Uncovering the LED's darkest secret

LED makers are eyeing up new markets – automotive headlights, large-area displays and general illumination. To some degree, declining prices will help LEDs to penetrate these areas, but this in itself cannot guarantee success. Chips will also have to deliver very high efficiencies at high drive currents. And this means addressing a highly publicized problem that goes by the name of droop.

This term describes the lower efficiencies of blue, green and white LEDs at higher currents. GaN-based LEDs typically hit peak efficiencies at just 10 A cm–2, and can fall to half that value by 100 A cm–2. This is a major concern because today s high-brightness chips already need to operate efficiently at current densities well beyond 10 A cm–2, and further progress is absolutely critical if the community is to stick to LED roadmaps.

Tackling this issue is clearly important, and it is attracting a great deal of interest from researchers in industry and academia across the globe. This includes our team at Rensselaer Polytechnic Institute in Troy, NY, which has been studying the origin of droop for the last few years. We believe that we now understand the cause and how to combat the problem with a radically different LED design.

Some of our work on understanding the cause of droop was carried out in collaboration with Mary Crawford s group at Sandia National Labs, NM. We focused on the influence of dislocation density on LED efficiency, discovering that while dislocations reduce low-current efficiency, they do not impact the droop-causing mechanism at high currents.

At low currents, carriers are often lost to a trap-assisted process – Shockley-Reed-Hall recombination, which becomes more severe as the dislocation density increases. Cranking up the current initially improves efficiency through an increase in spontaneous emission. However, efficiency then falls due to increased competition from an additional carrier-loss mechanism that causes droop (figure 1).

Leaky LEDs
We have also teamed up with Samsung Electro-Mechanics Company South Korea. This effort exposed the origin of droop – electron leakage from the active region, which is driven by the polarization mismatch between the quantum well, quantum barrier and electron-blocking layers (EBLs).

Our explanation can account for droop s bigger impact at higher currents: the increase in drive voltage causes a greater fraction of the injected electrons to escape from the active region and reach the LED s p-side, where they recombine non-radiatively with holes at the p-contact (figure 2).

This theory has been verified with experiments that compare the light output from LED structures under electrical bias and optical excitation (see box "Probing LED structures"). We showed that there is a recombination mechanism taking place outside the quantum well, before making the connection between carrier leakage and polarization mismatch with numerical simulations.

Interfacial issues
Our efforts have been restricted to LEDs grown on the conventional plane of GaN – the c-plane. These devices feature strong internal electric fields that create large sheet charges at interfaces (see box "Feeling the inner forces").

The interfacial sheet charges hamper LED performance on two fronts. They increase the barrier to electron injection into the quantum wells, while simultaneously reducing the barrier to electron leakage from the quantum well and over the EBL.

Our simulations back up the hypothesis that sheet charges degrade LED performance via electron leakage and suggest that carrier loss can be combated by cutting interface charges. The calculations also indicate that the lack of heavy p-type doping – particularly of the EBL – can add to efficiency droop.

It is well known that the heavy doping of the emitter compared with the base in bipolar junction transistors impedes minority carrier injections into the emitter. Similarly, in LEDs, hole injection into the active region can be hampered by low p-type doping levels in GaN and AlGaN layers. This in turn leads to greater electron leakage.

Our belief that electron leakage explains the LED-droop phenomenon is not universally accepted; in fact, several different mechanisms have been suggested. Prominent among these is Auger recombination, which was proposed by researchers at Philips Lumileds (Shen et al. 2007).

Lumileds carried out photoluminescence experiments on c-plane GaInN/GaN double heterostructures and observed droop at high optical excitation densities. Analysis with rate equation models led them to assign Auger recombination to the cause of this droop in MQW LEDs.

To reach this conclusion they introduced an effective recombination thickness, which was chosen to be smaller than the physical thickness of the double heterostructure due to reduced electron and hole overlap in the quantum wells.

Feeding this recombination thickness into the rate equations leads to a prediction for a higher spontaneous emission rate in quantum wells with an electric field than those without. This contradicts the fact that electric fields in GaN-based quantum wells actually reduce spontaneous emission. Consequently, we feel that Lumileds over-estimates the importance of Auger recombination in quantum-well LEDs at current densities close to the onset of droop.

The proof of the pudding is of course in the eating. To this end, we are now putting the results of our simulations into practice and building LEDs with AlGaInN barriers. Replacing the conventional GaN barrier with AlGaInN, and the conventional AlGaN EBL with AlGaInN, gives us the freedom to tune bandgap and polarization, and ultimately cut the polarization mismatch and sheet charges at the active region s interfaces.

For the quantum barriers, we would grow a quaternary that has the same bandgap as GaN and matches the polarization of a typical quantum well. This is a tall order, however, because it s very difficult to grow AlGaInN layers that are rich in both indium and aluminum. However, significant improvements in device performance can be realized by cutting the polarization imbalance. Similar arguments can also be used for the EBL.

Our simulations reveal a tweak that can deliver almost all of the benefit of complete polarization matching, while using barriers and wells that only cut the mismatch by half. The trick is to slightly reduce the bandgap of the barriers, which leads to additional carrier confinement in the active region.

Boosting wall-plug efficiency
These modifications have produced some encouraging results. Light output increased by 20% at high currents, thanks to the suppression of droop (figure 3). Forward voltage is also cut, due to a shortening of the height of the barrier to carrier injection into the quantum wells. This benefit – which is another consequence of the reduction in sheet charges surrounding the quantum well – helps to bolster wall-plug efficiency by 25%. Other benefits include a cut in wavelength shift with drive current, thanks to a lower electric-field strength in the quantum wells.

What is clear is that the growing focus on the causes of droop is paying dividends. Our understanding has improved and designs now exist that look capable of solving the problem. Commercial devices with this feature should be in the pipeline, opening up new markets including the ultimate prize – a worthy replacement for the lightbulb.

Further reading
N F Gardner et al. 2007 Appl. Phys. Lett. 91 243506.
M H Kim et al. 2007 Appl. Phys. Lett. 91 183507.
M F Schubert et al. 2007 Appl. Phys. Lett. 91 231114.
M F Schubert et al. 2008 Appl. Phys. Lett. 93 041102.
Y C Shen et al. 2007 Appl. Phys. Lett. 91 141101.   

View pdf of article