The solid-state lighting revolution will be spurred by plummeting LED costs and improvements to the quality of emitted light. Success on these fronts could be aided by refinements to existing technologies; the introduction of GaN-on-silicon LEDs; a move to colour mixing of red, green and blue LEDs; and a switch from LEDs to lasers at the primary lighting source. All options were discussed at the International Conference on Nitride Semiconductors. Richard Stevenson reports

The Gaylord Conventional Centre, situated on the bank of the Potomac river, hosted the tenth International Conference on Nitride Semiconductors. This meeting, which was held from 25th to 29th August, attracted about 900 delegates.

It can be argued that the nitrides, like graphene, are worthy of the moniker ‘the wonder material’. Attributes of GaN and related alloys include incredibly efficient emission in the violet and blue, a very high electric field strength that has led many to tip this material as the future for power electronics, and a power density that is far greater than that produced with the likes of GaAs and InP.

However, that is not to say that the III-Ns are faultless. This family of materials have a significant lattice mismatch with one another, which hampers the development of many devices, particularly VCSELs and multi-junction solar cells. Another major impediment to device performance is the lack of a straightforward process for forming a boule of GaN, which could be sliced to form native substrates, while even the most impressive device to date, the LED, still needs to make headway to unlock the door to a revolution in solid-state lighting.

Developments that could spur this revolution were a prominent feature at the tenth International Conference on Nitride Semiconductors (ICNS), which was held in Washington from 25th to 29th of August. Delegates attending this meeting learnt of: aggressive cost reduction of GaN-on-SiC LEDs; the development of high-performance GaN-on-silicon LEDs; improvements to the efficiency of green and yellow LEDs, which will aid the development of high-colour quality white light sources; the opportunity for lasers to replace LEDs and make solid-state lighting more efficient; and advances in the efficiency of blue and green lasers.

Cutting costs on SiC

One company that is already making lighting products with its own LEDs is the US firm Cree. At ICNS, one of Cree’s senior scientists, Hua-Shuang Kong, detailed improvements in the efficacy of the company’s LEDs. This firm is also trimming production costs, which will help to make a more compelling case for solid-state lighting. The fall in the cost per lumen realised by this vertically integrated firm – Kong described Cree’s activities as spanning “from dust to device to market” – was 45 percent in 2011, 40 percent in 2012, and should be 43 percent this year.

This increase in the bang-per-buck has been assisted by a continual improvement in the efficiency of the LED. Back in February, Cree claimed that it had broken the efficacy record for an LED driven at 350 mA, with a device emitting 276 lm/W at a colour temperature of 4401K.  And the performance of LEDs coming off of the production line is not that far behind, with the MK-R series that was launched in late 2012 delivering 200 lm/W.

In Kong’s opinion, the high level of performance stems from an excellent chip architecture and a high-quality SiC substrate, which forms a better platform than the more widely used material, sapphire. At Cree, LEDs are grown on single crystal 4H SiC that combines a high thermal conductivity – it is 4.2 W cm-1 K-1 in the a direction and 3.7 W cm-1 K-1 in the c direction – with excellent transparency in the visible. Thanks to an absorption coefficient of less than 1 cm-1, these substrates enable the fabrication of an LED with a very high extraction efficiency.

To increase light extraction, Cree’s engineers at use a ray-tracing model to optimise chip geometry. By considering various aspects of chip design, including bevel cuts, internal mirrors, flip-chip geometries and surface texturing, engineers have increased the light extraction from a 1 mm2 chip to 88 percent. According to Kang, state-of-the-art room-temperature performance of blue LEDs emitting at 447 nm is now an output of 782 mW and an efficiency of 79 percent at a 350 mA drive current.

Cree’s blue LEDs are combined with yellow phosphors to make cool-white emitters, while warm-white variants use this chip to pump yellow and red phosphors. Both forms of white emitter are inserted into lighting products, where efficacy is compromised by thermal, optical and driver-related losses, according to Kong. He revealed that a chip emitting 115 lm/W at a colour temperature of 2700K delivers just 75 lm/W in a fixture, while a 6000K LED with a figure of 160 lm/W on the data sheet creates a fixture producing only 104 lm/W.

Fortunately fixture efficacies, like those of the LEDs, are improving, while purchase prices are falling. Back in 2008, Cree’s LR24 Troffer retailed for $400, had a colour-rendering index (CRI) of 90, and produced 3200 lm at 65 lm/W. Fast forward to today, and the same light output is possible at 100 lm/W, while the purchase price has plummeted to $159.

Increases in the efficacy in lighting fixtures are expected to continue, driven by gains in LED efficiency. By 2015, Kong expects LEDs with colour temperatures of 6000K and 2700K to produce 210 lm/W and 151 lm/W, respectively, while fixtures based on them will deliver 174 lm/W at a CRI of 75 and 125 lm/W at a CRI of 83, respectively.

Silicon-based LEDs

One promising option for accelerating the cost reduction in LEDs, and ultimately speeding the introduction of far more affordable solid-sate lighting, is to switch to growing these devices on a silicon substrate.

“To replace the fluorescent lamp, we must reduce the cost [of LEDs] more and more,” proclaimed Youngjo Tak from Samsung Advanced Institute of Technology, Korea. In his talk, he argued that that although the cost-per-lumen for LEDs grown on sapphire is falling, it will plateau in 2015. So, in the long-term, alternative platforms are needed to make light-emitting chips more affordable.

Switching from 4-inch or 6-inch sapphire to 200 mm silicon is an attractive option, because it can cut costs by 31 percent and 48 percent, respectively. That's partly because the sapphire substrates are not that cheap – according to Tak, it is typically $500 for 6-inch sapphire and $1500 for an 8-inch equivalent – and also because LEDs grown on silicon can be processed in depreciated silicon fabs.

The development of GaN-on-silicon LEDs is no longer in its infancy. Initially, the efficiency of these devices fell a long way short of the incumbents, but the difference is closing fast. “[It’s now] less than 10 percent,” claimed Youngjo Tak.

He is not the only researcher to claim that this gap is minimal: “GaN-on-silicon LEDs have comparable efficiency to GaN-on-sapphire,” revealed Martin Albrecht from the Leibniz Institute for Crystal Growth, who has been working with engineers at Osram Opto Semiconductors and the MPIE, Germany.

Albercht unveiled details of performance of this team's latest LEDs. They require just 2.91 V to operate at 350 mA, and when combined with phosphor technology, they produce 104 lm/W of warm-white light with a colour-rendering index of 83. The efficacy increase over last year's devices is mainly due to a 6 percent fall in output voltage, which has stemmed from improvements to the epitaxial layers and the quality of the quantum wells.

Tak, in contrast, did not disclose the latest results from his group. But he did reveal that the performance of these LEDs, which are grown on 200 mm silicon, is much higher than that of the devices announced at the 2011 ICNS meeting. Back then, Samsung's researchers extracted 580 mW from a 1 mm by 1mm LED grown on 4-inch silicon. The device was driven at 350 mA under a forward voltage of 3.2 V.

Highlights of Tak’s talk included insights in the difficulties of growing GaN-on-silicon LEDs, and how to overcome them. One of the biggest challenges stems from the significant lattice and thermal mismatches between the substrate and epitaxial layers. These differences can cause the wafer to bow, or even crack.

A common approach to addressing this is to turn to a buffer structure that fine-tunes the stresses and strains in the epistructure. According to Tak, with this approach Samsung's engineers can realise a bow of less than 30 μm. They have found that in order to realise a very low bow, it is critical to start with a silicon substrate with a low degree of warp.

One downside of the buffer structure is that in order to be effective, it can have to be quite thick, and that adds substantially to the time and cost of producing an LED epiwafer. But Samsung’s engineers are addressing this: Their latest LED epistructures are about 5 μm-thick, compared with 8 μm of growth for the previous generation of devices. “They have a similar processing time to sapphire,” said Tak.

ICNS featured three posters sessions, which were all well attended.

Enhancing colour quality

It is possible to build white light sources with a high colour quality by switching from a phosphor-based approach to the mixing of the output of red, green and blue sources. Unfortunately, this form of lighting system, which can be used to make projection sources, is currently held back by the efficacy of the green LED, which is far behind that of its blue and red counterparts.

 

This weakness is known as the green gap. While GaN LEDs can be very efficient in the blue, and GaAs-based devices can produce high efficiencies in the red, neither maintains their performance when the composition in the quantum wells is adjusted so that it emits in the green (or the yellow). The arsenide material system will never emit efficiently in the green, because carrier confinement is so weak in this spectral range, but it is possible that nitride LEDs could emit more efficiently at longer wavelengths.

One of the weaknesses of GaN-based devices is that they are plagued by strong internal electric fields, but this can be mitigated by turning to thinner quantum wells that increase electron-hole overlap and therefore enhance carrier recombination. The performance of these LEDs is also hampered by deterioration to material quality that occurs when indium content is increased in the well, so that its emission is pushed to longer wavelengths.

The encouraging news coming out of ICNS is that a novel active region can combat the green gap. Speaking on behalf of the Toshiba’s Corporate R&D Centre, Rei Hashimoto revealed that a modified active region – it contains 3 nm-thick InGaN quantum wells capped with a 1 nm-thick AlGaN barrier, prior to the deposition of a 10 nm-thick InGaN barrier – enables the fabrication of very efficient yellow-emitting LEDs. Driven at 20 mA, these devices produce a peak emission at 570 nm, and emit 8.4 mW and an external quantum efficiency of 19.3 percent. Indium content in the well is estimated to be about 25 percent, and less than 1 percent in the barrier.

Hashimoto argued that one of the key benefits of the new design of active region is an improved surface flatness. He and his co-workers optimised the growth conditions for the quantum wells and barriers, and by selecting the ideal growth temperature they eliminated indium-rich clusters at the surface.

An alternative approach to forming a solid-state, green-emitting source was outlined by Thomas Lehnhardt from Osram. He revealed the results of a project involving using a blue LED to pump a green-emitting active region made from InGaN quantum wells. This approach culminated in the construction of 1 mm by 1 mm device emitting at 535 nm with a wall-plug efficiency of 22 percent. Efficacy for this device is 127 lm/W, and could rise to 138 lm/W by inserting Osram’s latest LEDs, which have a lower forward voltage for the same drive current.

Lehnhardt made a strong case for using this form of complex, green-emitting structure in preference to a direct-emitting green LED. He pointed out that using a blue-pump source makes this device less susceptible to droop, the decline in the efficiency of a nitride LED at higher current densities. What’s more, he explained that the blue LED has a higher electrical efficiency than a green-emitting variant, despite its higher bandgap; and that the green emitting, optically pumped structure does not suffer from an electron-hole imbalance, which is detrimental to a conventional green LED.

Trials of this novel green emitter have involved the growth of epistructure featuring a 40 period multi-quantum well. This is estimated to deliver an external quantum efficiency of 50 percent. To boost green emission and prevent significant output in the blue, three of these multi-quantum well structures were stacked on top of one another.

From LEDs to lasers

A radical alternative for solid-state lighting is to use lasers, rather than LEDs, as the primary light source. Jonathan Wierer from Sandia National Laboratories, Albuquerque, outlined this proposal, which could use a laser to pump a phosphor. He explained that he and his co-workers have tried this, and found that the colour quality is comparable to that produce by LED pumping: “Despite spiky spectra, you get good colour rendering.” Wierer also reminded the audience that this approach is already being pursued by the German automaker BMW, which is developing laser-based headlights.

The attraction of lighting with lasers, rather than LEDs, is that the former class of device is not plagued with droop. This opens the door to high efficiency at high current densities, so fewer chips are needed in a fixture, and those that are employed require less thermal management. Although Weirer calculates that LEDs will get more efficient, he believes that lasers will also improve, so laser-based lighting will continue to have the potential to offer the greatest efficacies in the future.

Delegates had an insight into the efficiency of the latest lasers in a talk given by Osram’s Adrian Avramescu, who revealed significant advances in the performance of the company's green and blue laser diodes developed in the lab.

According to Avramescu, the company’s green lasers, which emit around 520 nm, now deliver a kink-free output up to 250 mW and have a wall-plug efficiency of 8.7 percent at 150 mW. Meanwhile, the company's blue lasers produce an output up to 4 W, and at 1.6 W they have a conversion efficiency of 30 percent.

Avramescu pointed out that the main applications that Osram intends to target with its green lasers are small projectors, speciality lighting, assistance with surveying and head-up displays. Small projectors, which could soon feature in mobile phones, combine red, green and blue lasers with a two-axis mirror to form a technology known as flying-spot laser projection. “You build your image up pixel by pixel,” explained Avramescu.

Requirements for lasers used in small projectors include an output of about 80 mW, or 20 lumens. Blue lasers that the company launched in 2009 meet this, while the direct green variants that hit the market in 2012 fall short of that mark, producing 50 mW, equating to 12-13 lumens. In addition, the emission profile from these chips is not the ideal Guassian profile that is preferred for projection applications, due to interactions with the substrate.

All these weaknesses have now been remedied in the lab, due to various efforts at understanding the behaviour of the green laser diode. First-generation devices produced a sub-linear output above 50 mW, and investigating this decline in efficiency led the engineers at Osram to discover that the laser's differential gain varies with temperature. Self-heating and a high threshold current are partly to blame, along with imperfection injection.

Research efforts were also directed at material issues. Dark spots were found with dimensions of 1-20 μm, alloy fluctuations were uncovered at 50 nm to 500 nm length scales, and transmission electron microscopy uncovered defects in the quantum wells, such as dislocations.

Improving the quality of the material and making proprietary modifications to laser design enabled the fabrication of higher output green lasers with improved beam quality. At 80 mW – the output power required for small projectors – efficiency is 7.5 percent and 7 percent at 25 °C and 60 °C, respectively.

Avramescu also explained that Osram's more powerful blue laser is suitable for projectors in homes and offices, which require 2000 lumen sources. In 2012, the company launched a 1.4 W blue laser in a TO56 package, and now it has raised the bar to 2.5 W at 25 °C. Packaging is very important at these power densities, and by optimising this it is even possible to crank the output up to 4 W. However, Avramescu explained that this involves driving the chip in an “over-stressed” regime.

It will be interesting to see how laser performance improves over the next year, and whether efforts to use this source for general lighting take off. Progress occurring over this timeframe will not be reported at ICNS, because this is a biannual meeting, but the international nitride community will get together next year at the International Workshop on Nitride Semiconductors. This will be held in Poland, in the last week of August, 2014.