by James Montgomery, News Editor
May 15, 2007 – The new crop of tomorrow’s DRAM and NAND flash memory fabs will be twice as big in 2009 as they were just five years earlier, and the costs to equip these behemoths will simply squeeze most chip companies out of the business unless they can hook into one of the growing number of industry alliances, according to a new report from Strategic Marketing Associates.
Average capacity of a 300mm DRAM or NAND flash fab was about 40,000 wafers/month in 2004, increasing by 50% in 2006 to 60,000 wafers/month, and will reach 80,000 wafers/month by 2009, notes analyst firm Strategic Marketing Associates in its Quarterly Fab Report. Fabs beginning production in 2007 will have a value of about $31 billion, with more than half of them being DRAM and NAND, with that percentage being even higher next year. More than $100 billion in DRAM and NAND flash fabs will come online in the next three years, notes SMA president George Burns. “This growth would not be possible without these alliances.”
With ever-bigger fabs comes ever-higher investments to get them up and running, and for all but the most deep-pocketed and determined chipmakers (notably Samsung and Promos), that means bringing along a partner to share costs and risks. The poster child for monster memory fab investments is the “Flash Alliance” JV between Toshiba-SanDisk, which has already ramped its Fab 3 to ~150,000 wafers/month, and is poised to spend $10 billion to fully equip their new 300mm Fab 4, with capacity of 210,000 wafers/month. That puts the megafab, in terms of national fab capacity, in eighth place behind France and ahead of Ireland, quipped SMA president George Burns, in a statement.
With the exception of the Flash Alliance and the partnership involving Qimonda/Inotera/Nanya/Winbond, most of the fab alliances are fairly new, Burns notes. The IM Flash alliance joint venture with Intel and Micron was formed at the end of 2005, Rexchip was formed last year and the Hynix/SanDisk alliance this year. This speaks to the economics and risk-aversion of the new age of memory production.
Such enormous investments in memory capacity, coming at a time when prices are in a horrible slump (well under $2 for DRAM, with customers asking for ~$1.50 prices for certain components, according to a recent Gartner report), summons a Catch-22 problem. No memory chipmaker wants to blink first and give up marketshare when they believe in rabid demand from various markets 1-2 years out, when these fabs will be up and running, but these firms also need the money from profits to be able to fund these projects until they come online. On the other hand, falling prices make these memory devices more attractive to new applications, which boosts demand and spurs the need for even more capacity.
“It’s a very delicate balance,” Burns explained in a recent interview with WaferNEWS. Firms have to believe that there will be surging demand for memory e.g., replacing hard disks with flash, and that these new markets will open up faster if prices make the devices attractive enough. To absorb this hit, most companies are hedging their bet by forming alliances, helping to share the burdens not only of the fab costs but also development costs. One exception is Samsung, which is big enough to assume that risk on its own (~80% of Samsung’s spending is on memory). Another memory firm that is still going it alone: Promos, which does some joint technology work with Hynix, but is building its own fabs, and Burns thinks the company’s solo fortitude is something to watch.
Meanwhile, Toshiba and SanDisk “have the flash religion” through their mega-JV, believing demand from new applications will justify their huge investments. Plans for Fab 5 are already nearing the construction phase, with eventual capacity roughly equal to that of Fab 4 (~210,000 wafers/month), which would put the JV’s total capacity with the three fabs easily exceeding a jaw-dropping half a billion wafers/month.
Indeed, the Flash Alliance duo is expected to roughly triple their JV’s collective capacity through 2010 (see chart). Promos, too, is eyeing 3x capacity increases over the next four years, while two other groups will more than double their capacity (Qimonda/Inotera/Nanya/Winbond, and Powerchip/Elpida’s JV, dubbed Rexchip). In fact, the only firm expected to increase memory capacity by less than 50% by 2010 is the Intel/Micron JV, IM Flash — which had the third highest capacity of any memory firms in 2006, but will slip to 6th place by 2010.
Could the same shakeout that drove Japanese firms out of the DRAM market in the late 1990s, happen again in NAND flash — and again at the hands of Samsung? “If there’s an extended period of oversupply and no profitability, that could indeed drive some companies out again,” Burns noted, saying that Hitachi, NEC, and others recall those painful DRAM days, and will eventually “grow tired of those losses.” Another result of a protracted oversupply/unprofitable market slump is a memory industry shakeout, but Burns doesn’t see that as a possibility until maybe 2009, after maybe a year or more of small or no profits.
Ultimately, those who take on the risks of new 300mm memory fab investments need to be able to weather the storms of overcapacity and severe price & profit fluctuations. Samsung has a big wallet, as does Nanya living under a conglomerate (Formosa Plastics). Qimonda is somewhat insulated by shareholders, as is Winbond, and both of these companies share some risk through their JV Inotera. — J.M.