Executive Insight: Lucio Lanza

Lucio Lanza, managing director of Lanza techVentures, a former Intel engineer, and the 2014 Phil Kaufman Award winner, sat down with Semiconductor Engineering to talk about the shrinking number of startups, future investments, new opportunities in EDA, Moore’s Law and the Internet of Things.

SE: You’re one of the last VCs still actively investing in EDA. Why?

Lanza: There are several individuals still doing it—, Jim Hogan—and some angels. I like to call them fallen angels. Often it’s me and other people and it’s a rifle-shot type of fund. I get people involved in a specific investment where we all feel that they can contribute to a company. They understand the business and are a good resource for that particular company. They’re not just connected to the company with a receive-only line. They’re a resource. There are only 5 or 10 people who invest that way.

SE: The number of startups in the United States and Europe has really fallen off in semiconductors and EDA. Will that change?

Lanza: Yes. I believe that when people say, ‘It’s slowing down, the excitement is finished,’ what they’re really saying is, ‘I’m slowing down, I don’t see the future, there’s no more excitement for me.’ It’s very easy to take that position. As human beings, it’s nice to think you’ve done what you set out to do. So the point is not whether the industry is slowing down. It’s whether the people investing are slowing down. In the 1980s, people in the communications industry, including some people from Bell Labs and definitely those at AT&T, would tell me how the telephone business is done and there was no excitement. I remember making a presentation 13 years ago about people saying the semiconductor industry was slowing down and the application of would not continue at the same pace. And I remember Sony saying there would be no bigger TVs. People don’t always see the future clearly. But if those people have enough influence they can stop the future, and that’s not okay. That’s what I’m always concerned about. Some industries are that way while others are not. Anything to do with semiconductors cannot be allowed to stop the future. With pharmaceuticals, you have a situation where what is financed by that industry—not the government, just the industry—is growing exponentially, and the number of drugs discovered is going down linearly. That cannot happen in silicon. If you are working under the disruption of silicon, you will be pushed aside.

SE: Where does EDA fit into this?

Lanza: One of the problems we have is EDA says, ‘Here are the best tools. If you speed things up, we will sell a lot fewer licenses, so don’t speed up. And Lucio, if you back a company that speeds up something, we’re going to sue you.’ So I won’t do that again. I’m going to do something around it. But EDA is not only impacted by the laws of silicon and physics and Moore’s Law. It is impacted by a situation in which the definition of EDA has to change because what you’re really designing is different.

SE: So what are we designing now, and what will be designing in the future?

Lanza: Let’s start with a misnomer, and this is important because names can stop progress. An SoC, or system on chip, should be called an HoC. The hardware is on the chip. The system is not. The semiconductor is the easy part. The problem is that we have more people designing software today than designing hardware. What are you doing to help people designing software? In my mind, that should be done by EDA. I ask them, ‘What are you doing to solve the problem.’ The answer I get is, ‘We’re putting more people on it.’ But can you find a way to utilize the resources and the power so the Moore’s Law revolution can help you? They create more, they use more, they create more, which allows you to create more. You have to find out where the positive impact is. Very few people are looking at the ‘no man’s land’ between the SoC and software. Everyone says, ‘My chip is 15% less power than the other guy’s CPU.’ Who cares? You are throwing more power in misses to your cache. That’s the way software works. You are throwing things away. It’s not about how fast you can run multiplication. How much multiplication do you really do? None.

SE: Software is a big stack. You’re talking about the bottom of that stack, right?

Lanza: Yes, the one that’s on top of the HoC. If you can let people design that layer quickly, then we can utilize it and design the next layer. It is as modularized as the HoC. What kind of cache do you need? Look where you’re spending time and money and find out how you can reduce that. And that’s very simple and very easy. You don’t have to be a genius for that. I’m not sure this will happen, but 10 years from now people may not count transistors. They may count how many hundreds or thousands of processors you have. Assume there is a day when you say, ‘My cloud—the one in my house—is 1 million processors.’ How do you use them? Sometimes I dream about the way chips will be in X years. Maybe they will be 3D structures built with advanced concepts. A simple post will provide all the services through a set of layers, each well designed for what it’s supposed to be and unique to that layer.

SE: What you’re talking about is stacking platforms, right?

Lanza: Yes, it’s stacking platforms. Do we have the tools today to design that? No. But systems are going to be more complex—not complicated, complex—built with more elements. We will need a set of methodologies and tools and software and understanding to do that. Now we’re just on chips. If you think about the possible change in the role of computers—servers have won the battle of how you manage the G&A of a corporation. It’s done. But they’re not part of the development process. In order for computers to be part of the development, we have to move to point where we have machine learning.

SE: That intuitive approach is artificial intelligence, and computer companies have been talking about that for 40 years. We still haven’t gotten very far. What will make these devices more intelligent?

Lanza: To be smart is not to be faster in a dumb activity. Now we need to add layers to make computers smarter. If you can have the interpretation of things, and significant numbers of different interpretations, and then use human beings to judge what they have done, that is the way the intelligent computer will be used.

SE: You’re talking about where IBM’s Blue Gene is moving?

Lanza: Absolutely. And we will have millions of servers, with tens of thousands of processors per chip.

SE: Now that we’re getting to the point where compute power, storage and bandwidth are available to make this a reality, what happens next?

Lanza: You will have to be able to design a generation of these computers quickly to get to the next generation quickly. That is not going to happen by doing things the same way. You will need a set of methodologies and tools and software that allows you to design the next system that is more complex than the existing one—and quickly. Then you start attacking different areas.

SE: This is something that EDA has been attacking on a micro scale. Does this move out into other areas?

Lanza: Yes, it has to. EDA has been moving out, but they have been resisting talking about it. I was talking to Joe Costello about a presentation he made to his sales force in 1993 about why IP was the future of EDA. That’s 21 years ago. We still don’t recognize that IP is part of EDA. It is.

SE: But all the big EDA companies do sell IP.

Lanza: They’re beginning to recognize that because the only area that is growing quickly is IP. If they had grasped this early on and driven it with a passion, they would be the absolute leader in this area and they would have changed their business model to the point where they would be making much more money than they do today. This is what Robin Saxby did with ARM. He came up with that business model at the same time. And Artisan came up with that business model after ARM, and they had to change their name from VLSI Libraries to Artisan Components because they wanted people to stop thinking about this as a library. There is only one knob to turn otherwise, which is cost. Eventually that doesn’t work. You come up with zero. The point of your libraries is that you see components that are yours. That’s IP. Those are your components on a chip. There’s a pretty big list of people who told me that was an idiotic business model.

SE: As intelligence moves from machines into the cloud and the edge of networks, what happens to technology?

Lanza: The transformation into information is where you can use the intelligence of people to help other people to use it, rather than having to invent it. There were moments in the past in which the technology was its own value.

SE: So the is just one slice of what you’re talking about here, right?

Lanza: Yes. The accent is on things is a good one What you’re trying to do is look at those things from something that understands only electronic signals and captures the behavior of those things and impacts those things. Every time you are trying to understand new things, you need analog. You have an enormous number of new analog designs.

SE: This is basically rethinking Moore’s Law, right? You’re talking about how the data moves and the information.

Lanza: Yes, and there are two ways to rewrite Moore’s Law. One is very narrow. I call it a challenge. The industry that performed around that observation took the challenge of taking something that was growing exponentially and made it linear. On the side of that, on the West Coast of America only, were a bunch of people who decided to attack a different challenge. They took the challenge of allowing you to design twice as many transistors in the same time and for the same cost. So EDA was the community that accepted the challenge, irrespective of how many transistors, to allow you to design a chip in roughly the same time and roughly the same cost. The EDA people met it. And if you add the IP, and think about the different branching of the silicon, you would have to have different branches in EDA. In the Internet of Things, it isn’t the number of transistors that count. You can control your home, your alarm system. When I’m traveling in Europe I can turn on my lights and open my door. Of course, so can the burglar. The other side is trying to control different phenomena. Some of that data are inviting an enormous set of application software behind it. So you get the data from your own body. What are you going to do with it? You can go to the hospital or not. Your liver may have these characteristics and this genome and you may have to consider these changes in diet or lifestyle. The moment this happens, the Internet of Things will have a much bigger impact. Once you bring in signals you bring in problems and the ability to do something meaningful, and suddenly you have another market. It’s expanding out of computers to all of these different areas.

SE: Does that allow EDA to begin attracting young talent because this is the cool stuff?

Lanza: That’s what I think is going to happen. First of all, there is no man’s land. Then there will be all the new stuff.

SE: Does the lack of startups hurt, or do we get startups in new areas?

Lanza: Yes, there will be startups in new areas. We will have software startups in machine learning, in specific applications like the automotive industry. On the other side, there will be startups to make the fundamental platform more powerful and easier to access. People will be able to think at a higher level. It will make learning much faster. You’ll be able to toss out 100,000 computers and put in a million every five years.

SE: What’s the new term for EDA then?

Lanza: It has be something involving systems and applications of computers. It’s not the automation. The concept has to be how we take the responsibility of packaging the intelligence. The intelligence of computers will be the new Moore’s Law. We don’t know how to measure that yet.