Droop Draws The Crowds At ICNS-9
Causes of LED droop and the progress of green lasers were two of the big topics at this year’s nitride conference. Richard Stevenson reports.
Aside from the weather, the delegates that went to Glasgow for the ninth International Conference on Nitride Semiconductors received an incredibly warm welcome. Maybe this is not that surprising – after all, Scotland’s most populous conurbation has worked hard to get itself known as ‘The friendly city’. However, this is not always the case, and at times the city’s reputation can be tarnished by the fans of the two big football clubs – Celtic and Rangers – who will never ever see eye to eye. As expected, the delegates were on far better terms than these rival supporters, but that didn’t mean that there wasn’t a whiff of confrontation in the air at some of the sessions.
ICNS-9 was held on 10-15 July at the Scottish Exhibition and Conference Centre, Glasgow.
One topic that continues to court controversy and encourage researchers to vigorously defend their own position is that of the origin of droop, the mysterious phenomena that accounts for the decline in nitride LED efficiency when the current in this device is cranked up. Talks on this topic were very well attended, and while there seems to be a growing consensus of opinion that droop involves some form of Auger recombination (a non-radiative process involving either two electrons and one hole, or two holes and one electron), the debate is far from over.
In contrast to the last two ICNS meetings in Korea and the US, none of the leading LED manufacturers spoke about droop during the conference. This may be because these leading chipmakers feel that they now have a good enough understanding of the causes of droop and they cannot justify the funding of further research into its intricacies. However, that is probably only part of the reason – in the past few years the efficiencies of state-of-the-art white-emitting LEDs have shot passed those of compact fluorescent bulbs, and chipmakers are now focusing on improving the manufacturing process of these nitride-based devices, so that they can drive a revolution in general lighting. In their absence, academics made the running in the droop debate, sometimes aided by commercial modelling software.
Droop: the case for Auger
Some of the most compelling evidence for Augerrecombination as the primary cause of droop was givenin a talk by Wolfgang Scheibenzuber from FraunhoferIAF. He and his co-workers, including members ofNicolas Grandjean’s group from EPFL, Switzerland,have determined the extent of Auger recombination innitride laser diodes, the cousins of nitride LEDs.
Scheibenzuber began his talk by describing a pair ofequations that he and his colleagues have used to model carrier and photon populations in the laser diodes. He then went on to explain that this work employs the well-known ABC model for determining the evolution of charge carriers – in this widely used model, which is defined in terms of the carrier density n, the carrier recombination rate is described as the sum of three terms: the Shockley-Reed-Hall non-radiative recombination rate, An; the radiative recombination rate, Bn2; and the Auger non-radiative recombination rate, Cn3.
According to Scheibenzuber, one advantage of working with lasers, rather than LEDs, is that it is possible to determine the injection efficiency of the device using optical gain spectroscopy. After extracting this value – in this case it was 68 percent – it is possible to separate carrier leakage from the recombination rate.
The European collaboration’s next step was to characterise their laser under very low driving currents and obtain a value for the A coefficient of 4.2 x 107s-1. They then studied the dynamics of the laser, such as relaxation oscillations and turn-on delays, and were finally able to extract values for the B and C coefficients of 3 x 10-12cm6s-1and 4.5 x 10-31cm6s-1.
Scheibenzuber concluded his talk by pointing out that the value obtained for the C coefficient agrees with the value calculated by Chis van de Walle and co-workers from the University of California, Santa Barbara (UCSB). These West-coast theorists believe that the forms of Auger recombination that dominate LED droop involve phonons and alloy disorder.
Further support for Auger recombination as the primary cause of LED droop came from a presentation by Ted
Thrush from the University of Cambridge. He presented electroluminescence intensity plots for a commercial LED driven at current densities from 0.0001 A cm-2to 100 A cm-2at temperatures ranging from 77K to 385K.
Thrush and his colleagues have characterized this LED with a transmission electron microscope: Its threading dislocation density is 1.4 x 109 cm-2and its active region features five wells with thicknesses of 3.2 nm, sandwiched between 4.8 nm-thick barriers.
The ABC model has been used to provide a good fit to the electroluminescence intensity plots, using the same B and C coefficients at all temperatures. Thrush said that consistency of the C coefficient over this temperature range indicated that the droop mechanism was not due to traps, leakage or a direct Auger process. He argued, however, that it was consistent with an impurity- or phonon mediated Auger process, as suggested by the theoretical work of the UCSB group.
Auger recombination was also blamed as the major culprit behind LED droop in a paper given by Dmitry Zakheim from Ioffe Physico-Technical Institute, Russia, who has been working with researchers from Epi-center and STR-Group. Through a combination of theory and experiment, this partnership from St Petersburg has shown that LED efficiency can be increased by switching from a conventional active region to one based on a short-period superlattice.
Zakheim and his co-workers have used the STR Software SiLENSe 5.0 to model electron and hole distributions in a conventional LED featuring five, 3 nmthick quantum wells sandwiched between 10 nm-thick barriers. This model – which includes drift and diffusion effects and can account for carrier delocalisation in the active region – revealed that the holes are not uniformly distributed through the active region, but predominantly loacted in the well nearest the p-type region (see Figure 1). According to Zakheim, this high degree of carrier localisation leads to high Auger recombination losses, and ultimately LEDs that suffer from significant droop.
Figure 1. A partnership between Ioffe Physico-Technical Institute, Epi-Center and STR-Group has modelled electron and hole distributions in two types of LED: (a) a device with a conventional active region, containing five, 3 nmthick quantum wells sandwiched between 10 nm-thick barriers (b) a device with a short-period superlattice active region comprising 2.5 nm-thick wells and barriers.
Modelling indicates that a far more uniform hole distribution is possible with a superlattice active region comprising 2.5 nm-thick wells and barriers. Such a structure is far better at combating droop: A standard flip-chip LED with a p-type contact formed from ITO and silver had an efficiency at 1 A that was 52 percent of its peak value; in comparison, a similar device with a superlattice active region delivered 76 percent of its peak output at 1 A.
It is also possible to suppress droop by improving the capability of the electron-blocking layer. According to Russell Dupuis from Georgia Institute of Technology, who has been working with Fernando Ponce’s group at Arizona State University, one way to do this is to replace the conventional p-type InGaN layer with p-type InAlN.
This US partnership fabricated a series of LEDs with In0.18Ga0.82N electron-blocking layers of differing thicknesses, and found that device performance peaked for a 15 nm-thick blocking layer. The team have simulated device performance with a modified version of the ABC model that includes a term for electron spillover. In their work, they found a good fit to their experimental data using values for A, B and C of 1 x 107s-1, 2 x 10-10cm3s-1and 4.5 x 10-29cm6s-1.
One of the great strengths of the widely used ABC model is its simplicity. However, it has its weaknesses too, according to Mary Crawford from Sandia National Laboratories, Albuquerque, NM. She explained that the radiative recombination in real devices is not actually proportional to Bn2 at high carrier densities, and claimed that more realistic modelling is possible by using momentum resolved carrier distributions, which allow direct implementation of bandstructure properties into the rate equations. This provides a more accurate description of carrier-carrier and carrier-phonon interactions, includes a treatment of carrier leakage and capture, and derives radiative recombination via bandstructure and carrier distributions. One tremendous benefit of this approach is that it can be used to consider contributions not present in the ABC model, such as recombination at defects with a microscopically determined, carrier-density-dependent A coefficient.
Droop: The case for carrier overflow
Going against the grain, Akio Kaneta from KyotoUniversity, Japan, did not attribute any of the droop heobserved in his light-emitting nitride samples to Augerrecombination. Instead, he pointed the finger at carrieroverflow for the droop in blue LEDs, and blamed carrierdiffusion from radiative recombination domains to nonradiativeones for the decline in green LED efficiency athigher drive currents.
Kaneta and his colleagues came to these conclusions after scrutinizing the photoluminescence produced by blue and green single-quantum-well structures with a scanning near-field optical microscope, a form of scanning probe microscope that can realise incredibly high spatial resolution in the optical domain by illuminating a sample through a sub-wavelength aperture formed at the apex of an optical fibre.
The researchers illuminated both samples through the aperture at two excitation powers, differing by a factor of 100, and recorded the local emission intensity. The two samples featured domains of strong photoluminescence that were several hundred nanometers across, surrounded by areas of weaker emission.
With the blue-emitting sample, the researchers found that the emission from the brighter areas saturated when they collected the photoluminescence through the aperture of their probe. This was not the case when a lens, which collects photoluminescence over a greater sample area, captured the emission. The conclusion of Kaneta and his co-workers: Photo-excited carriers overflow from localisation centres and are not captured by non-radiative recombination centres, thanks to the potential barrier surrounding them. This explanation is consistent with blue LED droop caused by a decline in carrier injection efficiency. In contrast, photoluminescence mapping of the green-emitting sample showed saturation everywhere, indicating that carriers move from radiative domains to areas that are non-radiative, probably due to threading dislocations.
Lasers for displays
Another highlight of ICNS-9 was the talks fromrepresentatives of Osram Opto Semiconductors and theUCSB spin-off, Soraa, which provided updates on theircompany’s performance of their laser chips for displayapplications, including picoprojectors. According toJames Raring from Soraa, lamps and LEDs arealternatives for the light source, but they deliver aninferior optical throughput, typically by a factor of three.
Raring explained that the vast majority of commercial green lasers on the market today and capable of serving picoprojectors employ some form of frequency doubling of an infrared source. Replacing such devices with single, green-emitting chips will lead to improvements in efficiency, compactness, ruggedness and speckle. However, according to Uwe Strauss from Osram OS, in order to produce an image with sufficient brightness, these green laser chips must have: An emission wavelength of at least 515 nm; output power of 50 mW or more; a minimum wall plug efficiency of 5 percent; and, in both the lateral and vertical directions, a single mode output. If a shorter wavelength source is used – for example, a 505 nm laser – the power output requirements are higher.
Osram can exceed these requirements. Its lasers, which it has developed on the conventional c plane, produce 70 mW at 522 nm with a wall plug efficiency of 5-6 percent, and have a spectral width of 1.8 nm. What’s more, reliability – defined as the time taken for the operating current to increase by 30 percent – is more than 1000 hours. In comparison, the continuous-wave output of Soraa’s 516 nm green lasers that are grown on unconventional planes of GaN now exceed 100 mW. Other characteristics of these packaged diodes included threshold currents and voltages of 125 mA and 5.9 V, a slope efficiency of 0.4 W/A and a wall-plug efficiency that peaks at 4.1 percent.
In Raring’s talk, he explained that one of the benefits of using semi-polar and non-polar planes is an increase in the radiative recombination rate, which stems from increased overlap of the electrons and holes in the quantum wells. These orientations also aid hole injection, thanks to a reduction in the effective mass of this carrier. However, he claimed that the most exciting aspect of these novel planes is the far greater design freedom that they enable.
Soraa’s efforts have not been limited to increasing output power – the West-coast start-up has also focused on improving the beam-quality of its single-mode green lasers. Divergence along the fast axis is 14-22 degrees, and in the slow axis it is 10-14 degrees. Similar improvements have been made to the divergence of the company’s single-mode blue lasers, which in packaged form have a slope efficiency of more than 1.6 W/A, and a threshold current and voltage of 30 mA and 3.9 V.
When emitting 500 mW, these diodes deliver a wall plug efficiency of more than 20 percent and hit 22 percent at 200 mW. Soraa has also made more powerful, multimode variants that can produce 1.4 W and have a wallplug efficiency of more than 23 percent.
Making affordable, semi-polar LEDs
Another company turning to novel nitride planes todevelop new products is the Korean outfit LG Electronics. High-power, semi-polar green LEDs is the goal, which will be built on r-plane sapphire due to thehigh cost of GaN substrates and their incompatibilitywith mass production systems.
Yoon-ho Choi from LG explained that there are two options for forming a-plane GaN on r-plane sapphire: A planar approach involving interlayers; or selective growth, which tends to involve lateral overgrowth. “We tried to optimise and combine these strategies," explained Chio.
The result is a process that begins by forming GaN seeds on the sapphire surface. Three-dimensional growth follows to form large islands, before an interlayer is added that creates a continuous GaN film across the surface. After optimising this approach, the dislocation density in the epilayer is 2-3 x 109 cm-2, and the stacking fault density is 1-2 x 105 cm-1.
According to Choi, these values are not good enough for making LEDs, so the engineers then apply a lateral overgrowth step to these wafers with a hexagonal pattern. Engineers from LG have made some very basic LEDs on this platform, using an architecture that Choi claims to be around 20 years off the pace – for example, it has no electron-blocking layer. Compared to an equivalent LED on sapphire, the a-plane variant is 40 percent brighter. However, Choi says that this improvement is partly caused by the air gaps in the underlying structure that increase light extraction.
The development by LG of green LEDs, as well as the improvements to green and blue lasers by Soraa, indicates that it may not be long before semi-polar and non-polar devices start making a significant commercial impact. So it will be interesting to see how much progress has been made when ICNS-10 comes around in Washington DC in 2014. And by then, maybe, just maybe, the community will be closer to agreeing on the cause of droop. But don’t bet on this just yet.
© 2011 Angel Business Communications. Permission required.
Toyota gears up HEMTs for HEVs
Toyota is developing vertical HEMTs for incredibly efficient, high-powerswitching in hybrid electric vehicles (HEVs), and lateral equivalents for lowerpower applications in cars. The motivation behind these efforts, accordingto Tetsu Kachi from Toyota Central R&D Labs, Japan, is that the siliconelectronics used today to convert DC power from the battery to an ACform to power the motor throws away 5-10 percent of the energy, whichis wasted as heat. Managing this requires a water cooling system, andthe amount of energy lost could increase because the trend is towardshigher and higher powers in HEVs. The first generation Prius used a 30kW motor, but second and third generations have increased this figure to50 kW and 70 kW. Even higher figures are being used by Lexus, whichhas a motor with a power of more than 150 kW in its model LS600h.
Losses in the DC-to-AC conversion process results from imperfect switching. The ideal: No current would be drawn in the off-state; in the on-state there would be no resistance; and switching speeds would be instantaneous. Switching losses promise to fall by a factor of about six by replacing silicon switches with those made from GaN. But that’s not the only benefit – switch to GaN and the electronics no longer needs its own dedicated water-cooling system, explained Kachi. Instead, it can tap into the one used for the engine, which one day might just require air cooling.
Smaller power modules operating at a few kilowatts are also needed in HEVs for air-conditioning, and emergency and outdoor applications.
Toyota is developing two types of GaN HEMT for the HEV: vertical devices for high-powers, which have the merits of high current density, high breakdown voltage and a high on-resistance; and lateral transistors for lower powers, which combine high frequencies with a low resistance, and high breakdown voltage and low cost. The vertical devices feature a novel U-shaped trench that is formed by dry etching with an inductively coupled plasma, followed by wet etching for 60 minutes at 85 °C. The sidewall forms a metal-oxide-semiconductor channel that works well up to 300 °C. Normally-off operation is realised with this device, which has a breakdown of 180 V. The goal is to increase this to 600 V. Efforts on the lateral devices have created transistors that are normally off above 3 V, and have issues related to current collapse and reliability of the gate insulator. Engineers at Toyota will work to improve this device and its vertical cousin.
Toshiba addresses the green gap in LEDs
One of the biggest problems facing the nitride community is the ‘greengap’ – the rapidly declining efficiency of green light emitters at longerand longer wavelengths. But this issue can be combated, according toToshiba’s Tamonari Shioda, by inserting thin AlGaN layers in the activeregion of a conventional device. This approach can increase the outputpower of green LEDs by a factor of almost ten.
Shioda explained that there are several issues associated with propelling LEDs to longer wavelengths: Deterioration of the crystal structure and increased phase separation, which can be addressed by improving the growth process; and an increase in electron-hole separation via the quantum-confined Stark effect, which can be mitigated by switching the growth platform to a semi-polar or non-polar orientation.
Toshiba wants to improve its green devices on c-plane sapphire, and to do this its engineers have worked to improve the band structure of the device. The primary goal of this effort has been to increase radiative recombination efficiency through greater electron-hole overlap.
Initial efforts in this direction involved the growth of multiple quantum well structures featuring 1.5 nm-thick AlGaN layers with a range of aluminium compositions up to 30 percent. Cross-sectional transmission electron microscopy analysis on this set of samples revealed no degradation in any of the structures. And probing these structures via photoluminescence showed that the greater the aluminium composition in the layer, the greater the suppression of the decline in efficiency at longer wavelengths.
Shioda and his co-workers have produced 600 μm by 600 μm LEDs with an active region featuring AlGaN layers, which were grown at the same temperature as the InGaN quantum wells. The output power of these LEDs increases with the proportion of aluminium in the interlayer. Driven at 20 mW, a 532 nm LED incorporating an Al0.3Ga0.7N layer produced an output of 12 mW at an external quantum efficiency of 25.9 percent. One downside of this structure is its higher operating voltage – insertion of this aluminium layer increases the forward voltage from 3.5 V to 4.6 V.
Switching faces improves high-speed HEMTs
Researcher Umesh Mishra from the University of California, SantaBarbara, opened the ICNS meeting by highlighting the performance ofGaN HEMTs and detailing approaches to take these transistors to a newlevel.
He kicked off his talk by saying that LEDs had made great progress in recent times, and now it was the turn of the electronics sector to make significant commercial headway.
Mishra then provided a brief overview of where the RF performance of HEMTs stand today: 13.7 W/mm at 30 GHz, 10.5 W/mm at 40 GHz, and 2.4 W/mm at 60 GHz. He also reminded the audience of the records for the fastest devices – a cut-off frequency of 220 GHz and a maximum oscillation frequency of 400 GHz. “I’m convinced that you will see 500 GHz in the next five years. It will happen," added Mishra, who explained that only a few years ago such frequencies were unthinkable.
However, he pointed out that these devices have a major weakness – the gain falls fast as frequency rises. This stems from parasitic capacitances and resistances that are too high. According to Mishra, the best way to address these issues is to switch from a conventional HEMT, which is produced using the gallium face, to a variant based on the nitrogen face. Making contacts to this novel device is easier, and drastically reduces the contact resistance.
One downside of HEMTs with a nitrogen face is that they have traps at the interface between AlGaN and GaN. However, by doping this region it is possible to lift these trap levels out of harms way. Mishra’s team have built MBE-grown, nitrogen-face devices with optically defined gates that produce 5.7 W/mm at 10 GHz. And MOCVD-grown variants fabricated on off-cut SiC, deliver 20.7 W/mm at 4 GHz, and 16.7 W/mm at 10 GHz.
The team has started to move to higher speeds, which is where the benefits of the nitrogen-face really come into play. Initial results include a cut-off frequency of 163 GHz.
The West-coast academic also spoke about the promise of GaN transistors for digital applications – his team has produced a 60 nm gate length HEMT with an InAlN back barrier that produces 2.77 mA/mm.
Activities at Transphorm, the company that Mishra co-founded in 2007, were briefly covered towards the end of the presentation. Normally-off HEMTs produced by this start-up are converted into a normally-on mode with an additional circuit, and can be used to make products for power switching. This includes three-phase inverters for photovoltaics that can operate at efficiencies of more than 98 percent.