Open-Silicon adds Silicon Logic Engineering—for a good reason

SoC design practice has changed profoundly in recent years. While EDA vendors, IP suppliers, and the marketing departments at FPGA companies seem to think that every new SoC requires a $250 M start-from-zeros design effort, in fact SoC design has bifurcated into two distinct flows. One flow creates a platform design: a completely new SoC to serve a new application. That effort really does start from nearly scratch, and it requires an increasing wealth in people, time, and money. The other flow modifies the platform slightly to create a derivative design. The derivative flow exploits the platform as much as possible, sometimes just replacing one block in the physical design without changing the rest of the chip at all. Consequently, a derivative design may only require a dozen engineers and a couple of million dollars.

As global recession clamped down in 2008, many companies made hard choices based on this bifurcation. The platform team contains the crown jewels: the application specialists, architects, verification gurus, and new-design trouble-shooters. The derivative team, perhaps unfairly, can look like a group of routine implementers—that is, they can look dispensable. So with the wolf at the door, the derivative teams were vulnerable to outsourcing.

That’s where Open-Silicon enters the picture. The company was originally conceived as a turn-key back-end ASIC design shop, with management and technical customer relations in the US and a sophisticated back-end process in India. That model was highly successful in good times, and Open-Silicon built a strong customer base among fabless start-ups. But in hard times, start-ups aren’t the best folks to have in your accounts-receivable ledger. So president and CEO Naveed Sherwani set out to build a base among larger companies. Just as the larger companies started looking to outsource their derivative design efforts.

The result is that Open-Silicon now finds itself doing significant business in executing derivative designs. This transition required additional expertise. "To do derivative chips successfully, you need to have front-end design expertise," Sherwani explained. The derivative team has to know the platform well enough to grasp the implications of the changes, which can go two or three levels into the platform.

Initially, Open-Silicon met this need by gathering a cloud of specialist design partners to assist them. But Sherwani felt the company needed in-house expertise at the architecture and RT levels. "It really helps to be involved in the original platform design, or at least to have the expertise to understand it," he said. Particularly this is true for verification. Planning ahead on the platform can limit the scope of influence of changes made later in derivative designs. And it is vital to avoid having to reverify a large portion of the platform during a derivative flow.

Fortunately, an opportunity appeared. Silicon Logic Engineering, a full-range chip design team, spun out of Cray Research in 1996. The team was later purchased by Tundra Semiconductor, which was itself bought by IDT. IDT suggested that the team—still working away in Eau Claire, Wisconsin—was available, and Open-Silicon bought it. So now, relying on its newly-acquired architecture, RTL, and implementation team, Open-Silicon has the ability to engage fully with a platform design, either during its development or later on, and to produce derivatives to spec.

This may be an indication of yet another step in the gradual disaggregation of design teams. It appears that SoC design shops are eroding away until they retain only their differentiating applications, architecture, and accelerator expertise, allowing all the other facets of silicon design to flow into outsourcing. Whether this trend holds up—or for that matter strengthens—with increasing financial pressure and complexity is going to be an interesting question to watch.