The "city of sin" saw leading chip makers go on the offensive with claims of class-leading devices, and GaN-substrate developers championing breakthroughs in large-diameter free-standing material. Richard Stevenson reports from the 7th International Conference on Nitride Semiconductors.
Las Vegas oozes a party atmosphere and the beat wasn't lost on the delegates at ICNS. Great talks inspired a feel-good factor amongst the attendees and this positive outlook was only reinforced by registration figures that exceeded all expectations.
Klaus Ploog, the legendary MBE crystal grower and former academic of the Paul Drude Institute, Germany, kicked off the meeting by extolling the versatility of nitrides, although he did temper this by pointing out that many applications demand more development. While GaN LEDs are obviously a great commercial success, Ploog reminded everybody that work is needed to improve the efficiency of green emitters, as is the case for a whole host of other uses of the semiconductor.
Ploog believes that edge-emitting lasers are being held back by the lack of affordable GaN substrates, but said that recent progress in this area had been tremendous. The absence of a GaN VCSEL is also a blot on the nitride copybook and the key issues to address here are improvements to the mirror's reflectivity and conductivity.
On the RF side, problems such as high HFET drain-currents have been solved. However, low production yields are a concern, along with high chip costs that result from the use of expensive SiC substrates.
Ploog's themes were picked up by many speakers during the conference. As expected, the representatives from several of the leading LED makers – including those from Nichia, Cree and Lumileds – did not hold back in stressing the strength of their research and commercial devices, particularly in the context of white devices for general illumination.
If the key battleground is efficacy at 350 mA, then Nichia came out on top with a 134 lm/W white LED. This 1 × 1 mm chip can produce 361 lm at 97 lm/W and 1 A, and features a blue (λ~450 nm) LED.
Nichia's Yukio Narukawa explained that the latest LED also beats the company's previous record at 20 mA, which was announced last fall, and raises this efficacy benchmark from 138 to 169 lm/W. The previous record-holding chip was the result of efforts to boost extraction efficiency and featured an ITO contact in place of translucent metal and a hexagonally patterned substrate. The latest gains have come from research focused on lowering the forward voltage and increasing current-spreading through improvements to the device's epitaxial quality and design. The new record-breaking chip is better at maintaining efficiency with increasing drive current and has a wall-plug efficiency of almost 40% at 350 mA.
Although these performance figures are impressive, Narukawa believes that there is plenty of room for improvement. According to Nichia's calculations, phosphor-pumped LEDs made with blue emitters have a theoretical limit of 263 lm/W. This is significantly higher than the maximum for a violet 405 nm pumped equivalent, which is predicted to top out at 203 lm/W.
Cree also revealed its latest efficacy figures, which are just a whisker behind those of Nichia. Company co-founder John Edmond announced that Cree's best result now stands at 133 lm/W for an LED with undisclosed dimensions. This is a slight improvement over the 129 lm/W announced in September. Edmond also said that it should be possible to make light bulbs from a single GaN chip, although thermal issues must be overcome, and took the opportunity to slam mercury-containing fluorescents, which he described as "vile and poisonous".
The influence of chip size on performance was discussed in a talk by Lumileds' Frank Steranka. He explained that the company's LED design produces 115 lm/W at 350 mA from a 1 × 1 mm chip and also delivers 142 lm/W from a 4 × 4 mm chip. Steranka's underlying message was clear: claims of record efficacy need to include details of the chip's size.
Steranka also discussed the progress required to hit the US Department of Energy's stiff future targets, which include the goal of 150 lm/W at 2 A by 2012. According to him, hitting this milestone will require a hike in internal efficiency to 90%. Phosphor technology must also improve, as one-fifth of the emission can be lost in conversion.
Although the 2012 target is demanding, Steranka believes that reaching it holds the key to LED adoption in solid-state lighting. At present, incandescent and fluorescent lamps have a cost-per-lumen ratio of $0.03–0.05 and $0.06, respectively. To compete, the figure for LEDs needs to drop by a factor of 20. But increasing drive current to 2 A and efficacy to 150 lm/W will deliver a nine-fold improvement, which means that we can get into the right ballpark if we can halve the cost of making LEDs.
Although the blue and ultraviolet LEDs used to make white-emitting devices are delivering good efficiencies, their green equivalents are lagging behind. This problem, known as the green gap, is not well understood but is receiving considerable attention throughout the community.
One possible explanation for the decline in efficiency at longer wavelengths is the increased strength of the internal fields that separate electrons and holes in the quantum wells. However, Hiroshi Amano from Meijo University, Japan, told delegates that simulations with Semiconductor Technology Research software have revealed that turning to narrower quantum wells – the common approach for making these devices – produces LEDs that are only weakly affected by the built-in fields. In this type of structure, Auger recombination heads the loss mechanisms when the threading dislocation density is below 108 cm–2.
Amano said that a promising route to reducing non-radiative recombination involves the introduction of a GaInN layer beneath the quantum wells to reduce the lattice mismatch in the active region. Although ZnO is a potential candidate in terms of the crystal structure, it sublimes at nitride growth temperatures and Amano says that a better option involves the growth of a thick layer of InGaN as a different substrate. The epitaxial layer overgrowth technique can then be used on this epilayer, using growth conditions that estimate the effect of the underlying GaN.
Better substrates
Significant recent progress has been made in the field of GaN substrates, including the first 2 inch material produced by the ammonothermal method. Fumio Kawamura from Osaka University, Japan, explained that he and his co-workers produced this material by adding about 1 mol% of carbon to the standard mix of materials in the reactor, which substantially reduced the proportion of polycrystalline material. The 2 inch GaN, which is 4 mm thick and has a dislocation density of 2.3 × 105 cm–2, was produced with a new reactor design that features mechanical stirring and thermal convection to ensure a homogenous solution.
In the same session Hitachi Cable detailed its growth of 3 inch GaN using void-assisted-separation technology. Yuichi Oshima, from the company's advanced electronic materials research department, explained that the process used to make a substrate begins by depositing a 300 nm-thick GaN layer on sapphire at 1000 °C. A thin TiN layer is then added, which forms a nanoscale-sized mask when the wafer is annealed in a nitrogen and ammonia atmosphere.
A 600 µm-thick GaN layer is grown on this mask by HVPE at 1050 °C. Once growth is complete, this film self-separates from the sapphire during cooling to form a free-standing substrate. In trials, each of the four attempts at producing 3 inch material has been successful and all of the substrates have been free from cracks and a specular surface.
Non-polar substrates were also covered at ICNS, with Mitsubishi revealing its process for making 10× 10 mm m-plane material. Company representative Kenji Fujito explained that manufacture begins with HVPE-growth of c-plane GaN at 270 µm/h, typically for 40 h. The resulting GaN crystal boule is vertically sliced to form pieces of m-plane GaN. These substrates have a dislocation density as low as 4.4 × 105 cm–2, very few stacking faults and a thermal conductivity of 1.9–2.5 Wcm–1K–1, which is much higher than that associated with many types of GaN substrate. "Our activity is focused toward 2 inch diameter m-plane GaN," said Fujito, but it is clear that a different approach is needed to scale to that size.
One major beneficiary of Mitsubishi's low-defect-density material is the University of California, Santa Barbara (UCSB). One of their researchers, Jia Zhongyuan, explained that this low-defect-density material has held the key to the development of non-polar 400 nm LEDs producing 250 mW at 200 mA (driven in pulsed mode with a 1% duty cycle). Recently she has been fabricating longer-wavelength unpackaged versions that have produced 2, 1.5 and 1 mW at 450, 470 and 480 nm, respectively.
UCSB researchers have also been making LEDs on semi-polar material manufactured by Mitsubishi. Epilayers on this type of substrate have built-in fields that are typically just one-fifth of the strength of those associated with c-plane GaN and have the tremendous advantage of growth conditions that are very similar to those required for polar material. Pulsed results, to date, at 100 mA drive current include a 411 nm emitter producing 92 mW at 100 mA, and a 444 nm version delivering 67 mW.
Electrically pumped VCSELs
For several years GaN VCSELs have just been a distant dream, but this could soon change. Eric Feltin from École Polytechnique Fédérale de Lausanne, Switzerland, revealed that his groups' collaboration with the University of Southampton has produced electroluminescence from a VCSEL with an external quantum efficiency of 0.7%. The emission had a peak wavelength of 445 nm and a linewidth of 0.3 nm. The laser's cavity had a Q-factor of 1500, which is the highest Q-factor reported under electrical injection, said Feltin. According to him, if this figure can be doubled, there is a very good chance of producing a VCSEL that lases.
Although much of the focus was on optoelectronics, RF devices were also covered. Masahito Kanamura from Fujitsu detailed the company's development of alternative high-k dielectrics to SiN for improving transconductance and power gain in HEMTs. He revealed that the dielectric Ta2O5 can suppress leakage current and has led to the first MIS-HEMT with an output power over 100 W. Fujitsu's 38.4 mm device produced 143 W with 16.4 dB of linear gain at 2.14 GHz, and showed excellent RF power stability during a 150 h lifetime test.
Despite improvements in GaN RF performance, pessimism permeated through the rump session discussing the commercial future of these types of device. Replacing silicon LDMOS in cell-phone base-stations has been identified as the target market for many years, but this rival's continual performance improvements have thwarted sales wins for GaN. However, other potential markets for the technology, such as in the electronics circuit for light bulbs and cable TV amplifiers, were highlighted in this session. Philippe Roussel from Yole Developpement predicts that the total GaN market could be worth $101 million in 2010, but this equates to only 10,000 4 inch wafers per year, nowhere near enough to support the growing number of SiC suppliers. Lack of progress in scaling GaN to higher frequencies was also noted, with one wise-cracker claiming that the best root to progress was to replace nitrogen with arsine. But this session aside, the mood at ICNS was unquestionably upbeat.
Comments