It is very tricky to come up with a watertight explanation for the cause of droop. However, it is certainly possible to combat this mysterious malady, which causes LED efficiency to decrease at high current densities, by: turning to better electron-blocking layers made from InAlN; and improving the injection of holes into the wells, plus their distribution throughout the active region, argues to Jae-Hyun Ryou from the University of Houston, Russell Dupuis and P. Douglas Yoder from Georgia Institute of Technology and Fernando Ponce from Arizona State University.

LED light bulbshave many attractive attributes: lifetimes of 50,000 hours, negligible warm-up times, the absence of mercury, and higher efficiencies than incumbent sources.

However, the retail prices of these lamps are too high to tempt the majority of the public to invest in solid-state lighting, partly because the gains in efficiency over compact fluorescents are not yet to be that alluring. What’s needed is for the LEDs that are used in the bulbs start to delivering the eye-watering efficiencies that they do in the lab. If that happens, bulbs based on these chips will have far lower running costs and sell for much less than they do today.

When it comes to efficiency, chips in the lab have produced 276 lumens-per-Watt (lm/W), which is three-to-four times that of compact fluorescent lamps (60-80 lm/W) and vastly higher than incandescent sources (11-17 lm/W). However, according to reports coming from the US Department of Energy, the efficacies of the ‘well-made’ warm-white and cool-white LED lamps are only slightly higher than 100 lm/W and 130 lm/W, respectively. The culprit for this massive difference between the lab record and the efficacy of commercial LEDs in bulbs is a mysterious malady known as droop, which causes a decline in the efficiency of an LED as the current density through it reaches very high levels. What this means is that the peak quantum efficiency of the LED occurs at a lower current than the value that it is driven at in a light bulb, compromising its efficacy.

If droop could be eliminated – or even trimmed substantially – this could be a game changer for solid-state lighting, moving this industry so that it is not just serving the early adopter in the home and the lighting engineer who thinks about all the costs associated with lighting large buildings, but selling to the general public. That’s because LEDs with far less droop wouldn’t only be more efficient and thus cut electricity bills: They could be also driven at far higher current densities, because their greater efficiency translates into less heating, and that ultimately means that far fewer chips would be needed in a bulb, cutting its cost substantially.



Figure 1.  Quantum efficiency (QE) versus current density for blue LEDs without an electron-blocking layer (EBL), with an Al0.2Ga0.8N EBL, and with an In0.18Al0.82N EBL.  Inset shows light output versus current (L-I) characteristics of LEDs without an EBL, with an Al0.2Ga0.8N EBL, and with an In0.18Al0.82N EBL.  An alternative InAlN EBL significantly mitigates the efficiency droop with the lowest efficiency droop ratio of ~18 percent.  (Reprinted with permission from Appl. Phys. Lett. 96221105 (2010). Copyright 2010 American Institute of Physics.)

A little history

Droop is clearly distinct from the thermal roll-over seen in laser diode and VCSEL plots of output power as a function of current. Until the mid 2000s, droop was understood – without much controversy – as an inevitable phenomena associated with III-nitride materials. Nearly every commercial LED is grown on a lattice- and thermal-mismatched foreign substrate, such as either sapphire, SiC, or more recently silicon, using a technique commonly referred to as strained heteroepitaxy.

This leads to a high defect density, with threading dislocations typically higher than 108 cm-2. It is possible that these dislocations are not that harmful for device performance – but droop is the price that you’ll have to pay for this.

The reasoning behind this view is that the surprisingly high level of radiative recombination in such defect-ridden structures is a result of indium-rich, quantum-dot-like, localized states in InGaN quantum wells. At low currents, these states screen detrimental effects from crystalline defects, leading to a high quantum efficiency. But as the current through the LED is cranked up, more carriers overflow from the localized ‘shelter’ states to recombine non-radiatively in dislocations, causing the device’s quantum efficiency to plummet.

Today, this explanation of droop has fallen out of favour. That’s partly because it can’t explain why LEDs with much lower dislocation densities, which are formed on free-standing GaN substrates, are plagued by droop. However, it is also because many other conjectures for droop are being offered, due to mechanisms such as: Auger recombination (including direct band-to-band and indirect defect- or phonon-assisted recombinations); electron spill-over out of the active region; inefficient hole injection and transport in the active region; and several other theories, which all have their champions.

If you look at the academic papers that detail these conjectures, you’ll find that the data presented in each set of theoretical studies and experiments is fairly logical, and it supports the proposed mechanism; however, the findings and claims are not consistent with one another, and in some cases they can even be contradictory. This reveals that there is yet to be a unified, watertight explanation detailing the dominant mechanisms responsible for droop. Instead, prejudice abounds, with conclusions drawn that may heavily depend on a pre-emptive model. This state of affairs may even hamper efforts to get to the bottom of droop: It might be governed by several of the proposed mechanisms, which are inter-related and coupled to one another.

Fathoming the cause of droop is critical for advancing the understanding device physics, and it is one route towards the development of droop-free LEDs. But it is not the only way: It is also possible to come up with droop-busting designs without uncovering a universal, unquestionable explanation for this energy-sapping mechanism.

Droop and carrier dynamics

If you peruse the academic literature, you’ll find that all the leading conjectures for the origin of droop are related to carrier dynamics. Droop has been blamed on electron leakage, which is related to unsatisfactory carrier confinement; it’s been claimed to stem from poor hole transport into the active region, which depends on the injection of carriers and their concentration in each well; and droop has been linked to Auger recombination, which heavily depends on carrier density, so is influenced by injection efficiencies and carrier concentrations. Hence, tracking and understanding the injection, distribution and concentration of carriers will help with efforts to identify the origin of droop and possibly uncover ways to combat this malady.

It is critical that efforts to try and combat efficiency droop do not neglect the absolute value for peak quantum efficiency. Droop tends to be characterized by comparing the peak efficiency to that found at a high current density. It is possible to diminish droop by sacrificing the peak quantum efficiency, but that approach is not the right one to take, because the goal is to learn how to take LEDs that are really efficient at low current densities and replicate that performance at really high current densities.

Our US research team, a partnership between Georgia Institute of Technology, Arizona State University and the University of Houston, has focused our efforts at combatting droop on engineering carrier dynamics via alternative layer structures. Our modifications do not involve adjustments to the multi-quantum well active region, because experiments by other groups suggest that improvements in droop brought about by this come at the expense of the peak quantum efficiency of the LEDs (or even at the expense of quantum efficiencies over a wide range of current densities). Instead, we investigated how changes to the electron-blocking layers could influence electron confinement and the injection and transport of holes in the active region. As we looked at various different designs, our strategy was to: confine electrons in the active region as much as possible; inject as many holes into the active region as possible; and distribute, as uniformly as possible, both carriers among the wells within the active region.

Our first modification was to adjust the electron-blocking layer so that it is better at confining this carrier in the active region. AlGaN is the standard material for making the electron-blocking layer, which is sandwiched between a p-type layer and the active region and reduces the number of electrons that spill out of the quantum wells. Moving to a material with a wider bandgap for the electron blocker promises to improve the confinement of this carrier, but recent studies show that a switch to wider-bandgap AlxGa1-xN is not that effective.

A more promising replacement is In0.18Al0.82N. It combines the opportunity for lattice-matching with GaN with a wider energy bandgap than AlGaN and a larger conduction-band offset. What’s more, strained – and especially in-plane compressive-strained – InxAl1-xN electron-blocking layers offer unique features for visible LEDs.  They change strain in the blocking layer, which can offset the interface charges induced by spontaneous polarization between InAlN and GaN in the barrier of the active region, and also mitigate the quantum-confined Stark effect in multiple quantum wells.

When we replaced the ‘standard’ AlGaN electron-blocking layer with In0.18Al0.82N, we found that this reduced the droop in the LED (see Figure 1). In the absence of an electron-blocking layer, there is a rapid efficiency droop: It is 69 percent, when defined as the decline between peak quantum efficiency and efficiency at a current density of 360 A cm-2. In comparison, the LED with an Al0.2Ga0.8N electron-blocking layer has a droop of 30 percent, while that with InAlN has 18 percent droop. However, although that last figure represents an improvement, droop is still significant. That might be because other mechanisms besides electron spill-over contribute to efficiency droop, or that the suppression of electron spill-over is not complete, even with a wider-bandgap In0.18Al0.82N electron-blocking layer.

Adding an electron-blocking layer can actually be a double-edged sword. While it creates a barrier that stops electrons from leaking out of the device, it also can form a potential barrier for the injection of holes from a p-type layer (refer to the inset of Figure 2). Especially at low current densities, this barrier may limit hole injection, leading to lower device efficiency. 



Figure. 2.  Light-current characteristics of LEDs with In0.18Al0.82N EBLs of various thicknesses.  Inset shows equilibrium electronic band diagrams.  These curves suggest that both hole-blocking and electron-confinement effects of the EBL should be qualitatively considered when addressing peak efficiency and efficiency droop for LED operating at high current densities.  (Reprinted with permission from Appl. Phys. Lett. 101161110 (2012). Copyright 2012 American Institute of Physics.)

The In0.18Al0.82N layer is perfect for studying electron confinement and hole injection simultaneously. All that is needed is to alter its thickness (see Figure 2). That’s not the case for AlGaN, because changes in aluminium richness don’t just change the barrier height of this blocking layer – they also influence p-type doping efficiency and strain.

Measurements of light output at different current densities with this series of LEDs provided an insight into device behaviour. Below current densities of 300 A cm-2, an LED without an electron-blocking layer produced a higher quantum efficiency than a similar device with a 5 nm-thick electron-blocking layer. Meanwhile, the variant with a 20 nm-thick electron-blocking layer generally emitted less light than the device with a 15 nm-thick electron-blocking layer. Such results are impossible to explain when only considering the electron blocking effect of the InAlN layer.

To gain an insight into what is really happening in these devices, we extended the widely used ABC model, which features rate equations and efficiencies for various recombination paths, to include carrier spill-over and hole-injection effects. This led us to carry out the first ever theoretical and experimental study to determine the electron spill-over and the hole-blocking contributions to efficiency droop and limitations in peak quantum efficiency.

Modelling efforts revealed that, as expected, more spill-over leads to higher droop, and it also showed that it produced a small decrease in the peak quantum efficiency, which occurred at a lower current density (see Figure 3 (a), (b) and (c)). Hole blocking levels also impact droop, but not as significantly as electron spill-over (see Figure 3 (d), (e) and (f)). However, the current density that the peak efficiency occurs at is found to be more dependent on hole injection than electron over-spill.



Figure 3.  Calculated QE versus injection current densities of LEDs for various spill-over currents ((a), (b), and (c)) and hole/electron concentration ratios ((d), (e), and (f)).  The QE curves for the LEDs with high spill-over current show more pronounced droop.  For the effects of hole injection, the peak QE depending on hole concentration seems to be more pronounced than that depending on spill-over.  (Reprinted with permission from Appl. Phys. Lett. 101161110 (2012). Copyright 2012 American Institute of Physics.)

These results show that the injection of holes into the active region, plus their transportation through it, play a major role in the efficiency droop and peak quantum efficiency. To mitigate droop, every well within an active region must be populated with a uniform distribution of electrons and holes with the same concentration. Changing the distribution of these carriers alters the electronic band structures and radiative and Auger recombination rates. 

It is not that challenging to realise a uniform distribution of electrons among multiple quantum wells, but when it comes to holes, this is very tough. In a conventional LED, the concentration of holes differs from well to well, and it increases the closer the well is to the p-side. To work towards the development of an LED that combines uniform hole distribution with efficient hole injection and effective hole transport in the active region, we fabricated devices emitting at three different wavelengths, to provide an experimental evaluation of the hole distribution within the active region. Devices were also produced with different indium contents in the p-type InGaN layers, because this changes the height of the potential barrier and enables a study of the hole ‘reservoir’ effect.

Increasing the hole barrier by switching from p-In0.015Ga0.985N to p-In0.035Ga0.965N led to an increase in the uniformity of the intensities of three luminescence peaks, with the well located furthest from the p-type region emitting more light (see Figure 4). This stems from improved hole transport, leading to a greater uniformity of this carrier within the active region.



Figure 4.  Electroluminescence (EL) spectra of triple-wavelength-LEDs with p-In0.015Ga0.985N and p-In0.035Ga0.965N layers.  The lower injection efficiency for the LEDs with a higher hole barrier diminishes with increasing current. Injected holes overcoming the higher potential barrier can then be transported farther, resulting in more uniform hole distributions among the MQWs

A valuable discussion of hole dynamics has to distinguish between hole injection efficiency and effective hole transport in the active region. Hole injection efficiency is governed by the total number of holes injected into the active region – it does not depend on which well captures them – while hole transport gives an insight into the distribution of holes among the wells with the total number of holes injected. At the interface between the p-type GaN and the undoped GaN barrier, there can be a reservoir for holes that fills up, prior to injection into the active region (see Figure 4). If the LEDs have a higher barrier, hole injection efficiency may be lower, but once holes overcome the potential barrier, they will gain energy. This increase in energy when entering the active region will influence the capturing efficiency of holes in each of the wells, which have different energies.

It is also possible that the potential barrier can limit hole injection efficiency, especially under low injection conditions. But this effect will diminish as the current is cranked up, and is expected to become negligible under high injection conditions. In that regime, injection efficiency is not strongly influenced by barrier height, but if the holes have overcome a higher barrier potential, they can be transported farther, leading to a more uniform hole distribution within the active region.

Our studies show that it is not essential to produce an unequivocal explanation for droop – which will hopefully come soon – to mitigate droop in LEDs and ultimately increase sales of light bulbs based on this technology. What we do show is that it is possible to combat droop by: reducing the carrier concentration in each well, so that Auger recombination does not kick-in at high injection conditions; making the carrier concentration in every well high enough to maximise the radiative recombination rate, while maintaining negligible Auger recombination; ensuring that in every well, the concentrations of electrons and holes are ideally identical; and trying to enable a uniform distribution of electrons and holes within the multiple quantum well.

Implementing those requirements is far from trivial. It may require increasing the number of wells in the LED and increasing the confinement of electrons in the active region. In addition, holes will have to be injected efficiency into the active region and transported across it very efficiently, so that this charge carrier has a fairly uniform population across the multiple-quantum-well region, even if it contains many wells.

Performing further fundamental studies and engineering of LED structures will help to uncover a route to such efficiency-droop-mitigating devices and spur the solid-state lighting revolution.

The authors wish to thank Jeomoh Kim, Suk Choi, Hee Jin Kim, Mi-Hee Ji and Md. M. Satter from Georgia Institute of Technology, Yong Suk Cho from the University of Houston, and Alec M. Fischer from Arizona State University for their contributions to this study of LED droop.

Further reading

J. Kim et. al.IEEE Photon. Technol. Lett. 251789 (2013)

S. Choi et. al.Appl. Phys. Lett. 101161110 (2012)

J.-H. Ryou and R. D. Dupuis Opt. Express 19A897 (2011)

S. Choi et. al.Appl. Phys. Lett. 96221105 (2010)

J.-H. Ryou et. al.IEEE J. Sel. Top. Quant. Electron. 151080 (2009)

J. P. Liu et. al.Appl. Phys. Lett. 93021102 (2008)