SoC Ecosystems Become More Tightly Integrated

By Ed Sperling
SoC ecosystems are changing. Quality and focus are replacing volumes of names as companies that fund them begin to narrow down which partners add the most value and which markets they need to target.

Establishing a ring of allies is nothing new, of course. IBM had its circle of most trusted software partners back in the 1970s when mainframes were the dominant computing platform, and Microsoft built its fortune on creating the longest list of applications developers in the PC industry—to the detriment of IBM’s OS/2 and various flavors of Unix. But in SoC development this kind of approach has been much less structured, in part because the relationship between companies was never as well defined and in part because disaggregation is, at least by historical standards, a relatively new phenomenon.

“What’s changed is that the ecosystem is not horizontal anymore,” said Art Swift, vice president of marketing at MIPS, which has been hiring experts from different segments such as networking and broadcast to understand which companies should be included in the new vertical push. “We’ve got a lot of spending around verticals right now. It’s about two to three times what it was two years ago, and last year it was up another 50%. We’re trying to understand how products are used and what is the software ecosystem that’s required. But it’s tied to a specific application. We’re no longer interested in the biggest and broadest ecosystem.”

That sentiment is being echoed across the semiconductor industry as companies begin to grapple with complexity and figure out what they do internally and what should be outsourced, either with IP or services.

“In the past the attitude was that you could integrate with our products if you want,” said Shay Benchorin, director of marketing for Mentor Graphics’ embedded software division. “That’s no longer the case. The question being asked now is how fast you can bring the customer to a full working system. It forces you to focus because you cannot do everything. The big realization for us has been that a horizontal solution won’t cut it anymore. The same thing is happening on the hardware side where silicon vendors now have to provide more software. And even inside the SoC the blocks are richer in content than they were in the past.”

He noted that the integration task is becoming much more difficult, which was the impetus for some recent bundling around Mentor’s Nucleus real-time OS. “From a consumer point of view this is good news. The price is going down as silicon integration increases. But the window of opportunity is shrinking.”

All the major EDA and IP companies are watching the same trend unfold. “At 28nm and below, we are no longer just looking at IP without a deep understanding of the process technology,” said Neil Hand, group director for product marketing in Cadence’s new business group. “There’s a whole chain of requirements and you need to make sure people can drive these requirements.”

Cadence has shifted its IP strategy away from relying on third parties because of that complexity. “It’s not practical working through these issues with an ecosystem,” Hand said. “Once you get away from the bleeding edge, you can go to an ecosystem approach. And we haven’t completely abandoned our open integration. But we are more selective. We have a second tier of partners that we use for collaborative development, but we also do a lot of stuff ourselves.”

Changes in 3D
The relationship between IP vendors and other companies is even fuzzier as the industry heads into 2.5D stacking using interposers and 3D stacking using through-silicon vias.

For one thing, the number of choices is skyrocketing. “Everyone is sourcing more IP,” said Ajay Lalwani, vice president of strategic sourcing at eSilicon. “The complexity is exploding and there’s no one-stop shop in IP. We’ve seen this in EDA tools where there is increasing specialization for analog, RF, and microprocessor core hardening. And we’ve seen it in the manufacturing supply chain. Over the course of the past 12 months we’ve added 20 new suppliers. Managing complexity is a big challenge, and that complexity means managing more ecosystem partners.”

The problem gets even worse in 2.5D and 3D, where there is the potential to fuse together almost any combination of chips and functionality, from analog to MEMS.

“We need to share knowledge about interfaces to make all of this work in 3D,” said Navraj Nandra, senior director of marketing for Synopsys’ DesignWare Analog and MSIP Solutions Group. “This changes the relationship. It literally requires a knowledge transfer through experts who can do the brokering translation between experts. There are few companies with expertise in various areas. We need people with modeling expertise and people who understand quality issues like thermal/electrical stress for 3D connections.”

The problem is that in a new field such as stacked die, there also needs to be a fair amount of experimentation. Companies learn from mistakes, but these are problems that are not easily outsourced. “It requires a lot of high touch with some very smart people,” Nandra said. “This expertise is not common, which is why we’re seeing middlemen in 3D. These are 3D experts working in startups. There are lots of people thinking about this on Sand Hill Road.”

Those experts may be necessary to overcome reluctance to share their corporate secrets, which will be essential in the development of new IP. By all accounts the amount of IP in designs is growing. Naveed Sherwani, president and CEO of Open-Silicon, said that last year the company was using five pieces of IP from two vendors in derivative chips. Now it is using 10 pieces of IP, valued at $5 million to $10 million.

“What we’re looking for is IP vendors who give more of a road map about what they’re doing,” Sherwani said. “Derivatives require planning. This is all becoming much more planning-oriented.”

Sherwani noted that Open-Silicon has always had an IP acquisition group, but now it requires much more work to ensure the IP will work in designs. The IP also is much more oriented toward vertical markets. “This is much more challenging at 28nm, though. There is more IP from different vendors using different nomenclatures, which means there is a higher chance of making a mistake.”

Conclusion
Ecosystems are certainly not going away, but they are maturing as the problems that need to be solved increase in complexity. That maturation puts more of a burden on all of the players in the ecosystems, and it means there will be a perpetual weeding out of those companies that don’t add real value to the companies that are at the center of that ecosystem and footing at least part of the cost of maintaining it.

People are only as good as their partners, and as those relationships become deeper the dependencies continue to grow. The growing consensus is that more companies will need better partners. The challenge is choosing your partners wisely.