Foundry model could be key to InP industry future

Michael Lebby says that the InP industry is at a crossroads. The co-founder of Ignis Optics is currently putting together a roadmap highlighting several key recommendations to improve conditions in a sector that is shackled by the heavy overheads of running internal InP fabs with excess capacity.

His central message is this: optoelectronics companies need to drive a common foundry business model, just like the silicon industry did 20 years ago. But how similar are the silicon and InP industries, and just how exactly would such a foundry model be set up?

Michael Hatcher: What's the idea behind the InP industry roadmap?

Michael Lebby: I'm adding a lot of urgency and vibrancy to an organization [OIDA] that has huge potential. The OIDA represents the optoelectronics industry, and I'm beginning to tackle what I think are some of the industry's biggest issues. One of them is the InP foundry situation, and I'm trying to address a real interest in this hotly-debated issue.

MH: How strong are the parallels between silicon and InP?

ML: They're really strong. The silicon guys have been through their woes with foundries and the message is that we're just like they were 20 years ago. One of the reasons why silicon went to a foundry model was that they were going through a down cycle and there was overcapacity - a lot of fabs were empty and they had huge overheads. So they were driven to look at creative and innovative solutions to the problem.

I think we're in a similar situation with InP. We now have this overcapacity and it's going to drive us to think creatively and innovatively. The situation is a little bit worse [for InP], in that the different silicon CMOS processes were fairly similar, whereas we still have a uniqueness problem in our device and material design. I think it will be more difficult [to agree on common processes], but there are some parallels that we can learn from.

MH: What is your key message?

ML: The roadmap sounds a wake-up call for the industry. We need to act soon to address these issues and accelerate the development of an InP foundry model. The OIDA will be working with its members and the community as a whole to make strides in developing a common foundry for the marketplace and identifying viable solutions to simplify the situation.

MH: What's holding the InP industry back at the moment?

ML: There's definitely a reluctance from the [InP chip] suppliers to consolidate. Also, they have to be vertically integrated to have a reasonable business model. They really want to be horizontally integrated (i.e. supply a wider range of products) but they can't afford to, and there doesn't seem to be any route to do that. The big question is how to get the traditional vertical integration models to work on a horizontal basis. That's what is driving the roadmap.

The next thing to address is how to simplify the complexities of the materials, devices, packaging, intellectual property (IP) and manufacturing processes. This is a huge issue. Some lasers have got two re-growth steps, others have got three. The IP is just incredibly detailed and there are probably 500 different patents on DFB lasers alone.

We also have an inflexible customer base. The big communications system companies have got too many vendors to choose from. It doesn't matter if there's a standard or a multi-source agreement (MSA), they just choose the ones that they think have got the best price and the best performance. And if you don't meet it, then tough. Those inflexible customers are costing the livelihood of a lot of the vendors. The vendors have got no margins so they're driven down on price and way too hard on specifications. When you get driven on specifications you need unique devices, and you can't really work together.

MH: Will the industry support such a radical switch?

ML: From the engineering level they all understand it. At the CEO level they understand it too, but the thing that's holding it back is that if you go out of the vertical integration mode then you jeopardize your business. And you can't do that when you're not profitable. And so the horrible dilemma [for chip/module vendors] is that until some of these companies become profitable and break even from a business standpoint, they're going to have to move slowly towards the horizontal business model.

Only the brave companies are going to move in that direction right now, and I don't see many of them doing that. The question [for chip vendors] is how long can you work with high overheads and inefficiency until you get to the point where you're forced to say "I've got to find a different solution"?

MH: How could an InP foundry model be established and how long would it take?

ML: For a foundry to be successful, the [chip] suppliers have got to run the show, because they're the guys that need to simplify things. What I see is baby steps over the next two years. I don't see a real, proper commitment until around 2009. Then the technology has to be transferred to the foundry. That will take 18-24 months. It took Bookham probably about 18 months to transfer its fab from Canada to the UK - and they did a good job. So you're not going to see a real silicon foundry-like operation for InP until 2010 or 2011.

But if a good plan was put together, the capital markets in New York would fund it. And there will also be some help from the financial markets in terms of consolidation.

MH: What happens next?

ML: Until 2007, not going much at all. By 2009, there will be fewer, stronger, InP vendors - call it an oligopoly if you like. Those that figure out how to be profitable will have fab overheads to match the business they're running.

In 2009 the customer base will also be more flexible because there will be fewer suppliers to choose from. If there are fewer suppliers then customers have to go with the market price, and have to accept MSA specifications. That will reduce the complexity of devices and the difficulties of establishing a foundry.

MH: Who will be left standing?

ML: I guess the guys that have got lots of money. That means Agilent and JDS Uniphase [Ed note: the future of Agilent's semiconductor business is uncertain and JDSU has recently bought the test and measurement company Acterna, which could signal a change in priorities]. The guys that don't have any money have got some really smart technology, and good differentiation, but they're stuck in this vertical business model just to keep the business going. Opnext and Mitsubishi are well positioned, because they have other technology running through their fabs, which reduces the cost of telecom component manufacture.

MH: How can the OIDA help?

ML: What you're seeing now is a message to the community saying that we're committed to go help you figure this out. Now, we'll play whatever role you want us to play. We have a lot of influence with government agencies who have a lot of money to spend, and I'm pretty well connected with the venture-capital community. So if there's a good solution, OIDA can help finance it and help companies go and do it. Our roadmap is going to be composed by industry experts who have a really good understanding of the technology and business issues.

• Michael Lebby will be presenting an invited talk on this subject at the CS-MAX conference later this year. See compoundsemiconductor. net/csmax for details.