August 13, 2010 – DRAM memory "turbulence" might rear its head in 2H10 with supplies falling short of demand, because of manufacturers’ problems in migration to newer processes and unable to obtain key pieces of leading-edge equipment, according to iSuppli.
Memory shipments could reach 16M 1Gbit equivalent units this year, up nearly 50% from the prior year, with most of that expansion coming in the final two quarters of the year (11% in both 3Q and 4Q), says Mike Howard, senior analyst for DRAM at iSuppli, in a research report.
|DRAM bit shipment forecast 2009-2010, in millions of 1Gbit-equivalent units. (Source: iSuppli)|
This concentrated growth over just half a year will put a strain on suppliers, who are fighting two issues that could keep them from keeping up with that demand:
– Tool availability bottlenecks. Lithography tool supplier ASML has been well-understood to be swamped with orders — its backlog for immersion tools had been stretched to nearly a year (now it’s reduced to about six months), and even KrF tools are soaring. (Even EUV, which is probably more than a year away, will see a handful of tools delivered in the current year.) Nevertheless, "while ASML appears capable this year of delivering an additional 33 immersion scanners, it will not be enough to resolve the bottleneck," Howard writes.
– Sub-50nm yields. This could be an even bigger problem (though related since immersion is needed at these nodes). The big DRAM players (Samsung, Hynix, Micron) have shifted to immersion litho thanks in part to lessons learned on the NAND flash side. But others who may not have those resources and are just now making the transition will find the road bumpy. Elpida, for example, was expected to shift from its 6Xnm process to 45nm in the second quarter of this year. Any hiccup from that big step could significantly impact not only Elpida but also Rexchip, its JV with Taiwanese partner Powerchip Semiconductor. "Such dislocation could have far-reaching repercussions," Howard warns — possibly denting overall bit growth by as much as four percentage points, i.e. down to 45% annual growth instead of 49%+. (And that’s for an industry already straining to meet demand.)
And losing more bit growth would probably mean increasing prices for DRAMs — which is an advantage for those aforementioned firms who’ve already made the transition over to sub-50nm and immersion litho, as rising DRAM prices will swell their coffers and widen their market leads.