1 Introduction

Since 2020, the immediate actions required by the Covid pandemic pushed governments, private companies, and research institutes to focus on the development of several disinfection systems and different technologies able to contain the spread of SARS-CoV-2 virus (severe acute respiratory syndrome coronavirus 2). Ultraviolet (UV) light, in particular in the UV-C range, is well known since 1877^[1] to have a strong antiviral and antimicrobial effect, and it has been identified, within the first months of the Covid pandemic^[2] to be effective also toward the inactivation of SARS-CoV-2. The UV light dose required to achieve inactivation, the correlation between the wavelength, and a comparison between different lighting technologies has been recently reviewed by our research group;^[3] the results indicate that deep UV light-emitting diodes (LEDs) with a wavelength of around 265–280 nm represent the state-of-the-art technology for compact disinfection systems. However, compared to visible devices based on indium gallium nitride (InGaN),^[4] deep UV LEDs, based on aluminum gallium nitride (AlGaN), are affected by relatively low efficiency, mainly ascribable to: 1) lattice mismatch leading to dislocations, 2) critical quality of AlGaN with high Al content, and 3) high absorptive p-GaN contact layer. Despite a relatively low wall plug efficiency, in the range of 3–5%, UV-C LEDs have been demonstrated to be suitable for disinfection systems, particularly for small objects,^[5] thus representing a potential breakthrough in antiviral and antibacterial systems. Unfortunately, the main factor limiting their adoption in commercial disinfection systems is their limited reliability.^[6] With this work, we will focus on the impact of reliability on the usability of commercial-grade UV-C LEDs. As a disinfection device, the LED is required to provide a certain optical energy that corresponds to a certain number of disinfection cycles. The amount of energy that a device can emit is, therefore, the main parameter that should be assessed during the degradation: not only can it be accepted that a device has half of its lifetime when doubling its optical power, but this could also be beneficial, since it would allow a faster disinfection time. The impact of driving current as an acceleration factor is the key in this aspect, and while it has been analyzed on research-grade UV-B devices, no clear effects of current on reliability and usage are present in the literature for commercial-grade LEDs. The aim of this work is then to study the effects of driving current on the degradation of the optical, electrical, and spectral characteristics of commercial UV-C devices.

2 Experimental Section

In this work, we want to compare the degradation rate of commercially available UV-C LEDs at current levels that are two and three times higher than its absolute maximum current. For this reason, we selected a UV LED model with good emission performance (12 mW at 100 mA), a nominal wavelength of 277 nm, a chip size of 0.25 mm², a promising lifetime, and a low cost of about $4 per device; devices are provided on a hermetic ceramic based package equipped with a transparent quartz window. Several samples of this model had been acquired, soldered, and mounted on the same type of printed circuit board (PCB) to had the same thermal dissipation. From the datasheet, the samples have an absolute maximum current of 150 mA (60 A cm⁻²) and thermal resistance of 22 K W⁻¹, which led to a junction temperature during the stress of about 55, 88, and 126 °C at the currents of 150, 300, and 450 mA, respectively; these values were also confirmed by dynamic evaluation of temperature carried out according to the JESD51 standards.

The experimental stress setup was composed of a thermo-electric plate, to regulate the stress and measurement temperatures, a source meter, a photodiode, and a charge-coupled device (CCD) array spectrometer. During the aging tests, we carried out current density versus voltage (J–V) and optical power versus current density (L–J) characterizations at different temperatures, from 15 to 75 °C with steps of 10 °C, whereas power spectral density (PSD) characterization was performed at the begin and at the end of the stress tests, only at 25 °C. The maximum measurement current had been set to the stress current level for each device.

3 Results and Discussion

In Figure 1, we reported in linear scale and in semilogarithmic scale the electrical characteristics (J–V) of the devices before and after the aging tests. As expected, the initial characteristics of three LEDs are superimposed, except for the devices stressed at 300 mA that showed a slightly lower leakage current below the turn-on voltage. After the stress, we could observe a decrease in drive voltage for all samples, possibly ascribed to the activation of magnesium (Mg) in the p-doped region of the device,^[7] that is, confirmed also by the decrease in series resistance. About this result, it is worth to notice that we cannot compare the absolute value of the decrease in series resistance, but we can only compare the trends during the aging time, because this parameter was not extrapolated in the same region for all the devices, but it was calculated as the slope of the J–V curve in the voltage region before the current compliance imposed for each stress.

All devices showed an increase in sub-turn-on current after stress; we ascribed this effect to both the increase in the defectiveness of the active region,^[8] also confirmed by the increase in ideality factor,^{[9, 10]} and to the increase in lateral passivation-related parasitic paths,^[11] that led to the increase trap-assisted tunneling (TAT) conduction processes.^[12]

In Figure 2, we reported LED optical characteristics before and after the stress tests in linear and logarithmic scales. As shown, the decrease in optical power with stress times is more prominent for the device stressed at 450 mA at all current levels, due to the higher generation of defects caused by the stress^[13] and the decrease in injection efficiency in the active region, already reported in previous studies.^[14] Moreover, all the samples presented an evident breaking point in the optical power (OP) characteristics, where there was a change in the slope of the curves.

In Figure 3a, we reported the slope extrapolated from the logarithmic OP versus current curves at high and low current levels. When measured at high current, all devices presented a slope of about 1 before the stress, which can be associated with the regime where the radiative recombination prevails; as the stress time proceeds the slope tends to increase, and at higher stress currents the increase is steeper, suggesting a correlation with the growth of Shockley–Read–Hall (SRH) recombination rate.^[15] The slopes measured at low current levels have an initial slope of 2 or higher, which indicates that the devices are operating in the regime where prevails nonradiative recombination processes^{[16, 17]} and possibly asymmetrical injection.^[18] The low current L–I slopes presented an increasing trend during the aging similar to the high current ones. It is worth noticing that the decrease observable at longer stress times is probably caused by an inaccurate fit of the data due to the achievement of the noise limit of photodetector.

In Figure 3b, we analyzed the optical power emitted by LEDs measured at high (100 A cm⁻²) and low (80 mA cm⁻²) current levels during the aging test. By observing the results we can comment that: 1) a faster decay of the OP is detected when the devices are measured at lower currents, indicating that an increase in the nonradiative recombination is likely taking place during the degradation, as suggested by the analysis of the slopes in Figure 3a; 2) the optical degradation becomes more prominent with the increase in stress current, possibly due to the generation of SRH recombination centers and the decrease in the injection efficiency in the active region, since both processes are current- and thermally-dependent;^[19-21] 3) when measured at high currents the kinetics of the devices show a significant recovery in the first few hours of operation. This recovery is present for all the stress currents, and technology computer-aided design (TCAD) simulations indicate that this behavior is related to a change in $e^{-}-h^{+}$ wavefunction overlap during the aging of the samples, but further analyses are undergoing to confirm this hypothesis. Finally, 4) the L70, L80, and L90 have been extrapolated for the devices stressed at 150 and 300 mA by means of an empirical fitting where the L80 lifetime is the time required by a device to reach an OP equal to 90% of its initial value. The formula used for the fitting is a properly modified logistic decay function, which can be used to fit several optical decays of UV-C LEDs (extrapolation is reported in the dotted lines in Figure 3b).

At the beginning and at end of the stress tests, we evaluated the PSD of the devices at different current levels. In Figure 5a, we reported the measurements at low current levels (100, 400 μA cm⁻²) where we are able to identify other two parasitic bands and peaks, besides the main peak. In particular, we found the presence of a parasitic peak between 319 and 380 nm and a broad parasitic band between 415 and 575 nm, whose behavior is reported in Figure 5b. The first band had an energy of about 3.5–3.6 eV, slightly higher than the bandgap of GaN used in p-side, as reported in refs. [27, 28]. It could be ascribed to recombination in two possible regions: or at the interface between EBL and p-side through Mg deep-acceptor levels due to carrier escape or overflow from the quantum wells, or in the QW region due to recombination through defects (Figure 5c). The spectral changes are therefore in possible agreement with an increase in nonradiative recombination centers in the AlGaN material of the QW and/or its barriers, and this behavior has been ascribed to vacancies complexes, in particular, V_Ga^[29] or V_Al^[30] (gallium or aluminum vacancies). The hypothesis that these defects could be generated by Auger excited carriers is then supported by an increase in shoulder emission, in particular, at higher stress current. The second band could be related to the classical GaN yellow luminescence, already reported in several previous works,^[31] and possibly localized in the p-GaN side of the device. The increase in second parasitic band could be ascribed to the increase in defectiveness in the device, especially in the region where these recombinations occur. In contrast, the decrease in the first parasitic band is possibly ascribed to a decrease in the injection efficiency in the active region that led to a lower accumulation of carriers outside the QWs. Finally, the total energy delivered by the LEDs during their useful lifetime has been evaluated. Figure 6 reports the table and the plot of the total energy emitted by the LED over a lifetime. This was obtained by integrating the optical power during the operation for different lifetime intervals; three-lifetime intervals have been evaluated between zero and L70, L80, and L90, respectively.

Results confirm that: 1) as the current increases the total amount of energy delivered by the LED decreases sensibly. As an example, doubling the current from 150 to 300 mA approximately halves the number of photons emitted by the device during its lifetime, 2) by considering different lifetimes, the effect of driving current on the energy emitted is almost unaffected, as shown in the inset of Figure 6b where the normalized plots are overlapped, and 3) increasing the stress current to high values strongly reduces the energy provided by the devices, an increase in the current from 300 to 450 mA reduces the optical energy by 3.5 times.

The main conclusion of this part of the work is that an increase in operation current results in (1) a sublinear increase in optical power; (2) a $J^{-3}$ variation in the device lifetime. As a consequence, increasing the LED current to minimize the number of LEDs in a disinfection system is not an optimal solution. The increment in radiant power obtained by increasing the current is counteracted by the substantial decrease in the device lifetime, eventually reducing the amount of UV energy emitted during the useful life of the LED. It should be noted that since the duration of the treatment reduces as the current increases, a system-level analysis should be considered by factoring in the different conditions.

4 Conclusions

In this study, we analyzed the reliability of commercial off-the-shelf DUV LEDs with a nominal emission peak of 275 nm. These devices are typically used for disinfection systems where they emit a certain light dose (or energy) to fulfill the disinfection task. The balance between the emitted optical power and the lifetime of the device is therefore crucial to correctly design the most effective disinfection system; it is ultimately related to the operating current of the LED. Results from stress at different current densities indicate that after an initial recovery, the devices undergo a gradual degradation of the optical emitted power which becomes more intense as the stress current is increased. The effect of current on lifetime is approximately proportional to the inverse of the third power of the current density, thus indicating that the energy released by Auger–Meitner recombination is a possible cause for the degradation. A stronger degradation at low current levels and an increase in the L–I slope suggest that the degradation is related to an increase in SRH recombination, while spectral analysis suggests a possible localization of the defects in the active region and in the p-GaN layer.

Finally, we provide evidence that—for disinfection purposes—increasing the operating current to reduce the number of adopted LEDs is not an optimized strategy. Due to the strong dependence of reliability on current, an increase in the operating current strongly reduces the device lifetime, and consequently, the dose provided by the device during its useful life. This result provides relevant information for the development of advanced and reliable sterilization systems.

Conflict of Interest

The authors declare no conflict of interest.