Experts at the table, part 1: From business models to the death throes of Moore’s Law; and the software challenge to the evolving value chain.
Semiconductor Engineering sat down to discuss upcoming challenges and hurdles to overcome for the semiconductor industry with Vic Kulkarni, senior vice president and general manager, RTL Power Business at Ansys; Chris Rowen, Fellow and CTO, IP Group at Cadence; Subramani Kengeri, vice president, Global Design Solutions at GLOBALFOUNDRIES; Simon Davidmann, CEO of Imperas Software; Michael Buehler, senior director of marketing for Calibre Design Solutions at Mentor Graphics; and John Koeter, vice president of marketing, IP and prototyping at Synopsys. What follows are excerpts from that discussion.
SE: Let’s give it 10 years: What are the biggest challenges the semiconductor industry needs to overcome?
Kengeri: We do invest billions of dollars in R&D, exploration, pathfinding, trying to find the next best device to continue to scale. There are definitely challenges. I think there are both technical and economic challenges. If you really look at Moore’s Law, it’s really the notion that the density will double every 18 or 24 months — all that has been good for so long. But if you look at the technical challenges, at the end of the day, if you really want to continue to do that at any cost, and just get there technically, there are some things that you can push: lithography scaling mostly. There are still challenges with middle-of-line, back-end, integration, and other things. But then, like I said, if you just want to get there, double at any cost, it is still doable. There are technical limits but for the next two or three generations, there’s no problem. The real issue is we all have visibility maybe for the next two nodes — 10nm, 7nm, maybe a little bit to 5nm — but then the problem is the lithography scaling, that’s number 1. Number 2 is electrically, we all have challenges in terms of the middle-of-line and the back end — it’s not scaling. People always talk about the device, and there’s a lot of focus on the device, which is important, but even if we were to scale the device and find new switches, the problem is the back end of the metal. That’s not innovating at the same level.
Buehler: It’s ‘ands’ — This is not technology but the business model that goes with it. The notion that there is exploding demand — they won’t claim they’re on allocation but everybody is pretty tight or full — and people say, ‘Of course, you’re 14nm!’ But it’s everything. Bringing it back to IoT concepts, IoT is a great word but really when we look at it, it’s engendering awareness of all the established node technologies as part of it. You need to manage your entire analysis across a supply chain. Our job as an ecosystem — the EDA side of this — the whole way the industry looks at how they’re going to grow and support their business, we’re seeing a change. Even if you look at the investment, $6 billion on a new fab — it’s not on 180, it’s not anything on 8″, and that will choke the apps. It’s not like the world is going to end, but we need to change how we are thinking. Just like thinking, ‘I’ll just go to the next node.’ 2.5 and 3D are options, but it really starts back to the business model of how to serve this because you need to think of it across multiple process node, and holistically across the whole system. That is an opportunity and a challenge to us.
SE: Because it requires a shift in how things have been done in terms of the ‘Build it and they will come,’ attitude?
Buehler: The point is everybody mouths the word ‘partnership,’ but partnership has always been about who has the leverage. And if I could defend the foundry, everything said when they’re in capacity constraint, they hate it because they don’t like not being able to give customers what they want. But there’s still test, and how do we improve test, and make things more efficient, is part of the discussion. How you design more efficiently, how you look at DFM, how you look at test, thermal analysis, power — not because it is mandated, but because it helps you get more net good die per wafer when you are only going to get X number of wafers, or X number of test slots.
Rowen: I’m so amused when people talk about the end of Moore’s Law, not because I think it lives forever but because, from an EDA industry perspective, we should be so lucky. In fact, people go on innovating and coming up with new devices, often surrounded by new complex rules which create this infinite opportunity to innovate within the tools because there are yet more constraints and types of geometric, types of logical, types of physical, types of complexity-driven verification that needs to be applied at every level or optimizations within the physical design. Certainly the death throes of Moore’s Law are such a marvelous opportunity for innovation within EDA, and I’m obviously speaking facetiously when I say the death throes because I think that plenty of people talk about it, but I remember when I was at Intel in the early 80s and there was deep concern about the end of Moore’s Law.
SE: But there is a limit to the physics of scaling, right? And that means Moore’s Law will end.
Kengeri: It will stop scaling in one dimension but it will put pressure on different parts of the value chain in all aspects but the key here is, if you take a step back, i think it will go toward cost per function rather than cost per device, and then how you get that could be through design optimization, innovation, new ways of integrating — end of the day, that’s what will become more important. People are even talking about cost per user experience. The key here is as long as you can continue to monetize those.
SE: So even if it’s not sticking with the strict technical, physics definition of Moore’s Law, does it really matter as long as innovation is occurring?
Rowen: I think that’s a very good point, and I think one of the ways to look at it is not Moore’s Law scaling but Dennard scaling — the voltage scaling. That died a long time ago in the same sense that people say Moore’s Law is dying today, and what it’s done is shifted the domain of innovation much more to the system level in terms of how you’re going to drive down power; how you’re going to exploit parallelism; how you’re going to use voltage management in novel ways; how you’re going to do innovative algorithms — and there’s been more cool stuff that has been done at the architectural domain in the last 10 years in response to that that really shows any time you bring enough creative people together and you pose a problem, you’re going to get some pretty wacky and some really clever ideas coming out of it.
Kulkarni: I would even assert that Moore’s Law is in overdrive, in fact. There are three things which I look at: mobility, energy efficiency and big data in terms of the next 10 years of what the problems are that people are going to be solving. In the case of mobility, Apple and Samsung are driving the node as opposed to the Wintel duopoly in the old days — the feature sizes were not driven by Microsoft, it just happened that the tick tock method of Intel created that effect. So it’s an experienced-based Moore’s Law. It’s not physics-driven per se — physics can be endless as new quantum tunneling and other ideas come in the picture. Already they’re talking about 3nm in 2020s — so if you break down the problem of what people are trying to solve we look at it as Moore’s Law, Moore On Moore, and then More Than Moore–we see these as vectors going at extremely high speed. Moore On Moore is really the mobility we talked about, and that will be endless with all of the connectivity and mobility applications. Energy efficiency will be another very important opportunity for all of us. People are trying to reduce power from wherever they are at. The processor guys are reducing from 125 watts to 100 watts; the disk drive guys are creating hybrid technologies for disk drives and for massive data storage, NAND Flash and solid state storage. And then we see big data analytics as a key driver when you look at sensor node networks and think about 50 billion devices connected, and about 1 trillion sensors by 2025. Energy harvesting: an incredible challenge we can all participate in with the ecosystem; new standards of security; and so on. We have to look at Moore Than Moore. We have seen devices today which are really systems — sensor systems. Some of the variable things from our company — we have actually started to simulate (we are a multi-physics company) — we have seen sweat sensors on the same substrate as the RF, and high speed digital processing from video — all coming together on a system for IoT — all within 2mm max.
SE: What ties all of this together is software. The software challenge is enormous.
Davidmann: As far as the big challenges that need to be solved for the next 10 years, I think the industry has to evolve to where it can help the most. The people we’re trying to help are building electronic products — it’s not just the semiconductor world — that’s where we’ve come from. And if we look at the evolution, we’ve gone from the simplest schematic capture, place & route, and laying it out — all the way through verification and IP. I think the next stage of evolution of the industry is that we have to help the people building the products and a large component of what they do is the software, and the interconnection of software, the security of software, so close to the silicon and moving up is the huge software industry — and we have to evolve that, and I think that’s where the industry has to change because there’s no point getting silicon out if actually the product just doesn’t function and can’t get to market. If software is a large component of the effort and the time, then the race to get the chip done in six months is not as important. It’s going to take three years to get the software stack, so we have to address the software.
Koeter: There’s an interesting dichotomy in the semiconductor industry right now, which is, in one camp there is the ‘win at all costs’ sockets because of the massive high volumes that every foundry goes after and that’s being bolstered by the ways of consolidation where it just becomes more and more important to win individual sockets more accounts for the ecosystem parameters but if you believe that the IoT has merit — and I do — I actually see a whole different value chain emerge. You’re going to see not just the single, high-volume sockets, you’re going to see a series of highly innovative products coming out of smaller startups and potentially for a completely different value chain so you might see systems companies getting in; you might see the re-emergence of the ASIC vendors. In fact, you might see, for example, a Google getting into a hardware product. You might see a completely different business model in the semiconductor industry, which is not related to how many dollars you get per chip, but is actually — as Xiaomi has said publicly — hardware should be free, and I want to monetize it through a user experience and content. Could we be at a shift where there is a very significant business model change in the semiconductor industry — it’s interesting.