While the optical communications market has begun to rebound, most component suppliers continue to find profitability elusive. The overhead that in-house indium phosphide (InP) fabrication facilities represent-facilities that sources say almost all vendors operate at much less than maximum capacity-have hampered vendors’ pursuit of black ink. The Optoelectronics Industry Development Association (OIDA) believes it has the answer: develop a foundry model similar to that enjoyed by the semiconductor industry, and provide a path for optical component vendors to go fabless.
In conjunction with the Defense Advanced Research Projects Agency’s (DARPA’s) Photonic Technology Access Project, OIDA held a forum on the subject Nov. 2-3, 2005, in Arlington, VA. The conference drew more than 70 participants, who listened to approximately 25 speakers make observations, analyze trends, and, in some cases, tout their companies’ foundry services. Workshop organizers say the event succeeded in getting the issue on the table for discussion, as well as highlighted some of the obstacles that must be overcome before a foundry model can be made practical. However, certain basic questions-including, in the minds of some participants, whether a foundry model is even a good idea-remain unanswered.
Certainly there appeared to be general agreement that the industry possesses significantly more InP fab capacity than the current modest level of demand. A presentation from Agilent estimated that new wafer starts this year will number 40 per week; another from CyOptics estimated the optical communications market would require between 10,000 and 15,000 InP wafers per year. The same presentation put the number of InP wafer fabs at 20 to 30.
Michael Lebby, executive director of OIDA, believes that companies with too much InP fab capacity on their hands have limited options, particularly if they are looking to increase fab usage. They could wait for market Darwinism to take out competitors, but meanwhile they’ll still be burning cash. They could accelerate such Darwinism by acquiring competitors and thus expand their product lines or market share; yet significant capital as well as a favorable deal structure would be required.
They also could run other InP devices for nontelecom applications on the same line. Japanese companies in particular have succeeded with this model. Presentations at the workshop pointed out that the structure of major Japanese companies-large firms active in a wide variety of enterprises-make it more likely that they could find multiple uses for the same fab. Most U.S. companies in the telecom space don’t have that wide a breadth, despite efforts over the last few years to diversify. Lebby says that telecom companies looking beyond their core markets face significant barriers to entry that make such expansion difficult.
With this backdrop in mind, Lebby says a foundry model that would enable companies to divest their withered fabs makes sense. “The trick now is to try and gain confidence in the main suppliers to go work with foundries to reduce their overhead,” he says.
It is this trick Lebby hoped to pull off-or at least lay the groundwork for-at the workshop. The requirements for a successful foundry generated significant discussion. According to Lebby, it became clear that a successful foundry needed to focus exclusively on its niche, rather than potentially compete with its customers by producing its own optical components. The foundry will probably require expertise in more than InP, suggests Bill Ring, managing director at foundry service provider WSR Optical Device Solutions LLC (Ringoes, NJ), who moderated part of the workshop and who is summarizing the results of the event in a report. In fact, according to Lebby, OIDA has expanded its foundry discussion to include gallium arsenide as a result of this point. The foundry also will have to establish firewalls that separate the intellectual property of its customers, who potentially will be competitors.
The foundry model probably won’t look like the one for silicon, the workshop made clear. “If you look at silicon, you’re not really growing multiple quantum-well structures and different SCH regions and different doping levels and etch-stop layers and that kind of thing. So when you make the comparison between InP and silicon, there’s this whole area of epitaxial growth that has a different impact on the final design,” Ring explains. Rather than a model that comprises growth, fab, and some kind of reliability packaging test, the model might contain two foundry levels, epitaxial suppliers and fab houses, he says. “We couldn’t really come to any conclusions, other than that’s a major difference. And therefore that impacts how you set up a business model for this.”
According to Tom Hausken, the director of the components practice at Strategies Unlimited (Mountain View, CA) who has written on this subject in the past (see Lightwave, July 2005, page 27), OIDA has the right idea here. “The idea of comparing lasers with the silicon industry is totally flawed,” he asserts. “Oh, sure, they’re all done on wafers and a lot of the processes are similar, but the similarity stops there.”
While epitaxial concerns represent one element InP has that silicon does not, silicon has a few attributes that InP would require to make a foundry model realistic. Standard processes are one major lack in the InP world; it seems there are as many different ways to make a DFB laser, for example, as there are laser developers. “What the industry is really looking for is a common platform or family of devices to support the majority of system applications,” says Lebby.
The optoelectronics industry may be closer to common processes than people think, Lebby believes. “Clearly, some companies have their own secret sauce in their processing. But a lot of the process steps are turnkey these days, whether it’s putting down gold, Schottky barriers, ion implant, CVD deposition of dielectric layers, or even producing holograms to produce gratings and DFBs-a lot of these things are actually turnkey.”
Ring envisions a foundry association, perhaps similar to SEMATECH, that would help establish de facto standards for InP and related processes. Lebby suggests that such standards may begin as multisource agreements among vendors.
Once standard processes arrive, the industry will need software tools to model them. While companies such as RSoft Design Group (Ossining, NY) and others offer photonic modeling software, Ring and Lebby agree that more extensive tools will be required. Lebby anticipates that OIDA’s next workshop on photonic foundries will focus on this need.
However, the biggest obstacle to creating a photonic foundry model is that most companies don’t want to give up their in-house fabs. “One of the big issues in our industry because it’s vertically integrated is that all of the suppliers believe that they get their differential advantage through their device performance and fabrication technique,” Lebby admits. “That’s one of the reasons why a lot of these guys are unwilling to take that technology to a foundry, because it becomes a common platform.”
Certainly several presentations from companies such as Bookham, Emcore, Mitsubishi, Apogee Photonics, and Infinera touted the advantages of vertical integration. Besides holding the perceived key to product differentiation, in-house fabs enable their owners to react quickly to changes in market demand and reduce time-to-market. They also protect intellectual property.
Hausken sees their point. “To me, there were more pros to vertical integration than cons,” he says. “If you get to that point where you have the right number of suppliers, then there’s no reason to go to this fabless model.”
Indeed, both Lebby and Ring agree that even were it to reach fruition, the foundry model wouldn’t work for everyone. “Some of the devices will definitely go the common platform route. Other ones, the super-high-performance, super-high-spec [devices], may not be ready for this-and they may remain in a vertically integrated format,” Lebby concedes. “But there is common acceptance in the industry that standard device platforms need to be standard.”
The workshop may have addressed at least one foundry-model concern component vendors may have brought with them to the event. “There’s a big perception in the industry… that if you change your source, you change anything on your device-whether it’s your materials, your spec, where the device comes from-then there is the perception that you will lose your socket at your customer,” Lebby says. “What we discovered in the workshop was that this is not necessarily the case. If you can work with your customer base and say, ‘Hey, we’re going to move this technology to a foundry,’ in all cases the customer will say, ‘Hey, let us know ahead of time and we will work with you on this and we don’t have to go to one of your competitors.’ The perception is false.”
Ring and Lebby also pointed to a pair of startups, Syntune AB (Stockholm) and Eblana Photonics (Dublin, Ireland), who aim to overturn the notion that one can’t enjoy product differentiation in a foundry model. Both startups have developed InP-based optical devices whose fabrication they will outsource. Syntune will make tunable lasers and other technology, with CyOptics offering foundry services. Eblana will develop DFB-like light emitters using a “photon bandgap” technology that Vitesse will produce on its electronics lines. James O’Gorman, president of Eblana, presented his company’s vision (see Lightwave, May 2005, page 19) at the workshop, and Ring was impressed. “They’re a very good example of potentially how to make the model work,” he comments.
Both Ring and Lebby-and, by extension, OIDA-remain committed to establishing a foundry model. “I go back to the classic statement, the emperor has no clothes,” Lebby says. “The industry is well aware of its shortcomings. The industry is aware that it has a lot of spare capacity; the industry is aware that there has never been really been a successful cross-licensing program among the big players in this technology.”
“The OIDA needs to really work to pull together this foundry idea, to push it forward in the U.S., because the companies that are based here are really struggling-or, if not struggling, then their operating incomes for the last few years show that they’re having a great deal of difficulty supporting themselves,” Ring adds.
Lebby foresees OIDA’s role as a catalyst in this effort-an effort that will need patience and time. “It’s going to take 5 to 10 years to really get this thing up and running properly,” he concludes. “It took silicon 20 years, so why should we assume it’s going to take anything shorter?”