Growing GaN LEDs on amorphous SiC buffer with variable C/Si compositions

The epitaxy of high-power gallium nitride (GaN) light-emitting diode (LED) on amorphous silicon carbide (a-SixC1−x) buffer is demonstrated. The a-SixC1−x buffers with different nonstoichiometric C/Si composition ratios are synthesized on SiO2/Si substrate by using a low-temperature plasma enhanced chemical vapor deposition. The GaN LEDs on different SixC1−x buffers exhibit different EL and C-V characteristics because of the extended strain induced interfacial defects. The EL power decays when increasing the Si content of SixC1−x buffer. The C-rich SixC1−x favors the GaN epitaxy and enables the strain relaxation to suppress the probability of Auger recombination. When the SixC1−x buffer changes from Si-rich to C-rich condition, the EL peak wavelengh shifts from 446 nm to 450 nm. Moreover, the uniform distribution contour of EL intensity spreads between the anode and the cathode because the traping density of the interfacial defect gradually reduces. In comparison with the GaN LED grown on Si-rich SixC1−x buffer, the device deposited on C-rich SixC1−x buffer shows a lower turn-on voltage, a higher output power, an external quantum efficiency, and an efficiency droop of 2.48 V, 106 mW, 42.3%, and 7%, respectively.


Structural and Compositional Analyses of the Nonstoichiometric Si x C 1−x Buffers. Figure 2(a)
shows the C/Si composition ratio of the nonstoichiometric Si x C 1−x films as a function of the [CH 4 ]/[CH 4 + SiH 4 ] fluence ratio by the X-ray photoelectron spectroscopy (XPS) analysis. When the [CH 4 ]/[CH 4 + SiH 4 ] fluence ratio is enlarged from 70% to 92%, the C/Si composition ratio of the nonstoichiometric Si x C 1−x increases from 0.51 to 1.83. It corresponds to a composition transferring from Si-rich to C-rich condition, as shown in Fig. 2(b). The decomposed energy is inevitably decreased to supply each molecule at constant RF plasma power and higher molecular density. The molecules obtain less energy to make the decomposition rates of SiH 4 and CH 4 molecules become gradually indistinguishable with each other. This phenomenon can contribute to the reduction of SiH 4 decomposition rates. Therefore, the nonstoichiometric Si x C 1−x film grown with high [CH 4 ]/[CH 4 + SiH 4 ] fluence ratios reveals the high C/Si composition ratio. Moreover, the composition of the nonstoichiometric Si x C 1−x can be detuned from Si-rich to C-rich condition by varying the [CH 4 ]/[CH 4 + SiH 4 ] fluence ratios.
With high-resolution field emission transmission electron microscopy (FETEM) analysis, neither Si nor C clusters can be found in all nonstochiometric Si x C 1−x samples. The structural phase of these Si x C 1−x samples are almost amorphous, as shown in Fig. 2(c). However, some blurred diffraction rings can be observed from the selected area diffraction (SAD) analysis of selected area in the stoichiometric and C-rich Si x C 1−x . The diffraction rings in the SAD pattern almost fade out for Si-rich Si x C 1−x when its structural phase structure is completely amorphous. By contrast, the SAD patterns for stoichiometric and C-rich Si x C 1−x samples exhibit at least two diffraction rings as a supporting evidence of nano-scale SiC clusters. With the detailed analysis on the diffraction ring, the two diffraction rings for stoichiometric SiC are 0.307 nm and 0.226 nm, which are contributed by (110)-oriented 3C-SiC (0.311 nm) and (200)-oriented 3C-SiC (0.227 nm) phases, respectively. It indicates that the stoichiometric SiC possesses a 3C-SiC like bonding structure. In contrast, the C-rich Si x C 1−x film reveals d-spacings of 0.373 and 0.222 nm, which are attributed to the (004)-oriented 6H-SiC and the (102)-oriented 4H-SiC, respectively. It also represents that the both 6H-and 4H-SiC based nano-scale phase structures may co-exist in the C-rich Si x C 1−x . Figure 2(d) shows the X-ray diffractometer (XRD) spectra of the Si x C 1−x films grown at different [CH 4 ]/[CH 4 + SiH 4 ] fluence ratios to observe their amorphous phase structure. In comparison with the glass background, there is no XRD signal to present the crystallinity for all nonstoichiometric Si x C 1−x films. This indicates that the phase structures of all nonstoichiometric Si x C 1−x samples are amorphous from macroscopic observation. However, the nano-scale grains may be still existed in the nonstoichiometric Si x C 1−x when the selective-area structural analysis observation is performed by using SAD and TEM.
Structural and Device Performance of the GaN LED grown upon Si x C 1−x buffer. The voltage-current (V-I) curves of the GaN LED deposited on the Si x C 1−x /SiO 2 /Si substrate by varying the C/Si composition ratios of the buffered Si x C 1−x film are shown in the lower part of Fig. 3(a). When the C/Si composition ratio of the Si x C 1−x buffer increases from 0.51 to 1.83, the turn-on voltage of GaN LED slightly decreases from 2.61 V to 2.48 V. It indicates that the C-rich Si x C 1−x buffered layer can effectively release the strain of GaN layers. The lattice constants of the 4H-and 6H-SiC (a= 0.3073 nm) move closer to that of the GaN (a= 0.3186 nm) to further release the strain between n-GaN and Si x C 1−x films. Therefore, the carriers suffer from less scattering process and easily transport in the n-GaN layer to reduce the turn-on voltage. Moreover, the strain relaxation can effectively decrease the resistance from 3.388 Ω to 2.694 Ω when the buffered Si x C 1−x film is changed from Si-rich to C-rich condition. The upper part of Fig. 3(a) reveals the power-current (P-I) curve of the GaN LEDs grown on Si x C 1−x buffers with different C/Si composition ratios. The maximal EL power of the GaN LED increases from 93 mW to 106 mW with the corresponding P-I slope increasing from 0.84 mW/A to 1.02 mW/A when the Si-rich Si x C 1−x film replaces to the C-rich Si x C 1−x . That is also attributed to the contribution of the releasing strain between the n-GaN/C-rich Si x C 1−x interface. The structure defect related scattering or trapping probability decreases accordingly. Because of the defect reduction in n-GaN layer, more carriers can be effectively injected into LEDs to improve the tunneling probability under the same bias current. It enables the increasing internal quantum efficiency of the GaN LED grown on the Si x C 1−x /SiO 2 /Si substrate.
To characterize the interfacial defect distribution, the capacitance-voltage (C-V) characteristics of GaN LED are measured and shown in Fig. 3(b). The almost unchanged hysteresis during the detuned sweeping of the biased voltage corroborates that nearly zero C-V hysteresis is contributed by defects and doping impurity in depletion region of the GaN LED grown on all kinds of the Si x C 1−x buffers. The charge storage of LED can be determined by integrating the C-V hysteresis area over the voltage sweeping range with the following formula 26 , where Δ N carrier denotes the density of stored carriers in defects of GaN LED with an active area of A. The number of carriers stored per defects is defined as the ratio of the integrated area charge density to the estimated area density of defect. As a result, the area defect density is decreased from 2.87 × 10 10 cm −2 to 1.093 × 10 10 cm −2 when the buffered Si x C 1−x film is changed from Si-rich to C-rich condition in the inset of Fig. 3(b) because the released strain can effectively decrease the defect generation in the GaN LED grown on the C-rich Si x C 1−x buffer. In general, the defects generated by the threading and edge dislocations co-exist in the GaN film. These two dislocations can be observed from the TEM analysis. In our case, the defects in the GaN LED grown on the a-Si x C 1−x buffer are mainly contributed by the edge dislocations, which can be confirmed by the high-resolution TEM images.
In addition, such high density of defects obtained from C-V analysis is mainly attributed to the quality of GaN MQW LED epitaxial upon the a-Si x C 1−x buffer. In comparison with the crystalline SiC buffer, there is still observable lattice mismatch as well as strained dislocation occurred during the epitaxy of n-GaN film at very beginning, which inevitably contributes to structural defect generation within grain boundaries so as to scatter or trap the carriers before their radiative recombination. Consequently, the external quantum efficiency (EQE) of GaN LEDs can be estimated by using the following formula to corroborate with the numerical results given by Z-parameter analysis, where λ denotes the EL peak wavelength of the GaN LEDs. Figure 3(c) reveals the EQE of the GaN LED grown upon the Si x C 1−x buffers with different C/Si composition ratios. The maximal EQE of a GaN LED enhances from 38.8% to 42.1% by transferring the Si x C 1−x buffer from Si-rich to C-rich condition because the strain relaxation between n-GaN and Si x C 1−x buffer can be effectively improved by detuning the C-rich Si x C 1−x recipe. Apparently, the decreasing strain at n-GaN/Si x C 1−x interface also varies the droop of EQE performance. The EQE droop occurs as the recombination mechanism in the GaN LED grown upon Si-rich Si x C 1−x starts to transfer from radiative to Auger recombination by increasing bias current beyond 10 mA. In particular, the droop of EQE significantly reduces from 15% to 7% when the C/Si composition ratio of Si x C 1−x buffer increases from 0.51 to 1.83. It indicates that the Auger recombination probability is distinctly suppressed to promote the better radiative recombination mechanism with larger EQE at higher biased condition. In principle, the EQE droop of GaN LED is also originated from the polarization field, the poor hole transport, and the current spreading. The hole injection in GaN LED is usually hindered as compared to the electron injection because the active region in device is intrinsic or undoped, which increases the poor hole transport and stands alone the material polarization and sheet charge at interfaces 27 . In our case, the GaN LEDs grown on a-Si x C 1−x buffers with different C/Si composition reveal the same behavior, which exhibit the similar EQE droop induced by the poor hole transportation. This rules out the possibility of hole transportation which dominates the droop variation. In addition, the polarization field in GaN LED is dependent on the growth of GaN film 28 . The stronger polarization field in c-plane GaN LED growth may contribute to the larger EQE droop. In our case, the polarization field in all GaN LEDs remain almost the same as these their n-GaN layers are all grown on a-Si x C 1−x buffers. Hence the polarization field variation induced EQE droop maybe trivial in the GaN LEDs epitaxial on different a-Si x C 1−x buffers. On the other hand, the surface roughness induced non-uniform current spreading is also one factor to affect on the low carrier concentration for varying the EQE droop of the GaN LED 29 . From the analysis on EL intensity distribution which represents the uniformity of current spreading, the non-uniformly distributed region of EL intensity shrinks when transferring the buffered Si x C 1−x film from Si-rich to C-rich condition, where the density of interfacial defect traps gradually reduce to improve the EQE droop. In addition to the Auger recombination, the current spreading uniformity may considerably be another mechanism for the EQE droop variation. Therefore, the lower EQE droop for the GaN LED grown a-SiC buffer is mainly contributed by both the weaker Auger recombination and the better current spreading in devices.
To confirm, the dominated recombination mechanism in a GaN LED can be determined by using the Z-parameter analysis 30 . In principle, the net recombination of the LED in steady state can be described as a function of the injected current where I and V a denote the injected current and active volume, respectively. In general, the Z-parameter varies between 1 and 3 for different recombination mechanisms. Typically, the Z parameters are 1, 2, and 3 for the defect, radiative, and Auger related recombinations, respectively. In experiment, the Z-parameter can be calculated by plotting the P-I curve of a LED as the derivative of the ln(I) vs. ln(P 1/2 ). Goddard  where I defect , I rad , and I Auger are the defect, radiative, and Auger injected currents, respectively. The EL power is proportional to the current (I rad ) contributed to the radiative recombination, as described by When one of the recombination mechanisms dominates the injected current, the injected current can be approximated by When the Z-parameter plot can be experimentally obtained from the P-I response of the GaN LED, the Fig. 4(a) shows the Z-parameter curves as a function of biased currents for the GaN LEDs grown upon the Si x C 1−x with different C/Si composition ratios. At low bias currents, the dominated recombination mechanism in the GaN LED grown on all kinds of Si x C 1−x /SiO 2 /Si substrates is mixed of the defect and radiative recombinations with corresponding Z-parameter ranged between 1 and 2. In all cases, the Z-parameter gradually approaches 2 with increasing biased currents to 1 < I inj < 50 mA (with ± 10% deviation). It indicates that the radiative recombination dominates the GaN LED. At the bias current beyond 50 mA, the Auger recombination significantly enhances in the GaN LED grown with Si-rich Si x C 1−x buffer. Nevertheless, the Auger recombination can still be suppressed in the GaN LED grown upon buffered C-rich Si x C 1−x film even under the bias current of up to 100 mA. Such the suppression is because of the released strain when it induces the piezoelectric and spontaneous polarization to generate large local spatial field 31 . The field varies the momentums of electrons and holes and contributes to a large spatial separation between the wave functions of carriers, which further decreases the radiative recombination rate and internal quantum efficiency. If the strain can effectively be released in the devices, the generating probability of the Auger recombination can be reduced to further decrease the droop of quantum efficiency. In general, when the Z-parameter is 2, the radiative recombination dominates all recombination mechanisms in the GaN LEDs. However, Goddard's group also pointed out that the Z-ptarameter is equivalent to 2 when the co-existed defect and Auger recombinations dominate in GaN LEDs concurrently. In addition, the leakage current and the high level injection are non-ideal effects to make Z-parameter result complicated. If defect and Auger recombinations dominate the Z-parameter at Z = 2 condition, the efficiency of the radiative recombination should be decayed to an extremely low value as both mechanisms are non-rediative processes 32 . Owing to the decreased external quantum efficiency by these two non-radiative phenomena, the luminescent power would be greatly attenuated and associated with a relatively large efficiency droop. This is definitely not in our case, as the measured efficiency droop of the GaN LED on a-Si x C 1−x buffer is relatively comparable with others reported in previous works in Table 1 2 38 .
To evaluate the luminescent characteristics of the GaN LED grown on different Si x C 1−x buffers, the Fig. 4(b) shows the EL spectra of the GaN LED grown upon the Si x C 1−x films with changing the C/Si composition ratios. Under the bias current of 10 mA, the EL peak wavelengths of the GaN LEDs grown with the different C/Si composition ratios are all located at 450 nm. Because of the band filling effect enhanced by enlarging the bias current 39 , the EL linewidth is slightly enlarged from 19 nm to 26 nm when the Si x C 1−x buffer is transferred from C-rich to Si-rich condition (which requires higher bias). This mechanism elucidates why the GaN LED grown upon the  Si-rich or nearly stoichiometric Si x C 1−x film easily contributes to a broader spectral linewidth due to higher biased condition. The EL power of the GaN LED decays when the Si content in the Si x C 1−x buffer increases. The EL spectra of the GaN LED also red-shifts its central wavelength from 446 nm to 450 nm when the stoichiometry of the Si x C 1−x buffer is changed from Si-rich to C-rich condition, as shown in Fig. 4(c). In general, the blue-shifted wavelength is correlated with both the electric field in InGaN/GaN quantum well and the associated quantum confined Start effect (QCSE) under high biases 40 . In previous work, Xu et al. compared the InGaN/InGaN LED with the InGaN/GaN LED and observed the mitigated blue-shift on peak wavelength when the electric field in InGaN/InGaN quantum well reduces 40 . When the Si-rich Si x C 1−x buffer layer induces higher stain to result in more defects in the InGaN/GaN quantum well, the GaN LED grown upon the Si-rich Si x C 1−x buffer needs higher bias to overcome the defect trapping for achieving same EL power, which inevitably results in higher internal field in the InGaN/GaN quantum well so as to induce a larger blue-shift on peak wavelength. In principle, the strain relaxation suppresses the piezoelectric components with different polarizations to further decrease the QCSE 41 . As a result, when the Si x C 1−x buffer is transferred from Si-rich to C-rich condition, the spectral blue-shift almost diminishes as strain and structural defects decrease concurrently.
To characterize the distribution uniformity of the mapped EL intensity as well as bias current 42 , the 2D EL intensity contours of the GaN LED grown upon the Si x C 1−x buffers with different C/Si composition ratios were measured under the same biased current. It helps to observe if the injected carrier transportation is uniformly distributed between contacts, as shown in Fig. 4(d). When Si-rich Si x C 1−x film is used, the color of non-uniformly distributed EL intensity contour on the GaN LED surface widely spreads from orange to red, as shown in the upper part of Fig. 4(d). It represents that the carriers are unable to effectively transport along the region ranged between anode and cathode. This phenomenon is mainly attributed to the defect trapping effect as varied by the C-V analysis of the GaN LED. When the defects trap carriers in the lower EL region, the carriers require higher voltage to overcome the potential well around traps. This is because the biased current may contribute to the non-radiative and Auger recombinations. In particular, the Auger recombination will dominate the EL intensity under high biased current. The variation among the EL intensity distributions for all GaN LEDs grown upon different Si x C 1−x buffers is less distinguishable when the radiative recombination dominates under lower biased currents, as shown in upper part of Fig. 4(d). In contrast, the Auger mechanism arises to degrade the recombination process under highly biased operation. It effectively decreases the EQE to decay the EL intensity. From the lower part of Fig. 4(d), when GaN LEDs grow upon the buffer Si x C 1−x under the biased current as high as 100 mA, the Auger recombination occurs to differentiate the variation on EL intensity distribution. With Si-rich Si x C 1−x buffer, the non-uniform distribution of the EL intensity is observed with its contour color broadly spreading from blue to red. Especially, the strongest EL intensity is close to the anode region because the strain induced Auger recombination greatly suppresses the injected carrier transportation within a very short distance. In comparison, the C-rich Si x C 1−x buffer essentially releases the interfacial strain, decreases the Auger recombination, and improves the radiative recombination. Such a non-uniform distribution of the EL intensity shrinks when the Si x C 1−x buffer is transferred from Si-rich to C-rich condition, where the trapping density of the interfacial defect gradually reduces to improve both the intensity and uniformity of the EL distribution. In the lower part of Fig. 4(d), the EL intensity distribution can smoothly spread from anode to cathode by improving carrier transportation with suppressed Auger recombination. When the strain is released by growing the GaN LED upon the lattice matched buffer, the EL becomes more efficient to provide higher power with smaller droop, as confirmed from the Z-parameter analysis. Figure 5 shows the TEM images of the a-Si x C 1−x buffers on SiO 2 /Si substrate (first row), the n-GaN layers on different a-Si x C 1−x buffers (second row), the InGaN MQW structures on n-GaN/a-Si x C 1−x /SiO 2 /Si (third row) and the high-resolution interfacial images between n-GaN layers and a-Si x C 1−x buffers (fourth row). From the first row of Fig. 5, it is seen that all a-Si x C 1−x buffers on SiO 2 exhibit amorphous structure; however, the Si-rich a-Si x C 1−x further non-uniformly stacked along the surface normal, indicating its randomly distributed grains at different regions, as shown Fig. 5(a). In contrast, both the stoichiometric SiC (see Fig. 5(b)) and C-rich Si x C 1−x (see Fig. 5(c)) buffers present relatively fine nano-grains in the whole layer to facilitate the epitaxy of n-GaN. In second row of Fig. 5, the TEM images reveal the structural perfection of n-GaN layers grown upon a-Si x C 1−x buffers. Figure 5(d) shows that the n-GaN layer exhibits significant grain domains with abundant boundary imperfections owing to the misfit lattice induced dislocation loops. On the contrary, the stoichiometric SiC and C-rich Si x C 1−x buffers provide more small nano-grains to help the epitaxy of n-GaN with better crystalline quality. Therefore, few structural defects in the n-GaN layer grown on the stoichiometric SiC (see Fig. 5(e)) and C-rich Si x C 1−x (see Fig. 5(f)) buffers are observed after synthesis. In the third row of Fig. 5, the TEM images of the same p-GaN/InGaN-MQW/n-GaN layers epitaxial on Si-rich, stoichiometric and C-rich SiC buffers are shown. When Si-rich Si x C 1−x film serves as a buffer, the epitaxial GaN LED shows amorphous structure initially, which inevitably contributes to the structural defect generation such that the performance of the GaN MQW LED is somewhat degraded accordingly, as shown in Fig. 5(g). In comparison, both the epitaxial n-GaN MQW LEDs grown on stoichiometric SiC (see Fig. 5(h)) and C-rich Si x C 1−x (see Fig. 5(i)) buffers exhibits clearer MQW layer structure with fewer structural defects but some inhomogeneous shadows can still be observed among layers. In more detail, as the a-Si x C 1−x buffer changes to C-rich condition, the lowest row in Fig. 5(j) shows that the epitaxial n-GaN film upon Si-rich a-Si x C 1−x remains disordered lattice at beginning, and the distribution uniformity on composition of the n-GaN is the worst among all cases. With the use of stoichiometric SiC buffer, the crystallinity of n-GaN near interface slightly improves so as to effectively reduce the generation of structual defects, as shown in Fig. 5(k). The relatively ordered lattice structure is observed in n-GaN layer shown in Fig. 5(l) by using C-rich a-Si x C 1−x buffer, which leads to the lowest strain as well as structural defect density so that the dislocation can be minimized in the GaN MQW LED grown upon the n-GaN layer. This is the evidence attributed to the improved V turn-on , EL, EQE, and droop property of the GaN LED on C-rich Si x C 1−x buffer with the best strain relaxation.
At last, to further characterize if the performances of the GaN LED is degraded after lift-off process, the I-V and P-I curves of the GaN LED on a-Si x C 1−x buffer lifted-off on the copper substrate are measured and shown in Fig. 6(a). The lifted-off GaN LED on a-Si x C 1−x buffer exhibits a turn-on voltage of 2.8 V, an EL power of 35 mW at bias of 100 mA, and the corresponding P-I slope of 0.35 W/A after transferring onto the copper substrate. The Fig. 6(b) shows the EL spectrum obtained at a bias of 10 mA, which reveals the peak wavelength at 446 nm. In particular, the long-term lift-off process in BOE solution could slightly damage the GaN LED to induce more surface states and defects. As compared to the original GaN LED, the lifted-off GaN LED requires higher turn-on voltage but delivers lower optical power. Such an operation also leads to a higher internal electric filed across the MQW structure to induce the stronger quantum confined Stark effect, which effectively contributes to a slightly blue-shifted peak wavelength of EL for the lifted-off GaN LED.

Discussion
The epitaxy of GaN LED upon the a-Si x C 1−x buffer grown by low-temperature PECVD on the SiO 2 /Si substrate is demonstrated. When the [CH 4 ]/[CH 4 + SiH 4 ] fluence ratio is increased from 70% to 92%, the Si x C 1−x film increases its C/Si composition ratio from 0.51 to 1.83 with the corresponding fractional index (x) reducing from 0.66 to 0.35. The XRD analyzed that the phase structures of all nonstoichiometric Si x C 1−x samples are amorphous; however, the SAD analysis reveals some blurred diffraction rings from the stoichiometric and C-rich Si x C 1−x . Two SAD rings diameters observed for stoichiometric SiC are 0.307 nm and 0.226 nm, as contributed by (110)-and (200)-oriented 3C-SiC structures, respectively. In contrast, the C-rich Si x C 1−x film reveals d-spacings of 0.373 and 0.222 nm as attributed to the (004)-oriented 6H-SiC and the (102)-oriented 4H-SiC, respectively. When the Si x C 1−x buffer changes from Si-rich to C-rich condition, the turn-on voltage of GaN LED slightly decreases from 2.61 V to 2.48 V. It indicates that the C-rich Si x C 1−x buffer can effectively release the strain of GaN epitaxial layers when the lattice constants of the co-existed 4H-and 6H-SiC in C-rich Si x C 1−x move closer to that of the GaN. Moreover, the C-V analysis declares that the area density of defects is decreased from 2.87 × 10 10 cm −2 to 1.09 × 10 10 cm −2 when Si-rich SiC buffer replaces to C-rich one when the released strain effectively reduces the defects in the GaN LED grown on lattice matched C-rich Si x C 1−x buffer. The maximal EL power of the GaN LED increases from 93 mW to 106 mW, and the corresponding P-I slope also increases from 0.84 mW/A to 1.02 mW/A when the Si x C 1−x buffer is transferred from Si-rich to C-rich condition. The n-GaN/Si x C 1−x interface defect related scattering or trapping probability decreases accordingly. The Z-parameter analysis shows that the Auger recombination significantly decreases with released stain in GaN LED. Therefore, the maximal EQE of a GaN LED enhances from 38.8% to 42.1%, accompanied with its droop suppressing from 15% to 7% and its EL spectra red-shifted from 446 nm to 450 nm by transferring Si x C 1−x buffer from Si-rich to C-rich content. When the Si x C 1−x buffer transforms from Si-rich to C-rich condition, the spectral blue-shift becomes insignificant as the structural defects decrease. When Si-rich Si x C 1−x film is used, the non-uniform distribution of the EL intensity contour on the GaN LED surface widely spreads from orange to red. It represents that the carriers are unable to effectively transport along the region ranged between anode and cathode. This non-uniform distribution of EL intensity shrinks when the Si x C 1−x buffer is transferred from Si-rich to C-rich condition, where the trapping density of the interfacial defect gradually reduces to improve both the intensity and uniformity of the EL distribution. This GaN LED grown upon SiC/SiO 2 /Si substrate can be easily lifted-off for applying to the flexible and transparent LED applications.

Methods
Fabrication and Measurement of GaN LED on Si x C 1−x /SiO 2 /Si substrate. For the buffered Si x C 1−x deposition, the 500-nm thick Si x C 1−x film with different C/Si composition ratios was grown upon the 1-μ m thick thermal SiO 2 coated Si substrate by using a hydrogen-free low-temperature PECVD system. This synthesis was employed with a mixed gaseous recipe of Argon-diluted silane (Ar-diluted SiH 4 ) and methane (CH 4 ) without hydrogen carriers. The Ar-diluted SiH 4 fluence was controlled at 150 sccm. The fluence ratio defined as R SiC = [CH 4 ]/([CH 4 ]+ [SiH 4 ]) was detuned as 70%, 90%, and 92% to fabricate the Si-rich, the stoichiometric, and the C-rich buffer Si x C 1−x , respectively. The working pressure and the RF plasma power were set at 0.3 torr and 100 W, respectively. In addition, the substrate temperature was fixed at 500 °C, which is below the criterion for synthesizing crystalline SiC during deposition. The bonding information and composition ratio of the Si x C 1−x buffer were analyzed by using the XPS with an Mg K α -line radiation at 1256.3 eV. The structural phase of the nonstoichiometric Si x C 1−x film was characterized by using an XRD (X'Pert PRO, PANalytical). The preferred orientations of nano-grains in nonstoichiometric Si x C 1−x were determined by using the high-resolution FETEM (JEM-2100, JOEL).
To fabricate the GaN LED on Si x C 1−x /SiO 2 /Si substrate as shown in Fig. 7(a), the GaN LED structure was grown on the Si x C 1−x /SiO 2 /Si substrate by using a MOCVD system. The epitaxial procedure was entrusted to a local GaN LED manufacturer. The GaN LED structure consists of a 2-μ m thick n-type GaN bottom layer at first, a 0.2-μ m-thick InGaN/GaN multiple quantum well active layer (10 periods with 4-nm thick i-In 0.18 Ga 0.82 N and 16-nm thick i-GaN), and then a 0.3-μ m-thick p-type GaN top layer. The transparent Ni/Au (5 nm/5 nm) contact layer was formed by using e-beam evaporation to serve as the ohmic contact on the p-type GaN layer. Subsequently, a 200-nm thick Au contact was patterned on top of the semi-transparent layer to serve as the p-electrode. On the other hand, the Ti/Al/Ni/Au with thickness of 10 nm/200 nm/30 nm/100 nm was successfully deposited onto the exposed n-type GaN layer to serve as the n-type electrode. The cross-sectional SEM analysis of the GaN LED grown upon Si x C 1−x buffer was performed to scale the thickness and to characterize the uniformity of device structures, respectively. From the SEM image, the thicknesses of the GaN LED, the Si x C 1−x film, and the SiO 2 layers are determined as 2.5 μ m, 0.5 μ m, and 1 μ m, respectively, as shown in Fig. 7(b). This SEM image also reveals very clear n-and p-typed LED layered structures with excellent uniformity. It indicates that the GaN LED can successfully deposit on the amorphous Si x C 1−x covered SiO 2 /Si substrate. Such uniform epitaxy relies strictly on the formation of SiO 2 layer via the wet-oxidation of the Si substrate, whereas any other kinds of SiO 2 films formed by gas-or solid-phase deposition cannot meet the demand of broad-area uniformity. Figure 7(c) shows the microscopic photograph of the GaN LED grown on the different Si x C 1−x buffers with changing C/Si composition ratios. The interdigitated electrode structure was designed to enhance the carrier transportation, in which the anode has three fingers and the cathode has two fingers so as to increase the transportation paths. In addition, both the anode and cathode electrodes were allocated on the top surface to avoid the inconvenience caused from the LED package. In order to transfer the GaN LED onto other substrates, the GaN LED on a-Si x C 1−x buffer needs to be lifted off from a SiO 2 /Si wafer, as shown in Fig. 8. The entire GaN LED/a-Si x C 1−x /SiO 2 /Si sample was immersed in a buffer oxide etching (BOE) solution to remove the SiO 2 buffer. Afterwards, the GaN LED on a-Si x C 1−x buffer was clearly separated from the SiO 2 /Si wafers and floated upon the aqueous BOE solution. By gradually diluting the aqueous BOE solution with pure deionized water, other substrates can be used to hold the floated GaN LED on a-Si x C 1−x buffer subsequently. To distinguish the carrier  recombination mechanisms from one another, the I-V curves of GaN LEDs grown upon the buffered Si x C 1−x with different C/Si composition ratios were measured by programmable electrometer (2400, Keithley). To realize the interfacial defect feature, the capacitance characteristic of the GaN LEDs was measured by a capacitance meter (4280A, HP) with the operating frequency of AC signal at 1 MHz. The EL spectrum ranging from 300 to 850 nm was measured at a biased current of 10 mA, and the EL power was measured using a calibrated power meter with an integral sphere containing a large-area Si detector.