Over the past two decades, alongside the surge in activities related to 2D materials, many related fabrication techniques have been developed, demonstrating great potential for large-scale industrial manufacturing. In this section, we present an overview of cutting-edge techniques for fabricating integrated photonic devices incorporating 2D materials, highlighting their potential and limitations in industrial manufacturing. It is divided into four parts, including large-scale integration, precise patterning, dynamic tuning, and device packaging.
A. Large-Scale Integration
Producing high-quality integrated wafers, such as the silicon-on-insulator (SOI) wafers, is the initial step in manufacturing integrated photonic devices. Current fabrication techniques, such as chemical vapor deposition (CVD) and physical vapor deposition (PVD), have already enabled industrial-scale manufacturing of high-quality integrated wafers in different sizes (
e.g., 4, 8, and 12 inches). [
40,
41,
42] In this part, we focus on discussing methods for integration of 2D materials onto these wafers. To achieve this, the process typically involves two steps: synthesizing 2D materials and transferring them onto integrated substrates. In some cases, 2D materials can be directly synthesized on integrated wafers, thus simplifying the overall process.
Synthesizing high-quality 2D materials is essential for implementing hybrid devices with high performance. Mechanical exfoliation, liquid-phase exfoliation (LPE), and chemical vapor deposition (CVD) are three main methods for synthesizing 2D materials. Mechanical exfoliation provides a simple and efficient way to synthesize 2D materials without using complex facilities. However, for large-scale industrial fabrication, it suffers from limitations due to the inconsistent thicknesses and small sizes of the 2D flakes produced by this method. LPE is a solution-based method to synthesize 2D materials, allowing for cost-effective production of large volumes of 2D nanoflakes. Similar to mechanical exfoliation, it also faces limitations induced by small sizes of the exfoliated 2D flakes. In contrast to the above exfoliation methods, CVD offers an attractive approach for growing high-quality 2D films over large areas with precise thickness control. This makes it appealing for industrial production of high-quality 2D films. Despite this, the state-of-the-art CVD method for synthesizing 2D materials still faces challenges such as limited efficiency, film contamination, and complex process control.
In addition to the above three main methods, some other methods for synthesizing 2D materials have also been investigated, such as molecular beam epitaxy (MBE) [
43,
44] and PVD. [
45,
46] MBE is a precise atomic layer by layer growth technique used to grow single crystal films epitaxially on an existing crystal substrate. It was invented for the growth of compound semiconductors in the 1970s and has been successfully applied to the growth of metals, oxides, and topological materials. [
47] Recently, this technique has also been employed to grow high-quality monolayer graphene, TMDCs, and h-BN. [
48,
49,
50] The main limitation of current MBE technique is that it requires significantly higher vacuum levels than other deposition methods to achieve comparable impurity levels in the grown films. [
51,
52] The PVD method involves a process where material transitions from a condensed phase to a vapor phase, and then recondenses into a thin film. Sputtering and evaporation are two widely used PVD techniques for coating thin films of metal, glass, and polymers in industrial manufacturing. Recently, the PVD method has been employed for fabricating thin hBN, MXenes, and perovskite films. Compared to CVD and MBE, it provides higher deposition rates, but this comes with the trade-off of larger 2D film thicknesses (typically > 30 nm). [
45,
46,
53]
After synthesizing 2D materials, it is usually needed to transfer them onto the target integrated substrates to construct hybrid devices. Dry transfer methods using transfer stamps [
54] was widely employed for transferring mechanically exfoliated 2D flakes in the early research of this field. [
12,
55,
56,
57] However, several factors restrict the applicability of this method to laboratory environments instead of industrial production. These mainly include challenges in processing flakes with large lateral size, low fabrication yield, and reliance on complicated supporting facilities. For 2D flakes synthesized by LPE, solution dropping techniques such as drop casting, spin and spray coating are commonly used for their on-chip transfer. [
58,
59] Although these methods provide a fast way to prepare 2D films over large areas, they face limitations such as low film uniformity (typically > 10 nm [
60]) and large film thicknesses (typically > 100 nm [
61]), restricting its use for fabricating films with low thicknesses.
For 2D films synthesized by the CVD method, a high growth temperature is usually needed. As a result, they are typically deposited on metal substrates or foils, rather than dielectric substrates used for integrated photonic devices (
e.g., silicon, silicon nitride, silica). Consequently, subsequent film transfer processes are required. Unlike the dry transfer techniques, which are commonly used for small exfoliated 2D flakes, wet transfer techniques are typically employed for transferring CVD-grown 2D films with large lateral sizes. [
62,
63] They are easier to operate and do not require sophisticated facilities, resulting in higher success rates and improved transfer efficiency than the dry transfer techniques. Despite these advantages, problems like stretching, wrinkling, and bending can easily occur during the on-chip transfer process, even though CVD-grown 2D films can achieve a high uniformity on the original substrates. This makes it challenging to achieve highly uniform coating or conformal coating on integrated waveguides and metasurfaces, potentially impairing device performance in some applications. [
64]
In recent years, several representative works have been reported to address the limitations of the previously mentioned transfer techniques, as summarized in
Figure 4. In
Figure 4(a), success transfer of CVD-grown graphene films onto 4-inch silica/silicon wafers was achieved by introducing a transfer medium polymer (polyvinyl alcohol mixed with sorbitol molecules), [
65] where the adhesion between the polymer and graphene layers was adjusted through freezing-induced crosslinking, enabling crack-free film transfer over large areas. In
Figure 4(b), a modified solution dropping method was demonstrated by using electrochemical intercalation to prepare high-quality molybdenum disulfide (MoS
2) solutions, [
66] which allowed the coating of large-area MoS
2 films with low thicknesses (~3.6 nm) and high uniformity. In
Figure 4(c), a new solution-based film coating method relying on self-assembly of GO monolayers was demonstrated. [
67] This method enabled not only layer-by-layer film coating over large areas on dielectric substrates like silicon, silicon nitride, and silica, [
68,
69] but also conformal coatings on integrated waveguides and metasurfaces. [
28,
70]
In
Figure 4(d), a tri-layer transfer medium (polymethyl methacrylate (PMMA)/Borneol/graphene) with gradient surface energy was employed to improve the wet transfer process, allowing reliable adhesion to and release from the target substrates. [
71] In
Figure 4(e), to maintain 2D films with flat, intact and clean surfaces during the transfer process, supporting films were introduced by incorporating oxhydryl groups-containing volatile molecules into PMMA. [
72] These films could deform under heat, thus enabling controllable conformal contact with different 2D films such as graphene and MoS
2. In
Figure 4(f), a modified CVD method, combined with a bubble transfer method, was developed to fabricate a 30 × 30 mm single crystal hBN film on a silica/silicon wafer. [
73] During the fabrication process, the hBN was delaminated from liquid Au substrates by using hydrogen bubbles generated by electrolysis. This enables the fabrication of high-quality hBN films free from grain boundaries that significantly influence the film uniformity and mechanical strength.
By combining the precise positioning of dry transfer techniques and the high efficiency of wet transfer techniques, a hybrid technique called semi-dry transfer has been developed. For instance, in
Figure 4(g), three large graphene sheets of up to 50 × 50 mm [
2] were transferred to a silicon wafer with over 99% yield. [
74] This was achieved by using functional tapes with their adhesive forces controlled by ultraviolet light, along with electrochemical delamination. In
Figure 4(h), the semi-dry transfer method was employed to fabricate a monolayer tungsten disulfide (WS
2) on a silicon wafer. During the transfer process, the film was first exfoliated by thermal release tape assisted with a Ni film, followed by releasing the tape/Ni/film on the target wafer. Subsequently, the tape was removed by annealing, and the Ni was etched away using FeCl
3. [
75] In
Figure 4(i), direct growth of monolayer MoS
2 films onto 12-inch silicon wafers was achieved by pre-positioning amorphous Al
2O
3 films as interlayers. This provides an effective way for in-situ growing 2D films onto integrated wafers without the film transfer process. [
76]
Table 1 summarizes several synthesis and transfer techniques with high potential for large-scale manufacturing (some of which were discussed in
Figure 4), along with the relevant publications that utilize these methods. Although the current methods represent significant progress toward industrial manufacturing, challenges still remain. The modified dry transfer method in
Figure 4(a) has difficulties in achieving high fabrication yield and accurate film thickness control. The modified solution dropping method in
Figure 4(b) has limitations related to reproducibility and solution contamination. The self-assembly method in
Figure 4(c) face challenges for fabricating thick (>200 nm) films due to its time-consuming nature, as well as increased scattering loss in thicker films. The performance of the modified wet transfer techniques in
Figures 4(d) − (f) can be compromised by challenges such as residual solvents induced by the chemical solution, precise alignment during the transfer, and variability in thickness control. The semi-dry transfer approaches in
Figures 4(g) and
4(h) are currently limited by the complexity of the transfer procedures and potential contamination from chemical solutions. The introduction of Al
2O
3 interlayers for the direct growth method in
Figure 4(i) was used for fabricating TMDCs, and its applicability to other 2D materials remains to be investigated. In the future, fabrication techniques for large-scale integration of 2D materials can be further developed on the basis of these methods, along with the introduction of new techniques, to ultimately achieve the goal of industrial manufacturing.
B. Precise Patterning
Precise device patterning is crucial for engineering the functionalities of integrated photonic devices and optimizing their performance. In industrial manufacture of bulk integrated photonic devices, the patterning of dielectric patterns (
e.g., waveguides) is mainly performed using photolithography, followed by etching processes such as inductively coupled plasma etching and reactive ion etching. On the other hand, the patterning of metal (
e.g., for electrodes and plasmonic devices) is primarily achieved through photolithography, followed by electron beam evaporation and the lift-off processes. Given that the patterning techniques for fabricating bulk integrated photonic devices has been relatively mature, [
9,
81,
82,
83] here we focus on discussing the methods for patterning 2D materials in hybrid integrated devices. These include not only conventional techniques employed for bulk integrated photonic devices, but also new methods specifically designed for 2D materials.
Photolithography is a dominant device patterning technique in the IC industry. In the photolithography process, the designed layout is transferred from a mask onto photoresist coated on an integrated substrate. Depending on the employed light wavelength, it can be categorized into ultraviolet (UV), deep ultraviolet (DUV), and extreme ultraviolet (EUV) lithography, which allow different levels of patterning resolution. [
84,
85] Unlike electron-beam lithography (EBL), which achieves a higher resolution at the expense of considerably longer exposure times, photolithography is more widely employed for industrial manufacturing. In
Figure 5(a), 50-nm-wide graphene nanoribbons featuring straight, clean, and parallel edges were fabricated by using a multi-patterning process that involved photolithography and bottom-up self-expansion. This method enabled a patterning resolution < 100 nm, which is one order of magnitude lower than the traditional lithographical resolution. [
86] In
Figure 5(b), a 50-µm-long GO film was patterned on a doped silica micro-ring resonator (MRR). The film placement and coating length were precisely controlled via photolithography on a photoresist, followed by GO film coating (through self-assembly) and lift-off processes. [
87]
Nanoimprinting is another widely used patterning technique in integrated device fabrication. Compared to photolithography, it has similarities in steps like resist coating, film attachment, and resist removal, but differs by using an imprint mold to pattern the resist layer instead of photolithography. In
Figure 5(c), graphene nanoring arrays were fabricated on a 9-cm [
2] silica substrate via nanoimprinting. The fabricated rings had a width of <15 nm and a thickness of <0.7 nm. [
88] In addition, nanoimprinting was utilized to engineer the strain of 2D MoS
2 films in
Figure 5(d), allowing for precise control of the strain magnitudes and distributions at a low cost and high throughput. [
89]
Compared to photolithography and nanoimprinting, laser patterning provides a quick and simple method for directly patterning 2D film without using any masks, molds, photoresists, or chemical solutions. In
Figure 5(e), a GO flat lens was fabricated by using laser patterning to convert the GO into reduced GO (rGO) via photoreduction, achieving a focusing resolution of ~300 nm and an absolute focusing efficiency >32% over a broad wavelength range from 400 to 1500 nm. [
90] In
Figure 5(f), laser patterning was used for fabricating a flat lens on monolayer TMDC, achieving a subwavelength lateral resolution of ~400 nm and a high focusing efficiency of ~31%. [
91] In
Figure 5(g), a noncontact laser patterning method was proposed by using a two-photon 3D printer with a scanning pulsed laser. 2D materials like graphene, MoS
2, and platinum diselenide (PtSe
2) were patterned with a subwavelength resolution (
e.g., ≥100-nm hole diameter) at a high throughput (
e.g., ∼3s to clear a 200 μm × 200 μm area). [
92]
As a modified solution dropping technique, inkjet printing enables the on-chip transfer and patterning of 2D films in a single step. Its fast and in-situ patterning allows for industrial fabrication of large-area patterns, with their position and shape controlled
by program. In
Figure 5(h), additive-free MXene ink for was used for inkjet printing, achieving a line width of ~80 µm, a gap width of ~50 µm, and a spatial uniformity of ~3.3%. [
93] In
Figure 5(i), inkjet printing was used for patterning MoS
2 and graphene in large areas, with an optimal drop spacing of ~40 μm on silicon/silica substrates and ~45 μm on glass and polyethylene terephthalate. [
94]
Although the aforementioned methods have been extensively used for patterning 2D materials in laboratory research and offer appealing advantages for industrial fabrication, challenges still remain. For example, photolithography has limitations induced by chemical residues after the etching or lift-off process, damage in 2D films due to high-energy exposure or aggressive chemical solutions, and reduced effectiveness on non-planar substrates. The laser patterning process can also cause damage to 2D films. In addition, the minimum feature size of laser patterning is limited by the spot size of the focused laser beam, resulting in a relatively low patterning resolution that is typically > 300 nm. Nanoimprinting face challenges related to its flexibility in changing the device pattern and in achieving a high film uniformity across large areas. These make it better suited for fabricating relatively simple and repetitive patterns. The imprinting process can also cause mechanical damage to the delicate structures of 2D materials, thus degrading their performance for some applications. For inkjet printing, the patterning resolution is relatively low (typically > 1 µm), and it also face challenges similar to other solution dropping methods, such as low film uniformity and large film thicknesses.
Some other patterning methods, such as focused ion beam (FIB) milling, [
95,
96,
97] scanning probe lithography (SPL), [
98,
99] adhesion lithography, [
100] and self-assembled-mask lithography (SAML), [
101,
102] have also been employed for patterning 2D materials. Each method has its own distinct advantages, such as the high patterning resolution for FIB milling (
e.g., down to 10 nm [
95]) and the rapid processing speed for SPL (
e.g., ~20 mm/s [
103]). However, either relatively low fabrication efficiency or limited applicability to different 2D materials restrict their current use to laboratory research rather than industrial-scale production. With continued progress in micro/nano fabrication techniques, it is anticipated that more advanced patterning methods aimed at low-cost and highly efficient industrial manufacturing would be developed to overcome existing limitations in the future.
C. Dynamic Tuning
Dynamic tuning underpins the operation of many active integrated photonic devices such as lasers, [
10,
104] optical switches, [
105,
106] and electro-optic modulators. [
107,
108] Even for passive integrated photonic devices, introducing dynamic tuning is useful for optimizing their performance [
87,
109] and extending their applicability across varying conditions. [
110,
111] For commercial bulk integrated photonic devices, dynamic tuning is typically achieved by introducing co-integrated thermo-optic micro-heaters [
112,
113] or PN junctions [
114,
115] to change material refractive indices. The former has typical response times in the millisecond to microsecond range, whereas the latter enables faster tuning with response times on the order of picoseconds or even faster. In laboratory research, dynamic tuning can also be achieved through laser-induced photothermal effects or nonlinear optical effects. [
116,
117] For hybrid integrated photonic devices incorporating 2D materials, dynamic tuning of the properties of 2D materials is also crucial for enabling functionalities and optimizing performance. In this part, we discuss the mechanisms for dynamic tuning of 2D materials in hybrid integrated photonic devices, which are classified into gate tuning, laser tuning, thermal tuning, and strain tuning.
2D materials exhibiting metallic behaviors, such as graphene and TMDCs, are susceptible to external electrical fields due to their exceptional Fermi-Dirac tunability. This forms the basis of electrostatic gate tuning methods, which allows for efficient, reversible, and real-time modulation of their electrical and optical properties. In
Figure 6(a), a graphene/ion-gel heterostructure was integrated on a silicon nitride MRR with source–drain and top gating, which enabled gate tuning of the Fermi level of graphene and hence the chromatic dispersion of the hybrid MRR. [
118] In
Figure 6(b), a monolayer WS
2 was patterned on a silicon nitride MRR with ionic liquid cladding. The WS
2 film was doped by applying a bias voltage across the two electrodes through the ionic liquid, resulting in a significant refractive index change of ∆
n = 0.53. [
119] In
Figure 6(c), ferroelectric gate tuning was achieved by co-integrating a ferroelectric P(VDF-TrFE) top layer over a 2D molybdenum diteluride (MoTe
2) film on a silicon/silica substrate. By using a scanning probe to control the polarization of the ferroelectric polymers, lateral p–n, n–p, p–p, n–n homojunctions were constructed in the MoTe
2 film, allowing simple and arbitrary definition of the carrier injection with nanoscale precision. [
120]
Lasers can be used to trigger optical effects like saturable absorption, photon-excited carrier transport, and photothermal effects in 2D materials, which can, in turn, be harnessed to dynamically tune their properties. In
Figure 6(d), a graphene was integrated on a silicon photonic crystal cavity, where the optical absorption of graphene was tuned via exposure to a continuous-wave (CW) laser at 1064 nm, achieving a resonance wavelength shift of ~3.5 nm and a quality factor change of ~20%. [
121] In
Figure 6(e), the light absorption in a graphene-silicon hybrid nanowire waveguide was tuned by exposure to a 635-nm CW laser. During the tuning process, the photon-excited carriers in silicon were injected into graphene via the Schottky diode junction, resulting
in an increased carrier concentration and change of the Fermi level. [
122] In
Figure 6(f), by injecting a CW light into GO-silicon hybrid waveguides to induce reversible photo-thermal changes in 2D GO films, three functionalities were demonstrated across broad wavelength ranges, including all-optical control and tuning, optical power limiting, and non-reciprocal light transmission. [
123]
Apart from laser-induced photo-thermal changes, co-integrated micro-heaters can also induce photothermal changes in 2D materials. Unlike the hybrid tuning approach in
Figure 6(f) that combines laser and thermal tuning, the use of micro-heaters to cause temperature-dependent changes is a typical thermal tuning method. In
Figure 6(g), a ring-shaped metallic microheater was fabricated on top of polypropylene, with a MoS
2 film placed at its center. By applying a bias voltage to the microheater, the temperature of the polypropylene surface was changed, enabling precise and reversible tuning of the refractive index of MoS
2. [
124]
Strain tuning provides an effective method to modify the properties of 2D materials in a continuous and reversible manner by applying mechanical tension or compression. Localized strain can be utilized to finely control and adjust material properties such as optical emission and photoconductivity in 2D materials. However, in comparison to gate or optical tuning, the response time of strain tuning is relatively slow. In
Figure 6(h), single photon emission in a 2D hBN film were controlled via local strain tuning induced by atomic force microscope indentation. By tuning the indentation parameters, indents sites with various lateral sizes were obtained to fully activate the single-photon emitters. [
125] In
Figure 6(i), single-photon emitters were achieved in a WSe
2 film by using strain tuning through nanoscale stressors combined with defect engineering via electron-beam irradiation. This method not only allowed precise tuning of the emission sites but also improved the yield, purity, and operational temperature of the emitters. [
126]
Although the above dynamic tuning methods of 2D materials offer attractive benefits, several challenges still need to be addressed for their use in industrial fabrication. For instance, traditional gating tuning methods often struggle to achieve precise modulation in small, localized regions. This lack of precision can limit the effectiveness of tuning, especially for nanoscale devices. For laser tuning method, the tuning lasers are usually not co-integrated on chips, which hinders its wide use for commercial products. In addition, improper laser intensity or wavelength can result in localized overheating or photodegradation of 2D materials, leading to irreversible damage such as ablation or defect formation. Some strain tuning methods are prone to damage the 2D materials during the tuning process such as cracking, folding, or tearing, especially when the strain exceeds the material elastic limit. The relatively low response speed of strain tuning also limits its use in high-speed tuning applications. To address these challenges, it is expected that future advancements in dynamic tuning methods would be developed with more robust, flexible, and versatile control.
D. Packaging
Device packaging typically represents the final stage of integrated device fabrication, where the device is encapsulated in a protective case that safeguards it from physical damage and corrosion. This step is essential for the industrial manufacturing of commercial products with stable operation. After over 50 years of development, the IC industry has already established well-developed packaging techniques, such as pin grid array (PGA), land grid array (LGA), ball grid array (BGA) and 2D & 3D customized solutions. [
127,
128,
129,
130] With respect to PICs, some advanced packaging techniques have been adapted from those used for ICs, while also taking into account new factors such as fiber-to-chip coupling, chip-to-chip coupling, high-density electrical and optical interconnections, hybrid integration of photonic chips, multi-chip modules, and thermal management.
Figures 7(a) ‒ (c) show some images that showcase state-of-the-art packaging techniques for integrated photonic devices.
Figure 7(a) shows a hybrid tunable laser butterfly packaged by the LioniX International corporation, which includes active gain medium and a passive external cavity. [
131]
Figure 7(b) shows a packaging technique developed by PHIX corporation, which utilizes off-the-shelf building blocks and provides an open architecture that accommodates devices with various sizes and across different integrated platforms. [
132]
Figure 7(c) shows the first-ever fully integrated optical compute interconnect chiplet co-packaged with a Central Processing Unit (CPU) from the Intel corporation. [
133]
For hybrid integrated photonics devices incorporating 2D materials, in addition to packaging integrated chips, encapsulating 2D materials is also necessary to prevent material degradation and ensure stable operation. Due to their large surface area and ultralow film thicknesses, many 2D materials (e.g., BP and TMDCs) are highly sensitive to environmental factors such as temperature, humidity, mechanical stress, and chemical exposure. To improve their stability in hybrid integrated devices, various encapsulation materials, such as dielectric materials, organic polymers, and robust 2D materials, have been used to effectively shield 2D materials from the surrounding environment.
In
Figure 7(d), by using 3-nm-thick inorganic molecular crystal Sb
2O
3 as a protective layer, BP exhibited significantly enhanced structural stability for over 80 days, in contrast to the rapid degradation of unprotected BP within hours. [
134] In
Figure 7(e), a 6-nm-thick Al
2O
3 encapsulation layer was deposited on a BP film via atomic layer deposition, enabling the realization of air-stable BP-based humidity /chemical sensors. [
135] In
Figure 7(f), the stability of BP was improved by using a 60-nm-thick dioctylbenzothienobenzothiophene (C8-BTBT) organic film, which provided protection against oxidation under ambient conditions for more than 20 days. This approach can also be applied to other 2D materials such as TMDCs. [
136] In
Figure 7(g), a monolayer of MoS
2 was embedded within an elastomeric waveguide chip made from polydimethylsiloxane (PDMS), which not only ensured mechanical and environmental protection of the MoS
2 film but also enhanced its photoluminescence (PL) performance. [
137] In
Figure 7(h), BP films were plasma-treated and passivated with a ~300-nm-thick PMMA polymer cover layer, which not only eliminated chemical degradation of the oxidized BP surface but also provided protection against water and oxygen molecules in the air. [
138] In
Figure 7(i), a monolayer graphene film was encapsulated between two hBN layers (~30 nm on the bottom and ~10 nm on the top) with high thermal, chemical, and mechanical stability, allowing for broadband and low-noise terahertz photodetection at room temperature. [
139]
Although the above encapsulation methods provide many advantages in protecting 2D materials from environmental degradation, they still present some limitations in terms of broad applicability and long-term performance needed for commercial products. For instance, physical encapsulation with polymers like PMMA suffers from limited durability and will degrade rapidly when exposed to organic solvents. In addition, the encapsulation with dielectric materials such as Al2O3 via atomic layer deposition method and chemical functionalization with organic solutions is destructive, leading to inevitable degradation of the properties of 2D materials. The encapsulation methods such as dielectrics deposition and polymer coatings may also compromise the sensitivity and functionality of 2D materials in integrated photonic devices, resulting in a trade-off between protection and performance. Therefore, it is expected that more non-invasive encapsulation methods would be developed in the future, not only providing enhanced protection for 2D materials against environmental degradation, but also allowing their potential to be fully exploited.