Print  |  Send  |   

DRAM debate: Is boom cycle over?

August 10, 2010 // Mark LaPedus

DRAM debate: Is boom cycle over?

There are mixed signals in the DRAM market right now. Some say the boom cycle is over. Some say that this boom period is just the beginning and product shortages will linger for the foreseeable future.

Page 1 of 1

Others say the boom cycle will continue, but it will be somewhat limited, due to the lack of fab equipment. As reported, ASML Holding NV and Nikon Corp. are unable to keep up with huge demand for their respective 193-nm immersion lithography scanners for advanced DRAM production.

DRAM makers also failed to invest in new capital in recent cycles. As a result of this and other trends, overall DRAM bit growth projected for 2010 could come in from 2 to 4 percentage points lower than expected. This could reduce the projected annualized DRAM growth rate from 49 percent to as low as 45 percent, warned iSuppli Corp. in a new report.

What's in store for DRAMs now and the future?
Some are pessimistic in the short term. ''In fact, two-thirds of current (fab capacity) plans are centered on memories. The problem is dominantly in DRAMs, where as recently as May, everyone thought the consolidation would limit over-expansion. In less than one quarter, we have gone from a window of endless shortages to excess,'' said G. Dan Hutcheson, CEO of VLSI Research Inc., in a recent report.

Others disagreed. "We believe the DRAM cycle has a ways to go due to the enterprise upgrade cycle (these last few years, not quarters), and DRAM cycles are currently muted (more predictable supply growth and average selling price stability) unlike in the 80s and 90s," said Hans Mosesmann, an analyst with Raymond James & Associates.

"DRAM players are seeing 13-15 percent bit demand growth in 3Q '10 and an expectation of similar quarter/quarter growth in 4Q '10. Some OEMs are indicating 100-plus percent bit growth in 2011, which is obviously not an indication of overall bit demand growth but rather an indication that in enterprise-centric applications (servers) the new Intel Nehalem CPU architecture requires significantly more DRAM," Mosesmann said.

"A commodity profoundly susceptible to the variable dynamics of supply and demand, DRAM is expected to ship 15.9 million 1Gbit-equivalent units in 2010, up 48.6 percent from 10.7 million units last year," added Mike Howard, senior analyst for DRAM at iSuppli.

"Most of the years growth is forecasted to occur in the second half of the year, with each of the final two quarters of 2010 expected to post sequential bit growth of approximately 11 percent," Howard said. "In comparison, bit growth in the first two quarters of 2010 topped out at far below the 10 percent mark. Such high levels of growth, concentrated in a six-month period, will strain the production capabilities of DRAM suppliers."

Two major issues
However, two issues potentially might negatively impact second-half DRAM availability, according to iSuppli. As reported, ASML cannot ship enough 193-nm immersion scanners in the market as a whole. Nikon is behind the curve in the arena.

"While ASML appears capable this year of delivering an additional 33 immersion scanners, it will not be enough to resolve the bottleneck," iSuppli believes.

Secondly, without 193-nm immersion scanners, DRAM makers will have trouble scaling beyond 50-nm. Samsung, Hynix and Micron have obtained 193-nm immersion scanners and other tools to move beyond 50-nm.

"However, for resource-constrained companies or for those currently negotiating the transition, difficulties accompanying such a move might reduce their total output, negatively impacting the industry's overall bit growth in the process," according to the firm.

For example, Elpida Memory Inc. is expected to move from 6xnm processes to 45-nm--"a considerable lithographic jump that presents confounding yield problems," according to the firm.

Still, the question is clear: How far can DRAM makers scale their products? The DRAM will scale "as far as people have the money to do it," said Bob Merritt, an analyst at Convergent Semiconductors LLC. "It's becoming more and more expensive. It will become a cost-prohibitive slope."

Like the NAND crowd, the DRAM players hope to extend 193-nm immersion lithography as far as possible. But for next-generation DRAM production, Samsung is pushing hard for extreme ultraviolet (EUV) lithography. EUV tools could cost $100 million or more per unit. Immersion scanners are $50 million or more.

There are some material issues as well. Some DRAM makers moved from hafnium oxide to zirconium for the capacitor dielectric starting at 50-nm.

Then, there is a huge debate taking place in the industry. Some DRAM makers will extend zirconium for the capacitor dielectric, while others are looking at a new material called strontium, said Tod Higinbotham, executive vice president and general manager of microelectronics at ATMI Inc., a materials maker.
DRAM makers may be able to extend zirconium for "two generations," he said. Then, strontium may be required, which could get the DRAM down to the 2xnm-node, he said.

Beyond that, memory makers are looking at several alternatives, including 3-D devices, universal memories, among others. Instead of scaling, chip makers have proposed the idea of developing 3-D devices, based on a stacking technology called through-silicon-vias (TSVs).

However, there are still a lack of EDA tools and standards for TSVs. The cost model is also unclear. And to date, there is still no way to test devices based on TSVs, said Merritt.

Many are also scrambling to develop the replacement for a DRAM or a so-called universal memory. The main candidates are FeRAM, MRAM, phase-change, RRAM, among others.

The next-generation memory types are still in their infancies. Some will not succeed. Some will find homes in a new class of devices, such as SSDs, smartphones, tablet PCs and other products, Merritt said.

The real problem is the business model for next-generation memory types. The DRAM became viable, simply because it became the main memory type for the PC. The PC allowed DRAM makers a mass market for their products, which, in turn, drove down chip prices.

That is not the case for the next-generation memory types-at least for now. "What we need is a memory technology that focuses on a replacement for the DRAM business model-not the DRAM cell," Merritt said.


All news

Memory & Data Storage

Follow us

Fast, Accurate & Relevant for Design Engineers only!

Technical papers     

Linear video channel


Read more

This month Ambiq Micro is giving away five of its 'Apollo EVB' evaluation boards, worth 9 each for EETimes Europe’s readers to assess the capabilities of their cutting-edge Apollo sub-threshold microcontroller.

The new suite of Apollo MCUs is based on the 32-bit ARM Cortex-M4 floating point microcontroller and redefines 'low power' with energy consumption that is typically five to...


Design centers     

Infotainment Making HDTV in the car reliable and secure

December 15, 2011 | Texas instruments | 222901974

Unique Ser/Des technology supports encrypted video and audio content with full duplex bi-directional control channel over a single wire interface.


You must be logged in to view this page

Login here :