Shedding Light On The Mystery Of LED Droop
Increasing sales in the LED market have been driven by improvements in chip performance that have enabled these devices to target new applications. These blue, violet and green emitters - which can also emit white light when used in combination with yellow dye - have already generated billions of dollars through lighting the keypads and displays of handsets, and they are now starting to generate additional revenue through deployment in the backlights of TVs, car headlights and general illumination.
Replacing the light bulb with an LED has been identified as a very important, long-term goal within this industry. Tremendous progress has already made, but one of the issues that remains is the realization of very high efficiencies at high current densities. According to early reports, LEDs tend to produce their peak external quantum efficiency (EQE) at current densities below 50 A/cm2 (a current density that corresponds to a drive current of 500 mA for a 1 mm x 1mm chip), and monotonically decreases thereafter. This decrease in efficiency, which has been given the moniker “droop", even occurs when the LED is driven with a low duty cycle, pulsed current that prevents device heating.
The origin of droop is attracting tremendous interest from researchers in industry and academia, including our team from Virginia Commonwealth University and Kyma Technologies Inc. Our studies have led us to conclude that one of the strongest candidates for droop is electron overflow - or spillover – that is caused by asymmetric carrier injection that stems from far more electrons being pumped into the device than holes.
Semiconductor growth facilities at VirginiaCommonwealthUniversity
Is Auger to blame?
Many ideas have been proposed for the cause of LED droop, and there is currently no consensus behind its origin. One of the first explanations for LED droop was carrier loss due to Auger recombination, a non-radiative process involving the interaction of an electron, a hole and a third carrier. Researchers at Lumileds deduced an Auger coefficient of 1.4-2.0 x 10-30cm6/s for quasi-bulk InGaN layers by fitting a recombination rate equation to photoluminescence data.
In 2009, computational scientists at the University of California, Santa Barbara, calculated an Auger recombination coefficient of 2x10-30cm6/s that emanated from the presence of a 2.5 eV upper conduction band. Interestingly, these simulations indicated that Auger recombination would be effective only in a narrow range of wavelengths around 500nm (~2.5 eV).
Other theoretical work, however, disagrees with the claim that Auger recombination is the dominant cause of droop. Efforts led by Jörg Hader, a University of Arizona theorist, led to a far smaller Auger coefficient of 3.5x10-34cm6/s. This calculation employed fully microscopic many-body models, and concluded that intrinsic Auger recombination should not be the major mechanism for the efficiency loss.
More recently, a publication by Han-Youl Ryu and coworkers from Inha University and Hanyang University, Korea, cast further doubt on whether Auger recombination can account for LED droop. These researchers found that in order to account for the large efficiency droop in LEDs, the required Auger coefficient is too large to be reasonable. It would have to be in the range of 10-27-10-24 cm6s-1, at least three orders of magnitude higher than the other reported values. The implication of their work is that Auger recombination is insufficient to solely explain the droop in InGaN LEDs.
Resonant photoexcitation has been used by several research groups that are trying to fathom the origin of LED droop. This measurement involves the probing of samples with a laser that is tuned to ensure photon absorption in the quantum wells only. Equal numbers of electrons and holes are formed in the wells, and it is possible to then determine the proportion of carriers recombining radiatively and non-radiatively. The efficiency degradation has not been observed at carrier generation rates comparable to electrical injection levels, indicating that efficiency degradation is most likely to be an electrical problem. It might be related to the carrier injection, transport, or leakage processes.
Our team believes that LED droop stems from electron overflow, which we also refer to as spillover. This is caused by relatively low hole injection, which may combine with the poor transport of this carrier resulting from its large effective mass. The term “spillover electrons" refers to the electrons that escape the active region without any form of recombination, and tend to wind up recombining in the p-GaN region or the pcontact. In this p-doped region carrier lifetime is incredibly short, due to magnesium doping. Our hypothesis is supported by our efforts that show the mitigation of the efficiency degradation in LEDs with thinner barriers.
The vast majority of studies on LED droop have been restricted to investigations of conventional, polar devices. In comparison, we have carried out a wider investigation, and looked at the effect of a p-type electron blocking layer (EBL) in InGaN LEDs on both c-plane sapphire and nonpolar m-plane bulk GaN substrates. Regardless of the polarity of the growth platform, the omission of the EBL leads to a reduction in the electroluminescence intensity by a factor of four to five (see Fig. 1).
We have also performed resonant optical excitation measurements on all of these LEDs using a laser that excites the carriers into the quantum wells only. This series of experiments, which were performed at a range of excitation intensities, show that the EBL has essentially no impact on the internal quantum efficiency of the LED.
This suggests that the lower electroluminescence intensity for the LEDs without an EBL has its genesis in carrier spillover (i.e., electron overflow triggered by poor hole injection, among others, and poor hole transport inside the multi-quantum well region).
It is also worth noting that substantial carrier spillover occurs in both non-polar and polar devices that do not have an EBL. This suggests that the polarization charge is not a major factor responsible for the efficiency degradation observed, particularly at high injection levels.
At the recent MRS Fall meeting that was held in Boston we announced that an additional efficiency droop could result from current crowding. This would mainly affect LEDs with lateral current conduction in the pcontact/ epilayer region. We found that the design of the contact architecture is not the only factor affecting droop – the choice of p-contact materials also plays an important role.
We arrived at these conclusions after comparing the performance of LEDs with a gallium-doped ZnO (GZO) contact, and those with a semi-transparent Ni/Au (5nm/5nm) contact. The results revealed two major benefits of the GZO contact compared to the metal one: an increase in light extraction by almost a factor of two, thanks to far greater light transmittance through the contact; and a significant reduction in droop. The device with the GZO contact had a droop of about 27 percent up to 3500 Acm-2, compared to a droop of about 37 percent for the LED with the thin Ni/Au contact.
The reduction in droop is caused by elimination of current crowding. This crowding is responsible for the nonuniform light emission seen in LEDs with a Ni/Au contact driven at high current densities. When devices with a GZO contact are driven equally hard, they produce uniform emission.
Junction heating can also contribute to the efficiency droop at high currents. This can be combated by surrounding the chip with a heat-extracting package, an approach that is already well developed by the leading players in the LED industry. One of the consequences of junction heating is a decline in internal quantum efficiency that results from an enhancement in non-radiative processes, including Shockley-Read-Hall recombination. Heating can also degrade contacts, leading to an increase in series resistance that drives down quantum efficiency and power conversion efficiency.
Our studies also show that m-plane LEDs can outperform their conventional counterparts. They produce higher electroluminescence intensity, and the efficiency droop with increasing current is smaller than it is for c-plane, polar LEDs (see Fig. 2).
Possible explanations to account for the negligible droop in m-plane LEDs are enhanced hole carrier concentration and lighter holes in m-plane orientation, leading to enhanced hole transport throughout the active region, and the lack of a polarization-induced field. But whatever the cause, the combination of a high quantum efficiency and its retention at high current densities makes the m-plane LED a very promising candidate for general lighting.
The work at Virginia Commonwealth University is supported by grants from the Air Force Office (ARO) of Scientific Research and the National Science Foundation. Partial support by ARO under Phase II W911NF-07-C-0099 contract for non-polar bulk development at Kyma Technologies, Inc., is acknowledged. The study of the GZO contact is partially supported by a grant from the Department of Energy, Basic Energy Sciences, through a subcontract from the University of Wisconsin.
X. Ni, et al, Applied Physics Letters, 93, 171113 (2008)
X. Li, et al., Applied Physics Letters, 95 121107 (2009)
J. Lee, et al, Applied Physics Letters, 95 201113 (2009)
J. Hader et al. Applied Physics Letters, 92, 261103 (2008)
H.-Y. Ryu et al. Applied Physics Letters, 95, 081114 (2009)