Flexible integrated photonics: where materials, mechanics and optics meet [Invited]

: While the vast majority of integrated photonic devices are traditionally fabricated on rigid substrates, photonic integration of both passive and active photonic devices on flexible polymer substrates has been demonstrated in recent years, and its applications in imaging, sensing and optical interconnects are being actively pursued. This paper presents an overview of the emerging field of mechanically flexible photonics, where we examine material processing and mechanical design rationales dictated by application-specific optical functionalities. The examples include semiconductor nanomembranes which serve as the key enabling material for hybrid inorganic-organic flexible active photonics, and monolithically integrated passive photonic structures fabricated from semiconductors, polymers, or amorphous materials. Technical challenges and further research opportunities related to materials engineering and device integration on flexible substrates are also discussed.

The requirement of mechanical flexibility poses unique constraints on the material processing and mechanical engineering of flexible photonic devices in addition to their optical design considerations. Polymers have long been a preferred material for flexible photonic devices given their inherent mechanical flexibility. Besides serving as the flexible substrate material, polymers are also used for waveguiding [23,24], filtering [25,26], light emission [27][28][29][30], and optical modulation [31]. Besides polymers, new materials are progressively being introduced onto flexible platforms to offer new functionalities. Many of these materials are traditionally regarded as rigid or even brittle in their bulk form; however, free-standing nanomembranes made of the same materials can be tightly bent without cracking, since surface strain induced by bending is inversely proportional to the membrane thickness. In particular, semiconductor nanomembranes (NMs) present an attractive material platform for active photonic device integration on polymer substrates [32,33]. Semiconductor NMs retain a single crystalline structure essential for efficient optoelectronic device operation, while their small thickness imparts large mechanical flexibility, a salient feature that their bulk counterparts usually do not possess. The mechanical design rationale aiming to reduce strain level in flexible photonic devices is also applied to other material systems. One example is the recent demonstration of highly flexible, high-index-contrast glass-based photonic devices [34].
In this article, we present an overview of the rapidly developing field of flexible integrated photonics in light of the latest material and mechanical design innovations. The paper is organized as follows: we first examine the mechanical design principles that enable large mechanical flexibility in planar photonic structures and the material selection criteria for active and passive photonic components. Device designs and processing methods defined by these general principles as well as specific application needs are subsequently reviewed. Finally we conclude with a discussion on the emerging trends and opportunities in this field.

Nano-mechanical engineering: the path towards ultimate mechanical flexibility
Mechanical flexibility, measured by the maximum possible deformation (bending radius, relative elongation, etc.) without structurally damaging the photonic devices, is determined by yield strengths of materials that constitute the device as well as the device geometry which specifies the strain exerted on the devices during deformation. Therefore, the basic design recipe for improving flexibility is to minimize the strain of the photonic device layer at a given mechanical deformation configuration. This is particularly important for photonic devices made of rigid materials such as semiconductors or glasses. Here we will focus on the design strategies to enhance mechanical flexibility of photonic devices when they undergo two categories of deformation: bending/folding/rolling/twisting (note that twisting of a thin, non-stretchable membrane is equivalent to single-axis bending) and stretching/compression. (a) Pure bending of a multi-layer structure, whose the top and bottom surfaces undergo tensile and compressive strain, respectively. The neutral plane position is specified by Eq. (1). Strains vanishes at the neutral plane, and thus maximum bending flexibility is achieved when the photonic devices are located at the plane; (b) finite element simulated through-thickness strain distribution in structures bent to a radius of 1 mm: the blue curve corresponds to strain in a polyimide (10 μm)-silicone (50 μm)-polyimide (10 μm) tri-layer structure, and the red curve plots strain in a single polyimide layer of the same total thickness (70 μm). The large, threeorders-of-magnitude modulus mismatch between silicone and polyimide results in significant shear strain in silicone which invalidates Eq. (1) and effectively relieves the strain on the multi-layer stack surfaces.

Design approaches to increase bending flexibility of photonic structures
In the case of simple bending, strain minimization can be attained by adopting the so-called "neutral plane" configuration. At the neutral plane, the strain vanishes. Therefore, positioning thin film photonic devices at the neutral plane inside a multi-layer stack minimizes their strains during bending. For a multi-layer stack with the 1st layer on top and the n th layer at the bottom ( Fig. 1), the neutral plane position from the top surface is given by: where E i and d i are the modulus and thickness of the i th layer, respectively. The equation assumes linear mechanics, pure bending, and no-slip boundaries between the layers, which suggests zero shear deformation in the layers [35]. These assumptions are usually valid in the case of small bending (bending radius ). Based on the simple bending theory, elastic strain in the stack linearly increases with the distance away from the neutral plane. As a result, nanomembrane devices with an ultra-thin layout contribute to significantly reduced strain when they are strategically positioned at the neutral plane location. This approach has been successfully applied to demonstrate foldable silicon electronic circuits [36] and polymer waveguides with improved bending properties [37]. According to Eq. (1), the neutral plane design requires the photonic elements to be embedded inside the multi-layer structure. While such a configuration is generally compatible with electronic circuits, it may not meet the fabrication requirements or application needs of integrated photonics. For example, encapsulating the photonic components deep underneath a thick top cladding layer prohibits evanescent wave interactions with the external environment which is essential for chem-bio sensing and evanescent optical coupling. One solution to this challenge is to fabricate photonic devices on the top surface of a relatively thin (< 50 μm) flexible substrate with low elastic modulus, which reduces the strain level in the device layer according to the multi-layer bending theory despite that the device layer is not aligned with the neutral plane [35]. This approach has been most frequently implemented in conjunction with transfer printing of semiconductor NMs [1,38,39] and typically allows bending with > 10 mm radius. Further improved bending flexibility can be achieved by engineering the multilayer structure. It is worth noting that Eq. (1) may become invalid due to the elastic modulus mismatch between layers in the case of large bending (i.e. small bending radius). For example, when an elastomer layer with low Young's modulus (~MPa) is sandwiched between two glassy polymer layers with much higher moduli (~GPa), large shear deformation takes place in the elastomer layer to minimize elastic deformation energy in the two rigid glassy polymer layers. Figure 1(b) and 1(c) illustrates such an example where a 50 μm thick silicone elastomer layer (E = 2 MPa) is sandwiched between two polyimide (PI) layers (E = 2.5 GPa) of 10 μm thickness each to form a tri-layer structure. When bent, the silicone layer practically decouples the in-plane strain in the two PI layers and serves as an effective strain relieving agent which reduces the strain in on the multi-layer stack surface by a factor of 5 compared to that in a single PI layer of the same thickness ( Fig. 1(b)). We have recently employed this mechanical design to demonstrate highly flexible (minimum bending radius < 0.5 mm) photonic devices made of chalcogenide glasses, despite the poor mechanical strength of the glass materials [40].

Stretchable photonics: lessons learnt from stretchable electronics
Mechanical stretching and compression offers a convenient way for broadband tuning of the operation wavelengths of photonic devices [13]. In addition, the ability to accommodate stretching is necessary for photonic sensor systems integrated with biological tissues [41]. Stretching and compression are distinctively different from bending in that there is no "neutral plane" during deformation, and thus the aforementioned vertical ("out-of-plane") multi-layer stack engineering approach does not apply. However, the same design rationale of reducing strain at critical points when the devices locate still holds. Local strain in the devices resulting from stretching can be suppressed through in-plane patterning of the photonic structures. Interconnect lines used in stretchable electronics on elastomer substrates provide an excellent example for the design principle. While the metal material constituting the interconnects is hardly stretchable, metal lines patterned into serpentine shapes are able to withstand large elongation without fracture [41]: the mechanism is similar to that of stretching a helix shaped spring. Introducing a polymer material with higher Young's modulus than the elastomer substrate as a support for the meandering lines can further improve robustness of the structure [42]. This design approach is readily applicable to produce stretchable photonic wire waveguides, which can be used to connect photonic components fabricated on stiff, non-stretchable islands. Other geometries, such as wavy buckled structures [43], may also be used to create stretchable components from stiff materials.

Material selection and processing for flexible photonics
From the previous section, we note that through ingenious geometry designs, the deformation capacity of photonic structures can be decoupled from its constituent material's mechanical properties. The insight allows designers to focus on fabrication compatibility and optical properties of materials in the material down selection process for flexible photonic devices. To date, most flexible photonic components are fabricated either using transfer printing or direct monolithic patterning. In this section, we will first discuss the two processing routes. Material design options will then be evaluated based on processing constraints and device functional requirements for two main components of a flexible photonic circuit: passive photonics and active optoelectronics. Transfer printing, alternatively termed stamp printing, has several variants but all of them involve release of the functional layers (e.g. crystalline semiconductor NMs) or structures (e.g. fabricated devices) from the starting substrate, and transfer of the functional layers or structures onto a foreign flexible substrate. Figure 2 shows a general transfer printing process flow. The source material (e.g., SOI, GeOI, III-V multi layers with a sacrificial layer) is first being patterned into membrane (or strip forms) down to the sacrificial layer ( Fig. 2(a)-2(b)). Top membrane layers are then released by undercutting the sacrificial layer ( Fig. 2(c)), with fully released membrane settles down on the handling substrate via van der Waals force ( Fig.  2(d)) (note: this in-place bonding case only applies to thin sacrificial layers. For thick ones, please refer to another paper from Rogers' group [44]). The wet etch step either completely releases the devices, which subsequently settle down and are weakly bonded to the silicon substrate via van der Waals force, or partially undercut the supporting oxide layer beneath the devices to form pedestal structures. The latter method prevents lateral shift of devices during the wet etch process [13,45]. Finally membrane transfer process can take place with either a direct flip transfer or stamp-assisted transfer processes ( Fig. 2(e)-2(f)). In the direct flip transfer process, an adhesion layer is applied on the host (foreign) substrate first. The membrane to be transferred can be picked up by the host substrate directly to complete the transfer. The adhesion layer (e.g., glue) can be dissolved afterwards, if needed. In the stampassisted transfer process, a stamp (e.g., PDMS) was used to press toward the handling substrate, and lift-up the membrane to be transferred. Then the membrane attached to the stamp was picked up and attached to a new host substrate, which can be coated with glue, if needed. Slowly peeling off the stamp or removing the stamp with shear force, the NM can be left on the new host substrate [46].

Transfer printing of semiconductor nanomembrane structures and devices
By releasing high-quality flexible materials from SOI and other source substrates and transfer-printing them to a new host substrate, integration of dissimilar material systems becomes feasible on both rigid and flexible substrates, with individually optimized material structures, energy-efficient and flexible integration process, and intelligent functions. The transfer-printing steps can also be repeated to produce stacked multi-layer photonic structures [47]. Besides Si [48], many materials including III-V semiconductors [38,39], metals [49], and organic semiconductors [50] are all amenable to transfer printing processing. Based on transfer printed NMs on flexible substrates, various high performance photonic and optoelectronic devices have been demonstrated, including flexible photodetectors, LEDs, and solar cells, flexible Si Fano resonance filters, and mostly significantly, fast flexible electronics, with record speed of RF performances, owing to high electron mobility in single crystalline semiconductors [51][52][53][54][55][56][57][58][59][60][61][62]. Many excellent results have also been reported by Lagally et al. [51,53,69] and Rogers et al. [33,41,43] on the unique electronic, photonic, and thermoelectronic, and mechanical properties associated with this new class of inorganic flexible semiconductor membrane material system. To improve the yield of the printing transfer process, Xu et al. reported an innovative stamp printing technique with an adhesion controllable suspended configuration [45]. To address the alignment challenge over the large area flexible substrate, a local alignment scheme in combination with more accurate Si-NM transfer measures for minimizing alignment errors was reported [57]. By realizing 1 μm channel alignment for the Si-NMs on a soft plastic substrate, thin film transistors with a record speed of 12 GHz maximum oscillation frequency were demonstrated (Fig. 3). Recently, based on the strained Si/SiGe/Si-NMs transferred onto flexible substrates, another record was set for fast flexible transistors, with maximum oscillation frequency of 15 GHz demonstrated [63]. These results indicate the great potential of properly processed Si-NMs for high-performance flexible optoelectronics [55].
High quality photonic crystal (PhC) structures have been successfully transferred onto PET substrates based on NM transfer processes, as shown in Fig. 4. PhC air hole structures were first fabricated with the target wavelength of 1550 nm on 260-nm thick silicon-oninsulator (SOI) wafers using e-beam lithography and plasma dry-etching processes. Shown in Fig. 4(a) are a schematic (top) and scanning electron micrographs (SEMs, bottom) of the patterned PhC SOI structures, respectively. The patterned PhC SOI structures were subsequently immersed in aqueous diluted HF solution (49% HF: DI water = 1:4) for several hours to etch away the BOX (buried oxide) layer selectively ( Fig. 4(b)). Once the top patterned PhC Si-NM was completely released, it was rinsed in DI water and transferred onto PET plastic substrates (Fig. 4(c)). A micrograph is shown in Fig. 4(   Compared to the hybrid transfer printing approach, direct monolithic patterning of passive photonic devices on flexible substrates offers a much simplified alternative with improved yield, large-area patterning capability, and potential compatibility with roll-to-roll processing. The typical process flow of monolithic photonic device fabrication on plastic substrates is schematically illustrated in Fig. 5. The flexible polymer substrate is first attached to a rigid handler substrate to facilitate handling in subsequent processing steps, followed by deposition of the thin film device layer via Physical Vapor Deposition (PVD), Chemical Vapor Deposition (CVD), or solution processing (spin-coating and dip-coating). It is critical to ensure the flatness and low roughness of the substrate and film surfaces, since surface height variation can lead to defocus and image distortion during projection lithography, or stitching error during electron beam lithography. Patterning of the thin film into functional photonic structures is accomplished through plasma etching (note that the RF power should be kept low to reduce plasma damage to the polymer substrate [16]), lift-off [34,40], nanoimprint [64], or nanostencil lithography [65]. The monolithic fabrication route has been applied to create optical waveguides, resonators, gratings, and plasmonic nanostructures on flexible substrates. As an example, an optical micrograph of a micro-ring resonator fabricated using thermal nanoimprint in As 2 Se 8 glass films on a PET plastic substrate is shown in Fig. 6(b) [64]. The process starts with making a master for stamp fabrication, which comprises thermally reflowed photoresist lines on a rigid substrate with atomically smooth surfaces. PDMS stamps which inherit the smooth surface finish were fabricated via replica molding, and the photoresist master can be repeatedly used for PDMS stamp fabrication. Photonic devices on flexible substrates were subsequently patterned using single-step nanoimprint at temperatures above the As 2 Se 8 glass transition temperature. The method produced highindex-contrast photonic waveguide structures with a low RMS surface roughness of ~1 nm and optical propagation loss as low as 0.8 dB/cm, comparable to the performance of state-ofthe-art devices on traditional semiconductor substrates.

Materials design considerations for flexible photonics
Heterogeneous integration is currently the most promising approach to high performance photonic devices (also applicable to high-speed electronics), considering the difficulty to engineer indirect bandgap materials (in particular, Si). This argument is evidenced by the recent successes on the hybrid Si evanescent laser, the photonic devices based on the InPbonding-to-Si structure [66,67]. With recent advances in nanostructured material systems, the enabling materials for flexible photonics have expanded into inorganic crystalline and hybrid material systems, such as semiconductor quantum dots, nanowires, tubes, and thin film NMs on flexible substrates. Inorganic crystalline semiconductor NMs (free-standing sheets with nanometer to submicrometer scale thicknesses) have, in the last several years, demonstrated great potential to become a disruptive technology, driven primarily by the successes shown with Group IV crystalline NMs transferred and stacked onto foreign substrates, including both rigid (e.g. silicon and glass) and flexible (e.g. plastics and polymers) substrates. The drivers for this potential are the inherently novel electronic and mechanical properties of these sheets; their flexibility, conformability, biocompatibility, and transferability to other hosts; the ability to introduce strain (and thus novel properties associated with strain) in ways not possible with bulk materials; and the ability to integrate membranes of different materials because of the much better bondability of membranes than bulk material.
Owing to the stackability of nanomembranes based on PDMS transfer printing processes, originally developed by Rogers et al. [46], heterogeneous integration of dissimilar materials becomes very promising without compromising their structural integrity and optical properties. Since active optoelectronic components, in particular light emitters such as LEDs and lasers generally require using pristine single crystal materials to suppress defect-assisted carrier recombination, this unique feature justifies transfer printing as the processing route of choice for active optoelectronic device fabrication. High quality single crystalline Si-NMs have been transferred onto various foreign substrates, such as glass, flexible PET (polyethylene terephthalate) plastics, etc., based on low temperature transfer and stacking processes, developed by various groups [51][52][53][54][68][69][70].
In contrast, optical properties, rather than electronic or optoelectronic properties underlie the material choice for most passive photonic devices. Three types of materials are currently being explored for passive photonic components on plastic substrates: optical polymers, single crystal semiconductors, and inorganic amorphous materials, as all of them can fulfill the basic requirement of low optical loss for passive applications. Optical polymers, in particular fluorinated polymers, exhibit low optical loss down to 0.01 dB/cm in the 1310/1550 nm telecommunication wave bands [71]. In addition, polymers are highly versatile for device processing: polymer waveguides can be fabricated using conventional lithography and etching, direct photo patterning, micro-molding, or imprint/embossing, and they are fully compatible with plastic substrate integration. However, polymer materials by themselves generally offer relatively low index contrast. In comparison, single crystalline semiconductor NMs and inorganic amorphous materials possess high refractive indices suitable for highdensity integration, and both types of materials have been used for low-loss flexible planar passive device fabrication [13,16,40,45,55]. Since passive applications do not require perfect crystallinity, amorphous materials feature unique processing and integration advantages over their single-crystalline counterparts: they are known to exhibit very low optical loss [72], and can be monolithically integrated on polymer substrates free of lattice matching constraints. Multi-layer vertically integrated photonic structures can be readily fabricated as well by multiple sequential deposition and lithographic patterning steps with superior throughput and alignment accuracy [34]. It is important to choose a deposition process whose thermal budget is compatible with the polymer substrate material. For example, low temperature Plasma Enhanced Chemical Vapor Deposition (PECVD) was used to deposit a-Si films on heat stabilized polyester substrates at 200 °C for flexible photonic strain sensor fabrication [16]. The low CVD deposition temperature, however, compromised the optical quality of films and partially accounts for the high optical loss (~100 dB/cm) in the fabricated devices. A recent paper by Rangarajan et al. reported the deposition of low-loss (< 3 dB/cm) silicon oxynitride films via PECVD at 150 °C, which suggests a possible new route for high-performance flexible passive photonic device fabrication provided that plasma damage to the polymer substrate can be somehow circumvented [73]. Evaporation and sputtering are among the most common PVD deposition techniques, and amorphous transition metal oxides (e.g. TiO 2 [74]) and chalcogenide glasses [75] are both amenable to these lowtemperature PVD deposition methods. It is worth noting that sputtering deposited Si 3 N 4 and a-Si films exhibit high optical losses due to presence of silicon dangling bonds and are thus not suitable for flexible photonics applications [76,77]. Finally, solution processing (sol-gel, spin-coating, dip-coating, etc.) offers promising low-cost, large-area alternatives to PVD and CVD deposition techniques, although optical attenuation resulting from residual solvents and microstructure inhomogeneity needs to be mitigated through careful processing optimization [78][79][80].

Flexible photonic devices: function dictates form
A broad spectrum of photonic structures and devices integrated on flexible substrates have been demonstrated using the aforementioned fabrication techniques. In this section, we highlight several device application examples where mechanical flexibility either constitutes the key feature enabling the target applications, or contributes to significantly improved device performance.

Optical links for board-level interconnects
The ever increasing need for high-bandwidth data traffic between computer chips in High Performance Computing (HPC) has pushed electrical interconnects to their functional limit. Several optical interconnect schemes have been demonstrated as potential solutions to resolve this "interconnect bottleneck". The examples include designs based on on-board polymer waveguide interconnects (i.e. the "optoelectronic printed circuit board" configuration) [81], multi-mode fiber ribbons [82] or multi-core fibers [83], and flexible optical waveguides [4][5][6][7]. Since chips are usually assembled onto the circuit board using "pick-and-place" surface mounting technologies which have limited spatial placement accuracy (typically in the order of 10 μm or larger), mechanical flexibility of optical interconnects is essential to accommodate the limited chip-to-chip relative positioning accuracy. Compared to other approaches, flexible optical interconnects also allow significant deployment flexibility and scalability to ultra-high bandwidth density.   (Fig. 7(b) and 7(c)). Guidotti et al. fabricated and tested flexible optical waveguide interconnects directly coupled to edge-viewing lasers photodetectors [5]. Since the optical link only interfaces with the chips electronically, no optical alignment is required during packaging. In this sense, the flexible optical interconnects can be viewed as the planar counterpart of active optical cables, although their planar design offers much higher data bandwidth scalability to well beyond 10 Tbps, as the interconnect bandwidth density is only limited by the physical sizes of on-chip contact pads and solder bumps, and not by the limited optical alignment accuracy or optical coupler size. Reach of such planar flexible optical links, however, is ultimately limited by optical waveguide loss. IBM and Dow Corning recently reported flexible polymer waveguides with optical losses as low as 0.05 dB/cm, which is adequate for chip-to-chip and even board-to-board interconnects [7].

Broadband tunable photonic devices
Flexible photonics offers a unique platform to exploit optomechanical coupling in integrated photonics systems, and realize broadband photonic tuning where the wavelength tuning rage far exceeds what can possibly be achieved using conventional electro-optic or themo-optic tuning. The optomechanical coupling originates from two mechanisms: the material response characterized by strain-optic coefficients dn/dε, and the device shape change as a result of mechanical deformation. For instance, optomechanical coupling manifests itself as resonant wavelength shift dλ in a resonant cavity under mechanical strain ε [16]: where Γ i and (dn/dε) i are the optical confinement factor and strain-optic coefficient in the i th cavity material, L is the cavity length, and n g and n eff denote the group index and effective indices, respectively. In Eq. (2), the first term on the right hand side (RHS) represents the photoelastic (i.e. strain-optic) material response, the second term manifests the cavity length change, and the third term results from the cavity cross-sectional geometry modification. Compared to the original formalism used in Ref. 16, Eq. (2) separates the contributions from materials and cross-section geometry change. As an example of the latter effect, optical properties of a coupled waveguide system can be efficiently tuned through varying the gap width between the two coupled waveguides [85]. Optomechanical coupling in flexible photonic resonators has been investigated by several groups [16,40,86]. For resonators made of the same material and of the identical dimensions, Eq. (2) suggests the same dλ/dε coefficient regardless of the flexible device configuration. This conclusion was experimentally validated by our in situ transmission measurement of flexible micro-disk resonators [40]. Figure 8(a) shows a flexible micro-disk resonator chip under testing: optical transmission through the bus waveguides was monitored in situ using fiber end fire coupling as the flexible chip is bent. The resonant wavelength shift as a function of bending radius was measured on several different samples with different top/bottom cladding layer thickness combinations and is plotted in Fig. 8(b). While Fig. 8(b) indicates that the resonant wavelength shift may differ significantly in both magnitude and sign as the device configuration changes, re-plotting the same set of data as a function of local strain at the micro-disk resonators (Fig. 8(c)) clearly shows that all data points nearly fall on a straight line which agrees well with Eq. (2).
In the example shown in Fig. 8, the main contribution to resonant wavelength shift comes from the strain-optic material response. Due to the relatively small strain-optic coefficient of most engineering materials, tuning relying on photoelastic property change of materials (i.e. the first RHS term in Eq. (2) is inherently limited in tuning range. Broadband tuning has been accomplished on flexible photonics platforms by mechanical deformation of the cavity geometry (i.e. the second and third RHS terms in Eq. (3). Yu et al. demonstrated a stretchable photonic crystal cavity with over 60 nm wavelength tuning range near 1550 nm telecommunication wavelength (Fig. 9). The PhC consisted of a hexagonal array of Si rods, and the line defect cavity was formed by reducing the diameter of 19 rods along a line. The PhC was embedded in PDMS elastomer which rendered the structure mechanically compliant. When stretched, the cavity length changes leading to a resonant wavelength shift. The high yield limit of PDMS accounts for the large 67 nm wavelength tuning range, which well exceeds other existing tuning mechanisms. The maximum tuning range in a stretchable PhC cavity is ultimately limited by degradation of optical confinement and cavity Q-factor as the PhC geometry is deformed. The same mechanism has also been implemented to tune the optical resonant wavelength of metamaterials [88] and surface plasmon polaritons [89]. Besides resonant wavelength tuning, mechanical deformation was also used to modulate optical coupling by varying the gap width between photonic elements. The technique was applied to tune Mach-Zehnder interferometers and micro-ring resonators by Chen et al. [13].

Flexible detector/LED arrays
LED array integration on plastic substrates is a key enabling technology for flexible displays used in consumer electronics. Mechanical flexibility, on the other hand, allows the assembly of non-planar imaging arrays with hemispherical Field of View (FoV), low aberration, and conformal integration capability on complex curvilinear surfaces [2,90].
We have developed a frame-assisted membrane transfer (FAMT) process for the transfer of large-area crystalline semiconductor NMs, especially for the fragile material systems such as GaAs and InP compound semiconductors used inflexible detector and LED fabrication [38]. Large area flexible photodetectors, solar cells and LED arrays all have been demonstrated experimentally, as shown in Fig. 10, based on transferred InP nanomembranes. The measured flexible InP p-i-n photodetector performance characteristics under bending (bending radius = 32.3 mm) are shown in Fig. 11(a). Dark current of 1 μA was obtained for a 3 × 3 mm 2 device. The measured current-voltage curve for the flexible InP solar cell under bending (bending radius = 42.1 mm) is shown in Fig. 11(b). We obtained photovoltaic solar cells with open circuit voltage of 0.62 V, and power efficiency of 1.44%. Higher efficiency cells are feasible with optimized structural design by optimal absorption, light trapping, and photon recycling.  The thin PECVD SiO 2 layer with total thickness of 600 nm was universally deposited to passivate all devices. The devices are patterned and access holes of both top and bottom contacts are formed by photolithography and passivated layer is etched by using the RIEetching. Finally, the interconnect layer was patterned to connect all of the devices together forming the device arrays. A completed flexible InP LED arrays were formed, with top and bottom contacts (Fig. 12(a)). The SEM image of top ring contact and interconnect line connecting to the stripe to form the arrays and laying on the SiO 2 passivation layer is shown in Fig. 12(b). The InP LED was transferred onto flexible PET substrate as shown in Fig.  12(c). In Fig. 12(d) shows the device was under probing test and Fig. 12(e) illustrates the device was transferred onto gold coated PET substrate.

Sensing systems integrated on biological tissues
Many reports appeared over the last year, in the area of NM-based bio-integrated flexible devices and systems, where high-performance semiconductor device functionality can be achieved in forms that enable intimate, conformal contacts to flexible/conformal surfaces. Kim et al. presented a flexible/stretchable systems of InGaN microscale transfer printed inorganic LEDs (m-ILEDs) with wireless powering schemes, for implantable devices that could be used to accelerate wound healing, activate photosensitive drugs, or to perform imaging and spectroscopic characterization of internal tissues [91][92][93]. One of the more sophisticated examples of flexible optoelectronics in biology was demonstrated in a recent paper, where tiny LEDs were injected into the brains of mice to provide wireless control over behavioral responses through optogenetic techniques [94]. Along the direction of biointegrated optoelectronics, a physically transient form of silicon electronics was proposed [95], where nanomembrane devices are built on bio-resorbable substrates (e.g. silk) and the complete material system disappears physically at prescribed times and at controlled rates. An implantable transient device that acts as a programmable non-antibiotic bacteriocide was demonstrated and sealed in silk packages.

Challenges, opportunities, and future trends
Even though flexible integrated photonics is still in its infancy, many exciting applications have already been demonstrated as noted in the previous section. New materials and processing methods, as well as innovative optical and mechanical device designs are expected to further improve the performance of flexible photonic systems. To name a couple of examples, roll-to-roll manufacturing of flexible photonic structures will enable device fabrication over large substrate areas inaccessible using conventional processing techniques on rigid substrates, and integration of flexible photonic elements with MEMS actuating systems can lead to novel tunable photonic devices exploiting their singular strain-optical coupling behavior.
While flexible photonics offers a unique platform to explore strain-optical coupling for photonic tuning, such coupling is undesirable or even harmful for other applications where stable device operation independent of substrate geometry is essential (e.g. for sensing systems integrated on biological tissues). Controlling the coupling via nano-mechanical designs (e.g. multi-layer designs implemented in Fig. 8(b), or locally stiffened islands to minimize deformation of devices) is thus critical for practical applications of flexible photonic systems.
Optical coupling into and out of flexible photonic devices (in particular waveguide-based devices) presents another remaining technical challenge for flexible photonics. Grating coupling and butt coupling have both been employed, although both coupling scheme severely limit the deployment degrees of freedom of flexible photonic systems. Complete photonic integration of both active and passive elements (and possibly with power sources) on flexible substrates will significantly simplify system packaging and is highly desirable for many of the aforementioned applications, although careful thermal designs are required to mitigate the heat dissipation issue especially for free-standing flexible photonic systems where heat sinks are not readily accessible.

Concluding remarks
In this paper, we reviewed the material design, mechanical engineering, and photonic applications of flexible integrated photonics. We show that the requirement for structural flexibility and stretchability can be decoupled from its constituent material mechanical properties through ingenious nano-mechanical designs. Therefore, mechanically compliant structures can be made out of materials that are rigid in their bulk form. For example, crystalline semiconductor NMs, the key enabling material for flexible electronics, are also the material of choice underlying flexible active optoelectronic device applications, while amorphous glasses are promising candidates for passive optical components. Based on these novel materials and innovative device designs, flexible integrated photonics has found niche applications for interconnects, sensing, imaging, and photonic tuning. We anticipate that identification of new application fields where mechanical flexibility either constitutes the key feature enabling the target applications or contributes to significantly improved device performance will certainly accelerate penetration of the technology into diverse market sectors in the future.