UVB LEDs Grown by Molecular Beam Epitaxy Using AlGaN Quantum Dots

: AlGaN based light emitting diodes (LEDs) will play a key role for the development of applications in the ultra-violet (UV). In the UVB region (280–320 nm), phototherapy and plant lighting are among the targeted uses. However, UVB LED performances still need to be improved to reach commercial markets. In particular, the design and the fabrication process of the active region are central elements that affect the LED internal quantum efficiency (IQE). We propose the use of nanometer-sized epitaxial islands (i.e., so called quantum dots (QDs)) to enhance the carrier localization and improve the IQE of molecular beam epitaxy (MBE) grown UVB LEDs using sapphire substrates with thin sub- µ m AlN templates. Taking advantage of the epitaxial stress, AlGaN QDs with nanometer-sized ( ≤ 10 nm) lateral and vertical dimensions have been grown by MBE. The IQE of the QDs has been deduced from temperature dependent and time resolved photoluminescence measurements. Room temperature IQE values around 5 to 10% have been found in the 290–320 nm range. QD-based UVB LEDs were then fabricated and characterized by electrical and electroluminescence measurements. On-wafer measurements showed optical powers up to 0.25 mW with external quantum efficiency (EQE) values around 0.1% in the 305–320 nm range. 1) A of A/cm A/cm saturation of the minimum for A/cm 2 and 70A/cm 2 ; and an of the higher J (above A/cm the abrupt A/cm 2 in the case of LED B is mainly attributed a change of Fabry–Perot mode (see Figure


Introduction
Nitride based light emitting diodes (LEDs) can emit in the ultra-violet (UV) range from the near UV [1], which is a part of the UVA region (400 nm ≤ λ UVA ≤ 320 nm), down to the UVC region (280 nm ≤ λ UVC ≤ 100 nm) reaching a minimum wavelength emission of 210 nm achieved by using an AlN p-i-n junction [2]. Compared to the other existing UV sources, LEDs offer strong advantages such as low power consumption, compactness and long lifetimes. Furthermore, they are an environmentally-friendly solution to the commonly used mercury (Hg) lamps, which bring toxicity and recycling issues. Indeed, the replacement of Hg lamps has been planned to start from 2020 by the Minamata Convention on mercury [3]. Aluminum gallium nitride (Al x Ga 1−x N) based LEDs are expected to be the new UV technology and their development has attracted very important research activities over the last decade [4]: In particular, the fabrication of UVB and UVC-LEDs represent major topics. The UVC spectral region is targeting strategic applications for sterilization, water purification and surface disinfection, by inducing bacteria cell inactivation [5]. Another region of interest is the UVB for medical and environmental applications, such as skin treatment and plant growth [4]. In this spectral region, external quantum efficiencies (EQEs) of 4.4% and 2.2% at a current of 20 mA and 50 mA were demonstrated recently for LEDs emitting at 295 nm [6] and 310 nm [7], respectively. Therefore,

Results
In the first part of this section, theAl0.2Ga0.8N (n.c.) QDs main structural and optical properties are presented. The second part is dedicated to the investigation of the QD-based LEDs including PL, I-V, and EL measurements.

Al0.2Ga0.8N QDs Structural and Optical Properties
A typical AFM image of Al0.2Ga0.8N (n.c.) QDs grown at the surface of an Al0.7Ga0.3N layer is shown in Figure 2a. Similarly to previous results [28,33], the QD densities determined from AFM measurements have been found between 2 × 10 11 and 6 × 10 11 /cm 2 and their diameter varies between 5 and 15 nm. From previous transmission electron microscopy measurements [28], the average QD height is estimated to be of the order of the Al0.2Ga0.8N (n.c.) deposited amount ranging from 2 nm (for 8 MLs) to 3 nm (for 12 MLs), with a height distribution of ±20%. In Figure 2b are presented PL spectra at 300 K of a series of samples of Al0.2Ga0.8N (n.c.) QDs with deposited amounts ranging from 8 to 12 MLs. Each spectrum is characterized by a broad peak originating from the QD emission: Decreasing the amount of Al0.2Ga0. 8N  To investigate the optical properties of Al 0.2 Ga 0.8 N QDs, samples composed of a buried Al 0.2 Ga 0.8 N QD plane grown in an Al 0.7 Ga 0.3 N cladding layer were grown and the influence of the Al 0.2 Ga 0.8 N deposited amount was investigated by changing it from 8 to 12 MLs. In addition, an Al 0.2 Ga 0.8 N QD plane was grown at the surface of the top cladding layer. The surface QD plane was investigated by atomic force microscopy (AFM) to study the QD morphology. Concerning the optical characteristics, they were studied by photoluminescence (PL) and time-resolved PL (TRPL). The AFM measurements have been done in tapping mode using a diamond coated tip with a typical radius of 5 nm. The PL measurements were carried out in a closed cycle He cryostat using a frequency-doubled Ar laser at 244 nm (5.08 eV) and an excitation power of 20 mW. TRPL measurements were performed at 8 K using a mode-locked frequency-tripled titanium-sapphire laser with a wavelength of 260 nm (4.77 eV). The laser excitation power was about 200 µW.
After the growth, LEDs with a squared geometry were processed by using photolithography and reactive ion etching. The mesas are covered by a thin Ni/Au (5/5 nm) semi-transparent current spreading layer plus a Ni/Au (5/200 nm) contact as top electrode to p-GaN. The contact on n-Al 0.7 Ga 0.3 N consists of a stacking of Ti/Al/Ni/Au (30/180/40/200 nm). LEDs with a squared mesa size of 310 µm (corresponding to mesa areas of~96,000 µm 2 ) have been studied. The LEDs were characterized at room temperature by performing on wafer current-voltage (I-V) and electroluminescence (EL) measurements. The EL was measured from the backside of the LED chips by collecting the light extracted from the sapphire substrate side in an UV-grade optical fiber connected to a linear CCD spectrometer with a spectral resolution of 2.5 nm. The optical power (P opt ) was also measured from the backside of the chips by using a calibrated UV enhanced Si photodiode.

Results
In the first part of this section, theAl 0.2 Ga 0.8 N (n.c.) QDs main structural and optical properties are presented. The second part is dedicated to the investigation of the QD-based LEDs including PL, I-V, and EL measurements.

Al 0.2 Ga 0.8 N QDs Structural and Optical Properties
A typical AFM image of Al 0.2 Ga 0.8 N (n.c.) QDs grown at the surface of an Al 0.7 Ga 0.3 N layer is shown in Figure 2a. Similarly to previous results [28,33], the QD densities determined from AFM measurements have been found between 2 × 10 11 and 6 × 10 11 /cm 2 and their diameter varies between 5 and 15 nm. From previous transmission electron microscopy measurements [28], the average QD height is estimated to be of the order of the Al 0.2 Ga 0.8 N (n.c.) deposited amount ranging from 2 nm (for 8 MLs) to 3 nm (for 12 MLs), with a height distribution of ±20%. In Figure 2b are presented PL spectra at 300 K of a series of samples of Al 0.2 Ga 0.8 N (n.c.) QDs with deposited amounts ranging from 8 to 12 MLs. Each spectrum is characterized by a broad peak originating from the QD emission: Decreasing the amount of Al 0.2 Ga 0.8 N from 12 MLs to 8 MLs induces a UV shift of the main PL maximum intensity from 324 to 308 nm (3.83 to 4.02 eV). The PL spectra full width at half maximum (FWHM) are similar and found between 12 and 15 nm (≈150 to 190 meV). This broadening reflects alloy disorder in (Al,Ga)N and QD sizes inhomogeneity, i.e., confinement energies and Stark effect inhomogeneities as discussed in Section 4 below. Next, TRPL measurements are presented in Figure 2c, along with the temperature dependent PL measurement in insert (measured in a temperature range between 8 K and 300 K). At first, it is found that the PL transients exhibit a two-exponential decay behavior. This particular decrease of the PL intensity is attributed to the presence of non-radiative recombination channels unsaturated in the photo-injection conditions. As detailed in [26,34], the time dependence of the PL intensity I(t) can be fitted by considering fast (τ fast ) and long (τ slow ) decay times using the equation: with A fast and A slow the fast (non radiative) and slow (radiative) decay coefficients, respectively. This bi-exponential behavior is attributed to two different types of regions within the QD plane, i.e., regions with predominant radiative channels and other ones with additional competing non-radiative recombination centers [35]. τ slow defines the radiative lifetime, while τ fast is a combination of both radiative and non-radiative lifetimes. Values in the range of 2-2.2 ns and 16.8-18 ns have been determined for τ fast and τ slow , respectively, while A fast and A slow have values in the 930-970 and 170-185 range, respectively. From these data, it is possible to determine the IQE of Al 0.2 Ga 0.8 N (n.c.) QDs in Al 0.7 Ga 0.3 N cladding layers which will be used as the active region of UV LEDs presented in the following section. The IQE at low temperature (LT) can then be calculated as follows: Crystals 2020, 10, 1097 5 of 14 and IQE-LT values between 24% and 28% are obtained. In order to estimate the IQE at room temperature (300 K), temperature dependent PL measurements have also been realized and the evolution of the integrated PL intensity as a function of temperature determined (insert of Figure 2c). As discussed in a previous study [28], the IQE at 300 K is obtained by considering the relation: with I PL (300 K) the integrated PL intensity at 300 K and I PL (LT) the integrated PL intensity at 8 K.
The ratio I PL (300 K)/I PL (LT) is found to be between 0.2 and 0.3, giving IQE values at room temperature for Al 0.2 Ga 0.8 N (n.c.) QDs in Al 0.7 Ga 0.3 N cladding layers in the range of 4.8-8.4% in agreement with previous results and confirming the robustness of QDs in structures with high dislocation densities (>10 10 cm −2 ) [28]. Using similar active regions, QD-LED structures (LED-A and LED-B) were then fabricated as described in Figure 1 and Section 2 and are presented in the next section.
185 range, respectively. From these data, it is possible to determine the IQE of Al0.2Ga0.8N (n.c.) QDs in Al0.7Ga0.3N cladding layers which will be used as the active region of UV LEDs presented in the following section. The IQE at low temperature (LT) can then be calculated as follows: and IQE-LT values between 24% and 28% are obtained. In order to estimate the IQE at room temperature (300 K), temperature dependent PL measurements have also been realized and the evolution of the integrated PL intensity as a function of temperature determined (insert of Figure  2(c)). As discussed in a previous study [28], the IQE at 300 K is obtained by considering the relation: with IPL(300 K) the integrated PL intensity at 300 K and IPL(LT) the integrated PL intensity at 8 K. The ratio IPL(300 K)/IPL(LT) is found to be between 0.2 and 0.3, giving IQE values at room temperature for Al0.2Ga0.8N (n.c.) QDs in Al0.7Ga0.3N cladding layers in the range of 4.8%-8.4% in agreement with previous results and confirming the robustness of QDs in structures with high dislocation densities (> 10 10 cm −2 ) [28]. Using similar active regions, QD-LED structures (LED-A and LED-B) were then fabricated as described in Figure 1 and Section 2 and are presented in the next section.

QD-Based LEDs Characteristics
The LEDs have a squared mesa size of 310 µm and were characterized at room temperature by performing on wafer I-V measurements as well as EL determined from the backside of the device (i.e., extracted from the sapphire substrate).

Photoluminescence
Before processing the heterostructures, their optical properties were studied at room temperature using the frequency-doubled Ar laser at 244 nm. The results for LED-A and LED-B structures are shown and compared in Figure 3. In each spectrum, the photoluminescence covers a very large spectral range spanning wavelengths from 280 nm to above 550 nm. More precisely, the emission is separated between two regions: The most intense one is made of multi-peaks in the UV range with a maximum peak intensity at 314 nm for LED-A and at 320 nm for LED-B. The FWHM of

QD-Based LEDs Characteristics
The LEDs have a squared mesa size of 310 µm and were characterized at room temperature by performing on wafer I-V measurements as well as EL determined from the backside of the device (i.e., extracted from the sapphire substrate).

Photoluminescence
Before processing the heterostructures, their optical properties were studied at room temperature using the frequency-doubled Ar laser at 244 nm. The results for LED-A and LED-B structures are shown and compared in Figure 3. In each spectrum, the photoluminescence covers a very large spectral range spanning wavelengths from 280 nm to above 550 nm. More precisely, the emission is separated between two regions: The most intense one is made of multi-peaks in the UV range with a maximum peak intensity at 314 nm for LED-A and at 320 nm for LED-B. The FWHM of this emission is very broad, with a value of 50 nm and 67 nm (i.e., 630 and 810 meV) for LED-A and LED-B, respectively. This is much broader than the PL bands of single QD planes shown in Figure 2b, and reflects in our opinion the QD size and composition variations between the QD planes along the growth direction. In fact, for both structures, the emission ranges from about 280 nm to 400 nm. Yet, the band gap of Al 0.2 Ga 0.8 N is estimated at an energy of 3.89 eV that corresponds to a wavelength of 319 nm [32]. Therefore, the broadening of the PL spectrum is attributed to the QD sizes and/or Al compositions Crystals 2020, 10, 1097 6 of 14 fluctuations [24]. Regarding the presence of multi-peaks, it is mainly due to the Fabry-Perot effect resulting from the high refractive index contrasts in the heterostructures. Regarding the emission in the visible (blue-green) range, it is attributed to defects/impurities below the Al 0.7 Ga 0.3 N cladding layer band gap, p-and n-type, as commonly observed in Al x Ga 1−x N materials [36] and corroborated by the EL measurements for which no emission is observed in this wavelength range due to the electrical injection of carriers in the QD active layers only (not shown).
this emission is very broad, with a value of 50 nm and 67 nm (i.e., 630 and 810 meV) for LED-A and LED-B, respectively. This is much broader than the PL bands of single QD planes shown in Figure  2b, and reflects in our opinion the QD size and composition variations between the QD planes along the growth direction. In fact, for both structures, the emission ranges from about 280 nm to 400 nm. Yet, the band gap of Al0.2Ga0.8N is estimated at an energy of 3.89 eV that corresponds to a wavelength of 319 nm [32]. Therefore, the broadening of the PL spectrum is attributed to the QD sizes and/or Al compositions fluctuations [24]. Regarding the presence of multi-peaks, it is mainly due to the Fabry-Perot effect resulting from the high refractive index contrasts in the heterostructures. Regarding the emission in the visible (blue-green) range, it is attributed to defects/impurities below the Al0.7Ga0.3N cladding layer band gap, p-and n-type, as commonly observed in AlxGa1−xN materials [36] and corroborated by the EL measurements for which no emission is observed in this wavelength range due to the electrical injection of carriers in the QD active layers only (not shown).

Electrical Properties
The I-V curves of both devices are presented and superimposed on a semi-log plot in Figure 4. Starting from the characteristics under forward bias conditions, similar turn-on voltages are found around 12 ± 1 V and voltage values of around 15 ± 1 V are measured at 20 mA (i.e., at a current density (J) of 20.8 A/cm 2 ). The high operating voltages are attributed to low conductivities in the doped Al0.7Ga0.3N and Al0.8Ga0.2N layers, in particular the p-type ones characterized by very high activation energies (≥400 meV) of the Mg dopants [37]. Parasitic resistances have also been extracted and series resistance (Rs) values at around 60 Ω have been determined for both LEDs with a slightly higher one in the case of LED-B. They are the result of excessive contact resistances. Under reverse bias conditions, a leakage current of 1 µA (1 mA/cm 2 ) is found at -6.5 V for LED-A and at -3.3 V for LED-B. The significant reverse saturation current are due to leakage paths in the devices which could be related to the high dislocation densities in our devices (>10 10 cm −2 ). An unusual behavior of the I-V characteristics at low voltages (between ± 1 V) is observed with important variations of voltage values at low currents (<1 µA) attributed to unwanted transport mechanisms caused by structural defects. In addition, sub-threshold turn-on can be seen, in particular in the case of LED-B, which is caused by carrier transport (leakage) caused by surface states or deep levels and resulting in a shunt effect.

Electrical Properties
The I-V curves of both devices are presented and superimposed on a semi-log plot in Figure 4. Starting from the characteristics under forward bias conditions, similar turn-on voltages are found around 12 ± 1 V and voltage values of around 15 ± 1 V are measured at 20 mA (i.e., at a current density (J) of 20.8 A/cm 2 ). The high operating voltages are attributed to low conductivities in the doped Al 0.7 Ga 0.3 N and Al 0.8 Ga 0.2 N layers, in particular the p-type ones characterized by very high activation energies (≥400 meV) of the Mg dopants [37]. Parasitic resistances have also been extracted and series resistance (R s ) values at around 60 Ω have been determined for both LEDs with a slightly higher one in the case of LED-B. They are the result of excessive contact resistances. Under reverse bias conditions, a leakage current of 1 µA (1 mA/cm 2 ) is found at -6.5 V for LED-A and at -3.3 V for LED-B. The significant reverse saturation current are due to leakage paths in the devices which could be related to the high dislocation densities in our devices (>10 10 cm −2 ). An unusual behavior of the I-V characteristics at low voltages (between ±1 V) is observed with important variations of voltage values at low currents (<1 µA) attributed to unwanted transport mechanisms caused by structural defects. In addition, sub-threshold turn-on can be seen, in particular in the case of LED-B, which is caused by carrier transport (leakage) caused by surface states or deep levels and resulting in a shunt effect.

Electroluminescence
Electroluminescence (EL) measurements of the two types of LEDs as a function of the input current have been performed and a comparison of the results at four different current values is presented in Figure 5. As observed by PL (Figure 3), a multi-peak emission in the UV spectral range is found, in particular at lower current values for which a wavelength spectral range from nearly 280 Electroluminescence (EL) measurements of the two types of LEDs as a function of the input current have been performed and a comparison of the results at four different current values is presented in Figure 5. As observed by PL (Figure 3), a multi-peak emission in the UV spectral range is found, in particular at lower current values for which a wavelength spectral range from nearly 280 nm to 400 nm is covered. In addition, a shift towards shorter wavelengths of the maximum intensity EL peak is occurring when the input current increases.

Electroluminescence
Electroluminescence (EL) measurements of the two types of LEDs as a function of the input current have been performed and a comparison of the results at four different current values is presented in Figure 5. As observed by PL (Figure 3), a multi-peak emission in the UV spectral range is found, in particular at lower current values for which a wavelength spectral range from nearly 280 nm to 400 nm is covered. In addition, a shift towards shorter wavelengths of the maximum intensity EL peak is occurring when the input current increases. In order to have more insight into the dependence of the EL emission as a function of the input current density, the variation of the average EL wavelength is reported in Figure 6 and the variation of the full width at half maximum (FWHM) in Figure 7. In fact, since the EL spectra are broad and made of several peaks, the emitted wavelength has been arbitrarily defined as the arithmetic mean between the two wavelength values measured at full width at half maximum of each EL spectrum. For both LEDs, a similar behavior of the wavelength dependence is observed, which involves three distinct regimes (labelled with numbers from 1 to 3 in Figure 6): 1) A fast reduction of the wavelength at lower J (i.e., from 0.1 A/cm 2 to 40 A/cm 2 ); 2) a saturation of the wavelength at a minimum value (λmin) for J between 40 A/cm 2 and 70A/cm 2 ; and 3) an increase of the wavelength at higher J (above 70 A/cm 2 ). Actually, the abrupt wavelength jump at 40 A/cm 2 in the case of LED B is mainly attributed to a change of Fabry-Perot mode (see Figure 5b). In order to have more insight into the dependence of the EL emission as a function of the input current density, the variation of the average EL wavelength is reported in Figure 6 and the variation of the full width at half maximum (FWHM) in Figure 7. In fact, since the EL spectra are broad and made of several peaks, the emitted wavelength has been arbitrarily defined as the arithmetic mean between the two wavelength values measured at full width at half maximum of each EL spectrum. For both LEDs, a similar behavior of the wavelength dependence is observed, which involves three distinct regimes (labelled with numbers from 1 to 3 in Figure 6): (1) A fast reduction of the wavelength at lower J (i.e., from 0.1 A/cm 2 to 40 A/cm 2 ); (2) a saturation of the wavelength at a minimum value (λ min ) for J between 40 A/cm 2 and 70 A/cm 2 ; and (3) an increase of the wavelength at higher J (above 70 A/cm 2 ). Actually, the abrupt wavelength jump at 40 A/cm 2 in the case of LED B is mainly attributed to a change of Fabry-Perot mode (see Figure 5b). Regarding the FWHM, two regimes can be distinguished between lower and higher J. At low J (<40 A.cm −2 ), a difference is observed between the two types of LEDs since the FWHM strongly decreases with J for LED-A by 40 nm (from 70 nm to 30 nm), whereas it remains very high (around 50-60 nm) and fairly current independent in the case of LED-B. Actually, below 40 A/cm 2 , the EL spectra of LED B is mainly structured by two Fabry Perrot modes, and only one at higher current −2 Figure 6. Variation of the average EL peak wavelength value with increasing input current density J for (a) LED-A and (b) LED-B. The dashes lines are guides for the eyes, including an exponential fit of the wavelength variation at lower current densities (region 1). which leads to the strong difference in the spectra FWHM. At high J (>40 A.cm −2 ), identical behaviors are found for both LEDs characterized by an almost independence of the FWHM with J and a stabilization at a minimum value around 25 ± 3 nm. This characteristic implies a severe reduction (drop) of the FWHM for LED-B from 51 nm to 24 nm for a variation range of J from 37 to 47 A.cm −2 . In addition, this FWHM narrowing corresponds to an important UV shift of the EL emission from 322 nm to 304 nm (corresponding to the minimum wavelength λmin). This shift is enhanced by the structuration of the spectra by interferences. Regarding the FWHM, two regimes can be distinguished between lower and higher J. At low J (<40 A·cm −2 ), a difference is observed between the two types of LEDs since the FWHM strongly decreases with J for LED-A by 40 nm (from 70 nm to 30 nm), whereas it remains very high (around 50-60 nm) and fairly current independent in the case of LED-B. Actually, below 40 A/cm 2 , the EL spectra of LED B is mainly structured by two Fabry Perrot modes, and only one at higher current which leads to the strong difference in the spectra FWHM. At high J (>40 A·cm −2 ), identical behaviors are found for both LEDs characterized by an almost independence of the FWHM with J and a stabilization at a minimum value around 25 ± 3 nm. This characteristic implies a severe reduction (drop) of the FWHM for LED-B from 51 nm to 24 nm for a variation range of J from 37 to 47 A·cm −2 . In addition, this FWHM narrowing corresponds to an important UV shift of the EL emission from 322 nm to 304 nm (corresponding to the minimum wavelength λ min ). This shift is enhanced by the structuration of the spectra by interferences.

Optical Power and External Quantum Efficiency
The performances of the LEDs as a function of J have been further investigated by analyzing the light power-current density (L-J) characteristics of LED-A and LED-B. The measurements are presented in Figure 8. For both LED, P opt is in the µW range for current densities of 2 A·cm −2 or below, around 100 µW in the 40-50 A·cm −2 range and reaches 230-250 µW at 100 A·cm −2 . However, this increase is not constant as a function of J. Actually, P opt varies as a function of J at a constant power m, i.e., it is proportional to J m , and the value of m depends on the current density range. At lower J, below 40-50 A/cm 2 , a linear variation of P opt is found (m = 1), whereas at higher J an under linear variation (m < 1) is observed. In the case of LED-A, a value of 0.8 is determined for J above 50 A·cm −2 . In the case of LED-B, a much smaller value of 0.2 is found for J above 80 A·cm −2 . However, the behavior is more complex for this device, since between 50 and 80 A·cm −2 , a super linear dependence (m > 1) of P opt is observed with a value of 1.56, corresponding to an increase of the emitted power from 100 µW to 230 µW.
Crystals 2020, 10, x FOR PEER REVIEW 9 of 14 variation (m < 1) is observed. In the case of LED-A, a value of 0.8 is determined for J above 50 A.cm −2 .
In the case of LED-B, a much smaller value of 0.2 is found for J above 80 A.cm −2 . However, the behavior is more complex for this device, since between 50 and 80 A.cm −2 , a super linear dependence (m > 1) of Popt is observed with a value of 1.56, corresponding to an increase of the emitted power from 100 µW to 230 µW. In Figure 9, the EQE of both LEDs has been reported as a function of J, following the well-known equation: with h the Planck constant, c the speed of light, λ the emitted wavelength, e the electron charge, and I the injected current.
As a general result, EQE values between 0.03% and 0.15% have been found with higher ones at lower J (below 4 A.cm −2 ). The decrease of EQE at high operating current density is commonly observed in LEDs and known as the "efficiency droop", whose origin is mainly attributed to Auger recombination and to a lesser extend electron leakage contribution [38]. Differences are observed between both LED EQE characteristics, with higher EQEs found for LED-B at J below 4 A.cm −2 whereas LED-A shows higher ones for J between 4 A.cm −2 and 60 A.cm −2 . For J above 60 A.cm −2 , fairly similar values are obtained around 0.06-0.07%.  In Figure 9, the EQE of both LEDs has been reported as a function of J, following the well-known equation: with h the Planck constant, c the speed of light, λ the emitted wavelength, e the electron charge, and I the injected current.   As a general result, EQE values between 0.03% and 0.15% have been found with higher ones at lower J (below 4 A·cm −2 ). The decrease of EQE at high operating current density is commonly observed in LEDs and known as the "efficiency droop", whose origin is mainly attributed to Auger recombination and to a lesser extend electron leakage contribution [38]. Differences are observed between both LED EQE characteristics, with higher EQEs found for LED-B at J below 4 A·cm −2 whereas LED-A shows higher ones for J between 4 A·cm −2 and 60 A·cm −2 . For J above 60 A·cm −2 , fairly similar values are obtained around 0.06-0.07%.

Discussion
From the LED characteristics presented above, several points deserve to be commented. At first, the PL emission coming from the QDs, which intensity is maximum in the UVB range at 314 nm and 320 nm for LED-A and LED-B, respectively, is covering a broad emission range from 280 nm up to 400 nm. This feature is in agreement with the EL spectra which show large FWHMs up to 60-70 nm at low J. It is the consequence of fluctuations in the QD active regions in terms of QD size (i.e., mainly the QD height since lateral confinement effects in Al y Ga 1−y N QDs are weak compared to those along the <0001> direction due to a height over diameter ratio of the order of 4 [28]) and Al composition including the intrinsic alloy inhomogeneous broadening as discussed in previous works [24,32]. Regarding the EL and the PL intensity modulation, it is influenced by a Fabry-Perot effect resulting from the high refractive index contrasts at the AlGaN/sapphire and the air/AlGaN interfaces.
It should be noted that the broad spectral range emission is also a direct consequence of the presence of an internal electric field (F int ) in the heterostructures. This field is originating from the interfacial polarization discontinuities, leading to the quantum confined Stark effect (QCSE) in Al y Ga 1−y N/Al x Ga 1−x N heterostructures. Indeed, the large values of F int , that can reach several MV/cm in GaN-based QDs [26,39], have been shown to lead to an important red-shift of the QD emission, i.e., at much longer wavelengths than the QD material band gap energy and up to the visible spectral range [25]. However, as presented in Figures 6 and 7, the QD emitted average wavelength as well as the emission range (FWHM) strongly depends on the injected carrier density at lower J (<40 A·cm −2 ): A blueshift from 345 nm to 315 nm for LED-A and from 345 nm to 305 nm for LED-B is observed, together with a reduction of the FWHM from more than 60 nm down to 25 nm. In fact, for J typically above 30-40 A·cm −2 , the LEDs emit, as expected, in the UVB range below 320 nm. This result means that under these conditions, carriers are injected in the main Al y Ga 1−y N QD population with y close to the n.c. of 0.2, since the band gap energy of Al 0.2 Ga 0.8 N corresponds to a wavelength of 319 nm. The QD emission below the band gap energy of Al 0.2 Ga 0.8 N is the consequence of the quantum confinement effect together with a minimized influence of F int as the Al concentration of Al y Ga 1−y N QDs increases (due to the reduction of the piezoelectric polarization originating from the lattice-mismatch between Al y Ga 1−y N QDs and the Al 0.7 Ga 0.3 N cladding layer). Whereas, at lower J, carriers are preferentially injected in QDs with lower Al concentrations which represent deeper localization potentials, i.e., larger band offsets with the Al 0.7 Ga 0.3 N cladding layers. Differences in the LED characteristics between the two devices, that are attributed to the design of the active region made of 3 QD planes (LED-A) or 5 QD planes (LED-B), are observed at lower current densities (<40 A·cm −2 ). In this regime, the wavelength and EL spectrum FWHM values are characterized by a strong dependence on J (which varies from 0.1 to 40 A·cm −2 ). For both LEDs, a progressive shift towards shorter wavelengths is observed, which is the consequence of: (i) The partial screening of F int due to an increase of the electron and hole concentrations in the QDs [26,40], (ii) the progressive injection into higher Al concentration Al y Ga 1−y N QDs at larger J and (iii) a band-filling effect in the QDs [41][42][43]. The main difference between LED-A and LED-B is seen in the FWHM dependence of the EL spectra as a function of J: In the case of LED-A, a progressive reduction of the FWHM is observed (from 70 nm down to 30 nm) whereas for LED-B is remains large (between 50 and 60 nm) and fairly independent of J. This feature is attributed to a reduced value of the internal electric field in the active region of LED-B as a consequence of thinner barrier layers (5 nm) compared to LED-A (10 nm), limiting the influence of its partial screening when increasing carrier densities in the QDs with the use of larger J. Then, as J is increased above 40 A·cm −2 and the carriers are mainly injected in Al y Ga 1−y N QDs with y close to the n.c. of 0.2, a strong reduction of the FWHM down to around 25 nm is observed. It should be noted that the minimum FWHM value of 24 nm, compared to 25.5 nm (for similar J) in the case of LED-A, can also be seen as a consequence of a smaller value of F int in the case of LED-B inducing a smaller inhomogeneous broadening of the QD EL emission [44].
Regarding the variation of the optical power as a function of the current density J, once again similarities and differences are observed between the two types of LEDs. Two or three different regimes are seen in Figure 8 from the variation of P opt as a function of J at a constant power m. This parameter m is related to the recombination mechanisms involved in LEDs [45]. At lower J, i.e., up to 40 A/cm 2 , a (quasi-)linear dependence (m = 1 ± 0.1) of P opt is observed for LED-A and LED-B. Such a variation indicates that the IQE of the LEDs is constant in this regime. At larger J, two different behaviors are obtained. A specific regime is seen in the case of LED-B with a super linear dependence (m > 1) of P opt between 40 and 80 A/cm 2 . It is attributed to an increase of the hole concentration in the active region due to a hole activation mechanism as the LED junction temperature increases with higher J. Above 80 A/cm 2 for LED-B and 40 A/cm 2 for LED-A, m is found below 1, i.e., the LED output power increases sub-linearly as a function of J. While in the case of LED-A, the saturation of P opt is rather limited with a value of 0.8, it is much stronger in the case of LED-B with a value of 0.2. Since this characteristic coincides with the redshift variation of the EL peak as a function of the injection current ( Figure 6), it is related to important thermal effects caused by low injection efficiencies and poor current spreading [32]. More insight in the LED performances can be obtained by considering the EQE of the LEDs presented in Figure 9. This value is the product of the IQE, the IE, and the EE, and since the EE is constant and independent of J, the variation of the EQE gives additional information on the current density dependence of the IQE and IE. Therefore, at lower J when the IQE has been found to be constant, the EQE increase in the case of LED-A is related to an increase of the IE whereas in the case of LED-B its decrease involves on the opposite a decrease of the IE which is maximum at the minimum J value of 0.1 A/cm 2 . The EQE of LED-A then reaches a plateau above 0.08% between 1 and 10 A/cm 2 and starts decreasing for higher J as a consequence of the presence of non-radiative mechanisms attributed to defect states localized in the cladding layers or band gap states in the QD active layers leading to a reduction of the IQE, and thermal effects at larger J leading to a decrease of the IE. For LED-B, an unconventional behavior is observed since after a first decrease of the EQE until 40 A/cm 2 , a sudden increase is observed between 40 and 80 A/cm 2 , which is the consequence of a higher IE as discussed above. Then, a strong reduction is found at larger J which is caused by a drop in the IE. Finally, a similar EQE value of 0.06% at 100 A/cm 2 is measured for both devices giving a maximum optical power of 225-250 µW.
The sublinear dependency of the light output at higher current densities indicates the presence of non-radiative mechanisms which are directly related to the high TDDs in the heterostructures and reduce the IQE. In fact, considering the high operating voltages of the LEDs (i.e., in the 10-20 V range) heat is certainly the main reason for the EQE decrease observed at 2-3 A/cm 2 , whereas the high TDD is the main origin of the overall low IQE. Therefore, using AlGaN/AlN heterostructures with lower TDDs should enable to further improve the IQE of QD-based LEDs. Concerning the active region design, the insertion of 5 QD planes together with a reduction of the capping layer thickness seem to be an interesting approach to favor carrier injection in the QDs as well as to increase the light output, as observed for LED-B (in the intermediate J regime between 40 and 80 A/cm 2 ). Finally, improvements are required for thermal management, in particular at larger J since a severe drop of the IE has been observed above 80 A/cm 2 . Indeed, considering average values for the different efficiencies in our devices, i.e., with an IQE of 20%, an EE of 8% (estimated from the extraction cone at the sapphire/air interface) and an EQE of 0.05-0.1%, it is then possible to estimate the IE which is in the range of 3 to 6% only. Inefficient hole injection due to low hole concentration and mobility in high Al concentration Al x Ga 1−x N layers leads to electron leakage and overflow out of the active region and will require different layer designs such as tunnel junctions which have been shown as an attractive solution to improve the IE and avoid the use of an absorbing p-type GaN contact [46,47].

Conclusions
Al 0.2 Ga 0.8 N QD-based LEDs emitting in the UVB range have been fabricated by MBE using two different designs for the active region: 3 QD planes separated by 10-nm thick barrier layers (LED-A) and 5 QD planes separated by 5-nm thick barrier layers (LED-B). Despite TDDs above 10 −10 cm −2 , IQE values around 5-10% have been estimated on reference samples as a consequence of the 3D carrier confinement in the QDs. Electro-optical characterizations have pointed out at some differences in the LED performances: (1) Higher EQE below 1A/cm 2 and above 40 A/cm 2 , an emission at shorter wavelengths and an increase in the carrier injection efficiency in the QDs as well as in the optical power (between 40 and 80 A/cm 2 ) were observed for LED-B compared to LED-A; (2) higher EQE between 5 and 40 A/cm 2 and a reduction of the EL FWHM between 0.1 and 40 A/cm 2 were found for LED-A. At current densities above 80 A/cm 2 , similar characteristics were found with EQE around 0.06% and optical power in the 0.2-0.25 mW range. Further improvements in the IE together with a reduction of the TDDs are suggested for future research directions to reach higher performances. Funding: This research was funded by ANR Project (ANR-14-CE26-0025) "NANOGANUV" and the support from GANEX (ANR-11-LABX-0014) is also acknowledged. GANEX belongs to the publicly funded "Investissements d'Avenir" program managed by the French ANR agency.