“You Can’t Run A Company Based on Hearsay”: A Rare Interview with Marvell’s Hands-On CEO, Sehat Sutardja

1/19/11Follow @wroush

Marvell Technology Group (NASDAQ: MRVL) is one of Silicon Valley’s unsung success stories. The chip design firm’s low-power electronics are found inside key media and computing devices from disk drives and smartphones to network controllers and Blu-ray players. But it wasn’t until after I’d won an interview with co-founder, CEO, and de facto chief engineer Sehat Sutardja that I realized how lucky I was to get an up-close look at this company. Sutardja, a Berkeley-trained electrical-engineering PhD, who’s known for his creative touch with everything from to project schedules to chip layouts, sets aside time for only five press meetings per year, company spokesman Daniel Yoo told me.

But while Sutardja spends less time in the limelight than many other Silicon Valley CEOs, it would be a mistake to overlook his company. In just 16 years, Marvell has evolved from a tiny maker of hard disk controllers—a market it still dominates—into a behemoth with $3 billion in annual revenues from a range of so-called “mixed-signal” technologies, meaning chips designed to handle a combination of analog and digital functions. Traditionally, building mixed-signal semiconductor devices required expensive materials and complex manufacturing techniques. Marvell’s genius, and Sutardja’s, has been finding cheap, scalable ways to build these devices using everyday CMOS (complementary metal oxide semiconductor) fabrication methods.

Today the company builds products for storage, switching, network, communications, cellular, and Wi-Fi applications, including many “system-on-a-chip” devices with embedded central processing units to provide more control. Its Armada processors, built around the low-power microarchitecture known as ARM, are used in e-readers, smartphones, portable game consoles, and set-top boxes, and are envisioned as the foundation for a new generation of inexpensive tablet computers.

But the company builds absolutely none of this stuff on its own: It was one of the first “fabless” chipmakers in Silicon Valley, keeping its own staff busy with design, engineering, and sales while outsourcing manufacturing, assembly, and packaging to fabs in Taiwan and elsewhere. That’s a widespread practice now, but back when Marvell was getting started in the mid-1990s, Sutardja says, one of its key challenges was convincing disk-drive manufacturers that the company could deliver real products, despite its lack of fabrication facilities.

Sutardja and his co-founders—his wife Weili Dai and his brother Pantas Sutardja—attracted some unwanted media attention back in 2008, when the SEC accused Marvell and a number of other Silicon Valley firms of backdating options grants in order to offer more competitive compensation packages to employees. Marvell eventually settled the case, admitting no wrongdoing but paying a $10 million fine. And while the Sutardjas themselves hadn’t profited from the alleged backdating, Dai agreed to step down as the company’s chief operating officer.

Since then, according to Yoo, Dai has remained actively involved at the company, taking the lead in creating Marvell’s Moby platform, a reference design for a new family of tablet computers intended for classroom and educational use. Marvell hasn’t yet found an equipment manufacturer to build an actual Moby tablet. But the work it’s doing to integrate low-power processors, display controllers, and wireless chips could eventually bear fruit in the form of sub-$100 tablets like the planned XO3, a laptop successor planned by the Cambridge, MA-based One Laptop Per Child Foundation (OLPC).

In my hour with Sutardja, I had time to ask the Indonesian native to describe Marvell’s innovation style—both how the company carved out a niche for itself, and why he believes so strongly in centralized, hands-on R&D management. We also talked about OLPC, the role that low-cost computing devices could play in economic development around the world, and Sutardja’s plans for leading Marvell to the next level of growth. That’s all recounted in the edited transcript below.

I also had time to ask Sutardja about his beginnings as an engineer, why he founded Marvell as a fabless semiconductor company, how the company played off its strengths in disk drive controllers to build Ethernet chips and other devices, and why the company is so committed to the ARM architecture. But frankly, that material will be of interest mainly to serious Marvell students and analysts, so I’m going to save it for a supplemental piece tomorrow.

Xconomy: Back in 1995, no one had heard of Marvell. You decided to build chips to control the read channel on disk drives, which were already a commodity business back then. What did you do to make your product stand out?

Sehat Sutardja: When we built our first silicon, we rated it to run at 200 megabits per second [of input/output volume]. That sounds like nothing now, but at the time, people were shipping disk drives that ran at 20 to 30 megabits per second. So we were building a chip that ran at 8 to 10 times the performance of what was shipping at the time. We overshot the target by about three years, not even realizing it. We were clueless—we didn’t have any marketing guys, so we had no idea that nobody needed this. But then Seagate came along and said, “If anybody else uses this first, we are going to be out of business.”

X: What, if anything, does that episode illustrate about how Marvell runs today?

SS: Over the years, whenever we have tried to enter a new business, we have been more successful when we entered by trying to obsolete what is available. Despite the fact that people know us in certain industries, there are always people in new industries who have never heard of Marvell. So if we cannot come up with a differentiating solution, what is the point for them to talk to us?

Also, we have built a lot of knowledge and instinct about what can be combined. Certain things are easier in analog and certain things in digital. When you are in the gray area, the analog guys get confused and the digital guys get confused and nothing happens. Everybody waits until somebody shows them how to do it. So whenever there is this gray area, that’s what we are always good at.

X: You are known for staying very closely involved in engineering decisions inside Marvell. How can you do that and still run a 5,000-person company?

SS: It’s actually closer to 6,000 now. I have come to believe that in semiconductors, you cannot afford not to know what is going on in a technical area. A lot of the things that a CEO has to make decisions on will be stuff that requires taking risks—risks on future investments and which path to take. Once you take a certain path, it’s a year or two of time potentially wasted. If you make too many of those mistakes, you end up going nowhere. Many companies that have been in this business longer than we have have made this mistake, where they do not know what’s going on in their own R&D labs. The R&D people can just say whatever the boss wants to hear. “Everything is peachy, we are the leaders.” Okay, nice, maybe for this year, but how about next year? Marketing will say, “Yes, this is the product plan.” But based on what information?

Everybody makes plans based on hearsay. I don’t believe you can run a company based on hearsay. A company has to be run based on real, scientific data. And it’s the job of the CEO. If the CEO is not designing the chip or writing the software, the CEO has to know what people are working on and whether people are on the right path, and make the call if they have to change direction.

X: But if the CEO has to be personally involved in every project, doesn’t that put an upper limit on the number of projects a company can pursue?

SS: It will be fewer, theoretically. But practically, actually, we have more projects than any company twice our size, because [in other companies] there are too many projects that go nowhere.

For Marvell to propagate this culture where the leader needs to be hands on, we could basically create a number of mini-me’s. They don’t have to have as much knowledge as I do. I happen to have a lot of knowledge, so I can make a lot of decisions. But if each vice president has one quarter of my knowledge, and if their knowledge overlaps each other, then the union of this knowledge will not have holes. But if the management has no knowledge, or only PowerPoint knowledge, then when you put things together you have holes everywhere, and when the time comes to make a call, nobody knows how.

X: Marvell is the largest single supporter of the One Laptop Per Child initiative. Why is that project important to the company?

SS: When OLPC started out with a target of a $100 laptop, it was aggressive, but nonetheless reasonable. It was maybe just a few years ahead of its time. But I believe that now it’s possible to build this thing for less than $100. For the first time in the history of mankind, we can now enable billions of people to have access to computing technology for $100 or less. So for the first time in history, we will truly be able to bridge the gap between the haves and the have-nots, between the poor and the rich. We can’t say the same thing about cars yet—we cannot build a car that costs $500 or $1,000. So the gap has been narrowed drastically.

I’m excited about this because if OLPC’s dream becomes a reality, suddenly we will have billions of additional users in the world, and I am hoping that out of this, maybe one out of a million people happens to be some genius living in a village in the forest, and has access to this knowledge and becomes the next Einstein. If it’s even one in a million, then out of the next billion people we will have 1,000 Einsteins. And if we have 1,000 Einsteins we maybe able to solve the energy problem, the disease problem, and all the problems that the rich people have created.

X: So with the Moby initiative, are you saying that you want Marvell processors to be inside those sub-$100 machines?

SS: We’re not just talking about the processors alone. Networking technology is getting to a very high maturity, so we have high data rates over local area networks, whether you’re talking about Wi-Fi or 3G or 4G, with LTE. Processors, local area networking, wide area networking, communications, storage—all of these functions can be put into a single device and provided to billions of people for less than $100. Everything you need can be put into one or two chips. And that’s not the end of it—Moore’s Law is not dead yet. All of these things will get smaller and more functions can be put into the same size chips. So the rich will have more functions, and the poor will have pretty good functions.

This technology is not a luxury item. Actually, the challenge will be for the rich countries to understand. In the rich countries, people are well off to begin with, and they don’t understand that as the next billions of people get access to knowledge and technology and people in the rich countries are ignoring the technology, we will be at a disadvantage, despite having money. This $100 PC can enable the poorest people in the world to be educated while people over here with $1,000 PCs aren’t getting educated. But it’s going to happen—there’s nothing to stop it. We should not want to stop it, either. We want to enable everybody on the planet to have a better life.

X: How do you plan to take a $3 billion company and grow it into a $6 billion or a $10 billion company? Do you think you’ll need to change anything fundamental about how Marvell runs?

SS: Chips are becoming more complicated. They are useless without software. As the company grows, we have to be able to resolve some of these challenges, so that things will work more seamlessly, and our customers can spend less time making the products work. That is the problem everybody has. But at the end of the day, the way to get from $3 billion to $6 billion is just to sell two or three times as many chips. It’s as simple as that.

There are many markets where we’re not playing yet. We don’t ship to many parts of the world, as a practical matter. So it’s also a matter of us building a presence in those regions, whether it’s South America or Africa or Eastern Europe. We have very little presence in India. Yet the technology doesn’t know boundaries. People in India want the same things we want. That’s easy to say, but not so easy to execute, because you need people on support teams, application teams, field engineering teams, sales teams. So those are things we need to build all the time.

Wade Roush is a contributing editor at Xconomy. Follow @wroush

By posting a comment, you agree to our terms and conditions.