Photoluminescence pinpoints Auger as the cause of LED droop

A revolution in lighting has just begun. Individuals and businesses are now buying LED lamps, thanks to their great attributes. These solid-state sources can span a wide range of lumen outputs; they produce a colour quality that has improved dramatically in recent years to a level that now rivals that produced by halogen lamps; and they deliver lifetimes and efficiencies that are superior to most conventional sources. So why, given all these wonderful characteristics, isn't a battalion of LEDs now found at the heart of every installed light source?
Well, as always, many answers could be given. And in this case, one of them is quite simple: The costs are still too high to be widely accepted in the consumer market.
To understand the reason for this, one must take a close look into the physics of the LED. One of its features is that its light output depends on its drive current, so doubling its current does not produce an increase in brightness by the same factor. Instead, in many classes of LED, the brightness increase is lower than this "“ and particularly so for InGaN-based devices, where the effect is known as droop.
For most classes of LED, it is pretty clear which physical effect is behind the rollover in efficiency. For example, in short wavelength InGaAlP-based LEDs, greater carrier overflow is behind the fall in efficiency at increasing drive currents. But with InGaN-based LEDs, the rather unspecific term "˜droop' is used, because the root cause for this phenomenon has not been discovered, despite intensive research into this subject for many years.
The droop behavior in a typical blue-emitting LED produces a sub-linear rise in light output as a function of drive current (see Figure 1). The decrease in efficiency that results is often shown in plots of the external quantum efficiency versus drive current. Typically, efficiency peaks at about 20 mA for a chip with dimensions of 1 mm by 1 mm, and drops at higher currents. In addition, efficiency decreases at lower currents. However, this is not a major concern, because it corresponds to current densities that are well below those that would be used in a typical high-brightness LED. These devices often operate at current densities of around 40 A/cm�, which, as can be seen from Figure 1, is a regime where the device efficiency is significantly below its peak value.

Figure 1: Current dependency of the optical output power (red) and the external quantum efficiency (EQE, blue) of a typical blue-emitting LED. At operating conditions (350 mA), the light output increases sub-linearly with increasing input power, which can be seen in comparison to the linear function plotted in grey. The associated decrease in EQE towards high currents is commonly known as droop

So why not simply drive the LED at the current density that produces the peak external quantum efficiency? Well, because in order to target the same light output, the area of the chip will have to increase by an order of magnitude, but because the cost of the chip scales with its size − and the chip is a major cost contributor to the device − this would result in more expensive LEDs. So one has to choose between very efficient LEDs at very high costs, or less expensive LEDs with lower efficiency.
Fortunately, there is a third alternative, which is to address droop. This will allow the chip size to be reduced without losing efficiency, and it will ultimately enable cost to be brought down, speeding penetration of the LED into the general lighting market.
By far the best way to combat droop is to begin by understanding its origin, and then go on to develop LED designs that are not plagued by this malady. This approach sounds straightforward, but it is not − getting to the bottom of droop is far from easy. Over time, a handful of hypotheses have been developed to account for droop, and this has led to various physical models based on the likes of carrier leakage by overflow, defect-mediated losses and Auger recombination (these are illustrated in Figure 2). Some researchers have even drawn on the combination of hypotheses and blamed droop on mechanisms such as Auger-enhanced leakage.

Figure 2: An illustration of light generation in comparison to proposed droop mechanisms: a) One electron recombines with one hole and the energy is released via a photon (light) b) Self-screening of defects from quantum well carriers (here electrons); once a certain density is reached the carriers overcome the potential barrier and recombine non-radiatively c) Carriers (here electrons) overflow the quantum well(s) and do not contribute to the light generation d) Energy transfer from non-radiative recombination of electron with hole to a third charge carrier (here an electron)

Leading theories
With the model that focuses on defect-mediated loss, the quantum well can be compared to a pot that is not watertight, but has holes in its wall. When water is poured in, it remains within the pot until its level reaches that of the holes. Then, at that point, the water starts to leaks through them, causing filling efficiency to fall.
The leakage-via-defects model can be explained in a similar way, with charge carriers staying in the well when the carrier concentration is low. But when current density increases and carrier concentration rises, the defects in the quantum well become active, behaving as non-radiative recombination centres. Those supporting this view believe that these defects are probably V-pits, which are named after their apparent shape in a cross-sectional view of the crystal. It is argued that V-pits are surrounded by a repulsive potential, which hinders carriers to reach them and recombine there. However, this potential is surpassed as carrier density increases, enabling the activation of V-pits as recombination centres.
Charge carrier overflow can also be explained with the water-in-the-pot metaphor. This time the pot is tilted, so part of the water jet flows beside the pot and fails to contribute to its filling. It is also possible to imagine that if the water is flowing fast, it enters into the pot with such an impact that it splashes out again. Translating this behaviour to that for the LED suggests that the injection efficiency diminishes, due to charge carriers overflowing the quantum wells without recombining radiatively.
The third popular explanation for droop, Auger recombination, involves a process associated with the interaction of carriers to promote one of them to a higher energy. The additional energy can be given to either an electron or a hole by a process referred to as either "˜electron-Auger' or "˜hole-Auger'. The charge carrier does not maintain its high-energy state for long, but rather relaxes by emitting energy in the form of heat.

The case for Auger
When investigations into droop first began, it was felt that Auger recombination is negligible in blue LEDs, because the probability for this process plummets as the band gap increases. However, when researchers carried out detailed simulations "“ which yielded higher Auger coefficients by taking into account phonon interactions and alloy fluctuations − it appeared that Auger recombination could be the dominating factor for LED droop. Efforts to build more accurate models for the LED continue to this day.
Experimental investigations to identify the role of Auger have also become increasing sophisticated. Earlier this year a partnership between researchers at the University of California, Santa Barbara, and Ecole Polytechnique observed high-energy electrons from an electrically driven LED. Their interpretation of these interesting experimental results, which are currently subject to controversial discussion, is that the high-energy electrons are the result of an electron-Auger process.
One of the reasons why the debate on droop has gone on for so long is that not only is it challenging to distinguish between electron leakage and Auger recombination, but these processes can be coupled. In addition, it is possible to reproduce the droop curve with models based on leakage and Auger recombination, and the Auger carriers that are generated within the LED have a high enough energy to overcome barriers and thus contribute to leakage.
To try and distinguish between injection and Auger loss, our team at Osram Opto Semiconductors devised an experiment that can exclude leakage by exciting the carriers directly in the quantum well.� Our approach, like that of the UCSB-Ecole Polytechnique collaboration, exploits the fact that the Auger process generates highly energetic charge carriers (they are also known as "˜hot' carriers). However, in our independent work, we employ a novel structure that converts these hot carriers into highly energetic light. The central idea of our approach is to capture hot carriers with a tailored quantum well, and record the spectral output that results from their recombination and subsequent photon emission. To carry out this experiment, we measure the photoluminescence emitted by an InGaN-based structure containing neighbouring ultraviolet (UV) and green quantum wells (see Figure 3). We pump this heterostructure with a blue laser, so absorption only occurs in the green wells.

Figure 3: The approach taken by scientists at Osram to visualize Auger processes and correlate them to the droop. Ultraviolet quantum wells are used to capture hot carriers generated by Auger processes. Luminescence from the UV wells can be attributed to Auger recombination taking place in the drooping green wells

If we use a low intensity laser beam and generate a low carrier density, we can neglect Auger processes and should only expect to see green emission; but if the intensity of the optical pumping is sufficient to create a carrier density comparable to that found in the droop regime, hot carriers should be created via Auger recombination, before they are captured by the UV wells, which act as detectors. So, in other words, if there is luminescence from the UV wells, it will prove that Auger recombination is taking place in the green-emitting wells, and thus in an LED. One great merit of this photoluminescence-based approach is that no macroscopic electric fields are present in the structure. Consequently, it is possible to rule out the generation of hot carriers by transport-related effects.

Experimental evidence
Our expertise associated with the growth of InGaN LEDs enables us to fabricate structures with high material quality. That's essential, because if the crystal quality is poor, other non-radiative processes are stronger than droop. After fine-tuning our experimental set-up, we acquired our first photoluminescence spectra, which exhibited the expected UV peak (see Figure 4). This was a breakthrough: It was the first time we had detected highly energetic charge carriers generated in an InGaN-based structure under resonant photoluminescence excitation. What's more, the intensity of the UV light was even higher than we had expected.

Figure 4: A blue laser excites a sample containing neighbouring green and ultraviolet wells up to charge carrier densities where droop occurs. The resulting spectra reveal the expected green emission, plus luminescence originating from the UV wells. Since the UV luminescence is only present in structures where green and UV wells are combined, its origin must be Auger processes in the green wells

It is obviously important to rule out that the UV emission stems from artefacts arising from the green wells. To do this, we measured the photoluminescence spectra produced by our pair of reference samples, which contained solely green or UV wells. Emission from these samples did not feature a high-energy peak, implying that the high-energy peak in our structure with green and UV wells did not originate from artefacts from either the green wells or from direct excitation of the UV wells via two-photon absorption. Thermal effects and free carrier absorption were also ruled out with further experiments, leading us to conclude that only Auger processes could provide a full, consistent explanation of the data. Our experiment also offered a unique insight into the nature of Auger recombination. Since luminescence requires both carrier types in the UV quantum wells, our experiment reveals that the rates for electron-Auger and hole-Auger processes are both relevant.
Although we had detected the presence of Auger, we had yet to determine if it was the main contributor to droop. Could it just be a feature that is present and detectable, but one that does not play a significant role in the efficiency of the LED? To answer this key question, we took a closer look at the correlation between Auger processes, represented by UV luminescence, and charge carriers lost to the droop mechanism. This led us to observe a steep rise in UV intensity towards the excitation densities where droop occurs (see Figure 5). We note that droop and UV emission scale with the cube of the charge carrier density in the green well, implying that the number of green photons lost to the droop phenomena is directly proportional to the rate of emitted UV photons. Armed with this relationship, we could also prove that at least 1 percent of all lost charge carriers can be attributed to the Auger effect.

Figure 5: A measure for the internal quantum efficiency of the green wells is given by the ratio between detected green photons and exciting laser photons. Towards higher excitation density. where droop occurs, a steep rise in UV luminescence was observed. This was ascribed to Auger processes.

At first glance, this value of 1 percent seems very low, and it certainly doesn't follow that Auger is the sole cause of droop. But care is needed in interpreting this result. After thinking carefully about generation and recombination processes going on in this experiment, we can offer a sound argument that Auger is playing a dominant role in the behaviour of an LED. First, it is obvious that most Auger charge carriers, which are generated in the green quantum wells, will not be captured by the UV wells "“ instead, they will relax back into the energetically favourable green wells. So it is reasonable to assume that only a small fraction of them reach the UV wells. In addition, since every electron needs a hole to emit a UV photon, any discrepancy in the Auger generation rates for holes and electrons will diminish detection efficiency. Taking both of these factors into account, we conclude that the intensity of the detected UV emission strongly suggests that the Auger effect is the dominant cause of droop.
Developing a fully quantitative understanding of the contributions of Auger recombination to droop will require determination of electron- and hole-Auger rates. We have looked into this, and have just published a paper in Applied Physics Express showing that the transfer of energy to an electron, rather than a hole, is the more common process. This finding is in stark contrast to the claims of leading theorists, which conclude that the hole Auger process is stronger.
Another question that remains is why Auger recombination should play such a prominent role in wide-bandgap materials, such as GaN and its related alloys. Theoretical work points to phonons and/or alloy fluctuations, which facilitate momentum conservation and allow high Auger rates. However, no experimental results have been published related to these conjectures.
From a practical point of view, our findings let us draw one very important conclusion: LED development must now focus on structures that extenuate the impact of the Auger effect and disregard other hypotheses. This finding is tremendously valuable, because epitaxial structures have a vast number of degrees of freedom, and it is very helpful to know that at least some of them have limited potential to mitigate the impact of droop. At present, the best practical measures for reducing the Auger effect are still to be uncovered. However, they will be revealed in further, detailed engineering and scientific work. This will help to define droop more and more, enabling the manufacture of brighter LEDs and ultimately the accelerated adoption of solid-state lighting.

Further reading:

J. Iveland et. al. Phys. Rev. Lett. 110 177406 (2013)
M. Binder et. al. Appl. Phys. Lett. 103071108 (2013)
A. Hangleiter et. al. Phys. Rev. Lett. 95 127402 (2005)
Y. C. Shen et. Al. Appl. Phys. Lett. 91 141101 (2007)
E. Kioupakis et. al. Appl. Phys. Lett. 98 161107 (2011)
R. Vaxenburg et. al. Appl. Phys. Lett. 102031120 (2013)
B. Galler et. al. Appl. Phys. Express 6 112101 (2013)