Partnering to make room for maximum value-add

Texas Instruments’ CTO and senior VP of silicon technology development talks about outsourcing, the power that equipment suppliers hold and the impact of foundries.

By Ann Steffora Mutschler, Senior Editor -- Electronic News, 7/27/2007

Dr. Hans Stork, CTO and senior VP of silicon technology development at Texas Instruments, sat down with Electronic News/Electronic Business to talk about the company’s decision to outsource its process development, how decisions are made about which foundry to use, the power of semiconductor manufacturing equipment suppliers and the positive impact that foundries have had on the industry as a whole. What follows are excerpts of that conversation.

Q: Since TI made the move earlier this year to outsource its process development, what pieces has the company handed over so far?
Stork: 2007 is a complex year because our 45-nm development activities were, in some programs, close to finished; in some other parts of the program, still early. Our low power technology usually comes early and that program was far along, so we’re handling that as if we had completely qualified it, like we have done in 65- and 90-nm technology, meaning that as we approach the time when we start running silicon for qualification, we would get connected with the foundries, set up some engineering agreement that they would deliver, a technology with the same performance for us, with a schedule that was close to, but usually slightly behind, our internal development. Now we are going through that same process with the difference that we won’t actually continue internally but fully depend on the technology maturing in the foundry. On the other side of the spectrum, in the most advanced, high-performance technology it was still early, we were still making decisions – in fact, this year in the first quarter – regarding what choices of high-k, metal gate, materials, integration sequence, and we are using those decisions now to continue the process at one of the foundries, at least for the time being, and then fan that out as we see fit. We are ramping down our activity internally, and through the second half of this year, it will eventually go to zero. We are doing this based on milestones that existed and as we hit those milestones, we’ll change the place where we run the wafers from Dallas to Taiwan.

Q: What are TI’s foundry relationships like?
Stork: We’re working with UMC and TSMC and have slightly different arrangements with each one.

Q: Are they both working on the same processes or different ones?
Stork: On our low-power technology, where our need for having multiple foundries is essential, they are both working on it. For the other technologies, it is by choice (from us) right now.

Q: How do you make that choice?
Stork: It’s a mixture of technology conditions, cultural mix and business requirements. It would be too simple to just make it quantitative and one of the key end points that we want to arrive at, is having foundries that are within the same ballpark of capability. It’s never a perfect level playing field but we don’t want to have a system so skewed that we really have no choice.

Q: Are there certain cases from a technology standpoint that one of those foundries meets the needs a little bit better in terms of your requirements?
Stork: If the value we are seeking is say, a broadening of our IP capability, then obviously the bigger foundry has a clear advantage because they have more IP that they have access to or have developed internally.

Q: When did it become clear that a change needed to be made in terms of how TI developed its processes?
Stork: In the second half of 2006. This was a combination of changing business conditions, not just for TI, but across different segments, as well as our outlook on technology capability and what is driving that.

Q: What are some of those business conditions and the challenges going forward?
Stork: Very generically, because these advanced CMOS technologies support very complex systems-on-a-chip, the number of decisions at the system level including the partitioning, architecture, software support, and so forth, are increasingly dominating the schedule to the point that when you then look at when you need what level of technology to support that, or to the degree that you can take advantage of the technology, that was starting to become limiting. I know I can make that comment across the board for all the applications that TI supports. At the same time, you look at the supplier side of the technology and recognize how much of an impact –say- lithography has on what people can do. The whole industry pretty much depends on 1+ lithography supplier, and their latest and greatest machinery is available within six months for everybody, so from that perspective – which is a very big factor – you equalize to one another within six months. That doesn’t mean there is no differentiation possible, but it is at different levels, so we are trying to put our investments in those things.

Q: It sounds like you are saying that the semiconductor manufacturing equipment makers have such a high stake that they are calling a lot of the shots, is that correct?
Stork: They are always polite to us and they make us believe we do it. The reality is that they do have a significant impact and of course they seek out working with the foundries and IDMs as to what the requirements are because they don’t have the insight on what eventually needs to be built, but yes, they seek a common denominator to justify their investments from the development point of view. And with a limited supply base, because the granularity of the investments is so high, that in many cases, you may only have a few suppliers that are competitive and then indeed there is little room to stand out from the crowd. This has not been a sudden change; this is something that has continued to grow over the years.

Q: When did you start to notice in the industry in general that things were changing in terms of process development?
Stork: The transition to 300-mm was a big factor in all of this. It really drove a lot of consolidation, a lot more focus in the investments, both from an equipment supplier and from an integrator point of view, and left the playing field available to fewer and fewer players, drove the formation of consortia much stronger. In history, I think we are going to look back and say, “That’s what did it.”  And it took a while to play out, but that was a big factor in doing this [process development outsourcing].

Q: How far back do these changes stretch?
Stork: If you look far enough back, our industry was totally vertically integrated, when semiconductor companies in fact many of them provided end products and built their own wafers. Even the separation to a pure semiconductor player didn’t happen until the timeframe of the 1970s. And then in the 80s, the equipment supplier industry started to separate out. In the 90s, the EDA supplier industry started to separate out as its own base. You see an increasingly amount of specialization and the foundries played a key role in it as well. The foundries have, on one hand, forced the IDMs to focus more on what unique value they provide in their semiconductor products. On the other hand, they’ve also allowed smaller companies to still have access to very advanced technology. From an ecosystem point of view, [the foundries] have done a healthy thing for the industry and it’s just a shift in boundaries that lets companies focus on their best value add. At the same time, things have gotten so much more complex, it’s not like we are looking at the same problem now that we did 20 years ago; its orders of magnitude more complex. I don’t know how else we would have solved the problem but create more specialization to address these things. We do have an EDA industry because that is so intensely complicated just to support that part by itself that it would be a tremendous diversion to have all these different things running around in different companies. There’s a natural place where as a layer of technology becomes sufficiently complex, it justifies to have its own industry and focus on that activity with a couple of key players that then can deliver their products to a large base of customers, rather than have every company do it internally slightly differently than the other. You see this also happening at the process integration level where the ability to integrate certain things is going to be more common between players and so the challenge becomes how to differentiate your house. Everybody knows how to use the wood and the concrete, but there’s still a lot of furniture you have to put inside and that’s how you differentiate one house from another. We’re not in the business of building houses anymore but we are still trying to get the maximum value for the whole house. We share some of the building or get it done by a subcontractor and we’ll focus on the furniture.

Q: Can you give an example of where this dynamic is clearly apparent?
Stork: One place we still see a lot of that is in analog. Again, some intricacies of most of the analog products is that they are more simple in nature, they are more circuit-design or component-sensitive so having unique components available to your designer can be a real benefit so we keep doing those things internally.

Q: Looking ahead to next generation 32-nm and beyond process development, what is the interplay like between TI and its foundry partners?
Stork: We started our internal activities – early definition and so forth - on 32-nm. In the timing of our decision to take some action in our process development model, one of the key factors was that it is time to start making some serious capital investments for 32-nm. We do not plan to make those capital investments internally but are already working with both foundries to define and start on early activity for 32-nm with them. We will count on their capabilities to pull that off in the right timeframe with the right kind of specifications that we think we need for our customers. We largely align with their internal objectives, which is no surprise because we’ve already worked with both foundries for multiple generations and they know where we are coming from and I’m sure have been anticipating our next steps as well. It’s not perfect but those discussions are happening and we are trying to make the right kind of choices. For 22-nm we are not that far along in mapping that out yet.