critical for enabling the widespread adoption of neuromorphic e-skin technologies.⁴⁷ In a similar vein, Qin et al. have developed a double-network hybrid hydrogel that incorporates polyacrylamide, poly(acrylic acid), MXene, and PEDOT:PSS. This hydrogel demonstrates exceptional flexibility, self-healing properties, and mechanical stability. Notably, the addition of a protective MXene/PDMS encapsulation layer enables the PPMP hydrogel to serve effectively as a wearable strain sensor, capable of detecting nuanced movements such as facial expressions, blinking, and elbow joint motion. An important highlight of this research is the hydrogel's ability to maintain high electrical conductivity even under significant deformation, positioning it as an excellent candidate for durable and comfortable wearable e-skin applications.²⁵¹ These findings illustrate the promising potential of advanced hydrogels in paving the way for the development of next-generation soft, self-healing electronic devices.

Further, advancements in energy-efficient technologies and low-power neuromorphic hardware are crucial for the development of self-powered neuromorphic e-skins. The integration of energy harvesting systems, particularly triboelectric and piezoelectric nanogenerators, offers a sustainable solution by converting mechanical energy such as human motion into electrical energy. This minimizes reliance on external power sources and enhances the autonomy of these devices.^252–255 In the context of healthcare and biomedical applications, neuromorphic e-skins must embody essential characteristics, including lightweight construction, extended operational lifetime, wearability, flexibility, biocompatibility, and stable electrical performance. These features are vital for applications like continuous medical signal monitoring in patients, where comfort and reliability are paramount. Traditional batteries, however, have significant limitations. They often come with a finite service life, considerable weight, potential toxicity, and various environmental concerns. Piezoelectric and triboelectric nanogenerators have emerged as promising alternative energy sources for addressing these challenges. These nanogenerators can harvest renewable energy from diverse environmental stimuli including biomechanical movements, airflow, and mechanical vibrations and convert it into usable electrical power. The scientific advantages of piezoelectric and triboelectric nanogenerators include their simple energy transduction mechanisms, scalable and cost-effective manufacturing processes, and impressive energy conversion efficiency. Additionally, their lightweight and durable nature, combined with the ability to provide stable and high-output electrical performance, make them ideal candidates for powering wearable, flexible, and self-sustaining neuromorphic e-skins. Ultimately, these attributes demonstrate that nanogenerators not only serve as viable replacements for conventional batteries but also represent a transformative technology for advancing the field of sustainable and energy-autonomous biomedical devices. This shift could significantly improve the quality of life for patients while reducing the environmental impact of medical technology.²⁵⁶ Complementing these energy-harvesting strategies, the development of low-power neuromorphic hardware is essential for efficient data processing in e-skins. Memristors, known for their ability to emulate synaptic functions, offer a promising approach to in-memory computing, thereby reducing energy consumption and latency. A significant advancement is the multi-layered triboelectric nanogenerators (M-TENGs) developed by Park et al., which feature multiple friction layers and enhance output, achieving a maximum charge of 7.52 nC compared to 3.69 nC from single-layered TENGs.²⁵⁷ In this work, they integrated a M-TENG with neuromorphic hardware which showed promise in creating hybrid artificial synaptic devices that replicate learning behaviors, such as spike-timing-dependent plasticity (STDP) with low-power supply. These systems have low power consumption and high performance, making them suitable for real-time applications. Further innovations could involve exploring materials and architectures like memristors to enhance in-memory computing and reduce latency. Also, we should focus on optimizing M-TENG designs to improve energy conversion efficiency and durability. A proof-of-concept demonstrated a robotic hand mimicking long-term synaptic plasticity through continuous memory training. Future research should aim to expand the capabilities of e-skin devices through multi-modal sensing and adaptive learning, paving the way for applications in healthcare, robotics, and wearable technology.

Enhancing sensory integration and multimodal processing capabilities represents a pivotal avenue for advancing e-skin technologies. A critical focus for future research involves incorporating diverse sensor types, including pressure, temperature, and humidity sensors, into a unified e-skin platform. Such integration would enable more sophisticated multimodal sensing, expanding the range of applications in robotics, healthcare, and human–machine interfaces. To process the diverse and complex data streams generated by these sensors, the use of neuromorphic algorithms in conjunction with machine learning techniques holds significant promise. For example, a study conducted by Imke Krauhausen and colleagues demonstrated that bio-inspired multimodal learning using organic neuromorphic circuits allows robots to interact intelligently with their environment. The study highlights how biological systems adapt to their environment by learning through sensory signals, forming internal representations to guide decision-making. Specifically, they designed a robotic system capable of processing multiple types of sensory information to enhance object handling. Using organic neuromorphic circuits, the robot effectively learned to avoid hazardous objects, underscoring the potential of bio-inspired materials to elevate intelligent robotic systems.²⁵⁸

Furthermore, adaptive calibration of e-skin sensory systems to accommodate user-specific requirements or environmental conditions is essential for achieving optimal performance. For instance, Jan Klimaszewski and his team introduced an innovative robot-based calibration procedure for graphene-based electronic skin. This approach addresses challenges such as sensor sensitivity variability and load inconsistencies, particularly in configurations with large sensor arrays. The study demonstrates how an industrial robot equipped with a reference force sensor can automate the calibration process, significantly improving speed and repeatability. The methodology incorporates data preprocessing, sensor modeling, and performance evaluation of nonhomogeneous sensor matrices, providing a robust framework for maintaining consistent sensor performance.²⁵⁹ Future research should prioritize the development of advanced multimodal e-skin platforms that seamlessly integrate diverse sensor types with adaptive calibration and bio-inspired neuromorphic processing. Such innovations would enhance the functionality, reliability, and scalability of e-skin technologies, paving the way for transformative applications in next-generation robotics and beyond.

Advancements in neural interfacing are essential for fostering a seamless integration between neuromorphic e-skin and biological systems. Prioritizing the biocompatibility and stability of neural interfaces is critical for enhancing their compatibility with human tissues, which can lead to more effective and lasting applications. One promising avenue of research focuses on the development of hydrogel-based electrode arrays. Given their tissue-mimicking mechanical properties, volumetric capacitance, and customizable conductivity, hydrogels are well-positioned to establish conformal contact with neural tissues. These characteristics can facilitate improved signal transmission while minimizing immune responses. Recent studies have indicated that hydrogel-based neural interfaces possess the capability for long-term, high-quality detection of neural activity, showcasing significant potential for application in neuro-prosthetics and bioelectronics.^122,260 Another important area for exploration involves the development of closed-loop systems that effectively connect sensory feedback from neuromorphic e-skin with neural stimulation. Such systems hold the promise of enabling realistic and responsive control of prosthetics, thereby enhancing user interaction with their environment. Recent progress includes the creation of a monolithic soft prosthetic e-skin capable of multimodal perception and neuromorphic pulse-train signal generation, which has been successfully integrated into neuromorphic sensorimotor loops.^32,121 Additionally, neuromorphic hardware for somatosensory neuro-prostheses has demonstrated potential advantages such as increased information bandwidth, parallelization, and portability.⁴⁶ These innovations highlight the potential of closed-loop neuromorphic systems in enhancing prosthetics. Future research should aim to improve neural interfaces for better integration with the human nervous system, focusing on long-term stability and biocompatibility. Key goals include enhancing mechanical robustness and electrical performance, and optimizing algorithms for sensory feedback. Addressing these challenges can lead to improved e-skin systems, benefiting prosthetics, human–machine interfaces, and biomedical applications.

Advancing the scalability and standardization of neuromorphic e-skin manufacturing represents a vital opportunity for moving from laboratory prototypes to commercially viable products. By developing scalable fabrication methods like roll-to-roll printing and laser-assisted processes, there is great potential for producing large-area e-skins without compromising performance. These innovative techniques can significantly accelerate the commercialization process while lowering manufacturing costs. For instance, the successful scalable batch fabrication of ultrathin flexible neural probes using a silk-perylene bilayer has shown the possibility of creating high-performance, flexible electronic components that integrate seamlessly into e-skin systems.²⁶¹ Alongside these fabrication advancements, establishing standardized protocols for testing and evaluating the performance of neuromorphic e-skin is crucial, especially for applications in prosthetics. Standardization will promote consistency, reliability, and safety across devices from various manufacturers, facilitating regulatory approval and broader market adoption. A significant example is the development of standardized material testing protocols for prosthetic liners, which provide a solid framework for characterizing materials based on clinically relevant properties. Such protocols could serve as an inspiration for creating similar tailored standards for e-skin technologies.²⁶² Future research should focus on refining scalable fabrication processes for e-skin devices to ensure functional integrity and uniformity during mass production. This includes optimizing materials and manufacturing methods for high performance across large areas, as well as developing comprehensive testing standards to evaluate mechanical durability, electrical functionality, and biocompatibility. Establishing these standards will support regulatory compliance and enhance consumer trust in e-skin reliability and safety. Addressing these challenges will facilitate the adoption of neuromorphic e-skins in applications like prosthetics, robotics, and wearable health monitoring, ultimately advancing human–machine interfaces and biomedical engineering.

The exploration of decentralized neuromorphic architectures, wherein computation is distributed across sensor nodes, represents a transformative pathway for advancing neuromorphic e-skin systems. By enabling local processing of sensory information, this approach significantly reduces latency and improves system response times, minimizing reliance on centralized data processing. Furthermore, decentralized architecture enhances fault tolerance by allowing individual sensor nodes to independently perform critical functions. This capability is particularly advantageous for robotics, where maintaining system functionality in dynamic and unpredictable environments is crucial.²⁶³ Incorporating decentralized learning algorithms, such as localized reinforcement learning, offers an additional avenue for advancement. These algorithms empower neuromorphic e-skins to autonomously adapt to environmental changes, enhancing the intelligence and flexibility of robotic systems. For instance, recent studies have demonstrated the feasibility of localized reinforcement learning through distributed synaptic circuits, enabling robots to learn and navigate complex environments effectively. Such advancements could lead to more autonomous and adaptive robotic systems, broadening their application in dynamic real-world scenarios.²⁶⁴

Overall, future efforts should focus on refining decentralized architectures to enhance local computation capabilities further while optimizing energy efficiency and scalability. Additionally, advancing localized learning algorithms to improve adaptability and developing robust neuromorphic sensorimotor systems to enable seamless human–machine interaction will be pivotal. By tackling these challenges, neuromorphic e-skins could revolutionize applications in various sectors, particularly robotics, prosthetics, and human–machine interfaces, making them more intelligent, efficient, and adaptable to dynamic environments. Addressing these ambiguities will be critical for unlocking the full potential of neuromorphic e-skin technology.

6. Conclusions

In conclusion, neuromorphic e-skin technologies represent a significant advancement in the integration of sensing capabilities with intelligent processing, bridging the gap between biological and artificial systems. By emulating the complex functionalities of biological neurons, these innovations enhance the sensory perception and responsiveness of machines, thereby improving human–machine interactions across various applications, including robotics, prosthetics, and healthcare. The diverse architectures of neuromorphic devices, such as memristor networks and CMOS-compatible designs, are instrumental in facilitating efficient sensory processing while maintaining low power consumption and a compact form factor. The incorporation of advanced materials, particularly 2D materials like graphene, has further improved the flexibility, durability, and sensitivity of e-skin, making it capable of operating in real-time under various environmental conditions.

Neuromorphic e-skin represents a transformative technology with profound implications for sensory applications across robotics, prosthetics, and human–machine interfaces. By integrating flexible sensors and neuromorphic components, neuromorphic e-skin closely emulates the human skin's sensory and processing capabilities, thus enabling real-time perception and response to environmental stimuli. The two principal approaches, direct and indirect, each offer unique advantages: while direct systems provide swift, localized processing ideal for applications demanding immediate feedback, indirect systems leverage external processors to achieve more complex data handling and computational capacity. Both approaches underscore the versatility and adaptability of neuromorphic e-skin technology, but their practical implementations present distinct technical challenges. Direct approaches must contend with constraints in data processing capabilities and scalability, whereas indirect methods may face issues around latency and integration with flexible, wearable designs. Future advancements in materials science, flexible electronics, and neuromorphic algorithms will likely address these challenges, enhancing the functionality, durability, and responsiveness of e-skin technologies.

While the potential applications of neuromorphic e-skin are vast, several challenges must be addressed to realize its full capabilities. Key issues include the need for large-scale manufacturing processes that meet stringent quality and efficiency standards, as well as the development of energy-harvesting solutions to create self-powered systems. Furthermore, enhancing compatibility with established electronic frameworks will be crucial for integrating these technologies into internet of things (IoT) ecosystems, facilitating widespread deployment in numerous sectors, including environmental monitoring and personalized healthcare.

Future research should prioritize interdisciplinary collaboration among materials science, neuroscience, and electronics to unravel the full potential of neuromorphic e-skin. By refining these technologies, we can empower machines to perceive, learn, and adapt in ways that closely mirror biological organisms, paving the way for transformative advancements in human–machine interactions. The advancement of neuromorphic e-skin not only holds promise for technological innovation but also offers significant societal benefits, from improved health monitoring and rehabilitation to enhanced safety in robotic systems. Through continued exploration and development in this exciting field, we stand to gain profound insights and practical applications that could revolutionize the way humans and machines coexist and collaborate in daily life.

Author contributions

Chandrashekhar S. Patil: conceptualization, investigation, writing – original draft, Sourabh B. Ghode: investigation, visualization, Jungmin Kim: investigation, visualization, Girish U. Kamble: investigation, visualization, Somnath S. Kundale: investigation, visualization, Abdul Mannan: investigation, visualization, Youngbin Ko: visualization, Muhammad Noman: investigation, Qazi Muhammad Saqib: investigation, Swapnil R. Patil: visualization, Seo Yeong Bae: writing – review & editing, Jin Hyeok Kim: supervision, Jun Hong Park: supervision, Jinho Bae: conceptualization, supervision, investigation, funding acquisition, writing – review & editing.

Data availability

The data used to support the findings of this study are included within the article.

Conflicts of interest

The authors declare no conflict of interest.

Acknowledgements

This work was financially supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (2020R1A2C1011433 and RS-2024-00405278). This research was supported by the Korea Institute of Marine Science & Technology Promotion (KIMST) funded by the Ministry of Oceans and Fisheries (20220171).

References

1. C. Wang, X. Xu, X. Pi, M. D. Butala, W. Huang, L. Yin, W. Peng, M. Ali, S. C. Bodepudi, X. Qiao, Y. Xu, W. Sun and D. Yang, Nat. Commun., 2022, 13, 5216.
2. G. Pacchioni, Nat. Rev. Mater., 2022, 7, 847.
3. F. Sun, Q. Lu, S. Feng and T. Zhang, ACS Nano, 2021, 15, 3875–3899.
4. Z. Guo, G. Liu, Y. Sun, Y. Zhang, J. Zhao, P. Liu, H. Wang, Z. Zhou, Z. Zhao, X. Jia, J. Sun, Y. Shao, X. Han, Z. Zhang and X. Yan, ACS Nano, 2023, 17, 21518–21530.
5. P. Zhu, Z. Li, J. Pang, P. He and S. Zhang, Soft Sci., 2024, 4(17), 1–38.
6. A. Costa Cornellà, D. Hardman, L. Costi, J. Brancart, G. Van Assche and F. Iida, Sci. Rep., 2023, 13, 20004.
7. J. Xu, Z. Luo, L. Chen, X. Zhou, H. Zhang, Y. Zheng and L. Wei, Mater. Horiz., 2024, 11, 4015–4036.
8. W. Wang, Y. Jiang, D. Zhong, Z. Zhang, S. Choudhury, J. C. Lai, H. Gong, S. Niu, X. Yan, Y. Zheng, C. C. Shih, R. Ning, Q. Lin, D. Li, Y. H. Kim, J. Kim, Y. X. Wang, C. Zhoa, C. Xu, X. Ji, Y. Nishio, H. Lyu, J. B. H. Tok and Z. Bao, Science, 2023, 380, 735–742.
9. X. Liu, D. Li, Y. Wang, D. Yang and X. Pi, APL Mach. Learn., 2023, 1, 031501.
10. S. Dai, Y. Dai, Z. Zhao, F. Xia, Y. Li, Y. Liu, P. Cheng, J. Strzalka, S. Li, N. Li, Q. Su, S. Wai, W. Liu, C. Zhang, R. Zhao, J. J. Yang, R. Stevens, J. Xu, J. Huang and S. Wang, Matter, 2022, 5, 3375–3390.
11. A. Vitale, E. Donati, R. Germann and M. Magno, IEEE Sens. J., 2022, 22, 19490–19499.
12. L. Mao, ACS Sens., 2023, 8, 2896–2897.
13. C. D. Schuman, S. R. Kulkarni, M. Parsa, J. P. Mitchell, P. Date and B. Kay, Nat. Comput. Sci., 2022, 2, 10–19.
14. A. Shrestha, H. Fang, Z. Mei, D. P. Rider, Q. Wu and Q. Qiu, IEEE Circuits Syst. Mag., 2022, 22, 6–35.
15. F. Liu and C. Liu, Presented in part at the Proceedings of the 55th Annual Design Automation Conference, San Francisco, California, 2018.
16. P.-K. Zhou, X.-L. Lin, H.-L. Yang, B.-J. Chen, H.-H. Li, Z.-R. Chen and S.-T. Zheng, J. Alloys Compd., 2022, 925, 166783.
17. S.-O. Park, H. Jeong, J. Park, J. Bae and S. Choi, Nat. Commun., 2022, 13, 2888.
18. A. Handler and D. D. Ginty, Nat. Rev. Neurosci., 2021, 22, 521–537.
19. L. R. Hochberg, D. Bacher, B. Jarosiewicz, N. Y. Masse, J. D. Simeral, J. Vogel, S. Haddadin, J. Liu, S. S. Cash, P. van der Smagt and J. P. Donoghue, Nature, 2012, 485, 372–375.
20. S. N. Flesher, J. E. Downey, J. M. Weiss, C. L. Hughes, A. J. Herrera, E. C. Tyler-Kabara, M. L. Boninger, J. L. Collinger and R. A. Gaunt, Science, 2021, 372, 831–836.
21. G. Valle, A. Mazzoni, F. Iberite, E. D’Anna, I. Strauss, G. Granata, M. Controzzi, F. Clemente, G. Rognini, C. Cipriani, T. Stieglitz, F. M. Petrini, P. M. Rossini and S. Micera, Neuron, 2018, 100, 37–45.
22. K. Lee, X. Ni, J. Y. Lee, H. Arafa, D. J. Pe, S. Xu, R. Avila, M. Irie, J. H. Lee, R. L. Easterlin, D. H. Kim, H. U. Chung, O. O. Olabisi, S. Getaneh, E. Chung, M. Hill, J. Bell, H. Jang, C. Liu, J. B. Park, J. Kim, S. B. Kim, S. Mehta, M. Pharr, A. Tzavelis, J. T. Reeder, I. Huang, Y. Deng, Z. Xie, C. R. Davies, Y. Huang and J. A. Rogers, Nat. Biomed. Eng., 2020, 4, 148–158.
23. J. Shi, Z. Wang, Y. Tao, H. Xu, X. Zhao, Y. Lin and Y. Liu, Front. Neurosci., 2021, 15, 662457.
24. B. C.-K. Tee, A. Chortos, A. Berndt, A. K. Nguyen, A. Tom, A. McGuire, Z. C. Lin, K. Tien, W.-G. Bae, H. Wang, P. Mei, H.-H. Chou, B. Cui, K. Deisseroth, T. N. Ng and Z. Bao, Science, 2015, 350, 313–316.
25. J. Lee, S. J. Ihle, G. S. Pellegrino, H. Kim, J. Yea, C.-Y. Jeon, H.-C. Son, C. Jin, D. Eberli, F. Schmid, B. L. Zambrano, A. F. Renz, C. Forró, H. Choi, K.-I. Jang, R. Küng and J. Vörös, Nat. Electron., 2021, 4, 291–301.
26. G. W. V. Vidal, M. L. Rynes, Z. Kelliher and S. J. Goodwin, Scientifica, 2016, 2016, 8956432.
27. J. E. Niemeyer, Lab Animal, 2016, 45, 359–361.
28. I. Iturrate, R. Chavarriaga, L. Montesano, J. Minguez and J. D. R. Millán, Sci. Rep., 2015, 5, 13893.
29. Karen A. Moxon and G. Foffani, Neuron, 2015, 86, 55–67.
30. M. S. Willsey, S. R. Nason-Tomaszewski, S. R. Ensel, H. Temmar, M. J. Mender, J. T. Costello, P. G. Patil and C. A. Chestek, Nat. Commun., 2022, 13, 6899.
31. P. Zhu, S. Mu, W. Huang, Z. Sun, Y. Lin, K. Chen, Z. Pan, M. G. Haghighi, R. Sedghi, J. Wang and Y. Mao, Nano Res., 2024, 17, 6550–6559.
32. X. Wang, H. Yang, E. Li, C. Cao, W. Zheng, H. Chen and W. Li, Small, 2023, 19, 2205395.
33. C. M. Boutry, M. Negre, M. Jorda, O. Vardoulis, A. Chortos, O. Khatib and Z. Bao, Sci. Rob., 2018, 3, eaau6914.
34. X. Zhao, H. Zou, M. Wang, J. Wang, T. Wang, L. Wang and X. Chen, Adv. Mater., 2024, 36, 2403444.
35. P. S. Zarrin, F. Zahari, M. K. Mahadevaiah, E. Perez, H. Kohlstedt and C. Wenger, Sci. Rep., 2020, 10, 19742.
36. J. M. Lee, S. W. Cho, C. Jo, S. H. Yang, J. Kim, D. Y. Kim, J.-W. Jo, J. S. Park, Y.-H. Kim and S. K. Park, Sci. Adv., 2024, 10, eadp9885.
37. K. Kang, K. Kim, J. Baek, D. J. Lee and K. J. Yu, Appl. Phys. Rev., 2024, 11, 021328.
38. P. Zhu, S. Mu, W. Huang, Z. Sun, Y. Lin, K. Chen, Z. Pan, M. G. Haghighi, R. Sedghi and J. Wang, Nano Res., 2024, 1–9.
39. A.-C. Bunea, V. Dediu, E. A. Laszlo, F. Pistriţu, M. Carp, F. S. Iliescu, O. N. Ionescu and C. Iliescu, Micromachines, 2021, 12, 1091.
40. Y. Lee, J. Y. Oh and T. W. Lee, Adv. Mater. Technol., 2022, 7, 2200193.
41. V. Bhaltadak, B. Ghewade and S. Yelne, Cureus, 2024, 16.
42. E. Covi, E. Donati, X. Liang, D. Kappel, H. Heidari, M. Payvand and W. Wang, Front. Neurosci., 2021, 15, 611300.
43. A. S. Raikar, J. Andrew, P. P. Dessai, S. M. Prabhu, S. Jathar, A. Prabhu, M. B. Naik and G. V. S. Raikar, Artif. Intell. Rev., 2024, 57, 318.
44. C. Bartolozzi, A. Glover and E. Donati, Handbook of Neuroengineering, Springer, 2021, pp. 1–31.
45. J. Tu, M. Wang, W. Li, J. Su, Y. Li, Z. Lv, H. Li, X. Feng and X. Chen, Soft Sci., 2023, 3(25), 1–15.
46. E. Donati and G. Valle, Nat. Commun., 2024, 15, 556.
47. T.-M. Jang, W. B. Han, S. Han, A. Dutta, J. H. Lim, T. Kim, B. H. Lim, G. J. Ko, J.-W. Shin, R. Kaveti, H. S. Kang, C. H. Eom, S. J. Choi, A. J. Bandodkar, K.-S. Lee, E. Park, H. K. Cheng, W. H. Yeo and S. W. Hwang, Sci. Adv., 2024, 10(36), 1–10.
48. A. S. Nittala and J. Steimle, Next Steps in Epidermal Computing: Opportunities and Challenges for Soft On-Skin Devices, CHI Proceedings of the 2022 CHI Conference on Human Factors in Computing Systems, pp. 1–22 DOI:10.1145/3491102.3517668.
49. R. Thapa, S. Poudel, K. Krukiewicz and A. Kunwar, Measurement, 2024, 114123.
50. Z. Tang, B. Sun, G. Zhou, Y. Zhou, Z. Cao, X. Duan, W. Yan, X. Chen and J. Shao, Mater. Today Nano, 2023, 100439.
51. Q. Yuan, Z. Ma, S. Li and L. Pan, J. Semicond., 2019, 40, 1–2.
52. S. Kim, Y. Lee, H.-D. Kim and S.-J. Choi, NPG Asia Mater., 2020, 12, 76.
53. H. Abaci, Z. Guo, Y. Doucet, J. Jackow and A. Christiano, Exp. Biol. Med., 2017, 242, 1657–1668.
54. K. Qian, V. C. Nguyen, T. Chen and P. S. Lee, J. Mater. Chem. C, 2016, 4, 9637–9645.
55. R. Dahiya, N. Yogeswaran, F. Liu, L. Manjakkal, E. Burdet, V. Hayward and H. Jörntell, Proc. IEEE, 2019, 107, 2016–2033.
56. M. R. Ahmed and B. Sujatha, A Review on Methods, Issues and Challenges in Neuromorphic Engineering, International Conference on Communications and Signal Processing (ICCSP), 2015, pp. 0899–0903 DOI:10.1109/ICCSP.2015.7322626.
57. H. Patel, A. Vanarse, K. D. Carlson and A. Osseiran, Bringing Touch to the Edge: A Neuromorphic Processing Approach For Event-Based Tactile Systems, 2023 IEEE 5th International Conference on Artificial Intelligence Circuits and Systems (AICAS), 2023, pp. 1–5 DOI:10.1109/AICAS57966.2023.10168592.
58. J. H. Koo, Y. J. Lee, H. J. Kim, W. Matusik, D.-H. Kim and H. Jeong, Annu. Rev. Biomed. Eng., 2024, 26, 331–355.
59. J. C. Yang, J. Mun, S. Y. Kwon, S. Park, Z. Bao and S. Park, Adv. Mater., 2019, 31, 1904765.
60. T. Wang, J. Meng, X. Zhou, Y. Liu, Z. He, Q. Han, Q. Li, J. Yu, Z. Li, Y. Liu, H. Zhu, Q. Sun, D. Zhang, P. Chen, H. Peng and L. Chen, Nat. Commun., 2022, 13, 7432.
61. F. Chen, S. Zhang, L. Hu, J. Fan, C.-H. Lin, P. Guan, Y. Zhou, T. Wan, S. Peng, C. Wang, L. Wu, T. Furlong, N. Valanoor and D. Chu, Adv. Funct. Mater., 2023, 33, 2300266.
62. L. Gao, M. Wu, X. Yu and J. Yu, Int. J. Extreme Manuf., 2023, 6, 012005.
63. K. Ishida, N. Masunaga, Z. Zhou, T. Yasufuku, T. Sekitani, U. Zschieschang, H. Klauk, M. Takamiya, T. Someya and T. Sakurai, IEEE J. Solid-State Circuits, 2009, 45, 249–259.
64. D. B. Strukov, G. S. Snider, D. R. Stewart and R. S. Williams, Nature, 2008, 453, 80–83.
65. W. Sun, B. Gao, M. Chi, Q. Xia, J. J. Yang, H. Qian and H. Wu, Nat. Commun., 2019, 10, 3453.
66. L. Chua, IEEE Trans. Circuit Theory, 1971, 18, 507–519.
67. M. Riordan, L. Hoddeson and C. Herring, Rev. Mod. Phys., 1999, 71, S336.
68. W. F. Brinkman, D. E. Haggan and W. W. Troutman, IEEE J. Solid-State Circuits, 1997, 32, 1858–1865.
69. M. Pejović and M. M. Pejovic, Different Types of Field-Effect Transistors: Theory and Applications, BoD–Books on Demand, 2017.
70. G. V. Angelov, D. N. Nikolov and M. H. Hristov, J. Electr. Comput. Eng., 2019, 4792461.
71. R. Pritchard, Proc. IEEE, 1998, 86, 150–162.
72. W. Cao, H. Bu, M. Vinet, M. Cao, S. Takagi, S. Hwang, T. Ghani and K. Banerjee, Nature, 2023, 620, 501–515.
73. M. Zeng, Y. He, C. Zhang and Q. Wan, Front. Neurosci., 2021, 15, 1–23.
74. S. Zhu, B. Sun, G. Zhou, T. Guo, C. Ke, Y. Chen, F. Yang, Y. Zhang, J. Shao and Y. Zhao, ACS Appl. Mater. Interfaces, 2023, 15, 5420–5431.
75. J. Y. Kwon, J. E. Kim, J. S. Kim, S. Y. Chun, K. Soh and J. H. Yoon, Exploration, 2024, 4, 20220162.
76. Z. Tang, B. Sun, G. Zhou, Y. Zhou, Z. Cao, X. Duan, W. Yan, X. Chen and J. Shao, Mater. Today Nano, 2024, 25, 100439.
77. J.-K. Han, J. Oh, G.-J. Yun, D. Yoo, M.-S. Kim, J.-M. Yu, S.-Y. Choi and Y.-K. Choi, Sci. Adv., 2021, 7, eabg8836.
78. R. Khan, N. U. Rehman, S. Iqbal, S. Abdullaev and H. M. Aldosari, ACS Appl. Electron. Mater., 2023, 6, 73–119.
79. W. Chen, L. Song, S. Wang, Z. Zhang, G. Wang, G. Hu and S. Gao, Adv. Electron. Mater., 2023, 9, 2200833.
80. H. Lee, P. Chen, T. Wu, Y. Chen, C. Wang, P. Tzeng, C. Lin, F. Chen, C. Lien and M.-J. Tsai, Low power and high-speed bipolar Switching with a thin reactive Ti buffer layer in robust HfO2-based RRAM, 2008, pp. 1–4 DOI:10.1109/IEDM.2008.4796677.
81. H. Jiang, L. Han, P. Lin, Z. Wang, M. H. Jang, Q. Wu, M. Barnell, J. J. Yang, H. L. Xin and Q. Xia, Sci. Rep., 2016, 6, 28525.
82. S. Teja Nibhanupudi, A. Roy, D. Veksler, M. Coupin, K. C. Matthews, M. Disiena Ansh, J. V. Singh, I. R. Gearba-Dolocan and J. Warner, Nat. Commun., 2024, 15, 2334.
83. C.-H. Cheng, C.-Y. Tsai, A. Chin and F. Yeh, High-performance ultra-low energy RRAM with good retention and endurance, 2008 IEEE International Electron Devices Meeting, San Francisco, 2010, pp. 19.4.1–19.4.4 DOI:10.1109/IEDM.2010.5703392.
84. J. Tang, X. Pan, X. Chen, B. Jiang, X. Li, J. Pan, H. Qu, Z. Huang, P. Wang and J. Duan, Adv. Funct. Mater., 2025, 2412375.
85. A. N. Mikhaylov, E. G. Gryaznov, M. N. Koryazhkina, I. A. Bordanov, S. A. Shchanikov, O. A. Telminov and V. B. Kazantsev, Supercomput. Front. Innov., 2023, 10, 77–103.
86. J. Meng, Z. Li, Y. Fang, Q. Li, Z. He, T. Wang, H. Zhu, L. Ji, Q. Sun and D. W. Zhang, IEEE Electron Device Lett., 2022, 43, 2069–2072.
87. R. Li, S. Shreya, S. Ricci, D. Bridarolli, D. Ielmini, H. Farkhani and F. Moradi, Thermal-Induced Multi-State Memristors for Neuromorphic Engineering, 2023 IEEE International Symposium on Circuits and Systems (ISCAS), 2023, pp. 1–5 DOI:10.1109/ISCAS46773.2023.10182122.
88. M. Wu, Q. Zhuang, K. Yao, J. Li, G. Zhao, J. Zhou, D. Li, R. Shi, G. Xu and Y. Li, InfoMat, 2023, 5, e12472.
89. I. Boybat, M. Gallo, N. S. R. T. Moraitis, T. Parnell, T. Tuma, B. Rajendran, Y. Leblebici, A. Sebastian and E. Eleftheriou, Nat. Commun., 2018, 9, 2514.
90. M. Besrour, J. Lavoie, T. Omrani, G. Martin-Hardy, E. R. Koleibi, J. Menard, K. Koua, P. Marcoux, M. Boukadoum and R. Fontaine, arXiv, 2024, preprint, arXiv:2408.07734 DOI:10.48550/arXiv.2408.07734.
91. W. Wan, R. Kubendran, C. Schaefer, S. B. Eryilmaz, W. Zhang, D. Wu, S. Deiss, P. Raina, H. Qian, B. Gao, S. Joshi, H. Wu, H. S. P. Wong and G. Cauwenberghs, Nature, 2022, 608, 504–512.
92. S. Wu, X. Zhang, R. Cao, K. Zhou, J. Lu, C. Li, Y. Yang, D. Shang, Y. Wei and H. Jiang, Appl. Phys. Lett., 2024, 124(10), 102902.
93. S. Parveen, L. T. Manamel, A. Mukherjee, S. Sagar and B. C. Das, Adv. Mater. Interfaces, 2022, 9, 2200562.
94. F. Ponulak and A. Kasinski, Acta Neurobiol. Exp., 2011, 71, 409–433.
95. W. W. Lee, S. L. Kukreja and N. V. Thakor, Front. Neurosci., 2017, 11, 5.
96. Y. He, L. Zhu, Y. Zhu, C. Chen, S. Jiang, R. Liu, Y. Shi and Q. Wan, Adv. Intell. Syst., 2021, 3, 2000210.
97. T. Wunderlich, A. F. Kungl, E. Müller, A. Hartel, Y. Stradmann, S. A. Aamir, A. Grübl, A. Heimbrecht, K. Schreiber, D. Stöckel, C. Pehle, S. Billaudelle, G. Kiene, C. Mauch, J. Schemmel, K. Meier and M. A. Petrovici, Front. Neurosci., 2019, 13, 1–15.
98. V. Milo, G. Malavena, C. Monzio Compagnoni and D. Ielmini, Materials, 2020, 13, 166.
99. S. Srivastava and S. S. Rathod, Silicon neuron-analog CMOS VLSI implementation and analysis at 180 nm, 2016 3rd International Conference on Devices, Circuits and Systems (ICDCS), 2016, pp. 28–32 DOI:10.1109/ICDCSyst.2016.7570617.
100. K. Kang, S. Ye, C. Jeong, J. Jeong, Y.-S. Ye, J.-Y. Jeong, Y.-J. Kim, S. Lim, T. H. Kim and K. Y. Kim, Nat. Commun., 2024, 15, 10.
101. T. Wunderlich, A. F. Kungl, E. Müller, A. Hartel, Y. Stradmann, S. A. Aamir, A. Grübl, A. Heimbrecht, K. Schreiber and D. Stöckel, Front. Neurosci., 2019, 13, 260.
102. Z. Yu, A. M. Abdulghani, A. Zahid, H. Heidari, M. A. Imran and Q. H. Abbasi, Ieee Access, 2020, 8, 67085–67099.
103. S. Najmaei, A. L. Glasmann, M. A. Schroeder, W. L. Sarney, M. L. Chin and D. M. Potrepka, Mater. Today, 2022, 59, 80–106.
104. G. Cao, P. Meng, J. Chen, H. Liu, R. Bian, C. Zhu, F. Liu and Z. Liu, Adv. Funct. Mater., 2021, 31, 2005443.
105. W. Huh, D. Lee and C. H. Lee, Adv. Mater., 2020, 32, 2002092.
106. X. Lin, F. Li, Y. Bing, T. Fei, S. Liu, H. Zhao and T. Zhang, Nano-Micro Lett., 2021, 13, 1–14.
107. H.-S. Lee, J.-S. Ro, G.-M. Ko and H.-L. Park, Flexible Printed Electron., 2023, 8, 043001.
108. Y. Guo, X. Wei, S. Gao, W. Yue, Y. Li and G. Shen, Adv. Funct. Mater., 2021, 31, 2104288.
109. M. M. Makhoul-Mansour, J. J. Maraj and S. A. Sarles, J. Compos. Mater, 2023, 57, 659–678.
110. R. Bogue, Ind. Rob., 2024, 51, 697–703.
111. C. D. Flynn, D. Chang, A. Mahmud, H. Yousefi, J. Das, K. T. Riordan, E. H. Sargent and S. O. Kelley, Nat. Rev. Bioeng., 2023, 1, 560–575.
112. X. Sun, X. Guo, J. Gao, J. Wu, F. Huang, J.-H. Zhang, F. Huang, X. Lu, Y. Shi and L. Pan, Biomedicines, 2024, 12, 2307.
113. S. Pradhan, A. K. Brooks and V. K. Yadavalli, Mater. Today Bio, 2020, 7, 100065.
114. J. Wu, X. Yang, J. Chen, S. Li, T. Zhou, Z. Cai, X. Lian and L. Wang, Nanotechnol. Rev., 2024, 13, 20230181.
115. F. Sun, Q. Lu, M. Hao, Y. Wu, Y. Li, L. Liu, L. Li, Y. Wang and T. Zhang, npj Flexible Electron., 2022, 6, 72.
116. E. E. Yildirim, L. A. Panes-Ruiz, P. Yemulwar, E. Cihan, B. Ibarlucea and G. Cuniberti, MRS Bull., 2023, 48, 1208–1220.
117. S. H. Lee, H. J. Lee, D. Jeon, H.-J. Kim and S.-N. Lee, Nanomaterials, 2024, 14, 1501.
118. B. Walters, M. V. Jacob, A. Amirsoleimani and M. Rahimi Azghadi, Adv. Intell. Syst., 2023, 5, 2300136.
119. D. P. Sahu, P. Jetty and S. N. Jammalamadaka, Nanotechnology, 2021, 32, 155701.
120. M. Zahid, M. I. Sadiq, C. Jin, J. Wang, X. Shi, W. Liu, F. Aslam, Y. Xu, M. Tahir, J. Yang and J. Sun, Appl. Phys. Lett., 2024, 125, 193503.
121. W. Wang, Y. Jiang, D. Zhong, Z. Zhang, S. Choudhury, J.-C. Lai, H. Gong, S. Niu, X. Yan, Y.-Q. Zheng, C. C. Shih, R. Ning, Q. Lin, D. Li, Y. H. Kim, J. Kim, Y.-X. Wang, C. Zhao, C. Xu, X. Ji, Y. Nishio, H. Lyu, J. B.-H. Tok and Z. Bao, Science, 2023, 380, 735–742.
122. S. Cheng, R. Zhu and X. Xu, Commun. Mater., 2024, 5, 99.
123. F. Yin, H. Niu, E. S. Kim, Y. K. Shin, Y. Li and N. Y. Kim, InfoMat, 2023, 5, e12424.
124. H. Shim, S. Jang, J. G. Jang, Z. Rao, J.-I. Hong, K. Sim and C. Yu, Nano Res., 2022, 15, 758–764.
125. H. L. Park, Y. Lee, N. Kim, D. G. Seo, G. T. Go and T. W. Lee, Adv. Mater., 2020, 32, 1903558.
126. H. Tang, Y. Li, S. Liao, H. Liu, Y. Qiao and J. Zhou, Adv. Healthcare Mater., 2024, 2400562.
127. X. Sun, X. Guo, J. Gao, J. Wu, F. Huang, J. H. Zhang, F. Huang, X. Lu, Y. Shi and L. Pan, Biomedicines, 2024, 12, 2307.
128. F. Zhang, C. Li, Z. Li, L. Dong and J. Zhao, Microsyst. Nanoeng., 2023, 9, 16.
129. Q. Zhao, Z. Xie, Y.-P. Peng, K. Wang, H. Wang, X. Li, H. Wang, J. Chen, H. Zhang and X. Yan, Mater. Horiz., 2020, 7, 1495–1518.
130. H. Duan, D. Wang, J. Gou, F. Guo, W. Jie and J. Hao, Nanoscale, 2023, 15, 10089–10096.
131. S. J. Kim, H.-J. Lee, C.-H. Lee and H. W. Jang, npj 2D Materials and Applications, 2024, 8, 70.
132. D. Kumar, H. J. Chung, J. Chan, T. Jin, S. T. Lim, S. S. Parkin, R. Sbiaa and S. Piramanayagam, ACS nano, 2023, 17, 6261–6274.
133. F. Miao, Y. Cheng, Y. He, Q. He and Y. Li, Sensors, 2015, 15, 11465–11484.
134. S. Das, A. Chen and M. Marinella, Beyond CMOS, 2021 IEEE International Roadmap for Devices and Systems Outbriefs, 2021, pp. 1–129 DOI:10.1109/IRDS54852.2021.00011.
135. M. Xu, X. Chen, Y. Guo, Y. Wang, D. Qiu, X. Du, Y. Cui, X. Wang and J. Xiong, Adv. Mater., 2023, 35, 2301063.
136. J. Tang, F. Yuan, X. Shen, Z. Wang, M. Rao, Y. He, Y. Sun, X. Li, W. Zhang and Y. Li, Adv. Mater., 2019, 31, 1902761.
137. A. Mehonic, D. Ielmini, K. Roy, O. Mutlu, S. Kvatinsky, T. Serrano-Gotarredona, B. Linares-Barranco, S. Spiga, S. Savel’ev, A. G. Balanov, N. Chawla, G. Desoli, G. Malavena, C. Monzio Compagnoni, Z. Wang, J. J. Yang, S. G. Sarwat, A. Sebastian, T. Mikolajick, S. Slesazeck, B. Noheda, B. Dieny, T.-H. Hou, A. Varri, F. Brückerhoff-Plückelmann, W. Pernice, X. Zhang, S. Pazos, M. Lanza, S. Wiefels, R. Dittmann, W. H. Ng, M. Buckwell, H. R. J. Cox, D. J. Mannion, A. J. Kenyon, Y. Lu, Y. Yang, D. Querlioz, L. Hutin, E. Vianello, S. S. Chowdhury, P. Mannocci, Y. Cai, Z. Sun, G. Pedretti, J. P. Strachan, D. Strukov, M. Le Gallo, S. Ambrogio, I. Valov and R. Waser, APL Mater., 2024, 12, 109201.
138. D. Ielmini and S. Ambrogio, Nanotechnology, 2019, 31, 092001.
139. J. Zhu, T. Zhang, Y. Yang and R. Huang, Appl. Phys. Rev., 2020, 7, 011312.
140. C. García Núñez, L. Manjakkal and R. Dahiya, npj Flexible Electron., 2019, 3, 1.
141. R. Bao, S. Wang, X. Liu, K. Tu, J. Liu, X. Huang, C. Liu, P. Zhou and S. Liu, Nat. Commun., 2024, 15, 1327.
142. Y. Kim, A. Chortos, W. Xu, Y. Liu, J. Y. Oh, D. Son, J. Kang, A. M. Foudeh, C. Zhu, Y. Lee, S. Niu, J. Liu, R. Pfattner, Z. Bao and T.-W. Lee, Science, 2018, 360, 998–1003.
143. S. J. Yoon, J. T. Park and Y. K. Lee, Soft Sci., 2024, 4, 1–23.
144. L. E. Osborn, A. Dragomir, J. L. Betthauser, C. L. Hunt, H. H. Nguyen, R. R. Kaliki and N. V. Thakor, Sci. Rob., 2018, 3(19), 1–26.
145. M. Criado-Gonzalez, L. Bondi, C. Marzuoli, E. Gutierrez-Fernandez, G. Tullii, C. Ronchi, E. Gabirondo, H. Sardon, S. Rapino and M. Malferrari, ACS Appl. Mater. Interfaces, 2023, 15, 35973–35985.
146. R. A. John, N. Tiwari, M. I. B. Patdillah, M. R. Kulkarni, N. Tiwari, J. Basu, S. K. Bose, A. Ankit, C. J. Yu and A. Nirmal, Nat. Commun., 2020, 11, 4030.
147. X. Liu, J. Cao, J. Qiu, X. Zhang, M. Wang and Q. Liu, Front. Nanotechnol., 2022, 3, 1–12.
148. W. Gebhardt and A. G. Ororbia, arXiv, 2024, preprint, arXiv:2407.10028, pp. 1–21 DOI:10.48550/arXiv.2407.10028.
149. H. Wang, N. C. Laurenciu, Y. Jiang and S. D. Cotofana, IEEE Open J. Nanotechnol., 2020, 1, 135–144.
150. L. Chen, S. Karilanova, S. Chaki, C. Wen, L. Wang, B. Winblad, S.-L. Zhang, A. Özçelikkale and Z.-B. Zhang, Science, 2024, 384, 660–665.
151. B. S. Shvetsov, A. A. Minnekhanov, A. V. Emelyanov, A. I. Ilyasov, Y. V. Grishchenko, M. L. Zanaveskin, A. A. Nesmelov, D. R. Streltsov, T. D. Patsaev and A. L. Vasiliev, Nanotechnology, 2022, 33, 255201.
152. Z. Dong, Q. Hua, J. Xi, Y. Shi, T. Huang, X. Dai, J. Niu, B. Wang, Z. L. Wang and W. Hu, Nano Lett., 2023, 23, 3842–3850.
153. D. Pei, S. Yu, X. Zhang, Y. Chen, M. Li and C. Li, Chem. Eng. J., 2022, 445, 136741.
154. M. Li, Q. Shu, X. Qing, J. Wu, Q. Xiao, K. Jia, X. Wang and D. Wang, J. Mater. Chem. C, 2023, 11, 5208–5216.
155. J. Zhang, J. Li, W. Cheng, J.-H. Zhang, Z. Zhou, X. Sun, L. Li, J.-G. Liang, Y. Shi and L. Pan, ACS Mater. Lett., 2022, 4, 577–599.
156. L. E. Osborn, A. Dragomir, J. L. Betthauser, C. L. Hunt, H. H. Nguyen, R. R. Kaliki and N. V. Thakor, Sci. Rob., 2018, 3, eaat3818.
157. D. E. Snyder and E. D. Sapper, J. Compos. Sci., 2022, 6, 141.
158. M. T. Almansoori, X. Li and L. Zheng, Curr. Smart Mater., 2019, 4, 3–14.
159. L. Liu, M. Jin, L. Zhang, Q. Zhang, D. Hu, L. Jin and Z. Nie, Front. Neurosci., 2022, 16, 809657.
160. W. Zhou, Y. Yu, P. Xiao, F. Deng, Y. Zhang and T. Chen, Adv. Mater., 2024, 2403447.
161. C. Silva, J. Deuermeier, W. Zhang, E. Carlos, P. Barquinha, R. Martins and A. Kiazadeh, Adv. Electron. Mater., 2023, 9, 2300286.
162. M.-K. Song, J.-H. Kang, X. Zhang, W. Ji, A. Ascoli, I. Messaris, A. S. Demirkol, B. Dong, S. Aggarwal and W. Wan, ACS nano, 2023, 17, 11994–12039.
163. K. Udaya Mohanan, Nanomaterials, 2024, 14, 527.
164. Z. Li, L. Zhang, L. Guo, W. Hu, A. Yu and J. Zhai, Nano Res., 2023, 16, 11855–11861.
165. F. Liu, S. Deswal, A. Christou, Y. Sandamirskaya, M. Kaboli and R. Dahiya, Sci. Rob., 2022, 7, eabl7344.
166. M. Zeng, Y. He, C. Zhang and Q. Wan, Front. Neurosci., 2021, 15, 690950.
167. W. Lin, B. Wang, G. Peng, Y. Shan, H. Hu and Z. Yang, Adv. Sci., 2021, 8, 2002817.
168. Z. Xu, Y. Hao, A. Luo and Y. Jiang, Med-X, 2024, 2, 24.
169. S. Roh, T. D. Nguyen and J. S. Lee, Appl. Sci., 2024, 14, 1216.
170. M. Jung, J. Lee, S. K. Vishwanath, O.-S. Kwon, C. W. Ahn, K. Shin and S. Jeon, Flexible Printed Electron., 2020, 5, 025003.
171. H. Zhi, X. Zhang, F. Wang, P. Wan and L. Feng, ACS Appl. Mater. Interfaces, 2021, 13, 45987–45994.
172. W. Cao, Y. Luo, Y. Dai, X. Wang, K. Wu, H. Lin, K. Rui and J. Zhu, ACS Appl. Mater. Interfaces, 2023, 15, 3131–3140.
173. H. L. Park, H. Kim, D. Lim, H. Zhou, Y. H. Kim, Y. Lee, S. Park and T. W. Lee, Adv. Mater., 2020, 32, 1906899.
174. W. Wang, Y. Jiang, D. Zhong, Z. Zhang, S. Choudhury, J.-C. Lai, H. Gong, S. Niu, X. Yan and Y. Zheng, Science, 2023, 380, 735–742.
175. H. Bae, D. Kim, M. Seo, I. K. Jin, S. B. Jeon, H. M. Lee, S. H. Jung, B. C. Jang, G. Son and K. Yu, Adv. Mater. Technol., 2019, 4, 1900151.
176. S. Ham, M. Kang, S. Jang, J. Jang, S. Choi, T.-W. Kim and G. Wang, Sci. Adv., 2020, 6, eaba1178.
177. M. I. Tiwana, S. J. Redmond and N. H. Lovell, Sens. Actuators, A, 2012, 179, 17–31.
178. G. Sun, P. Wang and C. Meng, Nano Energy, 2023, 118, 109006.
179. S. Oh, Y. Jung, S. Kim, S. Kim, X. Hu, H. Lim and C. Kim, Sci. Rep., 2017, 7, 16963.
180. H. Niu, Y. Chen, E.-S. Kim, W. Zhou, Y. Li and N.-Y. Kim, Chem. Eng. J., 2022, 450, 138258.
181. A. J. Smith, ACM Comput. Surv., 1982, 14, 473–530.
182. F. Catella, D. Healy, J. A. Lawson and G. A. FitzGerald, Proc. Natl. Acad. Sci. U. S. A., 1986, 83(16), 5861–5865.
183. B. H. Rudall, Robotica, 1989, 7, 275–280.
184. L. Song, Asian-Pac. Econ. Lit., 2014, 28, 197–199.
185. N. H. Quang, N. P. Quang and N. Van Lanh, IJECE, 2022, 12(3), 2393–2404.
186. W. Lin, B. Wang, G. Peng, Y. Shan, H. Hu and Z. Yang, Adv. Sci., 2021, 8, 2002817.
187. M. Eltaib and J. Hewit, Mechatronics, 2003, 13, 1163–1177.
188. M. Park, B.-G. Bok, J.-H. Ahn and M.-S. Kim, Micromachines, 2018, 9, 1–35.
189. J. Chen, L. Wang, X. Xu, G. Liu, H. Liu, Y. Qiao, J. Chen, S. Cao, Q. Cha and T. Wang, Gels, 2022, 8, 356.
190. D. Li, J. Zhou, K. Yao, S. Liu, J. He, J. Su, Q. A. Qu, Y. Gao, Z. Song and C. Yiu, Sci. Adv., 2022, 8, eade2450.
191. O. Boursalie, R. Samavi and T. E. Doyle, J. Healthcare Inform. Res., 2018, 2, 179–203.
192. H. Abd and A. König, Chips, 2023, 2, 142–158.
193. J. P. Lee, H. Jang, Y. Jang, H. Song, S. Lee, P. S. Lee and J. Kim, Nat. Commun., 2024, 15, 530.
194. C. Zhao, J. Park, S. E. Root and Z. Bao, Nat. Rev. Bioeng., 2024, 1–20.
195. Y. Du, J. H. Kim, H. Kong, A. A. Li, M. L. Jin, D. H. Kim and Y. Wang, Adv. Healthcare Mater., 2024, 2303461.
196. S. Kazemzadeh, L. Dodsworth, I. F. Pereira and Y. van de Burgt, Adv. Electron. Mater., 2023, 9, 2200427.
197. S. Srikrajang, L. Kabir, S. Sagadevan, K. Wijaya and W.-C. Oh, J. Mater. Chem. B, 2025, 1–40.
198. J. Chen, Y. Yuan, H. Wang and R. Advincula, Nanotechnology, 2025, 1–15.
199. S. Wang, K. Chen, Q. Zhang, J. Ma, L. Zhu, Y. Sun, X. Liu, B. Wang, Z. Xu and X. You, Mater. Sci. Eng., R, 2025, 162, 100884.
200. K. Lim, H. Seo, W. G. Chung, H. Song, M. Oh, S. Y. Ryu, Y. Kim and J.-U. Park, Commun. Mater., 2024, 5, 49.
201. C. Chen, Y. Wang, H. Wang, X. Wang and M. Tian, Nanomaterials, 2024, 14, 438.
202. J. Liu, M. Liu, Y. Bai, J. Zhang, H. Liu and W. Zhu, Sensors, 2020, 20, 4009.
203. A. Pannone, A. Raj, H. Ravichandran, S. Das, Z. Chen, C. A. Price, M. Sultana and S. Das, Nature, 2024, 1–7.
204. M. Zhang, S. Gong, K. Hakobyan, Z. Gao, Z. Shao, S. Peng, S. Wu, X. Hao, Z. Jiang and E. H. Wong, Adv. Sci., 2024, 11, 2309006.
205. D. G. Pandey, Y. K. Sharma and N. Kumar, Revolutionizing AI with Brain-Inspired Technology: Neuromorphic Computing, IGI Global Scientific Publishing, 2025, pp. 411–428.
206. K. C. Rath, D. Mishra, S. K. T. Tripathy, B. K. Mishra and K. Muduli, Integration of AI, Quantum Computing, and Semiconductor Technology, 2025, pp. 415–440.
207. M. Azeem, M. Shahid, I. Masin and M. Petru, J. Text. Inst., 2025, 116, 80–95.
208. G. Balakrishnan, J. Song, C. Mou and C. J. Bettinger, Adv. Mater., 2022, 34, 2106787.
209. H. J. Shim, S. H. Sunwoo, Y. Kim, J. H. Koo and D. H. Kim, Adv. Healthcare Mater., 2021, 10, 2002105.
210. C. Jiao, J. Liu, S. Yan, Z. Xu, Z. Hou and W. Xu, J. Mater. Chem. C, 2025, 13, 2620–2645.
211. X. Qian, F. Zhang, X. Li, J. Li, H. Sun, Q. Wang, C. Huang, Z. Zhang, Z. Zhou and J. Liu, J. Semicond., 2025, 46, 012602.
212. D.-W. Park, A. A. Schendel, S. Mikael, S. K. Brodnick, T. J. Richner, J. P. Ness, M. R. Hayat, F. Atry, S. T. Frye and R. Pashaie, Nat. Commun., 2014, 5, 5258.
213. K. Tybrandt, D. Khodagholy, B. Dielacher, F. Stauffer, A. F. Renz, G. Buzsáki and J. Vörös, Adv. Mater., 2018, 30, 1706520.
214. P. Varaganti and S. Seo, Biomimetics, 2024, 9, 273.
215. N. Kutner, K. R. Kunduru, L. Rizik and S. Farah, Adv. Funct. Mater., 2021, 31, 2010929.
216. H. K. Raut, R. Das, Z. Liu, X. Liu and S. Ramakrishna, Biotechnol. J., 2020, 15, 2000160.
217. W. Hu, J. Jiang, D. Xie, S. Wang, K. Bi, H. Duan, J. Yang and J. He, Nanoscale, 2018, 10, 14893–14901.
218. M. M. H. Tanim, Z. Templin and F. Zhao, Micromachines, 2023, 14, 235.
219. Y. Lu, Transparent and flexible microelectrode arrays based on graphene for multimodal neural interfaces, University of California, San Diego, 2021.
220. A. Savchenko, D. Kireev, R. T. Yin, I. R. Efimov and E. Molokanova, Front. Bioeng. Biotechnol., 2023, 11, 1168667.
221. N. Ghoshal and B. Tripathy, Primer to Neuromorphic Computing, Elsevier, 2025, pp. 161–183.
222. A. Javanshir, T. T. Nguyen, M. P. Mahmud and A. Z. Kouzani, Neural Computation, 2022, 34, 1289–1328.
223. D. M. Alshangiti, T. K. El-Damhougy, A. Zaher and M. Madani, RSC Adv., 2023, 13, 35251–35291.
224. A. Devi VK, R. Shyam, A. Palaniappan, A. K. Jaiswal, T.-H. Oh and A. J. Nathanael, Polymers, 2021, 13, 3782.
225. P. Čvančara, G. Valle, M. Müller, I. Bartels, T. Guiho, A. Hiairrassary, F. Petrini, S. Raspopovic, I. Strauss and G. Granata, npj Flexible Electron., 2023, 7, 51.
226. K. Kang, S. Ye, C. Jeong, J. Jeong, Y.-S. Ye, J. Jeong, Y.-J. Kim, S. Lim, T. H. Kim, K. Y. Kim, J. U. Kim, G. I. Kim, D. H. Chun, K. Kim, J. Park, J.-H. Hong, B. Park, K.-B. Kim, S. Jung, K. Baek, D. Cho, J. Yoo, K. Lee, H. Cheng, B.-W. Min, H. J. Kim, H. J. Jeon, H. Yi, T.-I. Kim, K. J. Yu and Y. Jung, Nat. Commun., 2024, 15, 021328.
227. J. Y. Oh, D. Son, T. Katsumata, Y. Lee, Y. Kim, J. Lopez, H. C. Wu, J. Kang, J. Park, X. Gu, J. Mun, N. G. J. Wang, Y. Yin, W. Cai, Y. Yun, J. B. H. Tok and Z. Bao, Sci. Adv., 2019, 5, 1–9.
228. H. Z. Shouval, S. S.-H. Wang and G. M. Wittenberg, Front. Comput. Neurosci., 2010, 4, 19.
229. C. G. Núñez, W. T. Navaraj, E. O. Polat and R. Dahiya, Adv. Funct. Mater., 2017, 27, 1606287.
230. P. A. Merolla, J. V. Arthur, R. Alvarez-Icaza, A. S. Cassidy, J. Sawada, F. Akopyan, B. L. Jackson, N. Imam, C. Guo and Y. Nakamura, Science, 2014, 345, 668–673.
231. Q. Feng, Y. Wen, F. Sun, Z. Xie, M. Zhang, Y. Wang, D. Liu, Z. Cheng, Y. Mao and C. Zhao, Energies, 2024, 17, 638.
232. Y. Wang, D. Liu, Y. Zhang, L. Fan, Q. Ren, S. Ma and M. Zhang, ACS nano, 2022, 16, 8283–8293.
233. C. Xu, S. Solomon and W. Gao, 2023 Nat. Mach. Intell., 2023, 5, 1344–1355.
234. G. Cheng, E. Dean-Leon, F. Bergner, J. R. G. Olvera, Q. Leboutet and P. Mittendorfer, Proc. IEEE, 2019, 107, 2034–2051.
235. T. Seifert, M. Baum, F. Roscher, M. Wiemer and T. Gessner, Mater. Today: Proc., 2015, 2, 4262–4271.
236. J. Wang, R. C. Auyeung, H. Kim, N. A. Charipar and A. Piqué, Adv. Mater., 2010, 40, 4462–4466.
237. J. Y. Oh, D. Son, T. Katsumata, Y. Lee, Y. Kim, J. Lopez, H.-C. Wu, J. Kang, J. Park and X. Gu, Sci. Adv., 2019, 5, eaav3097.
238. W. Heng, S. Solomon and W. Gao, Adv. Mater., 2022, 34, 2107902.
239. C. Xu, S. A. Solomon and W. Gao, Nat. Mach. Intell., 2023, 5(12), 1344–1355.
240. Q. Guo, X. Qiu and X. Zhang, Materials, 2022, 15, 1–30.
241. M. S. Kim, A. S. Almuslem, W. Babatain, R. R. Bahabry, U. K. Das, N. El-Atab, M. Ghoneim, A. M. Hussain, A. T. Kutbee, J. Nassar, N. Qaiser, J. P. Rojas, S. F. Shaikh, G. A. Torres Sevilla and M. M. Hussain, Adv. Mater., 2024, 36, 2406424.
242. Y. Guo, X. Wei, S. Gao, W. Yue, Y. Li and G. Shen, Adv. Funct. Mater., 2021, 31, 2104288.
243. J. C. Yang, J. Mun, S. Y. Kwon, S. Park, Z. Bao and S. Park, Adv. Mater., 2019, 31, 1904765.
244. R. S. Dahiya, N. Yogeswaran, F. Liu, L. Manjakkal, E. Burdet, V. Hayward and H. Jörntell, Proc. IEEE, 2019, 107, 2016–2033.
245. X. Fu, W. Cheng, G. Wan, Z. Yang and B. C. K. Tee, Chem. Rev., 2024, 24(17), 9899–9948.
246. M. Soni and R. S. Dahiya, Philos. Trans. R. Soc., A, 2019, 378, 1–26.
247. K. Ding, A. Dragomir, R. Bose, L. E. Osborn, M. S. Seet, A. Bezerianos and N. V. Thakor, J. Neural Eng., 2020, 3, 1–22.
248. D. Son and Z. Bao, ACS Nano, 2018, 12, 11731–11739.
249. C. Shang, Q. Xu, N. Liang, J. Zhang, L. Li and Z. Peng, npj Flexible Electron., 2023, 7, 19.
250. P. Gomez-Romero, A. Pokhriyal, D. Rueda-García, L. N. Bengoa and R. M. González-Gil, Chem. Mater., 2023, 36, 8–27.
251. P. Qin, APL Mater., 2023, 031117.
252. G. Ding, S. T. Han, V. A. L. Roy, C.-C. Kuo and Y. Zhou, Energy Rev., 2023, 2(1), 1–15.
253. Q. Sun, in Handbook of Triboelectric Nanogenerators, ed. Z. L. Wang, Y. Yang, J. Zhai and J. Wang, Springer International Publishing, Cham, 2023, pp. 1843–1876 DOI:10.1007/978-3-031-28111-2_51.
254. J.-K. Han, I.-W. Tcho, S. B. Jeon, J. M. Yu, W. G. Kim and Y.-K. Choi, Adv. Sci., 2022, 9, 2105076.
255. H. Kim, S. Oh, H. Choo, D.-H. Kang and J. H. Park, ACS Nano, 2023, 17(17), 17332–17341.
256. E. Delgado-Alvarado, J. Martínez-Castillo, L. Zamora-Peredo, J. A. González-Calderón, R. López-Esparza, M. W. Ashraf, S. Tayyaba and A. Herrera-May, Nanomaterials, 2022, 12, 4403.
257. Y.-J. Park, Y. G. Ro, Y.-E. Shin, C. Park, S. Na, Y. Chang and H. Ko, Adv. Sci., 2023, 10, 2304598.
258. I. Krauhausen, S. Griggs, I. McCulloch, J. M. J. den Toonder, P. Gkoupidenis and Y. van de Burgt, Nat. Commun., 2024, 15, 1–9.
259. J. Klimaszewski, K. Wildner, A. Ostaszewska-Liżewska, M. Wladzinski and J. Możaryn, Sensors, 2022, 22, 6122.
260. J.-W. Hong, C. Yoon, K. Jo, J. Won and S. Park, iScience, 2021, 24, 1–29.
261. C. Cointe, A. Laborde, L. G. Nowak, D. N. Arvanitis, D. Bourrier, C. Bergaud and A. Maziz, Microsyst. Nanoeng., 2021, 8, 1–11.
262. J. C. Cagle, P. G. Reinhall, B. J. Hafner and J. E. Sanders, J. Biomech. Eng., 2017, 139, 4.
263. R. A. John, N. Tiwari, M. I. B. Patdillah, M. R. Kulkarni, N. Tiwari, J. Basu, S. K. Bose, A. Ankit, C. J. Yu, A. Nirmal, S. K. Vishwanath, C. Bartolozzi, A. Basu and N. Mathews, Nat. Commun., 2020, 11, 4030.
264. F. Liu, S. Deswal, A. Christou, Y. Sandamirskaya, M. Kaboli and R. S. Dahiya, Sci. Rob., 2022, 7, 1–115.

---