"When we talk to our customers," says Michael Maia, "what I say is, we're a firmware development house but we also sell semiconductors."

What Maia means is that the value PortalPlayer offers isn't contained primarily in the hardware design of its chips but in the software that the company cooks up to make the best use of the chip's capabilities.

"Firmware" is a fancy term for software that is embedded directly into hardware and generally does not get modified by the end-user. In PortalPlayer's case, the term refers to about a million lines of code, a miniature operating system that carries out all the tasks necessary for a state-of-the-art media player device. Such tasks include audio encoding, digital-rights management, power management, a database for handling thousands of files -- music, photographs and eventually video -- on a tiny hard drive, and hooks for all the other technologies built into the iPod.

Those million lines of code are PortalPlayer's competitive advantage, the intellectual property that makes songs on your iPod sound good and the device easy to use. The code is written by PortalPlayer's developers in Santa Clara and Kirkland, and its fully owned subsidiary in Hyderabad, India. Everything else that can be spun off, is.

PortalPlayer didn't even create the basic structure of its own chip. The first step in the chain is to license a microprocessor chip design from U.K.-based ARM, a company founded in 1990 by Apple and two other firms.

Starting from that ARM template, PortalPlayer's engineers then modify the original layout so that the hardware is optimized for the functions necessary to power a digital media player. This is a complicated, painstaking process. A chip is a piece of silicon with millions of transistors arranged in intricate, interconnecting patterns and layers. As PortalPlayer's engineers revise the original template pattern, the chip must go through a series of verification and testing steps to make sure the modifications work.

But even portions of that task are outsourced. PortalPlayer works with two other Silicon Valley semiconductor design companies, eSilicon and LSI Logic, who help do the nuts and bolts work of translating the design into workable hardware.

Once the final specifications are hammered out and the chip prototype works in the lab, the design is sent to the "wafer fab," a factory that etches discs of silicon with the multilayered patterns.

PortalPlayer relies on two different "foundries," the term for companies that specialize in made-to-order chips. Both are headquartered in Taiwan: the Taiwan Semiconductor Manufacturing Corp. (TSMC) and United Microelectronics Corp. (UMC). Here is where, Maia says, "the huge money" starts getting spent. Manufacturing chips is a process involving hundreds of steps and extraordinarily expensive machinery. (A state-of-the-art fab costs about $3 billion to set up.)

The fabs end up producing thousands of disc-shaped wafers, each of which is imprinted with hundreds of chips. The wafers then go to yet another tier of companies that specialize in testing and packaging. First the wafers have to be chopped up into individual chips, or "die," and then each chip must be tested to make sure it isn't a dud. Finally, it is coated in plastic and made ready for assembly. PortalPlayer uses two different companies for this part of the process, SiliconWare in Taiwan, and Amkor in Korea.

The finished product is shipped to a warehouse in Hong Kong, and then to an assembly plant in Shanghai where it is plugged into a circuit board and inserted into an iPod. And the last step? That iPod is loaded onto a FedEx jet for shipping to your front door.

And that, in a nutshell, is how a state-of-the-art Silicon Valley fabless design house operates.

But what does it mean?

Up until around the mid-'80s, a typical semiconductor company handled every step in-house, from design to final packaging. Such a company was described as "vertically integrated" -- a one-stop shop for semiconductor manufacturing. It was an industry that began in Silicon Valley.

Vertically integrated semiconductor companies still exist. Santa Clara's Intel, which alone accounts for some 15 percent of chip sales in the world, is the leading example. But in the late '80s, a different model -- the fabless design model -- began to emerge and has been growing ever since.

The internal dynamic of the industry forced the change. Over time, the capital investment necessary to operate a fab simply became too great for any but the most well-heeled to afford. And even then it was risky. Owning your own fab, explains Maia, is great when business is good. But the chip industry is intensely cyclical and when business goes sour, "They weigh on you like granite."

But if you didn't own a fab, how were you going to get your chips made? You couldn't exactly go to one of your competitors and ask them for help. For the fabless design niche to grow, it needed another industry niche to appear in concert. Enter the made-to-order foundry business, pioneered by Morris Chang, the founder of Taiwan's TSMC, initially a joint venture between the Taiwanese government and Philips Electronics.

Chang was born in China and worked at Texas Instruments in the U.S. for 20 years. Then he emigrated to Taiwan, where he figured out that if he focused on manufacturing made-to-order semiconductors for outside customers, he would protect himself from exposure to the market failure of any single customer. At the same time, he could insulate the design houses from the huge capital investment necessary for manufacturing. It was a win-win innovation. Both sides have flourished ever since. The fabless design sector of the chip industry now accounts for 16 percent of the overall semiconductor market and is growing every year.

But stripping wafer fabrication from the traditional structure of a chip company was just the first domino to fall in the chip production cycle. If it made business sense to rely on an outside foundry for chip manufacturing, then the obvious next question was what other portions of the supply chain could be shed?

Next on the chopping block: assembly and testing. What was the point of having your chip manufactured in Taiwan, then shipped to Santa Clara for testing and packaging, and then shipped back to Taiwan or Malaysia or China for final assembly into a CD player or cellphone or fancy toaster? It was obviously more efficient to rely on companies that specialized in the task and were located closer to the point of production.

And so it goes. The evolution of chip manufacturing has relentlessly followed a process clunkily dubbed "vertical disaggregation." One step after another in the supply chain gets spun off to those who can do it more cheaply and efficiently, leaving the companies at the top of the value chain increasingly focused on their core competency. One-stop shops have been replaced by what academics call "cross-national production networks."

The rise of vertically disaggregated, cross-national production networks can be seen as a persuasive demonstration of the free market in action. Individual entities, like PortalPlayer, constantly in search of a competitive advantage, dispense with any task that someone else can do more cheaply or efficiently and focus R&D on the particular job that it does best. The wide, immaculately groomed avenues that connect Silicon Valley's business parks are jammed with hundreds of other companies doing exactly the same thing.

But what happens when you've disaggregated everything. If India can do the design and East Asia handles the manufacturing, what's left for Silicon Valley? Lunches with venture capitalists? Meetings with reporters? Or empty office buildings?

Recent Stories