Frontier Exclusive Visionary Interview for hardware, software, system related business and and academia
Frontier Journal (FJ): Frontier Journal is interviewing Dr. Aart de Geus, co-founder, Chairman and CEO of Synopsys. Aart, my first question would be, back to GE research days, how was Synopsys started after you spun off from GE Research and why Synopsys could achieve such huge success?
Aart de Geus (Aart): Okay. Well, in 1985-86 I was at General Electric working with a team of physicists and many other things and General Electric decided to get out of the semi-conductor market. Because of that I needed to find a new direction and I was able to convince General Electric let me and my group go and give us the technology in return for some investment in Synopsys. And it turned out that this was very good timing because the complexity of circuits was at the level where synthesis was necessary. At that time we were talking about five micron transistors and a hundred to two, three hundred gates. So the synthesis that we have created was special because it could not only create a correct network of gates but it could also automatically optimize for speed. And many of the designers that used it initially...but after they studied the results they were very positive because the results were about 30% better than what they could do manually. And as the circuits then became more and more complex, synthesis became more and more important...design on circuits of many tens of hundreds of millions of gates and the company has been able to grow with the technology.
FJ: I see. Now the success of, the commercial success of logic synthesis is probably due to two factors. One is the synthesis technology. Another is language, which is HDL. Could you elaborate a little bit about during later '80's and early '90's, the widespread of HDL language? At that time was probably invented later on acquired by Cadence and that twenty years ago now, as of today there are higher-level languages. For example, SystemVerilog and SystemC, higher level languages be beneficial to the success of next generation synthesis? For example, behavioral synthesis or something like that?
Aart: That's a very good question. Historically decided to invest in Verilog because, as you said, Gateway had developed it as a simulation language. It was a very clean and very fast language and we had decided we would support Verilog and VHDL, because both were available. The reason that over the years people did not just focus on Verilog is because Cadence did acquire Gateway and then tried to close the language from others. And that forced all the other EDA vendors to say "well then we'll have to support VHDL." Today if you look at much of the design in the world it's probably about 40% VHDL and 60% Verilog. The sole reason for that is that Cadence tried to close off the market and we therefore invested in both languages. Today, we see a new evolution. In the last three years, the coming about of system dialogue was a response to the need to go to a higher level and at the same time have a more efficient, more powerful language? We are seeing very, very rapid adoption of Verilog today. And we are seeing very, very verification and we are seeing it also broadly used for verification and for synthesis. It is better than System C because System C does not have enough understanding of the RTL behavior whereas System Verilog does get very good results and is very fast. And so I suspect that in the coming years we will see the adoption of System Verilog accelerate even more.
FJ: Okay. I see. What is the exact interpretation of predictable success claimed by Synopsys? Coorectnes-through-construction or search-and-refine or anywhere in between?
Aart: Okay. That's a good question, because as you know designing complex chips is a very, very difficult task set. And moreover there is no single correct answer because you can always say "I will make the circuit a little bit slower and then I can be smaller" or "I will try to give up on a little bit of power and then the circuit can be faster." So there are many, many solutions that all trade off against each other. We put together the term predictable success under the Synopsys logo. What we meant to express is that Synopsys is the best company at getting you from your intentions on a design to getting to really good results, not only because of the tools we provide but also because we have specialists in our support team that can help the designer for timing or for low power or for signal integrity issues; in other words, on the many things that can go wrong. And as much as we are a software company, I think in reality we really want to be a successful customer company and that is where the predictable success comes from.
FJ: Now in terms of EDA success at least we have Of course, we may have different designs. Now as a the semi-conductor process becomes more and more advanced which means we have to be a more and more accurate model. For example in DFM sector we have to migrate from role-based DFM to model-based DFM. Something like that. On the other hand, for any complex system, or any other chips, our system becomes more and more complex which we need higher level abstraction which is much more inferior of abstraction. Now that means, what's the impact on EDA when this is an obvious dilemma in terms of abstraction. Advanced technology demands for much more accurate abstraction. System complexity demands lower and lower abstraction which means, we should say high level abstraction and low level abstraction. How do we deal with this abstraction dilemma?
Aart: I think it is not a dilemma for EDA it is a great, great opportunity because the value to the end design is of course...because the more things you can put on a chip, the more functionality you will get. Yes. But in order to do that you need to know so much more detail all the way down to the transistor or even sub-portions of the transistor, the physics and the electrical behavior. The reason I'm saying it is opportunity for EDA is because all EDA is really a succession of - at the very lowest level you try to have a accurate model of the process and then from there you try to get an accurate model of the device and from there you try to build an accurate model of the transistor in spite of the circuit level and from there you try to get an accurate model of the gate and from the gate you go to small macro and then from there you go to large IT blocks and from the large IT blocks you go to functional blocks that include software. And so what I am describing here is the hierarchy of abstraction which, if it is built well, allows a designer to work at any level without needing to know too much detail below. Now of course to do that well you need to have very good tools and a good approach to it. I'm not saying it's simple. But I'm saying that this is the history of EDA for the last forty years. If you look at the early days people worked only at the circuit level. So really they didn't have that much modeling. They had just transistor modeling and then, as you pointed out, in the '80's synthesis really say "Hey. You can work at the gate level." And then from the gate level to the language to RTL. And today Synopsys is actually one of the key providers of very large IT blocks where you want the user of the IT block, let's say USB core, PCI corp, or to not look inside because there's too much detail. And so it's a very good question you are asking because the entire strategy of Synopsys has been three steps. One is to automate as much as possible design for time and area and power of course and then push downward to the modeling of the physical issues, because they become more important with smaller things, and push upwards towards IT and system design for efficiency. And so in your question, you essentially described our entire strategy.
FJ: Is that Synopsys' agenda for predictable success?
Aart: Absolutely. And by the way, the predictable success is now a much longer road than it was 30 years ago. You start much higher, much more functionality and you need to go much lower, much more physical understanding.
FJ: That brings a follow-up question which is that design-productivity gap between design capability and market window. It's widening instead of being narrowed. So of course one of the reasons is the EDA challenge. My question is the problem that we do not understand the design problems fully or actually we almost understand all design but we cannot find a viable solution or anywhere in between for the design productivity gap problem.
Aart: If I may politely disagree a little bit with you. I must say that in 30 years, probably every two years I've gotten the question that e design-productivity is increasing. Every three years. And I think it is not. I think it is not a productivity gap. It is an economic gap. Meaning that if you are willing to spend more money on design and that because there is always a difference between what one can theoretically fabricate versus reasonable to design, there will always be an evolution of tools to try to make things faster or smaller. But every time that we make one of tools faster, so we reduce the productivity gap, immediately the user says "Oh. That's great. Let me design something that's more complex." And so I think that the design-productivity gap is almost a constant over all these years and it's really more an economic gap or, maybe better, an economic equilibrium.
FJ: So that means the end-user of consumers' desire can never be satisfied?
Aart: Absolutely. Today you buy a computer, tonight you'll feel great and by tomorrow you'll think "well, it's really a little too slow." The same is true with cameras. It's true now with video and the number of pixels. Today you buy this incredible video camera and in six months you say "well, you know, I really wish it was high definition".
FJ: And so you're a technology visionary turned entrepreneur you've been running Synopsys for over two decades. So from running a start-up with just around a dozen people to running a transnational with several thousand people how do you manage to scale up your management expertise?
Aart: You know, that's a good and difficult question. It goes in waves. Meaning that I have always been a strong believer that you need to question everything you do and keep learning with every opportunity. In reality, as you grow a company there are periods where things go very well and there are periods when things are really difficult and you're not sure you're going to make it. And during the periods when things do not go well is actually the highest learning. So during those periods I always try to talk to many people, inside and outside of the company. I do not try to resolve all the problems myself but really try to engage the rest of the team.
FJ: In any established company, in software or other business, you've got a huge infrastructure and larger customer installation base. That means it is not easy to be changed, for obvious reasons. In that case, how would you attempt to forge innovation in nurturing technology?
Aart: Well, first I don't think having a large install base is all negative. Because you're correct to say sometimes it's more difficult to bring about innovation while simultaneously having to maintain backwards compatibility. But the positive is if you are a leader in a field and you have a large install base that also means that your customers tell you their most advanced problems. And so I have always believed that innovation is not something that just happens in a vacuum alone. It is greatly influenced and the input of customers and having driving customers in high tech is the best way to stay at the leading edge in innovation.
FJ: This question you may sensitive. You may decline to answer. Synopsys started, I think around one decade ago, Synopsys started an internal pilot project adopting some key concepts in logical efforts. And for some reason it did not manage to make a commercial success. What happened during that period?
Aart: This goes back to your previous question, which is innovation. In order to innovate you typically invest in ten different ideas and maybe three or four make it into the product and one or two become very important to a product. And so we have an advanced technology group that constantly invests in all kinds of ideas. And very often I find that an idea that was maybe rejected four years ago tends to come back in a different form many years later. And so it is actually important in technology, and by the way in business, also to be willing to invest in many but you need to be able to have the decisiveness to kill the projects that are going nowhere. Not let them going and going and going but at some point in time you say is this a good idea, a bad idea or a so-so idea and kill the bad idea, ask more questions of the so-so ideas and put more investment around the great ideas.
FJ: I see. So Synopsys still has advanced technology there, right?
Aart: Oh yeah. We also have advanced technology but we also have the business units, as well. So we're a large R&D company, moving the state of the art in not only synthesis but physical design, design for manufacturing, ID, verification - those are many areas and so we have many just absolute world-class researchers in our team.
FJ: So you also have a central R&D center or something like that and you also have some to the R&D Are there any R&D efforts in reconfigurable computing, migrating from field programming computing and also some technology we call heterogeneous computing. For example, combining a GPU with a CPU some kind of for R&D efforts for EDA support of those technologies?
Aart: In general there's quite a bit of attention to how the environment is not changing. And so it is clear, if we look at the most advanced chip, that more and more chips are going to contain a multiplicity of cores, be it multiple processing cores or cores with graphic or specialized solutions. The reason configuring of computing has been less successful largely because it had not had the volume and the development efforts that the very, very large semi-conductor companies can put behind it. And so we are absolute believers that the computer paradigm is evolving and already today we have a number of our products that run on a combination of large farms or multi-core solution.
FJ: Okay. So I have two more questions. You are a supporter of open immigration policy. What's your opinion of first-generation immigrants, especially in the high tech sector, whether they come into the US as graduate students or as engineers. So a lot of first generation immigrants in shaping a better, stronger and more competitive America.
Aart: I am a first generation immigrant and I know that, if you look at Silicon Valley, about 50% of the companies were either started by or are run by immigrants. And so Silicon Valley has benefited greatly from attracting some of the best of the brains, typically coming in through an F-1 Visa and going to graduate school and then staying around. Or more recently, through immigration via H-1B Visa. So if I were to speak from a United States or a California perspective, there is no doubt in my mind that the more brilliant people that can be invited here, the better for this region. Now simultaneously I am a strong believer that we are seeing now a globalization of high tech and quite a number of these immigrants have gone back including to China and India and in the past to Taiwan and Korea. The benefit of the training they had got here has been very, very strong for the local economies. At this point in time I personally feel myself like I am a global citizen and not an American. I don't feel like a Dutch person either. I feel just as much at home in China or Silicon Valley or in the Swiss Alps.
FJ: Great. Great. You are a great musician, beside a technologist and also an entrepreneur -- and by the way, do you play golf?
Aart: I would have to play either a lot more or a lot less.
FJ: So let's say what is the similarity between playing musical instruments and some similarity?
Aart: Oh sure. Because there's both an individual component and a group component. I play sort of blues-jazz which is sort of improvisational. One of the first things to learn when you play in a band is not to play, it is to listen. If you listen to what the others are doing and as you increase your own skill you find that the most fun thing is to hear what somebody else is doing and then support and emphasize it a little bit more. So that's called "playing together." And I'm seeing so often that one of the weaknesses in many management teams is they're all individual. That a really good team does much better than a really good set of individuals. By the way, the same applies to many sports too. And so in addition, it's interesting that many engineers tend to also have a musical bent. I think there are some studies that say that some forms of creativity and of spatial understanding are very much related to how one perceives the music. And there's even people that have studied, let's say the music of Bach, and have shown that there are enormous amounts of mathematics de facto behind it. There are we may not fully understand.
FJ: Great insight. Aart, I enjoyed spending time with you and thank you very much.
Back to Home Page