Will Direct Rambus Rule the Roost? -- Some DRAM suppliers expect market segmentation.
John H. Mayer
Although it's clear that the next couple of years will see a major shift to new, high-performance DRAM architectures, opinions vary widely on whether the Intel-led Rambus juggernaut will dominate, or the market will be more segmented.
Even if there is some initial segmentation, it can't last long because the market usually swings one way or the other, according to Jim Sogas, director of DRAM marketing at Hitachi Semiconductor (America) Inc., Brisbane, Calif. "But I believe we're going to have definite segmentation over the next couple of years, because we're looking at radically new technologies that are not going to be easy to switch between."
Others, including Bill McClean, president of Scottsdale, Ariz.-based IC Insights Inc., believe that although there may be some segmentation, it won't be as much as some are expecting.
Driving speculation has been the introduction in the past several months of a slew of 64-Mbit double-data-rate (DDR) DRAMs that may be able to give Rambus Inc.'s Direct Rambus devices a run for their money.
In August, Samsung Semiconductor Inc., San Jose, sampled its first 64-Mbit DDR device. In the following two months, Fujitsu Microelectronics, Hitachi Semiconductor, Micron Technology, and Mitsubishi Electronics America introduced a range of DDR DRAMs targeted at high-end workstation, server, and PC applications. Last week, IBM Microelectronics announced shipment of its first 256-Mbit DDR samples. And other suppliers, including Toshiba America Electronic Components and Siemens Semiconductor, plan to announce high-density DDR parts in 1999.
"We have a significant num- ber of customers interested in these parts," said Keith Horn, director of the Memory Marketing Group at Fujitsu Microelectronics Inc., San Jose. Like most of its rivals, Fujitsu is offering 64-bit DDR SDRAMs in x4, x8, and x16 configurations, and plans to sample 128-Mbit devices in the first half of 1999.
DDR's virtues
DDR possesses some technical and economic merits that make it a very attractive solution for next-generation, high-bandwidth systems, according to Lane Mason, director of graphics/memory product strategy at IBM Microelectronics, Burlington, Vt. "The people who are true believers in DDR on the systems side are very strong proponents, have it worked into their roadmaps, and are working with all the vendors to make sure that it is a price/performance success," Mason said.
OEMs find DDR attractive because it offers a significant performance boost without requiring a major architecture shift from SDRAMs.
Boosting performance by transferring data on both edges of the applied clock, DDR offers twice the data rate of conventional SDRAM devices, yet can use the same TSOP package and test equipment.
"The key to [DDR's] popularity is that it represents a very evolutionary path for customers," said Cecil Conkle, assistant vice president of DRAM marketing at Mitsubishi Electronics America Inc., Sunnyvale, Calif.
On the cost side, there's little incremental investment needed to bring DDR to market, vendors say. "It's very easy for DRAM suppliers to produce this part because it's not a major redesign from a synchronous DRAM to a DDR," said Chee Ho, director of marketing at Siemens Semiconductor Inc., Munich, Germany.
The die premium for DDR over single-data-rate SDRAM is 2% or 3%, a negligible difference, Fujitsu's Horn said.
Moreover, memory-IC suppliers have been collaborating to develop the infrastructure needed to make DDR a success. For example, while discussion continues on an eventual 200-pin module, the 184-pin DIMM is now fully specified by JEDEC.
"We've been working to develop a common Gerber tape for the module, and we're working with other infrastructure vendors to push the development of clock drivers and logic as well," Horn said.
But there's still work to be done. "It's a slightly different bus than that used with the SDRAM," said Mike Seibert, marketing manager for DRAMs at Micron Technology Inc., Boise, Idaho. "[DDR] has a reference and a center termination. So we're working with the major customer base and discussing some of the implications of a main memory implementation of DDR where you're putting multiple DIMMs in a system and reviewing how you lay out the busing on a motherboard and what the timing budgets are."
The Intel factor
DDR's biggest obstacle to widespread deployment remains the absence of support from Intel Corp., which has thrown its weight behind the Rambus technology. "Anything can change, but it seems clear Intel will not be leading the parade to the altar of DDR," Mitsubishi's Conkle said.
Still, a number of third-party chipset suppliers are looking at the potential of DDR, according to vendors.
"As an industry group, the DRAM guys are talking to everyone about getting chipset support out there," Micron Technology's Seibert said.
There is currently no commodity chipset that supports DDR, but there are a lot of chipset vendors looking at it, so that may change, according to Kevin Kilbuck, memory applications manager at Toshiba America Electronic Components Inc., Irvine, Calif.
In addition, the absence of support from Intel has little relevance in the high-end server, workstation, and mainframe markets, suppliers point out.
"It's an obstacle for the mainstream PC market, where Intel's chipsets dominate," Fuji-tsu's Horn said. "But it's less of an obstacle at the high end because there the chipset is often a proprietary ASIC-type solution."
Graphics and networking applications also hold significant promise for DDR devices, according to suppliers. "For example, we've spoken to many of the graphics-chipset pro- viders, and the ease of implementation of a DDR interface on their product is a lot simpler than a Direct Rambus," Horn said.
George Iwanyc, an analyst at Dataquest Inc., San Jose, agrees that OEMs in certain markets will find DDR attractive because it offers a significant performance enhancement without forcing a major architectural redesign.
"PC-card vendors and OEMs putting graphics on the motherboard are looking for fairly mainstream memory solutions right now, and workstation and telecom OEMs are still not committed to Rambus," he said. "On the server side, OEMs will want to make sure they're using a highly reliable memory solution, so they probably won't want to go to a leading-edge technology like Rambus until it's well proven. That could offer an opportunity for DDR because it is looked on as an evolutionary change."
Iwanyc expects DDR will represent close to 10% of the DRAM market within the next two years.
Pricing will play a key role in the eventual extent of the DDR market. "As we move forward, the DDR and single-data-rate [SDRAM] parts will be options on the same die, so that gives you some indication that the die cost premium will be minimal," Mitsu- bishi's Conkle said.
Early on, there may be some premium because DDR will be less widely supported and the yields at high speed may not be stellar, but eventually they'll come close to SDRAM's, IBM's Mason said.
Others in the running
A couple of other high-performance DRAM technologies are also competing.
Virtual Channel Memory (VCM), an architecture introduced by NEC Electronics Inc., Santa Clara, Calif., in 1997, boosts performance by modifying the core of the SDRAM and storing data in multiple channels between the input/output buffer and the memory cells. This approach allows the memory chip to prepare other memory data requests in a separate channel while it is reading or writing current data.
For a small silicon overhead and little additional cost, VCM promises a performance boost of up to 25% while extending the life of SDRAM.
Several months ago, NEC and Advanced RISC Machines Ltd., Cambridge, England, announced a joint effort to develop an intelligent memory controller for the new VCM devices for portable computing, multimedia, and embedded applications. In August, JEDEC approved the VCM core technology as an industry standard, and shortly thereafter Siemens announced it would support VCM and provide a second source for NEC's devices.
In addition, three major PC-chipset vendors announced they would develop Socket 7 and Slot 1 PC chipsets to support the technology. But industry support has yet to come close to what exists for Rambus or DDR.
In the meantime, industry momentum appears to be evaporating for SLDRAM, the wideband-protocol, low-cost memory that a number of DRAM suppliers initially supported as a competitor to Direct Rambus.
"We're following the technology, but we have no active design or development work going on there," said Toshiba's Kilbuck, reflecting a view many DRAM suppliers appear to share.
"We're still working on a [SLDRAM] part, but we need someone to drive it from a systems standpoint," Siemens' Ho said. "We can drive it from a component standpoint."
Rambus' availability
If a highly segmented DRAM market does develop over the next several years, it's likely to occur because DRAM suppliers, system OEMs, and Intel failed to meet the plan to deliver Direct Rambus and the infrastructure to support it. Targeted at the massive PC main-memory market, Direct Rambus solutions will represent 5% of the total DRAM market by next year, 30% by 2000, and about 60% by 2001, according to Dataquest.
But if there's any question about availability, that could make everyone think twice, Dataquest's Iwanyc said. "So far, Intel and the Direct Rambus suppliers are hitting all the milestones they've set for themselves," he added.
To help ensure supply of Direct Rambus devices and accelerate their adoption, Santa Clara-based Intel announced in October its plans to purchase a $500 million equity stake in Micron Technology.
"It's really aimed at enabling some production ramp once we have a part," said Jeff Mailloux, DRAM marketing manager at Micron.
While a number of vendors have announced 64/72-Mbit Direct Rambus devices, most are using the lower-density parts to prove the functionality of the technology, while planning mass production of 128-Mbit devices. Micron, for example, is developing a 128/144-Mbit Direct Rambus device that it expects to sample in the first half of 1999. And Samsung was the first supplier to announce, in November, that it had completed development of a 128/144-Mbit Direct Rambus device. Fabricated in a 0.23-micron process, the part supports data rates up to 1.6 Gbytes/s, or roughly 10 times the capability of currently available PC-100 SDRAMs.
Samsung will produce 100,000 devices a month beginning in January, 1999, and plans to boost monthly production to 1 million units by the third quarter of next year.
Vendors credit pricing as much as any other factor for the focus on 128-Mbit devices. "Market prices eroded much faster than anyone wanted them to, making the price gap between a 64-Mbit and a 256-Mbit push out the 256-Mbit time to market," said Avo Kanadjian, vice president of memory marketing at Samsung Semiconductor. "Given that the biggest challenge to driving Rambus demand is making it competitive with SDRAM in terms of price, we're going to take the 128/144-Mbit device to mass production because the relative die-size increase with the Rambus interface will be smaller than on a 64/72-Mbit device," he said.
The expansion problem
Memory-IC suppliers are turning to 128/144-Mbit density for their first volume Direct Rambus devices to address another concern: system expandability. There's a limit to the number of devices a Rambus bus can handle per channel. While OEMs can expand the memory channel by using repeater chips and adding channels to the memory controller, the move to 128-Mbit density for the most part precludes that necessity.
"That was one of the biggest concerns of our OEM customers," Kanadjian said. "With the 64-Mbit generation, you can only support modules up to 128 Mbytes. By going to the 128-Mbit generation, we automatically expanded the module offering from 16 Mbytes to 256 Mbytes, in increments of 16 Mbytes. That's extremely important, because when system manufacturers advertise their systems, they like to specify the range of the memory a user can upgrade to."
The infrastructure for Direct Rambus is quickly falling into place. Smart Modular Technologies has developed Rambus In-Line Memory Modules (RIMMs) in 32-, 64-, and 128-Mbyte densities based on a 64-Mbit Direct RDRAM, and will begin production in the first quarter of next year. Module supplier Kingston Technology is working with Toshiba to bring to market Direct RDRAM RIMMs. And ATE suppliers Hewlett-Packard, Teradyne, and others have rapidly developed high-speed testers for the new technology. At the same time, IC and component suppliers are working on clock generators, memory controllers, and connectors.
Still, no DRAM technology has ever come off without a hitch.
"Even the move from DIPs to SOJs was fairly painful and costly for the industry," Micron Technology's Mailloux said. "Then, in the movement from EDO to SDRAMs, which also meant a movement from SOJs to TSOPs, we went through a period when it was very difficult; and we expect the same kind of challenges for any new technology," he said. "With Rambus, you not only have the high-speed-tester issue, but you're also moving the package from a TSOP to a BGA, which requires all new equipment in the back end."
"In spite of all the activity going on, you don't really get the best check of your technology until you start putting it in the hands of a lot of customers in order to find out how many ways there are to test the true capabilities and tolerances of the part in a wide variety of system operating conditions," Mitsu- bishi's Conkle said. "It's going to be a very challenging environment for manufacturers and users alike."
|
