The primary concern in this roadmap of vertical injection DUV-LEDs is the tensile strain-induced cracks for Al-rich AlGaN on GaN templates, which severely disrupt the device fabrication. Taking 280 nm DUV-LEDs as an example, a typical Al composition range of 50–60% is adopted for n-AlGaN27–30, suggesting ~ 1.2% in-plane lattice mismatch between AlGaN and GaN with a theoretical critical thickness of 30 nm for cracking31. Although the low-temperature Al(Ga)N interlayer allows the growth of crack-free AlGaN32,33, it is seriously worried that the low-temperature-induced rough surface would deteriorate the quality of the subsequent DUV-LED structures, and hence be detrimental to the device performance21,22. In order to juggle the strain and crystalline quality of DUV-LEDs, a decoupling strategy is proposed here, featuring the pre-crack and filling processes as shown in Fig. 1a (In-situ monitoring curve during MOCVD growth shown in Supplementary Fig. S1). Specifically, controlled pre-cracks are intentionally introduced through an AlGaN sacrificial layer on GaN templates, whose Al composition is as high as 80% to shallow down the cracks within a depth of ~ 100 nm (details shown in Fig. 2), laying a solid foundation for the following filling process (Supplementary Fig. S2). Subsequently, an AlGaN healing layer is employed to fill up the cracks, and thus recover the surface morphology. Meanwhile, as part of the optical propagation path, Al composition in the healing layer is reconciled to the emission wavelength, 65% here for 280 nm DUV-LEDs.
Surface morphology evolution of the decoupling structure, from pre-crack to filling, is characterized by atomic force microscopy (AFM) as shown in Figs. 1b and c. For the Al0.8Ga0.2N sacrificial layer (Fig. 1b), dense cracks are observed with a mean spacing of about 3 µm, which extend along the \({⟨\text{11}\stackrel{\text{-}}{\text{2}}\text{0}⟩}_{\text{AlGaN}}\) directions according to previous reports34–36, consistent with the lower surface energy of \(\left\{\text{1}\stackrel{\text{-}}{\text{1}}\text{00}\right\}\) cleavage planes against the \(\left\{\text{11}\stackrel{\text{-}}{\text{2}}\text{0}\right\}\) ones. While after the growth of the Al0.65Ga0.35N healing layer (Fig. 1c), no cracks can be identified any more, and the surface is recovered to the typical step-terrace morphology with a root-mean-square (RMS) roughness of 0.51 nm (10×10 µm2). Worthy of note is that the healing layer is nearly strain-free as demonstrated by X-ray reciprocal space mapping (RSM) for the \(\left(\stackrel{\text{-}}{\text{1}}\text{015}\right)\)-plane reflection (Fig. 1d), mainly attributed to the plastic relaxation by pre-cracks. It is then convinced that the DUV-LED epitaxial structure in the present roadmap is almost fully decoupled from the underlying GaN template (Supplementary Fig. S3), by which visually crack-free DUV-LED wafers can be obtained, typically as the 4-inch one shown in Fig. 1e. Further optical measurements (Fig. 1f) demonstrate that no surface cracks are identified except the edge exclusion region (EE, 3 mm), successfully driving the fabrication of DUV-LEDs into high-production 4-inch era.
Moreover, cross-sectional scanning transmission electron microscopy (STEM) is employed to reveal the healing of pre-cracks. Benefiting from the parallel direction between the incident-electron and crack extension (both along \({⟨\text{11}\stackrel{\text{-}}{\text{2}}\text{0}⟩}_{\text{AlGaN}}\)), a “buried” V-shaped crack with clear outlines can be identified in Fig. 2a, where the actual critical thickness of Al0.8Ga0.2N on GaN is determined to be less than 80 nm (Supplementary Fig. S4). The pre-crack is then filled along with growth of the Al0.65Ga0.35N healing layer, and eventually a flat and sharp n-AlGaN/Al0.65Ga0.35N interface is presented, consistent with the crack-free surface morphology in Fig. 1c. It is worth mentioning that two indicator lines are “buried” into the healing layer through the in-situ desorption-tailoring approach30, as enlarged in Figs. 2b and c, respectively. Specifically, the Al0.65Ga0.35N growth is intentionally suspended every 180 nm by stopping the precursor (TMAl and TMGa) supply, and then the desorption difference between Al and Ga atoms on the surface as well as on the inclined sidewalls of cracks (if exist) leads to higher Al composition with shallow contrast in the bright-field STEM image, directly indicating the filling degree of pre-cracks. At the filling thickness of 180 nm, there is an obvious bending (white arrows) in Indicator I (Fig. 2c), which corresponds to the crack sidewalls and hence demonstrates the existence of cracks. While increasing the thickness to 360 nm, Indicator II becomes straight and coherent, suggesting that the crack has almost been filled up.
As characterized by the indicator lines (STEM) and AFM, the thickness needed to fill the pre-cracks is found to be directly related to the Al composition of the healing layer. The higher the Al composition, the greater the thickness is required as shown in Fig. 2c, suggesting that Ga atoms play a key role in the filling process. Energy-dispersive X-ray spectroscopy (EDS) mapping is then adopted to reveal the atomic behavior of Al and Ga (Figs. 2e and f, respectively). Evidently, higher Ga composition inside the pre-crack is observed in comparison with that in the surrounding Al0.65Ga0.35N healing layer, and the composition difference gradually decreases until the pre-crack is filled up at the thickness of 360 nm (Indicator II). In general, the healing of cracks is attributed to the atomic migration along the inclined sidewalls37, it is hence convinced that there are more Ga atoms inside the filled pre-cracks since their diffusion length is much larger than Al ones. Further EDS line scanning across the pre-crack (Fig. 2g) demonstrates that the peak Ga composition inside the crack reaches 45%, transparent to the 280 nm emission light from the upper DUV-LED structure.
In addition to the tensile strain, potential quality degradation of DUV-LED structure is another essential issue in the novel roadmap, in particular the possible massive generation of threading dislocations (TDs) during the healing of pre-cracks. Cross-sectional TEM measurement is hence carried out as shown in Fig. 3a, where the TD density in the Al0.65Ga0.35N healing layer is roughly equal to that in the GaN template (details in Supplementary Fig. S5). Special attention should be paid to the pre-crack marked by the arrow (enlarged STEM image in Fig. 3b), where TDs nucleate at the sidewalls and extend up into the DUV-LED structure. Herein, a screw-type TD (labelled S) and an edge-type (labelled E) ones are identified according to the TEM measurements under two-beam conditions with \(\text{g = }\left[\text{0002}\right]\)and \(\text{g =}\text{ }\left[\text{11}\stackrel{\text{-}}{\text{2}}\text{0}\right]\text{ }\) (Fig. 3c). In consideration of the quite small surface coverage of the pre-cracks (Fig. 1b), it is convinced that the filling-induced TDs have less effect on the total TD density, as further verified by the X-ray rocking curves (XRCs) in Fig. 3d. XRCs of the \(\left(\text{0002}\right)\)- and \(\left(\text{1}\stackrel{\text{-}}{\text{1}}\text{02}\right)\)-planes are measured here for the GaN template and Al0.65Ga0.35N healing layer, respectively, while only slight broadening is observed for Al0.65Ga0.35N. The full width at half maximum (FWHM) values are then extracted, and the corresponding TD density is calculated to be 1.35×109 cm−2 in Al0.65Ga0.35N (294/387 arcsec)38, laying a solid foundation for the subsequent DUV-LED structure. Furthermore, the TD density in the active region of DUV-LED is estimated as 1.18×109 cm−2 in the plan-view STEM image (Fig. 3e), being approximate to that in the Al0.65Ga0.35N healing layer. As a feature, some of TDs are distributed along the straight dashed lines in Fig. 3e, suggesting that they originate from the filling process of pre-cracks.
The TD density in the active region would directly determine the radiative recombination efficiency39, which is a key factor in assessing device performance, and can be evaluated via the photoluminescence (PL) measurements. Figure 3f shows the high-angle annular dark field (HAADF) STEM image for the multiple quantum wells (MQWs) active region assembled by 1.8 nm-thick Al0.37Ga0.63N wells and 8 nm-thick Al0.5Ga0.5N barriers. The temperature-dependent PL is then performed from 10 K to 300 K (Fig. 3g), where the emission peak red-shifts with rising temperature and reaches 280 nm at room temperature. Assuming the non-radiative recombination centers frozen at 10 K, the MQWs exhibit a room-temperature internal quantum efficiency (IQE) of 70.9%, at the same level with those on AlN templates39–42. In addition, the dependence of IQE on excitation power is investigated and shown in Fig. 3h. It is found that the IQE monotonically increases with excitation power from 37.2% (5 mW) to 70.9% (37 mW), suggesting that the dominant recombination process gradually changes from the non-radiative recombination to the radiative one according to the Shockley-Read-Hall (SRH) model43,44. Since the IQE value doesn’t saturate here, even greater IQE can be expected under higher excitation power in PL measurements, or under higher injection current in DUV-LEDs44.
With the above two issues being solved, wafer-scale vertical injection DUV-LEDs with a wavelength of 280 nm are eventually fabricated as shown in Fig. 4. After preparation of the p-electrode, the epitaxial structure is crack-freely transferred from the sapphire substrate to a Si submount by means of wafer bonding and subsequent LLO (4-inch wafer in Fig. 4a, 2-inch one in Supplementary Fig. S6). It is worth mentioning here that there are two benefits by adopting the GaN templates instead of AlN: (i) the frequency-tripled Nd:YAG laser (355 nm) meets the requirement of removing the sapphire in this roadmap, which is low-cost and widely used; (ii) the absence of Al metal during LLO effectively avoids the occurrence of fracturing/cracking, ensuring the high yield of dies. Moreover, the decoupling strategy is supposed to provide a protection cushion against the thermal shock induced by laser irradiation, equally preventing the fracturing during LLO. Following the removal of sapphire, the GaN template must be sufficiently thinned by chlorine-based inductively coupled plasma (ICP) etching till the Al0.8Ga0.2N sacrificial layer is exposed, considering that GaN strongly absorbs the DUV emission light from the active region (Fig. 4b, details in Supplementary Fig. S8). Subsequently, KOH roughening is carried out to obtain the surface morphology of random hexagonal pyramids texture as shown in Fig. 4c16,18,20,25, which is expected to improve the light extraction efficiency by favorable scattering geometries45. The n-electrode is deposited as the final process, after windowing to the n-Al0.55Ga0.45N layer (Fig. 4d). It should be mentioned that it is still quite difficult to obtain Ohmic contact on the etched \(\left[\text{000}\stackrel{\text{-}}{\text{1}}\right]\)-plane of n-Al0.55Ga0.45N, leading to increased operating voltage as a consequence (Supplementary Fig. S9).
Wafer-scale fabrication of vertical injection DUV-LEDs is herein realized in both 2- and 4-inch wafers, and those with larger sizes are expectable as well, largely thanks to the decoupling structure. The 280 nm DUV-LED die with a size of 508×508 µm2 presents a light output power of 38.4 and 65.2 mW at 100 and 200 mA (Fig. 4f), respectively, much higher than the conventional flip-chip device with the same DUV-LED structure on AlN/sapphire template (23.6 and 42.1 mW at 100 and 200 mA, respectively, Supplementary Fig. S10 and S11). It is convinced that the improvement mainly benefits from the enhancement of the light extraction efficiency, owing to the surface roughening as well as the elimination of the total internal reflection at the epi/substrate interface. This leads to a peak external quantum efficiency of 9.63% at 20 mA, one of the highest values reported to date3. Meanwhile, the temperature distribution in vertical injection DUV-LEDs by infrared thermography is shown in the inset of Fig. 4f, where the die temperature is around 57.8℃ after an operation time of 5 mins at 100 mA, demonstrating better thermal management in the vertical injection devices than that in the flip-chip ones (59.3℃, Supplementary Fig. S11d). It should be noted that the poor n-contact as mentioned above inevitably results in more heat generation during operation, in other words, the die temperature could be further reduced with the issue of n-contact addressed in the vertical injection configuration. Moreover, the full far-field radiation pattern (Fig. 4g) is measured at variable emission angle θ and azimuthal angle φ, where θ = 0° and 90° correspond to the vertical and horizontal emission, respectively. A Lambertian radiation pattern is observed and the on-axis intensity is significantly enhanced in comparison with that in the flip-chip configuration devices (Supplementary Fig. S12), consistent with the result of the light output power.
In summary, a ground-breaking roadmap of AlGaN-based DUV-LEDs stacked on GaN templates is demonstrated to realize the wafer-scale fabrication of devices in vertical injection configuration, from 2 to 4 inches, and even expectably larger. The primary concern of the tensile strain-induced cracks in Al-rich AlGaN on GaN is addressed via the decoupling structure consisting of the strain sacrificial layer and healing layer, making the DUV-LED structure decoupled from the underlying GaN. Moreover, the decoupling structure provides a protection cushion against the thermal shock induced by laser irradiation, preventing the fracturing during LLO. 2- and 4-inch DUV-LED wafers are thus obtained without surface cracks, even after the removal of the sapphire substrates by LLO (355 nm frequency-tripled Nd:YAG laser). In terms of the device performance, the DUV-LED structure is demonstrated to roughly inherit the crystalline quality from the GaN template, leading to an IQE of 70.9% in the active region. It is more important that the 280 nm vertical injection DUV-LEDs in this roadmap exhibit a significant performance improvement, whose LOP reaches 65.2 mW at a current of 200 mA, largely thanks to the essential improvement of light extraction. This work will definitely speed up the application of DUV-LEDs featuring high performance and scalability. What’s more, beneficial from the substitution of AlN templates by GaN ones, ordinary MOCVD systems as well as mature LLO process for InGaN-based visible LEDs can be conveniently employed in the fabrication of DUV-LEDs, greatly promoting the development of this field.