Twelve Bar Blues is structured improvisation. A standard twelve-measure chord progression repeats tirelessly, and the experienced blues musician lays his soul over this monotonous harmonic structure like a fine linen drapery. Aart de Geus, President, CEO, and Co-Founder of Synopsys, the world’s second largest electronic design automation (EDA) company, is also an accomplished blues guitarist.
Blues in C
I (C) Measures 1-4:
The tonic orients the ear, providing a firm foundation of reference. In traditional blues, it is repeated for the first four bars of the sequence, setting up “home base” for the listener’s mind. When Aart de Geus co-founded Synopsys over two decades ago, logic synthesis was the tonic – the root chord upon which the engineering-centric company was founded. Logic synthesis set the key values of creativity, learning, and technological innovation that hardened the fledgling company so that it could survive and thrive in the tempestuous torments of two decades of The Moore’s Law blues.
“I played guitar starting around 14 years old,” begins de Geus. “It was very much around the campfire, scouts, gospel, Bob Dylan – those types of things. Then, somewhat by accident, I went to a couple of great blues concerts in Switzerland, where I grew up. It just hit me – a guy by the name of T-bone Walker, who was considered one of the fathers of Blues, was performing there. It was a big concert, not a small venue like a bar or anything, and after that I really never played anything else.”
De Geus may have found the Blues by accident, but he knew it felt right as soon as he was there. When we asked how he came to choose electronic design automation as a career, the answer sounded familiar.
 |
|
Aart De Geus
Photo by Laura Domela |
“I think many things have been accidental. When I was in Switzerland, working on my Master’s thesis, I did some design, and I needed to figure out if it would work. It’s a little embarrassing – I didn’t realize that simulators existed and so I wrote my own during my Master’s thesis. Later, I was working on my PhD and in parallel working at General Electric. I just sort of slid into the EDA side because of the job I had and because of the advisor I had.”
“Logic synthesis became the piece that really kicked my EDA interest into high gear. I had discovered that you could create more efficient logic by not using NAND and NOR gates but by using multiplexers. There were lots of problems with multiplexers, however, and one of them was that designers didn’t know how to design with them. I started on a program to automatically create a network of multiplexers, and, of course, that became a network of multiplexers plus NAND and NOR gates and inverters. That was called synthesis. I just didn’t know it at the time.”
If logic synthesis was the T-Bone Walker of Aart’s EDA interest, timing-driven optimization was his Gibson 335. The ES-335 was a seminal breakthrough, combining the concepts of the Les Paul solid-body electric with the acoustic resonances of the hollow-body. The integration of the warm tone of the hollow-body into the feedback-resistant stiffness of the solid-body was a landmark step in electric guitar evolution.
“I have probably a dozen guitars now, but I’m very partial to the Gibson 335,” explains de Geus. “I have a couple of original ones. Of course, I also have some of the more ad-hoc street-fighter sort of guitars like a Strat [Fender Stratocaster].”
In the early days of synthesis technology development, Aart found another pivotal breakthrough technology that would turn out to be a landmark evolutionary step – timing-driven optimization. “We were still doing just combinatorial logic, but we had one very very big step forward,” recalls de Geus. “We were not only optimizing the logic for area, but we had some concept of what the critical path was through it, so our synthesis was timing-driven. This distinguished it from other efforts at that time, and, as we moved forward, it really distinguished us from the pack. Still today, timing is the key in design.”
“In the early days, SiLC and Trimeter were the other two startups that were about a year ahead of us. Trimeter had a rule-based system. They had the concept that you could keep on adding more and more optimization rules and it would keep getting better and better. That is only true, of course, until you add one more rule and it gets worse. The fallacy was that you could just keep on going with a rule-based system with the user adding more rules. SiLC was really taken with behavioral synthesis, and their problem was that they didn’t really get competitive with what a human could do. On the other hand, we were extremely competitive with what humans could do. Sometimes, we’d have a circuit that the customer had worked on for weeks, and we’d come back 30% faster. We also had the situation where customers would say, ‘This is clearly magic; something must be wrong…’”
“My background was EE, but my talent was managing students more than anything. I discovered that I could hire a summer student and then two and then three and then more – at one point I had twenty-seven, and they needed tables and equipment and so on. Meanwhile, I had a stable of very motivated students and they did not know what isn’t possible, so they would try interesting things. GE discovered this, and I became a very involved recruiter for GE. I’d sit through 16 interviews in a day, and at the end I’d have to judge the people and refer them to the GE entity that might be interested. It was a superb training ground for me on how to quickly judge and locate top students as well as selling them on the company.”
De Geus took advantage of that technology and the enthusiasm, creativity and intelligence of those students and co-founded Synopsys in 1986. Synopsys came on the scene just as the EDA industry had settled into a happy stasis. At the time, ASIC designs were created with schematics using elaborate graphical editors, and then netlisted out so that simulators and layout tools could continue the trek to silicon implementation. Synopsys’s commercial-grade synthesis technology broke that mold and began to transform the industry from schematic entry to language-based design, with logic synthesis doing the detailed optimization.
IV (F) Measures 5-6:
After four bars, we feel like we’re cruising. Life is good. Now, it’s time to establish some motion – some tension. Without that harmonic directionality, we might be listening to modal music. In blues, conflict is introduced metaphorically by the subdominant chord. Most Westerners’ ears go there instinctively as the pre-dominant choice. The second four bars begin these turbulent times. In Blues lyrics, the struggle is often disclosed in this second line. Changes are a’comin…
Synopsys broke the industry’s momentum toward highly-integrated “frameworks” of schematic-based design tools. Customers wanted the productivity boost of language-based design far more than they valued the smooth integration of previous-generation technology at a lower level of abstraction. The “point tool” era was born and the EDA industry scrambled to adapt.
Innovation moved out of the big companies and into the domain of the startup. Big EDA companies optimized their sales channels and monitored the herds of startups, looking for the high-value innovations that could be acquired for a song and distributed in mass with the power of the big-company marketing machine. Many began to believe that big companies could no longer innovate and that the “acquire and assimilate” method was the model for the future of EDA.
I (C) Measures 7-8:
Two bars into the second line and we’re back to where we started – the tonic again. We’ve introduced change and tension, and then retreated to home – back to our roots.
De Geus and Synopsys disagreed. Sure, they did their share of acquiring, but de Geus believed that engineering innovation must remain a core competence of a large, successful EDA company. “We still do massive innovation,” de Geus explains. “The formal proof is that we’ve kept up with Moore’s Law, which hasn’t skipped a beat in forty years. As an exponential, we’ve held with it. I would dare to say that of the three components of Moore’s Law (one is manufacturing, one is EDA, and the third is the architectural ingenuity of using it all – the design) – of the two that are foundational, we’ve led the industry in an ongoing basis. The reason innovation doesn’t feel like it did in the early days when we went from zero to something is that everything still has to be backward compatible. Each design done in the last forty years has to still work today – and they do.”
Now, the acquire-and-distribute model is breaking down industry-wide. As the problems get tougher, the ability of a single point-tool technology to be useful diminishes. De Geus elaborates, “The systemic nature of the problem is what is now dominant. As a problem becomes larger, you go from optimizing one dimension to two to three… they are all interrelated. You can’t think about speed without power, power without signal integrity, integrity without yield… any time you optimize one thing, you may be screwing up some other place. We’re already vastly beyond the point tool era where innovation comes from startups developing point solutions.”
But, is logic synthesis now a solved problem? “I’ve been astounded that, after twenty years, there is such change still. Now, for example, inside synthesis we’ve sucked in very fast placement. Since the delay has moved mostly to the interconnect, the placement really counts for a lot. One thing you might say is, ‘Well, do a placement every time you make a logical change.’ Unfortunately, when you make a few trillion logic changes during optimization, you cannot do that. Instead, you try a very cheap, very fast approximation of placement, and it turns out you can get great results. It makes the synthesizer more complex but much more powerful. This has a very substantial impact with today’s designs.”
V (G7) Measure 9:
The final quartet of measures – nine through twelve, begins with the ultimate tension-builder – the dominant. The harmonic seventh’s 7/4 interval lands close enough to the classical equi-tempered dominant seven to pass on piano, but real blues artists bend the pitch down to its proper blue note on more flexible axes. The pull of the dominant back toward the tonic is almost irresistible, though, so the big jump to five leaves us with the almost insatiable inevitability of a return home.
Won’t RTL description eventually be replaced by a higher level of abstraction like behavioral code in something like C, C++, or SystemVerilog? “The abstraction rise is already done,” Aart retorts. “It’s called IP re-use. Of course RTL stays and logic stays and transistors stay and atoms stay. We have a hierarchy of abstractions throughout the design flow, and at each level we can optimize. In reality, though, RTL has long been replaced by IP re-use.”
Synopsys did venture into the behavioral synthesis arena early with their “Behavioral Compiler” product in the mid 90’s. “Behavioral Compiler was not sufficient to get the quality of results that a good architect could achieve manually,” de Geus reflects. “We’ve taken a lot of those capabilities and put them into logic synthesis – capabilities like retiming across register boundaries — but the notion of moving to a different building block size has actually had much more impact. As you move to bigger building blocks, however, those blocks become a lot more domain-specific and function-specific. You don’t do 48 different type USBs – there’s basically one.”
IV (F) Measure 10:
We coast through the subdominant one last time on the way back home to the tonic. This is the final release of tension. In Blues, the end of the lyrical stanza is often on this measure with the final two bars of the tonic reserved for instrumental solos.
Isn’t EDA now challenged by the shrinking number of ASIC design starts combined with the mind-numbing complexity required of tools in the latest and smallest process technologies? “The number of designers has not shrunk at all,” de Geus explains. “For complex designs, much more work is going into each project. What has shrunk since 2000 is the number of designs, although I think even that has reached a plateau. The designs we are now doing are dramatically more complex than they were in 2000. So much is going in at the high end into those designs and making them work – in terms of people years, complexity, sophistication of the tools used, and the number of checks done. That is one of the premises for continued growth for EDA, but these things are tied together. Everything in IC design has become much more interdependent, and that complexity is addressed only by EDA.”
I (C) Measures 11-12:
The 12-bar blues sequence ends where it began, providing a release of tension, a sense of closure, and a solid preparation for a new beginning, where we start the process all over again. Somehow, though, the tonic feels different now, in the aftermath of the journey, and we feel that the tonic represents both the ending and a segue to new beginning.
How does a company like Synopsys continue to succeed for over two decades in the face of all the chaos and change in the electronics industry and not fall into the trap of following a single technology or market off into oblivion as so many high-tech companies have done? Aart is ready for this one. “Certain things bind a team. There are some basic values that bind a team even when things get tough. These fundamental values are very deep. I actually teach them in the first three hours of our internal management training course. First, is ‘say what you do and do what you say’ – and if you just get that – that’s pretty good stuff. Ethics and integrity have different dimensions for different people, but they all end up in the same place with that kind of principle. This is part of the longevity of a good company. With Synopsys, we came to this field with an intensity and a love for technology and a desire to be best-in-class. We are definitely techno-nerds. Technology leadership is part of our DNA, and, for 21 years, Synopsys has always been close to the most advanced chips in the world.”
“There are other beliefs we do have that individuals – no matter where they are in life and no matter what they do – always have immense potential to grow beyond where they are and what they do today. If a company can offer an environment that supports that – and even account for the fact that individual development is in waves (and sometimes stuff has to hit the fan in order to get the wave to kick in), that is an environment where learning is viewed as a key value and a key reward. I’ve been a beneficiary of this myself, because many times I’ve been the bottleneck as much as the enabler of Synopsys’s success, and hopefully this has given many other people the opportunity to step forward.”
“Right now, we have entered a new phase of the industry where our team has to understand as much about economics as technology. Six of seven people that we started Synopsys with were summer engineering students on my team. If you look at our general managers today, every single one of those is one of the deepest people in their area of expertise. The management challenge for them is then step-by-step broadening their business acumen and business experience while still being able to explain three-dimensional cross-capacitance. Deep high-tech is really, really deep, and we’re at the heart- the deepest part of deep high-tech.”
“Deep high tech requires an understanding of the field and the technology that one cannot get quickly. Now, though, we are entering a phase that requires understanding of both technology and economics – I call it ‘techonomics’ – the intersection of technological discontinuity and economic discontinuity. The ‘techonomic’ understanding of what’s going on is much more crucial today in how we engage with our customers. We have to fully understand both the economic impact of what’s going on with them as well as the technology they need to solve their problems.”
“In technology terms, Moore’s Law had a sharp change in trajectory the minute power became a key issue, because every new process node did not immediately imply that everything goes faster. That tradeoff led to, for example, AMD and Intel completely revising their architectural trajectory and going to multi-core. That’s a good example of Moore’s Law after 40 years having an inflection. The other inflections are economic in nature – the cost to keep playing the game is much higher. The cost to develop a new technology node at 32nm is estimated at $1.5-1.6 billion just to develop the technology. You also need a fab, and they are $3.5B apiece. Right there, how many people can play? In the past, Moore’s Law was a no-brainer. Now, it’s definitely a ‘brainer.’”
“This is a phenomenon that has reset how investments are allocated. Companies like TI, Freescale, Sony, and many others all have gone ‘fab-light.’ The center of gravity of the market has gone to a much more consumer-centric place because computation and communication have merged. On the technology side, the mobile phone is really a massive computer that happens to have antennae. On the economic side, globalization has hit the market with sufficient force so that the number of mainstream customers has almost doubled, but the average socio-economic level is lower. This means that cost suddenly becomes predominant.”
As an EDA company, do you have to go through and re-work your whole product infrastructure, when you’ve been going along for years and trading off area versus performance, and suddenly, there’s area, performance, power, and manufacturability, and the order of importance is changing?
“Yes, particularly because coherence with the past is so important,” de Geus responds. “Depending on how good an architecture you have, you can potentially optimize more than just one or two constraints. We fundamentally have looked at what we do for a decade and a half as multi-valued optimization. You can see them coming -manufacturability, reliability, yield – those things are all coming up. I could summarize EDA as the intersection of physics and function. Function is moving into embedded software, but physics is coming up like crazy. It’s an exciting challenge.”
Bringing it all home:
Does your love of music impact the way you look at and manage the company?
“There are many levels to look at that. The left versus the right brain – many engineers are very creative – musically or otherwise, and they happen to have found a manifestation of that creativity in solving complex engineering problems. Many musicians are around the organization as well, and at a very fundamental psychological level, some interaction exists between the creative skills required for music and for engineering. In blues, you have fundamentally a chord structure that is the basis for which you get together. The structure forms the plan, but it is the improvisation that makes it interesting and valuable.”
“At a much higher level – an improvisational band is a very complex and interesting group behavior. The key skill is less the playing and more the listening. If you can pick up what somebody else does and add to it, you now have group creativity. Something better is created by the group over the individual. You also have to have discipline. If nobody shows up at the same time, you can’t practice. There is also the question of who is the leader. Every one of these dynamics in music I can point to in members of our executive teams going through these same challenges, so there is certainly a lot of correlation from the musical side.”
At Synopsys, the answer to the question of who should be the leader has been stable for quite some time. Aart de Geus has remained at the helm far beyond the point where most startup CEOs pass off the torch to board-appointed hired-gun executives. The company’s continued success speaks volumes of the versatility and vision of this man – technologist, musician, visionary and leader.
