Experts At The Table: Multi-Foundry Strategies

By Ed Sperling
Semiconductor Manufacturing and Design sat down with Walter Ng, vice president of the IP ecosystem at GlobalFoundries; John Murphy, director of strategic alliances marketing at Cadence; Michael Buehler-Garcia, director of Calibre design solutions marketing at Mentor Graphics; Bob Smith, vice president of marketing and business development at Magma, and Linh Hong, vice president of marketing at Kilopass. What follows are excerpts of that conversation.

SMD: Does a multi-foundry strategy work? And are there problems moving a design from one foundry to the next?
Murphy: There are definitely problems moving from one foundry to another. You need to look at where the handoff makes sense. Is it the ASIC model, or is it an implementation path into each one? If it’s the implementation path, then there is the ability to multisource to different manufacturing lines. When you try to do that at the GDS II level it becomes much more difficult because there’s differentiation in the foundries at the process level, which gets reflected in the GDS II.
Ng: It is difficult to multisource if you’re using those foundries that don’t want you to multisource. If you look at the reasons companies want to multisource, it typically involves assurance of supply. Those are usually the large customers. What’s required is a common process, a common DRC deck and common tech files with regard to design enablement, and then you need common IP. You build your design and come out with one GDS II. It’s not just a common process, though. It’s a fab-sync program. You need to make sure you’re within a certain level of tolerance. We’ve done this in the Common Platform. When you look at customers dictating their own process spec, it kind of works in the same way. But it certainly requires a lot more work on the part of the fabless companies to try to align manufacturers that are not naturally aligned. That requires a lot more work and coordination.
Buehler-Garcia: I would argue those customers aren’t really fabless. They’re IDMs without a fab, or fab-lite. When I was at Chartered we had a couple customers with more device engineers than we had at Chartered. They were looking for capital coverage by the foundry.
Ng: There are fewer and fewer of those customers. That’s a custom development. The ROI that a foundry has to get on the special custom development has to be quite significant.

SMD: Is there any change on the IP?
Hong: That’s one of the unique changes. You have to be a large IDM to have all the IP port from one foundry to another. But ecosystems are being built—GlobalFoundries has its Global Solutions and TSMC has its ecosystem. IP has been enabled where it’s pin-compatible. The internals of the GDS has to be the same to enable going from the Common Platform to other baseline manufacturing.
Ng: That’s a different level of multi-sourcing effort. That effort introduces risk, time-to-market challenges, and additional cost. It also was more common at the older nodes—180nm and 130nm—where you could do that without too much risk. At the newer nodes, the level of difficulty of doing this ‘rip and replace’ is not easy and it’s riskier. The process technologies are much more different at the leading edge. To rip and replace at 28nm is not realistic.
Hong: Up to 40nm, they have looked similar. At 28nm, with high k/metal gate, there is a divergence into two camps. But I believe there will be a convergence again.
Smith: We certainly see a big effort to not get too reliant on one supplier, especially where you’re looking at volumes in the millions. You may have the market and the know-how, but if you’re at the mercy of one supplier you can be in trouble. And it may not be just because of business. Look at what happened recently in Japan. There are no major foundries in northern Japan, but the supply chain for automobiles, for example, has really been impacted. But I’m also skeptical that one size fits all and that you can move stuff around because business dictates that the foundries differentiate. Maybe that’s the platform, but most companies will gravitate toward a secret sauce, whether it’s design rules, IP or something else.
Ng: I have been in a forum where someone accused GlobalFoundries of differentiating just because it can differentiate. That’s not even close to the case. If you look at the 28nm technology today and the development alliance we’re in with seven other companies, we came out with a statement about high k/metal gate even before others in the industry such as TSMC. They kept saying high k/metal gate was not necessary and exotic, and now they’re trying to rewrite history. We still believe our 32/28nm decision on gate first was the right decision for that technology node. There’s a 10% density advantage and there are some power advantages. When customers move from 40LP to 28LP, their design style is not impacted nearly to the degree that a gate-last process requires. The decision we’ve made on 20nm is a node-by-node decision. We’ve saved the customer the impact of one technology node.
Buehler-Garcia: Multi-sourcing depends on who you are and where you are. If you look at the ecosystem, our solutions have to consider all the different options. That’s all the way from a fab-lite company that can drive its own processes to a startup. You need a flow and design process that takes them there step by step. Plus, you have to look at what multi-sourcing means. Everything isn’t a single chip. The big guys are going to tape out 40 or 50 designs. They’re going to give 10 to one foundry and 10 to another foundry. That’s not all bad because too much of any one company is not a good thing for a foundry. The company can move somewhere else. What’s changed is that differentiation is no longer in tweaking the process. People now have to differentiate with the design, and that’s good for the ecosystem. There’s no canned answer of how you do this. If you put 10 customers in a room, their perception of what is multi-sourcing would be different. But they do want a system, a flow, the tools and the IP to let them do this.