Boosting flexible electronics with integration of two-dimensional materials

Abstract

Flexible electronics has emerged as a continuously growing field of study. Two-dimensional (2D) materials often act as conductors and electrodes in electronic devices, holding significant promise in the design of high-performance, flexible electronics. Numerous studies have focused on harnessing the potential of these materials for the development of such devices. However, to date, the incorporation of 2D materials in flexible electronics has rarely been summarized or reviewed. Consequently, there is an urgent need to develop comprehensive reviews for rapid updates on this evolving landscape. This review covers progress in complex material architectures based on 2D materials, including interfaces, heterostructures, and 2D/polymer composites. Additionally, it explores flexible and wearable energy storage and conversion, display and touch technologies, and biomedical applications, together with integrated design solutions. Although the pursuit of high-performance and high-sensitivity instruments remains a primary objective, the integrated design of flexible electronics with 2D materials also warrants consideration. By combining multiple functionalities into a singular device, augmented by machine learning and algorithms, we can potentially surpass the performance of existing wearable technologies. Finally, we briefly discuss the future trajectory of this burgeoning field. This review discusses the recent advancements in flexible sensors made from 2D materials and their applications in integrated architecture and device design.

INTRODUCTION

The development of flexible electronics^1–3 has raised the requirements for integrating electronic devices.^4,5 The next generation of electronic devices necessitates the integration of more functions into a microsystem or device, encompassing integrated sensing,^6,7 storage, and computation. This integration is crucial to achieving low power consumption,^7,8 fast computation, and enhanced convenience. In addition, the integration of multiple sensors^9–11 and data processing capabilities^12–14 enables the realization of advanced technologies such as human–computer interaction^15,16 and virtual reality.^17,18

Flexible electronics are manufactured by constructing electronic devices on flexible substrates such as polyimide (PI) and polyetherimide (PET), and provide greater flexibility than traditional rigid circuit boards. Flexible electronics encompass various components, including input devices, such as flexible touchscreens and pressure sensors, and output devices, such as electronic skins (e-skins), flexible displays, and flexible speakers. Additionally, they include energy management components such as triboelectric nanogenerators (TENGs), solar cells, and supercapacitors. This comprehensive range of components provides solutions for the development of new portable devices.

The atomic-level thicknesses^19,20 of two-dimensional (2D) materials endow them with excellent electronic,^21–25 optical,^26–30 and mechanical properties.^31–33 With the advantages of the high surface-to-volume ratio^34–36 and outstanding electronic mobility,^37–39 2D materials provide unparalleled advantages for high-frequency⁴⁰ and low-power microelectronic devices.^41–44 These exceptional properties indicate that the utilization of 2D materials in flexible electronic design^45–47 holds tremendous potential, and provides a broad platform for developing next-generation electronic devices.^48–51

The integration of flexible electronics with 2D materials, such as graphene⁵² and transition-metal dichalcogenides (TMDCs),⁵³ has become an important research focus in nanotechnology and electronic engineering.^54–56

The roles of 2D materials in flexible electronics are summarized in Table 1. Two-dimensional materials have become indispensable for various applications in flexible electronics. Consequently, numerous 2D materials have been developed and utilized in flexible electronics, including graphene, MXene, TMDCs, and black phosphorus (BP). These materials find utility in the design of pressure sensors for e-skins, photosensitive materials for artificial retinas, TENGs, and channel materials for transistors in digital circuits. The creation of heterostructures involving 2D materials^57,58 facilitates the functional integration of various devices, including transistors, logic gates,^59,60 and memristors. The vertical lattice-free mismatch characteristic^61,62 of 2D materials makes their combination and structural design more convenient.⁶³ By manipulating the stacking mode and interactions between different 2D materials, specific electronic energy-level structures^64–66 and band alignments⁶⁷ can be achieved, enabling the realization of more complex functionalities in flexible electronics. This heterostructure design based on 2D materials holds the potential to pave the way for future ultrahigh-density, ultralow-power flexible integrated circuits,⁶⁸ leading to unprecedented new functionalities.

TABLE 1 Roles of 2D materials in flexible electronics.

Species	Types of devices	Mechanisms of action	Roles of 2D materials in devices	Specific function	Reference
Graphene	Transistors	Auxiliary metal transfer	Material for electrode contact	Implementation of transistor arrays with uniform electrical characteristics	69
Graphene	Photodetectors	Infrared photoelectric charge injection	Photosensitive materials	Broadband imaging from 75 nm to 3.8 μm	70
Graphene	Brain–computer interfaces	Impedance decreases and transparency increases	Conducting electrodes	High spatial resolution recording of neural activity is achieved	71
MXene	Pressure sensors	Charge generation, charge trapping, and charge harvesting	Conductor	Health monitoring	72
MXene	Pressure sensors	Improved mechanical properties	Conducting electrodes for e-skin	Elongation of 2056.67% and tensile strength of 50.78 MPa	63
MXene	Triboelectric nanogenerators	Triboelectric effect	Triboelectric materials	Voltage of 163.7 V is generated for heating to promote wound healing	73
MXene	Supercapacitors	Provides attachment sites for Cu2+ and V3+ ions	Conducting electrodes	Energy density of 80.9 Wh kg−1 is obtained at a power density of 376.0 W kg−1	74
Black phosphorus	Photodiodes	Tunneling effect	Channel materials	Blackbody detection rate of 7.93 × 1010 Jones	75
WS2/WSe2	Photodetectors	Built-in electric field	Photosensitive materials	Response rate of approximately 107 mA/W and response time of 3–4 μs	76
MoS2	Photodetector arrays	Low-energy wireless transmission	Artificial retina by photosensitive materials	Energy consumption of 12.4 nW per pixel under 1 V	77
MoS2	Cascaded transistors	Digital circuits	Transistor channel	Extremely high accuracy of 98.8% and energy consumption of 11.4 W per recognition	78
PdSe2	Transistors	Carrier transconductance	Channel materials	Transport behaviors of carriers, p-type	79
WSe2	Memristors	Reconfigurable synapses	Channel materials	Versatile reconfigurability that provides both excitatory and inhibitory plasticity	80
2D perovskite	Photodetectors	Photoelectric effect	Semiconductor	Detection rate of 2.3 × 1013 Jones and response rate of 2.22 A W−1	81

There are several important reviews on flexible electronics⁸² that systematically introduce mechanical properties, synthesis methods, challenges, and solutions.^83–86 These reviews cover various topics including materials, namely conducting inks,⁸⁷ flexible electrodes,⁸⁸ semiconductors,^89–91 and simple devices, such as transistors,²⁴ and photodetectors.^92,93 Complex circuits were also updated including RF devices,⁹⁴ integrated circuits,⁹⁵ and straintronics.⁸⁵ Besides, the integrated systems were recently summarized, such as displays,³³ smart sensors,^87,96 soft electronics,⁹⁷ and sensor systems.⁴⁷

In this review, we comprehensively demonstrate the incorporation of 2D materials, such as MXene, NbS₂, PdSe₂, and PtTe₂, into fully integrated flexible electronics. We begin by introducing the synthesis of these 2D materials and the technological advancements achieved in flexible electronic processing. Through typical case studies, we then illustrate how 2D materials can be used to integrate multiple functions within a single device. This review provides a current perspective on the hybridization of flexible electronics and 2D materials (Scheme 1).

SCHEME 1. Typical applications of 2D material-empowered flexible and wearable electronics. (1) Flexible and wearable electronics. (A) Pulse temperature measurement. Reprinted with permission98 under the terms of a Creative Commons Attribution License. Copyright 2023. The Authors. Published by UESTC and John Wiley & Sons Australia, Ltd. (B) Rechargeable gloves. Reprinted with permission99 under a Creative Commons Attribution 4.0 International License. Copyright 2021. The Authors. Published by Springer Nature. (2) Flexible energy storage and conversion. (C) E-fabric for charging. Reprinted with permission99 under a Creative Commons Attribution 4.0 International License. Copyright 2021. The Authors. Published by Springer Nature. (D) Uses of mechanical and thermal energy to create sound. Reprinted with permission100 under the terms of a Creative Commons Attribution License. Copyright 2022. The Authors. Published by UESTC and John Wiley & Sons Australia, Ltd. (3) Display and touch. (E) Artificial retina (right). Reprinted with permission101 under the terms of a Creative Commons Attribution License. Copyright 2023. The Authors. Published by Wiley-VCH GmbH. (F) Dynamic trajectory extraction. Reprinted with permission102 under a Creative Commons Attribution 4.0 International License. Copyright 2022. The Authors. Published by Springer Nature. (4) Wireless communication. (G) Wireless smart sensing (right). Reprinted with permission103 under a Creative Commons Attribution 4.0 International License. Copyright 2022. The Authors. Published by Springer Nature. (H) RF signal transmission (Left). Reprinted with permission104 under a Creative Commons Attribution NonCommercial License 4.0 (CC BY-NC). Copyright 2019. The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. (5) Biomedical. (I) Thermal patches (right). Reprinted with permission105 under a Creative Commons Attribution NonCommercial License 4.0 (CC BY-NC). Copyright 2022. The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. (J) Electrocardiogram (left). Reprinted with permission106 under a Creative Commons Attribution-NonCommercial 3.0 Unported License. Copyright 2022. The Authors. Published by the Royal Society of Chemistry.

Our review emphasizes the utilization of 2D materials in nanogenerators for acoustic and motion sensors, textiles for memristors, fiber supercapacitors (FSCs), and integrated circuits for logic and in-memory computing. Furthermore, we provide updates on the progress in artificial intelligence, including piezoelectric–thermal object recognition, artificial retina, artificial throat, artificial ear, temperature, and pulse sensors. These updates are based on machine learning using the datasheet collected from multimodal sensor arrays. E-skin is capable of sensing, imaging, and hyperthermia, whereas neuro-inspired retinas enable the dynamic extraction of movement trajectories and information storage. In addition, state-of-the-art self-powered devices and multifunctional transducers, such as non-contact TENGs, zinc-ion FSCs, and multi-mode speakers, have been developed. Finally, these multifunctional integrated applications can provide a more efficient data processing speed and lightweight wearable device design, thereby promoting the commercialization of 2D material flexible electronics.

PROCESSING AND INTEGRATION TECHNOLOGY OF 2D MATERIAL-BASED FLEXIBLE ELECTRONICS

First, we examine the fabrication of 2D material-based rigid devices. Two-dimensional materials were transferred onto Si/SiO₂ substrates, and patterned channels were then fabricated using lithography, development, and plasma etching. Subsequently, metal electrodes were deposited onto the channels through masks by evaporation. Transistors and photodetectors were also fabricated.

The transfer of large-area 2D materials is prone to damage because of their atomic layer thickness.^107–110 It is therefore crucial to address transfer issues^31,107 to integrate 2D materials into flexible electronics.^111,112 Currently, the primary methods¹¹³ of 2D material transfer include wet transfer, dry transfer, and laser exfoliation. However, these techniques face various challenges, such as material damage during the transfer process, ^31,114 difficulties in controlling transfer accuracy,¹¹⁵ having to decide between single-crystal and polycrystalline 2D materials, and issues related to process cost.¹¹⁶ Polymer materials, such as PMMA, are typically used for the transfer of 2D materials. This process requires a soaking solution to remove the polymer support layer, which can easily be mixed with unnecessary impurities.^117,118 Gao et al.¹¹⁹ achieved peeling of more than 40 2D materials using a high-work-function metal, Au, as a support layer. Ly et al.¹¹³ used the adsorption force of ice to transfer MoS₂ and construct an ultraclean 2D material surface. In addition, intercalation peeling¹²⁰ can effectively overcome the adhesion between the layers of 2D materials and enable the industrial manufacturing of transverse large 2D materials. For example, Huang et al.¹²¹ developed a chemical scissor method that achieved a reversible transition from MXene to MAX. Various nanoscale patterned structures have been formed for electrode and device pathways.

Doping and plasma treatments are employed to regulate the physical properties of 2D materials. Doping can modify the density of states of 2D semiconductors, ^122,123 while plasma treatments lead to the passivation of defects in 2D materials, improving the subthreshold swing. In the case of p–n junction-based devices, it is crucial to establish an ideal band alignment at the heterostructured interface. Therefore, when designing flexible electronics, it is necessary to consider the effect of stress on heterostructured lattices. Furthermore, suitable materials must be developed for packaging these devices to improve their stability.

In flexible electronic applications, achieving the complete transfer of 2D materials onto flexible surfaces such as plastics, for efficient device fabrication,^31,33 poses a significant technological challenge in the integration of 2D material devices. Common techniques, including transfer printing,¹²⁴ inkjet printing, ^125,126 and microcontact printing,¹²⁷ facilitate the integration of 2D materials onto various substrates, such as plastics, ^128,129,116 paper, and even textiles.^130–132 This integration allows for the fabrication of flexible electronics with specific functionalities, including sensors,^133,134 memory devices,^135–137 and energy storage devices.¹³⁸ For example, Wang et al.¹³⁹ used inkjet printing to fabricate MXene-based interdigital electrodes for supercapacitors. Additionally, direct growth on the target materials is a feasible approach. For example, Palacios et al. achieved the direct growth of MoS₂ on wafer-scale silicon chips through a low-temperature process.¹⁴⁰

Currently, there is a pressing need to maximize the integration density of 2D material-based flexible electronics to achieve device miniaturization. This includes advancing solid-state lithium-ion batteries, solid-state zinc-ion batteries, and supercapacitors with high energy densities to further propel wireless charging technology. Additionally, the development of a series of multi-modal sensors coupled with the integration of machine and deep learning, aims to enhance the overall data accuracy of wearables and realize multi-directional monitoring of the human body. Simultaneously, efforts are directed toward the development of biocompatible wearable devices to improve human comfort and implantable applications.

APPLICATION OF INTEGRATING AND FLEXIBLE ELECTRONICS USING 2D MATERIALS

Two-dimensional materials can also be used in sensor designs. Graphene^141–145 and MXene^146,147 find applications in flexible circuits and energy storage devices owing to their excellent conductivity and mechanical flexibility, which enable the manufacture of bendable electronic devices. Dong et al.¹⁴⁸ employed MXene to design a superlattice structure embedded in a monolayer mesoporous carbon framework for energy storage in organic electrolytes containing numerous ions. This approach achieved a volumetric capacitance of 317 F cm⁻³ and an areal energy density of 0.10 mWh cm⁻². Shao et al.¹⁴⁹ constructed dense turbine-loaded graphene, achieving a volumetric capacitance of 234 F cm⁻³ and a power density of 14 kW L⁻¹. Two-dimensional metal oxides¹⁵⁰ and perovskites^151,152 exhibit good optical properties and find applications in solar cells, photodetectors, and flexible displays. Yang et al.¹⁵³ utilized 2D ferroelectric perovskite films to realize a UV photodetector with a high responsivity of 93 A W⁻¹. Two-dimensional TMDCs^154–158 and black phosphorous^159,160 are considered candidates for next-generation integrated-circuit fabrication, owing to their high carrier mobilities. Hao et al.¹⁶⁰ integrated BP with flexible ferroelectric materials to construct ferroelectric synaptic transistors with a mobility of 900 cm² V⁻¹ s⁻¹. Lee et al.¹⁶¹ used MoS₂ to construct a three-terminal hetero synaptic memory transistor to regulate defects. The application of different 2D materials is not limited; stacking a variety of different 2D materials enables the combination of their excellent properties to achieve the multifunctional integrated manufacturing of devices.

Two-dimensional materials drive the development of flexible electronics, owing to their high sensitivity^162–164 and low power consumption^165,166 in electronic and optoelectronic devices such as sensors,^167,168 transistors, ^169,170 and photodetectors. They also exhibit superior performance in energy storage and conversion in solid batteries^171,172 and solar cells.^173,174 Additionally, these materials hold promise in the biomedical field, ^35,175 particularly in drug delivery^176,177 and biosensing applications, ^178,179 because of their high surface area^57,180 and biocompatibility.^181,182 A multifunctional platform with self-healing capabilities^183,184 can be assembled by integrating these single-function devices.³⁰

Flexible wearable sensors

The rapid advancement of flexible electronics in the Internet of Things and artificial intelligence technology has greatly facilitated the management of personalized flexible wearable sensors. People have raised expectations for comfort and intelligence when wearing flexible sensors. Therefore, wearable sensors are designed to be miniaturized and multifunctional, ensuring continuous non-invasive monitoring of human signals, such as pressure and breathing. Herein, we initially explore various 2D materials and their applications in flexible wearable sensors, as detailed in Table 2.

TABLE 2 Two-dimensional material-based flexible sensors according to their sensing from different physical signals.

Signal types	2D materials	Structures	Features	Reference
Pressure	MXene	PDA/MXene/Fiber	Human physiological signal monitoring underwater	185
Pressure	MXene/MoS2	MoS2/MXene/Fiber	Joint-movement monitoring of the human body	186
Pressure	Graphene	Graphene/Fiber	Hyperthermia and exercise monitoring	187
Pressure	Graphene	Graphene/Aerogel	Muscle-stretch monitoring	188
Pressure	MXene	MXene/PDMS	Robotic haptics	189
Gas	MXene/MoS2	MoS2/MXene/Aerogel	NO2 detection in air	190
Gas	Graphene	Graphene/Fiber/Aerogel	Respiratory signal monitoring, encrypted communication	191
Gas	MoS2	Pt/MoS2/Polyaniline nanocomposite	NH3 detection in air	192
Gas	MXene	MXene/Graphene/Cyclic olefin copolymer	Respiratory monitoring, humidity monitoring	193
Light	CdPS3	Ag/CdPS3/ITO	Color recognition	194
Light	MoS2	MoS2/PET/Al2O3	Logical operations	44
Temperature	MoS2	MoS2/Al2O3/PI	Body-temperature monitoring	195
Light	MoS2	PDMS/Au/MoS2	Contact lens	196

As healthcare concepts transition from traditional hospital-centric diagnostics to patient-centric and personalized diagnostics, the materials and designs of sensors must adapt to the softness, elasticity, and curvilinear shape of the human body to ensure both comfort and functionality. Although hydrogels and liquid metals enable sensors to achieve mechanical properties similar to those of the human body, the seamless integration of sensor functions necessitates the introduction of 2D materials to improve performance.

Inspired by the design of human skin, the development of e-skin offers an alternative for sensory restoration.^197–199 This technology mimics the sensation and functionality of natural skin, delivering tactile feedback through actuators or electrohaptic technology.^200,201 For individuals with hearing impairments, the e-skin can transform sound vibrations into electrical signals, which are then relayed to the brain.^202,203 A photosensitive e-skin system was proposed to restore light perception in visually impaired individuals.^204,205 However, the signals generated by the e-skin require secondary decoding by the brain, which is not neurologically viable.²⁰⁶ Neurologically acceptable frequency coding of stimulus signals has been demonstrated.^207–209 Nonetheless, the design of frequency-coded sensory devices have features of significant complexity when achieving biological coherence under multimodal stimuli, for example, temperature,²¹⁰ pressure, ^210,211 and humidity.^212,213 The challenge lies in effectively separating dual signals using a single integrated device when overlapping mixes presented.²¹⁴ Promising sensing technologies, such as piezoresistive^215,216 and thermal sensing,²¹⁷ which respond to resistance changes, currently offer solutions. Leveraging their commonality enables dual-channel signal perception and decoupling, overcoming the design limitations of multimodal sensors.²¹⁸

Shen et al.²¹⁸ devised an MXene-based e-skin with multimodal and neuromorphic-encoded output. The device was manufactured by printing carbon black ink on a PET film to form a thermistor, using MXene as an intermediary conductive layer. Stress testing indicated a high sensitivity of 1319 kPa⁻¹ within a range of 100–300 kPa. To further mitigate the impact of pressure on the reduced precision of temperature monitoring, the team strategically placed a flexible Ecoflex with a low Young's modulus beneath the sensor.

Sensors were developed to perceive and detect pressure and temperature signals from a prosthetic arm and fingers. The signal acquisition and wireless transmission system, comprising a signal processing module, microcontroller unit, and Bluetooth transmission system, transmitted the detected signals to a computer terminal. The signals, acquired from these two modalities, were neuromorphically encoded to form three types of neurologically acceptable frequency signals interfacing with the nervous system. The static pressure and temperature signals were converted into positively correlated pulse spikes using a Python-designed algorithm. Dynamic components of the signal were neurally encoded into frequency signals using an Izhikevich-based neuronal model. Separate analyses of the pressure and temperature signals in the human brain were performed, and the two neural signals were encoded using Python. A 1D convolutional neural network, designed to mirror the processing of the human brain, was constructed to use machine learning for the analysis and recognition of the two signals (Figure 1). Testing with prosthetics demonstrated that the MXene-based temperature–pressure e-skin with decoupling capability exhibited tactile and object recognition abilities, revealing variations in the pressure and temperature signals for different materials.

FIGURE 1. Machine learning-assisted temperature–pressure electronic skin with decoupling capability (TPD e skin) enables object recognition. (A) Principle of using machine learning to recognize objects via e-skin. (B) Structure of a one-dimensional convolutional neural network for TPD object recognition. (C) Breakthrough in grasping objects made from 15 different materials by prosthetics. (D) Temperature–pressure frequency waveforms generated by prosthetic grasping of 15 different materials, realized by neuromorphic coding. (E) Visualization of 15 samples of signals of different frequencies by t-distributed stochastic neighbor embedding (t-SNE). (F) Confusion matrices for 15 types of object recognition. (G) Cognitive outcome waveform during expiration. (H) Identification and waveform of grasping thermoplastic bottles. Reprinted with permission218 under the terms of the Creative Commons Attribution License. Copyright 2023. The Authors. Published by Wiley-VCH GmbH.

Based on the pressure-sensing mechanism, e-skin can be prepared using three mechanisms: piezoresistive, capacitive, and piezoelectric. A strained material undergoing a resistance transformation under pressure is classified as piezoresistive. Piezoelectric e-skin is a favorable choice for dynamic pressure detection and self-powered sensing. In addition, the pressure–capacitive e-skin has a higher resolution for tactile perception.

Alongside the three methods of achieving pressure sensing through changes in the electrical properties caused by mechanical stress or tension, triboelectric electricity emerges as another viable method. Electric signals are generated through variations in the electric field resulting from electron or ion transfer during material contact or separation. The four types of 2D material-based tactile sensors are listed in Table 3.

TABLE 3 Four types of 2D material-based tactile sensors.

2D materials	Composites and interfaces	Types of mechanisms for e-skins	Applications	Reference
MXene	MXene/polyetherimide (PET)	Piezoresistive	Pressure sensing over a wide temperature range	219
MXene	MXene/agar/borax	Piezoresistive	Detection of human joint movement and writing on the surface of a flexible touch screen	133
MXene	MXene/PET	Piezoresistive	Temperature and pressure detection of prosthetic fingers	218
MXene	MXene/nylon fabric/polydopamine (PDA)	Piezoresistive	Human breathing and sleep status detection	220
MXene	MXene/polyurethane	Piezoresistive	Real-time monitoring of ECG signals	221
MXene	MXene/aerogel	Piezoresistive	Exercise monitoring	222
Graphene	PDMS/graphene	Piezoresistive	Detection of high-frequency muscle contraction movements	188
MoS2	MoS2/polyimide (PI) film	Piezoresistive	Real-time, continuous monitoring of skin temperature, ECG, and EMG signals	53
PdSe2	PdSe2/PI	Piezoresistive	Arterial temperature measurement	98
MXene	Zinc-ion hybrid fibers based on MXene	Capacitive	E-textile bracelet to charge a watch	99
Graphene	Graphene/polyethylene terephthalate/WO3/Au	Capacitive	Human sweat (L-cysteine) test	223
Graphene	Graphene	Capacitive	Detection of skin temperature and heat therapy	105
MXene	MXene/silk fibroin nanofiber	Piezoelectric	Cardiovascular health testing	224
MXene	MXene/COF-based composite fibers	Piezoelectric	Charging of watches and smart motors	225
Graphene	Ionic hydrogels/graphene	Piezoelectric	Flexible touch display	226
MXene	MXene/carbon nanofibers (CNFs)	TENG	Capture of long-distance motion tracks	227
Graphene	ZnP porous nanosheets/laser-induced graphene	TENG	Stretchable strain sensor	228
Graphene	Laser-induced graphene (LIG)/PI film	TENG	Dual-function acoustic transducers for machine learning-assisted human–robot interfaces	100

The piezoresistive effect involves a change in electrical resistance when force is applied to a metal or semiconductor device.²²⁹ For devices based on 2D materials, an externally applied force adjusts the band structure of electrons, thereby altering the bandgap width of the material, which, in turn, may limit the flow of charge carriers.^230,231 applied a mechanical bias to a MoS₂ film, using an atomic force microscope tip, and observed the resulting change in resistance. This observation, coupled with finite element simulation analysis, revealed that the MoS₂ band gap can be controlled by applying mechanical stress. Lemme et al.²³² developed a pressure sensor based on PtSe₂. By applying stress to increase the density of states and adjust the band gap, a canonical factor of −85 was achieved. The force also affects the spacing between the 2D materials, and the adjustment of the layer spacing changes the interaction between the layers.^233,234 Ghosh et al.²³¹ constructed a piezoresistive sensor using graphene as a substrate to adjust the graphene band gap from 0.19 to 2.46 eV through compression and tensile adjustment. Additionally, in the context of contact between two different 2D materials, mechanical stresses can modify the contact points present at their junctions.²³⁵ This adjustment may increase or decrease the contact point area or change the properties of the interface, subsequently influencing the overall resistance of the material. For instance, the contact resistance decreases with increasing pressure in flexible composites composed of polymers and 2D nanosheets.²³⁶ Compression enhances the contact between the nanosheets, creating more conductive paths. These mechanisms are not isolated but are interrelated within the material, working collaboratively to create a piezoresistive effect.

Typically, common piezoresistive sensors rely on external stress and geometric stretching to induce lattice deformation in the material, thereby causing changes in the conductivity. Changes in the magnetic field of the sensor, induced by the piezomagnetic effect, impact the magnetic resistance and alter the conductivity. Chen et al.²³⁷ used the ferromagnetism of In₂Se₃ to realize a photodetector for visual perception processing. This involved using a magnetic field to manipulate the resistor, resulting in a responsivity of 98 mA/W and a carrier mobility of 137.6 cm² V⁻¹ s⁻¹. Another category of piezoresistive sensors involves changes in the overall resistance caused by variations in the contact surface during contact. Yang et al.²³⁸ achieved a piezoresistive bioelectronic device with a sensitivity of 461 kPa⁻¹ by adjusting the resistance of the microstructure of MXene and the plush packing. This sensor can rapidly capture human movements.

Flexible wearable sensors have become an indispensable component of human–computer interaction devices, and play important roles in e-skin, heart rate monitoring, voice recognition, and medical fields. The good chemical stability and tunable surface properties of 2D materials contribute to enhancing sensor immunity to environmental changes such as temperature and humidity. As sensor functions become increasingly integrated, signal interference in multifunctional applications and the need for more efficient energy supply methods must be addressed. The utilization of machine learning and artificial intelligence algorithms for enhanced data analysis and processing capabilities allows individuals to monitor their health status through devices such as mobile phones and computers for personalized experiences.

Flexible bioelectric sensors

Bioelectronics are compatible with the human body and can be attached to the skin or implanted to achieve sensitivity to environmental stimuli. This technology has revolutionized healthcare, bioengineering, and neuroscience, among other fields. Soft bioelectronics find wide applications in the fields of e-skin, brain–computer interfaces, pacemakers, and artificial retinas. Two-dimensional materials are already widely used in bioelectronics and show transformative potential in many ways. As functional materials, these materials can not only improve the physical properties of devices, such as conductivity, mobility, and light-response speed, but also serve as carriers for drug delivery and the detection of specific molecules. In this section, we discuss bioelectronic applications of 2D materials.

Nan et al.²¹⁹ employed PET, renowned for its exceptional high-temperature resistance and tensile properties, in conjunction with MXene to create a 3D network structure for flexible resistive sensors. In this sensor design, PEI encapsulates MXene to form conductive channels. The porous network structure of the PEI enhances its mechanical performance. Moreover, the application of external pressure induces deformation, altering the current pathway and enabling current modulation.

This sensor demonstrated ultra-high sensitivity across a broad temperature range, with a detection limit of 9 Pa and a response time of 163 ms. It also exhibited remarkable durability at room temperature, enduring 10 000 cycles at 100°C and 500 cycles at −5°C. Additionally, it responded sensitively to external mechanical stimuli at temperatures of 150°C and in liquid nitrogen. By constructing a sensor array and detecting changes in the current at the array pixel points, Nan et al.²¹⁹ managed to localize the different chess pieces (Figure 2B). The sensor was further utilized to detect swinging robotic arms and soybean pressure, among other applications. This research offers a design solution for accommodating environmental variables, such as temperature, in wearable devices.

FIGURE 2. Application scenarios for piezoresistive sensors based on polyetherimide (PET)/MXene designs. (A) Wireless transmission system for MXene-based sensor signals. A Bluetooth module is used for signal transmission in response to pressure on the sensor. (B) Use of MXene-based sensor to detect the pressure of different chess pieces and thus locate them. (C) Pressure detection during the swing of a robotic arm. (D) Utilizing the brightness of an LED to reflect changes in pressure applied to the sensor. (E) Application of the MXene-based sensor to the skin for Joule heating experiments. (F) Temperature distribution of MXene-based sensors at different voltages. (G) Infrared thermal imaging of the MXene-based sensor at increasing voltage, corresponding to the test results in (F). Reprinted with permission219 under the terms of the Creative Commons CC BY license. Copyright 2022. The Authors. Published by Wiley-VCH GmbH.

Human body temperature serves as a noninvasive predictor and offers significant insights into physiological conditions. For instance, a rise in body temperature can signal inflammation or an ongoing viral infection.^239,240 Consequently, human body temperature holds substantial research importance in the medical field.^241,242 Thermal imaging ^177,243,244 relies primarily on infrared cameras for predicting and diagnosing various diseases.^245,246 However, this method requires expensive and cumbersome equipment,²⁷ and its detection accuracy is relatively limited.²⁴⁷ Flexible electronics are expected to facilitate precise real-time thermal imaging detection, ^248,249 presenting an innovative approach for advancing thermal imaging technology.

Nevertheless, current research on thermal imaging of wearable devices mainly employs single-point imaging.^250,251 For instance, Rogers et al.²⁵⁰ developed a wearable temperature sensor capable of single-point detection that facilitates temperature detection in diverse body parts by attaching multiple sensors to different skin areas. However, this single-point wearable sensor poses challenges in detecting skin diseases²⁵² that require a large temperature gradient, such as skin bruising and tumors.

Ahn et al.¹⁰⁵ proposed an integrated flexible skin patch based on a graphene design. The patch primarily consists of a graphene-based heater, flexible printed circuit board, and capacitive temperature sensor array, as illustrated in Figure 3. The device incorporates a wireless system and rechargeable battery to enable stable temperature and hyperthermia detection over extensive areas of the human body. The monitored signal can be read directly using a mobile phone. The attributes of graphene, including its mechanical flexibility and biocompatibility, satisfy design requirements. Furthermore, the use of graphene has resulted in a significant increase in sensor temperature and thermal capacity.

FIGURE 3. Arrayed flexible graphene thermal patches for patient skin temperature and hyperthermia monitoring. (A) Illustration of the process of detecting and sensing skin temperature using graphene patches. Human skin temperature perception and auxiliary heating are achieved by integrating array-based sensor patches into medical patches. The measured temperature signal is then processed through a readout circuit and transmitted to the phone via Bluetooth to detect the skin temperature signal in real time. Additionally, the heating temperature of graphene patches can be controlled through mobile phones. (B) Photos of a graphene capacitive sensor arranged in an 8 × 8 array composition. (C) Interface structure of the sensor array utilizing graphene as an electrode. (D) Partial enlarged view of a graphene sensor. (E) Enlarged image of the flexible substrate functional area of the graphene patch in (B). Active areas include a readout front end, an analog-to-digital converter, a microcontroller, an Xtal XO, Bluetooth low energy (BLE), a DC/DC converter, and a battery. (F) Framework diagram of the wireless measurement and heating system for graphene patches. The design allows precise temperature measurement by alternating between measurement and heating states. Reprinted with permission105 under a Creative Commons Attribution NonCommercial License 4.0 (CC BY-NC). Copyright 2022. The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science.

Ahn et al.¹⁰⁵ also tested the temperature mapping ability of graphene patches (Figure 3A). The sensor reached the preset temperature in 2.1 s and returned to the original temperature after 12 s. During this process, the sensor array displayed hotspots at the temperature. Further monitoring of the physical changes caused by strenuous exercise revealed that exercise-induced sweating had a negligible effect on the sensors. The device demonstrated excellent stability during long-term monitoring experiments. The precise control and stability of graphene-integrated flexible sensors are crucial for the treatment and rehabilitation of skin diseases. This is because graphene-assisted hyperthermia can enhance blood flow and improve metabolic behavior. The integrated graphene sensor patch enables targeted and stable hyperthermia and monitoring of specific skin areas, which is expected to enhance the treatment of skin tissue damage.

This study offers a more efficient approach to transdermal diagnosis and therapy. Although the clinical application of this graphene thermal patch has not been fully validated, the comparison with the measurement data from commercial medical thermometers indicated a consistent test result of 0.01°C between the graphene patch sensor and commercial thermometer, suggesting that the graphene-based sensor has potential value for further promotion.

Piezoresistive sensors, ^253,254 achieve intuitive detection of mechanical stress by turning to electrical signals,²⁵⁵ upon stretching 2D materials.^118,256 This stretching causes alterations in the electronic structure of the energy band, which subsequently leads to changes in the electronic mobility and conductivity, thereby enabling the high-strain-sensitivity detection of pressure signals.^257,258 TMDCs such as MoS₂²³⁰ and SnS₂²⁵⁹ have been effectively utilized in sensor designs, owing to their high strain sensitivity.²⁶⁰ These piezoresistive sensors, underpinned by TMDCs, have significant potential for applications in e-skins,^261–263 and human–computer interaction.^264–266

Palladium diselenide (PdSe₂), a recently discovered 2D TMDC material with notable stability, exhibits exceptional flexibility and piezoresistive properties.^64,267 Remarkably, single-layer PdSe₂ can endure up to 17% and 19% strain tension²⁶⁸ along the x- and y-axes, respectively. Given these attributes, PdSe₂ is particularly suitable for piezoresistive sensor design. Li et al.⁹⁸ grew a large-scale PdSe₂ film directly on a PI substrate and successfully created a large-area patterned piezoresistive sensor using direct laser writing. They leveraged plasma-assisted selenization to bolster the interaction between PdSe₂ and PI, thereby enhancing the strain sensitivity. The PdSe₂ sensor could detect small physiological signals with high sensitivity, such as finger curvature and throat vibrations. After 10 000 stretches, the device performance decreased by only 0.2%, demonstrating tensile stability.

Next, they employed the PdSe₂ sensor for pulse detection.⁹⁸ Constructing a 1D deep residual network and integrating the detection process with a deep learning algorithm, the arterial temperature was detected using the relationship between the amplitude ratio of the three arterial waves and skin temperature (Figure 4A,B). Li et al.⁹⁸ conducted physiological signal detection using wearable strain sensors, enabling flexible electronic devices for medical applications.

FIGURE 4. Electronic skin, artificial retina, and electronic nose designed with machine learning algorithms. (A) Deep learning based on PdSe2 piezoresistive sensors for pulse recognition and temperature readout. Deep learning steps for converting resistance to temperatures with the input of three distinct pulse signals. The detection of pulse temperature is achieved by employing the pressure signals of pulse beats. (B) Temperature–pulse curves representing distinct pulse shapes. (C) Stabilization of training and validation losses at low values after 500 training cycles. (D) Temperature readout through deep learning with 98% accuracy. (E) MoSSe-based artificial retina with integrated sensing, storage, and computing functions. (F) Graphene-based artificial nose identifying four different volatile organic molecules. (A–D) Reprinted with permission98 under the terms of the Creative Commons Attribution License. Copyright 2023. The Authors. Published by UESTC and John Wiley & Sons Australia, Ltd. (E) Reprinted with permission.269 Copyright 2021 American Chemical Society. (F) Reprinted with permission270 under the terms of the Creative Commons CC BY license. Copyright 2023 AIP Publishing.

Flexible electronics with next-generation biocompatibility and convenience are a rapidly growing technology. To date, in addition to neuromorphic devices related to e-skin, other types of highly integrated neuromorphic circuit devices have been developed (Figure 4). The memristor ensures the artificial retinal design of sensor, memory, and computing integration. Zhang et al.²⁶⁹ designed an artificial retina based on MoSSe to simulate the visual perception of the human body. Optoelectronic properties and ionic methods were used to manipulate the conductivity of the device. Synaptic remodelability was achieved by varying the gate voltage. The retina not only collected and converted photoelectric signals but also processed and stored signals to form visual memories. Olfaction can be tested in three ways: odor threshold, discrimination, and odor recognition.

Cuniberti et al.²⁷⁰ designed a graphene-based artificial olfactory system. They extracted 11 dynamic transient features from a gas-perception curve. Through the incorporation of a supervised machine learning classification algorithm (LDA), the sensor achieved 97.5% accuracy in odor recognition. The graphene sensor demonstrated efficient handling of binary mixed odors and successfully recognized the odor of four different volatile organic molecules in an air environment.

Biocompatibility must also be considered when developing wearable sensors. Combining the sensing and therapeutic functions of flexible wearable sensors provides a promising solution for higher precision treatment. In this section, we discuss several 2D materials and their roles in bioelectronic devices, as detailed in Table 4.

TABLE 4 Roles of 2D materials in bioelectronic devices and their features.

Structures	Roles	Features	Reference
MXene/hydrogel	Conductive electrodes	Repairing of muscle and nerve tissue	271
MXene/hydrogel	Scaffolds	Cochlear hair cell regeneration	272
MXene/PDMS	Patches	Artificial eardrum for sound detection and recognition	273
MXene/cellulose	Electrodes	Electrotherapy and hyperthermia that are applied to muscle tissue	274
MXene/PVDF	Triboelectric	Energy harvesting and tactile sensing	275
MXene/CNTs/enzyme	Supercapacitors	Converts lactic acid in sweat into bioelectricity	276
Graphene oxide	Patches	Eardrum repairing	277
Graphene/parylene-C	Phototransistors arrays	Contact lenses mimicking the retina	278
Graphene	Resistors for heating	Heating film for the treatment of Alzheimer's disease	279
Graphene/silk fibroin/Ca2+	Transistor arrays	Electronic tattoos for humidity, temperature, and motion detection	280
Graphene/hydrogel	Scaffolds	Soft actuators	281
Graphene/graphene oxide/polydopamine	Electrode arrays	Actuators and conformal patches for cardiac repair	282
MoS2/graphene	Phototransistor arrays	High-pixel flexible artificial retina	283
MoS2/PI	Photodetector arrays	Artificial retina	77
MoS2/MXene	Gas sensors	Artificial olfaction	284

Designs that combine conductive 2D materials as fillers with elastic polymers achieve mechanical compatibility between devices and soft tissues, thereby maximizing the flexibility of bioelectronics devices. For example, repairing eardrums and extracting and utilizing metabolic waste from the human body can be achieved through thin films with mechanical toughness. Flexible wearable sensors made of 2D materials have been widely used as a new type of portable electronic device in the fields of human–computer interaction, medical detection, and the Internet of Things.

Smart displays

Flexible display screens have been widely used in mobile devices such as smartphones, computers, and smart bracelets. The continuous maturity of the large-area and high-quality 2D material thin-film preparation process has enabled it to stand out in commercial display screen manufacturing. The substitution of traditional conductive materials such as indium tin oxide (ITO) with 2D materials has emerged as a new development direction.

The photocurrent output can be used as a display for information security applications. Image encrypters are capable of executing encryption procedures during image acquisition.^285,286 These encrypters typically employ encryption algorithms and key pairs to cipher content and merge the initial image data with keys to generate encrypted images. The encrypted images are then transmitted to the key authorizer via a secure channel.^287–289 Given the advancements in machine learning²⁹⁰ and quantum technology, the data transfer process between sensors and computing units is not as secure. Hardware cryptography, which provides keys, is believed to be one solution.²⁹¹ This strategy offers more concise design and randomness than software computing. However, the security key is at risk of duplication because the existing hardware security module is physically separated from the sensor.

2D materials are advantageous for image encryption applications, due to their large surface area,^292–294 high volume ratio,^295–297 high carrier mobility,^298–300 and tunable electronic properties.^301–304 Encryption within sensors using 2D materials allows the key to bind to the captured image, resulting in highly trusted encryption.

Chai et al.³⁰⁵ developed an array sensor with 256 phototransistors using MoS₂ to capture images and simultaneously generate a trusted cryptographic key. The uncontrollable plane disorder, resulting from the large-area MoS₂ uncontrollable growth process, was harnessed to enhance key security. A transfer curve was used to determine the threshold voltage, using 1 V as a reference. Based on the actual threshold voltage and its relative magnitude, the 256 transistor devices could be differentiated into two digital signals: 0 and 1 (Figure 5). Moreover, the lighting conditions affected the leakage current of the MoS₂ transistor. Considering the leakage current as a function of light stimulus (as an additional entropy source) and 18 μA as the reference threshold current, the 256 transistor devices could also be differentiated into two states: on (1) and off (0). Two binary system keys were realized through two physically unclonable functions (PUFs) of the threshold voltage and current. By combining these two binaries, signals 00, 01, 10, and 11 were produced, thereby realizing a system design of quaternary 0, 1, 2, and 3.

FIGURE 5. (A–C) MoS2-based field-effect transistor employed for information encryption. (D) Image captured by the MoS2 image sensor. (E) Transfer characteristics of the MoS2 transistor, determining binary values based on the intercept of the linear fitting. (F) Histogram of the gate-voltage intercept, with blue denoting 0 and red denoting 1; reference gate voltage is −1 V. (G) MoS2 transistor leakage current curve in the on/off state. (H) Histogram of drain current for binary data: 0 when ID [less than]18 μA and 1 when ID >18 μA. (I) PUF pattern. (J) Sequential steps of image encryption and decryption: the image, captured by the sensor, is encrypted with a PUF key and is subsequently decrypted with the corresponding key. Reprinted with permission.305 Copyright 2023 American Chemical Society.

Owing to the excellent randomness and uniqueness of the MoS₂ transistor sensor, its encryption key is highly reliable. In the image decryption process,³⁰⁵ the key used in the encryption process becomes imperative. An image can be encrypted using a photobinary PUF (Figure 5) and is subsequently decrypted using the same security key. The Vernam algorithm is then employed to conduct a quaternary multilevel encryption (Figure 5), further enhancing the high-quality randomness of the encrypted image.

An encrypted image sensor, based on a MoS₂ array design, can efficiently prevent key cloning. The 2D material image encrypter is advancing toward higher integration and miniaturization, incorporating additional functions, including image processing and data storage, for various scenarios.

Two-dimensional materials, such as graphene and WSe₂ are widely employed in information encryption with MoS₂ being a notable exception. Ye et al.³⁰⁶ designed a graphene-based PUF generated from two independent stochastic processes. They used O₂ plasma to improve the random localization of the graphene islands. Owing to the different numbers of graphene island layers, Raman mapping could produce a four-color image with encoding capabilities. The color image was then processed into a grayscale map to generate a 256-bit binary key. The Sobel and ORB key-point detection algorithms were used to extract the matching Raman image features. A database of 30 Raman image color layers was created to store key points and feature codes to verify the authenticity of the image. Even after 6 months of storage, the image analysis demonstrated a 93.6% match for the binary key. This graphene PUF contributes to enhancing security and anti-counterfeiting technologies.

Miao et al.³⁰⁷ fabricated WSe₂/h-BN/Al₂O₃ heterojunction non-amnesiac transistors. The key function in these transistors is implemented through gate-voltage regulation. The baseline current level of the device is changed by controlling the number of programmed optical and back gate electrical pulses. The concentration and type of channel carriers can be continuously or discretely adjusted by varying the magnitude and polarity of the applied gate voltage. Control of the source and drain currents is realized through coordinated control of the pulse and gate voltages, and the gate voltage is read to achieve logical encryption of “1” or “0.”

Non-contact touchscreens

Non-contact touchscreens have become significant innovations. The rapid rise of artificial intelligence and the Internet of Things³⁰⁸ has fostered significant advancements in human–robot interaction technology, fundamentally altering how humans interact^309,310 with their environments.^311,312 Traditional interfaces that require direct contact are prevalent in the interaction between electrical devices and systems but inherently have limitations.^313,314 Consequently, contactless interaction has garnered considerable attention in human–computer interaction, owing to its advantages^315–318 of convenience, sterility, and injury prevention. Touchscreens detect finger touches by sensing changes in capacitance caused by finger contact with the screen. Continuously monitoring these changes allows for the conversion of finger touch points into precise coordinates. Nonetheless, current non-contact interaction methods, ^313,319 including computer vision sensors, face disadvantages such as limited sensing distance and lack of robustness, which makes them unsuitable for practical and complex environments.

Shen et al.³²⁰ pioneered an integrated optical non-contact control system (ONCS) sensor using a PtTe_x/Si heterojunction photomultiplier transistor array. This innovative non-contact control system leverages the detection of elementary shadows to perceive distant objects. The photocurrent of the heterojunction responds to shadow changes. The ratio of the photocurrent change versus the time difference, defined as the photocurrent gradient, forms the basis for the digital output; a ratio exceeding the predefined threshold results in a digital output of 0, otherwise, the output is 1 (Figure 6D). The sensor can detect a wide range of visible and infrared wavelengths, enabling the perception of objects up to 0.15 m away. Non-contact operation of the sensor is achieved by converting the photocurrent changes caused by the umbra and penumbra into spatiotemporal sequence control commands. Furthermore, the non-contact control system exhibits robustness, thereby facilitating function selection, touch-free screen positioning, and real-time signal display.

FIGURE 6. Optical non-contact control system based on a PtTex/Si sensor array for human–machine interaction. (A) Scheme of a photomultiplier transistor. (B) Optical image of the sensor array. (C) Flowchart depicting the process of shadow encoding and recognition. Converting photocurrent into a discrete signal enables instruction retrieval. (D) Process of encoding a photocurrent signal using the gradient approach. (E) Output displaying the encoding of shadows. Reprinted with permission320 under the terms of the Creative Commons Attribution License. Copyright 2021. The Authors. Published by UESTC and John Wiley & Sons Australia, Ltd.

Shen et al.³²⁰ subsequently demonstrated the contactless control capabilities of the PtTe_x/Si heterojunction-based ONCS. The system can perform non-contact screen positioning and provide real-time gesture signal displays under varying shadow conditions. It detects shadows at different interference intensities, showcasing robust performance. The PtTe_x/Si heterojunction contactless sensor unveils the potential for future generations of highly integrated human–machine interface wearable electronics.

Advanced artificial vision systems^321–325 have the potential to propel the development of forefront applications such as smart homes, autonomous vehicles, ^321,326 and video content analytics. However, the design of such systems often necessitates multifaceted considerations.^327–329 A system is required to acquire high-quality image data,^330–332 which are subsequently preprocessed to eliminate noise and augment contrast, among other factors. Techniques such as machine learning,^{326,333–335} and training³³⁶ have been employed to train and optimize the acquired image features.³³⁷ Furthermore, the system must make decisions and output data based on the optimized images. Consequently, there is a pressing need for the design of artificial vision systems utilizing multifunctional and highly integrated image sensors to perceive, store, and process visual information.^{54,163,332,338}

Despite its significant application potentials, ^77,339 retina-inspired vision chips using Si CMOS technology involves complex biological vision systems, leading to substantial energy consumption.^340,341 Therefore, innovations in Si CMOS technology, such as low-power design technology using 2D material van der Waals heterojunction, are needed to enhance chip integration and address the challenges faced by retina-inspired vision chips.^{269,342–344}

Xu et al.³⁴⁰ introduced a neurally inspired optical sensor based on a NbS₂/MoS₂ heterojunction designed to cater to the requirements of neurally inspired vision systems for detecting and recognizing moving objects. The sensor, equipped with a 100-pixel photoelectric sensing array, utilizes the tunable charge-transfer capability of the effective gate at the heterojunction interface, offering tunable photoconductivity and facilitating multiple low-level sensory processing steps. Signal measurement is achieved by connecting the sensing unit to the corresponding digital circuit amplifier (Figure 7). The sensor amalgamates numerous functions including image sensing, storage, and contrast enhancement.

FIGURE 7. NbS2/MoS2-based neurally inspired optical sensor array for high-precision dynamic image recognition and single-point motion trajectory extraction. (A) Schematic diagram of the structure of the NbS2/MoS2-based vision system, which consists of a 100-pixel NbS2/MoS2 optical sensing array. (B) Scheme and circuit diagram of the NbS2/MoS2 optical sensing array. (C) Optical micrograph of a 100-pixel sensor array and scheme of a NbS2/MoS2 phototransistor. Reprinted with permission340 under a Creative Commons Attribution 4.0 International License. Copyright 2023. The Authors. Published by Springer Nature.

In addition, the sensor³⁴⁰ exhibits light-intensity dependence and time-resolution characteristics, empowering the NbS₂/MoS₂-based neurally inspired optical sensor to identify and track the trajectory of moving light spots. The sensor captures information on the 2D plane motion of the spot by differentiating the point outputs in the spatiotemporal dimension, which enables the restoration of the moving spot trajectory. This neurally inspired optical sensor, based on NbS₂/MoS₂, circumvents the data expansion problem encountered in image pixel processing of silicon-based CMOS chips. It simplifies visual system information processing by directly generating the detected spatiotemporal information. The neurally inspired optical sensor featuring the NbS₂/MoS₂ heterojunction offers a promising design for achieving high-performance artificial vision systems.

A touchscreen is inherently linked with a display. Capacitive and infrared touchscreens have a wide range of practical applications because of their high sensitivity and low manufacturing costs. ITO-based capacitive touchscreens are suitable for applications in rigid electronic devices; however, the brittleness of ITO limits its use in flexible electronics. Silver nanowires offer an alternative, owing to their high performance; however, they are difficult to manufacture at a low cost.

Printed electronics based on 2D materials have been used as a solution for achieving flexible touchscreens. Printed transistors based on 2D materials can be used to construct high-resolution touchscreens. Such 2D material-based transistors enable higher pixel densities and wider color ranges in display technologies, resulting in sharper and more vivid images and videos. Further advancements in developing 2D materials are essential, particularly in designing transparent conductive films to achieve high conductivity and light transmittance, while considering the resistance heat generation during usage.

Flexible energy storage

This section introduces flexible energy-storage methods using 2D materials, encompassing energy storage devices such as batteries and supercapacitors. Fibers^345,346 and textiles are ideal platforms for wearable electronics that are adaptable to various deformations including stretching and bending.³⁴⁷ These materials can be integrated into sensors, power sources,^348–350 and data processors.³⁵¹ An efficient energy storage system is crucial to ensure the stable operation and extended lifespan of wearable devices.^352–354 Fiber-based lithium-ion batteries³⁵⁵ and supercapacitors, ^356,357 especially hybrid supercapacitors developed through optimized metal-ion anodes, have garnered attention owing to their high energy densities and long lifespans.

The electrode materials in most hybrid FSCs are composed of composite materials, such as graphene and metal oxides, known for their high conductivity, good flexibility, and mechanical strength.³⁵³ For example, Lee et al.³⁵⁸ designed MXene-based FSCs that exhibit an energy density of 86.72 mWh cm⁻³ at a power density of 480.30 mW cm⁻³. Even after 20 000 charge–discharge cycles, the capacitance retention rate was 80%. Li et al.³⁵⁹ designed porous MXene-based FSCs with interfacial cross-linked structures, achieving an energy density of 77.6 mWh cm⁻³ at a power density of 401.9 mW cm⁻³. In this section, we evaluate the performance of supercapacitors based on 2D materials, as summarized in Table 5.

TABLE 5 Two-dimensional material-based flexible solid-state supercapacitors.

2D material types	Capacitance	Power density	Energy density	Rate range (mV/s)	Cycle number	Capacitance retention (%)	Reference
Graphene oxide	197.3 F cm−3 @500 mA cm−3			50–2000	7500	94.7	360
Graphene/hydrogel	205.2 mF cm−2	0.855 mW cm−2	48.2 μWh cm−2	5–100	1000	98.2	361
Graphene	160 F cm−3	3.4 W cm−3	97.9 mW h cm−3		25 000	98.1	362
MXene	1719 mF cm−2 @100 mA cm−2	502 μW cm−2	126 μWh cm−2	5–1000	10 000	99.8	363
MXene/CNT	550 F g−1	50 W kg−1	7.34 Wh kg−1	2–100	5000	99	364
MXene/Fiber	199.5 F cm−3 @0.1 A cm−3	1186.9 mW cm−3	17.7 mWh cm−3	5–200	6000	100	365
MXene/Zn–Co MOFs	144.7 F g−1 @1 A g−1			5–200	8000	80.78	366
MXene/W18O49	101 F g−1 @1 A g−1	900 W kg−1	45.4 Wh kg−1		10 000	85.4	367
MoS2/Co	1884.36 mF cm−2 @1 mA cm−2	168.3 mW g−1	102.7 mW h g−1		10 000	98.7	368

Hybrid FSCs have a higher energy density than conventional supercapacitors, which is attributed to the use of materials with high specific capacitance.²²⁵ In addition, their capability for rapid charging and discharging gives hybrid FSCs great application value in the field of fast-energy pulses.³⁶⁹

Flexible solid-state lithium-metal batteries^370,371 utilize solid-state electrolytes that are more chemically stable,³⁷² enabling them to achieve high energy density and specific capacity.³⁷³ Moreover, they can be combined with fibers to achieve a flexible design, offering great potential in wearable textiles.³⁷⁴ Rettenwander et al.³⁷⁵ used MHz pulse currents to reduce the critical current density and achieved an operating current density of 6.5 mA cm⁻². Gong et al.³⁷⁶ employed 2D fluorographene to enhance the performance of polymer electrolytes, to construct a solid-state lithium battery capable of stable cycling at 1°C with a coulombic efficiency of 99.5%. Certainly, the challenge of short circuits caused by the determination of flexible solid-state Li-metal batteries during charging and discharging must be resolved.^370,371 In this section, we evaluate the performance of flexible solid-state lithium-metal batteries based on 2D materials, as summarized in Table 6.

TABLE 6 Flexible solid-state lithium batteries from two-dimensional materials.

2D material types	Specific capacity (mAh g−1)	Conductivity (mS cm−1)	Cycle number	Capacitance retention (%)	Reference
Graphene oxide	120.2	0.22	250	84.6	377
Graphene	136.6	0.81	600	86.3	378
Graphene/MnO	860	8.5 Ω (Warburg impedance)	10 000	77.8	379
MXene/PAN	101	3.07 at room temperature	500	85.18	380
MXene/PEO	140	2.2 × 10−4 at 28°C	100	91.4	381
MXene/SiO2	141.8	0.46 at 20°C	250	88	382
MXene/PAN	131	0.217	2500 h	90	383
3D MoS2	1111 @ 10 A g−1	46.6	1000	>100	384

In addition, zinc-ion-based fiber-optic supercapacitors have significant application prospects, owing to their high theoretical energy density, excellent safety, environmental-friendliness, low cost, and abundant resources.

Shen et al.⁹⁹ designed a wearable zinc-ion hybrid FSC based on MXene material. In their design, MXene is uniformly coated onto the fiber as the positive electrode material using a dynamic multiple immersion method. Subsequently, with MXene as the core (cathode), the fiber is used as the shell (zinc anode), passed through the solid electrolyte, and woven on the MXene fiber surface. This woven structure improved the specific capacity and achieved an area capacitance of up to 214 mF/cm² and an energy density of 42.8 μWh cm⁻² at a scan rate of 5 mV/s. Even after 5000 charge and discharge cycles, 83.58% of the capacitance was retained.

Using the MXene coaxial capacitors, Shen et al.⁹⁹ also implemented a series and parallel design to widen the voltage window and enhance the energy density. By adjusting the output voltage and capacitance, the energy storage device could be better tailored to meet energy and power demands. The research team weaved a bracelet using a 1.5 m coaxial-structured FSC, enabling the watch to charge (Figure 8).

FIGURE 8. Supercapacitor woven bracelet based on the MXene coaxial structure for charging a watch. (A) CV curves of a single zinc-ion hybrid fiber supercapacitor (FSC) and two supercapacitors connected in series and parallel, respectively. (B) Plots of capacitance and energy of the supercapacitors versus length. (C) Bracelet weaving by consuming a 1.5 m coaxial FSC. The bracelet provides electric power for a watch and LEDs in a glove. Reprinted with permission99 under a Creative Commons Attribution 4.0 International License. Copyright 2021. The Authors. Published by Springer Nature.

Flexible energy conversion by nanogenerators

Flexible energy conversion mainly refers to nanogenerators. The energy conversion capacity of TENGs can be quantified and improved using multiple parameters. Although the energy conversion capacity of TENGs is limited by many factors such as dielectric loss, material wear,³⁸⁵ and discharge, the material can be improved by selecting materials with a high charge density,³⁸⁶ optimizing the surface structure, and increasing the contact area of the materials.^387–390 Notably, 2D materials such as graphene, MXene, and TMDCs have a high specific surface area, translating to a higher charge density during contact. Modifying the surface of 2D materials, improves the interfacial interaction, enhances the triboelectric effect, and enables coupling of multiple energy-harvesting mechanisms. In this section, we discuss the effect of 2D materials on the performance of TENGs, as summarized in Table 7.

TABLE 7 Comparison of the performance of TENGs before and after the incorporation of 2D materials.

Triboelectric materials		Performances
Without 2D materials	With 2D materials	Without 2D materials	With 2D materials	Improvement (%)	Reference
Nylon-12 and Ecoflex	Nylon-12/r-GO and Ecoflex/MoS2	Potentials of +350 V and −433 V.	Potentials of +903 V and −1352 V	258% and 352%	391
rGO/Nylon/PI	rGO/Nylon/PI/LIG	Charge density of 160 μC m−2	Charge density of 270 μC m−2	169%	392
Cu/Graphene/Acrylic substrate/Si	Cu/Graphene/Acrylic substrate/Si/Graphene oxide	Current output, 3 μA	Current output, 30 μA	Current output, 10 folds	393
		Retention ratio, 21%	Retention ratio, 95%	Retention ratio, 5 folds
PVDF/PA6	PVDF/PA6/Graphene	Peak power density, 78.2 W m−2	Peak power density, 130.2 W m−2	2 times	394
PVDF/Fiber	PVDF/Fiber/MXene	Circuit voltage of 15.2 V	Circuit voltage of 23.2 V	153%	395

Bae et al.³⁹¹ introduced an electrospun amino-functionalized reduced graphene oxide (A-rGO) combined with Nylon-12, serving as a high-friction cathode layer. They combined micro-patterned MoS₂ with Ecoflex as a friction anode layer. By embedding A-rGO and MoS₂ nanostructures in nylon and Ecoflex, the positive and negative surface potentials of TENG were significantly improved, with the potential of the positive electrode layer increasing from +350 V to +903 V, a factor of approximately 2.5, while the potential of the negative electrode layer increased from −433 V to −1352 V. Li et al.³⁹² constructed a self-powered pressure sensor based on reduced graphene oxide. They increased the transfer charge density of TENG from 160 to 270 μC m⁻² by introducing laser-induced graphene (LIG).

TENGs can also be used as sensors, finding application in various scenarios within flexible electronics, include health monitoring,^396–398 sports tracking,^399–402 smartwatches, and smart clothing.^403,404 These applications require the devices to maintain stability and high sensitivity across varied temperature environments.⁴⁰⁵ This is particularly crucial considering the fluctuating variables, such as temperature and humidity, in different scenarios including outdoor sports,⁴⁰⁶ extreme weather conditions,⁴⁰⁷ and medical monitoring.⁴⁰⁸ Therefore, ensuring effective device operation under these conditions is imperative. TENGs can convert external physical signals, such as pressure and vibration, into electrical signals. By measuring the changes in electrical signals, they can accurately reflect the nature and intensity of the stimuli. Thus, TENGs have evolved into multifunctional and flexible intelligence sensing platforms.

Self-powering systems include TENGs, supercapacitors, and sensors. Owing to their dual coupling effect of triboelectric and electrostatic induction,^409–412 TENGs hold immense potential for self-powered sensing devices.^413–416 In particular, they can convert power generated by various mechanical movements into electrical energy.^417–420 Wang et al.⁴²¹ constructed a TENG sensor for a water-level alarm capable of harnessing the energy from water waves, which yielded an output current of 15 mA and an output power of 24 mW. Shi et al.⁴²² designed a TENG sensor for oil–solid contact using PTFE and PI, featuring a charge density of 9.1 μC m⁻² and a power density of 1.23 mW m⁻², to detect contaminants in oil. TENGs can be integrated into lightweight and flexible clothing and wearable devices. When integrated with energy storage devices, such as supercapacitors, TENG-based devices can achieve stable self-power supply.⁴²³ Leveraging the self-powered capability of TENG^424–426 enables the powering of devices integrated into non-contact sensing devices.^427–429

Song et al.²²⁷ employed MXene-modified carbon nanofibers to prepare ink and developed a non-contact nanogenerator using a deep-trap layered structure and 3D printing technology. Unlike traditional designs that augment frictional charges through ion injection⁴³⁰ or dielectric-constant adjustment,⁴³¹ this CNF/MXene-based TENG controls the device output voltage via an electrostatic potential. The excellent electron-trapping ability of MXene is harnessed to achieve the long-term capture of induced charges, thereby enhancing the device output voltage. Under non-contact conditions, the charge density of this TENG²²⁷ is 900 mC cm⁻³, and after 900 s, it decreases to 600 mC cm⁻³. Even after 3000 cycles, the TENG exhibits a stable voltage output.

Furthermore, Song et al.²²⁷ assembled a single-electrode-mode non-contact TENG. The device achieves electrostatic equilibrium through triboelectric electrostatic charge shielding when a finger approaches or recedes from the device, causing the positive and negative charges within the device to vary. The TENG generates two contrasting output signals as the finger approaches or moves away. Maintaining a stable signal output after 3000 cycles of periodic approach and separation, this TENG facilitated the development of a pixelated TENG sensor array to capture the motion trajectory of a finger positioned above it (Figure 9A). By monitoring the voltage changes during finger movement, the height difference between the finger and the sensor can be derived (Figure 9B,C). The TENG sensor array illustrates its potential applications in human–machine interactions and medical devices.

FIGURE 9. Three-dimensional motion detection using an MXene-based TENG. (A) Scheme of a 4 × 4 TENG sensor array. The motion trajectory of a finger positioned above can be captured using the sensor array. (B) Perception of finger linear motion above the sensor. (C) Capture of the finger motion trajectory when moving in a curved path. (D) Blind navigation by installing the sensor on a walking cane. (E) Photograph of the non-contact sensor. (F) Voltage signal output when the finger moves above the six planes (A–F). Reprinted with permission.227 Copyright 2022 Elsevier Ltd.

Voice is a highly effective means of communication that is extensively employed in human-computer interface design.^432–434 Machine learning to extract and train speech features facilitates the efficient recognition of different sound characteristics.^435–437 The current focal point of interest lies in the development of acoustic devices with rich functionality or enhanced flexibility and wearability.^438,439 Triboelectric devices can also be used as transducers. For example, TENGs can convert acoustic signals into electrical signals for sound recognition and feedback. Triboelectric acoustic sensors enable efficient and natural human–machine interface design. Combining this transducer design with machine learning addresses the operating frequency limitations of traditional loudspeakers, enabling more accurate capture and emulation of the human voice.^440,441 Devices utilizing thermoacoustic effects emerge as potential materials for multifunctional speakers, owing to their wide operating frequency ranges and flexibility.^435,436 However, further research is required to explore the energy conversion capabilities of TENGs in acoustic equipment, aiming for integrated multifunctionality and more complex artificial intelligence in human–robot interactions.^442–444

Ren et al.¹⁰⁰ detailed a graphene-based dual-function acoustic transducer for machine-learning-assisted human–robot interfaces (GHRI) (Figure 10).

FIGURE 10. Graphene human–robot interfaces empowered by machine learning based on graphene acoustic transducers. (A) Auditory and vocal capabilities of the robot system empowered by the system. After training using convolutional neural networks, it can recognize different identities and emotional characteristics, allowing intelligent communication and responses. (B) Schematic representation of the graphene/PI/graphene structure formed through laser irradiation. (C) SEM image of graphene. (D) Human–robot interfaces attached to a robot. (E and F) TENG operating in microphone mode, detecting sound vibrations through surface-charge changes. (G) TENG in loudspeaker mode, generating acoustic waves through the thermoacoustic effect. Reprinted with permission100 under the terms of the Creative Commons Attribution License. Copyright 2022. The Authors. Published by UESTC and John Wiley & Sons Australia, Ltd.

The system exhibits dual functionality: First, it functions as an artificial ear via the piezoelectric effect of triboelectric acoustic sensing, and second, it serves as an artificial mouth through the thermoacoustic emission mechanism. In the microphone mode (Figure 10E,F), one-side-carved LIG with a PI film serves as the negatively charged triboelectric material of the TENG, attracting charges with the graphene backside acting as the electrode. In double-side-carved LIG, the bottom layer serves as the electrode and the positive material for the TENG. In speaker mode, the upper layer of graphene acts as a thermoacoustic source. Through the thermoacoustic effect (Figure 10G), graphene converts the input alternating electrical energy into periodic Joule heating. This process generates sound waves by causing air expansion and contraction due to changes in air pressure. Thus, electrical energy is effectively converted to sound energy.

Ren et al.¹⁰⁰ conducted sound tests on individuals of varying genders and moods using graphene-based human–robot interfaces. The results revealed differences in the frequency domain, time domain, and voice frequencies among individuals with varying moods. Leveraging these differences, convolutional neural network machine learning was used for precise semantic recognition, enabling the GHRI to respond intelligently to conversations through feature extraction.

Transistors and logic circuits with 2D materials

The scaling of silicon-based integrated circuits is encountering challenges,^445–448 and 2D materials have emerged as a fresh opportunity for non-silicon materials in advanced processes. ^39,69,449 The realization of transistor-channel transduction is made possible by processing 2D materials that are compatible with silicon-based chips.^39,450,451 Research on the design of advanced electronic circuits for 2D materials primarily targets nonfunctional SiO₂ substrate devices.^31,452,453 Nonetheless, some studies have integrated a single layer of graphene into silicon-based chips.⁴⁵⁴ Moreover, by developing a low-temperature growth process, MoS₂ can be grown directly on silicon CMOS chips.¹⁴⁰

Using multilayer h-BN for interconnections with silicon CMOS chips, Lanza et al.²¹ constructed a highly integrated memristor using patterning. The silicon CMOS chip is located at the 180 nm node. The multi-layer h-BN is resistant to damage during transfer, which enhances the yield of devices and circuits. By investigating the performance of standalone electronic memories, they established that the probability of locating defect clusters in small devices was low, leading to an increased dielectric breakdown voltage. Owing to current fluctuations and dielectric breakdown, the stand-alone device has a durability of only 100 cycles. However, by modulating the voltage of the CMOS transistor in one-transistor-one-memristor cells, they achieved precise control of the memristor current and a stability of 2.5 million cycles.

Lanza et al.²¹ employed the memristor for memory calculations, successfully demonstrating proof-of-concept calculations for “or” and “implication.” Subsequently, by modifying the device interconnection, they constructed a low-energy spiking neural network (SNN) electronic synapse based on nonvolatile resistive switching. This synapse comprised 784 input neurons, an excitatory layer composed of 400 neurons, an inhibitory layer consisting of 400 neurons, and plates for decision-making. The model was trained and tested using images from a Modified National Institute of Standards and Technology (MNIST) database (Figure 11). The benchmark results revealed that the SNN achieved an average accuracy of 90% after 50 Monte Carlo simulations.

FIGURE 11. Spiking neural network (SNN) structure based on a hybrid 2D-CMOS microchip. (A) SNN structure. The image from the Modified National Institute of Standards and Technology (MNIST) database is edited into a column vector with 784 input neurons. Pixel intensity is encoded by the firing pattern of the input neurons. Unsupervised training of neurons connecting the input and excitation layers results in labeled trained neurons. These, together with firing patterns, are transmitted to the decision block for feedback, allowing inference of the presented images. (B) Synaptic connection evolution training conducted on 400 excitatory and 400 inhibitory layer neurons. (C) Obfuscation matrix, which provides a visual representation of dataset accuracy. (D) 50 Monte Carlo simulations of 400 excitatory and 400 inhibitory layer neurons of SNN. After 50 iterations, the system accuracy reached 90%. (E) Schematic diagram of the neuron–synapse–neuron module circuit design based on h-BN. (F) SPICE-like simulation of synaptic signals from one-transistor-one-memristor cells. (G) Neuronal membrane potential simulated by SPICE simulation. Reprinted with permission21 under a Creative Commons Attribution 4.0 International License. Copyright 2023. The Authors. Published by Springer Nature.

Although the commercial application of 2D materials in high-density integrated microelectronic products faces limitations due to the local defects^455,456 in 2D materials, the results demonstrate the excellent performance of high-density integrated hybrid 2D-CMOS microchips with low power consumption.

The advancement of mobile devices and the Internet of Things necessitates a reduction in the size and power consumption of electronic components.^457–459 Memristors emerge as a promising solution, enabling enhanced circuit integration and lower power consumption.^460,461 Memristors, which serve as nonlinear passive bipolar elements, offer versatile electrical properties that can be utilized in various analog and digital computations within circuits.^461–463 However, the application of algorithms to memristor-based circuits is currently restricted, primarily owing to the significant differences between memristors and traditional electronic components.^464,465 Therefore, to realize memristor-based circuit designs, it is imperative to develop new algorithms or refine existing ones to align with the characteristics of memristors.^465,466

Cellular automata⁴⁶⁷ exhibit a capacity for multitask parallel computing, thereby significantly improving the computational speed. The computational rules of cellular automata are highly flexible and offer excellent fault tolerance. Implementing cellular automata in memristors presents an adaptive advantage, enabling the dynamic optimization of circuit performance. However, the implementation of cellular automata on traditional silicon-based chips is hampered by low parallelism and high hardware costs, which pose significant challenges.

Ren et al.⁴⁶⁸ introduced a circular logic operation scheme that utilizes a 2D transistor for optimizing cellular automata. They employed a cross-cutting structure, compatible with the transformation rules of cellular automata, to perform in-memory logic operations through multiplexing. However, this cross-structure computation necessitates that the input and output memristors be positioned within the same row and column, leading to superfluous calculations, and thus, increased hardware and power requirements. In contrast, the cyclic logic computation scheme aligns more effectively with the computational rules of the cell computer. In this scheme, the state and inverted state of the cell computer correlate with the resistance states of the two memristor circuits, and the states of its adjacent cells determine the state of each cell through a cyclic logic computation scheme (Figure 12B). The input signal is produced by employing Boolean logic formula operations and is divided into two parts—calculation and write modes—through mapping relations. The state calculation of the cell automaton is realized in calculation mode, whereas the storage mode is realized in write mode. This circular logic computation scheme allows the integration of data storage and computation within the device without requiring a data-reading step.

FIGURE 12. Comparison between the traditional cross-computing structure and the cyclic logic computing scheme. (A) Cross-operation structure. The input and output memristors are in the same row and column. (B) Circular logic computational structure. The state and inverted state of the cell computer are correlated with the resistance state of the two circuits of the memristor, and the state of its neighboring cells determines the state of each cell through a cyclic logic calculation scheme. (C) Schematic diagram of the design of the memristor array to implement the cyclic logic operation scheme. (D) Design diagram of the cyclic logic calculation scheme. The mapping scheme divides the input signal into two modes: calculation and writing. The state calculation of the cellular automaton is realized through the calculation mode, whereas the cell automaton storage mode is realized through the write mode. Reprinted with permission468 under a Creative Commons Attribution 4.0 International License. Copyright 2023. The Authors. Published by Springer Nature.

Additionally, Ren et al.⁴⁶⁸ used this circular logic operation scheme to design a basic logic circuit consisting of a memristor that integrates storage and computation, a MoS₂ transistor that isolates crosstalk from cell computations, and a resistor that serves as an input and output signal. The device achieved a resistance on/off ratio of 10⁵ to meet the basic requirements for circuit operation. Using the pulse signal applied to the corresponding node, simple NAND or AND logic operations can be performed using a memristor (Figure 13C,D). A and B are used as the input signals, Y and Y′ are the output signals, and X is the initial value of Y′. It is then possible to implement NAND operations for Y′ = AB and Y′ = ABX, or AND operations for Y = AB and Y = ABX. The group implemented the design of a symmetry-breaking small-transistor memristor circuit to eliminate the influence of crosstalk in cyclic logic computing schemes.

FIGURE 13. In-memory computing design for a hybrid logic circuit with a MoS2-based transistor and memristor. (A and B) I–V curves for memristors and MoS2-based transistors, respectively. (C and D) Simple NAND and AND logic operation verification using a memristor hybrid circuit. (E and F) Measurements of both NAND and AND logic operations. (G and H) Measurement of voltage deviation by 100 simulations using Vdd and VR. Reprinted with permission468 under a Creative Commons Attribution 4.0 International License. Copyright 2023. The Authors. Published by Springer Nature.

After establishing the basic logic operations, they progressed to a complex circuit design of the cyclic logic operation of the 1D cell automaton.⁴⁶⁸ In this logic, the neighborhood size of each cellular automaton is n = 2r + 1, where r is an integer. By defining two states S = (0.1) for a single-cell automaton, 256 logical rules were realized. These parameters were calculated separately for each cell and operated in parallel. The 110 rule was utilized to construct an equivalent Turing machine for operational demonstration, and most of the classification and edge detection algorithms were validated. Compared with field-programmable gate arrays, the hardware cost of the MoS₂-based cyclic logic computation scheme is reduced by a factor of 79 (Figure 14).

FIGURE 14. Demonstration of circular logic solutions for 1D cellular and basic cellular automata. (A) Optical image of a 1D cell automaton. (B) Circuit diagram of a 1D cell automaton. Green dotted coil indicates a basic unit. Each basic unit consists of a memory resistor and an auxiliary memory resistor. CAx and CAx′ represent the resistance value and resistance inverse value of the cellular automaton, respectively. (C) 110 logical operation of ECA rules that describe the corresponding operation in a circular logic operation. (D) Time series in which the 110 operation triggers the signal. (E) Evolution of memory resistor states under different conversion rules. Reprinted with permission468 under a Creative Commons Attribution 4.0 International License. Copyright 2023. The Authors. Published by Springer Nature.

Introducing a circular logic operation scheme to realize cell automata⁴⁶⁸ improves computational efficiency and reduces the hardware cost. This advances the potential of computational systems that use simple and highly parallel cellular automata combined with membrane resistors with special in-memory computing properties. The proposed design serves as a blueprint for implementing cellular automata, driving the development of edge-computing hardware devices.

Memory based on neuromorphic engineering

Electronic synapses^469–472 and neurons^473–476 are the basic building blocks of artificial and spiking neural networks.^477,478 However, challenges in achieving an efficient match between synapses and neurons, ^479,480 such as differences in materials and processes, have led to the development of memristor networks with reconfigurable capabilities.⁴⁸¹ E-textiles have an interwoven structure, which can effectively reduce the complexity of neuronal circuits and simplify the design of circuits with low energy consumption.^403,482 Smart e-textiles have advantages for integrating multiple functions, such as neuromorphic computing.^483–485

Chen et al.¹⁰² designed a memristor textile network with a Ag/MoS₂/HfAlO_x/CNT heterostructure. The neural network is based on the heterostructure of the MoS₂/HfAlO_x membrane as the basic unit for achieving synaptic and neuronal functions (Figure 15). State switching between the volatile and nonvolatile thresholds is realized through current control, and a postsynaptic cell body is used to integrate the obtained information. The action potential signal is then formed using an SNN to achieve the matching between different functions.

FIGURE 15. Schematic diagram of the structure of a memristor textile network with a Ag/MoS2/HfAlOx/CNT heterostructure. (A) E-textile memristor network with remodeled synapses at the upper layer and neuromorphic functions at the lower layer. (B) Unit device structure of the Ag/MoS2/HfAlOx/CNT heterostructure. (C) SEM images of the Ag/MoS2/HfAlOx/CNT heterostructure. (D) Artificial synaptic function simulation using the reconfigurable memristor. Reprinted with permission102 under a Creative Commons Attribution 4.0 International License. Copyright 2022. The Authors. Published by Springer Nature.

The fiber memristor¹⁰² uses 1.9 fJ per spike neuron energy, which is only 10⁻³ of the energy consumption of biological neurons, significantly improving the practicability of the design. In addition, functional integration was achieved by combining e-textiles, synaptic neurons, and heating resistance. Synaptic and neuronal connections in biological systems have been studied to simulate the complete neural networks required for complex life activities. Overall, this low-power, reconfigurable, neuromorphic computational textile system may open new avenues for bioinspired smart textile electronics.

SUMMARY AND OUTLOOK

In this comprehensive review, we delve into the utilization of existing 2D materials for the creation of integrated flexible electronics. First, we outline the current requirements for 2D materials in the flexible electronics sector and discuss the progress in these materials. Following that, we elaborate on integrated processing technology for flexible electronics using 2D materials. Additionally, we scrutinize innovative flexible electronics incorporating 2D materials, including flexible sensors, image displays, flexible energy-storage devices, and transducers. Crucially, we explore the integrated design strategies of 2D material devices that utilize a combination of neuromorphic algorithms in machine learning and cyclic logic computing.

Despite the significant advancements in flexible electronics composed of 2D materials, many challenges remain to be addressed.⁸²

First, the synthesis of 2D materials remains challenging. Although 2D materials meet these requirements owing to their small size, realizing their advantages necessitates progress in large-area synthesis, ^486,487 transfer to the device, establishment of good electrical contact, and regulation of electrical properties.⁴⁸⁷ Challenges persist in the use of 2D materials in conventional electronics. Controlled heavy doping is difficult to achieve in the thin layers of 2D materials, and the main strategies include Coulomb scattering and charge transport manipulation. The absence of suspension keys on the surfaces of 2D materials allows them to be manipulated by van der Waals interface engineering. For example, by controlling the interfacial contact in the device, a high-dielectric-constant material can be introduced to enhance the field-effect emission of carriers, or an edge contact can be used to reduce the interfacial resistance of the device. However, 2D materials often need to be grown at higher temperatures to introduce numerous defects and vacancies that can significantly affect the threshold voltage and contact resistance of the device.³¹ Therefore, new technologies must be developed to control and reduce the defects generated during these high-temperature synthesis processes while maintaining the properties of 2D materials.⁴⁸⁸

Second, the direct CVD synthesis of 2D materials on flexible substrates remains an unsolved problem. Flexible substrates determine the mechanical properties of 2D material flexible electronics.^46,489 These substrates must balance manufacturing process compatibility, low cost, high optical clarity, and excellent mechanical flexibility.⁴⁹⁰ Currently, the substrates used mainly include PET, PDMS, and PI, and the resistance temperature of these polymers is usually lower than 350°C. Low thermal stability leads to a significant limitation on the ambient temperature when 2D materials are used in wearable devices.^491,492 In addition, at high temperatures, the coefficients of thermal expansion of polymer substrates are higher than those of 2D materials, which can lead to film stress and cracks in the materials during high-temperature processing.⁴⁹³ It is necessary to solve the rigidity problem of integrated circuit components and integrate multiple electronic components into flexible substrates.⁹¹ The sensor unit converts the external physical signals into electrical signals for analysis. Graphene, TMDCs, and black phosphorus are ideal high-efficiency sensing materials.⁴⁹⁴ However, it is essential to analyze the use of a suitable sensing unit to facilitate the reading of the signal for a specific application, and then integrate a 2D material sensor with different functions into the same device.⁴⁹⁵ The design of high-resolution wearable devices requires processing high-density arrays of flexible materials over large areas. The current processing technology for 2D wearable devices is still dominated by mixing and integration with traditional rigid materials. Therefore, we must go through a development process from the hybrid development of 2D and rigid materials to the full integration of 2D materials for flexible electronics.

Third, issues arise during the transfer of 2D materials onto flexible substrates. In flexible electronic applications,⁸² 2D materials are often slowly oxidized in the environment, owing to their high surface activity,⁸² or their properties are degraded by moisture. Compared to rigid substrates, flexible substrates generally have poor gas barrier properties, in particular a weak ability to block water vapor, which may accelerate the degradation of 2D material properties. Although flexible polymer substrates offer excellent flexibility, they are less resistant to heat, which limits the direct high-temperature growth of 2D materials on flexible substrates.⁴⁵² The process of transferring 2D materials to flexible substrates involves complex manufacturing steps, and it is difficult to ensure that the transfer process does not compromise the material properties. The mechanical stresses that flexible substrates can withstand can also cause physical damage to 2D materials. In addition, the rougher surface of the flexible substrate may cause unwanted carrier scattering, which significantly reduces carrier mobility.

Significant opportunities exist in the integration of 2D materials with flexible electronics, as depicted in Scheme 2.

SCHEME 2. Current situation and future development trend of flexible electronics. Materials. Reprinted with permission.53 Copyright 2022 Wiley-VCH GmbH. Energy storage and conversion. Reprinted with permission.496 Copyright 2020 Elsevier B.V. Data processing. Reprinted with permission340 under a Creative Commons Attribution 4.0 International License. Copyright 2023 The Authors. Published by Springer Nature. Sensing technology. Reprinted with permission497 under the terms of the Creative Commons Attribution-NonCommercial license. Copyright 2021 The Authors. Published by the American Association for the Advancement of Science. Device comfort. Reprinted with permission498 under the terms of the Creative Commons Attribution-NonCommercial license. Copyright 2021 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. Biomedical application. Reprinted with permission499 under the terms of the Creative Commons CC BY license. Copyright 2023 The Authors. Published by Springer Nature.

Materials

The development of high-quality, low-cost, large-scale 2D material manufacturing technologies is a crucial aspect of advancing flexible electronics. Unfortunately, to date, there is no effective strategy for controlling the size and thickness of the 2D materials synthesized and mechanically exfoliated via hydrothermal synthesis.¹²⁰ Mainstream preparation strategies for large-area 2D materials include CVD, PVD, and MBE, and the preparation of 2D materials by vapor deposition inevitably requires substrate support.⁵⁰⁰ Further research is needed to understand the potential mechanism between the subjective dynamic process of 2D material growth and the competitive relationship between the thermodynamic barrier caused by the substrate surface structure.^31,501 Wang et al.⁵⁰² realized wafer-level monocrystalline MoS₂ epitaxial growth along a sapphire step using a tangential substrate design. A polymer sacrificial layer with special properties was used between the substrate and the 2D material to prevent the damage caused by the transfer of the 2D material required for device fabrication. Suspension and liquid–interfacial growth techniques.⁵⁰³ Artificial intelligence-aided design and modeling tools are used to achieve high-throughput screening and predict the performance of materials. Modifying the surface of 2D materials with organic functional groups enhances the detection sensitivity for specific parameters. Improving the resistance of 2D materials to environmental factors, such as humidity and temperature, is essential for realizing device applications in extreme environments. Furthermore, improving the thermal conductivity and durability of 2D materials through defect control and doping has been proposed.⁵⁰⁴ The thickness and spacing of the 2D materials are designed to achieve precise control of light absorption and emission. Despite these advancements, substantial challenges persist in achieving ubiquitous large-scale manufacturing of 2D materials and devices that meet the standards for flexible electronics, thereby demanding disruptive synthesis technology innovation.

Energy conversion and storage

The equipment should not be limited by traditional energy-supply methods, thereby allowing for more flexible usage. Efficient conversion and utilization of human mechanical energy and biological energy are essential for achieving greater energy self-sufficiency or even permanent power supply for sensors.^505,506 Currently, the power supply of flexible electronics mainly relies on supercapacitors³⁵⁴ along with self-powered technology and environmental energy harvesting. Although TENGs provide an innovative power solution for wearable devices, the instability of their output power is a major problem. TENGs are often required in combination with supercapacitors to achieve a stable electrical output and improve the mechanical toughness of the device. Using 2D perovskite materials, high-efficiency solar photovoltaic combiner panels can be fabricated, which are critical for the collection and utilization of solar energy. Flexible FSCs can be used as wearable fabrics; however, their safety when in close contact with the human body needs to be fully recognized.³⁷⁰ In addition, wireless energy transmission functions can be realized based on the excellent optoelectronic and mechanical properties of 2D materials. For example, highly efficient RF antennas using conductive 2D materials, such as graphene and MXene, can serve as receiving or transmitting terminals in wireless energy transmission systems to convert RF signals into direct current. Palacios et al.⁵⁰⁷ converted RF signals, including Wi-Fi signals, into electrical energy using a flexible rectifier antenna based on MoS₂. Furthermore, compact resonators designed using 2D materials can be integrated into clothing or bracelets to enable wireless energy transfer at specific frequencies. Ultimately, integrating these power supply technologies with sensors will enable not only power but also motion monitoring and medical care. Such an integrated design will make future wearable devices more efficient, portable, and capable of seamless integration into the Internet of Things,⁵⁰⁸ no longer constrained by traditional energy supply methods.

Sensing integration technology

The sensing integration technology of 2D material-based flexible electronics is evolving to improve the naturalness of interaction, enhance intelligent processing capabilities, and achieve multifunctional integration. The development of 2D material e-ink printing will drive a more intuitive user interface design, allowing screens to bend or even curl without compromising display quality. Multiple functions, such as sensing, display, and communication, are being integrated to improve the interactive perception experience related to vision and touch. By combining different types of 2D materials; reducing the number of components; and designing integrated devices capable of simultaneous data processing, communication, and perception functions, along with self-sustaining energy capabilities, we are realizing the miniaturization and weight reduction of equipment. The volume of monitoring data will persistently increase, and the development of software and algorithms will enable efficient and low-power data computing. Exploring interaction patterns between users and devices becomes crucial to provide a more natural and intuitive user experience. Through the integration of voice recognition, touch sensing, and gesture control technologies, users can interact more efficiently with devices. Machine learning and artificial intelligence algorithms play a crucial role in elevating the intelligence level of data processing. This enables sensors not only to accurately monitor and analyze data but also to provide personalized feedback and recommendations tailored to the specific needs of the user. Integrating multiple functions, such as environmental monitoring, biological health monitoring, and motion tracking, onto a single 2D material sensor platform enhances the application value and user convenience of the device.⁵⁰⁹ Developing solutions that seamlessly integrate with existing technology equipment creates an interconnected technological ecosystem.

Usability

Ergonomics must be considered in the design to ensure that the shape, size, and texture of the device are compatible with the body structure of the user, providing a comfortable and natural wearing experience. The device should also provide a clean and intuitive user interface that can be easily operated regardless of age.⁵¹⁰ Ensuring the security and privacy of user data has become paramount. Adhering to hardened data encryption and international privacy standards must be followed to ensure that user health data are adequately protected. Developing contactless energy transfer technologies, such as the use of triboelectric and piezoelectric effects, reduces the need for direct contact between the device and skin, thereby reducing skin irritation and enhancing the overall comfort of the wearing experience. Optimizing wearable device materials to reduce the weight helps alleviate the physical burden on users. Moreover, choosing cost-effective and easily producible materials can reduce manufacturing costs, making these devices affordable to a wider range of population.⁵¹¹

Biomedical healthcare

Bioelectronic medical care will eventually form a comprehensive system, including a portable power supply, high-sensitivity perception, intelligent data collection, diagnosis and treatment capabilities, and timely information feedback and treatment. To achieve this goal and ensure the standardization of experimental research and commercialization, more detailed and specific standards must be developed internationally. Prior to commercialization, wearable devices must be validated through rigorous clinical trials to ensure their reliability in real-world medical environments.⁵¹² Simultaneously, the equipment must be compatible with modern medical systems to ensure seamless health record monitoring between different devices, facilitate data integration and analysis, and achieve a holistic assessment of human health. The 2D materials used in wearable devices must be biocompatible and exhibit low toxicity, which is critical for designing devices that can be implanted into the human body. Long-term implantation of 2D material devices in vivo may have lasting effects on human organs; therefore, their long-term health effects need to be evaluated.⁵¹³ Developing reliable wireless charging technology is essential to ensure the continuous operation of wearable devices. An efficient data transmission solution should be built to minimize the risk of data loss during dynamic monitoring while connecting to telemedicine services for timely medical consultation and collaborative treatment. Wireless detection can reduce the iatrogenic injury of patients. This involves the placement of implant sensors with integrated power supply and wireless transmission into the human body, often in subcutaneous tissues. These sensors are implanted through minimally invasive surgery to achieve real-time monitoring of various physiological signals, such as blood oxygen and blood glucose.^497,513 The use of 2D materials^514,515 to construct biodegradable miniaturized devices can autonomously manage drug release and realize drug-targeted therapy.^516,517 Artificial intelligence and big data algorithms have been developed for the real-time analysis and collection of human health data. These technologies provide personalized health insights and early warning of potential disease. Continuous patient monitoring and remote sensing medical care is facilitated through wireless transmission detection technology.⁵¹⁸ Simultaneously, a broad spectrum of information is collected for early disease screening and the construction of a global disease database. The interconnected network of numerous sensors supports the concept of the Internet of Everything.⁵

In the future, we anticipate the development of a universal strategy for growing 2D materials on flexible substrates, using strain engineering and heterogeneous integration. These 2D materials will provide better flexibility and integration for a broader range of applications. More miniaturized and modular flexible electronic devices will be tailored to user needs and preferences, resulting in more personalized services and improved user experiences. Biodegradable materials improve the environmental-friendliness of the equipment, whereas implementing privacy protection measures will increase user trust in these devices. We eagerly await further innovations and breakthroughs that enable performance improvements and the development of new features for flexible electronics based on 2D materials.

ACKNOWLEDGMENTS

This work is partially supported by National Key Research and Development Program (No. 2022YFE0124200), National Natural Science Foundation of China (No. U2241221). J.P. thanks the Natural Science Foundation of Shandong Province for Excellent Young Scholars (YQ2022041), and the fund (No. SKT2203) from the State Key Laboratories of Transducer Technology, Shanghai Institute of Microsystem and Information Technology, Chinese Academy of Sciences for support. W.Z. thanks the Major Scientific and Technological Innovation Project of Shandong Province (2021CXGC010603), NSFC (No. 52022037) and Taishan Scholars Project Special Funds (TSQN201812083). The Project was supported by the Foundation (No. GZKF202107) of State Key Laboratory of Biobased Material and Green Papermaking, Qilu University of Technology, Shandong Academy of Sciences. M.H.R. thanks NSFC (No. 52071225), the National Science Center and the Czech Republic under the European Regional Development Fund (ERDF) “Institute of Environmental Technology—Excellent Research” (No. CZ.02.1.01/0.0/0.0/16_019/0000853) and the Sino-German Center for Research Promotion (SGC) for support (No. GZ 1400).

CONFLICT OF INTEREST STATEMENT

The authors declare no conflict of interest.