Enhanced light extraction efficiency of far-ultraviolet-C LEDs by micro-LED array design Open Access

Light emitting diodes (LEDs) emitting in the ultraviolet (UV) spectral range are of great importance in a wide range of applications due to their multiple advantages over conventional UV light sources. They are used, among others, in disinfection, sensing, production, and many more (see Ref. 1 and references therein). Recently, LEDs emitting in the far-UVC range with wavelength below 240 nm have attracted particular attention since they can be used for the skin-safe deactivation of pathogens such as methicillin-resistant staphylococcus aureus (MRSA)² and for the detection of nitrogen oxide gas.³ Despite considerable progress in the development of UV LEDs, their efficiency is still low compared to LEDs emitting in the visible range. This is particularly true for short wavelengths, with maximum external quantum efficiencies (EQE) below 1% for LEDs emitting below 250 nm.⁴ The highest reported EQE values for 233–235 nm LEDs are in the range of 0.28%–0.57%^5,6 on bulk AlN substrates and ∼1% on AlN-sapphire templates.⁷ 

One of the key obstacles limiting the EQE is the light extraction efficiency (LEE). Since the light is extracted through the transparent sapphire substrates for the vast majority of UVB and UVC LEDs on the market, the LEE is limited by absorption in semiconductor layers with smaller bandgap than the emitted light [usually p-(Al)GaN cap layers], absorption in contact metals, and total internal reflection at interfaces between the semiconductor and the substrate as well as the interface between the substrate and the surrounding air. Despite enormous efforts to increase the LEE, e.g., by surface patterning, substrate removal, reflective contacts, and encapsulation in polymers, the overall performance is still comparably low.⁸ One additional reason for the low LEE particularly for far-UVC LEDs is the fact that due to band structure reordering, the emitted light is not anymore dominantly TE-polarized with the electric field vector perpendicular to the c-direction of the wurtzite crystal semiconductor layers and therefore parallel to the wafer surface. Instead, the light is of mixed or dominantly TM-polarized character with a significant part of the emitted light traveling parallel to the quantum wells, thus having a high probability for being trapped in the chip due to total internal reflection and subsequent absorption.⁹ While it is possible to influence the degree of polarization up to a certain point by adjusting the strain in the active region¹⁰ or by modifying the lattice constant of the AlN template,¹¹ enhancing the LEE remains challenging.

One technology that is promising to allow for a significant enhancement of the LEE is the use of micro-structured mesas. While for dominantly TE-polarized light emission in the UVA and UVB range, nanopixels in combination with a reflector material such as aluminum can lead to large improvements,¹² for light emission with a significant amount of TM-polarized light, as can be found particularly in the far-UVC range, an approach with a tilted mesa is more promising.

Micro-LEDs (also called μLEDs) have been in the focus of the development for applications in the visible spectral range, in particular, for displays for virtual reality and augmented reality.¹³ In the UV range, applications such as non-line-of-sight communication¹⁴ and biosensing¹⁵ have been addressed as potential candidates for micro-pixel light sources. Recently, a strong focus has been placed on technological issues such as size-dependent efficiency, parasitic surface recombination, and passivation techniques such as wet chemical treatment and ALD-deposited passivation layers.^16,17 So far, the majority of these investigations focused on micro-LEDs emitting in the visible wavelength range. A study on the size-dependent efficiency of 275 nm LEDs showed an increase by about 20% when reducing the mesa diameter from 300 to 20 μm accompanied by a wavelength shift and an increased optical power density, which was attributed to a more homogeneous current distribution.¹⁸ Floyd et al. studied 275 nm LEDs with slanted or vertical mesa sidewall with diameters down to 5 μm and covered with Al₂O₃-Al, which showed an enhancement of peak EQE by a factor of 1.83.¹⁹ The enhancement is attributed to an improved LEE due to the extraction of TM-polarized photons and an improved heat extraction. In this work, we studied the output power and EQE of far-UVC LEDs emitting at 233 nm and compared arrays of micro-LEDs with slanted mesa side walls and different mesa diameters. Two different types of insulator material deposited on the inclined mesa sidewalls were studied, and numerical simulations were conducted to obtain the expected UV-reflectance of the mesa sidewall.

The LED heterostructures for this study were grown by metalorganic vapor phase epitaxy on 2″ single side polished c-plane sapphire substrates with an off-cut of 0.5° toward the m-direction. It was designed to emit far-UVC radiation at around 233 nm and consists of a 2.5 μm thick AlN base layer, an 850 nm thick n-Al_0:84Ga_0:16N n-side, a threefold AlGaN-AlGaN quantum well active region, a thin AlN interlayer, and a graded, non-intentionally doped AlGaN layer that acts as a polarization doped p-side (more details can be found in Ref. 7). This stack was capped with a 25 nm thick Mg doped GaN layer, and the p-conductivity of the wafers was thermally activated.

After the epitaxial growth, the wafers exhibit a strong convex bow with a typical radius of curvature of 2.5 – 4.0 m. In order to be able to perform the front-end process with high resolution and accuracy, this bow needed to be reduced to a radius of more than 10 m. This was achieved by an internally focused laser process in which the focus depth, number of scribe lines, and pitch of laser lines were adjusted to the initial bow.²⁰ Furthermore, the backside of the wafers was covered with a titanium layer to avoid back reflections from the wafer carrier during projection lithography. This step is proved to be important since the entire layer stack is transparent to the UV light used for photolithography. Later in the process, the Ti layer was removed again.

In the first step of the front-end process, the mesas of the micro-LED arrays were defined by plasma etching, and the n-side was exposed. A positive photoresist was used to generate resist columns with a slightly slanted sidewall. This pattern was transferred into the AlGaN by using a two-step chlorine-based inductively coupled plasma reactive ion etching process²¹ for smooth mesa side walls with a ∼45°–50° inclination angle. The mesas had diameters of 1.5, 2, 5, 10, 20, and 50 μm, a height of approximately 400 nm, and were arranged in square arrays. A reference structure with several 110 μm wide, connected finger-shaped mesas was also included. The effective area based on the top-diameter of the micro-LEDs is summarized in Table I.

After the resist stripping, the vanadium-based n-contact and the platinum p-contact were fabricated by liftoff technique and annealed in nitrogen ambient at 850 °C (n-contacts) and 500 °C (p-contacts), respectively. The n-contacts showed non-linear behavior with contact resistivities in the range of 0.07–0.1 Ω cm² at a current density of about 10 A cm⁻². Next, 200 nm thick insulator layers were deposited by plasma-enhanced chemical vapor deposition (PECVD) and opened above the p-contact and below the n-pad by fluorine-based plasma etching. Either SiN_x or SiO₂ was used as insulator materials. While the SiO₂ has a refractive index of approximately 1.52 at λ = 233 nm²² and is fully transparent in the entire UV range, the SiN_x has a higher refractive index of about 2.4 at λ = 233 nm and is absorbing in the UVC range.²³ A scanning electron microscope (SEM) image of a typical micro-LED after insulator processing is shown in Fig. 1.

An aluminum layer was sputtered onto the LED wafers with the SiO₂ insulator layers to act as an additional reflector, and thick Au-based pads were deposited on all wafers for later soldering to submounts. The wafers were then thinned from the backside to a thickness of approximately 200 μm, polished, laser scribed, and diced into individual chips, which were mounted in flip-chip geometry on planar AlN ceramic submounts.

We conducted transfer-matrix calculations to simulate the effect of the mesa sidewall angle and the insulator properties on the reflection and absorption in the LED structures using refractive indices of 2.7,²⁴ 2.4 + 0.07i,²³ 1.52,²² and 0.13 + 2.3i²⁵ for AlGaN, SiN_x, SiO₂, and Al, respectively. It was found that total internal reflection occurs at the interface between the AlGaN semiconductor and the SiO₂ layer if the angle of incidence is above 34°. For an angle of incidence of 45°, the reflectivity for the p-polarized TM-mode at the SiO₂-AlGaN interface is above 95% if the insulator layer thickness is 100 nm or more, while for the s-polarized TE-mode, it is above 99% for SiO₂ layer thicknesses of more than 25 nm. This also means that almost no light reaches the Al reflector on top of the SiO₂ insulator. On the other hand, strong light absorption occurred in the 200 nm SiN_x layer.

For the electro-optical measurements, the flip-chip-mounted micro-LED-array chips were mounted on heat sinks with an active cooling to 20 °C. Their spectrally resolved output power was then measured in an integrating sphere in 2π configuration. Initially, the LEDs were subjected to a burn-in-process at a nominal current density of 70 A cm⁻² for 5 min since some devices showed inhomogeneous electroluminescence distribution upon first operation. This behavior might be attributed to the migration of hydrogen in the devices at the beginning of operation.²⁶ By applying the burn-in, stable operation with a homogeneous light emission from all pixels of an array was achieved. After the burn-in, the voltage and spectrally resolved output power were measured in a current-controlled regime until roll-over of the output power was observed.

The emission spectra of all LEDs showed single peak emission with a maximum at around 233–234 nm and only little parasitic luminescence around 420 nm, which had at least two orders of magnitude lower intensity than the far-UVC emission and was excluded from the power measurements. We did not observe any dependency of the peak wavelength on the structure size. The main emission peak had a full width at half maximum (FWHM) of 9–13 nm.

Figure 2 shows the operation voltage of the LEDs as a function of the dc current. As can be seen, at 50 mA, all devices show operation voltages between 7 and 11 V. Arrays with smaller mesa diameters show higher voltages as had been expected from the overall smaller p-contact area (see also Table I). Furthermore, the operation voltages of devices with SiO₂ insulator exhibit higher operation voltages than their SiN_x counterparts. This might be linked to the different deposition techniques of these layers (e.g., deposition temperature and gas phase content), resulting in differing hydrogen concentrations in the insulator materials. These could, in turn, influence the p-conductivity through hydrogen migration and passivation of Mg in the p-side.²⁷ 

Figure 3 shows the spectrally integrated output power of micro-LED arrays with different mesa diameters and with either SiO₂ or SiN_x insulator as a function of the applied current. It should be noted that the overall effective current-carrying, thus light emitting area, differs depending on the mesa diameter, i.e., different current densities are reached for the same current. Therefore, the point of roll-over is reached at lower currents for LEDs with smaller mesa diameters and smaller overall active area. Furthermore, the point of roll-over is reached at lower currents for the SiO₂-based micro-LED arrays than for the SiN_x-based ones with the same mesa diameters, which is attributed to the higher temperature due to the higher operation voltages shown in Fig. 2. As can be seen, for the LEDs with SiN_x insulator, the output power at a fixed current below roll-over increases only slightly with decreasing diameter. In contrast, for LEDs with SiO₂ insulator, the output power for the smallest mesa diameters is significantly increased.

This becomes even more obvious when the EQE is compared for both insulator types as a function of the applied current density, calculated based on the top-diameter of the micro-LED's etched mesa (see Fig. 4). While the peak EQE for the LEDs with SiN_x increases from about (0.4 − 0.5)% to ∼0.6% for diameters below 5 μm, the peak EQE for LEDs with SiO₂ increases more significantly with a maximum of 1.62% for a diameter of 1.5 μm. This is an improvement by a factor of three as compared to 10 μm mesa diameters (peak EQE: 0.54%) and by a factor of four as compared to the standard large area devices (0.4%, gray curve in Figs. 2–4). The peak EQE of 1.62% corresponds to a wall plug efficiency (WPE) of 0.88% and was achieved at a current of 20 mA and a nominal current density of approximately 100 A cm⁻². Furthermore, it is the highest reported EQE for an LED emitting in this wavelength range^4–7 so far.

The origin of the different current densities at which the peak EQE occurs might be connected to the fact that the nominal current-carrying area of a pixel is derived from its mesa diameter at the top, while the p-contact metal is slightly smaller due to manufacturing. Therefore, the effective area for current flow and photon generation could differ from the value used in the evaluation, depending on the current spreading length. This effect should be stronger for smaller devices.

In Fig. 5, the peak EQE of the devices with SiN_x or SiO₂ insulator is shown as a function of mesa diameter. As can be seen, the LEDs with SiN_x insulator show only a weak dependency of peak EQE on diameter. This is expected since the SiN_x is absorbing the UV light, and the refractive index contrast to AlGaN is low so that only little radiation is reflected at the mesa edge. In contrast, the peak EQE for LEDs with non-absorbing and low refractive index SiO₂ insulator shows a strong dependency on the mesa diameter, indicating a large enhancement of the LEE.

There are two potentially counter-effective mechanisms influencing the EQE of the micro-LED arrays. While the internal quantum efficiency (IQE) can be reduced by non-radiative carrier recombination at the etched mesa sidewalls, the LEE can be increased if photons are redirected upon reflection at the mesa and can leave the chip. Both effects should scale with the mesa diameter. The IQE should drop with decreasing mesa diameter, since more and more carriers can reach the mesa edge by drift or diffusion and recombine non-radiatively via states from non-passivated surface defects or dangling bonds.¹⁶ This effect could be different for the two investigated insulator materials. In contrast, a shorter distance from the place of photon generation to the point of possible reflection at a mesa edge increases the probability that a photon is redirected and leaves the chip before it is absorbed.¹⁷ This requires that the mesa sidewalls are inclined by about 45°, and that there is a non-absorbing, reflective layer deposited on the mesa sidewall. This case is true for the 200 nm of SiO₂ used in our LEDs due to its transparency and high refractive index contrast to AlGaN. Since we see a strong increase in the peak EQE for smaller diameters with no sign of saturation, it can be concluded that the LEE enhancement dominates over the potential IQE reduction.

In summary, we fabricated arrays of micro-LEDs emitting in the far-UVC region at around 233 nm with a record peak EQE of 1.62% and WPE of 0.88%. By reducing the mesa diameter and using slanted mesa side walls in combination with a UV-transparent, low refractive index SiO₂ insulator deposited on the mesa, the dominantly TM-polarized light was redirected toward the chip backside, decreasing the probability for light trapping by total internal reflection, and thus enhancing the LEE and EQE by a factor of four as compared to conventional LEDs with large area mesas. Further work will be addressing the potential issue of carrier recombination on the mesa sidewall.

The authors would like to thank Sylvia Hagedorn for providing the AlN/sapphire templates used in this study, Tamukanashe Anthony Musengezi for mounting the chips on submounts, and Huyen Nguyen for supporting in the measurements. This work was partially supported by the Federal Ministry of Education and Research (BMBF) through Contract No. 03COV10E.

The authors have no conflicts to disclose.

Jens Rass: Conceptualization (lead); Formal analysis (lead); Investigation (lead); Methodology (equal); Project administration (equal); Resources (equal); Software (supporting); Supervision (equal); Visualization (lead); Writing – original draft (lead); Writing – review & editing (lead). Hyun Kyong Cho: Investigation (equal); Methodology (equal). Martin Guttmann: Investigation (equal); Methodology (equal); Resources (equal); Software (lead); Writing – review & editing (equal). Deepak Prasai: Investigation (equal). Jan Ruschel: Investigation (equal); Writing – review & editing (equal). Tim Kolbe: Resources (equal); Writing – review & editing (equal). Sven Einfeldt: Conceptualization (supporting); Funding acquisition (lead); Project administration (equal); Supervision (equal); Writing – review & editing (equal).

The data that support the findings of this study are available from the corresponding author upon reasonable request.