Executive Insight: Lip-Bu Tan

Semiconductor Engineering sat down with , president and CEO of Cadence, to discuss his outlook on EDA, Moore’s Law and his strategy for investing in startups around the world. What follows are excerpts of that conversation.

SE: What’s worrying you these days?
Tan: There are a couple of things. One is the complexity of chip design at advanced nodes. It’s not just the chip itself, which involves performance and yield. It’s also time to market and the design cost. When you move to 16/14nm, there are double patterning and questions about EUV. There is a lot of consolidation and there are only a few big platforms. This is not an easy game.

SE: Do you see this as a threat to EDA?
Tan: There will be more and more consolidation. At 16/14nm and 10nm, a lot of companies are doing test chips rather than real production. It’s a very different story than in the past. A lot more innovation is needed and there are not many EDA companies left—and venture capital is not backing many. In semiconductors we still see some very good ones, but there are not many in EDA. So it falls on the existing players. That’s why I’ve been working very hard in the past few years to build an innovating culture. Our latest tools are all organic innovations. We have to do a lot more of that. That is the challenge for the semiconductor industry and for EDA. VCs have given up because it is too complex to make money, but it’s needed.

SE: Let’s dissect some of these points. Where’s the opportunity for EDA in all of this?
Tan: As chips get more complex, they need EDA partners more. There is more collaboration and more co-investment. Verification will become more critical. And as you move to smaller geometries in digital designs, there will be massive parallelism and 3D-IC and scalablity to 100 CPUs. You can’t just get by with 8 CPUs. How do you interconnect these designs? There also will be more IP. Some of the complex SoCs have up to 125 IP blocks. How do you put them all together and optimize them? Power is a big issue. has slowed down, so now we need massive parallelism, more cores, and power and interconnects become critical. Those create opportunities and risk.

SE: Doesn’t it simplify things a bit? In the past, you couldn’t really work closely with 20 or 30 companies, but with less than 10 it’s a different story.
Tan: Yes, and in some ways that was needed. Large IP vendors, foundries and ODMs (original device manufacturers) all have to work more closely together now. You can’t rely on a serial approach where I hand it off to you and you hand it off to someone else. IP vendors, tools vendors, foundries, the end customer and the software developer sometimes have to work in one trusted environment. Time to market is so critical that you can do this in a serial way. It has to be done in parallel. This is why we’ve created hybrid verification (emulation plus virtual prototyping). It can save six months. The other side of this is that quality has become very important. You need a first-time pass. It’s smart engineering. At 28nm/16nm/14nm you hear numbers like $250 million, but I’ve seen startups do it for $25 million.

SE: Is that just the NRE?
Tan: No, it’s the full silicon die. It can be done, but it has to be first-pass silicon. You can’t do another spin and another spin. One company I’m involved with—Ambarella—every node is first-time pass. That’s why it’s so capital efficient. There are a few examples of that. One startup did a very complex chip in a very capital-efficient way, but it was first-time pass.

SE: A lot of the tools developed at advanced nodes are for very specific problems. Can you roll back some of this technology to older nodes and improve efficiencies there, as well?
Tan: Yes. Some of the things you learn at 16/14nm and 28nm you can apply.

SE: Are you seeing a big push into 2.5D because of this complexity?
Tan: It’s still in the early stage. As an EDA vendor, you always have to be there ahead of the customer. 2.5D and 3D can be done. It’s more of the cost. But we are seeing some customer engagements at 2.5D.

SE: A lot of companies seem to be ready for a lot of options, whether it’s stacking die or shrinking features, but they’re not moving yet. Is that correct?
Tan: Exactly. We talk to the foundry and the packaging guys. They’re getting ready and waiting for someone to take the lead in terms of volume. The cost then will come down enough to handle the volume. It’s chicken and egg. At the end of the day, we follow the customer.

SE: At some point there has to be something on the menu, though, whether it’s chicken or eggs, right?
Tan: Yes, and we’re working on 3D-IC and advanced nodes. Our tools will be optimized, IP will be optimized, and when customers are ready we’ll be ready to work with them and help them solve the challenges.

SE: What do you think about moving beyond 10nm? Will it happen? We’ve got high mobility materials, multipatterning, issues with interconnects. Is it worth investing in?
Tan: Right now we’re focused on 16/14nm and 10nm. By next year we will see real production. Yield is still a challenge. Beyond that, [chief strategy officer] Chi-Ping Hsu is focused on foundry and IP strategies for these advanced nodes. We can see 10nm. Beyond that, we need to think about what will replace EUV if it’s not there. There are still a few challenges with the EUV light source, and the cost will be high even with EUV. We need to look for alternatives because we need to tell the customer when they ask us how they should plan their products. We’re talking with our foundry partners and the equipment manufacturers. We are reaching out to the supply chain and they are reaching out to us. We’re trying to figure out double and triple patterning and how to address this. My gut feeling is that 28nm will be a long node for volume. Next year we will know what kind of yield we will have at 16/14nm. And for 10nm, we are engaging with foundry partners. At 7nm, we are talking to people but not putting a lot of money into it yet. We are taking one step at a time. We want to be a little bit ahead, but not too far ahead.

SE: That’s a new strategy, right? In the past, EDA vendors wanted to be way out in front.
Tan: Yes. In the past, when you moved to the next geometry you saw benefits in terms of performance and power and cost. Now you may not see advantages in cost. It’s increasing instead of going down. That becomes a tradeoff. If you go down, what’s the benefit and do you really need it? Can you get away with smart engineering design at 28nm and 20nm? Our customers are very worried that if they put a lot of effort at one node, what happens if someone else comes in at the next node? Do they become less competitive in terms of cost or performance or power?

SE: Let’s swap topics here. Where is the startup investment money going these days?
Tan: We see a lot of focus on vertical systems. If you look at the recent investments I’ve made, a lot of them are in medical. That’s become very exciting for me, and it’s meaningful. There is a lot of diagnostics, monitoring, and even drug discovery, which is using semiconductor automation. Some of the startups had been very focused on cloud storage and networking, and a lot of these are ARM-based. A lot of people are talking about the . One issue that needs to be addressed is low power. It’s critical. A lot of new startups try to address that. It’s a humongous issue. If it’s machine-to-machine, it has to be able to last a month or a year. A lot of people tried to rush into the market using mobile components and found the power is too high. Power has to be extremely low, and there is a lot of innovation in that area. At least the last five startups I’ve seen in this area are addressing low power. There are a lot of techniques in this area.

SE: Where else are you seeing investments?
Tan: A lot of companies are addressing the next killer app. The killer app will be more medical—EKG or health monitoring. Communication is key. Programmability will be very important. That’s the reason we bought Tensilica. It’s very low power and programmable, so you can fill the gap quickly. All opportunities have their own challenges and requirements. That’s the way opportunity comes, and where I’ve been investing. There is a lot of talent and innovation in Israel. There also is a lot out of China and India.

SE: What is happening in China? It’s harder to get a grasp on what’s going on there than the other countries.
Tan: China is a big one. The government is setting aside billions of dollars for the industry. They’re smart. They look at a $300 billion industry and they already consume $161 billion—more than 50%. They project they will consume 65% by 2020. They view semiconductors as very strategic for them, and they feel that for national security they need to be self-sufficient. They’re willing to pour a big number ($19.2 billion, according to China Securities Journal) into this. And it’s not just the central government. Four provinces also are investing heavily. They are actively acquiring companies. They acquired Spreadtrum, RDA, Onetouch, they just made an offer for Omnivision—they’re going after some of the components that are important for them so they can bring back some R&D into China. That will be very important. Ten years from now, the top 10 customers I have will be from China. That’s why I go to China so often. If you only go every quarter, you lose track of what’s going on because they move people around. You need to be very current on who are the leaders, who are the decision makers, and which companies are getting mainstream government support. That’s very important because they get a lot of grants from the government. I call it ‘Follow the money.’

SE: So are most of the startups happening there?
Tan: Most of the startups are in the United States. And they are for very technically difficult problems, such as low power, medical, the Internet of Things, cloud and storage. Not many VCs are investing in semiconductors, so most of the startups are from my own network. A lot of my co-investors see the need for innovation and they work with me on dealing with the risk. We need to get some of the brightest students to come into this industry instead of going to Google and Facebook. If we don’t, we’re toast. We need new blood and new ideas. We proactively reach out to the best Ph.D. students. We need a lot of computer science capability to look at problems differently. When you put them together with people with experience, it becomes something great.

SE: Are there startups coming out in EDA, and do we need more?
Tan: Not many, and the VCs are not backing them. Most of them are funding by individuals. We keep track of all the individual investors and founders. We have a very robust M&A and investment team tracking startups. We stay in touch with them and track them. But the barrier is very high, and not many startups are willing to do it.

SE: It goes back to the complexity, right?
Tan: Yes. That makes it harder to do it standalone. It’s not impossible, but it’s not easy. And a lot of customers are asking whether we can provide the whole solution.

SE: What does that do to your business model? EDA has been feeding off of relatively low-cost startups. Doing internal development is much more expensive.
Tan: It’s not easy, and I’ve been giving a lot of thought to how to innovate these ideas. You can’t have an ‘A’ team and a ‘B’ team, where the ‘A’ team gets more stock options. We’re trying to drive a high-performing culture. Usually you cannot have a startup idea in a big company, so you have to use a small team and really focus on it with a very cost-effective way. One advantage we have as a big company is that we can try new ideas with our customers and they’re willing to help. A startup company can’t do that. But big companies also do things in a big-company style. That’s why we have an incubation center. We don’t want to just do a one-off, though, so the moment they have a breakthrough we want it to be part of a new tool. By doing just a standalone tool, how much value you can get is limited. So in some cases we’re doing spin-ins. But whatever they do, from day one we want to make sure it’s integrated. We are working through that.

SE: What does Cadence look like in five years? Is it an EDA company or something else?
Tan: We’ve moved from EDA to system design enablement. The foundation is core EDA. It’s the foundation of doing anything, and if that isn’t strong the whole thing will collapse. We have doubled and tripled down on digital. We have a series of products that are new, with massive parallelism—these are complete rewrites with big performance improvements. In a way we’re building up an entire wave to change your approach to design and implementation. So that foundation has to be strong. There is a lot of room to grow. Simulation has a lot of room to grow. Verification has a lot of room to grow. A second piece for system design enablement is IP. In the last 2.5 years we’ve done quite a few acquisitions.

SE: You’re number four in IP, right?
Tan: Yes. Our partner is ARM, so we won’t do anything in the CPU space. We’re focused on memory, which is why we bought Denali. We’re also focused on high-speed connectivity—SerDes, USB, PCIe. And with Tensilica we’re focused on audio, video and the Internet of Things. It’s programmable and low power. And in some cases, we’ve teamed up with ARM in smart meters and medical. And then the other part is the PCB business. We bought Sigrity to help up with system integrity and power. We have all three pieces for supporting the system companies. A lot of our systems customers are now vertically integrated. We provide the tools, the packaging and the IP. That system analysis becomes very important them.

SE: System analysis is one of the big holes in all of this, where you can do tradeoffs up front to consider what’s the best solution, right?
Tan: Yes, you need to be able to look at the whole system design requirements so that when the chip comes out the power isn’t too high and you need a re-spin and miss the market. And the system companies aren’t just in the United States. Those become our growth engine. It’s a much bigger market and we still have a lot to learn. We have developed a culture of constant learning. As a VC I also spend a lot of time with startups, which come up with brilliant ideas. Then I take those ideas back and ask my team, ‘How can we solve these problems, whether it’s power or performance?’