Electronic News: What’s changed in the foundry business?
Nave: With the increased cost of capital, more and more companies cannot afford to build their own fabs. We see more and more companies that used to be vertically integrated, which increases the volume aimed at foundries even from companies like Intel, Motorola and TI. It’s not worth it for them to build their own fabs for certain technologies, and they now outsource peak volume. The foundry business, as a whole, is growing at roughly twice the pace as semiconductors. Semiconductors are growing about 7 percent or 8 percent a year, while foundries are growing at about 16 percent a year.
Martin: We would probably start further back in the information chain and look at the markets that have to be served and how information needs to flow from the markets into process development and into the offerings for the customer base. We look for the systems-level knowledge being built into the process, and we are doing that with our partnership with IBM, Samsung and Infineon to make sure the processes meet the requirements. Economics is an extremely important element, and we believe partner economics provide substantial benefit for technology as well as the broad base of customers. Development can be done by combining resources — CapEx, as well as engineering resources — to get efficiencies that can be presented in the process and lower the risk on the process. We’re also providing customers with multiple sources for products.
Byers: There’s a great deal of complexity being added to the role of the foundry. There are business considerations and partnership considerations. There are also easy access considerations. The discussion in this business is time to market, which is being driven to a great extent by consumer applications. We’re seeing consumer products playing a new role. One of my favorite data points is who is the largest consumer of technology today. It’s a 13-year-old girl. We have to determine how to get products to market faster, what services are out there, what reference flows are there, what are the libraries, what are the IP offerings and how broad is that? That has changed from when we started out and were able to offer advanced process technologies. Now we have to determine how we’re going to get customers to market as quickly as we can.
Electronic News: Is it a volume game, or is it now a custom-chip issue? Or, to put it another way, how much life does Moore’s Law have for many companies?
Byers: The message we’re trying to get out there is high value, low barriers to entry, and easy access. But we’re not seeing people getting off the road map. We’re seeing a lot written about people getting off the road map. The foundry industry’s job is to make sure there are high-value processes that can be easily accessed with low risk.
Martin: The number of companies looking for advanced technologies is growing. The opportunities are really substantial at advanced technologies levels. The implementations of 90 nanometer and 65 nanometer technologies are for substantial volumes. The number of companies with new designs is growing. I would reject the hypothesis that people are getting off the road map.
Nave: I’ve heard a different story, and it’s primarily related to NRE [non-recurring engineering costs]. The development costs are skyrocketing largely because of verification. With system on a chip, it’s very complex. The more you condense things, the more complex it gets and the chances for getting silicon right the first time are getting lower. Either it’s several iterations, which cost you in time to market, or it’s tons of verification. This gets exacerbated by the fact that with new technologies, the interaction between the design and the manufacturing makes it more difficult to take into account all the side effects to get it right the first time. Even the mask costs are in the order of $1 million at 90 nanometers. Only so many products can afford to amortize the NRE. Definitely, there are Pentiums and platforms for cell phones that will sell in so much volume that they can stay at advanced nodes. But there are other products that cannot afford it and have to stay at older generations, where the balance between NRE and the overall cost is lower than at advanced nodes.
Electronic News: Is it the volumes that are increasing on advanced process nodes, or is it the number of companies looking to churn out products at advanced nodes.
Byers: It’s both. It’s not an either/or.
Martin: I agree. We see a lot of companies interested, as well as a broad spectrum of parts.
Electronic News: Are we seeing a split in the industry, then? We’re getting two different stories here.
Martin: We see a trend in the industry for a number of IDMs moving to the ‘Fab Lite’ strategy. That builds need and demand for foundry services, as well as the continuing growth of fabless companies. The combination is very strong there to support the greater than overall industry growth. Of course, there are still very substantial companies that do it all on their own, and they continue to be successful. So there is a partitioning of the industry.
Electronic News: We don’t see anyone doing it completely alone anymore. Even Intel and TI use foundries.
Nave: I’ve seen several companies that were manufacturing in-house now looking to shut their fabs or go outside for their next-generation technologies. I would bet other companies would see even more.
Byers: This trend has been in place for five years. Motorola [now Freescale] announced a strategic plan to move one-third, and then 50 percent of their production to foundries. This is a trend that has been in place for five years and I think we’re beginning to see it show up in the numbers, not only in terms of the percentage of output the foundries account for, but the revenue growth of the individual foundries.
Electronic News: Because of the growth of the foundries, there has been a lot of concern about who actually owns the intellectual property in chips and how that is shared. Is it changing?
Martin: I believe the intellectual property ownership rules have not changed. Each of the companies that develops IP will take steps it feels is necessary to protect that property. The basis continues to be the same. The investment in intellectual property may be shifting from what used to be solely IDMs to much more activity in the foundry space.
Electronic News: But with the separation between manufacturing and design, the flow of information isn’t as free as it used to be.
Nave: There are still cases where the foundry and the fabless companies collaborate. Sometimes they decide it’s better to fix in the design and sometimes it has to be fixed in the process. But clearly, if someone brought the IP with them it belongs to them. It becomes interesting when the fabless company is asking the foundry to develop some tweaking of the process or some optimization. Then the question is whose IP it is, the one that came up with the idea or the one that implemented it, and how much exclusivity do they deserve. They may say, ‘I’ll develop it for you and you will have a year of exclusivity, but after that I will sell it to anyone else.’ It’s very intricate.
Byers: Just because a company has gone fabless or ‘Fab Lite’ doesn’t mean that communication is suddenly difficult. Communication with the foundry is a 24 x 7 proposition now. There are also some economic incentives, both on the part of the customer and the IP vendor and the foundry to make it work. Bonus incentives are a good way to get people to talk to each other and collaborate and make it work. IP ownership is now thought of ahead of time, it’s negotiated ahead of time and it really is dependent on the parties and the business conditions and the market dynamics as to who is going to own what and when. Everyone’s different.
Electronic News: Is that recent? As of a year ago, it seemed those types of discussions were still embryonic.
Nave: It’s evolving, but there are more and more of those cases so it’s becoming a common practice.
Byers: It’s negotiated.
Martin: There’s nothing that prevents multiple companies from coming together to agree on a framework or process. That can be done quite reasonably. In our partnership activities, we’ve set up the rules and they work quite well. We know what each person brings to the activity and how it should be handled and who owns what.
Byers: One of the values of the pure-play foundry model is the protection of the customer’s IP. We take that extremely seriously, and have elaborate security measures both inside the fab and outside the fab, on the business and the technical side of the company. The role of the foundry is to manufacture other people’s designs. There is no intent inside the pure-play model to change that.
Martin: I concur. That is one of the extremely important elements of a foundry’s success — protection of a customer’s IP.
Nave: You don’t try to compete with your customers, and you don’t try to play the role of an IP provider. You also have to establish trust with your customers. When they put their jewels in your custody, it has to be taken care of.
Electronic News: Let’s look at this a different way. How much weight is given to the IP of the foundries versus the designers, because chips do have to be manufacturable. Who sets the rules?
Byers: Lacrosse has rules. Manufacturing has dialog. You’re dealing with an interesting concept of mass customization, which may sound like an oxymoron. What applies to Case A this hour may not apply to Case B. This goes again to easy access. What will help you move your product out in this window. There is a dialog that has not been seen since the industry’s founding. This is the new set of skills that are going to be required to be successful. What’s changed is that the foundry is now a catalyst in pulling the whole supply chain together.
Electronic News: But is the differentiation coming from the foundry or the chip designer?
Martin: There is continuing differentiation and innovation from the foundry and the chip designer, as well as the EDA folks. We’re certainly in a very dynamic situation. We look for improvements and benefits from each area that have to be melded together.
Electronic News: If you are in what TSMC calls mass customization, does the price of creating chips come down?
Byers: We need to differentiate between price and cost. Price is part of cost. It’s not the whole thing. Cost can include other factors like hitting a market window. Moore’s Law has been built on cost. You can shrink designs so you can get more and more on the same size wafer. In the next 12 months, there will be a lot of announcements from a number of companies, which are filled with very bright people. They’re not going to do anything to throw their costs out of whack.
Martin: Cost per function also continues to go down. That drives the ability to provide more functions at a very economical level.
Electronic News: Still, the raw dollars it takes to develop a new chip have gone up significantly. Have you done something to knock that down?
Martin: You probably need to talk to the designers about that. From the process standpoint, we’re doing everything we can from the partner economics on developing technology to the implementation of technology to provide competitive economics for manufacturing the advanced technology nodes. Even with the daunting macro figures for implementing 300mm fabs, the individual wafers can be manufactured with mature yields much faster than what they could have been done with technology from two generations or three generations ago. While a new design may have several millions of dollars invested, it has a much higher probability of first-time success. The output of the factory is much more controlled and much more precise at an earlier point in time to guarantee the success of new designs.
Nave: The incentive of going to a new generation of technology could be performance, if you need the raw speed. It could be form factor, if you need to squeeze a lot more into one chip. Or it could be power or cost. On the first two, there’s no argument. If you need the performance at 5 gigahertz, then you go to that node. On power, we see more leakage, so power is actually going up and creating more challenges for the designer. On cost, it only works if you have huge volumes to amortize your development cost. The hit rate is still much higher on earlier generations than on the more modern one. If you can afford the older generation and you don’t have to go to newer ones, you’re better off with time to market. And unless you sell huge volumes, you’re better off because of the development cost.
Byers: There’s another alternative, which is half nodes. At 0.18, there is an opportunity to get more per wafer from going to 0.15. That is all aimed at bringing down the cost.
Electronic News: But do those half nodes still work when we get down into several atoms of thickness in insulation?
Byers: We’re looking at taking 90 nanometers down to 80 nanometers. Many of those designs will be in the consumer area. The half node has proven to be a very good answer for people who want the most out of their investment.
Electronic News: Still, something very fundamental has shifted here. In the past, it was designs that changed. Now it’s a matter of what can be manufactured. Who’s gaining in influence?
Nave: I don’t think there’s a shift there. What is changing is the applications and the markets, which have been shifting toward consumer. That is driving the opportunities. Once you’ve identified the opportunities, then you have to ask yourself, as a customer, what does it take to implement it? The foundries and IDMs are driving the technology road map, but the success stories come from those companies that have identified an application and opportunity and transformed it into a product they can sell for millions of copies and make a profit. It could be iPods or digital cameras or RFID. The foundries are the enablers. If you go back 10 years, it would have taken a huge investment to implement it. The foundries are enabling that.
Martin: In the 1980s, the IDMs were the primary source of design and manufacturing. In the 90s, we went through a disaggregation in which various parts of the design and manufacturing could be done independently. We presently see some reaggregation, at least from the communication and partnering elements. There is a much closer link from the entire chain from design to manufacturing to delivery of product. We’re reaggregating the information flow.
Byers: This is the win-win-win. It is going to be collaborative and parties are going pull together. Those that do it successfully will enjoy the fruits of their efforts.
Electronic News: One of the industry segments that has not used the foundry business very heavily is analog. Is that changing?
Nave: There is analog and there is mixed signal. If it is pure analog, they will continue developing their own processes. But more and more, you see analog elements integrated onto the chip. Those are becoming part of the offering through IP vendors or the foundry or customers putting their own IP on the chip. More and more analog elements are migrating onto the chip.
Byers: We’re seeing the popularity of the mixed-signal RF business. That’s where we’re seeing analog.
