LED Droop Continues To Bamboozle Researchers
As a fast-growing and dynamic country, Korea was an ideal host for the most recent International Conference on Nitride Semiconductors (ICNS-8). This popularity of this conference series is clearly on the up: almost 1000 delegates flocked to the meeting, and 27 different industrial exhibitors made their way to Jeju Island from across the globe, continuing the trend of growth seen from previous conferences in Las Vegas (2007) and Bremen (2005).
Despite the recent economic recession, the leading LED manufacturers - such as Cree, Osram, Philips Lumileds, and Samsung - dominated the conference program presentations. Their strong presence also provided further evidence of the robust nature of the nitride-based device market, which is expected to grow strongly, thanks in part to the emergence of solid-state lighting (SSL). Similarly, sustained worldwide governmental support for the development of energy-efficient SSL ensured that a high proportion of European, American and Asian academics were able to attend.
Thanks to these shared goals and interests, it was no surprise to see that the majority of contributions from academia and industry focused on light-emitting devices. Key themes that emerged during the week were the unsolved problems of ‘efficiency droop’ and the ‘green gap’.
Although LED efficiencies initially tend to rise with increasing current densities, they subsequently level off and then drop as the operating current increases. This ‘roll-over’ and the ‘droop’ that follows it are major obstacles to the development of high-efficiency, highbrightness LEDs for low-cost SSL. Unfortunately, while droop isn’t such a problem for near-UV devices, it becomes increasingly important for longer-wavelength blue and green emitters, which contain high proportions of indium in the InGaN active regions of the device. This increase in indium content tends to be correlated with drastic reductions in device efficiencies even at the optimum operating current, limiting device performance as emission wavelengths move into the green and yellow spectral regions.
Unfortunately, coming from the other end of the spectrum, the efficiencies of AlInGaP devices - traditionally used to create red, yellow and orange emitters - also falls off as wavelengths shorten and move towards the green, hence the term ‘green gap’. It’s a gaping hole that desperately needs to be plugged, especially if devices such as green lasers for projection applications are to reach the market. With this in mind, Steve DenBaars of the University of California, Santa Barbara (UCSB) opened the conference with a plenary talk on devices made using semipolar and nonpolar crystal orientations, which can help reduce or eliminate the piezoelectric fields that plague conventional c-plane (polar) nitride devices. The audience was especially keen to hear his views on the best film orientation for making green laser diodes.
DenBaars began by recapping recent achievements throughout the globe. The leading successes on c-plane material are a 526 nm laser from Osram and 515 nm version from Nichia. For non-polar lasers, UCSB and Rohm are leading the way with lasing at around 492 nm and 500 nm, while Sumitomo Electric recently produced a 531 nm laser on the unusual semipolar 2021 plane. Thanks to the rapid pace of progress using all of these orientations, no clear winner has emerged just yet; however, DenBaars says that the alternative orientations do offer some significant advantages. For example, although incorporating high concentrations of indium is still difficult, the UCSB group has shown that the absence of fields across the light-emitting InGaN quantum wells allows them to be grown thicker, and this in turn enables the elimination of the AlGaN cladding layer. Getting rid of this layer is a big plus point, because it means that the growth time for a 492 nm blue m-plane laser can be slashed from 12 hours to just 2 hours. Additionally, LEDs grown on semipolar orientations - where indium incorporation is easier - could help close the green gap: UCSB’s first unoptimized 562 nm LEDs showed a surprisingly high external quantum efficiency of 13.4 %.
Droop: the mysterious malady
However, UCSB’s nonpolar and semipolar LEDs still showed unexplained efficiency droop, despite the reduction of piezoelectric fields. Summarising the debate on droop, Shugo Nitta from Toyoda Gosei discussed the possible candidate mechanisms: Auger recombination, related to high current densities; electron overflow from the active region; poor hole injection; polarisation effects, which encourage carrier leakage; and problems related to defects and InGaN material composition.
Aurelien David confirmed Philips Lumileds’ existing position on droop, presenting both electrical and luminescence measurements that could be explained in terms of a composition-independent Auger process. Consistent with this finding, many groups reported success by reducing carrier densities in the active region: the Fraunhofer Institute, Germany, opted for wide quantum wells and low dislocation densities, while Cree’s “engineering solution" was to increase the chip area.
Taking a different approach, the group of Seong-Ju Park at Gwangju Institute of Science and Technology, which is working in association with Samsung LED, reduced droop by removing the AlGaN electron blocking layer that lies between the InGaN quantum wells and the p-type layer. This improved hole injection efficiencies at high current densities. However, Fred Schubert from Rensselaer Polytechnic Institute showed that p-doped electron blocking layers allowed good hole injection, while compositional control could be used to reduce the polarisation mismatch between the LED’s quantum wells and the barriers that separate them. This combination reduced the carrier leakage, thereby reducing droop. However, the strong droop observed in UCSB’s nonpolar LEDs suggested that polarisation couldn’t be the whole story.
No matter what, there were no reports of droop-free LEDs, regardless of the active layer design, film orientation, defect densities or measurement temperature – leading Jong-In Shim of Hanyang University and many others came to the conclusion that droop must be intrinsic to III-nitride materials, although it wasn’t clear why. At the packed rump discussion on this topic, most agreed that some loss mechanism occurred at high carrier densities, but the simple rate equation used to describe it might be inappropriate to describe all the complex, poorly understood recombination processes that seem to occur in the nitrides.
However, leading the discussion away from droop mechanisms, James Ibbetson of Cree said it was “time to move on from the name-calling" and instead think about “what comes next", arguing that high LED operating temperatures produced an equally important droop effect. Christian Fricke presented Osram’s route towards better thermal management, involving drilling tiny holes in the chip and creating buried p-contacts, permitting uniform current injection and light extraction.
However, he also emphasised the need to control costs and to concentrate on LED reliability and light quality, rather than limiting the focus simply to high-brightness devices, which are not appropriate for every application.
Slashing LED costs
One way to reduce costs is to increase the substrate size, cutting down on the wasted edge material and thus boosting yields. However, it can be harder to get uniform layers on larger substrates, due to wafer bow. With an eye on future trends, Aixtron demonstrated a new reactor configuration for 6-inch wafers, capable of producing GaN-on-sapphire epilayers with a reproducible thickness uniformity of 1%. Building up to full devices, the University of Cambridge group presented 455 nm LEDs with a high internal quantum efficiency of 58 % fabricated on 6-inch silicon, while IMEC showed considerable progress with their green LEDs on 4-inch sapphire.
An alternative route towards lowering costs is to increase LED efficiencies. This can be realized by reducing the high densities of dislocations – typically 108 cm-2 – that are produced when nitride films are grown on foreign substrates. While there were few surprises in this established area, the University of Cambridge did present data showing that dislocations could move and react within GaN films at growth temperatures, suggesting a future for suitably chosen annealing-based dislocation reduction processes.
Of course, low defect density bulk substrates would be an ideal solution to the dislocation problem, but their costs need to fall substantially. Fortunately, considerable progress towards commercially viable substrates is being made using several competing approaches: both the Chinese Academy of Sciences (with Suzhou Nanowin) and Unipress/TopGaN announced plans for mass production of substrates grown by hydride vapour phase epitaxy (HVPE), the latter using high-pressure solution growth on top of a HVPE-grown ‘seed’ crystal.
In contrast, Osaka University, Japan, announced breakthroughs in bulk GaN grown by the sodium-flux method, reporting inclusion-free 2-inch c-plane GaN wafers with dislocation densities of around 3 x 104 cm-2 and announcing their plans to scale up to 4-inch substrate production. However, the most impressive data came from the Polish company Ammono Ltd., who reported flat 2- inch c-plane GaN substrates grown by the ammonothermal method with dislocation densities as low as 5 x 103 cm-2. Despite the low growth rates associated with this method, high-volume production and commercial release of these 2-inch substrates is scheduled to occur, once a sufficient stock of seed crystals has been built up. All of these methods also have the potential to lead to the commercialisation of bulk nonpolar and semipolar GaN. Additional progress on low defect density c-plane bulk AlN was also announced by several groups, including the Russian start-up Nitride Crystals, who reported the sublimation growth of c-plane AlN using AlN-on-SiC seeds in a tungsten crucible. The resulting 2-inch substrates have a ‘block’ structure that enables the realization of areas with dislocation densities as low as 102 cm-2. Substrate sales will begin once seed crystal stock has been built up.
Although the availability of all large-area native substrates is currently limited and the cost is high, this route towards low defect density material looks set to be especially important for UV-LEDs. Although deep-UV LEDs were announced based on nonpolar (NTT Corporation), semipolar (University of South Carolina with NITEK) and c-plane (RIKEN and Sensor Electric) heteroepitaxial films, the high defect densities were linked to high forward voltages, low output powers and highly resistive p-type material. Elsewhere at ICNS-8, only modest progress was reported on the reduction of defect densities in heteroepitaxial AlGaN and AlN films, which is needed for high external quantum efficiencies and output powers in the UV.
Low defect density, insulating GaN and AlN substrates are also promising for GaN-based electronics, as dislocations are thought to lead to high leakage currents and degradation in GaN-based HEMTs, which are used in high-frequency and power switching applications.
Reflecting these concerns, many groups presented device reliability studies. One clear trend is towards the use of InAlN, and even InAlGaN, in order to circumvent some of the strain and defect issues associated with AlGaN-based transistors. New designs were also presented, including the ‘normally-off’ transistors based on nonpolar crystal orientations that were developed by Umesh Mishra’s group at UCSB, while Cree announced a series of improvements to their high-power, high-frequency transistor range, bringing the prospect of high-power near-terahertz applications for the nitrides within reach.
It will be interesting to see how much progress is made on this front, and also on the quests for an understanding of the green gap and LED droop when the nitride community reconvenes for the ninth meeting of this series in Glasgow in 2011. If this meeting is anything like the previous affairs, it promises to be a have a real buzz about it.