SAN JOSE, Calif. Fresh details have emerged about the second spin of double-data-rate DRAM, the newest contender for the throne of next-generation high-performance memory. Though DRAM makers and analysts were split over how significant DDR-II will be, a consensus was forming that the DRAM market is headed for changes based on diversifying memory types and a fundamental restructuring in PCs the memory market's major driver.
DDR-II is expected to offer a minimum bandwidth of 400 Mbits/second per pin based on 100-MHz signaling. With chip frequencies rising, a 150-MHz core could produce bandwidth as high as 600 Mbits/s per pin. The interface is expected to run at 1.8 volts, down from the 2.5 V of DDR-I, and it will demand new packaging at both the chip and module levels, as well as a new data-capture and synchronization scheme.
Nevertheless, "DDR-II is an evolutionary change for the industry, not a revolutionary change," said Joe Macri, a member of the technical staff at ArtX Inc. (Palo Alto, Calif.) and chairman of the DDR-II task group for the Joint Electronic Device Engineering Council (Jedec). Macri presented an update on his committee's work at the Platform 2000 conference here this week.
Industry unease over the future of DRAMs is magnified by a belief that the memory's chief user, the PC, is beginning to cede ground to an onslaught of information appliances, each with its own memory needs. Indeed, Semico Research Corp. (Phoenix, Ariz.) said that while 69 percent of all DRAMs went into PCs in 1999, only 46 percent will go there in 2004, with consumer and communications gear taking up the slack.
"What it does is force more diversification," said Jeff Mailloux, director of DRAM marketing for Micron Technology Inc. (Boise, Idaho). "We're defining more products based on the application."
The first spin of DDR technology doubled the performance of standard synchronous DRAM by pumping data bits on both the rising and falling edge of the clock cycle. Macri said the revised version doubles the total bandwidth again by increasing the data fetch from 2 to 4 bits. As a result, the same 100-MHz SDRAM core that became a 200-Mbit/s-per-pin DDR device will jump to 400 Mbits/s per pin with DDR-II.
With a different package and more bandwidth, DDR-II modules will have more pins than the 184 found in today's DDR-I dual-in-line memory modules probably more than 200 pins. Macri said there is a development push for a single DIMM that will support either chip, and that DDR-II chips will be fully compatible with their predecessors.
Current DDR chips use a thin small-outline package, but the successor will require some type of ball-grid array. Macri said the final decision has not been made, but leading candidates are chip-scale packaging and micro-BGA. That issue could be resolved by March, he said.
A rapid decision is important, because there will be an intermediary version of the technology that tucks a DDR-I die into the new packaging. Those chips will likely feature core frequencies of 150 or 166 MHz. While some have referred to this generation as DDR 1.5, Macri said a better name is PC300 or PC333, because nothing on the chip is changed except the packaging. This generation could be available by the second half of next year, and will provide a transition to DDR-II using the same packaging.
Macri said his Jedec committee is currently in the ballot process for the DDR-II standard, and that the entire spec should be finished by the end of this year. DRAM vendors say this schedule means they could begin producing the chips in volume by 2003.
The question that remains is who will manufacture and use the new technology. "We don't think there will be a need for DDR-II," said Bob Fusco, marketing manager for Hitachi Semiconductor. "The main problem with DDR-II is that it will need new DIMM modules and feature lower power, and that means OEMs will have to redesign their motherboards."
Moreover, Fusco said he expects DDR-I to enjoy a relatively long life. With standard SDRAM technology continuing to improve and frequencies expected to rise as high as 200 MHz in the not-so-distant future, the prospect of a PC400 DDR-I device has more appeal for some than a more complicated PC400 DDR-II chip. "We think DDR-I will meet the demands of OEMs for another three to five years," Fusco said.
However, David Pulling, executive vice president for sales and marketing at server chip set vendor ServerWorks (formerly Reliance Computer Co.), said his company is already using DDR-I and plans to advance to DDR-II as soon as it is ready.
Next logical step
"DDR-II is the next logical step in memory evolution," Pulling said. "We are always updating and changing our products, so redesigning a motherboard to accommodate a different DIMM or power supply shouldn't be considered such a major task."
DDR-II will compete head-on with Direct Rambus DRAMs, which are now available, albeit at a cost premium over SDRAMs. Some industry watchers say the superior speed of Direct Rambus currently listed as 800 Mbits/s per pin and projected to double over the next few years will ensure that RDRAMs remain a force in the market.
"DDR II-is doomed," said Peter Glaskowsky, a principal analyst with Cahners MicroDesign Resources (Sebastopol, Calif.). "It is not technically superior to Direct Rambus. And as a designer I would not want to develop something new just for sake of the effort."
Farhad Tabrizi, vice president of strategic marketing and product planning for Hyundai MicroElectronics (San Jose), questioned whether there is room for both DDR and Direct Rambus in the niche markets of high-end systems they both target. Indeed, Semico Research released a projection this week that shows Rambus eking out just 0.1 percent of the DRAM market in 2004.
"We believe the mainstream PCs will eventually shift to DDR, but there is no doubt that OEMs do have Rambus on order now," said Sherry Garber, memory analyst for Semico Research.
"The projected demand for RDRAM chips this year is 100 million units, but I think the real number will be more like 20 million to 30 million units, which is nothing compared to the entire DRAM market," Tabrizi said. "We expect to see more and more difficulty for RDRAM to penetrate the market as a high-volume product, because the high cost means it can only make sense in high-end PCs. That segment is less than 5 percent of the total market and shrinking."
Rambus Inc. (Mountain View, Calif.) demands a licensing fee from manufacturers of RDRAMs, and the technology requires a new manufacturing infrastructure. In addition, RDRAMs currently carry both a die-size penalty and low yields.
By contrast, DDR-I chips are now just slightly higher in price than SDRAMs, but use nearly the same manufacturing infrastructure and have no significant die-size penalty. Several DRAM vendors have said the DDR chips will reach price parity with SDRAM by the middle of this year. ArtX's Macri said DDR-II technology is designed with the goal of being equally cost-effective.
Tabrizi said Hyundai is finding it nearly impossible just to break even on 128-Mbit Rambus DRAMs priced at $50, and that the economics make it hard to imagine a healthy future for the technology. "I can't get the same return on a wafer as I can with PC100 SDRAM, even when Rambus is priced three times higher," he said.
Mailloux of Micron Technology sees DDR coexisting with Rambus, SDRAMs and whatever new memory types come from the recently formed Advanced DRAM Technology group, led by Intel Corp. and five major memory suppliers, including Micron.
"By the end of the year, DDR should hit a run rate that makes it about 10 percent to 20 percent of the total DRAM market," Mailloux said. "We expect that in the future, we will have to provide three to five memory architectures in volume." And in the immediate future, he said, that will definitely include Rambus.
Micron has yet to ramp either its Direct Rambus or DDR DRAMs into high volumes, but is sampling both types. The company plans to unveil its first 128-Mbit DDR SDRAM device on Monday (Jan. 31). "Our 64-Mbit DDR has been a debug vehicle and most people are focusing on 128-Mbit DDR as a volume vehicle."
The drama of the DRAM is being played out against the backdrop of what many see as an impending shakeout in the PC business. Some industry players said the results could further fragment the DRAM market, as consumer and communications systems try to enforce their own unique memory demands.
"The memory business as we know it is going to go through some big changes," said Dean Klein, vice president of integrated products for Micron, who hopes to spearhead the company's move into what he calls "embedded logic."
Klein's group aims to roll a 3-D graphics chip with 12 Mbytes of embedded memory next quarter and an integrated graphics/core logic chip later in the year. Ultimately he is seeking design partners with consumer logic expertise to tap what Micron thinks will be a broad embedded-memory market.
"I do think there is a new landscape emerging," said Michael Slater, founder of Cahners MicroDesign Resources, in a keynote speech at this week's Platform 2000 event. "Just as the Net exploded over the last five years, there will be an explosion over the next five years of new kinds of devices that access it."
Slater lambasted today's "consumer PC" as a hobbyist toy that "can do many things but can do none of them well." Consumers, he said, want simple tools that help them do a job without imposing "device complexity."
"The PC industry has to move quickly to address the ease-of-use issue," he said, "or information appliances will eat much of their potential for growth."
Slater was generally pessimistic about whether this will happen, noting that the PC industry remains focused on speed and its software upgrade cycle rather than those underlying problems. That opens the door to a new class of vendors, Slater said.
Stepping into that opening, representatives of Transmeta Corp. showed up at the conference toting prototype Web pads based on the company's low-power Crusoe microprocessor, launched earlier this month. "This enables a whole new class of platforms you could not have made before," said Daniel McKenna, director of technical marketing at Transmeta.
While Transmeta has no announced design wins, McKenna said notebooks using the chips are expected late this year, as are mobile Web-access devices.
Mailloux of Micron said the Advanced DRAM Technology group was discussing whether it should consider such systems as its target in crafting a new DRAM technology aimed at use in 2003.