234 nm far-ultraviolet-C light-emitting diodes with polarization-doped hole injection layer

The development of AlGaN-based light-emitting diodes (LEDs) with emission wavelength in the far-ultraviolet-C (far-UVC) spectral region (λ < 240 nm) is driven by various applications like monitoring of gas concentrations (e.g., NO, NH₃),^1,2 the measurement of nitrates in water,³ or the skin-friendly UVC-antisepsis.^4–6 Although these LEDs have been improved over the recent years, the maximum emission power and lifetime of the devices still need to be increased as well as the voltage decreased for commercial applications. Today, the highest emission powers of far-UVC LEDs are in the range of only a few mW, which corresponds to a maximum external quantum efficiency (EQE) of around 1%.^7–11 

The optimization of the semiconductor heterostructure is very important to realize efficient far-UVC LEDs. Major challenges are AlN/sapphire templates with a low dislocation density, an electrically well conducting and transparent Al_xGa_1−xN-based n- and p-side with aluminum mole fractions x > 0.80, a high injection efficiency, and a strong confinement of the carriers in the active region because of the small band offset between quantum well, quantum well barrier, and electron blocking layer (EBL).

In the past, for most UVC LEDs, a conventionally Mg-doped p-side was used.^12–14 However, especially for far-UVC LEDs with their high aluminum content in the p-contact layer, the high ionization energy of the magnesium acceptor in AlGaN materials limits the device performance.^15–18 Particularly, for far-UVC LEDs with a transparent p-side (aluminum mole fraction > 80%), Mg ionization energies of >450 meV are observed.^16–18 This means that only a small fraction of the acceptors is ionized at room temperature, which results in a low p-type conductivity and, thus, in LEDs with a high operating voltage. An alternative approach to realize p-conducting layers is to use polarization doping.^19–24 For this, a gradient of the aluminum content in the p-side AlGaN layer of the LED is necessary. For gallium-face growth, n-type conductivity is obtained by increasing the aluminum content along the growth direction and p-type conductivity for the opposite case of decreasing aluminum content.^25,26 This method works also well for AlGaN layers with a high aluminum content above 80% since no ionization of acceptors is involved. Therefore, polarization doping is a promising approach to produce far-UVC LEDs with a transparent p-side of high conductivity, which promises finally a high emission power and a low voltage of the devices.

Up to now, a detailed description of the optimization of a transparent AlGaN-based p-side in far-UVC LEDs exploiting polarization doping is missing. In this paper, the emission power, voltage, and leakage current of 234 nm LEDs with different polarization-doped hole injection layers (HILs), which replace the conventionally Mg-doped p-side of our LEDs, will be compared. The influence of the thickness of the polarization-doped layer (PDL), an additional Mg doping of the PDL, as well as a combination of a PDL with a conventionally Mg-doped AlGaN layer will be discussed. Finally, it will be shown that the polarization doping concept is well suited to realize far-UVC LEDs with a better performance compared to LEDs with a conventionally Mg-doped p-AlGaN heterostructure.

The semiconductor heterostructures of the 234 nm LEDs were grown on double growth and double annealed (DGA) 2-in. AlN/sapphire templates²⁷ by metal-organic vapor phase epitaxy (MOVPE). Trimethylaluminum, trimethylgallium, triethylgallium, ammonia, disilane, and biscyclopentadienylmagnesium were used as source materials. The templates were fabricated in two steps. In a first step, a 0.4 μm thick AlN-layer was sputtered on 2-in. (00.1) oriented sapphire substrates. Afterward, these templates were annealed for 3 hs at 1720 °C under nitrogen atmosphere in a high temperature oven. In a second step, these wafers were overgrown in a 11 × 2-in. planetary reactor with a 1.1 μm thick AlN layer at an elevated temperature and annealed again for 5 h at 1720 °C under nitrogen atmosphere. Detailed information about the template fabrication can be found in Ref. 28. On these AlN/sapphire templates, the LED heterostructures were grown in a 6 × 2-in. close coupled showerhead reactor. The sample structure is shown in Fig. 1. After deposition of a 500 nm thick AlN base layer, a 20 nm thick graded Al_1.00→0.84Ga_0.00→0.16N transition layer was grown followed by a 900 nm silicon-doped Al_0.84Ga_0.16N contact layer, a triple Al_0.72Ga_0.28N/Al_0.80Ga_0.20N:Si quantum well active region, a 2 nm thick undoped AlN EBL, a 40 nm to 100 nm thick HIL followed by a thin Al_0.82→0.00Ga_0.18→1.00N transition layer, and a 20 nm thick heavily Mg-doped GaN contact cap layer. The AlN base layers on the AlN/sapphire template show a typical full width at half maximum (FWHM) of the x-ray rocking curves of 60 arc sec for the (00.2) reflection and 170 arc sec for the (10.2) reflection. The corresponding threading dislocation density is about 2.1 × 10⁸ cm⁻².^29,30 The layer thicknesses were determined by in situ reflectometry, scanning transmission electron microscopy, and scanning electron microscopy measurements on cross sections of the heterostructures. The degree of strain relaxation and the layer compositions were determined by high-resolution x-ray diffraction using ω−ω/2Θ reciprocal space maps (RSM) of the (00.4) and (11.4) reflections.

In a first sample series, 40, 80, and 100 nm thick AlGaN PDLs were used as HIL [Fig. 1(a)]. The gallium flow was increased linearly over the whole PDL thickness, which corresponds to a composition gradient from AlN to Al_0.82Ga_0.18N. Additional Mg doping of the PDL was investigated in a second sample series [Fig. 1(b)]. For this, samples with a 80 nm thick PDL with a magnesium concentration of around 2.5 × 10¹⁸ and 1.5 × 10¹⁹ cm⁻³ were grown. Finally, a third sample series was analyzed where the HIL consists of a combination of a PDL with a conventionally Mg-doped Al_0.82Ga_0.18N layer of uniform composition on top [Fig. 1(c)]. Also, different magnesium concentrations in the p-Al_0.82Ga_0.18N layer are tested here ([Mg] ∼ 1.5 × 10¹⁹ and 5 × 10¹⁹ cm⁻³). The rest of the LED heterostructure was kept constant.

After MOVPE growth, the samples were annealed in a 70% nitrogen/30% oxygen ambient in order to activate the Mg dopants. LEDs were fabricated using standard chip-processing technologies, detailed information can be found in Refs. 31 and 32. Mesa structures were defined by inductively coupled plasma etching in order to expose the n-AlGaN surface. Platinum-based p-contacts and vanadium–aluminum-based n-contacts were deposited to form the p-electrode and the n-electrode, respectively. The electrical and optical characteristics of the LEDs were measured on-wafer under continuous wave (cw) operation. For that purpose, the wafers were placed epi-side up on a sample holder without active cooling. The emission spectra were measured by collecting the light emitted through the substrate with an optical fiber spectrometer and the optical power vs current (L-I) characteristics by using a calibrated silicon photodiode. Selected wafers were diced, and single chips were flip-chip mounted on planar ceramic packages. The mounted samples were measured under cw operation in a calibrated integrating sphere at a heat sink temperature of 20 °C.

In a first step, LEDs with different PDL thicknesses and, therefore, different AlGaN composition gradients are compared. The rest of the LED-heterostructure and all growth conditions were kept unchanged. Figure 2 shows the mean emission power and voltage of 234 nm LEDs with a PDL thickness of 40, 80, and 100 nm. The measurements were done on-wafer at room temperature at an operation current of 50 mA (cw). Within the scattering of data, measurement accuracy, and reproducibility, the PDL thickness has no significant influence on the emission power. Only for a PDL thickness of 100 nm, a slight decrease in the mean emission power of around 0.05 mW can be guessed. In addition, the voltage remains constant over the whole investigated PDL thickness range with 8.8 V at 50 mA.

However, upon closer inspection of the current–voltage characteristics of LEDs with different PDL thickness, differences in the region below the turn-on voltage can be observed. This is exemplified illustrated in Fig. 3, where the current is plotted as a function of the voltage of twelve representative LEDs for each layer thickness. It can be observed that most of the LEDs with the thin PDL (40 nm, black curves) show a high leakage current below the turn-on voltage. At a PDL thickness of 80 nm (red curves), the number of LEDs with a high leakage current is smaller. Finally, for a PDL thickness of 100 nm (green curves), only one out of twelve LEDs shows significant leakage current. One possible explanation for that behavior could be a lateral thickness fluctuation of the PDL. This, in combination with a roughness of the underlaying heterostructure, can result in additional leakage paths in devices with a too thin PDL. However, with an increase in PDL thickness, the influence of the surface roughness on the following PDL is reduced, which results in a lower number of LEDs showing leakage currents.

This can be also seen in the on-wafer yield, which is shown in Fig. 4 as a function of the PDL thickness. Here, the yield is defined as the percentage of LEDs that show leakage current less than 1 mA at 6 V. For LEDs with a 40 nm thick PDL, only 67% of around 1200 devices per wafer passed the yield criterium. For PDL thicknesses of 80 and 100 nm, the on-wafer yield increases to 82% and 90%, respectively. Therefore, the PDL thickness is one import factor to realize far-UVC LEDs with a high emission power and a low leakage current. In our experiments, a PDL thickness of 100 nm is optimum.

In a second step, the effects of adding a Mg doping to the PDL were investigated with the aim to further improve the LED performance. Therefore, LEDs with an 80 nm thick PDL and a Mg concentration of 0, 2.5 × 10¹⁸, and 1.5 × 10¹⁹ cm⁻³ were fabricated. All other parameters were kept unchanged. Figure 5 shows the mean on-wafer emission power and voltage for different magnesium concentrations of the PDL for an operation current of 50 mA (cw). The LEDs with the undoped PDL show the highest on-wafer emission power of around 0.72 mW. With the increase in Mg concentration of 2.5 × 10¹⁸ and 1.5 × 10¹⁹ cm⁻³ in the PDL, the emission power decreases to around 0.62 and 0.14 mW, respectively. However, a positive influence of the additional Mg doping in the PDL on the voltage or leakage current was not observed. Therefore, additional Mg doping of the PDL is counterproductive regarding the performance of far-UVC LEDs and has not been used for the subsequent sample series.

Finally, LEDs with a combination of a 40 nm thick undoped PDL and a 40 nm thick conventionally Mg-doped AlGaN layer were investigated. For this experiment, two different Mg concentrations of 1.5 × 10¹⁹ and 5 × 10¹⁹ cm⁻³ were used in the AlGaN:Mg layer and compared with a reference sample with a 80 nm undoped PDL [see Fig. 1(a)]. The corresponding mean emission power and voltage measured on-wafer at 50 mA (cw) are shown in Fig. 6. It can be observed that the emission power and voltage of the lower doped sample (1.5 × 10¹⁹ cm⁻³) are similar to those of the reference sample. In contrast, the sample with higher Mg doping (5 × 10¹⁹ cm⁻³) shows around 42% lower emission power and around 19% higher operating voltage. The reason for this behavior was not investigated in detail. However, a possible explanation for this effect is Mg back diffusion³³ into the PDL layer and/or self-compensating effects in the highly doped AlGaN:Mg layer.³⁴ By looking at the on-wafer yield, again defined as the percentage of LEDs that show a leakage current less than 1 mA at 6 V, it is noticeable that the samples with an additional p-AlGaN layer show a reduced yield of around 56% (reference wafer 90%). This yield is similar to that of the sample with the 40 nm thick PDL (without additional p-AlGaN layer), which shows that not the total thickness of the p-side but the thickness of the PDL affects the yield of the LED wafers.

Based on all these optimizations, 234 nm LEDs with a maximum EQE of 1% (at 20 mA, wall plug efficiency 0.63%) as well as an emission power of 4.7 mW and a voltage of 9.0 V at 100 mA (cw) were realized. The corresponding current–voltage and current–emission power characteristic as well as EQE of the mounted LED chip are shown in Fig. 7. The inset of Fig. 7 shows the emission spectrum at 50 mA. A nearly symmetrical emission spectrum with a typical FWHM of 12 nm was observed, which confirms a homogenous emission from the quantum wells and a transparent n-side of the LED heterostructure.

By comparing these LEDs to LEDs with a conventionally Mg doping from the literature (emission wavelength 233–237 nm, emission power at 100 mA < 2 mW),^9,10,35 the emission power of the polarization-doped LEDs, reported in this paper, is higher by more than a factor of two, showing that the polarization doping concept is well suited to realize high performance far-UVC LEDs with a transparent p-side.

The performance of far-UVC LEDs with different polarization-doped HILs emitting at 234 nm has been investigated. It could be shown that the design of the PDL has a strong impact on the emission power, voltage, and leakage current of the devices. Moreover, it could be shown that the polarization doping concept is well suited to realize far-UVC LEDs with a higher performance compared to LEDs with a conventionally Mg-doped p-side.

The PDL thickness has only little effect on the emission power and voltage of the LEDs but a large influence on the leakage current of the LEDs. In this respect, LEDs with a 100 nm thick PDL show the best performance. Additional Mg doping of the PDL reduces the emission power without influencing the voltage or leakage behavior. Finally, a combination of a 40 nm thick undoped PDL with a 40 nm thick conventionally Mg-doped AlGaN layer as a HIL does not improve the device performance but increases the voltage and increases the probability for leakage current. On the contrary, a too high Mg doping of around 5 × 10¹⁹ cm⁻³ results in a 42% decrease in the emission power and a 19% higher voltage of the far-UVC LEDs.

Based on these optimizations, 234 nm LEDs with a maximum EQE of 1.0% and a maximum wall plug efficiency of 0.63% (both at 20 mA) as well as an emission power and operating voltage of 4.7 mW and 9.0 V, respectively, at 100 mA (cw) were realized.

The authors would like to thank C. Neumann and T. Petzke for their technical support in epitaxial growth as well as T. T. H. Nguyen for her contributions in the device preparation and measurements. This work was partially supported by the German Federal Ministry of Education and Research (BMBF) under Project Contract No. 03COV10E.