No one is longing for calamity to strike again, but in Asia, memory makers are beginning to say there are only two scenarios that can haul the industry out of the pickle it's in. One is an act of God. The other is bankruptcy. "The 'V' [-shaped recovery] can only happen if there is attrition, if one of the companies goes belly up," said Bert McComas, founder and chief analyst at InQuest Research.

With informed observers saying a turnaround now seems unlikely until 2003, a war of attrition is playing out as South Korean and Japanese memory makers flood the world market with product amid the DRAM industry's worstever slump. "The current spot market is selling parts at unbelievable prices. I think this demand situation is the worst in my 20 years [in the business]," said NEC's Shimakura.

Il Ung Kim, vice president of memory marketing for Samsung, said memory makers will have boosted supply to a reasonable 5.5 billion units of 64-Mbit equivalents through 2001, to last year's 3.67 billion. But as output goes up, sales go down. Kim said Samsung's second-quarter sales amounted to 32.06 million units, down from the first quarter's 32.26 million.

"This is the worst-ever year in terms of bit growth," at below 50 percent, said Richard Gordon, principal analyst for semiconductors and memories at Gartner Dataquest. "The average is 68 percent and some years have been as high as 80 percent. The fact that it's less than 50 percent [now] is dire."

"There just isn't any upside," said Sherry Garber, memory analyst for market research firm Semico Research Corp. Garber predicted a 7 percent growth in PC sales for 2001 and a 35 percent decline for the memory market compared with last year.

"It's not a pretty picture," she said. "Inventories at DRAM makers are at four to six weeks; this time last year they were at two weeks." As for signs of strength amid the gloom, Garber said, "There's little things, like growth in the portable market, but there isn't anything to pull it up."

With things this bad, the technical debates that raged last summer over double-data-rate (DDR) technology vs. Rambus have become less relevant, sources said. RDRAM will achieve limited and persistent penetration of the higher end of the market, representing as much as 10 to 12 percent of DRAMs by year's end, said Sun Chung, a semiconductor analyst for Merrill Lynch & Co.

Intel Corp.'s delay in getting out Pentium 4 chip sets, technical difficulties with DDR and the fact that it cannot be sold at a premium have negated the importance of this memory-standards battle, at least for this year, Chung added.

Samsung's Kim claimed his company hit 55 percent of RDRAM share in the both the first and second quarters, and 50 percent and 35 percent of the DDR market in the same period. But the relatively tiny market share of DDR and Rambus, coupled with increased competition, means neither standard will likely have much impact on vendors' 2001 bottom lines.

"RDRAM does not make a profit for Samsung anymore, due to higher costs coupled with faster price erosion on the 128-Mbit [density]," said Soo Kyoum Kim, senior manager worldwide for memories and semiconductors at IDC Korea. Kim said Rambus prices tumbled from $16 in March to $9 in June. "And Nanya, Samsung and Micron have not found good demand on DDR. This year both DDR and Rambus will take about 7 percent of the total DRAM bit demand, which actually means no significant penetration."

That leaves industry observers on the equivalent of a death watch, with Hynix, formerly a unit of Hyundai, everyone's favorite candidate to go belly-up. Hynix, though, has stubbornly refused to buy into that scenario.

"I expect the companies that were expecting Hynix to exit the market should change their plans," said Farhad Tabrizi, vice president of worldwide marketing for memory products.

Hynix has faced $11.2 billion in debt following its merger with LG and is blamed by rivals for allegedly dumping parts on the spot market to eke out liquidity.

But the company has paid down $6 billion of that debt in two years and has invested more than $1.5 billion in capital expenditures, said Tabrizi. Hynix's recent GDR issue raised about $1.25 billion from 104 million shares, giving it all the liquidity it needs "until the end of 2002," he said.

This spring the Korean Government threw the company a couple of cushions, rescheduling Hynix's loan repayments and extending the deadline for settlement into 2003, said Brian Matas, an analyst at IC Insights Inc. (Scottsdale, Ariz.).

Sources now say that any recovery in the DRAM market will have to wait until the summer of 2003, when a combination of DDR and SDRAM chip sets and modules for the Pentium 4, new processors from Advanced Micro Devices, greater memory-density demand for a tested Windows XP operating system and renewed corporate purchasing should turn the market.

With relief that far out, a number of scenarios are possible. Until recently, 2002 was supposed to represent a new era of competition, as the biggest players battled for dominance with timely ramping of 12-inch fabs on the 0.13-micron process. This, for example, is the cornerstone of Elpida's strategy, with its purpose-build Hiroshima, Japan, fab leading the charge into profitability, said NEC's Shimakura.

Assuming second-tier companies do not exit, or Hynix does not collapse, the 0.13-micron process technology era will be the "critical turning point" for deciding who will survive and thrive in this volatile business, said Shimakura.

But this year's market is making it harder to cough up the $3 billion required for a 12-inch fab, said both IDC's Kim and Merrill Lynch's Chung. Tight finances might force others to follow Mosel Vitelic in pushing out ramps, they said.

The Infineon-Mosel Vitelic ProMOS Technologies fab could open as early as October. Samsung's Fab 12 at Hwasung could start a pilot run in December, said the company's Kim. Micron Technology's Utah fab and Elpida's Hiroshima fabs are scheduled to come online next year and Hynix's fab is to open in 2003.

However, 12-inch fabs are not the Nirvana they seem, said Samsung's Kim, especially with high wafer costs, pricey initial production and economies of scale taking years to work through. "Our 12-inch fab is ready. I could run a pilot if I wanted to. But it costs $3 billion to ramp, so it's not really realistic to add 12-inch to next year's supply growth," he said. In Kim's view, 12-inch production "will not impact until 2003."

The more sensible option, for many, is to push shrinks. Samsung is accelerating its shrink to 0.12 micron this December instead of next spring, Kim said. And Hynix is also following this strategy, Tabrizi said, through a "synergy technology" that Tabrizi claimed added 11 percent more die per wafer while cutting mask steps. "We will have the lowest-cost 0.15-micron memory technology in the world. It's very good for the 256-Mbit density," he said.

Doing 8-inch shrinks instead of 12-inch ramps could save the industry next year, claimed Samsung's Kim. Shrinking provides a maximum bit boost of 40 percent. If memory makers constrain themselves to this low-cost strategy and if the PC market grows next year, the supply-demand relationship could flip-flop, he said.

"The biggest potential comes with the P4 and [Windows] XP between the second and the third quarters. Almost every maker is cutting investments," Kim said. "If the supply bit growth stays at about 40 percent, and the PC market grows more than 10 percent, that will create bit demand of 50 percent. That could be the ignition point."

• Most memory types suffer downturn

Return to 2001 Midyear Forecast