An answer to LED droop? A major obstacle to solid-state lighting is so-called "droop" - the way that the efficiency falls at high drive currents. Its origins were unclear: until a recent spate of papers appeared, that is... The US Department of Energy's roadmap for white LEDs demands a rapid improvement in LED efficacy per dollar. To stick to this roadmap, manufacturing costs must fall whilst light output is increased – particularly at the high drive currents required for general lighting applications. But high efficacies at high drive currents are not something that we can associate with today's GaN devices. That's because LED internal quantum efficiency falls as current increases – a detrimental phenomenon known as "droop". Tackling droop starts with establishing a thorough understanding of its root cause, but progress in this direction has been slow. Quite simply, until recently nobody had provided a convincing explanation of the physics behind this phenomenon. However, this situation has now moved on, following the release of several research papers last fall that offered different explanations. In these papers, Philips Lumileds claimed that Auger recombination was the dominant cause for droop; Fred Schubert and his team from Rensselaer Polytechnic Institute (RPI) said carrier leakage was responsible; Yun-Li Charles Li and co-workers from National Taiwan University stated that thin quantum wells were to blame; and Bo Monemar from Linköping University pointed the finger at defects. Lumileds backs thicker wells Before Mike Krames and his colleagues from Lumileds identified Auger recombination as the dominant cause for droop, they sought to rule out several alternative explanations. Temperature-related effects were discarded first. "If you take a diode and run it in low-duty-factor pulsed operation you still observe this characteristic," explained Krames. The team then considered whether droop could be caused by carrier leakage from the active region, which would be driven by the electric field across the device. Early device results suggested that this might be the case, so Krames attempted to verify the findings with resonant photo-excitation measurements on a very simple structure – a thick InGaN layer (Appl. Phys. Lett. 91 141101). This experiment was expected to show a linear increase of light output above threshold. "To our surprise we didn't see that – we saw the droop," remarked Krames. "That, by itself, immediately rules out any sort of transport effect, any interfacial effect, or any quantum-well effect that may be caused by the presence of built-in fields." According to Krames, these results actually showed that droop was a bulk effect that occurs for epilayers grown in any orientation. The Lumileds team went on to consider whether defects were causing the droop by probing two samples with significantly different dislocation densities. But high-dislocation and low-dislocation density samples had very similar photoluminescence characteristics, ruling out the extended defect theory. Krames says that this meant that the most likely cause of droop was a highorder, carrier-density dependent physical phenomenon, specifically Auger recombination. This non-radiative process involves the transfer of energy from electron-and-hole recombination to a third carrier. Pushing out the efficiency peak Armed with this knowledge, Krames' team went on to fabricate a range of LEDs that are less susceptible to Auger recombination, which feature thicker wells to reduce the carrier density in the active region. Specifically, 9 and 13 nm thick wells with In0.14Ga0.86N and In0.16Ga0.84N compositions were used for the active region (Appl. Phys. Lett. 91 243506, 2007 for details). The researchers then compared the performance of these devices with conventional designs featuring two or more 2.5 nm thick InGaN quantum wells. Pulsed measurements with a 1% duty factor showed that the standard devices produced a peak efficiency at 10 A/cm2, with droop kicking in as current increased. In contrast, the LEDs with the thicker active region peaked at more than 200 A/cm2, backing up the Auger theory. The new designs emit at 430–445 nm and produce 20% more light at 1.8 A than the company's state-of-the-art quantum well structures. They can also deliver 2.3 W at 2 A. Figure 1 Krames' view that thicker wells can combat droop is backed up by research from Yun-Li Charles Li and co-workers from National Taiwan University (see Appl. Phys. Lett. 91 181113, 2007 for details). These scientists have measured the efficiency droop in 460 nm InGaN LEDs and have focused on quantum wells ranging from 0.6 to 1.5 nm thick. The droop is virtually eliminated as the quantum well thickness increases to 1.5 nm, but this improvement comes with the unwanted penalty of lower internal quantum efficiency. The elimination of droop is surprising in wells that are this thin, because commercial LEDs suffer from droop and are typically 2–3 nm thick. Li is also puzzled by this, but he says that droop could depend on the growth process for the electron-blocking layer, or the overall growth rate. Schubert goes for polarization-balancing The RPI team, which has been working with Samsung Electro Mechanics Corporation, Korea, has also been using optical measurements to search for the origins of droop. However, they have arrived at a different conclusion – that the effect results from carrier recombination outside the multi-quantum-well active region under forward bias (Appl. Phys. Lett. 91 183507, 2007). Figure 2 This team performed simulations that accurately matched the output of their devices at a range of currents, before making additional calculations for LEDs with polarization-matched electron-blocking layers and multiple quantum-wells (figure 2). "We're proposing a revolutionary design of the active region," explained Schubert. According to the calculations, droop can be eliminated by replacing the GaN/InGaN multiple quantum wells and the AlGaN electron blocking layers with polarization-balancing quantum barriers and blocking layers made from AlGaInN. The scientists are now trying to produce this structure, but it's quite a challenge, as the barriers and blocking layers require well-controlled aluminum and indium content. Schubert's group has also been working with Sandia National Labs on a study that examines the relationship between defects and droop (Appl. Phys. Lett. 91 231114, 2007). Measurements show that conventional 440 nm InGaN LEDs with high-defect densities do not suffer from the droop that plagues lower-defect density equivalents. However, the trade-off is a low overall efficiency. "Obviously, that isn't the right way to solve the droop problem," said Schubert. Meanwhile, Linköping's Bo Monemar is backing yet another explanation for droop – carrier localization at defects (Appl. Phys. Lett. 91 181103, 2007). In particular, he points a finger at phonon-assisted transport of holes via tunneling at defect sites and threading dislocations that can induce a parasitic tunneling current. So who's right? What's clear is that these research teams have differing views and they can't all be right. So what do they make of each other's theories? Krames remains confident, saying that none of the other droop explanations have changed his view that Auger recombination is the dominant factor. While Monemar argues that the improved droop observed in the best non-polar LEDs is thanks to low epilayer defect densities (see Massive power boost for non-polar GaN LEDs), Krames believes that it actually results from the thicker quantum wells used. According to him, these thicker wells increase the recombination volume, which reduces carrier density, and make three-body interactions required for Auger recombination far less likely. The Lumileds man suspects that the problem with the RPI result is that insufficiently high excitation densities were used in those optical experiments. "That's a trap that we also fell into initially," admitted Krames, "before we invested a great deal of time and money in building a completely new optical set-up with an [high-power] alexandrite laser." Schubert, however, is concerned that recent Lumileds papers employ a new quantity, "recombination thickness", which is used to determine the carrier concentration in the quantum well. He believes that the way this quantity is employed could lead to an overestimation of both carrier concentration and Auger recombination. These differences in opinion do not seem to be generating any ill-feeling, however. "We have a wonderful situation at this time," said Schubert. "Many different mechanisms have been proposed and I believe that the scientific community will be smart enough to find out what are the dominant causes and the minor effects." Let's hope that he's right, for droop-free LEDs could be the key breakthrough to unlocking the general lighting market. About the author Richard Stevenson is features editor of Compound Semiconductor magazine.