原创 The squabble over 28-nm market share

In the Silicon Valley, there is a decades-old war of words between the two dominant suppliers of programmable logic that can at times make political bickering seem downright collegial.

The latest occasion for disagreement, which bubbled to the surface after each reported earnings in the past two weeks, is over market share at the 28-nm node. Both companies began shipping 28-nm parts earlier this year (though, not surprisingly, there is some nuanced dispute about who was first). Both companies also claim the early market lead at the node.

During its earnings call this week, Altera executives suggested the company has 28-nm FPGA market share of about 60 percent. Xilinx estimates its 28-nm market share at about 70 percent. For those of you scoring at home, that's 130 percent, total. Now there's the way to grow a market.

Xilinx and Altera have been playing the tit-for-tat marketing game for years. The two companies have dominated the programmable logic space since the 1980s and cumulatively hold about 85 percent share of the market (higher than that at the high end of the market). In some respects, it's no wonder that their marketing rhetoric smacks of the bickering and one-upmanship borne by sibling rivalry: for each company, the other is the only real competition.
Unlike Republicans and Democrats, however, there is one thing that Xilinx and Altera do agree on: Both have been claiming for some time that FPGAs are increasingly winning sockets that traditionally went to other types of chips, chiefly ASICs and ASSPs.

Moshe Gavrielov, president and CEO of Xilinx, has long championed the existence of a "programmable imperative" enticing customers to move from ASICs and ASSPs to FPGAs to save on the growing non-re-occurring engineering investment required to build the more traditional chips. During the Xilinx earnings call earlier this month, Gavrielov said 28-nm design win momentum continues at "an unprecedented pace" and that the company continues to see customers moving away from ASICs while Xilinx FPGAs integrate and displace "evermore" ASSPs.

Not to be outdone, John Daane, Altera's chairman, president and CEO, followed that up last week by proclaiming that Altera was seeing "great success" for 28-nm products. "We continue to open larger sections of the ASIC and ASSP markets for PLD displacement," Daane said.

Could these guys have been separated at birth?

Xilinx CEO Moshe Gavrielov (left) and Altera CEO John Daane. Separated at birth?

Gavrielov and Daane have been humming similar tunes for years. Though some holdouts still dispute the claim, mounting evidence—such as increased revenue for the FPGA vendors in recent years—supports the argument. And analysts have bought in, too.

"FPGAs continue to increase the overall addressable market with investments like 28-nm and embedded processor initiatives," said Ian Ing, an analyst at Lazard Capital Markets LLC, in an email exchange. (Though Ing added that communications infrastructure requirements are rapidly evolving and may tip the balance back in favor of ASICs and ASSPs in certain applications).

Why don't the numbers add up?
There is, by the way, some reasonable basis for the incongruity of the two firms' 28-nm market share estimates. According to Jon Olson, Xilinx' chief financial officer, Xilinx believes it has been winning at least 70 percent of available sockets for 28-nm programmable platforms. But, according to Olson, Xilinx believes it has a much broader portfolio of products at 28-nm than its smaller competitor, including the Xilinx' Zynq FPGAs, which integrate an ARM Cortex-A9, and the Virtex-7 H580T 3-D heterogeneous FPGA.

Altera is "defining a universe that is much smaller than our universe," Olson said in an interview. "We have two vectors to our portfolio that they don't even play in."

For its part, Altera believes it is taking more than two-thirds of the market at 28-nm. But Daane acknowledges that the figure he is using is not a straight number of sockets, but the dollar value of the sockets—based on the number of parts customers say they will buy—a number that is more vunerable to revision. "If you look at just the number of sockets, you can win a lot of sockets but none of which really matter in terms of value," Daane said.

In other words, both companies are estimating their market share at 28-nm in a subjective way that is highly dependent on what they consider their total available markets, which apparently are not quite the same.

For the record, Xilinx reported about $10 million in 28-nm FPGA revenue for its most recently concluded quarter, while Altera estimated about $8.2 million for the second quarter and more than $16 million cumulatively since the year began. Both companies expected 28-nm to increase to more than $20 million in the current quarter. (Daane said Altera's second quarter revenue would have been higher if not for a tight supply of 28-nm manufacturing capacity).

Delving into more depth on a direct competitor than you will typically hear from a CEO on a financial earnings call, Daane told analysts that Xilinx gets the majority of its 28-nm revenue from prototyping programs that deliver early revenue that does not become a reoccurring stream. By contrast, Altera's 28-nm revenue is from production design chips, nearly all for high-end applications, he said.

Olson strongly denied Daane's claims about the makeup of Xilinx' 28-nm revenue. "It is categorically not true," he said.

Unfortunately (or fortunately, depending on your point of view), unlike the presidential candidates, Xilinx and Altera won't be participating in any internationally televised debates later this year where their differences can be crystallized for the FPGA buying public. Customers will continue voting with their purchase orders. Polling data, for the 28-nm node and elsewhere, may not be reliable.