Spansion, the flash memory division of AMD, on Monday plans to announce a new generation of its MirrorBit flash memory chips for cellphones and networking equipment. These new chips, the company says, will help expand its new and unusual lead in the sector over old rival Intel.
When AMD took over Spansion, which started life as a joint venture between the chipmaker and Fujitsu, AMD vaulted past Intel in terms of market share, according to some estimates.
The company has continued to grow by taking customers away from Intel, in part by keeping costs lower, said Bertrand Cambou, chief executive of Spansion.
In the NOR segment of the flash market, AMD accounted for 24 percent of production last year, compared with 23 percent for Intel, according to iSuppli. In 2004, the research company estimates that AMD will be at 28 percent and Intel at 17 percent. AMD, by its own estimates, says it has passed Intel in market share for flash memory in cellphones, 42 percent to 25 percent.
"We are way ahead of Intel on cost per bit" of memory, Cambou said; "25 percent -- that is still too much."
Intel heartily disputes the notion that it has fallen to No. 2 in cellular memory, but it acknowledges that it needs to do better.
"Intel has been very public as to our flash memory missteps last year," a representative said. "We have addressed the problems, and at this point, we'll let our business do the talking."
Additionally, the cost and performance of the new MirrorBit chips should help AMD land design wins that would normally go to NAND flash makers like Samsung and Toshiba.
Flash comes in two varieties. NOR (which stands for "n or", a reference to how data is stored) is more reliable and is used to store applications and operating systems. NAND ("n and") is slightly more prone to data errors, but it's denser and less expensive. It gets used in memory cards to store songs and digital cameras.
Existing MirrorBit chips can hold two bits of data per cell and generally contain between 16 and 128 megabits of memory.
The new chips will still contain two bits per cell but will hold up to 256 megabits and contain a number of performance enhancements. In all, the changes will make them dense enough and cheap enough to compete against NAND chips, Cambou said.
In a few years, AMD expects to advance the technology further by coming out with a MirrorBit chip that contains four bits of information per cell, which will again double the density of the memory.
Other manufacturers have expressed doubt that memory with four bits per cell is possible, but Cambou said MirrorBit's architecture will make it possible, after AMD shifts to the 65-nanometre manufacturing process. AMD is slated to start making processors on the 65-nanometre process in 2007, but work on flash typically lags behind such shifts.
"With four bits per cell, we will bury NAND," Cambou said.
The potential to upstage NAND memory with its MirrorBit flash presents a big opportunity for AMD, said Jim Handy, an analyst at Semico Research. The grand total of NAND, measured in megabits shipped, grew by more than 200 percent last year. Japanese cell phones began to incorporate NAND cards, while the worldwide shipment of Universal Serial Bus flash drives, which also use that technology, grew from 5 million in 2002 to 25 million in 2003.
"Between the two, they required 50 million more parts than people expected," Handy said.
But NOR isn't a bad business, either, thanks to stable pricing and to shipment increases that, in some cases, have even led to shortages. Meanwhile, an excess of manufacturing capacity means that NAND prices have been cut in half since January, a decline that has neutralised the potential revenue gains that should have occurred because of increasing shipments.
PDAs will likely be the most visible NAND-NOR battleground this year, Handy said, since some designers of personal digital assistants are looking at replacing their current NOR chips with NAND or with newer, cheaper types of NOR.
Intel also has plans to cross into the NAND markets, company executives there have said.
Spansion turnaround
Cambou, who came to AMD from Motorola at the request of Hector Ruiz, AMD's chief executive and a former Motorola executive, described the surge in the chipmaker's flash fortunes as a matter of execution. When AMD and Fujitsu ran the operation as a joint venture, the group was "dysfunctional," he said, with teams from the two companies undercutting each other's sales efforts.
By combining the two groups into one corporation, AMD eliminated several redundant positions but at the same time expanded capacity, a process that is continuing. "Between now and next year, we will double capacity," he said.
AMD also managed to hire more outsiders, including Sylvia Summers from Cisco Systems, to help run Spansion. Flash now accounts for more AMD revenue per quarter than do microprocessors.
The reorganised Spansion first targeted wireless handset makers. Intel had long enjoyed a lead in the market, and several of its long-term agreements were winding up in 2002, Cambou said.
Intel also announced plans to raise prices by up to 40 percent on some chips, a move that Cambou says was intended to encourage Intel customers into signing long-term contracts. Intel denies this.
Whatever the reason, AMD, Samsung and others began to land contracts with Intel flash customers in 2003.
Battles over flash are largely an overseas phenomenon. Japan accounts for around one-third of AMD's flash sales. South Korea, China and Europe are the group's other three target markets. Only about 5 percent of AMD's product gets shipped to US manufacturers.
