Plenty of Opportunity in Billion Dollar HB LED Materials Market
High brightness LEDs’ need for exotic substrates, metalorganics, high-temperature packaging materials and custom phosphors means that materials may account for up to 30% or more of packaged device cost, according to some estimates. That’s in what Strategies Unlimited figures is already a $4.9 billion market for HB LEDs, projected to see 19% average annual growth long term, to exceed $12 billion by 2013.
Total LED materials cost estimates range all over the map, as information is closely held and producers’ processes still differ widely, with different substrates and especially different packaging solutions. But a 20% estimate for the bill of materials means the HB LED market is currently somewhere around $1 billion, with realistic expectations for double-digit growth. Naturally, materials prices will trend down with sector maturity and growing volumes— but the push for improved performance and yields to enable a real high-volume solid state lighting market means plenty of opportunity exists for major innovation as well. Chip makers are looking for larger wafers and exploring ways to engineer the wafer surface for better epitaxial growth. Suppliers have also recently introduced improvements ranging from replacing onboard gas cylinders with central gas deliver systems to finding ways to standardize phosphors.
Volume Production Moves to 4-inch Wafers, Looks at 6-inch
Volume producers are moving from 2-inch to 4-inch wafers and are talking about 6-inch within two to three years. Adoption of 4-inch sapphire wafers will be widespread in 2010, reports Rubicon Technology (Franklin Park, Illinois)VP of engineering Sunil Pahtak. He notes that 6-inch wafers are already in pilot volumes, starting production in 2010, and will see widespread adoption by 2011-2012. Rubicon has shipped more than 30,000 epi-polished 6-inch substrates in the last two years. Prices for the larger wafers have looked daunting, compared to the $10 or so for 2-inch sapphire, but suppliers argue that higher volumes, and some sector agreement on standard specifications, should help bring down costs.
Equipment suppliers report that most producers aiming at high volumes for LCD backlight markets are going to 4-inch wafers. They say Samsung and Toshiba are starting production at 4-inch and looking seriously at 6-inch, and most LCD makers in Taiwan making the HB LEDs for their internal use are starting at 4-inch or planning to convert.
A batch of 3-inch-diameter SiC wafers is loaded into a high-temperature oxidation furnace in Cree’s new Advanced Device Clean Room facility in Research Triangle Park, NC. (Photo: Cree)
Users Look to Surface Processing the Substrate to Improve Yields
The next option is likely some sort of surface processing of the substrate to allow growth of better GaN films. Bulk GaN might be the ideal substrate to eliminate all lattice and thermal mismatch, but it still costs a fortune. So chip makers are trying innovative ways to treat the surface to ease the MOCVD process instead. “Bulk GaN gives a huge advantage in performance,” says Keith Evans, CEO of Kyma Technologies (Raleigh, North Carolina). “But the world doesn’t have much experience in making it, so we’re still coming down the volume learning curve. But in the meantime there’s a driving force for an intermediate solution.”
Source: Kyma Technologies
The first intermediate solution now getting a serious look is patterned sapphire. Texturizing the surface with different schemes of wet or dry etch improves the quality of the epi layers grown on top, for better light extraction efficiency, though why this works and how best to do it remains unclear. Also a possibility is adding a layer of AlN layer on top of the patterned sapphire, with initial results reportedly showing increased brightness and wavelength uniformity. “We strongly suspect this may bring down costs too,” says Evans, “since it can eliminate some of the MOCVD steps.” The AlN-coated sapphire can be ramped to temperature faster and needs no buffer layer, so the epi process can skip straight to high temperature deposition of the GaN, improving throughput of the reactor. The next logical step would be adding the GaN layer as well, though high volume HVPE equipment for depositing the GaN is not yet commercially proven. Kyma is working with a tool vendor on more stable and higher volume processing, though it is no longer considering getting into the tool business itself. “We hope and plan to have a first volume supply of AlN and GaN on sapphire substrates in 2010,” says Evans.
But long term, he argues, bulk GaN costs are likely to come down significantly with volume production, driven largely by strong growth in demand from other applications, such as power electronics for everything from hybrid cars to server farms, and major improvements in the process tools for production. Evans estimates that continuing on the current cost curve, a 2-inch bulk GaN wafer is on track to drop from $2,000 currently to perhaps $50 by 2015.
Still a Possibility of Radical New Approaches
In addition, there’s the more radical alternative of putting a GaN surface on the sapphire, then taking away the sapphire and using just the freestanding GaN foil alone as the substrate for growing the LED layers. Goldeneye (Carlsbad, California) grows a 30-micron layer of GaN by HVPE, then removes it by laser liftoff. These thin GaN foils (a centimeter square) can be very quickly ramped to growth temperature and cooled back down, allowing an epi growth cycle as fast as 30 minutes, claims Scott Zimmerman, VP of technology. He sees a potential 10X increase in throughput by eliminating the need for nucleation, n-layer and ramp time.
GaN on GaN foil from Goldeneye (Photo: Goldeneye)
Goldeneye needed to buy a light source for an application it was developing for a customer. However, the company discovered that it could not get the quality or quantity of LEDs it needed, so decided to make its own. After some work growing device layers on both templates and foils in commercial reactors, it decided foils provided much better yields, but existing reactors were not well suited to the task. So it began work with a supplier on a new version, without the usual rotating disc, designed specifically for very rapid temperature change. The new tool is being delivered this month. “We’re about to turn it on and see what happens,” says Zimmerman. The first goal is to demonstrate good throughput and yield numbers from the reactor, then to get back to making light sources.
The more appropriate model for LEDs for general lighting is probably not the semiconductor industry, but the thin-film solar industry, with its big areas and low costs, argues Zimmerman, as products that not requiring sub-micron resolution don’t need to be processed as wafers. Goldeneye’s GaN-on-GaN foil device ends up looking more like an inorganic version of an OLED, and the company is working on eliminating other back-end processing costs as well, with a transparent conductive oxide layer also grown with MOCVD, and ceramic phosphor that attaches directly to the device.
But radical changes get harder to make as the current technology infrastructure gets more established. “Customers are telling us, and we believe as well, that if silicon or some other substrate doesn’t emerge as viable within a couple of years, the train will have left the station,” says Veeco Instruments’ Jim Jenson, VP of marketing for the MOCVD business. Veeco is headquartered in Plainview, New York. Jenson notes that the market is exploding. One thousand MOCVD tools are in production and hundreds more being added each year, so it will get harder and harder to change as the incumbent technology has so much built up investment in development.
"With present SSL costs 20X to 50X higher than DOE and general lighting costs targets of $1.00/klumen, it is yet to be proven whether the incumbents using existing MOCVD reactor approaches will find themselves on the wrong train," counters Zimmerman.
IMEC (Leuven, Belgium), meanwhile, is working on controlling thermal mismatch of GaN grown on silicon with careful stress engineering, aiming to apply the same GaN on 200mm Si platform for both power electronics and LED lighting. “GaN on Si is the option that offers the best cost reduction possibilities in the end compared to any other choice,” says Marianne Germain, GaN program manager for IMEC. “Si is the only one that allows for large wafer diameters and lower substrate cost. GaN templates or bulk still require significant technology development for defect reduction and wafer size enlargement.”
The research organization uses an in situ growth monitoring system for growing GaN on 100-150mm wafers for both power devices and LEDs. It claims to have reduced total defect density by an order of magnitude, to _3E8/cm2, or roughly comparable to that on sapphire, by using several intermediate layers of AlGaN, and a SixNy interlayer between the Si and the GaN.
Gas Delivery Systems Can Increase Tool Uptime by 10-15%
Uptime of the MOCVD equipment can be increased 10-15% per tool by installing a central gas delivery system such as commonly used in semiconductor fabs, argues Joe Reiser, business director for metalorganic technologies at source material supplier Dow Electronic Materials. The LED epi reactors typically still use onboard cylinders for their precursor gases, but with today’s larger chambers and higher volume production, changing out the cylinders requires increasing downtime. And the individual onboard systems mean the recipes may need to be subtly tweaked for each individual tool.
A few fabs are starting to install such central gas delivery systems for their MOCVD processes, mostly in new facilities. But retrofits can also offer savings. “There are some upfront costs, but most agree payback would be about one to two years,” says Egbert Woelk, applications and technology development manager for metalorganic technologies, noting that the industry wastes millions a year shuffling cylinders around.
Purity and consistency have made major strides and are no longer much of an issue, argues the major supplier of group III trimethyl Ga, In and Al, and the once raging patent wars have quieted down, as cross licensing deals have been worked out.
More Standard Phosphor Solutions Can Simplify Production
Phosphors for enabling the white color and efficacy of LEDs for solid state lighting have headroom for perhaps 5-10% per year improvement in cost and performance, says Yi-Qun Li, EVP, CTO and co-founder of Intematix Corp. (Fremont, California), noting particularly the need for better sustained performance in heat, and better color rendering. More consistency in different producers’ HB LEDs would also help, so a custom phosphor solution did not have to be made for every customer. One big Asian LED maker reports keeping dozens of different phosphors on hand, so the appropriate phosphor can be chosen to adjust the color of each bin of chips to improve yields. “Our color is very consistent from batch to batch,” Li says. “But the same solution may turn out a different color for different customers, because their chips and packaging technologies are can be different.” The company has now figured out a way to simplify the customization by making only five different yellow phosphors, and mixing them in different custom proportions for different users, or customers can mix them at their manufacturing sites.
Li says the yellow phosphors used with blue LEDs (for cool white light for cell phones) have been improving by about 10%-15% a year. He says that they are now probably at almost 95 % of potential quantum efficiency, so future improvements will likely be only in the 2%-3% range, primarily to increase the absorption efficiency of blue and scattering loss of white emission light. Intematix R&D is focusing now on improving green and red phosphors which are critical for both high NTSC TV backlighting and high CRI warm white applications.
He notes that licensing costs are no longer much of an issue now that the major LED makers have licensing agreements with each other, and the others buy Intematix’s alternative versions instead. He figures phosphor costs are only about 1-3% of the packaged LED cost, or at most perhaps 5-6% of the final price, including license costs.
Ramping phosphor production for potential SSL volumes isn’t an issue, as Intematix already has an automated production line in China making phosphors for compact fluorescent lamps (CFL) and cold cathode fluorescent lamps. “The SSL market still uses few tons a year, but these other markets use tens of tons per month, and those technologies will be applied to LEDs,” he notes. The plant it established in 2008 in Suzhou, China, has capacity of 300 metric tons/year.
November 3, 2009