No room for Second Place
Other than the titanic struggle between Intel and Advanced Micro Devices, few semiconductor industry battles have generated more interest, excitement and perhaps sheer entertainment than the conflict between programmable-logic device (PLD) market leaders Xilinx and Altera.
Both contestants are a fight fan's fantasy: evenly matched, drawn from similar backgrounds and supremely confident. To make things even more interesting, there's also a bad-blood element at play. "It's a Hatfields-and-McCoys kind of thing," says Tom Halfhill, a senior semiconductor analyst at technology industry research firm In-Stat, a division of ELECTRONIC BUSINESS' parent company.
Nearly seven years ago, when EB last visited the battle between Xilinx and Altera, the issues included the first stabs at system-on-a-chip (SoC) PLDs and the novelty of fabless production. Much has changed since then, with SoC PLDs and fabless production now both mainstream concepts.
There has also been a change at the top for Altera. Rodney Smith, the company's austere chairman, president and CEO in 1999, is long gone. He was replaced by John Daane, a former design engineer, who joined the company from LSI Logic in November 2000. Meanwhile, Willem (Wim) Roelandts remains firmly in control of Xilinx, where he has held the posts of president and CEO since 1996 and board chairman since 2003. Roelandts joined Xilinx after a 30-year management career at Hewlett-Packard.
Historically, Xilinx and Altera have mirrored each other's activities and fortunes. Both got into complex PLDs (CPLDs) and field-programmable gate arrays (FPGAs) at about the same time. The companies also benefited greatly from the early 1990s PLD market boom. Yet, as the Xilinx/Altera bout continues through its third decade (Altera was founded in 1983, Xilinx a year later), the PLD market itself is evolving and moving away from its glue-logic roots. FPGA devices, for instance, are becoming more sophisticated and finding homes in more types of products, including consumer wireless products and multimedia systems.
According to statistics compiled by semiconductor industry research firm iSuppli, PLD revenue from consumer products is projected to climb from $354 million in 2005 to $477 million in 2010. Over the same time span, PLD revenue from wireless communication, also a prodigious DSP user, will rise from $260 million to $396 million. "There's a digital convergence in the high-tech industry," says Roelandts. "A few years ago, your cell phone was used just for voice communication; now you can take pictures and video and listen to music with it." He notes that product complexity and the rapid pace of change in many customer markets are making it increasingly possible to do hard designs. Daane agrees, observing, "This is the era of programmable logic."
As Xilinx and Altera both adjust their strategies to accommodate customers producing complex, fast-paced, consumer-driven systems, the companies also continue to compete against each other. Yet, for the first time, they are keeping an eye open for upstarts that could suddenly turn the market into a three-way or even wider race.
Dividing the pie
From a business standpoint, the PLD market is in the same general state that has existed for nearly as long as anyone can remember. In other words, the market consists of Xilinx, Altera and everybody else. "There are a few other small players, but they're so far down the list that almost nobody ever hears of them," says Halfhill. In fact, Xilinx and Altera accounted for a combined 83.4 percent of the PLD market in 2005, according to iSuppli. In 1999, when EB last looked at the companies, they almost equally divided a 63 percent share of the PLD market.
Back during the tech boom years, Altera topped the PLD market. Today Xilinx is the PLD market leader with 2005 revenue of $1.6 billion and a 50.3 percent market share. Second-place Altera had 2005 revenues of $1.1 billion, which allowed it to capture 33.1 percent of the PLD market. Altera's only possible blemish at press time was the threat of NASDAQ delisting its stock for a delayed report relating to questionable stock options practices.
The two companies' closest competitor, Lattice Semiconductor, posted 2005 revenue of $226 million, representing a mere 6.4 percent PLD market share.
"Lattice's first-generation FPGAs were frankly not on the cutting edge of FPGA technology," admits Stan Kopec, Lattice's corporate vice president of marketing. "We have several new developments that we think will change this situation." The PLD maker announced a partnership with Fujitsu in 2004 to develop new FPGA technology.
For much of the company's current fiscal year, Xilinx's growth has been flat. In fiscal 2006's first and second quarters, the company reported revenue of $405.4 and $398.9 million, respectively, each period, representing a 1 percent decline from the preceding year's corresponding quarter. Xilinx blamed most of the weakness on slow sales to large communications and storage companies that have manufacturing operations in Asia Pacific.
Revenue rebounded in the third fiscal quarter, hitting $449.6 million, a 27 increase over the previous year. A company statement attributed the revenue bump to "sales [that] were better than expected as a result of broad based product and end market strength." In guidance released in early March, Xilinx predicted that fourth-quarter revenues are expected to be up 1 percent to 5 percent sequentially. Altera, which follows a calendar-year reporting system, reported in early March that it expected 4 percent to 7 percent sequential growth in its first quarter.
In recent years, the Xilinx/Altera rivalry has been fueled by leapfrogging technologies, particularly in FPGAs. Xilinx socked Altera hard when it introduced its Virtex line, in 1998. The high-performance Virtex family helped the company extend its FPGA technology to a wider range of customers and gain the upper hand over Altera.
"When Xilinx introduced its Virtex product, it was thought that it had taken the technology lead, and that's one of the reasons it gained market share," says Jordan Selburn, iSuppli's core silicon principal analyst. "Then there was the thought that Altera, when it introduced its Stratix product, had a bit of an edge over Xilinx, and that's why it picked up a little bit." In May, Xilinx debuted its next generation Virtex 5 line, which promises enhanced performance at a lower price and may help the company tip the scale back in its direction.
Bryan Lewis, chief PLD analyst for Gartner Dataquest, a technology industry research firm, agrees that Altera's latest products, including its next-generation Stratix II devices, are boosting the company's revenue stream. "Xilinx really had the best products for a while, but Altera has gone after those, redesigned its products, really talked to the customers and come up with some pretty competitive products."
Still, over the past year or so, the Xilinx/Altera relationship has remained relatively static, despite both companies' revenue gains. Between 2004 and 2005, Xilinx saw revenue grow by 3.7 percent, whereas Altera's leaped by 7.6 percent. "Xilinx's market share was flat," says Selburn. "Altera did gain, but it didn't really seem to come at the expense of Xilinx." The PLD market grew 3.6 percent in 2005, hitting $3.27 billion.
Seesawing cutting-edge technologies have allowed Xilinx and Altera to hold on to their leadership positions over the years. But as the companies now plunge headfirst into DSP technology—a move designed to boost revenues by targeting consumer product manufacturers—they're entering an arena with several already well-established players and potentially exposing themselves to competition from small, pesky FPGA rivals working on ways to dislodge the Xilinx/Altera duopoly.
DSPs are used in a wide range of products, including MP3 players, cell phones and home entertainment systems. Key to Xilinx's and Altera's move into the field is that an FPGA can manipulate logic at the gate level. By simultaneously performing all of an algorithm's subfunctions, an FPGA can outperform a dedicated DSP by as much as 1000:1. With DSP technology critical to their future success, both companies are aggressively pitching an array of DSP-oriented FPGAs, cores, development kits and other products and services. "Moving into DSP; microprocessors; and analog blocks, such as transceivers, is a natural evolution for programmable logic," says Daane.
Halfhill sees the companies filling a market niche that has been left unaddressed by mainline DSP vendors. "By providing DSP functionality in FPGAs, Altera and Xilinx are competing to some degree with DSP vendors such as Texas Instruments, Analog Devices and Freescale Semiconductor, but they're not competing directly with those companies," says Halfhill. "They don't offer parallel products that directly replace DSPs in every instance."
Although none of the minor FPGA vendors are currently prepared to challenge Xilinx or Altera in the DSP arena, this situation is unlikely to last forever, given the market's propensity for rapid change. Halfhill points to Actel as a company that could cause trouble for the market leaders.
Actel was the No. 4 PLD maker in 2005 with a relatively minuscule $166 million in revenue and a paltry 5.5 percent market share. Although the company lacks sales and market presence, it has a potentially important innovation up its sleeve that its competitors lack. "It has a kind of FPGA technology that's a little bit different from Xilinx's and Altera's," says Halfhill. "One thing it can do is encrypt the logic that's programmed into its chips, and it can do this without really impairing the performance much."
This seemingly innocuous ability could pose a threat to both Xilinx and Altera on the soft-core front. A soft core is a block of digital logic that's designed to be inserted into an FPGA. The market leaders currently license their soft cores to customers for a nominal fee, with the expectation that they will then use them in their FPGAs. "The Xilinx soft core has to work in a Xilinx FPGA, and the Altera soft processor core has to work in an Altera FPGA," notes Halfhill.
Actel has used its technology to work out a deal with ARM Holdings, a Cambridge, U.K.-based company that provides low-power microprocessors to the makers of mobile phones, portable MP3 players and other leading-edge products that Xilinx and Altera want to target with their FPGAs. Under the arrangement, customers can purchase Actel FPGAs and get an ARM processor core without having to go to ARM and negotiate a separate licensing deal. In other words, Actel becomes something like a reseller of these cores for ARM," notes Halfhill. "In effect, when Actel sells you the FPGA, the cost of that ARM license is built into the chip—that's a potentially powerful combination."
"Partnering with an established player such as ARM allows Actel to offer its customers the benefits of the rich ecosystem of tools, support and experience that has been built up around the ARM architecture, simplifying the user's design task and increasing the likelihood of product success," says Jake Chuang, Actel's senior director of marketing for application solutions. The encryption technology was vital to securing ARM's cooperation, says Chuang. "ARM is confident that its valuable IP will not be tampered with, reverse-engineered or copied from an Actel FPGA."
To strengthen their market positions, Xilinx and Altera have created multiple manufacturing and technology partnerships. As fabless companies, Xilinx and Altera both depend on external foundries to manufacture their chips. Xilinx has historically used United Microelectronics Corp. (UMC) as its foundry but now, after a now-discontinued partnership with IBM, has begun sending some production to Toshiba. "The reason it's working with Toshiba is that Toshiba has a lot of internal consumption and is a big semiconductor player," says Gartner's Lewis. "Xilinx hasn't given up on UMC, but I think it's hedging its bets with Toshiba."
Altera, meanwhile, relies on just a single foundry partner: Taiwan Semiconductor Manufacturing Co. Daane says the single-partner approach keeps things simple. "Because it works only with us in the programmable field and we're working only with it, there's no concern about sharing intellectual property," says Daane.
On the R&D front, the companies are building a web of partnerships that are designed to help them find fresh FPGA applications and leverage existing technologies. Both companies have teamed with wireless, software networking and DSP partners and specialists in various other areas of interest. "We have literally hundreds of partners," says Roelandts. "FPGAs are becoming more of a system," he adds. "It almost impossible for a single company to build all the components of a system, so you have to build alliances and relationships." Roelandts notes that his company's move into the DSP arena has proved particularly challenging, requiring plenty of outside expertise. "DSP is a new market with a very different customer base," he notes. "You like to align yourself with people who can help you."
Daane notes that partnerships help Altera focus on its core competencies. "We are a company that focuses on doing a few things, so we that can do those things well, and then we partner with other companies and leverage their expertise," says Daane. "We try to partner with the industry leaders."
For his part, Daane sees potential in a new generation of power-thrifty devices. "Altera has been working on several novel 65-nanometer solutions to significantly reduce power consumption," he says. "At 65 nanometers, we'll be able to deliver higher performance and higher density at or below the power consumption seen in our 90-nanometer devices."
Roelandts claims that his company intends to push FPGA density to its limits. "Our goal is to provide the industry's highest density," he says. "Designers don't use PLDs just for glue logic anymore, and we want to provide the best of both worlds: the high functionality of ASICs with the flexibility of programmable logic."
Daane is confident that by embracing new technologies and adding new customers, Altera is on the path to gaining higher revenues and greater market share. "The formula has absolutely worked for the last several years, and we expect this formula to continue to work for us in the future," he says. "Over the past three years, we have been the fastest-growing PLD company and the fastest-growing FPGA company; we've had the highest return on investment for R&D dollars spent; and if you look at our market share in 90 nanometer and 130 nanometer, we have more than 50 percent market share."
Roelandts counters with his own statistics. "We have gained 20 points of market share in the last seven or eight years," he says. "In 1998, Altera was bigger than Xilinx, and today we are 60 percent larger than it is, based on the last quarterly results."
Xilinx's chief adds that he's certain his company will beat back any Altera challenge, by once again leaping over its archrival with innovative technologies. Roelandts notes that although Altera may occasionally develop innovative products, the company doesn't have the same commitment to R&D as Xilinx. "We really are the innovator in the industry," says Roelandts. "From all that I can see, Altera's strategies are different from ours in that respect."
And the bout goes on.
Can they co-exist or will one win out? Send your thoughts to firstname.lastname@example.org.
John Edwards is a freelance business-technology writer based in Gilbert, Ariz.