Opportunities in electrically tunable 2D materials beyond graphene: Recent progress and future outlook

Ever since the successful isolation of a single sheet of graphene (GR) from bulk graphite in 2004,¹ research into two-dimensional (2D) materials, where the interlayer interactions are governed by van der Waals (vdW) forces, has surged. Graphene, with its linear electronic dispersion, has been the most well-studied thus far, with significant progress toward commercialization.^2,3 Graphene, however, is just the tip of the iceberg in terms of vdW materials, where a plethora of materials have vdW bonding between the layers and thus can be isolated down to the atomically thin limit.⁴ Depending on the composing elements, the crystal structure, and even relative angles between two layers, these materials can display a wide variety of electrical properties, including semi-metallic, metallic, semiconducting, and insulating behavior. They even support exotic phenomena such as charge density waves and superconductivity.^5–10 Furthermore, magnetism at the monolayer limit has now been discovered,^11,12 opening applications in nm-scale magneto-optoelectronic and spintronic devices.^13,14 The transition metal dichalcogenide family (TMD) (MoS₂, WS₂, WSe₂, etc.) has been the most well studied after graphene, but large strides are being made regarding other 2D materials as well.

2D vdW materials display some truly fascinating properties that are inherently not possible with their thicker, three-dimensional (3D) counterparts. For one, it has been well-documented that the electrical, optical, and magnetic properties can dramatically change at the few-layer thickness regime, or even between even vs odd number of layers.^11,15–17 Additionally, 2D materials display extremely large mechanical flexibility, where large amounts of strain can be applied to tune their properties before rupture.^18,19 The vdW bonding between the layers allows for these materials to be grown or transferred onto a wide variety of substrates without the difficulty of epitaxial requirements, enabling easy integration with silicon technology that is technologically mature. Multi-layer heterostructures can be created with a variety of 2D materials, even with sub-degree control of the rotational alignment between the layers.^9,20

Another immense benefit for using 2D materials in next-generation electronics and opto-electronics, and the focus of this review, is that they are able to be strongly tuned using electric fields in ways that are not possible for 3D materials due to their reduced dielectric screening.²¹ Furthermore, the existence of 2D insulating materials such as few-layer h-BN allows researchers to bring materials within atomic-scale proximity to the backgate, allowing for extremely efficient tuning via electric fields. To fully reveal the emerging functionalities and opportunities of these atomically thin materials in practical applications, this review thoroughly details several representative and noteworthy research directions regarding the use of electrical means to tune both the physical and structural properties of layered vdW materials, with an emphasis on discussing the major advancements in 2D materials beyond graphene in the past 5 years and an outlook of what is to come in the next decade. A schematic overview of current and perspective applications of 2D materials beyond graphene is presented in Fig. 1.

First, a broad overview of the materials beyond graphene is presented, detailing crystal structure, optical and electrical properties, and other relevant features that are important for the review. Next, we outline the current status of electrical tunability in materials beyond graphene, including in applications for logic and memory devices, photonics and optoelectronic devices, and sensors (gas, bio, and strain sensors). We then turn our attention to more emerging areas and application opportunities that are showing exciting potential. These novel research areas include tunable quasiparticle dynamics, electrically controlled magnetism, piezo- and ferro-electricity, phase modulation, and electrically tunable twisted heterostructures.

Since the first discovery of graphene, 2D vdW materials have attracted immense research interest, thanks to their unique layered structures. As shown in Fig. 2,^22–58 common vdW materials can be divided into groups based on their crystal systems, structure, and tunable properties. Atoms are bonded with each other in each layer, but the interactions between neighboring layers are mostly dominated by vdW forces. Thanks to the weak vdW interactions, pristine surfaces of atomic, ultrathin layers can be easily preserved through mechanical exfoliation and other synthesis methods. Weak vdW interactions also allow the stacking of different 2D vdW materials without the limitation of lattice matching, which provides researchers more degrees of freedom in the design of complex vdW heterostructures with desired functionalities. In addition, the surface properties become extremely significant due to the ultrathin geometry of vdW materials and can be tuned with surface modification.

Transition metal dichalcogenides (TMDs) and Janus structures are of the type MX₂ (M = Mo, W, etc., and X = S, Se, or Te) where a layer of M-atoms is sandwiched between two layers of X atoms for each monolayer. While conventional TMDs possess the same chalcogen layer on both sides of the transition metal layer, Janus structures have a different chalcogen layer on either side. Based on the varied stacking orders of M- and X-atom layers, TMDs can be categorized into 2H (ABA) and 1T (ABC) phase. The 1T phase can be further divided into the 1T′ and the T_d phase as a result of structural distortion.

Hexagonal boron nitride (h-BN) is an insulator isostructural to graphene, where lattice points of the honeycomb structure are alternatively occupied by B and N atoms.

MXenes are layered structures resulting from selectively etching the A atom layer of the parent MAX phase, where M is an early transition metal, A is a group 13 or 14 element, and X is C or N.

Fe₃GeTe₂ consists of a Fe₃Ge layer sandwiched between two Te layers. The Fe₃Ge layer is composed of a hexagonal network formed by Fe atoms, which is covalently bonded to the remaining Fe and Ge atoms.

III–VI compounds are of the type A₂X₃ or AX, where A and X are group 13 and 16 elements, respectively. Among them, In₂Se₃ is one of the most well-studied materials. It consists of a Se–In–Se–In–Se quintuple layer, with distinct 2H- and 3 R-phases differentiated by the stacking of the quintuple layers. Similarly, InSe consists of Se–In–In–Se quadruple layers. For its β-phase, the second quintuple layer is rotated by 60° around the [001] direction.³¹ 

Elemental 2D materials include silicene, germanene, and black phosphorus (BP). Silicene and germanene are analogous to graphene. The major difference is that they are buckled structures, while graphene is flat. Black phosphorus has a pleated honeycomb-like structure.

Cr₂Ge₂Te₆ consists of CrTe₆ octahedra arranged in a honeycomb-like structure, where the center of each cell in the network is occupied by a Ge–Ge dimer.

Graphitic carbon nitride has a monolayer form which consists of an energetically preferred tri-s-triazine structure.

Group-IV monochalcogenides are of the type AX, where A and X are group 14 and 16 elements, respectively. They have a highly distorted rock salt structure, where each A atom is placed in the distorted octahedron and bonded with six X atoms.

Chromium trihalides are of the type CrX₃ (X=Cl/Br/I). For each monolayer, CrX₆ octahedra pairs with a shared edge are arranged in a honeycomb-like structure.

Transition metal thiophosphates are denoted by the chemical formula A^IIB^IIP₂X₆ or A^I(III)B^IIII(I)P₂X₆, where both A and B are transition metals, and X refers to group 16 elements. Among them, CuInP₂S₆ is one of the most important compounds. Its monolayer consists of a sulfur framework where the octahedral voids are filled by the Cu, In, and P–P triangular patterns.

Strong electrostatic tunability is the highlight feature of all 2D semiconductor-based field-effect devices. This is also the principle reason for their suitability as a channel material in extremely scaled transistors.^59,60 The suitability of vdW 2D channels for post-Si electronics or “more than Moore” electronics is well known and extensively discussed in reviews exclusively focused on transistor devices.^59–61 Here, we will focus on the following question: What does this electrostatic or gate tunability enable in 2D semiconductor devices that is unattainable or not demonstrated yet in other known semiconductors?

In the context of electronic devices, the impact of the electrostatic tunability of 2D materials can be looked at by first examining the simplest semiconductor device, which is the p–n junction. When a p–n junction is comprised of atomically thin semiconductors, it can be electrostatically tuned across both p and n layers that are vertically stacked resulting in unusual electrical characteristics. This was first reported on n-type 2D MoS₂ and p-type 1D carbon nanotube (CNT) junctions.⁶² Both a tunable rectification ratio of the diode and the first evidence of anti-ambipolar (Λ-shaped) transfer characteristics were demonstrated, with two OFF states at either extremes of gate bias and one ON state in the center [Fig. 3(a)].⁶² This observation of tunable rectification and anti-ambipolar transfer characteristics has since been generalized to several other 2D–2D,⁶³ 2D–organic,⁶⁴ all-organic⁶⁵ and other unconventional, mixed-dimensional^66,67 heterojunction systems, emanating from strong electrostatic modulation in 2D material-based device examples.

Numerous reports on various applications of such anti-ambipolar and tunable diode characteristics in 2D heterojunction devices have now been published, as summarized in other reviews.⁶⁸ One of the most prominent applications is in tunable, multivalued logic. A ternary inverter with three logic states has been demonstrated using an anti-ambipolar heterojunction between small-gap black phosphorus (BP) and larger gap ReS₂, which form a type III junction, connected in series with a p-type BP FET device⁶⁹ [Figs. 3(b) and 3(c)]. Similar results have also been obtained in gated type II p–n heterostructures of BP/MoS₂,⁷⁰ MoS₂/WSe₂,⁷¹ and SnS₂/WSe₂.⁷² Anti-ambipolarity provides both positive and negative transconductance with a steep transition between them that can be tuned either actively with dual gates⁷³ [Fig. 3(d)], as seen in MoS₂/BP self-aligned dual gated junctions, or passively with device geometry.⁶⁴ This allows frequency modulation and⁶⁶ multiplication applications, and opens room for constructing spiking neurons.^73,74

The above listed semiconductor combinations and device applications are promising for enhanced complementary metal–oxide–semiconductor (CMOS) or even neuromorphic and analog computing. However, all of them are comprised of materials that are still far from the maturity level of Si or III-V s. Among 2D materials beyond graphene, only MoS₂, WS₂, MoSe₂, WSe₂, and h-BN have been synthesized at wafer-scale uniformity on CMOS-compatible substrates.⁷⁵ Therefore, a more practical way to take advantage of 2D materials in heterojunctions is to combine them with commercial-scale 3D materials such as Si and III-V s. Along this direction, several recent breakthroughs have been made.^67,76 The earliest reports of gate-tunable 2D/3D heterojunctions include integration of graphene on Si to make gate-tunable Schottky diodes or barristors.⁷⁷ A key disadvantage of 2D/3D semiconductor junctions is that the 3D material does not allow significant electric field tunability or even a dual-gated structure. On the other hand, complementary doping type and density in 3D semiconductors are already well established and controlled, leading to high tunability of rectification and ON/OFF current ratios. Recent reports on junctions of MoS₂ with Si and GaN have established superior performance, as seen in Figs. 3(e) and 3(f).⁷⁸ The large rectification $(>106 times)$ and resistance $(>107 times)$ tunability suggest that these tunable heterojunctions can be dynamically configured to operate as both diodes and transistors, renewing the opportunity to explore the long-abandoned concept of diode-transistor logic in a new way.

Finally, 2D/3D junctions are also an important research avenue for tunneling field effect transistors (T-FETs). The controlled degenerate doping in the 3D case combined with strong gate-tunability in the 2D layer makes a compelling case for T-FETs. However, all T-FET demonstrations in this regard have thus far been limited by low ON/OFF ratios for sub-thresholds <60 mV/dec or low ON currents. To overcome this, several attributes are desired from the 2D/3D combination: (1) Near-intrinsic doping and clean semiconductors with large density of states at band edges, (2) clean and defect-free interfaces, and (3) superior coupling to the gate with a gate dielectric that is both thin and has high breakdown strength. Meeting all these criteria is challenging and therefore presents plenty of space for exploration of material combinations and interface design/engineering. Likewise, there is also opportunity for more complex designs, such as p–n–p or n–p–n bipolar junction transistors. While some attempts have been made in this regard, the performance and properties remain far from optimal.^79,80 Controlled interfaces and controlled doping will once again hold the key to higher performance and more mature devices.

As silicon-based, nonvolatile memory (NVM) devices are anticipated to approach their fundamental scaling limits, new technologies and devices are continuously being investigated to meet ever-increasing demands of high-density data storage. Resistive random-access memory (RRAM) devices have emerged as potential candidates in the pursuit of cheaper and smaller NVM devices with high write/erase speed and greater endurance and retention.^81–83 RRAM technology is foreseen to play a leading role in the future of memory industry, owing to its superiority over flash memory devices in terms of scalability without compromising on data storage, lower power consumption, and CMOS compatibility. So far, the most widely used functional materials for RRAM are based on insulating transition metal oxides; however, these materials are facing the bottleneck of scalability.^84–86 Therefore, extensive research efforts have been devoted to explore new functional materials for RRAM devices. Two-dimensional vdW material-based RRAMs have received tremendous research attention over the past few years,^87–89 with the advantage of highly scalable memory cells with low power consumption and fast switching speed. Several reports have explored the use of graphene as either a highly conductive electrode material or as an active layer in RRAM based NVM devices.^90–92 Likewise, other 2D materials such as TMDs have also been used in order to improve the performance of NVMs. Here, we provide a comprehensive overview of the most recent advancements in NVM devices based on 2D materials beyond graphene with a focus on electrical tunability of memory device properties.

Among 2D TMDs, MoS₂ is the most widely explored semiconductor as the channel material in NVM devices due to the combined effect of excellent mechanical flexibility and high charge-carrier mobility (>20 cm² V⁻¹ s⁻¹ for N ≈ 10¹¹ cm⁻²).⁹³ Memristive phenomena in monolayer MoS₂-based lateral devices through the inclusion of grain boundaries (GBs) have been investigated.⁹⁴ Resistance switching ratios of ∼10³ at small bias and small set fields (∼10⁴ V cm⁻¹) [Fig. 4(a)] were observed in devices with GBs connected to just one electrode. Owing to their atomically thin nature, these MoS₂ GB memristors exhibit gate-tunability of the set voltage from 3.5 to 8.0 V [Fig. 4(b)]. The same authors have shown multi-terminal memtransistors using polycrystalline (PC) monolayer MoS₂ for complex neuromorphic computing operations.⁹⁵ These memtransistors also exhibit high cycling endurance (≈475 cycles) and long retention times (24 h). However, these devices rely on GBs with site-specific orientation, which is difficult to control. Recently, a focused helium ion beam was used for precisely introducing sulfur vacancy-related, site-specific defects into TMDs (MoS₂). The device exhibited high endurance (>1180 cycles) and retention (>10³ s), as well as gate-tunable smaller set voltages.⁹⁶ Another recent study has shown a high gate tunability of the switching ratio from ≈10⁰ to 10⁵ using a few-layer MoS₂-based transistor.⁹⁷ 

A quasi-nonvolatile memory utilizing a semi-floating gate architecture was also used, achieving ultrahigh writing speeds of ≈15 ns and refresh times of ≈10 s, which is comparable to or higher than commercial dynamic random access memory [Fig. 4(c)].⁹⁸ The high writing performance was obtained due to the presence of two different charge transport mechanisms, e.g., switch path and flash memory path, in the device. Furthermore, the gate tunability of the device resulted in a high resistive switching ratio exceeding 10³ [Fig. 4(d)]. In another report, a vertically stacked MoS₂/h-BN/graphene heterostructure exhibited a significantly low off-state current ≈10^–14 A, resulting in a ultrahigh switching ratio of over 10⁹.⁹⁹ 

Another prototype approach for obtaining tunable memory operation has been demonstrated by adding redox-active molecules on a 2D $MoS2$ channel material (≈7 layers). By controlling molecular configurations through the gate voltage, carrier concentration in the 2D $MoS2$ could be modulated, resulting in the multistate memory operation.¹⁰⁰ The device exhibited a reasonable switching ratio of ≈10³ but showed relatively poor endurance (≈50 cycles). Thus, further efforts to improve the retention and endurance of such kind of devices are required to be useful for memory applications.

Recently, research in the area of NVM devices has focused on designing 2D ferroelectric field effect transistors (FeFETs), where a ferroelectric layer with high polarization field and dielectric constant is used as the gate dielectric material for producing efficient gating effect in the semiconductor channel layer.^101,102 Due to switchable electric dipoles and large retention properties, FeFETs are considered as potential candidates for building NVM with low‐power consumption and ultrafast logic operation. Earlier studies on 2D FeFETs were focused on using graphene as the conducting channel with bulk ferroelectric materials such as lead zirconate titanate (PZT) or organic poly(vinylidenefluoride‐trifluoroethylene) (PVDF‐TrFE) thin films. In these, the resistance of the conducting channel was effectively tuned by the reversible electrical polarization of the ferroelectric thin film.^92,103,104 In line with the conventional approach, integration of a graphene channel with an atomically thin 2D ferroelectric as the gate dielectric has also been demonstrated. The use of ultrathin ferroelectric materials increases the effective gate field, which reduces the writing or erasing voltage needed to flip the electric polarization and leads to a low‐power consumption memory device.^105,106

The fabrication of a 2D FeFET with graphene as the conducting channel and ferroelectric 2D α‐In₂Se₃ as the top gate dielectric has been recently demonstrated, where the resistance states of the graphene channel were efficiently modulated by sweeping the ferroelectric gate voltage, leading to switching of the polarization direction in the α‐In₂Se₃ layer. The device exhibited excellent endurance for ≈10⁵ switching cycles and retention performance for ≈1000s.¹⁰⁵ Another recent study achieved a giant electroresistance switching ratio ≈10⁶ through the modulation of Schottky barrier height and width via ferroelectric switching in large area α-In₂Se₃ grown on graphene using molecular beam epitaxy [Figs. 4(e) and 4(f)].¹⁰⁶ Devices with semiconducting 2D materials have also been investigated. Ferroelectric memory transistors using monolayer to few-layer MoS₂ as a channel and P(VDF‐TrFE) as the ferroelectric top gate insulator exhibited an ON/OFF ratio of ≈10³, concurrently with retention properties for more than 1000 s.¹⁰⁷ A similar approach has been adopted for BP¹⁰¹ and MoSe₂¹⁰⁸ FE-FETs, which showed similar or higher ON/OFF ratios, retention, and endurance. However, slow dipole dynamics of ferroelectric polymers and low thermal durability compared to its inorganic counterparts can hamper the practical applicability of organic ferroelectric-based FETs. Thus, FeFETs integrating 2D materials with inorganic ferroelectric materials were also investigated. Nonvolatile memory devices using mono‐ to few‐layer TMDs (WSe₂ for p‐type and MoS₂ for n‐type) as the channels and epitaxial PZT thin films as the FE layer have been fabricated [Fig. 4(g)].¹⁰⁹ The ferroelectric gate tunability of the device (3L-WSe₂) was demonstrated, and an ON/OFF ratio was tuned from ≈10⁴ to ≈10⁵ with positive gate bias [Fig. 4(h)]. Furthermore, the device exhibited endurance over 400 switching cycles and retention for up to 10⁴ s. Another recent work reported high-performance FE-FET using MoS₂ channel on top of AlScN dielectric and achieved a high ON/OFF ratio of ≈10⁶ and retention up to 10⁴ s.¹¹⁰ A 2D/2D vdW heterostructure FeFET for nonvolatile ReRAM applications has also been designed, wherein 2D CuInP₂S₆ was integrated on top of the MoS₂ channel as a ferroelectric insulator.¹¹¹ The MoS₂/CuInP₂S₆ FeFET showed an ON/OFF ratio of >10⁴ via ferroelectric polarization switching-induced changes in the band alignment between the metal and CuInP₂S₆.

So far, the most intensively explored device concept for FeFETs consists of a ferroelectric gate dielectric integrated with a semiconductor channel layer. However, this conventional approach suffers from charge trapping and gate leakage issues, resulting in short retention times. Recently, a few 2D vdW materials have been shown to exhibit ferroelectricity as well as semiconducting properties, e.g., α‐In₂Se₃¹¹² and CuInP₂S₆.⁵⁸ A few reports have realized FeFETs using ferroelectric 2D materials as the channel layer, where the resistance states of the devices were controlled by ferroelectric polarization switching.^113,114 In a recent study, both planar and vertical memristors using α‐In₂Se₃ were investigated, demonstrating memristive phenomena based on both in-plane and out-of-plane polarization.¹¹⁵ Both the devices exhibited a resistive switching ratio >10³ due to modulation of the Schottky barrier height with ferroelectric gate bias and showed excellent endurance and retention properties for over 100 cycles and up to 1000 s, respectively. Another recent study has also observed similar performance (ON/OFF ratio >10³) using an α‐In₂Se₃ FeFET.¹¹⁶ To provide an overview of the performance of different 2D material-based NVM devices, a comparison between the ON/OFF ratio and the retention time is plotted in Fig. 5.^{94–96,99–102,105,107–109,115,117,118}

Harnessing solar energy using photovoltaic (PV) cells is by far the most promising strategy to fulfill the ever-increasing demand of renewable energy. PV cells using Si, III–V compounds, and, very recently, organic/inorganic hybrid perovskites have been investigated, and significant conversion efficiency has also been achieved.^119–121 However, the fabrication of these devices is associated with certain complexities, such as the requirement of epitaxial growth techniques, limited abundance of some elements, and chemical stabilities, which raise concerns about their long-term economic viability. Recently, 2D TMDs, which have bandgaps with energies ranging from the ultraviolet to the near infrared (NIR) regions, have been considered as a new class of functional materials to potentially serve as thin absorbers for high performance PV cells. This is due to their high absorption coefficient and appropriate bandgaps in the bulk (1.3–1.4 eV) for attaining maximum power conversion efficiency (PCE) per Shockley–Quiesser (SQ) limit.¹²² Two-dimensional TMDs are inherently flexible, and the lack of dangling bonds at the surface encourages the creation of high-quality heterointerfaces, resulting in low-cost, flexible, and light-weight PV devices operating in a large region of the solar spectrum. PV cells based on 2D TMD heterostructures are also expected to exhibit high power conversion efficiency (PCE) as the maximum photovoltage is not limited by the built-in electrical potential energy, as is the case for conventional p–n junctions. The charge generation process in 2D TMDs is associated with strongly bound excitons instead of weakly bound electron–hole pairs. The exciton generation and dissociation occur simultaneously in a narrow region near the heterojunction, resulting in a large carrier concentration gradient that acts as a powerful PV driving force. Their atomically thin nature offers a unique platform for achieving electrically tunable PV properties for high efficiency. The tuning of carrier density in active layers of a p–n junction also has strong implications for photodetectors. For instance, increasing carrier density in a p–n junction results in a greater built-in potential; this leads to a smaller reverse saturation current, directly impacting the sensitivity and responsivity of the photodiode detectors. Conversely, decreasing the carrier density results in a lower depletion width and an increased reverse saturation current. In this section, we highlight p–n junctions based on 2D semiconductors where electric-field-induced tuning of the semiconductor induces modulation of the electrical response to light absorption.

PV energy conversion was first reported in lateral p–n homojunctions using several electrostatically doped monolayer 2D materials such as WSe₂, MoS₂, and BP.^123–126 Biasing a pair of gate electrodes with opposite polarities, the carrier type and density were controlled in the channel, resulting in the formation of a p–n junction. The maximum PCE achieved using such junctions is ≈0.5%. However, it can be further enhanced to ≈14% under a standard air mass (AM)-1.5 solar spectrum using multilayer (≈10 atomic layers) MoSe₂ crystals stacked onto dielectric h-BN.¹²⁷ Recently, using high-resolution angle-resolved photoemission spectroscopy, the surface PV effect has been observed in a β-InSe semiconductor that allows for further tuning by in situ surface potassium doping. This study can be further extended to engineer the photovoltaic effect in InSe-based p–n devices.⁴⁵ Furthermore, vertical p–n homojunctions have also been designed. A vertical homojunction using thin and thick flakes of MoSe₂ has been fabricated, and a gate tunable PV effect was obtained due to different gate modulation levels of the carrier densities in these flakes. The PV effect as a function of gate voltage (V_g) showed a maximum open-circuit voltage (V_oc) ≈ −0.24 V and PCE ≈ 1.9% [Figs. 6(a) and 6(b)].⁴⁶ Similarly, a gate-tunable PV effect has also been demonstrated in a vertical homojunction of n-type MoS₂:Fe and p-type MoS₂:Nb few-layer flakes by changing the charge carrier density through the Si-gate electrode.¹²⁸ 

In contrast to homojunctions, the vdW heterojunctions, made by stacking 2D materials of different bandgaps, are expected to exhibit enhanced PV effect. A type-II vdW heterojunction composed of MoS₂ and WSe₂ monolayers has been fabricated, with gate tunability of the short-circuit current (J_sc) and V_oc resulted in a PCE of ≈0.2%.¹²⁹ In another work, external quantum efficiency (EQE) ≈ 2.4% has been achieved using a monolayer MoS₂/WSe₂ heterojunction sandwiched in between graphene electrodes, which can be further improved by increasing the number of atomic layers in the MoS₂/WSe₂ heterojunction.⁶³ A polymeric gate was also used on top of MoS₂/InP to obtain an electrically tunable PV effect, leading to maximum PCE of ≈7.1%.¹³⁰ Recently, a vertical vdW heterostructure consisting of multilayer InSe and Te has been fabricated, exhibiting a PV effect as well as a broadband photoresponse with an ultrahigh photo/dark current ratio exceeding 10⁴ and a high detectivity of ≈10¹² Jones under visible light illumination [Fig. 6(c)].¹³¹ In a vertical vdW p–n junction between few‐layer p‐BP and n‐InSe, the band alignment and intrinsically high carrier mobilities in BP and InSe resulted in an EQE as high as 3%, indicating an efficient separation of the photogenerated charge carriers. The polarization‐dependent photoresponse of the device was also investigated through scanning photocurrent microscopy which showed a substantially higher polarization sensitivity (photocurrent anisotropy ratio ≈0.83) than those of traditional BP photodetectors.¹³² 

To scale up the process, CVD-grown large area vertical heterostructures have also been employed. A gate-tunable PV effect has been obtained in CVD-grown WS₂/MoS₂ and MoS₂/WS₂ vertical heterostructures sandwiched between graphene electrodes.¹³³ The I_sc and V_oc increased monotonically from 30 nA and 0.7 mV to 120 nA and 2.4 mV, respectively, by varying the gate voltage. However, this enhancement is smaller than what has been reported for mechanically exfoliated samples.

In another recent report, high performance PV effect was demonstrated in WSe₂/WS₂ heterostructures due to effective modulation of the junction transport properties as a function of gate voltage due to the ambipolar nature of WSe₂,¹³⁴ where a significant modulation of the output electrical power P_output of the device with gate voltage was achieved [Fig. 6(d)]. The heterostructure exhibited a maximum V_oc of ≈0.58 V and PCE of ≈2.4%. The modulation of the PV properties was correlated with the combined effect of channel conductivity and quasi-Fermi level tuning at the interface. Theoretically, it has been further proposed that a high PCE ≈ 30.7% can be obtained using a dual-gated semiconducting 2H-phase WTe₂/MoSe₂ vdW heterostructure.¹³⁵ By adjusting the dual-gate voltages, the photocurrents in the two subcells can be matched, leading to tandem-cell operation of the device.

Another research area is the study of the PV effect in 2D/3D heterostructures, which can theoretically absorb more of the incident light. By varying the gate voltage through an ionic polymer top gate in a MoS₂/GaAs heterostructure, the PCE was improved from 6.87% to 9.03%,¹³⁶ caused by the improved barrier height (ϕ_barrier) induced by the shift in the MoS₂ Fermi level [Figs. 6(e) and 6(f)]. Similarly, MoS₂/Si and MoS₂/GaN 2D/3D semiconductor heterojunctions have been fabricated for switching and rectification applications. By tuning the Fermi levels of MoS₂ via electrical gating, the devices exhibited over seven orders of magnitude modulation in the rectification ratio and an ON/OFF ratio exceeding 10⁷.⁷⁸ In another recent study, a large lateral photovoltaic effect (LPVE) with ultrafast relaxation time (≈2 μs) in a SnSe/p-Si junction has been reported. The diffusion of electrons laterally in the inversion layer formed at the SnSe/p-Si interface resulted in the large LPVE with ultrafast relaxation time.¹³⁷ 

In general, 2D vdW materials exhibit layer dependent optical bandgaps, large exciton binding energies, and high carrier mobilities, making them exciting candidates for designing novel optoelectronic devices such as phototransistors and LEDs.^{122,138–140} Electrical tuning is the most preferred approach to control optical excitations in 2D materials due to its fast action and easy on-chip integration in nano/micrometer scale. This has been implemented to control excitonic emission in 2D TMD-based LED devices.¹⁴¹ Electroluminescence (EL) emerging from desired exciton species with unique emission characteristics is of fundamental importance for the practical implementation of LEDs. In view of this, 2D TMDs represent a new class of functional materials for realizing electrically tunable, exciton-mediated LEDs. In this section, we summarize the progress in realizing electrically controlled electroluminescence in 2D TMDs and their heterostructures using different device configurations.

Gate-tunable EL has been reported in numerous TMD-based lateral p–n junctions, formed by either electrostatic gating or ionic liquid gating, where EL originating from different excitonic complexes is tuned by changing the carrier density.^{123,124,141–144} The recent research efforts in this direction have focused on using CVD-grown TMDs for fabricating large area LED devices. Lateral p–n junctions using CVD-grown, polycrystalline WSe₂ and MoS₂ monolayers have been realized by placing an electrolyte on top of the TMDs, where the electronic charges are induced by the formation of electric double layers.¹⁴⁵ The tunability of the EL intensity with the applied bias was demonstrated. However, the EQE obtained in such devices was significantly low (≈10^–3% for WSe₂ and ≈10^–5% for MoS₂), likely due to the poor quality of the CVD-grown samples.¹⁴⁵ Apart from the inorganic 2D-material LEDs, LEDs using 2D molecular semiconductors have also been fabricated, which hold great promise for ultrafast on-chip optical communication applications.¹⁴⁶ The lack of suitable bipolar Ohmic contacts for achieving high injection levels of electrons and holes in the 2D material is a bottle neck for obtaining enhanced performance from the lateral p–n junction devices. In a recent study, EL from a dopant-free two-terminal device was observed by applying an AC voltage between the gate and the semiconductor [Fig. 7(a)].¹⁴⁷ An excess of electron and hole populations simultaneously present in the monolayer TMD (MoS₂, WS₂, MoSe₂, and WSe₂) during the AC transient resulted in pulsed light emission at each V_g transition [Fig. 7(b)].

In lateral p–n junctions, emission is spatially localized only near the contact regions due to difficulty in achieving uniform carrier injection throughout the monolayer, leading to low quantum efficiency. An improved LED performance can be realized using vertical vdW heterostructures due to the reduced contact resistance, large luminescence area, and wider choice of TMDs, which can result in significantly improved EL efficiency. Light emission from p–n junctions composed of 2D/3D heterostructures such as n-type MoS₂/p-type silicon and¹⁴⁸ MoS₂/GaN¹⁴⁹ has been demonstrated, in which an electric field/current-driven tuning of the band alignment at the heterostructure interface or carrier redistribution was proposed as the origin of the EL tuning.

Recent research work has focused on investigating light-emitting properties of interlayer excitons (X_I) in 2D/2D type-II heterostructures, which show great promise for novel excitonic device applications. Interlayer exciton emission in electrostatically gated MoSe₂/WSe₂ heterobilayers has been investigated.¹⁵⁰ By biasing gate electrodes with opposite polarities (V_BG1 = −1 V and V_BG2 = 5 V), a p–n junction was realized and EL was obtained, which was dominated by interlayer exciton emission under forward bias condition. The EL peak was found to shift from 1.35 to 1.38 eV with a vertical electric field via the Stark effect. In another recent report, electrically tunable interlayer exciton emission was demonstrated, along with upconverted EL via Auger scattering of interlayer excitons by injecting high carrier density into a WSe₂/MoS₂ type-II heterostructure [Figs. 7(c) and 7(d)].¹⁵¹ 

An extremely high EQE has been achieved in quantum well (QW) structures comprising vertically stacked metallic graphene, insulating h-BN, and semiconducting TMDs. An EQE as high as ≈8.4% in a MoS₂-based, multiple QW structure has been obtained at low temperatures,¹⁵² and a high EQE (≈5%) at room temperature has been obtained using a WSe₂-based QW, in which a monotonic increase in the EQE as a function of bias voltage and injection current density was observed [Fig. 7(e)].¹⁵³ In a recent study, high-speed electrical modulation of light emission by integrating a photonic nanocavity (GaP) with a WSe₂ QW structure was reported.¹⁵⁴ The light emission intensity of the cavity-coupled EL peak enhanced ≈4 times compared to the cavity-decoupled peak as the bias voltage was increased. Furthermore, electrical modulation of the EL revealed fast rise and decay times of ≈320 and ≈509 ns, respectively, by turning the voltage on and off, resulting in fast-operational speed (≈1 MHz) of the device [Fig. 7(f)].

An ultralow turn-on current density of 4 pA·μm⁻², which is ≈5 orders of magnitude lower than that of the best single QW device, has been obtained in a metal-insulator-semiconductor (MIS) vdW heterostack comprising few-layer graphene (FLG), few-layer h-BN, and monolayer WS₂.¹⁵⁵ EL from the positively charged (X⁺) or negatively charged (X⁻) trions was reversibly controlled by electrostatic tuning of the TMD (WSe₂) into n- and p-type doping regimes under forward bias.¹⁵⁶ The EL intensity increased almost linearly with the tunneling current, and the lower-bound EQE was estimated to be ≈0.1% for X⁺ and 0.05% for X⁻. Furthermore, electrically driven light emission from multi-particle exciton complexes in TMDs (WSe₂ and WS₂) has been demonstrated using a MIS-type structure [Fig. 7(g)].¹⁵⁷ By tailoring the parameters of the pulsed gate voltage, an electron-rich or a hole-rich environment can be created in the 2D semiconductors, where the emission intensity from different exciton species is tunable [Fig. 7(h)].

Owing to inherent broken inversion symmetry, monolayer 2D TMDs have been exploited to obtain valley-polarized EL. Electrically controlling circularly polarized EL was first demonstrated in monolayer and multilayer WSe₂ using ionic liquid gating.¹⁵⁸ Due to large carrier injection capability with ionic gating, inversion symmetry breaking and band structure modulation can be obtained in multilayer 2D TMDs. Consequently, circularly polarized EL from monolayer and multilayer samples has been observed under forward bias. The EL intensity linearly increased with increasing bias voltage. Electrically tunable, chiral EL from large area CVD-grown monolayer WS₂ in a p⁺-Si/i-WS₂/n-ITO heterojunction was found to be dominated by negatively charged excitons, with an increase in the injection current from 2.5–4.0 μA due to imbalanced carrier injection under forward bias, resulting in an enhanced n-type doping. EL with a high degree of circular polarization ≈81% at 0.5 μA was reported in the same device [Fig. 7(i)].¹⁵⁹ Similarly, EL from MoSe₂ with circular polarization ≈66% has also been obtained.¹⁶⁰ 

Another popular approach for obtaining electrically controlled, valley-polarized EL is via spin-polarized charge carrier injection through a ferromagnetic semiconductor or electrode (Ni/Fe Permalloy) into the TMD monolayers due to the spin-valley locking effect. Circularly polarized EL at a (Ga,Mn)As/WS₂ p–n heterojunction and a WSe₂/MoS₂ heterojunction under forward bias has been obtained due to the imbalance of spin-injected carrier population at K and K′ valleys.^161,162

In addition to the direct-gap excitons, luminescence from localized defect states in 2D TMDs can be used to obtain single-photon quantum emitters, which are the fundamental building blocks for quantum photonics and quantum information technologies. The emission from such localized defect states can be controlled electrically, resulting in the realization of electrically controlled quantum emitters. Electrically controlled EL from the localized states has been demonstrated in graphene/h-BN/WSe₂/h-BN/graphene vertical heterostructures, where¹⁶³ a spectrally sharp emission (at ≈1.607 eV, V_b ≈ −2.15 V) corresponding to a single defect state (SDE7) was obtained. As the electrical bias was increased, additional broad features corresponding to emission from other localized states also emerged [Fig. 7(j)]. The full-width-at-half-maximum (FWHM) of the localized EL peak increased from 0.6 to 1.4 meV as the bias was increased above −2.3 V. In another report, electrically driven light emission from defects in WSe₂ using both vertical and lateral vdW heterostructure devices was obtained.¹⁶⁴ The vertical structure exhibited a broad peak due to defect-bound exciton states, including a narrow emission peak (≈1.705 eV) that was referred to as emission from a single defect. Electrical tuning of the emission from a single defect was demonstrated, where the emission was repeatedly switched on and off by sweeping external bias V_b from 1.9 to 2.1 V. In contrast, the lateral heterojunction device exhibited EL originating from several single defects having line widths <300 μeV and a doublet structure with ≈0.7 meV energy splitting. Similarly, electrically driven single-photon emission from localized states in mono- and bi-layers of WSe₂ and WS₂ has also been reported.¹⁶⁵ The electric current dependence of the EL from the quantum emitter showed clear saturation, whereas the emission from the unbound monolayer WSe₂ excitons exhibited a linear relation between emission intensity and injected current.

Extensive research efforts have been focused on designing ultrafast, low power, and compact optical modulators for a variety of applications, such as optical interconnects, environmental monitoring, security, and biosensing.^166–168 For designing high-performance optical modulators, materials with low optical-loss, large non-linear optical constants, broad wavelength operation, large tunability of optical constants, and ease of integration with various optical components are desired. In this category, 2D materials have attracted significant attention, owing to their atomic thickness, strong light-matter interaction, broadband optical response, and tunable opto-electronic properties.¹⁶⁸ 2D materials also offer advantages for low-cost and large-scale integration to the well-developed silica fiber and silicon-based technology. Thus far, graphene has been the leading candidate for obtaining high speed optical modulation in an extremely broad spectral range, extending from the ultraviolet to microwave regions, primarily due to its unique linear energy-momentum dispersion relation and high mobility.^169,170 However, other 2D materials such as monolayer TMDs and black phosphorus, which offer properties complementary to graphene, also exhibit similar potential and have been explored recently. The presence of a bandgap and parabolic band structure in 2D semiconductors can allow much higher tunability of the optical dielectric functions. Different tuning methods have been demonstrated to modulate the optical response of the 2D materials that can be characterized by the change in the dielectric constants or the complex refractive index of the materials.¹⁷¹ The electrical control of optical response is particularly desired for data communication link applications. Herein, we present state-of-the-art progress in electro-optic modulators based on 2D semiconductors beyond graphene.

For the electro-optic modulators, the application of an external electric field is desired to tune both the real and imaginary components of the refractive index such that both the amplitude and phase of the optical field can be modulated. In a recent report, a modulation of the refractive index of monolayer TMDs, such as WS₂, WSe₂, and MoS₂, by more than 60% in the imaginary part and 20% in the real part, around their excitonic resonance was demonstrated using electrical gating.¹⁷² The giant tuning in refractive index was attained by changing the carrier density, which broadened the spectral width of excitonic transitions and facilitated the interconversion of neutral and charged excitons. The gate dependence of optical absorption and optical constants of TMDs was also investigated by several other groups [Figs. 8(a) and 8(b)].^173,174 A modulation depth as high as ∼6 dB for visible light (red light ∼630 nm) has been reported.

The other popular approach that can be used for the manipulation of dielectric properties of TMDs is by the interaction of excitons with metallic plasmons. In one work, narrow MoS₂ excitons coupled with broad Au plasmons led to an asymmetric Fano resonance that was effectively tuned by the applied gate voltage to the MoS₂ monolayers.¹⁷⁵ The gate-dependency of the Fano resonance is strongly sensitive to the modulation of the exciton–plasmon coupling strength and can be controlled by the MoS₂ exciton absorption at different external gate voltages. Active control of light-matter interactions is also critical for realizing plasmonic nanostructure-based electro-optic modulators. An electrical control of exciton–plasmon coupling strengths between strong and weak coupling limits in a 2D semiconductor has also been demonstrated with electrostatic doping [Figs. 8(c) and 8(d)].¹⁷⁶ In addition to the coupling of the plasmonic modes to the neutral excitons, a strong coupling with negatively charged excitons was also obtained that can be switched back and forth with the gate voltage. In another recent study, electrical modulation of plasmon-induced exciton flux was demonstrated using Ag-nanowire waveguides overlapping with TMD transistors.¹⁷⁷ The laser-coupled-plasmon propagated through the Ag nanowire in the axial direction, which sequentially excited excitons of the TMD, and the exciton flux was modulated by the gate voltage. In most of the reports, electrical tuning of the optical properties of TMDs was achieved near their excitonic resonances only, where the refractive index and absorption can be modulated simultaneously at a maximum magnitude. In a recent report, electro-optic response of monolayer TMD (WS₂) at NIR wavelength regions was probed for integrated photonics applications.¹⁷⁸ Using an ionic liquid gate, high electron doping densities (7.2 ± 0.8 × 10¹³ cm⁻² at 2 V) were induced in the monolayer WS₂ and a large change in the real part of the refractive index (≈53%), and a minimal change in the imaginary part (≈0.4%) was demonstrated [Figs. 8(e) and 8(f)]. Also, a doping-induced efficient phase modulator with high |Δn/Δk| ≈ 125 and low propagation losses was achieved.

Apart from TMDs, black phosphorus (BP) is also an emerging candidate for designing electro-optic modulators in the mid-infrared frequencies owing to its smaller bandgap. Several theoretical and experimental investigations have shown that an external electric field can result in a shift in BP's absorption edge due to the interplay between different electro-absorption mechanisms, mainly the field-induced, quantum-confined Franz–Keldysh effect, and the Pauli-blocked Burstein–Moss shift.^179–181 These different mechanisms lead to distinct optical responses that are strongly dependent on the flake thickness, doping concentration, and operating wavelength. To isolate and define the working mechanism of different electro-absorption effects, a recent report used two different field-effect device configurations with different gating schemes, wherein the BP either floats electrically in an applied field or is in direct contact [Figs. 8(g) and 8(h)].¹⁸² In the electrically floating case, the dominant tuning mechanism is the quantum-confined Stark effect, while in the other case tunability is dominated by carrier concentrations effects, e.g., by the Burstein–Moss shift. Near-unity tuning of the BP oscillator strength and electro-optic tuning of linear dichroism over a broad range of wavelengths, from the mid-infrared to the visible, by controlling the thickness of the BP was reported.

While significant progress has been achieved with optical modulation using 2D materials, it is safe to say that, with the exception of graphene, the field is in its infancy. This is evident from Fig. 9, where we have compared the depth of modulation with the wavelength of modulation. Graphene-based modulators not only show exceptional modulation depth, but, due to its zero-gap nature, they can also operate well at telecom wavelengths, with performance comparable to Si, LiNbO₃ and III–V modulators. In contrast, TMDs are barely able to operate in the NIR range. The key issue at play for materials beyond graphene to serve in high-performance optical modulators is the lack of quality large-area material. Furthermore, most 2D semiconductors are in the 1.1–3 or <0.5 eV range of bandgap values. This makes it difficult to achieve modulation by means of a dominant photorefractive effect at a band or exciton edge in a 2D semiconductor in the telecom range (≈0.8 eV). Finally, it is equally important to attain good metal contacts to inject and extract carriers at high speed with minimal resistance or other parasitic circuit elements, particularly at high operation speeds.

Due to their unique physical properties, tunability, and versatility, 2D materials are particularly attractive for multiple sensing applications. Here, we consider their use for gas (primarily environmental), humidity, and biosensing as well as in strain and pressure sensing. The fact that their electronic structure is highly susceptive to and tunable by hybridization, chemical modification, or doping makes them an ideal platform for regulation of catalytic properties. This is particularly important for their applications as chemical and biosensors, often allowing fine control of sensing by electrical means. We specifically focus our attention on the electrical means of readout and control of such sensors. For another very powerful way to control such sensors, namely, optical or optoelectronic detection, we address the readers to the other reviews in the area.^195–197 

Based on the tremendous success of graphene-based sensors, as seen both in the published research (at the moment of the review submission (June 2021): Google search gave ≈36 500 000 results; Web of Science search – ≈27 600 publications) and first successful commercialization,¹⁹⁸ efforts in constructing novel sensing platforms led to exploitation of other, more versatile and tunable 2D materials, e.g., TMDs, carbon nitride, boron nitride, phosphorene, and MXenes, which will be discussed in this section of the review.

In this review, we consider the recent developments in non-graphitic 2D material sensors that have followed in the wake of graphene. We discuss the different classes of 2D materials in application to sensing as well as their figures of merits, such as Level of Detection (LoD), selectivity, response/recovery time, and detection range, and compare them with the performance of graphene sensors.

When discussing the physical mechanisms of sensor operation, we only consider non-covalent (and thus reversible) interactions between the target analyte and 2D material, which include vdW forces, hydrogen bonding, coordination, and π–π interactions. The type of interaction for each 2D material will depend on its chemical and electronic structure, distinct structural features, and surface chemistry. To be electrically detectable, any of these interactions should induce changes in physical parameters such as conductivity, work function, or permittivity. This is typically realized through fundamental mechanisms such as modulation of doping level and/or Schottky barrier, as well as the formation of dipoles and interfacial layers. For more details, see Meng et al.¹⁹⁵ and the references within.

In the majority of cases, the numerical output of a whole FET device is read out in the response to global gas exposure. Recently, Noyce et al.¹⁹⁹ provided an important insight into sensing mechanisms by realizing precise nanoscale control over the position and charge of an analyte using a charged scanning atomic force microscopy (AFM) tip acting as an effective analyte. The non-uniform sensitivity of an MoS₂ FET channel was demonstrated, showing time-stable, sensitive hotspots, where the signal-to-noise ratio was maximized at the center of the channel, and the response of the device is highly asymmetric with respect to the polarity of the analyte charge. The work reveals the important role of analyte position and coverage in determining the optimal operating bias conditions for maximal sensitivity in 2D FET-based sensors, which provides key insights for future sensor design and control.

Sensors serve an important role in the detection of common environmental gas pollutants including nitrogen dioxide (NO₂ at 21 ppb), sulfur dioxide (SO₂ at 7.5 ppb), hydrogen sulfide (H₂S at 5 ppm), ammonia (NH₃ at 20 ppm), and carbon monoxide (CO at 4 ppm). Each of these gases is considered to be toxic for human health although the specific exposure limits vary significantly. The recommended exposure limits (as set by the Gothenburg Protocol²⁰⁰ or the Paris Climate Agreement²⁰¹) for each of these pollutants are shown in parentheses above. See Buckley et al.¹⁹⁶ for more details. Additionally, carbon dioxide (CO₂) and methane (CH₄) are not toxic gases but are responsible for the greenhouse effect, leading to climate change. The exposure limits indicated in the parentheses above should be taken into account when performance of individual sensors is evaluated, as in a number of cases the demonstrated LoD and detection range are far outside of specific environmental requirements.

Among all 2D materials, TMDs and specifically MoS₂ have been the most intensely explored for gas sensing, where earlier works focused largely on mechanically exfoliated TMD materials of variable thickness.^202,203 MoS₂ FETs fabricated with 2–4 layers of MoS₂ exhibited better performance compared with monolayer MoS₂, which showed unstable response. For similar MoS₂ FETs, detection of both electron acceptors (NO₂) and electron donors (NH₃) under different conditions such as gate bias and light irradiation has been explored²⁰⁴ [Fig. 10(a)]. $MoS2$ FET exhibited better sensitivity, recovery, and ability to be manipulated by gate bias, e.g., sensitivity to 200 ppm $NO2$ exposure increased from ≈50% to ≈400% for bilayer ad 5-layer $MoS2$⁠, respectively, under $Vg=15 V$⁠. Additionally, for 5-layer $MoS2,$ resistance goes down when the sample is irradiated by green light, and the magnitude depends on the illuminated power density.²⁰⁴ It is noteworthy that the measurements were conducted in the high concentration range of NO₂, 10–1000 ppm. There have also been reports studying gas sensing of MoS₂ using more scalable material techniques, such as vapor phase growth of thin films²⁰⁵ and flexible transistor arrays using all solution-processable materials.²⁰⁶ 

Typically, the time-response of solid-state sensors is a challenging parameter, often showing unsatisfactory performance in ambient conditions. Poor response time and incomplete recovery at room temperature have often restricted the application of many 2D materials in high-performance practical gas sensors. Fast detection of NO₂ and reversibility have been demonstrated for MoS₂ gas sensors at room temperature.²⁰⁷ While incomplete recovery and a high response time of ≈249 s of the sensor were observed in ambient, MoS₂ exhibited an enhancement in sensitivity with a fast response time of ≈29 s and excellent recovery to NO₂ (100 ppm) under photoexcitation at room temperature. The effect was attributed to charge perturbation on the surface of the sensing layer under optical illumination. Driven by the same goal and using hybrid MoS₂ materials, Long et al.²⁰⁸ presented fast detection and complete recovery (both at <1 min) to NO₂ at 200 °C using MoS₂/graphene/hybrid aerogels. Similarly, using an MoS₂ nanosheet−Pd nanoparticle composite, Kuru et al.²⁰⁹ showed a room-temperature H₂ response/recovery time of 40/83 s, respectively. In a work by Park et al.,²¹⁰ a highly porous h-MoS₂/Pt nanoparticle (NP) hybrid, synthesized by pyrolysis, was used for H₂ detection, demonstrating 8/16 s for response/recovery time, respectively, for 1% of H₂. This response is among the fastest for 2D material-based H₂ sensors in a standard ambient environment.

While experimental detection of non-polar gases (e.g., CO₂, CH₄) with 2D material-based sensors is generally problematic due to the lack of strongly pronounced adsorption mechanisms, a modeling study using first-principles and Monte Carlo simulations revealed that single and double sulfur vacancies exhibit an excellent adsorption ability for both polar and non-polar gases.²¹¹ The simulation results showed that MoS₂ with a single S vacancy could absorb 42.1 wt. % of CO₂ and 37.6 wt.% of CH₄ under a pressure of 80 bar at room temperature.

As mentioned above, heterostructures combining a 2D material and some other active component(s) often demonstrate an enhanced performance for gas sensing compared to their individual counterparts. For example, MoS₂/Co₃O₄ thin film sensors were implemented for ultra-low-concentration detection of ammonia at room temperature,²¹² revealing high sensitivity, good repeatability, stability, selectivity, and fast response/recovery characteristics. Zhang et al.²¹³ demonstrated a hydrogen gas sensor based on a complex Pd-SnO₂/MoS₂ ternary hybrid. The experimental results showed a response within 1%–18% resistance change, swift response-recovery time (10–20 s), good repeatability, and high selectivity toward hydrogen gas in the range of 30–5000 ppm at room temperature.

In comparison to MoS₂, WS₂ has been significantly less explored for gas sensing applications, with the most notable works including complex 3D geometries assembled of 2D nanostructures. For example, a WS₂ nanoflake-based sensor showed a good sensitivity to ammonia (1–10 ppm) at room temperature with the response/recovery time of 120/150 s, respectively. Interestingly, the sensor also demonstrated excellent selectivity to formaldehyde, ethanol, benzene, and acetone.²¹⁷ A sensor made of WS₂ nanosheets with size of 10 nm exhibited selectivity toward NO₂ and H₂S,²¹⁸ where the presence of oxygen and elevated temperatures (160 °C) were necessary for H₂S sensing. An interesting approach to increase the surface area (and thus number of active sites for molecular adsorption) of atomically thin TMDs was demonstrated by synthesizing 3D WS₂ nanowalls combined with CdSe–ZnS quantum dots (QDs), with a demonstrated detection limit of 50 ppb for NO₂ and a fast response time of ≈26 s.²¹⁹ 

In general, many 2D disulfides have proven selectivity toward NO₂. A sensor based on SnS₂ platelets demonstrated high sensitivity (600 ppb measured) and superior selectivity to NO₂²²⁰ due to the unique physical affinity and favorable electronic band positions of SnS₂ and NO₂. Density functional theory (DFT) studies of NbS₂ also suggest its high selectivity toward NO₂²²¹ depending on its edge configuration. In a recent work, a monolayer Re_0.5Nb_0.5S₂ sensor demonstrated high sensitivity, selectivity, and stability toward NO₂ detection, accompanied by only minimal response to other gases, such as NH₃, CH₂O, and CO₂.²²² In the presence of humidity, the monolayer sensor showed complete reversibility with fast recovery at room temperature. Including other TMDs, the fast response/recovery rate accompanied by enhanced sensitivity under UV illumination was achieved in a p-type MoTe₂ gas sensor for NO₂ detection, demonstrating full recovery within 160 s after each sensing cycle at room temperature [Fig. 10(b)].²¹⁴ The sensitivity of the sensor to NO₂ was significantly enhanced by an order of magnitude under 254 nm UV illumination as compared to that in the dark, leading to a remarkably low detection limit of 252 ppb (as derived from noise measurements).

Phosphorene and blue phosphorus: Following the initial prediction of superior gas sensing applications using first-principles calculations,²²³ semiconducting phosphorene (or 2D black phosphorus) has been proven to have great potential for gas sensing of CO, CO₂, NH₃, and NO (see, e.g., Liu et al.²²⁴). Abbas et al.²¹⁵ experimentally demonstrated phosphorene-based FETs for NO₂ detection, achieving a lowest detectable concentration of 5 ppb [Fig. 10(c)]. Phosphorene functionalized with gold has been shown to display selectivity to NO₂ (compared with H₂, acetone, acetaldehyde, ethanol, hexane, and toluene).³⁷ Although pristine phosphorene is typically insensitive to H₂, functionalization with Pt NPs led to improved H2 sensing efficiency, as seen in the decreasing of the drain–source current and increase in the ON/OFF ratio at low concentrations of H₂.^37,225 An impedance-based phosphorene-based sensor has demonstrated a strong response to a low concentration of methanol vapor (28 ppm) at ≈1 kHz in impedance phase spectra, showing a high selectivity to methanol and absence of cross-selectivity from toluene, acetone, chloroform, dichloromethane, ethanol, isopropyl alcohol, and water, due to the different dielectric constants of these molecules.²²⁶ 

Sensing properties of another 2D allotrope of phosphorus—blue phosphorus—were studied by first-principles calculations with respect to the adsorption behaviors of environmental gas molecules, including O₂, NO, SO₂, NH₃, H₂O, NO₂, CO₂, H₂S, CO, and N₂.²²⁷ The calculations showed that O2 tended to chemisorb, whereas the other gases were physisorbed on monolayer blue phosphorus, showing different interaction strengths and distinct modifications to the bandgap, carrier effective mass, and work function.

h-BN: Applications of h-BN in gas sensing have remained challenging due to its inherently low chemical reactivity. Recently, a chemiresistor-type NO₂ gas sensor based on sulfate-modified h-BN has been investigated, demonstrating a linear response over a wide NO2 concentration range and low LoD of 20 ppb.⁴¹ Theoretical calculations predicted that the sulfate groups spontaneously grafted to the h-BN, effectively altering its electronic structure and enhancing the surface adsorption capability toward NO₂, in addition to a strong charge transfer between NO₂ and h-BN. Thus, applications of sulfate-modified h-BN in capturing environmentally hazardous exhaust from motor vehicles as well as combustion emissions monitoring have been proposed.

In a recent theoretical study, graphene/h-BN heterostructures were studied for detection of NO, NO₂, NH₃, and CO₂ gas molecules.²²⁸ The strongest interaction was observed for the case of NO_x, where NO and NO₂ molecules are more reactive at the interface regions of the heterostructure compared to the pristine ones, where large changes in the conductance with adsorption were seen. In addition, recent experimental work has successfully demonstrated NH3 detection with graphene/h-BN devices.²²⁹ The charge transfer from NH₃ to graphene strongly depends on the average distance between the graphene sheet and the substrate. Since the average distance between graphene and h-BN crystals is one of the smallest, the graphene/h-BN heterostructure exhibited the fastest recovery times for NH₃ exposure and revealed importance of substrate engineering for development of 2D gas sensors.

MXene: Recently a new class of 2D materials—transition-metal carbides (Ti₃C₂T_x MXene)—have received a great deal of attention for potential use in gas sensing, showing both high sensitivity and good gas selectivity. Integration of an effective superhydrophobic protection fluoroalkylsilane (FOTS) layer with MXenes has helped overcome one of the major problems typical for this class of material, which is environmental instability.²³⁰ FOTS-functionalized Ti₃C₂T_x displayed very good hydration stability in a humid environment and showed good tolerance to strong acidic and basic solutions. The experiments also demonstrated very good sensing performance (e.g., sensitivity, repeatability, stability, selectivity, and faster response/recovery time) to oxygen-containing volatile organic compounds (ethanol, acetone), as achieved in a broad relative humidity range of 5%–80%.

This review represents only a relatively small fraction of the research on the use of 2D materials sensors for gas detection. In a chart shown in Fig. 11, we summarize the recent figures of merits: LoD (a), detection range (b), and response/recovery time (c) for graphene (as well as GO and rGO) and other layered 2D materials based on the results presented here as well as recent, more specialized reviews by Buckley et al.¹⁹⁶ and Meng et al.¹⁹⁵ We limit this summary to NO₂ gas only as being generally one of the easiest substances for detection and thus the most intensely studied in literature. It is important to emphasize that in many cases, the measurements were performed in very specific physical conditions (such as temperature, illumination, etc.), which are not reflected in the chart. The red dashed/dotted lines on the chart [Figs. 11(a) and 11(b)] represent recommended annual mean limits of exposure as specified by the European Union (EU) and National Ambient Air Quality Standards (NAAQS), i.e., 21 and 58 ppb, respectively.^231,232 The area above the EU annual mean level (as highlighted by gray) demonstrates a potentially unsafe level of exposure for humans (note the exponential scale both for LoD and range). In terms of the response and recovery time, the results also vary significantly, with the recovery time being generally (in many cases, significantly) larger than the response one [Fig. 11(c)]. Thus, the conclusions driven from these graphs are unfortunately not very positive. While each individual study often shows an interesting research breakthrough and outlines favorable detection conditions, the overall picture shows that, in the majority of cases, detection is achieved in conditions which are not suitable for environmental monitoring (e.g., falls significantly above red lines), yet may still be relevant for sensors employed in the chemical or defense industries. Graphene is generally more suitable for environmental diagnostics compared to other 2D materials, both in terms of higher sensitivity to NO₂ and generally a more favorable detection range, although this might be due to the larger number of studies conducted on graphene. Focusing on non-carbon-based 2D materials, some of the results obtained on BP and MoS₂ appear to be the most promising in terms of the LoD and detection range. The comparison between non-carbon 2D materials and graphene seems to be more encouraging in terms of the response and recovery times, where semiconducting 2D materials with a bandgap (BP, MoS₂, SnS₂) are characterized by generally faster times and would be more suitable for the immediate response to changing environmental conditions. This analysis demonstrates a crucial need not only to improve performance of the sensor but also to align it with the relevant targets dictated by the environmental detection needs, which have been unfortunately disregarded in many studies.

In addition to their wide-spread application in gas sensing, 2D materials, especially TMDs, have exhibited great potential for humidity sensing. For example, monolayer MoS₂ devices on a flexible PDMS substrate have demonstrated a high humidity sensitivity >10⁴, with their mobilities and ON/OFF ratios decreasing linearly at RH = 0%–35%.²¹⁶ The authors showed an exponential increase in the resistance with a human finger moving closer to the MoS₂ device [Fig. 10(d)], with short response and decay times. In a similar approach, a hybrid ultrasensitive humidity sensor based on a MoS₂-SnO₂ nanocomposite revealed very good sensing parameters (e.g., response, fast response/recovery time and repeatability)²³³ compared to the pure MoS₂ and SnO₂ counterparts, and graphene. In another work, a complex device, which incorporated large-area WS₂ as a sensing element, graphene as an electrode and thin flexible and stretchable PDMS as a substrate was utilized for humidity sensing (up to 90%) with fast response/recovery times (in a few seconds).²³⁴ The sensor was then laminated onto human skin and showed stable water moisture sensing behaviors, enabling real-time monitoring of human breath. Another flexible humidity sensor based on VS₂ has represented a new concept of a touchless positioning interface, based on the spatial mapping of moisture.²³⁵ In general, these flexible skin-attached chemical sensors (electronic skin or e-tattoo) are of great interest for many applications. Several interesting works have also emerged exploring applications of phosphorene in humidity sensing.²²⁴ Yasaei et al.²³⁶ realized a humidity sensor made of phosphorene nanoflakes. When the sensor was exposed to flows of H₂, O₂, CO₂, benzene, toluene, and ethanol, it resulted in enhancement of the electrical response only in the presence of a water atmosphere.

The several 2D sensors summarized here have proved to be strong candidates for ultrahigh-performance humidity sensors for various applications and have demonstrated a clear path toward developing low-power-consumption, wearable chemical sensors based on 2D semiconductors.

We have shown that, similar to graphene, 2D material FET sensors bear the advantage of miniature size and good compatibility with CMOS technology. Moreover, the existence of the bandgap allows for an additional degree of electrical control of adsorption/desorption of targeted molecules. The bandgap becomes particularly useful for manipulation of the sensor's response/recovery time. It is important to note that, while individual comparison studies may confidently prove advantages of a certain 2D material to graphene for the main figures of merit, the cumulative comparison performed in this review for the case of NO₂ demonstrated an extremely broad spread of such parameters as LoD, detection range, and response/recovery time, which overall is not dissimilar for graphene and non-carbon 2D materials. Thus, an overwhelming advantage of 2D materials over graphene has not been statistically demonstrated at the current technological level.

The success of graphene-based biosensors has stimulated a large research interest in the implementation of other 2D materials in biosensing platforms. Here, we present a short overview of the current state of the art and outline the most promising directions in the area of biosensors, focusing our attention on the electrical schemes of a sensor's control and optimization. For more detailed and insightful reviews in this area, we refer to the recent works by Bollella et al.,²³⁷ Zhu et al.,²³⁸ Hu et al.,²³⁹ Meng et al.,¹⁹⁵ and Kou et al.²⁴⁰ 

Similar to gas sensing, among atomically thin 2D materials, TMDs have been the most intensely studied for electrochemical biosensing purposes.²⁴¹ Their high surface-to-volume ratio offers potential for the detection of large amounts of target analytes and low power consumption. The current state of growth is such that 2D TMDs can be readily synthesized on a large scale and can be directly dispersed in aqueous solution without the aid of surfactants, providing environmentally friendly and even biocompatible and biodegradable solutions.^{195,242–244} Figure 12 schematically summarizes the broad range of applications of 2D TMD biosensors for the detection of various molecules. Specifically owing to its high conductivity and a large number of active defects that provide sites for adsorption of biomolecules, MoS₂ is one of the most commonly used TMD materials for biosensing.¹⁹⁵ Either in its pristine form or as a part of hybrid structures/nanocomposites, MoS₂ has been used both as a platform for non-enzymatic sensing and as a biocompatible matrix for enzyme immobilization and development of both electrochemical sensors and biosensors [Figs. 12 and 13(a)].^237,245,246

MoS₂-based FETs have been successfully employed as a 2D platform for detection of various biomolecules, including streptavidin and biotin,²⁴⁸ ochratoxin,^249–251 dopamine,²⁵² anti-PSA,^253,254 TNF-α,^255,256 and bisphenol A.²⁴² We will further discuss these applications in order of increasing weight and complexity of biomolecules (Fig. 12). Specifically, for streptavidin and biotin detection (one of the strongest known binding reactions in biology), it has been demonstrated that $MoS2$-based sensors provide specific protein sensing at concentrations as low as 100 fM. Superior performance of a MoS₂-based FET biosensor to graphene counterparts has been proven, e.g., ≈70-fold better sensitivity of the MoS₂ biosensor was demonstrated^245,248 [Fig. 13(a)]. The detection of glucose, one of the most important human biomarkers, using atomically thin 2D materials has been widely explored in recent years.^257–259 In general, two common routes have been developed, implementing both enzymatic^258,260 and non-enzymatic^259,261 sensors. MoS₂ has also often been used as a platform for the development of hydrogen peroxide, H₂O₂, an essential compound involved in many biological processes.^262–266 Similar to glucose detection, enzymatic^263,264 (typically involving an electrocatalytic reaction with haemoglobin²⁶⁷) and non-enzymatic^268,269 routes have been explored.

TMDs also exhibit attractive properties for detection of small biomolecules (neurotransmitters, metabolites, vitamins, etc.). Detection and differentiation of DA (dopamine), AA (ascorbic acid), and UA (uric acid) have been demonstrated using MoS₂-sensors.^{260,270–274} TMDs were also employed as a sensing platform for detection of biomarkers, such as carcinoembryonic antigen (CEA).²⁵¹ Recently, MoS₂-based platforms were employed for detection of μ-opioid receptor, a synthetic opioid peptide, and specific μ-opioid receptor agonist.²⁷⁵ 

MoS₂-based sensors have also been utilized for detection of nucleic acids^276–278 and selective detection of dsDNA²⁷⁹ and ssDNA.²⁷⁸ where the detection mechanism was based on the different affinity of MoS₂ toward each DNA type.²⁷⁹ Liu et al.²⁴⁷ successfully used a MoS₂-Au NPs-DNA-functionalized FET-based biosensor for the screening of Down syndrome [Fig. 13(b)]. The MoS₂ FET biosensors were able to reliably detect target DNA fragments (chromosome 21 or 13) with a detection limit below 100 aM, a high response up to 240%, and a high specificity, which satisfies the requirement for the screening of Down syndrome. In another series of works, MoS₂-based FET sensors have demonstrated high sequence selectivity capable of discriminating the complementary and noncomplementary DNA,²⁸⁰ RNA and ATP monitoring,^281,282 and CEA detection.^251,283

A recent comprehensive study has highlighted important insights into the bioabsorption of CVD-grown monolayer MoS₂, including long-term cytotoxicity and immunological biocompatibility evaluations on live animal models.²⁴⁴ The authors presented MoS₂-based bioabsorbable and multi-functional sensors for intracranial monitoring of pressure, temperature, strain, and motion in animal models. A simple implantable electrical sensor based on monolayer MoS₂ was capable of monitoring intracranial temperature over a specified period before dissolving completely. They observed no adverse biological effects and verified that biodegradable MoS₂-based electronic systems offer specific, clinically relevant roles in diagnostic and therapeutic functions during recovery from traumatic brain injury.²⁴⁴ 

Although currently significantly less explored than its MoS₂ counterparts, WSe₂²⁸⁴ and WS₂ bioFETs have also been used for detection of glucose,^257,285 IgE,^286,287 steroid hormone (e.g., estradiol²⁸⁷), and nucleic acid aptamers (e.g., DNA^288,289 and micro-RNA²⁹⁰) Alternatively, working on a higher level of bio applications, wearable electronics, or skin tattoos based on PtSe₂ and PtTe₂ with medical-grade Ag/AgCl gel electrodes has been demonstrated.²⁹¹ Specifically, in terms of sheet resistance, skin contact, and electrochemical impedance, PtTe₂ outperforms state-of-the-art gold and graphene electronic sensors. The PtTe₂ tattoos show 4 times lower impedance and almost 100 times lower sheet resistance compared to monolayer graphene tattoos, opening exciting applications in the development of advanced human–machine interfaces.²⁹¹ 

An interesting comparative analysis of TMD- and graphene-based biosensors has been performed recently by Bollella et al.²³⁷ It concluded that MoS₂-modified graphene platforms have shown the best results in terms of sensitivity.^259,261 In the case of H₂O₂ detection, the best electrochemical sensing electrode was realized with a MoS₂-CNT nanocomposite, which shows a wide linear range, the lowest LoD and the highest sensitivity.²⁶⁸ The presence of incorporated metal NPs has been shown to improve the electrochemical performances of all sensors, where the MoS₂-based sensing platforms displayed the best results among the other 2D materials employed.²³⁷ Regardless, a comparative analysis (LoD and detection range) of a large number of graphene, GO, rGO, and MoS₂-based electrochemical glucose sensors (based on the results presented here as well as adapted from the comprehensive recent review by Meng et al.¹⁹⁵) do not demonstrate an obvious advantage of one material against the other [Fig. 13(c)]. The comparison shows that both the LoD and detection range vary significantly for each material. It is important to note that, although all compared materials demonstrated a generally low LoD, only a limited number of reported cases matched the clinically relevant detection range for glucose in blood.²⁹² Still, both carbon and MoS₂ based sensors could be a good platform for detection of glucose in urine, where significantly lower (only traces of the substance) levels are expected.²⁹³ This example demonstrates a supreme importance of targeting development of 2DM sensors with a clear view on the application niche and on the underlying requirements.

Compared to TMDs, other 2D materials, such as black phosphorus (BP), boron and silicon nitrides, Mxenes, silicenes, etc., have been significantly less explored for biosensing applications. In many cases, the research is still limited to theoretical studies and predictions.

Black phosphorus: BP is a promising candidate for biosensing due to its inherent conductivity, biocompatibility, and electrocatalytic properties. Recently, BP-based sensing platforms have been used for the detection of human immunoglobulin (IgG) and anti-IgG,²⁹⁴ H₂O₂ (through immobilization of hemoglobin),²⁹⁵ myoglobin (iron- and oxygen-binding protein),²⁹⁶ and leptin (a protein hormone, which is an important biomarker for liver diseases).²⁹⁷ However, the primary limitation for the use of BP in bio-applications is its relatively quick degradation due to moisture absorption and oxidation.

h-BN: Compared to BP, use of h-BN in biosensing provides the important advantage of chemical stability. Additionally, the good electrocatalytic performance of h-BN is highly beneficial for the development of electrochemical sensors. For example, h-BN sensors have been used for both enzymatic²⁹⁸ and non-enzymatic²⁹⁹ detection of H₂O₂. h-BN electrodes support a high overpotential required for DA oxidation and have led to successful detection of DA in the presence of UA.^300,301 Furthermore, the simultaneous presence and differentiation between DA, AA, and UA has been demonstrated^302,303 and complemented by the development of a non-enzymatic glucose sensor.³⁰⁴ A h-BN based biosensor has also been employed for detection of an important neurotransmitter called serotonin.³⁰⁵ 

MXenes: They have attracted significant research interest due to their metallic conductivity, hydrophilic surfaces, and good stability in aqueous environments. Ti₃C₂T_x-based FETs have been used for the detection of H₂O₂ (via immobilization of haemoglobin),^306,307 monitoring of hippocampal neurons (responsible for transmission of brain signals relevant to learning, emotions, and memory),³⁰⁸ and fabrication of an enzymatic glucose biosensor.³⁰⁹ Moreover, Pt-doped Ti₃C₂T_x biosensors have been demonstrated for detection of DA, AA, UA, and acetaminophen.³⁰⁷ Furthermore, MXene-Ti₃C₂T_x-based composites have been utilized for detection of nitrites in environmental water.³¹⁰ 

The current status of this rapidly developing field forebodes that 2D material-based bio- and electrochemical sensors can be widely used for detection of disease biomarkers in diagnostics and disease monitoring. Due to their advantageous physical and chemical properties (such as tunable conductivity, large surface area, biocompatibility, and electronic anisotropy), 2D materials are well suited for continuous and real-time monitoring of specific molecules, even in complex environments such as the interior of living cells or blood serum.²⁴⁰ Among other detection mechanisms, electrical schemes of detection remain the most robust and versatile for ON/OFF biosensing platforms. Most commonly, they exploit FET-based platforms, which allow an additional degree of freedom (typically either through a back gate or top liquid gate) to tune the device sensitivity and response.

There are still a significant number of obstacles to overcome, biological specificity being one of the most crucial, where a large number of interfering substances in biological fluids may impact the accuracy and specificity of detection. The critical and reliable evaluation of the toxicity and biocompatibility of 2D materials is essential for in vivo applications. On a physical side, the limited understanding of the influence of the structural and compositional defects on sensing properties complicates reproducibility and device optimization.¹⁹⁵ Future developments should include novel synthetic methods, with a large degree of structural control. The development of novel hybrid materials and composites, e.g., by addition of other electroactive components, such as metal oxides, metals, graphene, or conductive polymers, is a huge advantage and has already been widely incorporated into the design of a new generation of biosensors. The improved catalytic activity and low-cost of such complex 2D hybrid materials make them a useful biocatalyst for multiple applications in environmental chemistry, biotechnology, and clinical diagnostics.

Recently, the application of 2D materials in micro/nanoelectromechanical systems (MEMS/NEMS) and energy conversion devices, active flexible sensors, actuators, and more, has relied on the inherent piezoelectric property in their atomically thin crystal structure. Here, we discuss several fascinating developments, mainly related to pressure, strain sensors, and human-computer interfacing, where carrier generation, transport, recombination, or separation is tuned through electrical or mechanical stimuli. For comprehensive reviews in the area, we address the readers to recent publications,^197,311 see also Sec. IV C: Piezo- and Ferro-electricity in 2D vdW materials for details.

MoS₂ is probably the most explored piezoelectric semiconductor. The bandgap of MoS₂ is highly strain-tunable, which results in the modulation of its electrical conductivity and manifests itself as the piezoelectric effect. The piezoelectric effect is generally studied through application of external strain to devices, for example, through the use of flexible substrates, where the electrons can be driven to flow into an external circuit when stretching the substrate or flow back when releasing it.³¹² To quantify the piezoelectric response experimentally and measure the piezoelectric coefficient (e₁₁), a free-standing monolayer of MoS₂ has been investigated using a combination of AFM probe-based nano-indentation and a laterally applied electric field with e₁₁ = 2.9 × 10⁻¹⁰ C m⁻¹.³¹³ The effect of the number of layers (i.e., odd vs even) and orientation of crystals was also demonstrated. We further discuss applications of TMD materials in MEMS/NEMS, nanogenerators, and strain-based humidity sensors.

TMDs are model materials for MEMS/NEMS due to their atomically thin nature and coupling between electrical and mechanical properties (see, e.g., Manzeli et al.³¹⁶ or Wagner et al.³¹⁷). The calculated piezoresistive gauge factor was found comparable to state-of-the-art silicon strain sensors and higher than those based on suspended graphene. Electromechanical piezoresistive sensors were also realized in relatively little explored 2D PtSe₂.³¹⁷ In this work, high negative gauge factors of up to −85 were achieved experimentally in PtSe₂ strain gauges. Integrated NEMS piezoelectric pressure sensors with freestanding PMMA/PtSe₂ membranes have been realized and exhibited very high sensitivity superior to previously reported devices. The low temperature growth makes PtSe₂ compatible with CMOS technology, which is particularly attractive.

MoS₂ has also been widely studied for applications in nanogenerators,^318,319 providing a new way to effectively harvest mechanical energy for low power-consuming electronics and realizing self-powered sensors. For example, in one of the earlier works, it was demonstrated that a monolayer MoS₂ device on a flexible substrate under 0.53% strain produced a voltage of 15 mV and a current of 20 pA, corresponding to a power density of 2 mW m⁻² and a 5% mechanical-to-electrical energy conversion efficiency.³¹⁸ In a separate study, it was shown that under applied strain of 0.48%, the output power of an MoS₂ nanogenerator in the armchair orientation was about twice higher than that in the zigzag orientation.³²⁰ 

Realization of a novel type of MoS₂ humidity sensor was recently demonstrated, where exploitation of the piezoelectric effect allowed for a simple and stable way to enhance the sensor's sensitivity.^216,321 The authors showed that tensile strain generated in the sensor led to a larger current output and an enhanced sensitivity to humidity. The observed output current and humidity sensitivity were both enhanced when more electrons are moved to the conduction band under tensile strain at a positive gate bias.¹⁹⁷ The tunability of the sensor by strain was better achieved in a low humidity range, which was attributed to a better manifestation of piezoelectric effect when number of water molecules absorbed on the channel surface was small.

InSe: Although less explored than in TMDs, the piezoelectric effect in other 2D materials is an emerging field with many promising outcomes. Piezoelectric outputs up to 0.363 V for a few-layer α-In₂Se₃ device with a current responsivity of 598 pA for 1% strain were experimentally demonstrated, outperforming other 2D piezoelectrics by an order of magnitude. Self-powered piezoelectric sensors made of these 2D layered materials were successfully applied for real-time health monitoring [Figs. 14(a)–14(c)].³¹⁴ In another work, strain sensors produced from large-scale CVD-grown In₂Se₃ exhibited two orders of magnitude higher sensitivity (gauge factor ≈ 237) than conventional metal-based (gauge factor ≈1–5) and graphene-based strain (gauge factor ≈ 2–4) sensors under similar uniaxial strain.³²² Additionally, the integrated strain sensor array, fabricated from the template-grown 2D In₂Se₃ films, displayed a high spatial resolution of ≈500 μm in strain distribution, making this material platform highly attractive as e-skins for robotics and human body motion monitoring.

MXenes: Emerging biowearables, various human–artificial intelligence (AI) interfaces, and soft exoskeletons urgently require high-performance strain sensors satisfying multiple sensing parameters, such as high sensitivity, reliable linearity, and tunable strain ranges.³²³ Recently, a number of fascinating studies have emerged exploring application of MXene hybrids in wearable electronics. Cai et al. demonstrated that a percolation network based on Ti₃C₂T_x MXene/CNT composites could be designed and fabricated into versatile strain sensors.^315,324,325 The weaving architecture combined good electric properties and stretchability (attributed to the CNTs' network) and sensitivity of 2D Ti₃C₂T_x MXene nanoplatelets. The resulting strain sensor was characterized by an ultralow detection limit of 0.1% strain, high stretchability (up to 130%), high sensitivity (gauge factor up to 772.6), tunable sensing range (30%–130% strain), and excellent reliability and stability (>5000 cycles) [Figs. 14(d) and 14(e)]. The versatile and scalable Ti₃C₂T_x MXene/CNT strain sensors were proposed as a material platform for wearable AI, capable of tracking physiological signals for health and sporting applications in real-time. In a similar approach, wearable aerogel sensors that combined insulating 1D aramid nanofibers (ANFs) with conductive 2D MXene sheets demonstrated ultra-light weight, wide sensing range, and good sensing ability. The resulting MXene/ANFs aerogel sensor showed a wide detection range (2.0% to ≈80.0% compression strain), sensitivity (128.0 kPa⁻¹) in the pressure range of 0–5 kPa, and ultralow detection limit (0.1 kPa), opening applications in detecting human motions ranging from a light movement to vigorous loads in extreme sports. The MXene/ANFs aerogel with excellent integrated ability was proposed as a potential candidate for a human behavior monitoring sensor, as well as for sensing under extreme conditions. In a recent advanced approach, Yang et al. realized wireless Ti₃C₂T_x MXene strain sensing systems by developing hierarchical morphologies on piezoresistive layers and integrating the sensing circuit with near-field communication technology.³²³ The wireless MXene sensor system could simultaneously achieve an ultrahigh sensitivity (gauge factor > 14 000) and reliable linearity (≈0.99) within multiple user-designated high-strain working windows (130%–900%). The wireless, battery-free MXene e-skin sensing system was able to collectively monitor the multisegmented exoskeleton actuations via a single database channel and was successfully used to assist limb rehabilitation. Other promising recent examples include CNTs and Ti₃C₂T_x MXene for the monitoring of human activities,³²⁴ breathable Ti₃C₂T_x MXene/protein nanocomposites as a medical pressure sensor,³²⁵ and hetero-dimensional 2D Ti₃C₂T_x MXene and 1D graphene nanoribbon hybrids for machine learning-assisted pressure sensors.³²⁶ 

Graphene nitride: Anomalous piezoelectricity in 2D graphene nitride nanosheets (g-C₃N₄) has recently been demonstrated, where g-C₃N₄ was chosen because it naturally possesses uniform triangular nanopores and has advantageous piezoelectric properties (e.g., the linear relationship between piezoelectric response and applied voltage and the effective vertical piezoelectric coefficient of ≈1 pmV⁻¹).⁵³ 

Germanene: By means of first-principles calculations, it was shown that in an AlAs/germanene heterostructure, both electric field and strain could be used to tailor its electronic bandgap and dielectric function. Under a negative electric field and compressive strain, the material's bandgaps showed a near-linear decreasing behavior, whereas a dramatic and monotonic decrease in the bandgap as a response to a positive electric field and tensile strain was shown.³⁴ It was also predicted that the optical properties of the heterostructure could be improved by electric field and mechanical strain.

The experimental observations of piezoelectrical phenomena have yet to be fully demonstrated for all the materials predicted to be piezoelectric. In some advanced cases, such as 2D MoS₂ and MXenes, their piezoelectric properties have already enabled active sensing, actuating, and new electronic components for nanoscale devices. Further applications of other 2D materials are expected to emerge in self-power nanodevices, adaptive biosensors, e-skins, and tunable and stretchable electronics. Still, the remaining challenges remain substantial. Most importantly, the existence of interface states and disappearance of piezoelectric properties with thickness remain significant obstacles on the way to device realization. Another vital challenge is improving material stability, which should be addressed by further advances in materials processing and encapsulation technologies.

The presence of a bandgap in many 2D materials lends them much more promise for optoelectronic applications than the gapless graphene. In contrast to bulk semiconductors, these materials feature very weak dielectric screening and strong spatial confinement of charge carriers. A consequence of this is the rich variety of excitonic quasiparticles which exist in these materials, particularly in 2D Mo and W chalcogenides and perovskites. The high binding energies of these quasiparticles mean that they persist even at room temperature dominating the optoelectronic behavior. This makes them exciting candidates for a range of applications, including light detection and emission, as well as spin- and valleytronic devices.

Beyond the neutral exciton, quasiparticles observed in TMDs include the trion, or charged exciton,^327–329 as well as higher order excitonic particles, such as neutral and charged biexcitons.^330–334 The relative population of different excitonic species can be effectively controlled in gated devices, making them suited for electrically tunable optoelectronic applications. Additionally, the large spin–orbit coupling and circular dichroism in TMDs results in a spin splitting at the K and K′ valleys of opposite sign, enabling selective population of excitons in a particular valley by optically pumping with either right- or left-hand circularly polarized light.^335–339 

The basic physics of excitons in 2D materials has been studied extensively by other authors, and for more details, we point readers to other reviews.^139,340,341 In the rest of this section, we shall therefore limit our discussion to only more unusual excitonic quasiparticles: the interlayer exciton and the exciton–polariton.

A heterobilayer is formed of two distinct 2D materials, layered together in a vdW heterostructure. For TMDs, the range of different bandgaps and work functions means heterobilayers typically form vertical p–n junctions, with type II band alignment. This means that electrons and holes in the heterostructure have energy minima in different materials. Experimental measurements, with techniques including angle resolved photoemission spectroscopy and scanning tunneling spectroscopy, have confirmed that this is the case for a wide range of TMD heterobilayers.^342–345 The band structure of the heterobilayers is then further altered by hybridization of the band structures of its constituents, an effect which is governed by the momentum-varying interlayer hopping potential³⁴⁶ and by the superposition of a moiré period for non-lattice matched and rotated crystals.

In these heterobilayers, an interlayer exciton (IX, often referred to as an indirect exciton) can be formed by the creation of a regular intralayer exciton (DX, or direct exciton) in either layer, followed by interlayer charge transfer, leaving a Coulomb-bound electron-hole pair with each carrier in a different material. The resulting exciton configuration is shown schematically in the insets to Fig. 15(a). A great advantage of interlayer excitons is significantly enhanced lifetimes due to the spatial separation of the carriers.³⁴⁷ For charge transfer to be energetically favorable, the binding energy of the interlayer exciton must be lower than that of the intralayer exciton [Fig. 15(b)]. This means that interlayer excitons can be observed in photoluminescence (PL) spectra as a lower energy peak, arising from the heterobilayer [Figs. 15(a) and 15(c)].^347–352 

Similar to the case of intralayer excitons, valley polarization has been observed in interlayer excitons.^353–355 It has also been demonstrated that the charge transfer process, which converts intralayer to interlayer excitons, conserves the spin-valley polarization of the excited charge carriers.^356,357 However unlike for intralayer excitons, these valley-polarized states are orders of magnitude more persistent, with lifetimes on the order of microseconds. This makes interlayer excitons ripe for exploitation in valleytronic devices.

The properties of interlayer excitons can be further tuned by electrical gating. The broken inversion symmetry of a heterobilayer means that the relative band offsets of the constituent 2D materials can be adjusted via the field effect. This, in combination with the out-of-plane electric dipole possessed by interlayer excitons, allows effective external control of the exciton energy. Gated WSe₂/MoSe₂ devices have shown gate-dependent energy shifts ≈70 meV.³⁴⁷ Additionally, time- and polarization-resolved PL revealed significant gate-dependent changes in the lifetimes of interlayer excitons and polarized states [Fig. 15(d)].³⁵³ 

Beyond electronic tunability, it has recently been demonstrated that the properties of interlayer excitons can also be altered by the twist angle between two layers through modification of the period of the moiré pattern caused by the overlapping lattices.^{350,358–361}

The family of 2D-material quasiparticles is further expanded with the addition of exciton–polaritons. These are hybrid light-matter quasiparticles, resulting from strong coupling between an electromagnetic wave and the electric dipole associated with an exciton. This strong-coupling regime is enhanced by the large exciton-binding energies and sharp resonances in TMDs, meaning these exciton–polaritons persist even at room temperature. These materials are therefore interesting platforms to study strong light-matter interactions.

Strong coupling is typically achieved by placing 2D materials in a photonic microcavity between two mirrors [Fig. 16(a)], which concentrates the local light intensity by exciting a cavity resonance.^362–365 In this geometry, exciton–polaritons in monolayer WSe₂ have shown room temperature valley coherence, whose phase can be effectively controlled via the Zeeman effect, through application of a magnetic field.³⁶⁶ Additionally, in a cavity-embedded field-effect device combining WS₂ and MoS₂, gate-controlled, polariton-mediated energy exchange was demonstrated, between the excitons originating from each 2D material [Fig. 16(b)].³⁶⁷ 

An alternative method of achieving strong light-matter coupling is through the use of scanning near-field optical microscopy (SNOM). This uses tightly focused near-field light focused at the apex of an atomic force microscope probe to both excite and detect exciton–polaritons, enabling high-resolution spatial mapping [Fig. 16(c)]. In multilayer slabs of TMDs, thickness-dependent internal waveguide resonances can be excited, which can interact strongly with exciton–polaritons. SNOM studies of WSe₂³⁶⁸ and MoSe₂³⁶⁹ have shown that this results in exceptionally long polariton propagation lengths of over 12 μm [Fig. 16(d)], as well as thickness-tunable polariton wavelengths.

Various further methods have been demonstrated to induce exciton–polaritons, without the need for a cavity or SNOM tip. WS₂ nanodiscs have been shown to support internal Mie resonances and novel anapole states, which couple strongly with excitons, forming polaritons whose energy can be tuned via the disk radii.³⁷⁰ In monolayer WS₂, strong coupling has been demonstrated between excitons and waveguide modes, formed by patterning the TMD into a photonic crystal.³⁷¹ Additionally, WS₂ nanogratings fabricated on gold have shown strong coupling between excitons, cavity modes, and plasmon polaritons.³⁷² 

For further detail on the physics behind exciton–polaritons, we point readers to previous reviews on 2D material polaritonics.^373,374

Magnetism in 2D vdW materials has been sought after for engineering ultra-scaled magnetic devices and tunable magnetic phenomena in low dimensions. Theoretically, intrinsic magnetism in atomically thin materials was believed to be prohibited due to enhanced thermal fluctuations, as per the Mermin–Wagner theorem.³⁷⁵ Therefore, previous efforts were focused on extrinsically inducing magnetism in 2D vdW materials through defect engineering, doping, intercalation, and/or band structure engineering.^376–381 Recently, however, it was discovered that several atomically thin vdW materials do, in fact, sustain long-range magnetic order through the inclusion of magnetic anisotropy that opens up a spin wave excitation gap, thereby suppressing the thermal agitations and resulting in finite Curie/Neél temperatures.^13,14,384 The discovery of intrinsic ferromagnetism (FM) and antiferromagnetism (a-FM) in 2D vdW materials provides unprecedented opportunities for studying various exotic properties such as spin fluctuation-driven generation of new quantum phases and topological orders. In addition, it provides an ideal material platform for experimentally realizing ultrathin 2D spintronic devices for sensing, quantum information, and memory applications.^385–387 Magnetic 2D vdW materials are easily integrable into heterostructures without the need of lattice matching, making them suitable for exploring emergent interfacial phenomena such as multiferroicity, quantum anomalous Hall effect, and unconventional superconductivity.^385–387 The 2D heterostructures also provide access to various external stimuli such as optical, electrical, and mechanical tuning of their physical properties. In particular, electrical manipulation of magnetism in 2D vdW materials provides an exciting opportunity for realizing low-power and high-speed spintronic devices compatible with existing semiconductor technology. In this section, we particularly focus on the recent advances and current understanding of electrically manipulating magnetic order of 2D vdW materials and realizing new functional devices.

Electrical control of magnetism through electrostatic gating has been demonstrated in several recently discovered 2D vdW magnets. The carrier density or the Fermi level position of a 2D magnet can be modulated, which in turn affects the magnetic exchange interactions and magnetic anisotropy. However, this state-of-the-art approach is volatile as it requires persistent electrical control. Ferroelectric switching in magnetoelectric–multiferroic systems, where an applied electric field modifies the magnetization in 2D magnets by coupling through electrical polarization, is considered to be an effective approach for achieving nonvolatile electrical control of magnetism for practical applications.

Manipulation of magnetism through electrostatic gating was first demonstrated in few-layer, insulating 2D magnets, such as CrI₃⁵⁶ and Cr₂Ge₂Te₆.³⁶ In particular, atomically thin CrI₃ has been shown to exhibit intriguing layer-dependent magnetic order: each monolayer of CrI₃ has a FM ordering while the stacking between the layers is a-FM. The weak interlayer exchange interactions are easily susceptible to external electrical perturbation, thus enabling a unique route for controlling magnetic order in CrI₃.¹¹ Both the linear magnetoelectric effect (ME) and electrostatic doping have been reported for tuning magnetism using single-gated or dual-gated field-effect devices [Fig. 17(a)]. The magnetism was probed by the magnetic circular dichroism (MCD) or magneto-optical Kerr effect (MOKE) microscopy. For the linear ME effect, bilayer CrI₃ with graphene and h-BN as gate electrodes and dielectric, respectively, in a dual gated device was used.⁵⁶ An increase in both the magnetization in the a-FM stacking state and the critical magnetic field for a spin-flip transition (H_c) was observed with increasing gate voltage due to the linear ME coupling effect. A remarkable phenomenon of complete switching of the magnetic order from FM stacking to a-FM stacking has also been demonstrated near H_c through the application of an external electric field [Fig. 17(b)]. Tuning of the magnetic order by the linear ME effect is only possible in samples with an even number of layers that exhibit both broken time reversal and spatial inversion symmetries. In monolayer CrI₃, spatial inversion symmetry is still present; therefore, no tuning of the magnetism through linear ME effect can be observed.⁵⁶ 

In contrast to the linear ME effect, electrostatic doping can control magnetism in both even and odd numbers of layers of CrI₃ by controlling the doping density with gate voltage.⁵⁷ In monolayer CrI₃, the magnetic order was found to be strengthened by hole doping and weakened by electron doping, where significant tuning of the coercive force (H_c) up to ≈75%, saturation magnetism (M_s) up to ≈40%, and Curie temperature (T_c) up to ≈20% was achieved [Fig. 17(c)]. In bilayer CrI₃, H_c could be continuously decreased by increasing the electron doping density (n) until, above a certain critical density n ≈ 2.5 × 10¹³ cm⁻², a complete transition from an a-FM to a FM stacking state was attained due to a significant decrease in the interlayer exchange coupling [Fig. 17(d)]. It can be inferred that electrostatic gating induces both linear ME and doping effects simultaneously. Thus, the exact mechanism for tuning magnetism through electrostatic gating remained elusive, until a recent study investigated both effects independently.³⁸⁸ They found that electrostatic doping played a major role in controlling the magnetic order in bilayer CrI₃ [Fig. 17(e)]. This shows that doping could be an efficient and more general approach to control the magnetism in 2D vdW magnets.

Furthermore, both ionic liquid [N,N-diethyl-N-methyl-N-(2-methoxyethyl) ammonium bis(trifluoromethylsulfonyl)imide (DEME-TFSI)] as well as thermally grown silicon oxide gate-dielectric on a degenerately doped Si back gate electrode have been used for electrostatic tuning of magnetism in vdW $Cr2Ge2Te6.$⁵² The doping density with ionic liquid gating is usually $≈100$ times higher than what can be achieved with a solid-state thermally grown silicon oxide gate dielectric on a degenerately doped Si wafer which serves as the back gate electrode.³⁸⁹ A large tuning of the magnetism with ionic liquid gating was attained for thicker $(≈19 nm)$ $Cr2Ge2Te6$ samples, which could not be obtained with thermally grown silicon oxide gate-dielectric on a degenerately doped Si back gate electrode. An ultrahigh-sensitive MOKE setup was used to investigate the magnetization, and the saturation field (H_sf) was found to reduce by a factor of two at gate voltage (V_g) = −4 V, compared to that measured at V_g = 0 V [Fig. 17(f)]. However, for thinner ≈3.5 nm Cr₂Ge₂Te₆, solid Si gating was used. Both electron and hole doping resulted in enhanced M_s that was also consistent with first-principles calculations [Fig. 17(g)]. Moreover, electron doping resulted in greater magnetization than hole doping due to shifting of the Fermi level into the conduction band by filling Cr-d orbitals that gives rise to larger magnetizations than the p orbital for the valence band from Te atoms.⁵² 

Compared with CrI₃ and Cr₂Ge₂Te₆ ferromagnetic insulators, 2D magnetic metals such as MnSe₂,³⁸³ VSe₂,³⁹⁰ and Fe₃GeTe₂³⁸² exhibit high Curie temperatures. This provides an ideal material platform, wherein coexisting itinerant electrons and local magnetic moments enable interplay of both spin and charge degrees of freedom. As of yet, there are only a few reports that demonstrate gate tunability of the magnetic properties of 2D metallic magnets. Recently, it has been demonstrated that doping induced by ionic gating can elevate the Curie temperature from 100 K to room temperature in thin Fe₃GeTe₂ flakes [Fig. 17(h)].³⁹¹ Under gate voltage, a high electron doping density $n≈1014 cm−2$ was induced in thin flakes (3 layers) of Fe₃GeTe₂, that led to modulation of the Curie temperature up to $300 K$ accompanied by a large modulation in the coercivity (μoHc) ≈ 0.6 T.

Although MnSe₂, VSe₂, and Fe3GeTe₂ have been shown to be room temperature ferromagnets, as of yet, there has not been a report of a semiconducting 2D intrinsic ferromagnet at room temperature. Thus, there is a significant effort to impart magnetism to non-magnetic materials through the creation of dilute magnetic semiconductors. There are numerous theoretical papers focused on introducing magnetic dopants into various 2D materials. Some examples include predicted ferromagnetism in Mn-, Fe-, Co-, V-, Cr-, and Zn-doped MoS₂,^392–394 V-doped WSe₂,³⁹⁵ and Co-doped phosphorene.³⁹⁶ Specifically, for Co-doped phosphorene, the authors demonstrated that the exchange interaction and magnetic ordering could be tuned by adding holes or electrons to the system (e.g., by gating), which may result in a FM to a-FM transition. Experimentally, however, the challenges in synthesizing dilute 2D magnetic semiconductors lie in preventing interstitial substitutions, clusters, or alloy formations, which would result in drastically different electronic properties of the host material, and magnetism not inherent to the material. Recent reports have successfully demonstrated room temperature FM and confirmed uniform alloying through electron microscopy in Mn-doped MoS₂ (≈3 at. %), Fe-doped MoS₂ (≈0.5 at. %), V-doped WS₂ (up to 12 at. %), and V-doped WSe₂ (≈0.1 to ≈4 at. %).^397–401 However, thus far, the tuning of magnetism via electric fields has only been demonstrated for V-doped (0.1 at. %) WSe₂ by Yun et al.⁴⁰⁰ They studied the phase contrast of ferromagnetic domains under back-gate biases from −10 to 20 V using magnetic force microscopy, observing non-monotonic variations in the phase contrast between domains at different gate biases. Although growth of these TMD-based dilute magnetic semiconductors is currently limited to micrometer-sized individual domains, they provide the foundation to explore applications in room temperature spintronic devices.

a-FM semiconductors, such as MnPSe₃, FePS₃, etc., are theoretically predicted to exhibit transitions from a-FM semiconductor to FM half-metal with both electron and hole doping.^402,403 The spin polarization direction doping that can be controlled through an external gate voltage is opposite for electrons and holes [Fig. 17(i)]. Such an a-FM to FM transition may provide a new means of magnetization switching for memory devices.

To give an overview of possible magneto-optical applications, a comparison between the Curie/Neel temperature and bandgaps of the 2D magnetic materials explored so far, as well as other bulk ferromagnetic materials and insulators, is plotted in Fig. 18.^{52,57,383,390,391,404–418}