"Real men have fabs." AMD founder Jerry Sanders made the point—in blunt, politically incorrect fashion—that maintaining control over IC manufacturing was crucial for a top-tier semiconductor company. The fabless semiconductor community heartily disagrees. In fact, proponents of the fabless business model often quote Sanders to highlight the folly of not following their strategy. Even Sander's own AMD eventually spun out its manufacturing capability with the "asset light" or "asset smart" strategy that led to the creation of Globalfoundries.
Except for Intel and the memory vendors, semicos who didn't take up the fabless or at least fab-light mantra were ridiculed. Process technology began to be viewed as commoditized and unable to provide a competitive advantage. Continuing down the integrated device manufacturer path was a sure path to bankruptcy. At least that became the conventional wisdom.
Few would argue that Intel is a successful company. Their status as a semiconductor superpower can be attributed to their commitment to taking a longer term view. Intel leadership prefers innovation over cost cutting. They simply don't depend on suppliers or even partners to develop future technology platforms to manufacture their processors. Intel might be one of only a handful of entities that can actually afford to build new fabs, but they face the same temptation to dump chip manufacturing and pocket the savings as everyone else.
A company like Intel that continues to invest the resources required to build next generation fabs in parallel with development of future nodes is obviously taking a different view. The company has been around a while. There have been lots of chances to cut and run. Management had many opportunities to mortgage the company's future by maximizing current profit levels through shortsighted cost cutting. Instead, they chose to reinvest in continued technology development.
But Intel was not alone in shunning the fab-light trend. Many Japanese IDMs have maintained their manufacturing capability. That strategy was covered in the June edition of EE Times Confidential. Junko Yoshida's article, "Surviving the New Semiconductor Cycle," considered the possibility that the Japanese semicos might be well positioned compared to companies that felt safer in the fabless herd. As Yoshida pointed out in her article, the Japanese were "pilloried in the press and the markets for having avoided the tough choices." Despite not building out capacity for the latest nodes, the Japanese IDMs stuck around and maintained the development and manufacturing capability that will allow them to jump back into the game.
Faced with the multi-billion dollar price tag, building a new manufacturing facility is not even a choice for many companies. So what lies ahead for fabless chip companies? In perhaps overly simplistic terms, the landscape appears like this: A relatively small engineering team can assemble blocks of off-the-shelf IP and have their design manufactured by an Asian foundry. Can companies survive as chip vendors working like this? It seems doubtful if they plan to differentiate on hardware alone.
Great points. It's hard to argue that the fabless-foundry model has been anything but an unqualified success, Sanders' quote be damned. But Intel's resistance to taking this path has arguably been one of the biggest factors in its success. (Perhaps it will become an even bigger differentiator now that Intel has signaled its willingness to at least dabble in foundry work).
In the end it seems the key point that this blog entry makes is that the key to success is to have a clear strategy and stay with it. Once an IDM goes fabless or even fab-lite it is committed to that path and there's no getting the genie back in the bottle.
I think the car analogy proves the opposite point. There is a huge range in auto reliability, which suggests it isn't just the components. Clearly the system design, assembly and testing can produce very different outcomes.
The fab lite model can be viewed as investing in comparative advantage. A company with a mix of analog and digital products might shift out of digital fabs, invest more in analog fabs and analog and digital design, and wind up with a stronger market position.
I agree with Hank that the automobile business suggests that the fabless model should work perfectly. I almost left it out of this article. It's there because I couldn't resist the idea that an automobile brand can appeal to consumer emotions -- something that should be much less successful with cynical engineers with managers and accountants always looking over their shoulders.
Small point.....Jerry Sanders ( who looked great at the SIA Dinner earlier this month where he sat next to Craig Barrett )was long gone from AMD when they made 1) the decision to overspend for the acquisition of ATI and 2) the decision to go fabless after a couple years of site-sourcing for what ultimately became the GloFlo fab in Tech Valley, NY. Jetty's other famous quip was that "chips are the crude oil of the 80's"
It makes me wonder: Who will be left to build the next gen fab? Also, what really is the value add of the fabless vendor? It can't be solely IP, it must be the "mix": tools, IP, service, and lets not forget COST. How does a fabless vendor win new business unless it is based on cost? With the reduction in unique fabs and the subsequent sharing of the same fab between multiple fabless vendors what differentiates them, again I suggest that is is cost? In the end a good thing for the consumer but a hard place to be profitable for the fabless ones.
@robotics Developer: you ask who will be left to build the next gen fab...the answer is fairly simple: Intel, Samsung, and TSMC...maybe someone in Japan, Toshiba? maybe someone in Europe/Middle East, GlabalFoundries?...that is pretty much it...Kris
The recent article here at EE Times that reported the cost of EUV litho equipment at $120 million per system tells you all you need to know about who can afford next generation fabs.
The semiconductor industry was way more fun back in the 1980s, when dinosaurs like Jerry Sanders still roamed.
This blog is based on the premise that IDM’s Process Tech is something that is significantly differentiated from that offered by the Foundries. While certain IDM’s may be lucky enough to operate in a niche requiring little process innovation, it should be noted that Jerry Sanders also said, “Niches are for Roaches”.
It should also be pointed out that ~2 years ago TSMC produced an ATI/AMD GPU at 55nm before Intel had moved to 45nm. Of course Intel still preserves its process tech lead in CPU’s (and at a significantly higher capacity), but in the past it only had to stay ahead of the other IDM’s, the Foundries were nowhere to be seen – that is no longer the case. Now the Foundries are competitive with Samsung, Toshiba, Micron … and if you are an IDM Fabbing on process technology that is offered by a Foundry at a more advanced node and you *not* making similar developments, then you better be preparing for Fab shutdowns as soon as one of your competitors does take advantage to the Foundry developed process.
Preserving the old IDM model only benefits one main competitor in each sector … the leader … if you think that only process tech is critical. This is fine if you want a single monopolist dominating each sector. The future will be different where each design will succeed based on its own merit and not on whether or not the designer is also capable of getting the device into volume production. There will be more competition as the barrier to start-ups will be significantly reduced. We will have to live with it and the customers will benefit … and Intel can look back fondly to the days when it could keep its dwarf-like competitors at bay, by means of its long process-tech arms.
The article also ignores Intel’s use of all means at its disposal to crush competitors, so as to insure its market domination. This doesn’t work for every company in the semiconductor industry. So presenting Intel as if they could/should be emulated is fundamentally flawed … they are exceptional. If everybody behaved like Intel, everybody would have their own standards and refuse to recognize others architecture (unless fundamentally altered). Upgrading individual components would be impossible. And no architecture could survive into the next generation.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.