Qimonda makes the quality play

One quality play Qimonda will make is to seize the initiative in the production of fully-buffered DIMMs (FB-DIMMs), which Intel is prescribing for its upcoming Bensley server and Glidewell workstation platforms. It's been called "serial RAM," because FB-DIMM replaces the traditional parallel communications bus with a serialized system that enables point-to-point buffering. By reducing the number of memory "drops" that take place at once, throughput rates can increase dramatically, to a data rate Qimonda clocks at 4.8 GB/s.

It's the bigger guys in the mobile space, says Qimonda VP Bernd Lienhard, who are driving up price stability.

But the problem, as Tom's Hardware Guide discovered last October, is keeping these new modules cool. Intel requires something called an Advanced Memory Buffer (AMB), which controls the critical timing settings between the memory controller on the motherboard and all the memory modules in the system (FB-DIMM allows as many as eight). Timing is critical here not just for performance's sake, but for temperature as well. This is where Qimonda will make its play: As the company's North American division vice president Bernd Lienhard explained, at a "Fab Day" gathering last week at Qimonda's Richmond fabrication facility that included TG Daily, Qimonda's AMB will be capable of driving throughput connections of up to 6 GB/s per pin, using timing specifications it worked out last May with Intel and the JEDEC Solid State Technology Association.

Although neither Qimonda nor Infineon invented the AMB, Qimonda believes there's something to be gained from the perception of being the company that typically produces the advanced buffer - a perception of qualitative distinction, as opposed to being the low price leader. (Think Target as opposed to Wal-Mart.) This perception is designed to pervade the server market, and carry the Qimonda brand to a distinctly defined group of customers.

Qimonda intends to mass-produce this AMB device mainly through its fabrication facility in Dresden not only for its own FB-DIMM modules, Lienhard said, but for other memory producers as well, with the intention of capturing a full 50% market share in the near term. Last May, Qimonda acknowledged iSuppli's estimate of 6.8 million FB-DIMMs to be sold to the OEM server market this year, by way of reporting iSuppli's forecast of 68% share of that market for Qimonda by the end of this year.

Surprisingly, though, sizable market share isn't really Qimonda's key goal. As Qimonda North America President Henry Becker told us last week, "I don't think market share guarantees profitability." Infineon had a surge of DRAM market share once before, Becker said, back in 2003, closing to within two percentage points of then #2 player Micron, according to iSuppli. "Infineon gained market share," Becker said, "but didn't capitalize."

Capitalization, the way Becker sees it, means for Qimonda to use market share as an elevated platform to make its permanent value proposition, to specific groups of customers. This is what the company means when it says it's targeting applications of technology; German semiconductor analysts in recent weeks have mistaken this to mean it's moving away from DRAM and into logic units, such as dedicated ICs for handsets.

Qimonda President Henry Becker describes his as a "creative memory company." "Without creativity," he said, "Moore's Law would have run out already."

In an extraordinarily frank tone which was maintained throughout last Thursday's session, Becker and his colleagues basically acknowledged that Qimonda won't make Infineon's mistakes. One example he cited was Infineon's earlier push toward a broader portfolio that included logic units, following the path its competitors had blazed. As Joanne Itow, managing director of Semico Research, told Becker, generally memory fabrication facilities make the shift toward logic. Why is Qimonda's direction the other way around, she asked? The trend toward logic, Becker responded, typically gives semiconductor producers a way "to do something with their memory factories as they grow older." Micron, for example, had an aging 200 mm production facility that it shifted from memory to logic over time, just to keep it operational. Samsung, meanwhile, has the revenues to maintain a broad portfolio.

But Becker saw a "diverging set of synergies between our logic business and our memory business," he said, making a similar path for Qimonda not quite feasible. There's a significant cost advantage to be realized, he said, by concentrating Qimonda's 300 mm facilities - especially in Richmond - on DRAM. Meanwhile, its 200 mm fabrication unit continues operation within the same complex - in fact, just across the walkway - devoted to a variety of tasks, but staying largely within the framework of memory.

Another example of a mistake Qimonda will avoid, Becker said, is investing too much resources in a technology where it can't make a competitive play. Flash memory is that potential pitfall. "We're not competitive in flash memory," Becker said multiple times during the session. "It's not a significant part of our business yet." The company's revenue in 110 nm flash is fair, but not anything wonderful, and certainly not reliable. And since other "800-pound gorillas" are more than willing to go down that road, wherever it leads, then good luck to them.

Qimonda will work on developing a superior flash memory technology, using 75 nm technology. (That's not a typo; 75 nm is the lower number Qimonda pointed