Experts At The Table: Changes In The Ecosystem

Last of three parts: Issues for stacked die; chip development costs for ecosystems vs. IDMs; innovation and startups; process variation; older process nodes; future challenges.

popularity

By Ed Sperling
Semiconductor Manufacturing & Design sat down with Michael Buehler-Garcia, director of design solutions marketing at Mentor Graphics; Seow Yin Lim, group director for marketing at Cadence; Kevin Kranen, director of strategic alliances at Synopsys, and Tom Quan, director at TSMC. What follows are excerpts of that conversation.

SMD: Does the increasing collaboration in the ecosystem solve some of the thorny problems of liability for stacked die?
Kranen: You can solve more of the test problems more easily, once you put the chips together, and figure out diagnostically where the problem is. That’s the kind of work that’s going on today.
Buehler-Garcia: All the IEEE board stuff really matters now. But there’s also a business issue here. If you want to own it, then you take the risk. The question is why are the big IDMs jumping in? Is it for cost or for control? That’s a business decision. Being part of the ecosystem, we can’t guess what the reasons are. But we can get 80% of the stuff into the channel, with the other 20% being what differentiates companies. If you want it customized, then you have to fund it. We’re seeing that even with EDA tools with some of the decks. If they want to modify it, they have to pay for it—and they have to own the responsibility for a variant from what’s owned by the foundries.

SMD: Does a tightly integrated ecosystem make it more or less expensive to develop a chip?
Quan: It’s all about iteration. In a tight ecosystem, you get fewer iterations. That’s cost reduction right there.
Buehler-Garcia: If you want to back to an IDM model, that’s really expensive. You have complete control, but you also have really high costs. How much do you want to own? There’s a whole bunch of stuff you have to let go.
Lim: At this point, it’s really a matter of controlling what’s most important and how you improve the efficiency of the rest and optimize it. Ultimately what all customers want to get to is faster time to market, and what all of us ultimately want efficiency.
Buehler-Garcia: We all want ROI.
Kranen: Yes, but that comes from efficiency and expanding the total available market.

SMD: Is innovation easier or harder in an ecosystem?
Kranen: It’s better.
Quan: In the early phase, when we try to define new nodes, as foundries we think there’s a certain way to do it. But as we start working with our partners, their R&D team comes up with faster, easier and more efficient ways to do it.
Those cannot be done alone. It definitely helps with innovation.
Buehler-Garcia: We’re in as long as we trust we’re the only ones in. Getting a startup into the early discussion and having the resources is difficult. We’re all on the 15th node with TSMC. We’re all on speed dial. The number of unofficial meetings we have is probably 50x the number of official meetings. But innovation for companies here is good for us. It’s very hard for a startup to get involved.
Quan: There are fewer people in the ecosystem to work with. For us, the key is to have enough committed resources to solve a problem.

SMD: What happens to the startups?
Quan: From our standpoint, we go from the wafer up. What’s important to us are the things closest to silicon, such as DRC or DFM. There are only a few players there, and they happen to be the ones that have the tool flow. To invest in things closest to silicon, that’s a major undertaking with a huge investment. We see less of that, and most of the new innovations in that space come from the big partners.
Lim: The wafers are getting more expensive, so from an IP perspective, if you want to be a startup—especially in hard IP—the resources and effort needed for finFET tapeout is enormous.
Kranen: It has to be done speculatively, too. If you want to get to market when the customer will be there, you have to start months before and you can’t be sure the customer will even buy your IP.
Buehler-Garcia: But if you walk the DAC show floor, there are still a lot of companies there. This is where we have to come back to 2.5D, which will grow more innovation. The process is frozen, but the complexity of the designs at 180nm is a lot more complicated than when we first did them. There are multiple power rails, different IP applications, different plug-in tools, with thermal and stress concerns. There are pieces there, probably for best-in-class tools, where there will be opportunities. But you’d better be in it for the long haul.

SMD: How is process variation as we move to the next nodes? Is that still a problem?
Quan: At 20nm, the main variation is due to double patterning from when you split things into masks and then try to align them. We went through that and it’s manageable.
Kranen: We went through a whole cycle of planning for there to be substantial mismatch, and then we discovered at the end it isn’t so bad.

SMD: How about the next node?
Buehler-Garcia: It depends on how restrictive you make the rules. The wafers cost more, but you don’t have to worry about how big you make your standard cell. We’re swimming in area at 20nm and 10nm. We’ve got lots of room. We can fit extra grid points to get yield. We have seen some of the leading-edge companies with two sets of libraries, too. They have libraries at the early end, and then they reload as the process matures. A lot of data we’ve seen with more structured IP is that variation is going down, not up. It’s getting better. It’s not that the process has improved, but the rules we’re using are more structured.
Lim: It’s also learning from experience. We’re innovating around the tools.
Buehler-Garcia: The 80% we know about is fine. It’s the 20% we’ve never seen before that causes the problem. It blows up the DRC deck and the timing.
Kranen: And the router gets confused.
Buehler-Garcia: Yes, and then we ask what they did. That’s where the problem is. We can used the ecosystem for the baseline, but customers have to be part of this. You can’t get an ASIC for a foundry price.

SMD: What’s the resistance from the customers to getting involved in the ecosystem?
Buehler-Garcia: It’s a lot of work and money.
Kranen: It also means there are multiple parties that have to agree on things. So you enter in the process of making things work. That’s challenging, because everything they know isn’t always right.
Buehler-Garcia: And some of these guys aren’t planning on one spin. They’re planning on two. You have to get comfortable with working with multiple companies. It takes a mindset to say that’s okay.
Kranen: TSMC has flipped the cycle of process on its head. It’s much more externally focused. The foundries used to be the temple. Now it involves IP and the process, and if you don’t get it all right then it doesn’t work.

SMD: What happens when we start encountering quantum effects over the next couple of nodes?
Quan: When we went to double patterning, in the early days, we said no designer would do a lot of stuff that was necessary. Eventually we figured out a way to make it transparent. If you have new material or new issues at the bottom, you have to figure out a way so that the design flow can handle it. Otherwise no one will adopt it.
Buehler-Garcia: We’ve always figured this out before.
Lim: Yes, we have to. That’s where innovation comes in. If there is a problem, you solve it.
Kranen: And if your current device runs out of gas because of quantum effects or leakage, you come up with a different device. You go from planar to fins or carbon nanotubes.
Buehler-Garcia: We’re very interested in silicon photonics. Based on what our kids need for Facebook and the number of server farms required for that, we will need more nuclear power plants just to power these server farms. But if you can double the energy efficiency, there don’t need to be any additional power plants. That’s where 2.5D and III-V come in. There are new opportunities to do that. Photonics is like DAC 15 years ago. It’s coming, and there are more things like that. And it’s going to be a whole different generation of designers working with it.