MADISON, Wisc. — How many CEOs in the semiconductor industry today are so possessed with their inner technology questions that they actually spend weekends looking for answers?
Given their engineering backgrounds, many CEOs at chip companies can probably be found in the basement tinkering on weekends. Rarely, however, do we encounter an executive who not only discusses his personal obsession with technologies, but also reveals a solution he’s found after countless sessions of quiet experimentation — all alone at home.
Meet Sehat Sutardja, CEO of the Marvell Technology Group.
Last month, he disclosed in a keynote at the International Solid-State Circuits Conference (ISSCC) his weekend projects: the development of a new interconnect technology called MoChi, and the Final-Level Cache (FLC) memories, which he says can substantially reduce the amount of DRAM main memory needed in a system.
In a recent interview in Barcelona during the Mobile World Congress, Sutardja said, “For the last 30 to 40 years, the computer system architecture has remained the same. The advancements of computers have always depended on the bigger and faster CPU and more memory.”
“We are going to fix it for once and all,” he said, with Marvell’s newly proposed system architecture which uses only a fraction of DRAM and puts DRAM in deep sleep. Marvell plans to launch prototype chips — based on the MoChi interconnect and FLC memories — at the end of this year.
It’s probably not inaccurate to describe Sutardja as the ultimate geek, still a hands-on designer despite his lofty position, and an engineer’s engineer.
In an era when a CEO in the electronics industry often sees his role as financial engineering in search of a high corporate valuation, Sutardja — who talks like an absent-minded professor about his pet projects — is a breath of fresh air.
'This is my hobby'
Sutardja, however, made it clear in the interview: “No, developing this [new system architecture] is not my job. My day job is running the company. This was my hobby.”
But he acknowledged, “I’ve been thinking about this for decades: why a computer system has to be designed this way.”
Recalling the time, early in his engineering career in the 1990’s, when he was working at Integrated Information Technology, Inc. (which later became 8×8) designing graphics accelerator chips, Sutardja said, “We got killed on the market” by competitors using bigger DRAM. “Since then, I’ve always wondered why everything requires so much bigger memory. I’ve kept the problem in my head.”
[Calling all tinkerers: Feed your soul at ESC.]
Obsessing about unresolved technical problems has been a lifelong habit. “I’ve been that way since I was 12 years old.” He started his career as an analog circuit engineer, so, he said, “I still think about how to build a better A-to-D converter. I work on my new ideas at night, sometimes call a professor, and discuss the issue.”
How the idea hit him
Over the years, Sutardia kept returning to the idea of a new, less main memory-heavy system. As he explained to Rick Merritt, EE Times, in an interview last month, the idea for developing a kind of virtual main memory hit the Marvell CEO “while staring at a task manager screen on a PC,” he said.
The task manager monitors real-time running applications and started processes, while evaluating CPU and network usage.
Sutardja said, “Then it hit me. Most processes were idle — 0%. In aggregate these idle processes take a lot of expensive DRAM. I thought maybe it’s possible to cache only active processes in less DRAM and leave inactive ones in cheaper memory.”
His solution was to use FLC to reduce the DRAM footprint, while relying more on SSDs and hard disks. Of course, this was just an idea that popped into his head. He knew that proposing a 90-percent reduction in main memory was crazy. “But my instinct told me that it could be done.”
Typically, Sutardja prefers working on his crazier ideas alone.
Although he isn’t against teamwork, keeping harmony within a big group gets more problematic when the idea of changing system architecture is as radical as this. He said, “You don’t want to offend anyone” who might have a vested interest in keeping a certain structure the way it is. He said he’s also wary of compromising the solution for the sake of keeping everyone happy.
Thus, the Marvell CEO kept working on his idea on his own during weekends. It wasn’t until two years ago when Sutardja finally introduced it to the Marvell team. At that time, he said he knew he had a solution, but “I couldn’t explain it properly.”
To read the rest of this article, visit EBN sister site EETimes.