Can you spell Disaggregation?
The new industry model, mostly driven by economics and time-to-market issues, has semiconductor companies turning to others for product design, testing and packaging
By Karen D. Schwartz -- EDN, February 1, 2003
|
No one who worked at Conexant Systems Inc. three years ago would recognize it now. At that time, the Newport Beach, CA-based company was a semiconductor conglomerate that kept tight control on the design and manufacturing elements of its many disparate businesses.
When Conexant was spun off from what was formerly Rockwell Semiconductor Systems, Milwaukee, in 1999, it inherited a fab that produced specialty wafers for the wireless communications and optical networking industries. The company also inherited several marginally related businesses—everything from solutions for mobile communications and Internet-based networking to those focusing on broadband access, digital imaging and global positioning systems (GPS).
After trying to create synergy among its various businesses for two years, Conexant executives gave up. "We tried to get to the scale of a big IC conglomerate, but we couldn't justify keeping them together," says Matt Rhodes, president of Conexant's Broadband Carrier segment.
Instead, Conexant created a series of separate companies, each of which focuses on one of its business areas. The parent company retained a limited number of internal functions that it now considers its core mission. All told, Conexant's changes created a family of pure-play communications semiconductor businesses.
The dismemberment of Rockwell's semiconductor unit reflects a number of tough truths about the industry today. Most semiconductor companies can't afford the expensive fabs and staff necessary to develop 300-mm wafers and system-level IC designs, especially for the 90-nanometer (nm) and eventually the 65-nm process nodes. Add those pressures to a down economy and increased time-to-market requirements, and the result is the disaggregation of almost the entire industry. The famous quip two decades ago by former Advanced Micro Devices Inc. (AMD) CEO Jerry Sanders III about fabless companies, "Real men have fabs," can finally be laid to rest: Last year, Sunnyvale, CA-based AMD hooked up with Taiwan's United Microelectronics Corp. (UMC) foundry for future chips.
Shifting norms
The two-decade industry evolution from fab to fabless has reached its logical end point: Just about every function of a chip company is being outsourced or accomplished via a joint venture. Many experts say that, over time, nearly all electronics companies—with the possible exception of Intel Corp. of Santa, Clara, CA—not only will be fabless but will rely on others for product design, testing, packaging and other functions. Already, many cell phone OEMs rely on designs developed by others (ELECTRONIC BUSINESS, "Who knows who is inside," December 2002).
Although there are many reasons for the extent of disaggregation, economics and time-to-market issues are the main drivers, says Jerry Worchel, senior analyst in the Semiconductor Research division of In-Stat/MDR, Scottsdale, AZ (In-Stat is a division of ELECTRONIC BUSINESS' parent company, Reed Business Information). The soaring cost of fabs alone is helping fuel the trend, he says. A typical fab today costs about $3 billion, an amount few companies can afford to spend. An analysis by investment bank The Goldman Sachs Group Inc., New York, estimates that a device maker needs $7 billion in sales to justify a new fab. (See The Goldman Sachs Group chart, "Most ODMS Fall Short," below.)
But the cost of fabs isn't the only expense. Few organizations can afford to keep those fabs up and running around the clock. "There was a time when you could bring a fab up to 15% to 20% of capacity and hope that over the next year you could bring it up to 60% to 70% capacity," Worchel says. "But today, if you can't bring that $3-billion fab up to 70% to 80% of capacity immediately, you don't turn the power on."
The sluggish economy also has accelerated the move toward disaggregation and fabless manufacturing, forcing companies to eliminate redundancies and scrutinize every expenditure from both a technology and an economic standpoint.
"Even in the depths of the worst recession, there are meetings going on today where manufacturing managers are telling their CEOs that even though they have excess capacity at 0.13 and 0.18 micron, they need $2 billion of capital to compete in the 90- and 65-nm markets," says Bob Bailey, CEO of PMC-Sierra Inc., a Burnaby, BC-based fabless company. More often than not, he says, the CEO ends up contracting with a foundry that can handle the company's anticipated requirement instead of agreeing to the expenditure. From its launch in 1984, Bailey's firm was one of the first to rely on contracted fabs.
The growing complexity of technology, especially at 0.13 micron, 90 nm and 65 nm, also has forced semiconductor companies to reconsider the deployment of increasingly scarce resources.
"As you go to 0.13 micron, you have to cram more and more functionality into a smaller space at higher and higher speeds. There are packaging and test issues at those levels that never cropped up before," explains Chuck Byers, director of Worldwide Brand Management at Taiwan-based Taiwan Semiconductor Manufacturing Co. Ltd. (TSMC), the industry's largest foundry company.
These factors have led to disaggregation of another kind—outsourcing to highly educated design, packaging and test talent from low-cost regions throughout Asia. Oftentimes, engineers there are available at $30,000 or less—a fraction of the cost of similar professionals in North America.
The extent of outsourcing opportunities today offers top executives many tools to improve their companies' profitability. A company employing this new business model might have a cost of goods sold (COGS) that is 20% higher than its competitors still using the traditional business model, says Jack Harding, president of eSilicon Corp., a fabless provider of ASICs in Sunnyvale, CA. However, that same company will enjoy a higher net profit because it has fewer drains on gross profit.
"The economy is forcing CEOs to contract with foundries that can handle a company's anticipated requirement instead of investing more capital into their own manufacturing."
—Bob Bailey, CEO, PMC-Sierra Inc.
Companies employing the new business model "don't have a 15-cent-on-the-dollar R&D expense because [they are] leveraging the R&D of the global supply chain," Harding says. Sales, general and administrative expenses, including R&D, may account for 18% of revenues. The result, he explains, is an average 12% pretax profit, compared to an industry standard of 7% or 8% on a normalized multiyear basis.
Go with the flow
While small, cash-poor companies have relied on the fabless model for years, larger, more established semiconductor companies now also have embraced the disaggregation trend in an effort to remain competitive.
Motorola Inc., for instance, made a dramatic shift. The Schaumburg, IL-based company adopted a hybrid model it calls "Asset Lite," where facilities that handle differentiated and leading-edge technology are retained and others that are inefficient, underused or handle older technologies are shut down.
The company gradually has whittled down its wafer fabs and test and assembly sites from 27 to 10, accelerating the rate of shutdowns in concert with the economic downturn. In the last two years alone, Motorola closed seven wafer fabs, leaving a total of eight.
Motorola also plans to outsource as much as 50% of its 0.13–micron and 90-nm manufacturing—a big change from the 5% it outsourced about a decade ago. "We'll be focusing more of our outsourcing at the leading edge, where all the capital investment and risk is associated," explains Alex Pepe, vice president and director of strategy for Motorola's Semiconductor Products Sector group.
To mitigate the risk, Motorola recently struck a deal with Philips Semiconductors, Eindhoven, The Netherlands, and STMicroelectronics NV, Geneva, to build a 300-mm fab in France. "We thought it would be much better from a cost and leveraging standpoint," Pepe explains. "It will help keep us on the Moore's Law curve."
At the same time, there are some technologies Motorola won't outsource at all, like power management chips for cell phones. Those technologies, which the company considers core competencies and its differentiator in the marketplace, will stay internal, Pepe says.
Other large semiconductor companies are more reticent to close even a portion of their fabs or to take significant steps toward disaggregation. Traditionally, these companies—including Intel, Tokyo-based NEC Electronics Corp. and Samsung Group, Seoul, Korea—have been able to hold on to their traditional business models in large part because of a fairly narrow product focus coupled with a large market. This has allowed them to reach the economies of scale and maintain the technologies—and fabs—they need to stay competitive without going outside the company.
"We have the volume that requires multiple fabs per process generation, and that will remain true when we migrate to 300-mm wafers," says Sunlin Chou, a senior vice president in Intel's Technology and Manufacturing group. "Even at the 300-mm level we'll operate multiple fabs, so it's to our advantage to maintain our own fabs." The company does a small amount of fab outsourcing, mostly a continuation of the arrangements made by companies it has acquired, but that small amount of outsourcing isn't likely to grow larger, Chou says.
Many say that although these stalwarts can hold out longer because of deep pockets, the companies eventually will succumb to the disaggregation and fabless model being adopted by other firms. "If you've historically been burdened with large capital investments, you will be much slower to write that capital off and move to a global supply chain; but you'll get there eventually," claims eSilicon's Harding. "There is no choice.
"I've talked to many people who run large semiconductor companies, and they predict that within single-digit years, these major corporations will own two things: their own IP around their core competency and a channel for selling it to the customer," Harding says. "They will outsource everything else."
Karen D. Schwartz is a freelance writer specializing in technology and business issues. She has written articles for CIO, InformationWeek, Mobile Computing & Communications and Business 2.0. She can be reached atkaren.schwartz@bigfoot.com.
SEMICONDUCTOR BUSINESS MODELS
| IDM Model | Fab-Lite Mode | Fabless Model | ||
| Marketing/Sales (B2C) | Marketing/Sales (B2C) | Marketing/Sales (B2C) | ||
| Design/IP Systems | Design/IP Systems | Foundry Partners | Design/IP Systems | Foundry Partners |
| Manufacturing | Manufacturing | Manufacturing | Manufacturing | |
| Companies: IBM, Intel | Companies: Motorola, Infineon, ADI | Companies: Xilinx, Nvidia, more than 600 others | ||
| Pros: Control over their own roadmap | Pros: Have some control over process technology, yet chance to have access to leading-edge technology. | Pros: Ability to focus on value-added, while having wide access to leading-edge technology. Upside during capacity shortages, and no worry with utilization during oversupply. | ||
| Cons: Cost, risk, swings in utilization Prerequisite: Must be $7B+ to support aggressive manufacturing. | Cons: Once decision is made to reduce or stop investment, ability to reverse is difficult. | Cons: Dependence upon limited number of foundry partners. | ||
| Joint Venture Model Companies: AMD, TI, Japanese Companies | IP Model Companies: ARM, MIPs, Rambus | |||
| Source: FSA | ||||
Changing times, changing business models
When Jon Sreekanth, Thomas Riddle and Brian Folsom developed a semiconductor company dedicated to the emerging silicon-on-chip storage networking market in 2000, they knew they couldn't structure it as a traditional, vertically integrated company if they were to succeed financially and technologically.
"We've had to look hard at headcount needs and find ways to reduce cash outlay wherever possible."
—Brendon Howe, vice president of marketing, Trebia Networks Inc.
To make it work, the founders of Acton, MA-based Trebia Networks Inc. knew they had to create a company that could take a product from conception through production with as little cash outlay and personnel expenditure as possible.
This next-generation fabless company was designed with today's realities of economic efficiency and time-to-market pressure in mind. Through two rounds of financing, the founders have raised a total of $40 million to date. Despite a fat war-chest, their first decision was to go fabless. By outsourcing the manufacturing of its chips to TSMC and packaging to Amkor Technology Inc. of Chandler, AZ, Trebia can run more efficiently and focus on what it does best—developing its own designs for IP storage networking applications.
As the company's needs and the realities of today's semiconductor market became clearer, Trebia's executives chose to disaggregate even further, offloading the packaging and testing logistics to eSilicon, an ASIC provider that brokers deals for the design and operation of semiconductor-based products.
"While we chose to own the really important things—the design team, verification team and physical design team—we chose to outsource the packaging engineer we need two weeks per design cycle or the test engineer we need three weeks each year," explains Wayne Koch, vice president of Hardware Engineering at Trebia.
The decision to work with companies like eSilicon was driven largely by the current economy, says Brendon Howe, Trebia's vice president of marketing. "We've had to look hard at headcount needs and find ways to reduce cash outlay wherever possible," he says.
Achieving permanent success involves a willingness to change the business model as needed, Koch concedes. Next up, he predicts, is an eventual shift toward additional software development outsourcing, as more third-party vendors develop software drivers and products.
"We'll end up with more of a traditional chip company balance," Koch predicts, "where a big part of our development organization will become hardware-related."—K.D.S.
