Chip makers set for 'fab-lite' future?
The fate of the mega chip corp.
ISSCC By 2025, vertically integrated megacorps such as Intel will be endangered species. Or not.
On Tuesday evening, at the International Solid-State Circuits Conference (ISSCC) in San Francisco, the fate of such integrated device manufactures (IDMs) was debated by a group of industry heavyweights during a wide-ranging panel discussion entitled "The Semiconductor Industry in 2025."
As might be expected, Intel Senior Fellow Mark Bohr sang the praises of IDMs. With such a structure, he said, a product's elements can be "more carefully co-optimized" and system integration is more optimal as well. But after listing the wide range of technologies needed to bring a concept to market, even he admitted that "vertically oriented companies can't be experts in all of these technology pieces."
Bohr also noted that IDMs have higher development costs because their structure doesn't allow them to share such costs with other companies. Horizontally oriented companies can develop a product, then sell it to many other system integrators, thus distributing the development costs.
Bohr's vision of the successful 2025 semiconductor company blends the IDM and horizontal models: "As you might guess, it's maybe a modified version of the vertical integration approach, one that I call the 'core' vertical integration, where you have an IDM that is primarily a vertically integrated company, but they focus more on the technology pieces that are key to the products that they are making," such as process R&D, wafer manufacturing, and chip design.
In Bohr's core-integration future, IDMs will handle the leading-edge products - not the least because an IDM structure can typically develop products with a quicker time-to-market, since their divisions can communicate more openly and quickly than can horizontal companies. He envisions multiple horizontally integrated manufacturers producing what he referred to as "second-wave" products.
Wally Rhines, chairman and CEO of Mentor Graphics, however, pointed out the trend away from IDMs: "If you look at [top ten] companies in 2007, you notice that all them were IDMs. And yet, one year later, we have our first fabless company enter the top ten - Qualcomm - and, in fact, half of the top ten have expressed an intent to become either fabless or 'fab-lite'."
Sreedhar Natarajan, vice president of TSMC's Design Technology Canada, also weighed in - although his opinion might be considered biased because of his association with the self-described "world's largest dedicated semiconductor foundry." He said that most IDMs are moving away from major foundries and going fabless or "fab-lite" because of the growing complexity and expense of fab development with each process shrink.
When asked if IDM Intel would have any real competition come 2025, Rhimes responded: "You're assuming that vertical integration will continue to be a requirement to have that degree of market penetration and differentiation. That's not clear. Certainly the successful stars rising in the last 10 to 20 years have not, in fact, been vertically integrated."
But after giving a nod to IDMs such as Intel, Toshiba, and STMicroelectronics, Charlie Sodini of MIT said that merely thinking in terms of component development misses the point. In his vision of the future, the focus should be on the entire product value chain, from product definition to product design, to manufacturing and testing, to marketing and distribution, to service.
From Sodini's point of view, it makes little difference whether a company handles all those elements of product design, production, and release. There's a broad array of companies working in all those areas, he said. "Companies really get to choose which functions they want to work on. Some companies work on all of them - they're both an IDM company and a product company."
But that's not necessary for success: "Other companies pick things like product definition, marketing and distribution, and service. Apple has been an amazing company in this - its product definition, as we know, has been absolutely terrific. But another brainchild has been on the service side - the iTunes side coupled with the product."
Sodini thinks the vertical-versus-horizontal debate is moot. "The way I see it in 2025...it's not whether you just do an IDM or whether you're vertically or horizontally integrated, [it's] how do you pick which functions that you can do, which functions are built into the DNA of your company that you can do, and they don't necessarily have to be constrained to the component level." ®
A questioner asked the panel why an engineer would ever recommend to his children that they follow in his footsteps, seeing as how when adjusted for inflation the questioner's income hadn't increased since 1973. The panel's moderater, Tyfone CTO Siva Narendra, responded: "Moore's Law doesn't apply to salaries."
Sponsored: Hyper-scale data management