So as its second generation of FPGAs come to market, Tabula is trying a new business model. This time, the strategy is to make things as no-brainer as possible: Tabula is producing pre-built boards and systems, rather than waiting for its systems-level customers to come up with them.
“We find it’s a challenge to walk into a traditional FPGA customer with something as mind-bending as Spacetime,” says CEO Dennis Seger. “It is hard for customers to wrap their heads around this topic.”
The target market is data networking — high-end packet processing in the data path, specifically — and the results are on display at the Intel Developer Forum (IDF) this week. Tabula is showing 100-Gb/s bridge and switch boards as well as a full 1U switch design. The switch uses two of Tabula’s ABAX2 chips, the second generation of the company’s programmable chip, to push a purported 1.1 Tb/s of traffic.
Tabula says it’s shipping these products in low quantities, although it won’t disclose how many customers it has or how close they might be to launching their own products.
A 3-D Trip Through Spacetime
Spacetime gives a Tabula chip the illusion of having multiple stacked layers. Picture it as a stack of chips hovering above one another like the floors of a building, with vertical connections between floors.
Tabula doesn’t really build the chip that way, though. Tabula fakes it in software.
The company’s chip reconfigures on cue to a different design — one that you could think of as the next layer in the 3-D stack. All the signals that are in transit would think they’d climbed to the next level and would continue on their way.
What do you get out of that? The motivation for Steve Teig, Tabula’s founder and CTO, was to make use of idle gates. An FPGA has large areas of logic gates that can be programmed to do arbitrary digital functions, but it turns out these gates spend a lot of time waiting for an instruction to arrive. Teig wanted a way to assign other tasks to these idle gates. As the ABAX reconfigures itself to the next “floor” of the 3-D design, particular gates and transistors get set up to handle another set of tasks.
All this “3-D” stuff means an ABAX chip takes up a lot less space compared with normal FPGAs doing the same tasks. It also means fewer worries about power consumption and heat. FPGAs have not proven viable competitors against application-specific standard products (ASSPs) — the mostly-single-function chips that are a specialty of Broadcom, among others. Tabula thinks ABAX could finally swing the fight in FPGAs’ favor.
(If you’re wondering about Intel’s role in all this — the chips are built at Intel’s foundry, using manufacturing processes that are a bit radical themselves. And at IDF, the companies are showing a board using Tabula’s chips and Intel’s silicon photonics.)
But as the company now knows, Spacetime is too much to swallow in one gulp. The technology might be capable of previously unheard-of functions, but not every customer is ready to dream up those functions, let alone program them into Spacetime.
What systems vendors can do is pick up a mostly completed design, add some customization, and ship it. For Tabula, walking into a customer meeting carrying a working system is a way to overcome the sense that Spacetime is cornered into a geeky niche. “You’re able to bypass a lot of the traditional resistance,” Segers says.
Tabula is not becoming a systems vendor; the goal is still to sell lots of chips. It’s just that those chips are now being sold inside subsystems, with all the software pre-loaded for a particular function — a bridge card, for example.
From that point, the customer is able to peer into Tabula’s code to add customization. In some of these designs, the chips are only being used to about 50 percent capacity, officials said.
As of 2010, Tabula had amassed more than $100 million in three rounds of venture funding, from investors willing to be in this for the long haul. Segers won’t say what Tabula’s funding status is now, but at least some of those patient investors are still around, it seems, hoping that a new business model will help Spacetime finally get its footing.