Remote heteroepitaxy of GaN microrod heterostructures for deformable light-emitting diodes and wafer recycle

See allHide authors and affiliations

Science Advances  03 Jun 2020:
Vol. 6, no. 23, eaaz5180
DOI: 10.1126/sciadv.aaz5180


There have been rapidly increasing demands for flexible lighting apparatus, and micrometer-scale light-emitting diodes (LEDs) are regarded as one of the promising lighting sources for deformable device applications. Herein, we demonstrate a method of creating a deformable LED, based on remote heteroepitaxy of GaN microrod (MR) p-n junction arrays on c-Al2O3 wafer across graphene. The use of graphene allows the transfer of MR LED arrays onto a copper plate, and spatially separate MR arrays offer ideal device geometry suitable for deformable LED in various shapes without serious device performance degradation. Moreover, remote heteroepitaxy also allows the wafer to be reused, allowing reproducible production of MR LEDs using a single substrate without noticeable device degradation. The remote heteroepitaxial relation is determined by high-resolution scanning transmission electron microscopy, and the density functional theory simulations clarify how the remote heteroepitaxy is made possible through graphene.


Light-emitting diode (LED) displays (15) are transforming the landscape of display technology for higher dynamic range, improved readability under various illumination conditions, and unprecedented form factors applicable to surface-mountable electronics (6, 7), biomedical devices (8, 9), and transportation vehicles. The application-driven development of LEDs requires achieving high-performance and cost-effective manufacturing. However, the concurrent achievement of the main requests of future LEDs has been hindered by the preparation of high-quality materials limited to structurally commensurate substrates (10). Furthermore, the brittle, less flexible properties of LED materials in a thick-film form make it difficult to use them for extremely flexible free-form devices unless they are minutely diced and lifted off from wafer. One-dimensional nano/microstructures, such as inorganic nanowires (NWs) and microrods (MRs), have been considered as a solution to overcome materials compatibility problems because nucleation at the small area and strain relaxation along side surfaces of the elongated structure substantially reduce structural defects in the light-emitting medium (11, 12). Thus, the crystalline NW and MR LEDs can be fabricated onto various substrates encompassing homoepitaxial wafer (13), crystalline Si (14), polycrystalline metal foils (15), and amorphous glass (16). The ability of heterogeneous integration enhances the LED efficacy via direct control of driving current on circuits and improves durability through heat dissipation on efficient heat sinks, such as metal foils (17). Moreover, the geometry of NW LEDs offers freedom of LED packaging form factors because those can be manufactured on flexible substrates (18). Although the NW LEDs have the potential to be a universal solution to existing problems of LED development, there are still materials issues to be addressed.

Physical characteristics of NWs, beyond the crystallinity, are still affected by substrates. Stacking faults and strain near interface are generated from covalent (or ionic) bonds between NWs and a substrate (1921), which may result in deterioration of LED performances. Other issues are the formation of an unintentional interfacial layer (e.g., SiNx for GaN NWs grown on Si) and contamination of NWs via atom diffusion from the substrate (e.g., GaN NWs on Ti) (2224). Thus, fabrication of high-quality NWs integratable in LED architecture has become an important issue. For the use of graphene substrate, because graphene does not form the covalent bonds with the overlayer NWs, the interfacial strain can be efficiently released through the small footprint (25). Furthermore, the high chemical inertness and thermal stability of graphene do not cause the contamination to the NWs even above 1000°C (2628). Hence, graphene is one of the ideal substrate materials that can resolve the issues mentioned above for growing high-quality optoelectronic components.

The remote epitaxy is the epitaxy technique for obtaining a single-crystalline overlayer whose crystallographic registration can be dictated from wafer through an ultrathin atomic layer (i.e., graphene). It is a versatile solution of the materials issue of NW- and MR-based LEDs because the technology enables us to grow high-quality NWs on perfectly matched substrates and detach the NWs from the substrate without a destructive method to remove a sacrificial layer (2933). Thus, the remote epitaxy may allow the reuse of the wafer substrate, which is a considerable material cost for LED production. This versatile epitaxy technique has been recently applied to prepare unidirectionally aligned semiconductor MR arrays even on the recycled wafer after the MR overlayer is released (31, 34). Nonetheless, there remains a challenge to develop practical MR devices that are implemented with the strengths endowed from the remote epitaxy.

Vertically aligned MRs offer an ideal geometry of spatially separate device arrays that provide a desirable platform for use in flexible devices (7, 18, 3537) manufactured in economic ways. Among many semiconductor wires, GaN-based heterostructures have validated their practical use in high-performance optoelectronic devices, including LEDs (14, 3840), lasers (41), solar cells (42), single-photon sources (43), and red-green-blue subpixel display components (36, 44), which entirely take advantage of excellent physical properties of GaN, including complete compositional tunability of InxGa1–xN, high quantum efficiency, etc. Thus, there will be plentiful opportunities once the remote epitaxy is successfully adapted to fabricate GaN MR optoelectronic devices. To date, the remote epitaxy has dealt with homoepitaxial (30, 31) or nearly lattice-matching heteroepitaxial epilayer-wafer systems (34), which typically require the use of high-cost wafers. However, it is well known that the use of Al2O3 wafer with a large in-plane lattice mismatch of 16.1% is quite beneficial because of its large scalable and well-compatible processing for the conventional GaN growth. Meanwhile, Park et al. (28) have recently reported that the sapphire wafer is more stable than GaN substrate under harsh reaction conditions of hydrogen- and ammonia-rich high growth temperature of more than 1000°C when the surface is coated with graphene, and graphene is also robust in the same ambience. Thus, it is expected that such tolerance of Al2O3 is exploitable to allow recycling of the wafer after the release of the overlayer. Thus, it is imperative to achieve the remote heteroepitaxy of GaN vertical wires on Al2O3.

Here, we report on the remote heteroepitaxy of GaN MR heterostructures on c-plane Al2O3 wafer through graphene, which enables (i) fabrication of deformable LEDs and (ii) recycling of the underlying substrate after exfoliation of the MR overlayer. The LEDs are deformed and tailored to diverse shapes, and the LED performances are characterized in terms of electrical and electroluminescence (EL) properties with respect to repetitive bending cycles. High-resolution scanning transmission electron microscopy (HR-STEM) confirms the remote heteroepitaxial relationship across graphene, and Raman spectroscopy reveals the existence of graphene after high-temperature GaN growth. Density functional theory (DFT) calculations elucidate how the remote heteroepitaxy is made possible through graphene. After exfoliation, the native c-Al2O3 wafer is recycled, and the deformable LEDs obtained on the recycled wafers are characterized in terms of device performances.


Deformable LEDs were fabricated by remote heteroepitaxy of GaN MRs (Fig. 1A). The basic strategy for fabricating deformable LEDs begins with remote heteroepitaxy of MR LEDs on graphene-coated Al2O3 (0001) wafer via metal-organic vapor-phase epitaxy (MOVPE) (Fig. 1B), followed by transfer of the as-grown MR LED arrays onto a Cu plate (Fig. 2A). For substrate preparation, chemical vapor deposition (CVD)–grown polydomain single-layer graphene (SLG) was transferred twice onto c-Al2O3 wafer, which makes an ultranarrow gap for the epitaxy of GaN remotely from c-Al2O3. For clarity, it is herein noted that wafer is intended to refer to a single-crystalline substrate (i.e., Al2O3), while the substrate is a substance that growth was carried out on (i.e., graphene/Al2O3). The p- and n-electrodes of Ni/Au and Ti/Au were formed on top and bottom of MRs, respectively. The graphene layers remained on the wafer after the exfoliation process (fig. S1), unlike our previous results (31, 34). As shown in fig. S2, the annealing led to the intimate interface between graphene and Al2O3 wafer so that the graphene remained on the wafer after the delamination process. As a result, it was convenient to make an ohmic electrode of Ti/Au on the bottom of MR LEDs. In between two metal electrodes, an insulating polyimide layer was spin-coated, which electrically isolates the two electrodes and structurally supports the MR LED arrays as a deformable molding film. Then, the MR LED arrays embedded in polyimide film were transferred onto a conducting Cu plate using thermal release tape (Fig. 2A). As illustrated in Fig. 1B, every single MR includes the core n-GaN stem, three-period radial InGaN/GaN multiple quantum wells (MQWs), and outermost shell p-GaN layer, which collectively corresponds to core/shell MR p-n junction. Detailed procedures for growth, device fabrication (fig. S3), measurements, and theoretical computations are described in Materials and Methods.

Fig. 1 Remote heteroepitaxy of GaN MR heterostructures on c-Al2O3 across graphene.

(A) Photograph of EL light emission from MR LED in a bent form. (B) Cross-sectional schematic of MR heterostructures grown on graphene-coated c-Al2O3 wafer. (C) Annular bright-field (ABF) STEM image of remote epitaxial heterointerface of GaN/graphene/c-Al2O3 focused on graphene. The location of the doubly stacked SLG is marked with red wedges. Atomic-resolution filtered ABF STEM images of (D) GaN MR and (E) c-Al2O3 taken around the heterointerface area. (F) Tilt-view FE-SEM image of as-grown MR LED arrays of radial p-n junction heterostructures. (G) Photograph of EL light emission from 5 mm by 5 mm area MR LED in a flat form at 100 mA. Photomicrographs of MR LED (H) without and (I) with current injection of 100 mA, taken from the boxed area in (G). The off-state photomicrograph of (H) was taken under normal lamp illumination conditions.

Fig. 2 Diversely deformable MR LED.

(A) Schematics illustrating key procedures for fabricating the deformable LED, including remote heteroepitaxy of MR LED arrays and transfer onto conducting metal plate. (B) A series of photographs of cyan MR LED (λ = 500 nm) deformed in various shapes, such as twisted, 90°-folded crumpled, and 180°-folded forms, operated at 100 mA. The inset in the rightmost image of (B) is a schematic illustrating the geometry of MR arrays in the folded form. (C) Photographs of blue MR LED (λ = 450 nm) mounted on various surfaces, including a pen and thin wall edge of plastic box, operated at 100 mA. It is noted that the blue MR LED shown in (C) is produced from a recycled wafer. (D) Photographs of 10 mm by 10 mm MR LED (λ = 450 nm) tailored to be fitted to two back legs of a minifigure (left two panels); photographs of LED-adhered LEGO minifigure with different leg postures (right three panels). The LED was operated at 100 mA. Scale bars (D), 10 mm. Photo credit: Junseok Jeong, Sejong University.

Cross-sectional HR-STEM image of the remote heteroepitaxial GaN MR/graphene/Al2O3 shows the existence of doubly stacked SLG after the high-temperature growth of MR LEDs (Fig. 1C), which was also observed by Raman spectra (fig. S1). The delamination gap was induced between GaN and graphene/Al2O3 during the cross-sectional milling process for transmission electron microscopy (TEM) observation, which might evidence the weakly bound noncovalent heterointerface. The graphene interlayer was observed without aperture-associated discontinuity over the whole interface area in low-magnification TEM observation (fig. S4), implying that the epitaxy was not initiated from covalent heteroepitaxy through a graphene aperture. Otherwise, the use of graphene with apertures did not allow complete delamination of the MR overlayer because the MRs grown through apertures are strongly bound to Al2O3 due to the covalent bonds (fig. S5). To avoid the presence of these apertures, we doubly stacked SLG on the Al2O3 wafer to form double-layer graphene on the wafer (fig. S6), which enabled a nearly perfect delamination of the MRs from the native substrate. It is noted that the crystalline quality was degraded when using graphene thicker than triple-layer graphene.

In Fig. 1 (D and E), atomic-resolution STEM images show that GaN MR was grown along the Ga-terminated wurtzite-[0001] (or c+) direction on Al-terminated corundum Al2O3(0001) (of c+ direction), with the heteroepitaxial relation of (0001)[1¯01]GaN║(0001)[1¯1¯20]sapphire across polydomain graphene. This observation signifies 30° rotational in-plane epitaxial relationship without polarity inversion, similar to covalent heteroepitaxy. It is well known that such in-plane rotational epitaxy typically occurs in covalent heteroepitaxial systems with large lattice mismatch to reduce the surface dangling bonds and interfacial stress caused from the lattice mismatch (45). In contrast, domain aligned growth is preferably yielded without interfacial strain even for a large mismatching system when the growth is driven by the van der Waals (vdW) epitaxy (37, 46, 47): If the growth had been ruled by the vdW attracting force from graphene, the epitaxy of GaN must have shown the domain aligned relation with graphene. Accordingly, our TEM observation on remote epitaxial relation implies that the influence of the underlying wafer on the overlayer GaN was not obstructed by the graphene interlayer.

Field-emission scanning electron microscopy (FE-SEM) further revealed long-range ordering of hexagonal sidewall {101¯0} facets of vertical MRs over an area of at least 1 cm by 1 cm (Fig. 1F and fig. S7). If the graphene had ruled the epitaxial relation, the homogeneous MR-sidewall alignments must have been observed only within each domain (size ~5 to 20 μm) of graphene (25, 34, 37, 48). Thus, the SEM observation also corroborates that the heteroepitaxial relation was determined remotely from single-crystalline c-Al2O3 rather than polydomain graphene. The mean values of density, height, and diameter of MRs were measured to be (3.0 ± 0.3) × 104 cm−2, 16.1 ± 4.1 μm, and 14.4 ± 2.2 μm (± denotes SD), respectively.

As a comparative study, the GaN MRs were grown on the Al2O3 wafer partially covered with graphene, as displayed in fig. S8. Both the surfaces of graphene-coated Al2O3 and bare Al2O3 exhibited that the vertical MRs were grown with the same in-plane ordering of sidewall facets, irrespective of the existence of graphene. This indicates that the crystallographic orientation of overlayer GaN was exclusively ruled by the Al2O3 wafer for both the surfaces. Noticeably, the MRs grown directly on Al2O3 were not delaminated at all from the wafer because of strong covalent bonds.

The MR LED arrays grown by remote heteroepitaxy exhibited good EL light emission properties. For example, large-area EL emission with uniform emission color was obtained over the entire chip area (Fig. 1, A and G, and fig. S9), and the EL emission area also covered the entire MR growth area (fig. S10). One reason for the successful fabrication of large-area LED panels is the thick top electrode of Ni/Au (15/15 nm) with a low sheet resistance of 4.2 ohms per square that enables good current spreading for homogeneous EL emission across the entire LED panel area. However, because of the thickness, the optical transmittance was as low as ca. 51% at 500 nm wavelength. Photomicrographs show that almost every single MR LED exhibited EL emissions with uniform cyan color under forward bias (Fig. 1, H and I). This result signifies that the remote heteroepitaxy and thermal release tape–assisted transfer technique are viable for fabricating large-area bendable LEDs with a high device yield.

Because remote epitaxy uses the gap layer of graphene, the vdW interface formed by sp2 bonds of graphene easily allows the delamination of MR LEDs and transfer onto a surface of interest using thermal release tape, as schematically depicted in Fig. 2A. Thus, the MR LED arrays transferred onto a Cu plate were able to withstand large deformations such as twisted, right-angle inward/outward folds, random crumpling, and 180° folds without severe degradation to the EL emission area and color within at least several time extreme deformations (Fig. 2B and fig. S11). Noticeably, the MR LED arrays for blue EL, fabricated on a recycled Al2O3 wafer after delamination of other MR LEDs, were attached on various surfaces with different curvatures as well (Fig. 2C). The method and results of wafer recycle are described later. Moreover, the 10 mm by 10 mm MR LED was tailored with normal scissors to fit to the surface and shape of two small legs of a LEGO minifigure. Figure 2D shows the LED attached to the lower half of the minifigure, and the tailored LED was reliably operated in different postures of the minifigure’s legs. This outcome suggests that the MR LED can be tailored and attached to many static and even moving surfaces, such as articular parts in a robot.

The LED performances were evaluated by measuring the current-voltage (I-V) characteristic curves and EL spectra as a function of bending radius (Rb) between ∞ and 10 mm. In a flat form (Rb = ∞) without deformation, the I-V characteristic curve shows a typical diode rectification feature with an EL turn-on of ~4 V, while with the EL intensity increased rapidly with an increase of electrical current above the turn-on (fig. S12). Figure 3A compares the I-V characteristic curves measured at Rb = ∞ and 10 mm, showing comparable electrical rectification curves with an electrical threshold of ~4 V and a leakage current of ~5 × 10−3 A at −5 V. It is obvious that the electrical and EL properties are not as good as those of commercial thin-film LEDs. One of the main reasons is the double contact at the bottom on which both the core n-GaN and shell p-GaN has contact with n-electrode of (Ti/Au). This electrical shunt factor caused the leakage current to be higher than that of thin-film LEDs by two or three orders of magnitude. Hence, it is necessary to spatially and electrically isolate the p-GaN bottom from Ti/Au to improve the LED characteristics. Also, the structural uniformity as well as contact optimization is needed to improve the MR LED performance.

Fig. 3 Electrical and electroluminescent properties and repetitive bending cycle test.

(A) I-V characteristic curves and (B) EL spectra of the MR LED bent at radii of curvatures of ∞ (blue lines) and 10 mm (red lines). For the EL measurement of (B), the LED was operated at the same current of 100 mA. (C) I-V characteristic curves measured after repeating the bending cycles from 1 to 1000 times between radius curvatures ∞ and 10 mm. (D) EL peak position (red empty squares) and EL intensity (blue solid circles) measured as a function of bending cycle up to 1000 times. For the measurements, the MR LED was flattened and operated at 100 mA after the bending cycles. Insets are photographs of EL emission after bending cycles of 0, 500, and 1000 times. A.U., arbitrary units.

The EL turn-on was measured to be ~4 V at both curvatures. Furthermore, the light emission characteristics, such as the EL peak position (at 501 nm for cyan EL) and intensity, exhibited no appreciable differences against the bending deformation at a fixed injection current of 100 mA (Fig. 3B). Because the strain by stress on InGaN/GaN MQWs typically induces a considerable change of the luminescence spectrum (49), the absence of the EL peak shift suggests that the mechanical strain applied to the MR LED arrays was almost negligible for an Rb of 10 mm.

Regarding the origin of EL, the light emission is surmised to be emitted from the three-period radial InGaN/GaN MQWs formed on the MR sidewalls. Considering the MR geometry with moderate aspect ratio (mean diameter, 14.4 μm; mean height, 16.1 μm), it can be plausible that not only the sidewall but also the topside MQWs would emit the EL emission. In our TEM observation (fig. S13A), the typical p-GaN thicknesses of sidewall and topside were measured at ca. 2.4 and 5.5 μm, respectively, because of anisotropic growth for MR. Because the resistivity of p-GaN is much higher than that of n-GaN by three orders of magnitude, the current flows preferentially through the sidewall part rather than the topside for minimization of total resistance, as depicted by an equivalent circuit model in fig. S14. Thus, the EL emission was thought to mostly occur on the sidewall MQWs. To eliminate the unwanted EL from topside MQWs for higher EL color purity, it is necessarily desirable to remove the topside metal electrode (or MQWs) by using a chemical-mechanical polishing technique, as demonstrated by Tomioka et al. (19).

The full width at half maximum (FWHM) of EL emission was 54 nm for the cyan EL, wider than that of thin-film LEDs (36 nm) for the same emission wavelength of ~500 nm (fig. S15). The wider FWHM is tentatively due to the MR-to-MR deviations in the EL peak position, or another possible reason is the tall MR that causes the graded InGaN QWs along the sidewall (50).

In repetitive bending deformations, the EL and electrical characteristics were stably maintained for a bending cycle of more than 1000 times between the Rb of ∞ and 10 mm. As shown in Fig. 3C, the LED exhibited almost the same I-V curves after various repetitive bending cycles. Also, the key parameters for EL emission, including the integrated EL emission peak intensity and peak position, showed no noticeable degradation up to 1000 bending cycles (Fig. 3D). Hence, deformable LEDs fabricated via remote heteroepitaxy are reliably robust against bending deformations such that they can be mounted (or attached) on many curved surfaces without structural damage that typically causes malfunctions, such as the collapse of MR arrays and fracture of electrodes or MR/electrode junction.

It is imperative to explore how the remote heteroepitaxial relation was determined by the underlying wafer substrate despite the presence of the monolayer or bilayer graphene. To this end, the atomic configurations and charge density (ρ) distributions of SLG or bilayer graphene (BLG)/Al2O3 substrates were estimated through the DFT calculations. The in-plane alignment was set to be [1¯21¯0]BLG║[11¯00]sapphire with A-B–sequence-stacked graphitic BLG (Fig. 4), and BLGs with misaligned graphene layers were also simulated, as shown in fig. S16 (G to L). Figure 4A shows DFT-simulated atomic configurations in the substrate layout of the BLG adhered onto c-Al2O3, in which the average distance of bottom and upper graphene was 3.63 and 7.07 Å, respectively, from the surface of Al2O3. The surface ρ distributions of the bottom and upper graphene of BLG were simulated by mapping the charge density at an altitude of 4.4 and 7.3 Å, respectively, as marked with blue lines in Fig. 4A. In Fig. 4B, the plan-view tomographic ρ contour map at the surface of bottom graphene shows that the charge transfer occurred to exhibit regular patterns of honeycomb-shaped negative charge (red region) and triangular arrangement of local positive charge (blue spots). Specifically, the positive charge regions formed locally on every carbon atom placed on top of Al atoms of the Al2O3 surface, while the negative charge patterns occurred on carbon atoms that are close to O atoms. This indicates that a substantial amount of the charge was transferred across the noncovalent interface of the graphene/c-Al2O3 because of close distance.

Fig. 4 DFT calculation.

(A) Atomic configuration of BLG/c-Al2O3 substrate with in-plane alignment of [1¯21¯0]BLG║[11¯00]sapphire. Plan-view charge density (ρ) contour maps of BLG/c-Al2O3, tomographically sectioned at the heights of (B) 4.4 Å and (C) 7.3 Å apart from the topmost surface of Al2O3, as marked with blue lines in (A). (D) Cross-sectional ρ distribution contour map sectioned across the three Al atoms, as marked with a red line in (B) and (C). The maps were simulated by calculating ρBLG/sapphire − ρBLG to eliminate the background charge density coming from graphene layers. The ρ simulation was performed in the range between −8 × 10−5 e Å−3 and 8 × 10−5 e Å−3. The locations of Al and O atoms in the plan-view contour maps are denoted as large and small dot-line circles, respectively, as marked in the inset box.

The same trend of the surface ρ distribution appeared on the SLG/c-Al2O3 substrates with two differently aligned SLG, as shown in fig. S16 (A to F), implying that graphene alignment cannot affect the remote epitaxial relation between GaN and Al2O3 for the use of SLG. This elucidates the long-range ordering of MR hexagonal facet directions over the whole substrate area despite the use of polydomain CVD graphene (fig. S7). The charge of Al2O3 did not reach a height of 4.4 Å without the graphene, as simulated in fig. S16 (C and F). Hence, we surmise that the diametrically aligned pz orbitals on both sides of graphene intermediated the charge transfer from Al2O3 to such a long distance of 4.4 Å along the z direction.

On top of the upper graphene of BLG (Fig. 4C), the ρ contour map displays a nearly charge-neutral feature for the use of BLG. This feature was also observed in vertical cross-sectioned maps for the several representative alignment combinations of the bottom and upper graphene (Fig. 4D and fig. S16, H, I, K, and L). It was figured out that the transfer of the upper graphene on the bottom one hardly changed the overall shape of ρ distribution of the bottom graphene, as seen by comparing Fig. 4B and fig. S16B. The periodicity of alternate charge density on the bottom graphene was not severely distorted by placing the upper graphene (fig. S16, G to I and J to L). The amount of net charge formed on the bottom graphene was estimated to be much greater than that on the upper one by two orders of magnitude so that the electric attraction from the bottom graphene/Al2O3 can penetrate through the charge-neutral upper graphene. This explains how the remote epitaxy was made possible through the doubly stacked SLG. In other words, because of the strong influence of Al2O3 atoms, the remote epitaxial relationship of GaN and Al2O3 was the same as the covalent epitaxy. Considering the charge transfer attenuation across BLG, the use of thicker graphene cannot drive the remote epitaxy. In practice, we observed that the alignment of MRs was seriously degraded by using trilayer graphene or thicker.

The noncovalent, weakly bound heterointerfaces of the remote epitaxial GaN/graphene/Al2O3 structure allowed delamination of MR arrays for fabricating the deformable LEDs. After exfoliation of MRs, the surface of original Al2O3 wafer was clean and smooth, like a brand-new wafer (fig. S17B), but graphene remained on the surface of the wafer, as confirmed by Raman spectroscopy (fig. S1). The as-delaminated native substrate was refurbished to be reused for remote heteroepitaxy. For wafer refurbishment, organic solvent-based cleaning, thermal treatment at 900°C, and reactive ion etching were carried out to completely remove the remaining graphene as well as the residues inevitably left from the use of polyimide and thermal release tape. Then, graphene was transferred onto the cleansed wafer, followed by repeating the whole procedure shown in Fig. 2A. Figure 5A exhibits the homogeneous alignment of the hexagonal MR sidewall facet orientation on both the virgin and recycled wafers, and the average size and number density of MR were almost the same with an acceptable deviation range. For example, the average values of number density, height, and diameter of MRs prepared on the recycled wafer were measured at (3.3 ± 0.4) × 104 cm−2, 16.9 ± 2.8 μm, and 16.7 ± 4.6 μm, respectively, whose values are almost the same as those from the virgin wafer. This result indicates that the refurbished wafer provides comparable epi-ready quality surface for the remote epitaxial growth of MR.

Fig. 5 Wafer recycle.

(A) Plan-view FE-SEM images of remote heteroepitaxial MR LED arrays grown on graphene-coated virgin (left) and recycled (right) wafers. The yellow arrows indicate that the MRs on both the virgin and recycled wafers have homogeneous in-plane alignment of hexagonal symmetry of MR sidewalls. (B) I-V characteristic curves and (C) EL spectra of MR LEDs fabricated by using virgin (blue line) and recycled (red line) wafers. The insets in (C) are photographs of EL light emission from the MR LEDs fabricated with virgin and recycled wafers in a dark room. (D) Photographs of blue EL emission of the LEDs attached on the surface of finger-sized bottle in the bent form at an Rb of 10 mm. The EL spectra and photographs were obtained at the same applied current of 100 mA and the EL measurement conditions. Photo credit: Junseok Jeong, Sejong University.

We further compared the electrical and EL performances of MR LED produced from virgin and recycled wafers. The I-V curves and EL spectra present that the electrical and light emission properties of LED obtained from the recycled wafer were as good as those from the virgin one (Fig. 5, B and C). Noticeably, the EL wavelength (or color) was reliably reproduced on the recycled wafer under the same MOVPE condition, as displayed in Fig. 5C. The LED produced from the recycled wafer was also deformable without the device performance degradation (Figs. 5D and 2C). Accordingly, the ability to reuse the wafers through the remote epitaxy allows substantial savings on the use of high-cost compound single-crystalline wafers.


In summary, we have demonstrated the remote heteroepitaxy of GaN-based MR LED arrays on Al2O3 through graphene. The electron microscopy analysis revealed vertical MR LED arrays with 30° rotational in-plane epitaxial relationship with respect to the underlying Al2O3 wafer. According to the DFT calculations, carbon atoms in graphene facilitate long-distance transport of the charge density from Al2O3 to the SLG surface, leading to remote heteroepitaxy across the noncovalent interface. The weakly bound, remote epitaxial interface allows transfer of the entire MR LED arrays onto a Cu plate using a thermal release tape technique with polymeric molding while leaving the original Al2O3 wafer available for reuse. The transferred MR LED showed good electrical and EL performances with excellent deformability. Stable, reliable electrical and EL properties were evidenced by repetitive bending cycles. The LEDs, fabricated from both the virgin and recycled wafers, reproducibly showed almost comparable LED performances.

As an application example, we have represented the LED panels that are deformable and tailored in various shapes. Regarding the prospective applications, the remote heteroepitaxy could be potentially adapted to fabricate heterogeneous optoelectronics/electronics integrated circuits (OEICs) or micro-LED displays, owing to the facile, high-yield device transferability and wafer reusability. For feasible practical device applications, the configurable (or addressable) mass transfer is necessary for heterogeneous ICs in a designed way. This prerequisite may be fulfilled by site-selective remote epitaxy for which precise control of size and position of semiconductor MRs is controlled at the epitaxy stage, which will then improve the process compatibility with the standard microelectronics manufacturing process. We believe that the versatility of remote epitaxy will readily drive the epitaxy technology from yielding the rigid, brittle electronics into manufacturing the next-generation flexible, wearable electronics.


Remote heteroepitaxy of GaN MR heterostructures

The remote heteroepitaxy of GaN MR heterostructures was performed on graphene-coated Al2O3 (0001) wafer using MOVPE. The SLG film was synthesized on Cu foil using the CVD method. To obtain the nearly aperture-free quality graphene interlayer, the SLG was doubly transferred onto the c-Al2O3 wafer using a poly(methyl methacrylate) (PMMA)–supported etching-transfer technique, which yielded misaligned BLG-coated c-Al2O3 substrate. It is noted that there was limitation in graphene thickness for the remote heteroepitaxy, and graphene thicker than triple layer did not yield vertically aligned GaN MRs, tentatively due to the attenuated charge transfer across thicker graphene. For the MOVPE growth of GaN MR heterostructures, trimethylgallium (TMGa), trimethylindium (TMIn), and ammonia (NH3) gas were introduced as precursor reactants for Ga, In, and N, respectively, with high-purity hydrogen (H2) or nitrogen (N2) carrier gas. The n-type GaN MRs, as a core stem in MR LED, were grown using the precursors of TMGa and NH3 with flow rates of 20 and 10 to 50 standard cubic centimeters per minute, respectively, at 1050°C. No initial buffer layer was introduced as nucleation enhancement process [i.e., flow-rate modulation epitaxy (51), low-temperature buffer layer, etc.] before the growth of core GaN MRs. The hydrogen-diluted silane (SiH4, 200 parts per million in H2) was used as an n-dopant precursor. Then, three-period InGaN/GaN MQW layers were heteroepitaxially deposited on the n-GaN MRs, and Mg-doped p-GaN was coated as the outermost shell layer on the circumference of MR heterostructures using bis(cyclopentadienyl) magnesium (Cp2Mg) precursor at 1000°C. The cyan and blue light from MR LED arrays were obtained by adjusting the growth temperature of InGaN QW to 770° and 800°C, respectively. The growth time of InGaN/GaN QW was 105/180 s. After the MOVPE growth, p-type activation was performed via rapid thermal annealing treatment at 900°C for 5 min in N2 ambient.

Device fabrication

The LED devices were fabricated by making ohmic contacts of Ni/Au and Ti/Au layers on both the surface of the outmost p-GaN layer and the delaminated bottom surface of n-GaN MRs, respectively. The gap between the GaN MR heterostructures was spin-coated with a polyimide insulating layer, to isolate two different metal electrodes. For metal contacts on p-GaN, oxygen plasma ashing was performed to expose the top of MRs. Then, Ni/Au (15/15 nm) metal layers with a size of 5 mm by 5 mm (or 10 mm by 10 mm) were deposited on the exposed p-GaN surface of MR tips using electron beam evaporation with the oblique angle deposition technique, followed by annealing at 300°C for 3 min. To deposit n-electrode, polyimide-molded MR arrays were delaminated from the native graphene/c-Al2O3 substrate via the thermal release tape–supported delamination method, and Ti/Au (30/30 nm) metal layers were deposited on the delaminated bottom surface of polyimide-molded MR arrays. The released polyimide film with MR LED arrays was adhered to the Cu foil using an Ag paste, and the tape was removed by heating on a hot plate at 190°C. The exfoliated mother substrate was then reused for producing LED devices again. For regeneration of wafer, the substrate was primarily cleansed with organic solvents of acetone, methanol, and isopropyl alcohol in an ultrasonic bath and then was thermally treated at 900°C for 1 hour in a furnace under normal ambient air conditions. Then, reactive ion etching was carried out under ambient oxygen conditions. These procedures completely remove the residues or contaminants originated from the use of polyimide and thermal release tape.


Surface morphologies of the samples were observed by FE-SEM (Hitachi S-4700). The existence of the graphene was confirmed by Raman spectroscopic analysis (excitation laser with a wavelength of 514 nm and a power of 20 mW; Renishaw 2000). Crystal structures of the samples were analyzed using HR-STEM (JEOL JEM-ARM 200F). For the HR-STEM observations, samples were cross-sectionally milled with a 30-kV–accelerated beam of gallium ions using a focused ion beam machine (FEI Helios NanoLab). The incidence electron beam was directed along the GaN[12¯10]║Al2O3[11¯00] to determine the remote heteroepitaxial relationship between GaN MR and Al2O3 wafer across the graphene layers. The compositional profiles of the indium contents in the top plane and sidewall InGaN/GaN MQWs were recorded from EDX spectroscopy in the scanning TEM mode to confirm the InGaN QW layers. The electrical and light emission properties were characterized by measuring the I-V characteristics and EL spectra using a source meter (Keithley 2400) and a monochromator with a charge-coupled device detector (Andor iDus DV420A), respectively.

Computational methods

The DFT calculations were performed within a generalized gradient approximation (GGA) for the exchange-correlation functionals (52, 53), implemented in the Vienna ab initio simulation package (VASP) (54). The kinetic cutoff energy was set to 400 eV, and the projector augmented wave potentials were used to represent the electron-ion interactions (55). For the vdW corrections, Grimme’s DFT-D3 method based on a semi-empirical GGA-type theory was used (56). For SLG/c-Al2O3, both the (4√3 × 4√3) supercell of SLG on the (2√3 × 2√3) supercell of Al2O3 (0001) with an in-plane alignment relationship of [1¯21¯0]SLG║[11¯00]sapphire and the (7 × 7) supercell of SLG on the (2√3 × 2√3) supercell of Al2O3 (0001) with an relationship of [11¯00]SLG║[11¯00]sapphire were set to be calculated with minimized lattice mismatches of 3.92 and 5.00%, respectively. The BLG/c-Al2O3 was simulated by doubly stacked SLG with various in-plane alignment relationships of [1¯21¯0]SLG║[[1¯21¯0]SLG (referred a G1 on G1 in fig. S16), [11¯00]SLG║[1¯21¯0]SLG (G2 on G1), [1¯21¯0]SLG║[11¯00]SLG (G1 on G2), and [11¯00]SLG║[11¯00]SLG (G2 on G2), all of which were placed in parallel with Al2O3 [11¯00]. For the Brillouin zone integration, a (3 × 3 × 1) grid for graphene/Al2O3 was adapted in the Gamma-centered scheme. The atomic configurations were fully optimized by minimizing the Hellmann-Feynman forces less than 0.02 eV Å−1. To clarify the role of graphene for charge transfer and/or remote heteroepitaxy, the charge density (ρ) was calculated by ρgraphene/sapphire − ρgraphene, where ρgraphene/sapphire and ρgraphene are the charge densities of graphene (SLG or BLG)/Al2O3 and graphene, respectively.


Supplementary material for this article is available at

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.


Acknowledgments: Funding: This research was financially supported by the Basic Science Research Programs (NRF-2016R1D1A1B03931518; 2017R1A2B2010123), the Priority Research Center Program (2010-0020207), and the Global Research and Development Center Program (2018K1A4A3A01064272) through the NRF of Korea. We acknowledge financial support of the KIAT through the International Cooperative R&D program (N0001819). This work was supported by Laboratory Directed Research and Development and CINT, a U.S. Department of Energy, Office of Basic Energy Sciences User Facility at Los Alamos National Laboratory (contract 89233218CNA000001) and Sandia National Laboratories (contract DE-NA-0003525). Author contributions: Y.J.H. conceived and directed the main experimental idea; S.H. conducted theoretical simulations; M.J.K. directed electron microscopic analyses; J.J. and D.K.J. performed the remote epitaxial growth and LED fabrications; Q.W. and S.K. carried out the TEM analyses; J.C. and S.H. simulated the DFT calculations; D.H.S. and S.W.L. performed the delamination and transfer procedures; B.K.K. and W.S.Y. fabricated graphene and graphene-coated wafers; J.H.J., Y.S.C., J.Y., and C.-H.L. characterized electrical and luminescent properties; A.Z. and J.K.K. designed fabrication of semitransparent electrode and the oblique-angle electrode deposition; M.J.K., S.H., and Y.J.H. are responsible for TEM analyses, DFT calculations, and LED growth–fabrication parts, respectively, and co-wrote the manuscript. All authors discussed and commented on the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors.

Stay Connected to Science Advances

Navigate This Article