On-chip integrated photonic circuits based on two-dimensional materials and hexagonal boron nitride as the optical confinement layer

Currently, optical communications are widely used in interchip communication in information storage and high-speed computing on account of their considerably broader bandwidth, higher operating speeds, and lower energy consumption. The first commercial optical fiber communication system was successfully developed in the 1970s.¹ Optical fiber systems have high data rate and bandwidth and low transmission decay; hence, they increasingly replace electronic circuits and are regarded as the cornerstone of fifth-generation communication networks. In 2018, communication networks are expected to have a speed of 400 Gb/s in cloud data centers. Fueled by the high demand of cloud computing and data sharing, this is an area of exponential growth.

Optical communication bandwidth is highly suited to large-scale data processing. Complementary metal–oxide–semiconductor (CMOS) technology has reached the limits of Moore’s law, with transistor size being reduced below 10 nm. Currently, data processing speed is limited by the transmission speed of interconnectors. For high-data-rate (over 100 Gb/s) communications, such as cloud data center interconnects and chip-to-chip communication, silicon photonic devices and electronic transistors should be integrated onto the same chip because high-frequency metal interconnects connecting electronic to optical devices, with connection length above 10 mm^2,3 (which is the typical size of an electronic transistor chip), suffer from significant energy loss and time delay. Combining optical communication with electronics in optical devices should be the best strategy to meet next generation’s demand for high speed, wide bandwidth, and parallel computation. Although silicon photonic devices are excellent optical platforms for light propagation and modulation, a fully functional photonic device system requires active optical materials serving as on-chip light sources, optical modulators, and photodetectors (PDs) that are compatible with CMOS technology.

In 2004, a new class of optical materials, namely, two-dimensional materials, came into the limelight. Due to their unique properties, it is quickly realized that they have the potential to push the limits of silicon photonic devices if these can be grown on silicon. 2D materials were discovered by mechanical exfoliation of layered bulk crystals bonded by Van der Waals (VdW) forces.⁴ Graphene is the archetypical example of these, as it is a single-layer carbon atomic sheet with a honeycomb lattice that defines a unique linear electron dispersion relation.⁵ Its electron transport has the characteristics of 2D Dirac fermions, whose signature is the room temperature observable quantum Hall effect.⁶ The linear dispersion of Dirac fermions allows graphene to exhibit ultrafast and broadband optical absorption behavior, which has been successfully applied in ultrafast laser pulse generation.⁷ 2D plasmon polaritons have also been discovered in graphene 2D electron gas, whose polariton wavelength is in the mid-infrared,^8,9 suggesting a deeper subwavelength plasmon resonance than in conventional surface plasmonic structures.¹⁰ These unique properties have made graphene an interesting candidate of photonic material.

Group-VI transition metal dichalcogenides (TMDs) distinguish themselves from graphene by possessing optical bandgap that can be in the visible to IR range. The justification for using the monolayer material is that most bulk TMDs have indirect bandgaps (e.g., MoS₂ and WS₂), whereas monolayer TMDs are direct bandgap semiconductors with bandgap covering the visible and near-infrared.^11,12 This semiconductor family includes MX₂ (M = Mo, W, Ta, Ti, and Nb; X = S, Se, and Te). There are also TMDs with different chemical formula, such as GaX and In₂X₃, where X = S, Se, and Te. The lack of inversion symmetry in monolayer semiconductors lifts the degeneracy of the K and K′ valleys and give rise to valley- and spin-dependent optoelectronic properties, so that oppositely circularly polarized photons can resonate at different band valleys.¹³ Moreover, owing to quantum confinement, 2D electron gases can be formed in TMDs, and it can also have large exciton binding energy at room temperature.^14–17 The intrinsic strong index modulation potential of 2D materials originates from the lack of screening of the vertical electric field at the 2-D limit, and light-matter interaction can be further enhanced by waveguide mode engineering as well as metasurface fabrication. Despite the current bottlenecks in the direct growth of 2D materials on the silicon platform, there are strong interests to explore the use of 2D TMD as optical emitters and photodetectors on silicon platforms.

In addition to semimetals and semiconductors, 2D materials also include metals such as NbSe₂^18–22 and insulators such as hexagonal boron nitride (hBN).^23,24 These materials can be combined to form heterojunctions.^25–27 By engineering band structures, heterostructures can exhibit unique optoelectronic properties. For example, type II heterojunctions can be designed by selecting TMDs with the appropriate energy levels, leading to efficient separation of photoinduced carriers.^28–31 Other interesting interfacial effects, which may depend on the twist angles between the heterolayers, have been found in graphene on hBN, such as the strongly confined 2D plasmon of graphene,³² and the Hofstadter butterfly effect in moiré superlattices.^33,34 Owing to their atomic thickness, 2D TMD semiconductors usually exhibit excellent gate electrostatics.³⁵ The main optical applications of 2D functional materials are listed in Table I.

In this article, the potential applications of 2D materials in the emerging field of integrated photonic circuits (IPC) are discussed. In such a system, optical devices can be divided into two categories: (1) passive devices and (2) active devices. Passive photonic devices include waveguides, optical cavities, multiplexers/demultiplexers (consisting of arrayed waveguides with a wavelength-selecting panel), and optical grating couplers. Silicon provides high optical confinement, achieving low-loss optical propagation (less than 0.3 dB/cm⁹³) in waveguides and other optical structures. This article introduces ultrathin dielectric such as hBN as the optical confining material, which serves as the new on-chip dielectric system for 2D material integration. Active photonic devices include light sources, such as laser diodes (LDs) and light emitting diodes (LEDs), photodetectors, and optical modulators. Light sources or photodetectors commonly require light-emitting (or absorbing) material with direct bandgap to obtain high power-converting efficiency and speed. It is a long standing goal to develop IPCs on silicon-based platform to achieve highly efficient optical transmitters and receivers. However, the lack of a direct bandgap in silicon precludes the development of efficient optical emitters and absorbers. The arguments for using 2D-based materials are based on its strong material index modulation, strong light-matter interactions, and compatibility with vertical downscaling; however, the challenge is how these materials can be monolithically integrated onto the silicon platform. This article introduces IPC based on a hBN light-confining structure, which allows the 2D materials inserted in the middle of the hBN dielectric layer in order to achieve a high optical confinement factor.

Various integration schemes to achieve lasing from 2D materials have been extensively investigated. 2D materials with a direct bandgap provide strong light emission suitable for fabricating nanolasers. The challenge of making 2D nanolasers is the low quantum yield of 2D TMDs, which requires a higher lasing threshold to achieve a high spontaneous emission factor.⁹⁴ The quantum yield is limited to under 10% at room temperature primarily due to the presence of crystal defects,⁹⁵ exciton-exciton annihilation,⁹⁶ and Auger recombination.⁹⁷ Strategies to achieve lasing by photonic circuits commonly involve the following steps: (1) increasing the cavity quality factor to enhance the Purcell effect and (2) placing a 2D nanosheet in the cavity mode maximum to maximize the light-matter interaction.

The Purcell effect is defined as the enhancement of the spontaneous emission rate of the active material by highly localized optical cavity modes. Lasing materials experience the Purcell effect if placed inside optical cavities. For example, in Fig. 1(a), the dipole oscillations in the 2D semiconductor imbedded in the dielectric cuboid lead to photons emission that are collected and localized by the optical cavity mode, while it is also driven by the electric field of the optical cavity mode. This light-matter interaction is regarded as two-energy-level coupling: photons with two optical mode states (⁠ $|0⟩$ and $| h ω photon ⟩$⁠) coupled with conduction and valence band edges of the 2D semiconductor, as shown in Fig. 1(b). As 2D exciton emissions feed photons into the cavity, the cavity mode state density $| h ω photon ⟩$ is enhanced. In turn, according to Fermi’s golden rule, the conduction-to-valence band transition rate is enhanced by an increase in the optical mode density.

The optical confinement factor is analyzed by using the simulated cavity mode distributions. In a conventional III–V dielectric cavity, $Γ$ is usually more than 50%. For 2D monolayer integrated on top of conventional dielectric cavity, the mode energy overlap usually drops to less than 0.1% due to its atomic thickness.

Nanolaser with 2D monolayer emitter was first realized in a photonic crystal cavity with monolayer WSe₂ as the optically active material.³⁶ The photonic crystal cavity was made by using a suspended gallium phosphide with an air hole array as the optical cladding. Operating at 80 K increased Q from 1300 (at room temperature) to 2500. Optically pumped by a 632 nm continuous-wave laser at 80 K, lasing was realized above a power threshold of 1 W cm⁻². However, lasers were only achieved on III–V platforms. Another study integrated a MoS₂ nanolaser on SiO₂ microdisk and silica microsphere cavities.³⁸ The bilayer MoS₂ was sandwiched by two cavities with an optical confinement factor $Γ$ of 1%. This study demonstrated that even for low quantum yield material such as bilayer MoS₂, lasing can be realized. This was followed by the demonstration of a WS₂ laser in a microdisk cavity with a Q factor of 2604 for the lasing mode near the center of the WS₂ PL spectra.³⁷ In the vertical structure of the cavity, WS₂ was sandwiched by HSQ (a photoresist which can be regarded as SiO₂) and Si₃N₄, so that the confinement factor achieved was 0.11%, compared with 0.085% without HSQ coverage. This observation indicates that the confinement factor has to be increased to facilitate lasing, but interestingly, a confinement factor of only 0.11% can produce lasing, indicating an efficient coupling of the monolayer TMD to cavity modes. The ability to achieve lasing with a low confinement factor requires further research to understand the lasing mechanism in monolayer TMD.

A silicon nanobeam cavity had been previously fabricated for the near-infrared PL emission of MoTe_2.³⁹ Silicon nanobeam cavities are photonic crystal cavities with one-dimensional photonic bandgap formed by periodical air holes in a silicon waveguide on a SOI substrate. The cavity core is formed by adding air holes with gradually decreased diameters in the middle of the array. This cavity has a Q factor of 10⁵ owing to the high refractive index of silicon and the strong photonic confinement by the photonic crystal. MoTe₂ has an exciton energy of 1.08 eV or 1150 nm, which is approximately 30 nm longer than the bandgap of silicon (1.11 eV or 1117 nm). As silicon is transparent to MoTe₂ PL emission, it is possible to integrate MoTe₂ as optically active material in silicon photonic devices. However, the top dielectric coverage was not facilitated in silicon nanobeam cavities in this work, so the 2D MoTe₂ was placed on the cavity surface with relatively low confinement factor.

In addition to optical cavities based on conventional dielectric material, layered insulators such as hBN have been used as dielectric material for optical cavities.^100,101 hBN has been reported to have an in-plane refractive index of 2.3,¹⁰⁰ which suggests possible light confinement by hBN optical cavities placed directly on a silicon oxide substrate (refractive index less than 1.5 at visible and near-infrared wavelengths). Moreover, hBN is an excellent platform for VdW heterostructures, for example, hBN/TMD/hBN, so that a monolayer TMD optical emitter could be placed at the center of the optical cavity. Atomically smooth surfaces and heterojunction interfaces in such heterostructures reduced optical loss and led to high Q modes. The shortcoming of hBN lies in the presence of mid-gap energy levels caused by crystal defects, which reduces its optical transparency. The mid-gap energy levels can range between 0.7 and 4.8 eV, covering the TMD emission wavelengths.⁶³ Therefore, the performance of hBN based IPC is limited by the crystal quality of hBN.

The hBN optical cavities on SiO₂/Si are shown in Fig. 2. An hBN/TMD/hBN heterostructure (with a height of approximately 300 nm) is used for vertical optical confinement and optical emission. The lateral optical confinement is achieved by patterning the heterostructure into disk [Fig. 2(a)], ring [Fig. 2(b)], and 2D photonic crystals [Fig. 2(c)]. The hBN refractive index of 2.3 results in a large contrast index (≈0.9) at the hBN and SiO₂ interface for the disk and ring cavities. The SiO₂ thickness is less than 300 nm, which leads to optical absorption loss at the Si substrate. However, in the case of a photonic crystal cavity, the asymmetric out-of-plane index could introduce optical loss;¹⁰² thus, in this type of cavity, top SiO₂ layer coating can be used to achieve index symmetry.

These cavities allow electrical contacts in the TMD active material without damaging the optical modes. For the disk cavity in Fig. 2(a), the metal contacts are placed at a distance of $d contact >2λ/ n hBN$ from the disk edge so that the optical mode does not experience optical loss from metal absorption. It should be noted that the electrode can be connected to the TMD material by depositing on the slope edge of the hBN/TMD/hBN structure, forming a one-dimensional contact.¹⁰³ For the ring cavity in Fig. 2(b), the metal contacts can provide the plasmonic mode¹⁰⁴ or metallo-dielectric mode¹⁰⁵ in addition to being connected to the TMD material in an electrical channel on the two sides of the ring cavity. The contacts for the photonic crystal cavity in Fig. 2(c) should be placed away from the photonic crystal region because high Q confinement usually requires an array of 10–20 air holes in the dielectric slab without any metal absorption.

Electrically driven light sources are in high demand in silicon photonics because they simplify the photonic system by directly generating photons from electrical injections. In current silicon lasers, hybrid and epitaxial lasers are electrically pumped in III–V quantum wells. Currently, there are two designs for electrically pumped LEDs in 2D TMDs: (1) lateral p-n homojunctions in illustrated in Fig. 3(a) and (2) vertical p-n heterojunctions, as shown in Fig. 3(b).

P- and n-doped semiconductors are the fundamental components of LED devices. Under forward bias, carriers accumulate in the junction regions for recombination. The first 2D LED was discovered in 2013 in a metal/MoS₂ junction by depositing Cr/Au on MoS₂.⁶⁰ Although top gating (30 nm Al₂O₃ and Cr/Au) was applied to the MoS₂, the dopant concentration was insufficient to form a lateral p-n junction. Instead, a Schottky junction was formed between MoS₂ and the electrode for photon emission on the electrode edge. The absence of a p-n junction resulted in a low external quantum yield efficiency (EQE) of 10⁻⁵. In 2014, a lateral p-n junction LED was obtained for WSe₂ with back gates,¹⁰⁶ as shown in Fig. 3(c). The back gates consisted of a 10 nm hBN structure as the dielectric layer, and the top gate electrodes were separated by 300 nm. The overlapping of back gate electrodes and source/drain electrodes on both sides of WSe₂ helped reduce the Schottky barrier. The internal quantum efficiency achieved was 5%. Furthermore, the neutral exciton and charged exciton emission can be tuned by injection bias. Lateral p-n junctions were also realized by silicon nitride back gating¹⁰⁸ and ion gel top gating.¹⁰⁹ A lateral p-n junction using MoTe₂ was recently realized; the junction was transferred to a silicon photonic crystal waveguide to demonstrate the possibility of integrating LEDs in silicon photonic devices.¹¹⁰ 

VdW heterostructures can be made into 2D LEDs when p-doped and n-doped 2D semiconductors are stacked together. MoS₂ is a natural n-doped monolayer semiconductor and can be stacked with p-doped WSe₂, as shown in Fig. 3(d). The EQE of such a 2-D p-n junction can be up to 10%.⁶¹ LEDs made from multiple 2D heterostructures were obtained in 2015.¹⁰⁷ This device was a monolayer MoS₂ sandwiched by two membranes of few-layer hBN from top and bottom, and the other sides of the hBN structure made contact with graphene, forming a semiconductor/barrier insulator/metal heterostructure [Fig. 3(e)]. This structure supported carrier tunneling from graphene to MoS₂ through hBN, as the graphene Fermi level was doped away from the MoS₂ Fermi level by top and back gating. For example, if the back-graphene electrode was gated with Fermi level above the MoS₂ conduction band, the electrons could be injected to MoS₂ and recombined with holes from another side of the electrode, emitting electroluminescence according to the MoS₂ bandgap. In cryogenic temperature (50–150 K), EQE was found to be 1%. To further enhance EQE, the tunneling heterostructures (graphene/hBN/MoS₂) were repeated three to four times; this increased EQE to 8.4%. This design was also tested for monolayer WSe₂ at room temperature.⁶² Owing to the separation of dark excitons and bright excitons in WSe₂, more pump carriers recombined as bright excitons so that LED EQE was improved to 5% at room temperature, which is 250 times as high compared to previous MoS₂ devices. Moreover, the thickness of hBN was optimized to be two or three layers by weighting brightness and efficiency. A WS₂/hBN/graphene tunneling LED was also realized in 2017.¹¹¹ In that study, the Fermi level of n-doped WS₂ was tuned higher than that of p-doped graphene, resulting in efficient hole injection from graphene to WS₂ through an hBN barrier layer of thickness 3–6 nm. This thick hBN barrier limited the tunneling rate and lowered the operation current to a few pA/μm², with a LED EQE of approximately 1%. Another advantage of the thick hBN barrier was that the Fermi level difference of WS₂ and graphene could be as large as 4–6 eV for efficient carrier injection.

The defect states of 2D semiconductors can be used for making single-photon-emission LEDs. Single-photon LEDs have been successfully fabricated by stacking monolayer graphene, few-layer hBN, and monolayer or bilayer WSe₂ in sequence, forming a tunneling LED on an oxidized silicon substrate,⁶⁸ as illustrated in Fig. 3(f). The single-photon emissions originate from the quantized emissions of intrabandgap state transitions at crystal defects. These defect states are sensitive to the operating temperature. At room temperature, the EL spectra stretch to longer wavelength than that of PL, owing to the presence of charged excitons in EL. At 10 K, the localized exciton emission was more visible in EL than in PL. The second-order correlation function exhibited a normalized value of 0.29 at zero delay, which is close to 0 ideally, demonstrating good quantum single-photon emission. Therefore, 2D TMD-based LEDs indicated the feasibility of electrically pumped on-chip quantum light sources.

The ultimate test would be achieving lasing by electrical pumping of the LEDs into population inversion. In principle, nano-LEDs can be realized by integrating an existing 2D LED on an optical cavity. Although LEDs have been intensively investigated with high quantum yield and ultrathin structure, the current injection is difficult when 2D materials are embedded in the optical cavity. For example, the architecture of a WS₂ nanolaser consists of a silicon nitride/HSQ microdisk on a silicon post, sandwiching monolayer WS₂ in the microdisk, as illustrated in Fig. 1(e). To fabricate the LED based on this microdisk 2D laser, the electrodes are required to be added on the optical device. To contact WS₂ and inject carriers to WS₂, the electrodes need to be placed on the edge of microdisk, which would destroy the photonic modes. Electrical driving was further studied in the cavity-assisted LED, aiming to gain strong Purcell effect of electrically driven light source, and to achieve coherent light source. Cavity-assisted LED was made by transferring a GaP photonic crystal cavity membrane onto a tunneling heterostructure (graphene/hBN/WSe₂/hBN/graphene) LED on an oxidized silicon substrate. A lower Q factor (a few hundreds) than that of GaP photonic crystal nanolasers with WSe₂³⁶ was obtained because the cavity mode was adjacent to a light-absorbing graphene in the 2D LED. Therefore, to achieve higher Q factor and lasing in cavity-assisted 2D LEDs, it is better to fabricate cavities without light-absorbing material transferred to the cavity mode region. Importantly, III–V materials-based cavity such as GaP may lead to fabrication difficulties such as wafer bonding and suspending III–V membranes on silicon platform.

Here, we introduce a simple and functional cavity-assisted LED by using 2D LED devices integrated into hBN cavity on silicon oxide/silicon platform. In order for on-chip 2D LEDs to be practically useful, the LEDs should have high EQE and large area emission region that strongly couples with the cavity mode region. Lateral p-n junction LEDs have an emission region of hundreds of nanometers, which limits the potential for large-scale integration. For heterostructure tunneling LEDs, the emission region can be designed as large as possible in lateral space because the tunneling junction functions in the vertical direction. However, the presence of metal (graphene) causes optical absorption in the optical cavity. One strategy is to use electrostatic gating to dope the graphene Fermi level above photon energy and reduce optical absorption by Pauli blocking.¹¹² 

A proposed cavity-assisted LED is shown in Fig. 4. The cavity has a similar structure as the hBN/TMD/hBN disk cavity in Fig. 2(a), except that the middle active material is replaced by a TMD/hBN/G trilayer LED from Ref. 111. The trilayer stack has a ringlike region overlapping with WGM. In the trilayer, the top TMD and bottom graphene structures are extended from the ringlike region and connected to two electrode contacts separately, as shown in Figs. 4(a)–4(d). By applying voltage to the two contacts, carriers are injected from the contacts and recombined only at the trilayer stack. Moreover, graphene can be gated to reduce optical absorption. If the optical absorption is suppressed such that cavity Q factor reaches at least 5000, this device may operate as a laser diode. This disk cavity can couple with an hBN waveguide adjunct to the disk edge, forming the optical signal output system. Additionally, electrode contacts can be linked to silicon-based transistors on the same chip, providing additional electrical-to-optical conversion.

Graphene optical modulators are based on the principle of highly efficient refractive index modulation through electrostatic doping. The imaginary part of graphene’s refractive index determines optical absorption per unit area, leading to amplitude modulation. For graphene structures on oxide/silicon waveguides, as shown in Figs. 5(a) and 5(b), the waveguide mode amplitude can be tuned by electrostatic gating using capacitors formed by graphene/Al₂O₃ dielectric layer/doped silicon. Modulation depth is limited by the graphene active area, and bandwidth is determined by the capacitance and resistance of the capacitor. Such a design first appeared in 2011 with a modulation depth of 4 dB and a bandwidth of 1 GHz.¹¹³ The capacitor part was further improved by using a sandwiched graphene/Al₂O₃/graphene structure, and the bandwidth increased to 3 GHz.¹¹⁴ Another waveguide design consisted of a heavily doped silicon electrode connected to a lightly doped silicon waveguide with a graphene/Al₂O₃ dielectric layer/doped silicon capacitor, exhibiting an improved bandwidth of 5 GHz.¹¹⁵ 

Graphene modulators realize amplitude modulations and phase modulations of continuous optical signal. The amplitude modulations were realized with graphene modulators integrated in optical cavities, as illustrated in Fig. 5(c). Graphene in this design coupled with optical modes whose energy densities were higher than those of the waveguide modes so that modulation depth was enhanced to 15 dB with 30 GHz bandwidth.¹¹⁶ 

Moreover, the phase shift of the waveguide mode can be achieved by tuning the real part of graphene’s refractive index by gating. Based on phase tunability, optical modulators can be formed by using graphene on one arm of Mach–Zehnder interferometers (MZIs), as shown in Fig. 5(d). Gate-tunable phase differences between the two arms in the MZI lead to optical intensity modulation at the output waveguides. This leads to a modulation of 35 dB with a bandwidth of 5 GHz, and the gate voltage is decreased to 2 V¹¹⁷ compared with 10–50 V in amplitude modulators. Graphene contact resistance still dominates the overall resistance, and, thus, further bandwidth improvements remain an issue. The real-world application of 100 GHz bandwidth optical interconnects should require multiplexing of wavelengths using multichannels of graphene optical modulators.

A proposed optical cavity modulator is shown in Fig. 6 and has a similar structure to the cavity-assisted LED in Fig. 6. The graphene/Al₂O₃/graphene heterostructure provides gate-tunable depth modulation for the cavity modes. Compared to the device in Ref. 116, graphene active layers are integrated in the middle of the cavity, and, thus, they receive larger whispering gallery mode (WGM) optical energy density than when integrated on the cavity surface such that the modulation depth can be further enhanced by stronger light-matter interaction. Furthermore, the modulator is connected to silicon-based transistors by metal interconnectors on the same silicon chip; therefore, this design leads to practical applications of integrated photonic circuits.

Photodetectors (PDs) based on 2D materials have been extensively studied. Compared to 2D LEDs, the device structure of 2D PDs is usually simpler because photon-to-carrier quantum efficiencies are higher than carrier-to-photon recombination in 2D materials. In this section, PDs are reviewed with regard to their integration on silicon photonic platforms.

The photoinduced electrons and hole pairs in 2D materials are dissociated in the photo-absorbing material and conducted into electric circuits. The efficiency of photon-to-carrier conversion can be evaluated by photoresponsivity. Photodetection for practical applications requires large bandwidth for high-data-rate communications. Bandwidth is another important measure for evaluating whether PDs can be applied to silicon integrated photonic circuits for 400 GHz optical transceivers. PDs with 2D active material can be divided into two categories: (1) metal-semiconductor-metal (MSM) PDs and (2) photovoltaic PDs.

MSM PDs have been used in conventional silicon or III–V PDs, which convert incident photons to carriers in silicon and are separated by Schottky barriers of metal–semiconductor junctions at the ends of short semiconductor channels.^118,119 This design was first applied to monolayer MoS₂ in 2013.⁷⁷ A schematic of a MoS₂ MSM PD is shown in Fig. 7(a). A responsivity as high as 880 A/W was achieved owing to the high conductivity introduced by photoinduced trap states.¹²⁰ Conductivity could be further improved by the strong gating effect of ferroelectric polarization, with responsivity raised to 2570 A/W at zero bias.¹²¹ Other 2D semiconductors, including WS_2,¹²² MoSe_2,¹²² WSe_2,¹²² GaS,¹²² GaTe,¹²² ReSe_2,¹²² and In₂Se₃¹²² were also used as MSM PDs, with detection wavelengths from ultraviolet to visible and photoresponsivity reaching 1000 A/W but bandwidth limited to less than 1 kHz. A high photodetection bandwidth of 14 kHz was achieved in a MSM PD with few-layer MoS₂ (corresponding to a response time of 0.07 ms).⁸⁶ Multilayer SnS₂ with a thickness of 80 nm exhibited a photoresponse time of 5 μs (200 kHz bandwidth) and responsivity of 8.8 mA/W.¹²³ Even though TMDs have high photoresponsivity, their temporal response is limited by low mobility and high detection density. Further bandwidth improvements, thus, require high carrier mobility and crystal quality.

Metal-graphene-metal (MGM) is a special case of MSM PD. In MGM structure, the photons absorbed in graphene generate electron and hole transition in the linear energy band. The photoinduced carriers migrate under electric field between the two electrodes to form photocurrents. For high-bandwidth PDs, MGM has the advantages of high mobility and fast electron transition time. MGM was first realized in 2009, with a photoresponsivity of 0.5 mA/W and bandwidth of 40 GHz, by using symmetric metal contacts.¹²⁴ Low responsivity was attributed to the fast transition time and short carrier lifetime of the photoinduced carrier in graphene. Responsivity was further improved to 6.1 mA/W by metal contacts with different work functions and 16 GHz bandwidth.¹²⁵ Integration of MGM PDs to waveguides was achieved with a hBN/graphene/hBN heterostructure transferred on top of a silicon SOI waveguide, with responsivity improved to 0.36 A/W (bandwidth 42 GHz) owing to a metal/graphene junction placed near the waveguide mode region.¹²⁶ 

MSM was realized not only in the lateral structure but also in the vertical VdW heterostructure of graphene/TMD/graphene. The TMD layer thickness was on the order of tens of nanometers, which was shorter than the depletion length of a graphene/TMD junction, so that photon-induced carriers could be dissociated efficiently. Electrostatic doping to graphene was applied to enhance the potential difference on the two sides of the TMD multilayer, resulting in larger photoresponsivity, namely, 0.27 A/W.¹²⁷ This structure was further tested for ultrafast photodetection, exhibiting a photoresponse time of 1.6 ns in a graphene/few-layer-WSe₂/graphene device, suggesting a possible 625 MHz bandwidth (with a responsivity of 47 mA/W¹²⁸). A thin TMD middle layer significantly shortens the electronic channel, giving rise to a faster response time. However, one problem for ultrafast photodetection in MSM structures was that dark current was high owing to the increased conductivity in the short semiconductor channel. The high dark channel would lead to high power consumption and energy loss, thus rendering MSM unsuitable for low-loss high-efficiency photodetection.

A p-n junction PD (or photodiode) has a different detection mechanism. In a p-n junction detector, a photon is absorbed by the semiconductor in the p-n junction, and the photoinduced electron and hole pairs are dissociated by the built-in potential at the p-n junction depletion region. For a p-n junction with reversed bias, whose doping levels for both types are raised, the depletion region is enlarged with enhanced built-in potential. The diffusion of photoinduced carriers can be driven by built-in potential to form photocurrent. One advantage of reversed p-n PDs is suppressed dark current with reversed bias owing to the large depletion region.

N-type monolayer-MoS₂ and p-type monolayer-WSe₂ have been combined as p-n heterojunctions for fabricating a photodetector, as shown in Fig. 7(b). Owing to the atomically thin layers, the depletion region is absent, and photoinduced electrons (or holes) are driven by the differences of conduction band minimums (or valence band maximums) of MoS₂ and WSe₂. Due to the spatially segregated electrons and holes, the intralayer carrier recombination is quenched, resulting in efficient carrier conversion. Moreover, the top and bottom graphene layers were used as electrodes in the vertical p-n stack to reduce the lateral resistance and diffusion time in the p and n layers, and this produces a responsivity of 10 mA/W at visible wavelength.²⁶ A higher responsivity of 46 mA/W at visible wavelength was reported for monolayer-WSe₂/multilayer-MoS₂ vertical junctions at reversed bias owing to enhanced photon absorption in multilayer TMDs; the response time was less than 1 ms, but the exact bandwidth of the PD was not reported.⁶¹ Monolayer-MoS₂/multilayer-BP vertical junctions further achieved 418 mA/W responsivity at visible wavelength.⁹¹ Monolayer MoS₂/graphene/WSe₂ heterojunctions shows improved responsivity to 10⁴ A/W at visible and 1 A/W at near-infrared, with a bandwidth of 19 kHz.¹²⁹ VdW heterostructure photodiodes are promising as PDs in integrated circuits because each layer can be controlled at the atomic level. Monolayer graphene plays an important role in vertical heterostructures by providing broadband response and transparent electrodes. Further studies for practical applications should focus on ultrafast response time toward GHz bandwidth.

A cavity-assisted 2D PD is shown in Fig. 8. A ternary structure based on graphene/TMD/graphene is selected as active material owing to the high bandwidth for GHz applications. This graphene/TMD/graphene trilayer is integrated in the middle of the optical cavity for maximum light-matter interaction. Furthermore, as in the case of the proposed LED and modulator, this PD is linked to on-chip transistors so that optical-to-electrical conversion can be performed on the same silicon chip.

2D photonic integration can be achieved by a polymer transfer process in addition to the conventional CMOS process. Considerable attention has been given to optically pumped 2D nanolasers and 2D LED, but practical 2D nanolasers that are electrically pumped on integrated silicon platforms have yet to be achieved. In photonic integrated circuits, the waveguides couple and carry cavity emission signals to other optical devices. The output signals of the optical cavity can be modulated by electrostatically gated graphene as a tunable optical absorber on the waveguide or cavity. These multifunction 2D devices can be designed as periodic cavity/waveguide arrays. Compact integration can be realized by large-scale fabricated arrays with repeated 2D semiconductor/silicon nanostructures, with each individual component controlled by electrical signals via electrodes.

In 2D-material integrated devices, the insulator layer plays an important role. As the light emitter sits on a silicon waveguide/cavity,^39,110 a 2D semiconductor may form a 2D/silicon heterostructure at the interface, where carrier loss can occur. The current solution is to add a silicon oxide layer on the surface of the silicon waveguide/cavity,^113,130,131 thus placing the 2D material away from the photonic mode maximum. To this end, a hBN layer can be used in the waveguide and the cavity to localize the photonic mode in the insulator layer.¹⁰⁰ hBN serves as the dielectric material for optical confinement and signal guiding. The high refractive index of hBN allows optical cavities and waveguides to be fabricated directly on SiO₂/Si substrate. Moreover, hBN layers allow VdW heterostructures to be constructed with graphene and TMD, and the latter can serve different optical functions in hBN optical cavities.

An integrated photonic circuit using different 2D materials to serve various functions such as electroluminescence, optical signal guiding, optical modulation, and photodetection is proposed in Fig. 9. The laser diode with graphene/hBN/TMD LED and hBN optical cavity on silicon oxide (shown in Fig. 4) is driven by a DC electrical source and emits laser from a cavity. This laser signal is coupled to an inactive hBN disk cavity and is filtered to a single output wavelength. This output optical signal is a continuous wave that is guided to an optical modulator through an hBN waveguide. The optical modulator (Fig. 6) converts electrical signals from transistors to optical signals in the waveguide. The modulated optical signals may be processed by wavelength-division multiplexing (WDM) and coupled out to an optical fiber by optical grating. Finally, a cavity-assisted photodetector (shown in Fig. 8) receives optical signals from external optical fibers and converts them to electrical signals feeding back to transistors on the same chip. Therefore, multiple devices connected by optical waveguides and optical couplers can function as optical data processing devices, from photon emission to photon-to-electron conversion.

In conclusion, laser diodes, optical modulators, and photodetectors have been proposed based on 2D material heterostructures integrated in hBN disk cavities and connected to silicon-based transistors on the same silicon chip. The proposed system opens up the possibility of high bandwidth ultracompact on-chip optical interconnects on a silicon platform. Optical devices can be linked by hBN optical waveguides forming on-chip integrated photonic circuit. It is possible to confine an optically active 2D TMD within the mode of guided light in hBN disk cavities, allowing for the propagation of linearly polarized light in the plane of the monolayer. This may allow exciton states with different optical selection rules (e.g., dark and bright excitons) to be independently accessed using differently polarized light. However, the practical realization of these photonic circuits requires the monolithic growth of large area TMD and hBN on the silicon platform, which has yet to be achieved to date. Nonetheless, breakthroughs in large area TMD and hBN growth have been achieved on other substrates;^132–134 thus, future developments in automated wafer-to-wafer transfer techniques may help address the gap in materials integration.