Programmable Devices: Paul Sasaki
Faster than a speeding FPGA

DynaChip Corp. is a fast company, a grand gamble that a startup can create an FPGA of such speed that it could break into the fiercely competitive PLD market. Speed is even a personal motif at DynaChip, where four of the senior execs drive Corvettes. But Paul Sasaki has a Honda.

That is entirely typical of the understated, nearly shy director of architecture development. One contractor, who had known Sasaki in his earlier days at Signetics, spoke with him several times at DynaChip before finding out Sasaki was the chief architect there. Yet Sasaki's reticence conceals competitive determination and independence that have become vital to his new company.

"The founders of DynaChip had one thing in mind-to create an FPGA faster than anything Xilinx could build," Sasaki said. "Initially, they wanted to do this with a unique process. They tried that for a while, and eventually became convinced that it couldn't be done. So they turned to architecture."

That's where Sasaki came in. With a background in PLD architecture at Philips Signetics-the land where the PLA (programmable logic array) was invented-Sasaki was highly qualified to architect a new FPGA. But more important, he was the bearer of a new idea, not widely accepted at that time: that the industry leaders were looking for speed in all the wrong places.

"Conventional wisdom then was that FPGA performance came from fast logic circuits and from efficient logic cell design," Sasaki said. "So the FPGA industry evolved very fast ways of implementing lookup tables and flip-flops. And they all moved toward the three-input lookup table that Jonathan Rose's work at the University of Toronto said was most efficient."

New worry
But as circuit geometries decreased toward half-micron, another problem began to grow. Transistors were shrinking, reducing their ability to carry high currents. And interconnect impedance was growing-particularly for the long lines that were so frequently the critical elements in FPGA circuits.

FPGA vendors recognized the problem, and responded with retrofits to their existing segmented-metal routing architectures. Device designers strove for ever-lower ON-resistance in the pass transistors that switched signals to and from the metal segments. Chip designers sought cleverer ways to arrange the mixture of short, medium and long metal segments that made up the interconnect pool.

But what the existing vendors missed, in Sasaki's view, was that the ground rules had changed. In the new, deep-submicron processes metal was expensive, but transistors were nearly free. "There was the opportunity to completely rethink the way interconnect was architected, using active tristate buffers instead of passive switches," Sasaki said.

It was that kind of rethinking the DynaChip board was after. Sasaki had been consulting for some time after leaving Signetics, and was attracted to both the challenge and the environment at DynaChip. "Basically, the board only said three things," Sasaki remembered. "Go faster than Xilinx. Don't touch any of their patents. And whatever you do, make it process-independent."

Still, Sasaki thought twice about DynaChip's offer. "Going to a start-up is not for everyone. It can be a very demanding environment, in terms of time and energy. You have to think through whether you are willing to make that kind of sacrifice of your personal life."

Once on board, however, Sasaki set about proving his architectural idea. Working interactively, Sasaki, the chip design team and the tool designers began to put flesh on the conceptual bones, creating DynaChip's first FPGA.

"It was absolutely vital to work directly with the chip design people," Sasaki said. "Too often, you see architects work in isolation. They end up creating something wonderful but unimplementable. You need an optimization of the thing the designers can actually implement, not just an optimal idea."

When completed, the FPGA followed closely on the original concept. It used relatively simple logic cells with three-input lookup tables, like most other FPGAs. But in place of pass-transistor switches and myriad different-size wire segments, the interconnect array comprised chip-spanning vertical and horizontal metal runs, broken at intervals by tristate buffers. The technique provided interconnect de-lays that were both short and predictable.

And the design was process-agnostic. "It happened that we implemented the design first with bipolar tristate drivers," Sasaki said. "That was the best speed/power point for that particular process and foundry partner. But there is nothing in the architecture that requires BiCMOS. Indeed, in newer deep-submicron processes pure CMOS will probably be a better approach."

Taking the concept to production required not only a unique vision, but a particular personality, DynaChip observers said. Sasaki's combination of a quiet, attentive manner in the presence of new information and comfort in working with other engineers as peers made it possible to implement a new idea with a seasoned design team. Yet another, less obvious element of Sasaki's personality was necessary as well.

That facet was competitiveness. It might not be obvious in Sasaki, unless you have played golf with him-or, rather, against him. "I consider myself a competitive golfer," Sasaki admitted. "Although not as much as I used to be. I just haven't been able to get out as much lately.

Fire in the belly
"You have to have a real competitive drive in a startup," he continued. "You have to really be determined to win." The drive is necessary, Sasaki said, to keep going when the fatigue and unavoidable difficulties of a pioneering startup make it hard to continue.

Now that parts are shipping, and customers are verifying Sasaki's point about enhanced performance, there is still no time to sit back and enjoy victory. There are field reports to follow up, new designs to pursue, a constantly shifting market and competitive environment to watch. And, on the horizon, the goal of an initial public offering.

"Sure, you think about that when you commit to the demands of a startup," Sasaki said. "There is always the idea that the startup will succeed and that you will be set up very nicely."

Sasaki can envision himself coming out of the DynaChip experience with a valuable nest egg. "It's not money for its own sake that's the attraction," he said. "It's the freedom the money gives you. You don't have to take the next job that comes along. You can work at the projects you think are significant, maybe helping create the environment."

Would that mean Sasaki would like to run the business next time?

"Absolutely not. I don't have what it takes to be a CEO. If you are the guy at the top, you have to be willing to deal with the administrative stuff. And you have to be willing to beat up on people sometimes. That's not me."

So what is the long-term goal that draws a quiet rebel of an architect into the startup world?

"I think," he began tentatively, "that the bottom line is . . . I'd like to be remembered for accomplishing something. Something significant that not everyone else had done. That would be worth it."

In the meantime, the Honda runs just fine.

Return to: TimesPeople 98