Programmable Notes

Who's on First? And Why? Year-end financial results usually reveal a surprise or two; 1994 was no different. by Larry Waller

Right on schedule, the first wave of reports on last year's programmable logic financial results compiled by ace market watchers began to percolate during February. As always, circulation of these numbers accelerates to warp speed since they amount to an annual scoreboard eagerly awaited by PLD players and observers alike.

Also, some surprising developments that even this heavily networked business had not anticipated always seem to surface.

And so it goes this time. A new sales champ, Xilinx (San Jose, CA) somewhat unexpectedly has taken over the number-one programmable spot. With $321 million in sales, the CMOS FPGA pacesetter moved decisively ahead of the perennial PLD leader Advanced Micro Devices (AMD, Sunnyvale, CA), which came in with $280 million. Most observers believed that AMD, which still has substantial (though plummeting) bipolar business, would maintain its top slot for another year. The market consultant firm Integrated Circuit Engineering (ICE, Scottsdale, AZ) was the exception, calling it right on the nose with its Status 1995 Outlook estimate published in January.

Rounding out the top tier of PLD firms, defined as those with revenues upwards of $100 million, are Altera (San Jose, CA) at $199 million in 1994 sales, and Lattice Semiconductor (Hillsboro, OR), with $134 million.

Altera grew 42 percent, and has momentum that makes it a cinch to soon be a strong second. These results are part of an industry report being compiled by Pace Technologies (Scottsdale, AZ).

As for PLD market size, ICE estimates 1994 came in at $1.32 billion, up about 11 percent. Pace notes the star products were CPLDs and FPGAs, with sales of $307 million and $466 million, respectively, up 40 percent and 31 percent.

Even the most cursory analysis of this rundown of top PLD firms underlines the big programmable news of 1994, which portends profound changes. These top PLD outfits in the first tier, paced by Xilinx and Altera, are pulling away from the crowd. "We're seeing more stratification [between the tiers]," observes Rhondalee Rohleder, of Pace Technologies.

The reason for the widening separation between the top and second tier (about $25 to $75 million) is clear to her, and others who follow programmables. The firms continuing to attain supercharged growth are those that had high-density products (or devices with 40 pins or more and a multilevel architecture) early on and have pushed them hardest. "If you have these, you've done very well, and the more of them the better," Rohleder notes.

Xilinx and Altera are the models, not only bounding ahead at high double-digit rates, but more and more dominating the high-density programmable device business. In 1994, Xilinx and Altera together bagged about 65 percent of the high-density CMOS market, which accounts for some 70 percent of all CMOS PLDs. Moreover, with a full head of steam and new products coming up, Erik Cleage, marketing VP at Altera, predicts that "this share will go up again this year."

This powerful grip by established leaders on the best part of PLDs spells even more trouble ahead for the trailing members of the pack trying to close the gap. For one thing, many of them are forced to scrap it out against each other for the low-density business, or "jelly beans," as Xilinx CEO Bernie Vonderschmitt calls it. Demand for these simple PLDs already is tapering off faster than expected as designers opt for higher density parts that increasingly offer the performance and price features that enable handfuls of them to be replaced by a single device. Low-density devices made up less than 30 percent of the programmable pie in 1994, and according to several marketing officials, will drop to below 25 percent this year. Those firms that are attempting to keep up with the leaders by slugging it out with their peers in low-margin simple PLDs have, in Rohleder's words, "a tough row to hoe."

Analysis of 1994 results further signals that it was a watershed year for programmables, when automatic annual sales gains no longer could be taken for granted. Some firms, with Texas Instruments (Dallas, TX) standing out because of its overall prominence in semiconductors, actually experienced a revenue decline. (TI announced their exodus from programmables in February.) Others grew somewhat, but still lost market share. "Historically, this is unheard of," says Rohleder, who expects pressures, exerted across the board by keen competition, to further intensify.

Foremost among these pressures is price-cutting by the leaders, implemented to bring programmable tabs down to levels that spur market expansion. Some of the most popular Altera and Xilinx parts, for example, sold for some 40 percent less at year-end. This is a textbook strategy that is proving most effective for the leaders, as reflected by ballooning overall programmable sales. It further confirms Vonderschmitt's observation that "this market is very price-elastic."

What the continuing price-chopping also does, however, is to make it tougher for those of lesser size to match this pace. In particular, AMD suffered from this downward spiral. During 1994 they hit their unit sales target, which projected $100 million in revenues for the high-density MACH devices. The participation in the competitive price-cutting, however, reduced actual MACH sales to $60 million, according to marketing director Chris Henry. This shortfall, combined with the decline in bipolar revenues from $120 million to $90 million, caused AMD to topple from their sales leadership position.

Henry fully anticipates more price erosion, and is prepared to be in the vanguard this time around. He identifies this "big price battleground" in the 32-64 macrocell range upwards from 44-pin density. Prices for such devices as Altera's 732 and AMD's MACH 210 now run $5-$7 each, and he predicts that they will fall to $4-$5. As this testing of each company's process and financial resources moves steadily into even higher densities, when $30 device prices are halved, it will be the smaller competitors who will falter, he says. "The second tier of companies," Henry says, "can't cut it."

Most players think that these 1994 programmable trends appear fairly well established and should stay on track; however, if there is a wild card, it would be products based on the antifuse architecture. Their cost and speed advantages against the SRAMs (along with a superior design methodology) have been recognized all along, but a later start and some other problems have held them back. It didn't help antifuse's reputation, either, when their leading proponent, Actel (Sunnyvale, CA) suffered a production glitch in 1993 that flattened growth for a year. Once a low top-tier firm, Actel only last year got moving again, chalking up $76 million in sales, up from $60 million.

Now, however, all kinds of positive antifuse developments are happening. CMOS kingpin Xilinx is putting finishing touches on its antifuse line, which it plans to be sampling by midyear, "or maybe a month or two later," says Vonderschmitt. Actel marketing vice president Dennis Nye professes to welcome Xilinx's move, as "confirming the benefits of antifuse," and is introducing new high-density generations of his company's own ACT devices.

Also, a revitalized QuickLogic (Santa Clara, Calif.) is again in the game, with 0.6µm products fabbed by Cypress Semiconductor (San Jose, Calif.), which also has an antifuse line. QuickLogic has set its sights on a $20 million year, and expects to be profitable soon, according to CEO Tom Hart. All in all, if antifuse plans click, these products could become a big programmable story this year.

Larry Waller is a contributing editor to Integrated System Design.

To voice an opinion on this or any Integrated System Design article, please e-mail your message to: michael@asic.com.

integrated system design  June 1995