Degradation of AlGaN-based UV-C SQW LEDs analyzed by means of capacitance deep-level transient spectroscopy and numerical simulations

The Sars-CoV-2 pandemic has changed the world's view of health and safety. The disinfection of surfaces of medical equipment, the clinical environment, mobile devices, public spaces, and even garments has become more and more important. As a consequence, the development of efficient, compact, and low-cost UV irradiation systems for disinfection would enable alternatives to existing hygiene protocols. In this sense, UV LEDs have proven to be the right candidates for this purpose.^1–4 However, these devices still present reliability issues that need to be investigated to improve their long-term lifetime.^5–7 Typical degradation processes include p-side contact degradation, decrease in carrier injection efficiency, increase in series resistance, changes in spectral characteristics, and generation of defects within the active region of the device.^8–12 Most degradation processes are related to the presence and/or generation of point defects that cannot be easily identified.

To give a contribution to this field of research, we present an analysis of the physical processes responsible for UV LED degradation based on combined deep-level transient spectroscopy (C-DLTS), electro-optical characterization, and simulation. Remarkably, C-DLTS allowed us to detect the presence of two Mg-related traps, one possibly located at the n-side, and the other located closer to the p-side. Moreover, after 1000 min of stress, an additional third trap could be identified and ascribed to point defects generated by ageing. The analysis of the electrical characteristics indicates that constant current stress induced (i) an increase in the shunt current component, (ii) an increase in trap-assisted tunneling (TAT)-related conduction processes, (iii) a variation in the series resistance of the devices, and (iv) an increase in turn-on voltage. This last process was correlated with the passivation of Mg close to the p-contact, as confirmed by TCAD simulation of the forward I–V curves of the devices. Finally, the degradation of the optical characteristics is ascribed to the decrease in injection efficiency in the active region at high current levels and to the increase in non-radiative recombination at low current levels.

We analyzed SQW ultraviolet light-emitting diodes (UV LEDs) with a nominal emission wavelength of 265 nm. The devices were grown by metal–organic vapor phase epitaxy (MOVPE) with the epitaxial structure reported in Fig. 1(a). UV-LEDs are grown on high temperature annealed (HTA) epitaxially laterally overgrown (ELO) AlN on sapphire with a threading dislocation density of 9 × 10⁸ cm⁻² AlN/sapphire template.^13,14 On top of the template, the following layers are grown: a 400 nm AlN layer, a 1000 nm silicon doped Al_0.76Ga_0.24N current spreading layer graded in the last 100 nm to an Al mole fraction of 65%,¹⁵ and a n-contact layer of 200 nm, composed of Si-doped Al_0.65Ga_0.35N (N_d = 4 × 10¹⁸ cm⁻³). Then, the structure features a 40 nm first barrier layer in Al_0.63Ga_0.37N ([Si]: 5 × 10¹⁸ cm⁻³), a 1.4 nm Al_0.48Ga_0.52N SQW, a 10 nm Al_0.63Ga_0.37N barrier, referred to as last barrier, and a 10 nm Al_0.83Ga_0.17N Interlayer (IL), which are non-intentionally doped. The p-side is composed of a 25 nm Al_0.75Ga_0.25N electron blocking layer (EBL) (Mg: 1 × 19 cm⁻³), above which a 250 nm GaN p-contact layer is finally grown (Mg: 6 × 10¹⁹ cm⁻³). The analyzed structure is simpler with respect to a conventional commercial UV LED, in order to simplify the interpretation of the results without involving the effect of multiple QWs, superlattice EBL, a current-spreading layer, and an optimized underlayer to block the growth of defects toward the active region. At the same time, the main degradation mechanisms affecting more complete LED structures should be similar, even if they may be accelerated/slowed down due to the variation in the magnitude of specific driving forces, e.g., different operating temperatures due to a less optimized driving voltage, increased recombination-enhanced degradation due to the reduction of the active volume, or a more strongly localized degradation due to the absence of a current spreading layer.

The structure reported in Fig. 1 was simulated with TCAD Synopsys Sentaurus,¹⁶ to analyze the theoretical carrier distribution, the band diagram, and the junction electric field; such results are of interest to provide a more accurate interpretation of the defect spectroscopy analysis that will be presented later in the paper. In particular, in Fig. 1(b), we show the profile of the electric field (normalized to its peak in the interlayer IL, to allow an easy comparison). By combining these results with the simulated electron carrier density (not shown for brevity), we can conclude that by changing the applied bias down to −2 V, we explore with the boundary of the space charge region (SCR) from within the QW, down to the interface with the first barrier. With higher negative bias values, the boundary is swept through the first barrier as well. Finally, we note that at the p-side, at all voltage levels, we constantly modulate the charge located in the EBL. This phenomenon is important to understand the results of C-DLTS measurements since it indicates that with those measurements, we are investigating simultaneously both sides of the junction. This is because the junction is not unilateral, i.e., the doping levels on the two sides are of comparable magnitude.

To induce a sufficient amount of degradation in a reasonable amount of time, a device was aged under constant current stress of 100 mA (corresponding to 100 A cm⁻²), at the baseplate temperature of 40 °C for 12 100 min (about 200 h). During the test, electrical and optical characteristics were monitored at logarithmic time steps. Moreover, based on the previous stress experiments performed under similar conditions, we decided to carry out capacitance–voltage (C–V) and C-DLTS measurements at specific levels of optical degradation, i.e., when the optical power (OP) reached 100%, 92%, 85%, and 75% of its initial value.

In Fig. 2(a), we report the electrical characteristics measured during the stress test. Three main degradation processes can be observed: (i) an increase in the shunt current at low forward and reverse bias (V < 3 V), (ii) an increase in the magnitude of defect-assisted conduction mechanisms at voltages lower than the turn-on voltage (2.5 V < V < 4.5 V), (iii) an increase in series resistance at high current levels (V > 5 V), and (iv) an increase in the ideality factor of the diode, calculated from the exponential fit at high current levels. We ascribe the decrease in shunt resistance R_p to the generation of parasitic conduction paths probably located at the mesa edges of the device.^17–19 The second process (ii) is ascribed to an increase in the concentration of defects that contribute to trap-assisted tunneling (TAT); in particular, according to the results presented by Roccato et al.,²⁰ the observed current increase can be modeled by considering a generation of deep traps in the interlayer placed between the EBL and the active region. The increase in series resistance (iii) can be attributed to the increase in resistivity of the p- and n-sides induced by stress.²¹ Finally, the increase in ideality factor reported in Fig. 2(b), correlated with the variation of the turn-on voltage, suggests the degradation of carrier transport processes through heterobarriers.²² In particular, we modeled it as due to the degradation of first nanometers of p-contact, caused by the passivation of Mg and the consequent increase in Schottky contact barrier. This topic will be covered more in-depth in the following paragraphs of this paper.

By analyzing the optical characteristics measured in continuous wave (CW) during ageing reported in Fig. 3(a), we can observe a global decrease in optical power, which is more prominent at low current levels. The observed variations can be due to the combined effect of a decrease in injection efficiency (impacting both the high and low current regimes) and to an increase in the non-radiative recombination rate, impacting mostly the low current region of the optical power curves.

The decrease in injection efficiency can be caused by the redistribution/pileup of charged defects within or near the active region,¹² with the consequent generation of a potential barrier for carriers.^12,23 This is also confirmed by the increase in the EQE peak current,²⁴ not reported for brevity, which implies that a higher current is needed to reach the same radiative recombination rate, i.e., a similar carrier density.

The increased non-radiative recombination is ascribed to the increase in the concentration of non-radiative recombination centers (NRRCs) induced by stress:^25,26 this is partly confirmed by the increase in the slope of the L–I curve calculated from 1 and 4 mA, which increases from about 1 to 2.^27,28

To explain these two processes, we considered the interpretation proposed in Ref. 12 for the optical degradation of UV-C LEDs, which takes into consideration the generation of defects in the active region of the device. These defects are supposed to act simultaneously as NRRCs and as charged centers that impede the injection of carriers into the QW, thus leading to a decrease in injection efficiency.

To assess the stress-induced deep-levels, we carried out C-DLTS measurements before and during stress. This technique allows to easily investigate defects located within the SCR.³⁶ The measurements were done in the range: V_meas = –2 to V_fill = 0 V, with a filling time of t_fill = 100 μs. Based on Fig. 1(b) and on the previously done consideration, the device region explored with such parameters should correspond to QW, the last barrier, and EBL.

As depicted in Fig. 4(a), from C-DLTS measurements, we found the presence of two peaks in the rate windows plot, indicating the possible presence of two distinct traps within the SCR of the device. The primary peak, centered around T = 225 K in Fig. 4, is present during the entire duration of stress, whereas the second peak around T = 280 K in Fig. 4, appears only after 1000 min of stress. The first peak results from the superposition of two peaks: by performing a deconvolution with two Gaussian curves, we were able to extract two Arrhenius plots, reported in Fig. 4(b), associated with two associated traps, called T1 (⁠ $E a=474 meV$ and $σ s=2× 10 − 14 c m 2$⁠) and T2 (⁠ $E a=146 meV$ and $σ s=1× 10 − 20 c m 2$⁠). By comparing the Arrhenius plots with the signatures collected in our database of GaN and AlGaN defects, reported in the literature in Ref. 36, we found that T1 is compatible with Mg-related defects detected in n-type material,^37,38 and it might originate from Mg atoms back-diffused toward the active layer, consistent with Ref. 39. Another possibility is that it originates from the Mg level in the AlGaN EBL.⁴⁰ On the other hand, T2 can be related to defects due to Mg doping in p-GaN,^37,41 suggesting its location in the p-side of the device—according to our simulations [Fig. 1(b)] in the GaN:Mg p-contact layer close to the EBL or in the SCR of the Schottky type p-contact. In Fig. 4(a), we note that the sign of the C-DLTS peaks cannot be directly used to distinguish between majority and minority carrier traps, since (a) the junction is not unilateral, and (b) a parasitic Schottky junction is present also at the metal/p-GaN contact, and in series to the main p-junction, but with opposite polarity, having an impact on device capacitance. Finally, for T3, we extracted $E a=705 meV$ and $σ s=2× 10 − 12 c m 2$ (see Fig. 4) and is ascribed to point defects⁴² forming in the structure during the ageing. Noteworthily, the concentration of the trap T3, was found to increase, from the C-DLTS analysis in Fig. 4(a), indicating the generation/propagation of defects through the structure. Even though we could not yet identify the trap as well as the reasons for its formation/propagation, we think it is connected to the decrease in emission output power (increase in A-coefficient, decrease in injection efficiency). Further tests will be carried out to clarify a possible correlation between defect T3 and process explained for Fig. 3(e).

The decrease in defect T2 might be related to the stress-induced variation in the turn-on voltage and the largest impact can be expected for the active Mg acceptors close to the p-contact. Using TCAD, we model a decrease in the active Mg doping concentration within the first 20 nm of the p-contact layer (Fig. 5). During ageing, this reduction might occur through a partial passivation of Mg doping.⁴³ Specifically, we considered tunneling as the dominant process, and placed a dense (non-local) mesh from the metal contact to the first 10 nm of the p-GaN layer. To reproduce the degradation process, we changed the p-doping concentration (Mg at E_V + 150 meV) of the first 20 nm of p-GaN near the contact and fitted the I−V curves >5 V for different ageing times. This induced an increase in the width of the associated SCR, lowering the tunneling probability from the barrier, thus impeding carrier transport from the metal to the p-side. The simulations were found to match the experimental data accurately [Fig. 5(a) and Table I]. The widening of the Schottky-related SCR [Fig. 5(b)] also induced a decrease in the p-contact capacitance, which led to a decrease in the total capacitance of the device (for this reason, the amplitude of the C-DLTS signal cannot be directly converted into a defect concentration).

In this paper, we investigated the degradation mechanisms of UV-C LEDs by carrying out an extensive analysis based on C-DLTS, electro-optical characterization, and simulation. During the operation for 12 000 min at 100 A/cm², we observe a decrease in optical power and increase in serial resistance and ideality factor. Based on the analysis of the electro-optical characteristics, the reduction in optical power is ascribed to the combined effect of decreased injection efficiency and increased SRH recombination. C-DLTS measurements revealed that ageing induced the generation of a point defect T3 with an activation energy of 705 meV, which we believe is related to the decrease in optical power. Also we found two defects T1 and T2 (at 474 and 146 meV, respectively) present in the unaged sample and possibly decreasing in density over ageing time, likely associated with Mg-related traps in the EBL (T1) and the GaN:Mg layer below the p-contact. We modeled the observed increase in resistance and turn-on voltage as due to the passivation of Mg close the p-GaN–metal interface achieving an excellent quantitative agreement.

The authors would like to thank Sylvia Hagedorn and Markus Weyers (Ferdinand-Braun-Institut (FBH), Berlin, Germany) for providing the ELO AlN/sapphire templates for the LEDs. This work was partially supported by the German Federal Ministry of Education and Research (BMBF) within the “Advanced UV for Life” consortium.