1. Introduction

The family of light-emitting diodes (LED) has been expanding for decades, as its latest member — the deep ultraviolet (DUV) LED based on the AlInGaN system, with wavelength shorter than 300 nm — started gaining interests in recent years [1]. The ultraviolet (UV) light with peak wavelength at about 275 nm has been proved perfectly efficient for water sterilization, of which is in strong demand in some regions of the world [2]. The traditional UV source is the fragile, potentially toxic mercury vapor lamp, thus, about to be updated. Nonetheless the UV-LED, just like all sorts of new-born LEDs, faces two typical problems -low efficiency and short life-span. All these are the consequence of the immature epitaxial engineering. The luminous efficiency can be measured by the external quantum efficiency (EQE), defined as the internal quantum efficiency (IQE) times the light extraction efficiency. High point defect (PD) and dislocation density have detrimental effects on the IQE [1,3]. Karpov et. al. demonstrated via the first principle calculation that holes tended to be captured by dislocations cores where they recombined with electrons non-radiatively [4]. The efficiency droop effect, commonly found in visible light LEDs (vis-LEDs) [5, 6], is also severe in UV-LEDs. Unlike vis-LEDs of which the droop effect has multiple origins, the efficiency droop in UV-LED is largely due to the low injection rate of hole from the p-region. As has reported by Kneissl et. al., a by-product of higher Al composition in the p-region is the lift in the activation energy of Mg ions, which leads to a lower hole concentration and, as a consequence, a strong impairment in electron and hole at high current injection [7]. In term of long-term reliability, the performance of UV-LEDs is also inferior when compared to vis-LEDs. In our aging test for commercial blue LEDs, the EQE dropped by only 4% after a 3224-hour ultrahigh-current aging [8]. While for UV-LEDs, in general, the shorter the wavelength is, the faster their performance degrades [3]. In the work reported by Fujioka et. al., the 310-nm LED barely maintained the efficiency at working current and temperature after 1000 hours, while the efficiency of the 255-nm counterpart dropped to 60% even at lower current [9]. In addition to those problems related to efficiency, the room-temperature (RT) and low-temperature (LT) electroluminescence (EL) spectra of UV-LED are often observed to be accompanied with parasitic peaks which for vis-LEDs is rare and only occurs at LT [3, 10–12]. The shapes of the parasitic peaks so-far reported vary over literature, they nonetheless share some common features -one or more parasitic peaks extend from the low-energy side of main peak to the blue [11], or even the green spectral region [12], resulting in bluish or pure white light. A peak that resides very close to the high-energy side of main peak are occasionally observed, for example, by Santi et. al. [10] and our group. The precise origins of these parasitic peaks remain unclear but are commonly accepted to be associated to the radiative recombination via deep level defects, such as Ga vacancies in the active region [13], and Mg-dopants in the p-region [14]. Comprehensively, this defect-assisted radiative recombination is undesirable since it consumes the carriers through low-efficient emission at unwanted wavelengths. Therefore, it has attracted great interests. On one hand, meticulous investigations on the energy band of the radiative deep-level defect, as well as their perturbations on carrier dynamics, are required before they can be eliminated completely. On the other hand, since the energy levels corresponding to these parasitic peaks are similar to those of non-radiative recombination centers, the radiation spectra from deep level defects can be utilized for the evaluation of defect evolution during aging. Meneghini’s group monitored the increase in intensity of the green parasitic peak over the 256-hour aging and demonstrated the defect proliferation as the reason of luminous degradation of main peak [3]. Note that the radiative defects are of various types, dwelling at different energy levels, and with different activation energy. A powerful tool for analyzing the morphology and the evolution of defects is the full-range temperature measurement. That is to measure the optical and electrical performance of the sample that is subjected to a wide temperature range, commonly from cryogenic to room temperature or even higher ones. Plus, at each temperature point, the parameters are to be captured at various injection levels. Although the UV-LED samples may not be operating at such extreme conditions, e.g. 25 K, in real-life cases, measuring the LT emission at various current injection levels is still necessary. For example, a majority of deep-level emission from defects only occur at LT and low injections. The spectra of these emission, combining with the injection density at which they are subjected, reveals the physical properties of these defects [15, 16]. In addition, carrying out the measurement at a series of stages during aging tests will provide further information on the evolution of defects. Hitherto there have been few studies on the cryogenic measurement on DUV-LED [10], and no reports has been found on performing cryogenic measurement during aging. In literature, like the work done by the group of Meneghini [3, 17], the electrical and optical properties have been investigated mostly under room temperature over a long-term current stress. In the work of Santi, etc. [10], the authors introduced low-T measurement and numerical simulation into the investigation of the similar DUV-LED samples, and identified the origins of some typical parasitic peaks, which also appear in our work. On the basis of these work, we further introduced the tool of EQE-I curve for the aging analysis, and extended low limit of the temperature range to 20 K.

In this article, we subject several 275-nm DUV-LEDs to a 15-hour, high-current stress aging, during which the EL spectra, EQE-I curves, I-V curves are monitored under a broad temperature range spanning from 20 to 300 K, and wide injection ranges as well. After a systematic evaluation on the experimental results, especially the correlation between the parasitic peaks and the EQE-I curves, as well as the evolution of this correlation during the aging, we are able to present a detailed description on the carrier-lattice interplay and the possible origin of luminous degradation.

2. Experimental setup

The epitaxial structure of the 275-nm samples under investigation is similar to those employed in our previous work [18]. The chip, 1 mm × 1 mm in size, is encapsulated beneath an inorganic glass lenses, of which the degradation can be ignored. All the samples are experiencing a short-term aging of 15 hours, under 500 mA (Injection density of 50 A/cm²) at 300 K, and the optical and electrical properties of them are measured at 0h, 1h, 2h, 5h, 10h, and 15h. The measurement is conducted under both ambient and cryogenic temperatures. In the ambient measurement, the samples are mounted on a temperature-controlled heat sink system which contains a LED-850 (Instrument System, Germany), and a Keithley 2510 (Keithley, USA). The temperature is set at 300 K, and the EL spectra are captured by a spectrometer of Ocean Optics QE65pro, under a series of currents from 0.3 mA to 500 mA (0.03 A/cm² to 50 A/cm²) provided by the Keithley 2400 (Keithley, USA) which is also used to measure the I-V curves. In the cryogenic measurement, the samples are put into a vacuum chamber, of which the temperature is controlled by a closed-cycle helium refrigerator system. Under a series temperature from 20 K to 300 K with a step of 20 K, the similar measurement are conducted except the maximum current is only 20 mA (2 A/cm²). The reason is that, once the the current exceeds 20 mA, the temperature controller is not sufficiently powerful to remove the heat produced by the sample and thus fails to keep a set temperature. As will be mentioned latter, the EL spectra contain a bunch of parasitic peaks, which would lead to errors when calculating the EQE. We employed an asymmetry-exponential function, as shown in Eq. 1 to extra the main peak by curve-fitting.

In Eq. 1, the P(λ) and λ denote the intensity and wavelength, respectively. Other parameters are to be determined by the fitting. The R-squires exceed 0.99, which ensures the accuracy. The EQE (ζ) is then calculated out of the P(λ) and the forward current I_f, as shown in Eq. 2.

In Eq. 2, the e, h, and c denote the elementary charge, Planck constant, and light speed in vacuum, respectively. all the parameters are rendered in SI units. All the intensities of EL spectra are calibrated by re-measuring the sample in an integrating sphere (ISP 500, Instrument System, Germany) with a Spectro 320 (Instrument System, Germany), yielding absolute values.

3. Results and discussion

3.1. Analysis on data of the pristine sample

We start from the EQE-I curves and EL spectra, which are known to contain abundant information of the intrinsic properties of the device [19, 20]. The EQE and forward voltage at three temperatures are illustrated in Fig. 1, respectively. The data of 20 K [Fig. 1(a)] are captured at forward current ranging from 0.3 to 7 mA (0.03 A/cm² to 0.7 A/cm²). The data beyond 11.5 V is omitted since that the voltage reaches the breakdown voltage of the anti-parallel Zener diode. The EQE reaches the maximum at only 1 mA then starts to droop. As the temperature increases, the EQE onset current increases, while the EQE maximun decreases in general, as illustrated in Figs. 1(a)-1(c). Compared with the LT counterparts, the EQE-I curves and V-I curves measured at 300 K exhibit conspicuous deviations as illustrated in Fig. 1(c). First, EQE maximum drops from 1.19% at 20 K to 0.61% at 300 K. Second, the maximum-EQE current locates at 100 mA at RT, much higher than that at 20 K. Third, the forward voltage at which the EQE reaches maximum falls from 8.93 V at 20 K to 7.14 V at 300 K. In fact, the forward voltage falls with the increasing temperature because of the thermal activation. The EL spectra captured under the certain driving conditions at the same temperatures of Fig. 1 are illustrated in Fig. 2, which show a main peak (or band-edge peak) located at 275 nm. By a careful inspection, we distinguish three parasitic peaks located at 260 nm (peak 1#), 310 nm (peak 2#), and 400 nm (peak 3#), respectively. These peaks are similar to those reported by Meneghini’s group [10]. In this work, they considered the peak 2# as the result of radiative recombination via defects. In this work, we investigate the peak 3# in detail. In a LED, the carrier injection would split the unified Fermi level into two quasi-Fermi levels of the hole and electrons, between which the difference contributes largely to the forward voltage at this injection density. Therefore, at temperature around RT, at which the carriers are activated, the difference of quasi-Fermi levels (ΔE_f) can be estimated from the value of forward voltage. With this in mind, in Fig. 3(a), we plot the EL spectra captured at 0.3 mA (V_f = 4.20 V), 0.5 mA (V_f = 4.34 V), and 1 mA (V_f = 4.53 V), all at 280 K. At the 0.3-mA case, as the forward voltage is smaller than the bandgap of 4.5 eV (roughly estimated from the main peak), the quasi-Fermi levels lie within the bandgap and induce no band-edge emissions, as expected. Whereas, the EL spectrum contains a relatively strong peak 3# (400 nm) and a weak peak 2# (310 nm), which indicates an origin of deep-level defect recombination of these two peaks. In the 0.5-mA spectrum, as ΔE_f approaches band-gap, the 275-nm band-edge peak emerges with the enhanced peak 2#. As current increases to 1 mA, the band-edge emission surges while the increase in two parasitic peaks stalls. All these phenomena are in consistent with the nature of typical deep-level defect recombination. The defect-related recombination is commonly non-radiative, known as the Shockley-Read-Hall (SRH) recombination. In some UV-LEDs, there nonetheless exist some radiative deep levels (RDLs) which originates from the point defect and can cause the yellow emissions [12, 13, 21]. In the sample adopted in this work, there exist two types of RDLs, which are sketched in Fig. 4, as well as their interplays with carriers under different injections levels. As illustrated in Fig. 4(a), at the very low injection level, a majority of holes are concentrated on the last quantum well adjacent to the p-AlGaN, splitting the local Fermi level into two quasi-Fermi levels of holes and electrons, respectively. Due to the large hole concentration, the hole quasi-Fermi level lies in the valence band, while that of electrons in midst of the band-gap, close to the energy level of 400-nm RDLs, which are therefore filled with electrons and can be recombined with holes. This recombination corresponds to the spectrum with sole peak 3# at 0.3 mA in Fig. 3(a). At a slightly higher injection density, as sketched in Fig. 4(b), the separation between the two quasi Fermi levels becomes larger that the electron’s approaches energy level of the 310-nm RDLs, thus both these two sorts of RDLs are excited, corresponding to the 0.5-mA spectrum in Fig. 3(a) with two peaks at 310 nm (peak 2#) and 400 nm (peak 3#) respectively. As the forward current increases to 1 mA to raise the forward voltage to the level higher than the band-gap, the E_f, as illustrated in Fig. 4(c), surpasses the band-edge, resulting in strong band-edge emission, while the intensities of two RDL emissions have been barely changed. Though disappearing at the moderate current, the peak 3# reemerges at higher currents which, according to the EQE-I curve, are highly correlated with the EQE droop onset. At each temperature, once the EQE reaches the maximum and starts to droop as current increases, the peak 3# becomes apparent. Though the critical current, at which the peak 3# emerges and the EQE start to droop, varies with temperature, this correlation persists at each temperature.

 

Fig. 1 EQE-I curves and V-I curves for the pristine sample measured under (a) 20 K, (b) 80 K, (c) 120 K, and (c) 300 K, respectively. The values of voltage are recorded during capturing ELs under each current.

Download Full Size | PDF

 

Fig. 2 EL spectra of the pristine sample, captured under a series forward current under (a) 20 K, (b) 80 K, and (c) 300 K, respectively.

Download Full Size | PDF

 

Fig. 3 (a) EL spectra captured at three low forward currents at 280 K and the inset demonstrates the normalized version; (b) EQE-I curves at 180 K and 160 K; EL spectra captured at a series of forward currents (c) at 180 K and (d) at 160 K.

Download Full Size | PDF

The close correlation between the EQE droop and reemerge of peak 3# can be explained as follows. The peak 3# has two distinct sources of RDL, one located in the band-gap of active region (QWs) and other in that of the p-region. The former emits at very low current when the quasi-Fermi level approaches the energy level of related recombination centers. Therefore, the emission would commence at very low current, and as the current increases, the main peak surges, of which the emission would emerge the weak peak 3#, making it seem as if disappearing. On the other hand, to induce the emission of the RDLs in the p-region, the electrons should pass the EBL, entering the p-region, which requires the high quasi-Fermi level at high currents near EQE droop onset. As sketched in Fig. 4(d), once the electrons enter the p-region, they would recombine with holes there via defect levels, which emit light similar to peak 3# and thus reinforce it. This is why the peak 3# reemerges. It is therefore clear that both the EQE droop and the reemerge of peak 3# are caused by the electron overflow, which we consider is what underlies the close correlation between the two. It is the quasi-Fermi level that directs the electron distribution. Note that the relation between quasi-Fermi level and the injection current depends strongly on the temperature — the quasi-Fermi level increases as the temperature decreases at same injection current. In consequence, at low temperature, the quasi-Fermi level would be higher than the defect level in the QWs, even at low-limit of our current. Those peaks 3# illustrated in Figs. 2(a) and 2(b) at temperatures lower than 80 K are all associated to the RDLs in the p-region. The ELs all exhibit a weak peak 3# at the lowest current which, according to the related EQE-I curves in Fig. 1, is close to the EQE droop onset current that has induced electron overflow. As the current rises and surpasses the EQE droop onset, the relative intensities of peaks 3# in each figure increases. While at RT (300 K), the peak 3# associated to RDLs in active region would be quenched by the thermal activation. Thus, under the injection condition in this work, the low-current peak 3# and its disappearing could only be observed at the narrow temperature window, which, in this work, ranges around 160 to 180 K. As illustrated by the EQE-I curves in Fig. 3(b) which show that the EQE droop onset is at 13 mA (180 K) and 7 mA (160 K), we observe both the low-and high-current peak 3# at 180 K and 160 K and illustrate the EL spectra in Figs. 3(c) and 3(d), respectively. We believe that at lower temperature, if one can exert even lower current to draw down the quasi-Fermi level, it would be possible to observe the low-current peak 3#.

 

Fig. 4 Sketches of carrier dynamics under different injection levels. (a) Case of very low current injections with quasi Fermi level of electron lying on the 400-nm defect level; (b) Case of low current injections with quasi Fermi level of electron lying on the 310-nm defect level; these two are both off-band. (c) Case of over-band injection levels, with strong band-edge emissions; (d) Case of electron overflow, which to induce p-GaN defect emission as well as EQE droop.

Download Full Size | PDF

Also should be emphasized is the fact that the current range of each stage in Fig. 4 varies significantly with temperature. Taking the data measured under 280 K for example, the stages in Figs. 4(a) and 4(b), where the parasitic peaks 2# and 3# emerge with the absence of main peak, occur around 0.3 – 0.5 mA, as suggested by the measured ELs in Fig. 3(a). The stage of Fig. 4(c) corresponds to the spectra measured at 1 mA, which suggest that this stage occurs at current range higher than 1 mA and below the EQE droop onset. The stage in Fig. 4(d) where carriers overflow and EQE starts to droop will occur beyond the current limit of our low-T measurement. This stage can be observed for data measured at lower temperatures, as the onset of EQE droop increases from ~1 mA at 20 K, to ~13 mA at 180 K, and to ~100 mA at 300 K. As a consequence, the current range varies drastically with temperature.

We also investigate the temperature dependence of the other parasitic peaks. The peak 2# spans a wide spectral range across the whole UV-A. To extract it from the whole EL, we perform the multiple-peak Gaussian fitting by employing an empirical two-exponential-asymmetrical function to simulate the band-edge peak and the peak 2#. We pick the EL spectra measured at 5 mA, and at various temperature ranging from 20 K to 300 K for the fitting. The reason to select 5 mA is that this current should be a typical one that it excites all four peaks in all temperature range, and does not induce EQE droop at the lowest temperature of 20 K, since the electron overflows would complicate the carrier transition. The peaks 1# and 3# are three orders of magnitude lower than the band-edge peak, thus cannot be caught by the algorithm. We remove the main peak from the measured whole EL spectra by using the fitted band-edge peak, and the residual are technically the collection of parasitic peaks. However, due to the low intensity, the data from the short-wavelength side of the band-edge peak do not clearly show the peak 1# as expected, thus has been omitted. The residual spectra at long-wavelength side with its normalized version are respectively plotted in Figs. 5(a) and 5(b), which contain mainly the peak 2#, and peak 3# for some LT spectra, at which temperature the 5 mA reside on the EQE-droop region. As clearly illustrated by the plot of integrated intensity vs. temperature in Fig. 6(a), the main peak exhibits a near monotonic decrease with increasing temperature, except for a slight increase between 140 K and 200 K at which a plateau is observed. Although the peak 2# also shows monotonic decreases in intensity with the increasing temperature, it slows down at 140 K but speeds up at beyond 160 K. The coincide can be illustrated more clearly in the temperature dependent ratio between the peak 2# and main peak, as plotted in Fig. 6(b). The ratio increases at low temperature and then decreases after exhibiting a maximum at 140 K, which also applies to spectra captured at other currents. The general decreases of these two peaks are known as a result of thermal activation of carriers, but the intensity correlation between the two peaks can be associated with something beyond, as explained below. Except for the already-saturated 400-nm defects levels, there are two main paths leading to radiative recombination for an electron-hole pair ejected into the active region. Beside the strongest band-edge recombination, it could also be transited to the 310-nm deep-level defect, contributing to peak 2#. The energy structure of such deep-level RC can be revealed from the spectral shape of peak 2#. It barely changes over temperature. A few deviation at high temperature is rather due to the fitting error for low intensity than a real physical shift. As suggested by forward voltage that all larger than 5 V, all the ΔE_f have exceeded the band-edge. The radiative defect levels are therefore all uniformly occupied by carriers. As a result, the Gaussian-like spectral shape reveals that the defect states between band-gap follows the normal distribution, with the density peaks at 4.06 eV (325 nm), 443 meV lower than that of band-edge (4.50 eV). As the temperature increases from 20 K, the proportion of carriers participating in the RLDs out of all radiatively recombined carriers increases. The ratio peak at 140 K marks the maximum efficiency of carrier transition from band-edge to defect RCs. Beyond 140 K, the enhanced thermal activation facilitates the carriers to be captured by band-edge rather than by RDLs, thus the intensity of band-edge emission retains while that of defect emission fast decreases. In Fig. 7 of Santi’s work [10], the normalized intensity of the 300-nm peaks are also plotted upon temperature ranging from 100 K to 400 K. A monotonic decrease with the increasing temperature has been identified, which the authors attribute to the affect of hole injection. The behavior is roughly in consistence with those in the temperature region of 140 K to 300 K. As the lower temperature region has been explored in this work, our results complement to the previous one.

 

Fig. 5 (a) EL spectra captured at a series of temperatures, 5 mA, with the band-edge peak eliminated; (b) The normalized versions. The spectral regions of all ELs below 280 nm suffer intense vibrations due to errors in subtractions, thus, have been removed.

Download Full Size | PDF

 

Fig. 6 (a) Changing in the integrated intensity of peak 2# (left y-axis) and band-edge peak (right y-axis), both at 5 mA, upon temperature; (b) Changing in the ratio of intensity of peak 2# and band-edge peak at 5 mA upon temperature.

Download Full Size | PDF

In contrast to the peaks 2# and 3# that stem from the RDL recombination, the peak 1# resides on the high-energy side of band-edge peak, indicating its distinct origin. It can be observed in all temperature range, at most current density as long as the E_f reaches the energy level corresponding the peak 1#. Although increases with current, it’s ratio to the band-edge peak is current independent, unlike peak 3#. We suggest it may be caused by band-edge inhomogeneity, such as well-width fluctuation and/or Al content concentration.

As a summary to this subsection, we demonstrate four main radiative recombination paths, including the band edge emission (main peak), two types of RDLs (peak 2# and peak 3#) and one over-and edge-emission (peak 1#).

3.2. Analysis on the short-term degradation

All the above-mentioned properties are shifting during the 15-hour aging. Before discussing the optical properties, we first present the electrical ones. The I-V curves captured at 300 K and 20 K are illustrated in Figs. 7(a) and 7(b), respectively. We could divide the forward region into three sub-regions. For the 300 K data, the first is the cutoff region ranging from 0 to 1.40 V (0 V to 3.50 V for 20 K), of which the current merges in the background noise. The second one, ranging from 1.40 V to 3.95 V (3.50 V to 5.05 V for 20 K), exhibits a monotonic increase in current with aging, as plotted with black lines in Figs. 8(a) and 8(b), respectively. The third region, ranging beyond 3.95 V (beyond 5.05 V for 20 K), in which a decrease in current with aging is observed, as also plotted with red lines in Figs. 8(a) and 8(b). Considering the relation between ΔE_f and V_f, in the second region the ΔE_f lies within the bandgap among the deep-level defect. Therefore, the leakage current in this region are associated with the deep-level, which provide routes for recombination. As seen from the extremely weak intensity of peak 3#, these sorts of recombination are mostly non-radiative. A proliferation in the non-radiative deep-level defects, being fast in the initial 1 h and slowing down thereafter, is suggested in the plot of leakage current vs. aging. Such leakage current has been associated with the defect growing along the dislocations [22]. In the third region, as the ΔE_f rises beyond the band-gap, the equivalent resistance of LED greatly decreases. The decrease in current is thought to be ascribed to extrinsic sources, such as the Ohmic contact degradation.

 

Fig. 7 I-V curves measured over the whole aging at (a) 300 K and (b) 20 K, respectively.

Download Full Size | PDF

 

Fig. 8 Currents with aging (a) in region II (2V) and region III (5V) at 300 K, and (b) in region II (3.8V) and region III (11V) at 20 K.

Download Full Size | PDF

We then discuss the EQE-I curves and EL spectra. The EQE-I curves of 20 K and 300 K are plotted in Figs. 9(a) and 9(b), respectively, from which one could discern that some common features are shared. a) The EQE decreases dramatically, especially at the initial 1 hour, similar to the work in literature [17]; b) The current of EQE droop onset moves toward higher current, which is in consistent with green LEDs [23], but in contrast to blue ones [8]. We employed the ABC model for fitting the EQE-I curve at RT, in which the A coefficient is associated with the SRH non-radiative recombination [24–26]. The plot of A coefficient vs. aging time is illustrated in the inset of Fig. 9(b). The increasing A coefficient indicates the growth in the SRH nonradiative recombination rate. Meanwhile, the intensity of parasitic peaks is decreasing along with the band-edge emission. We choose the data captured at 20 K, 5 mA for analysis, which contains all four peaks. As illustrated in Fig. 10(a), the EL spectra decreases as a whole with aging time. The normalized spectra are illustrated in Fig. 10(b), which indicate the spectral shape are being modified by aging. Since the origin of peak 1# remains unclear, we concentrate on other parasitic peaks. As illustrated in Fig. 10(b), the peak 3#, in term of relative decay, is the more susceptible to the aging than peak 2#. The ratios between peaks 2# and 3# to band-edge peak respectively are all plotted upon aging time in Fig. 11(a). The 2# ratio (black line) first decreases slightly before 1 h, then increases monotonously with aging; The 3# ratio (red line) experiences a monotonic decrease. As mentioned above, the peak 3# stems from the p-region, and there are many processes can cause its decrease, such as reduction in the electron entering the p-region, and healing of the defect level by injection current. The former can be further dissolved into two sort of processes, i.e. the Mg activation that leads to increasing in hole injection; Reduction in electrons concentrations. Considering the increasing current in the region III of the I-V curve, we suggest that the Mg activation is unlikely. Were it true, the current should have increased, instead of the observed decrease. In addition, there is no clear evidence on the defect healing by current injection. On the other hand, the reduction in electron density is highly possible. The growth of the non-radiative centers (NRCs) in the active region leads to more electrons being trapped in the active region before they could enter the p-region. The peak 2# originates from the RDLs in the active region, where the defect related to low-current peak 3# and the NRCs are also accommodates. The low-current peak 3# also decreases during aging. The EL spectra at 0.3 mA, 260 K for 0 h and 15 h are illustrated in Fig. 11(b). Not only does the intensity decreases, but also the peak shifts significantly towards lower energy. It indicates that the current injection has modified the defect structure and therefore changed the energy levels. It is reasonable to conjecture that such modification could also occurs for some NRCs, resulting in its concentration growth. In addition, the NRCs and the RDLs share similar energy levels, possible couplings and interactions are expected when carriers are transited from band-edge. The increase in the NRC density would exert some influence on the carrier capture rate of RDLs, which we believe is the origin of the changing in intensity ratio of peak 2#. The quick decay within the initial 1 hour could be regarded to the fact that increasing NRC trapping more carriers that otherwise may be captured by RDLs. A detailed mechanism of this interaction is yet to be discovered.

 

Fig. 9 EQE-I curves measured over the whole aging at (a) 300 K and (b) 20 K. Inset in (b): the changing in A-coefficient upon the aging time.

Download Full Size | PDF

 

Fig. 10 (a) The EL spectra at 5 mA, measured over the whole aging at 20 K and (b) the normalized ones.

Download Full Size | PDF

 

Fig. 11 (a) the ratios of intensities between peak2# and band-edge peak (black, round) and between peak 3# and band-edge peak (red, rectangular); (b) The EL spectra captured at same condition (0.3 mA, 260 K) before (black) and after (red) the aging.

Download Full Size | PDF

4. Conclusion

In this work, we perform a thorough investigation the short-term degradation behavior of DUV-LEDs. By measuring the optical and electrical properties under a vast range of temperature before and during the aging, we gain a more detailed view on the defect distribution, carrier dynamics and reasons that jeopardize the efficiency and reliability. The following are conclusions we have come to. a) The 400-nm parasitic peak are associated with radiative recombination via deep-level defects that exist both in the active region and the p-region, as revealed by the EL spectra at LT, which show the 400-nm peaks at low current and high-current side. The latter coincides with the current at which the EQE starts droop. b) The 310-nm parasitic peak belongs to the emission of radiative recombination from another sort of deep level defects, which exists only in the active region, with energy level 400 meV lower than band-edge. c) The culprit of luminous decay is the proliferation of NRCs in the active region, which would capture most of carriers that otherwise may joint radiative recombination on band-edge or RCs related to 310-nm peak. One byproduct is that it reduces the number of excessive electrons entering the p-region, thus also mitigate the intensity of 400-nm peak originating from the p-region. There is no strong evidence on the Mg activation on p-region.

In term of methodology, this work shows the advantage of combining full-range temperature and injection measurement with the short-term aging, which is to create environments with drastically different quasi-Fermi levels and the degrees of thermal activations, where the nature of deep-level defects can be well inspected through their deviations in optical behaviors with the changing conditions. There have been two examples in this work. First, the oberservation of low-current parasitic peaks 2# and 3# in prior to the main peak, which requires not only a low current, but also, more importantly, a proper temperature that should be neither too low to let the quasi-Fermi level merge the defect level, nor too high to quench the already weak radiation. After a careful investigation we found that the temperature window resides on 260 to 280 K. Second, the identity of the two-sourced peak 3#, which show up at both low and high current regions. All these can only been discovered and investigated by the measurement under both wide ranges of tmeperature and current. As a powerful tool in the evaluation of devices properties, this method deserves further development and should be applied to other optoelectronics devices.