What Just Happened?

In the period of a year the semiconductor industry has transformed into something completely different.

popularity

Boy that went by fast. One minute, I’m waking up a little groggy on New Year’s Day, wondering whether the silicon industry is ever going to rebound. The next minute, it’s today and the industry had a good year, and is, in many ways, a completely different animal than it was 12 months ago.

Innovation is evolutionary, sure. But if you really think about 2013, you can make an argument that it was a watershed year in ways we won’t really appreciate for some time.

Going Up
Consider that FinFETs and 3D ICs really started to gain traction in 2013. The spring saw a flurry of announcements from fabs, EDA vendors, silicon and IP companies on 16nm FinFET technologies.

And on the business model level, we started to feel the impact of application-driven system design: Apple, Microsoft, Facebook, Amazon, Samsung and other big system OEMs are increasingly designing and building their own silicon.

Nimish Modi, senior vice president of marketing and business development at Cadence, who for years oversaw Intel server-processor development, put it simply:

“They’re designing their own system architectures because their business, storage, networking and power-management needs are so unique that that’s the most cost-effective path for them.”

Kurt Shuler, VP of marketing at Arteris, took it a step further during a Semico Summit presentation in November:

“The algorithms you need to do web search are different than what you need to do e-commerce are different than what you need to do social networking side.”

That’s big, and it has enormous implications for the IP-EDA-silicon-and-foundry ecosystem in the coming years.

What kinds of implications? Does it mean that these OEMs will someday start to build their own test and fab equipment as they did back in the 1960s? Probably not, but changes are afoot for design and integration-ready IP. Start with the role of silicon vendors in that design chain: If the system OEMs are designing their own silicon with IP from IP vendors or EDA vendors, how much less will they rely on silicon vendors for big parts of their system solutions? The answer isn’t 100%, but neither is it 0%.

Dream State
Tensilica founder, and now Cadence Fellow Chris Rowen, puts a poetic lilt to the conversation when he raises the question “What do applications dream about?” His answer in part is that applications (read engineers) dream bigger things today because of the degree of design freedom that Moore’s Law, EDA tool evolution and the rise of IP-based design have given them over 20 years.

The implication, at least to Rowen, is that EDA vendors must move “beyond EDA:”

“I would put it in crude terms of moving … past the focus of ‘this is how you do it’ much more to today ‘this is what you should do.’ Meaning, (EDA) is a lot more central to defining, architecting, building, programming these silicon platforms. And (it’s) less about taking the architect’s conception of it and doing the back-end implementation.”

This of course, like everything else in our industry, won’t happen overnight. It’s an enormous opportunity for EDA vendors, but industry analyst Gary Smith thinks EDA vendors aren’t being bold enough. EDA luminary John Sanguinetti (Forte Design Systems) put it to me earlier this year: the existing tool chain’s just too complex.

“It takes a long time for a tool to mature, especially one that occupies a big place in the transformation flow. If you’re doing a checker for something, that doesn’t take a great deal of time and effort. But for transformation like high-level synthesis, the time to maturity for those tools is really much longer than it used to be.”

It’s not pretty, of course. With additional complexity, engineering teams can no longer focus just on components of a larger design. That approach creates seams in the larger design, and problems arise where such seams come together, Modi says.
EDA vendors are trying to minimize those seams (that’s one reason you see so much IP-startup acquisition activity; it’s also one reason you’re seeing more complete hardware-software co-design solutions emerging this year).

Node to Joy
At the same time, we see enormous investment and early test-chip successes at 16/14nm nodes as the industry moves to FinFETs. But 40nm is still considered a mainstream node. More intriguingly, according to Geoff Lees, senior vice president and general manager for microcontrollers with Freescale, the 28nm node is a relatively inexpensive step from 40nm and will enjoy and a long and important run in designs for years to come.

“You could say that 28nm is the last simple node,” Lees said, while presenting at a technical forum earlier this fall.

You can expect all of this and more to evolve as we pick up the stray Champagne bottles on New Year’s Day and gird for battles ahead. Semico President Jim Feldhan sees a year in which semiconductor sales rise as much as 10T. That would be nice. We’ll see.

What I can guarantee you is that the relentless pursuit of productivity will continue to alter the silicon and system-design landscape as we know it (or think we’ve known it)…well, until I write another post like this 12 months from now.

What do you think December 2014 is going to look like?