I was looking over the EE Times Rambus articles since the beginning of the year this weekend. Looking back it makes me wonder how anyone could have been surprised by the problems that showed up in September.
(The search I did was for "Rambus AND Samsung")
Memory makers face decisions on Direct Rambus DRAMs
01/08/99 eetimes.com
Bullish analysts predict about 250 million units of the Direct RDRAMs will ship in 1999
"But a lot of testers are needed. One estimate is that an investment at the back end of about $8 million to $10 million is needed to produce 1 million units of Direct RDRAM.
the question is whether the industry will invest the $2.5 billion or so needed to support the 250-million-unit Direct RDRAM scenario."
My personal view is that we will see shipments of about 150 million to 200 million Rambus memories this year."
Stuart Atkins, a marketing analyst for module vendor Kingston Technologies (Irvine, Calif.), said he's heard that Direct RDRAM premiums will be in the 10 to 25 percent range over conventional SDRAMs.
The high cost of those testers, the need for chip-scale packaging and increased testing for Rambus-in-line memory modules (RIMMs) will result in a stiff premium - as much as 60 percent - at the device level for the Direct RDRAMs later this year.
_____________________________________________________
PC makers may see sparse Rambus DRAMs supplies this year
01/15/99
eetimes.com
One source at a systems vendor said that his company is designing a Rambus-based system that emphasizes videoconferencing capabilities, using ASDL and taking advantage of the higher memory bandwidth
Subodh Toprani, marketing vice president at Rambus Inc. (Mountain View, Calif.), said the back-end requirements for Direct RDRAM will require an investment of about $8 million to $10 million per million units per month - or several hundred million dollars total - to get the packaging and test equipment in place this year.
at least one PC vendor said the price is 40 percent higher than standard 64-Mbit SDRAMs, and some DRAM vendors have quietly said that 50 percent is more like it.
"The PC OEMs have two concerns," said Dean McCarron, principal analyst at Mercury Research (Phoenix). "Adequate supply of the Camino chip sets from Intel, and the ability of the DRAM vendors to accomplish a steep ramp on the RDRAMs. The DRAM industry has to get to volumes on the RDRAMs, realistically speaking, by the end of this year, and that is a steep ramp over three quarters."
Sheri Garber, memory analyst at Semico Research (Phoenix), said that as of this month, several OEMs had not received enough chip sets and RDRAMs to begin their testing programs. Dell Computer (Austin, Texas) was one of the few that had, leaving its competitors chomping at the bit.
"As of last week, Dell was probably in a unique situation," Garber said. "Our info is that the other major OEMs, including Compaq, IBM and HP, were still waiting to get all the components they needed.
"By 2000, perhaps in the second quarter, things will be better," she said. "But for this year I think we are looking at a painful situation."
One of the biggest problems DRAM makers face is that the Camino chip set will arrive just as they start to shift to the 128-Mbit generation. Some vendors don't feel it's worth their while applying the new technology to a memory density that will soon be displaced.
Kanadjian said Samsung is putting all its resources this year into the 128/144-Mbit generation. The 144-Mbit parts include one extra bit, per byte, for error correction. Kanadjian said Samsung is finding that about half of the demand is for the 144-Mbit configuration. Some say error correction is more needed in the Rambus architecture than in SDRAMs.
__________________________________________________
Intel invests $100 million in Samsung to aid Rambus output
1/21/99 eetimes.com
30% of PC's this year (1999), 50% next. (Probably talking "by the end of the year").
Samsung plans to produce 500,000 72- and 144-Mbit Rambus chips a month in the first half of this year, and will expand production to 5 million chips a month by the end of the year. Samsung expects 30 percent of new PCs this year will ship with Direct Rambus. By next year more than 50 percent of systems will use Rambus, the company said.
________________________________________________________
Vendors scramble to keep pace with Rambus
01/27/99 eetimes.com
Eight DRAM makers have started to offer samples of 800-MHz parts, and the industry is still on schedule for the production of 64/72-Mbit devices, said Subodh Toprani, marketing vice president at Rambus. Toprani cited In-Stat Inc. figures predicting that Rambus memories will take 10 percent of the DRAM market this year, almost 40 percent next year and more than 50 percent in 2001.
All told, memory vendors will have to invest $7 to $8 million for every 1 million RDRAMs produced in monthly volumes. So a DRAM manufacturer that wants to output 5 million RDRAMs a month will have to spend about $40 million, Toprani said.
__________________________________________________
Rambus ramblings
2/17/99 eetimes.com
But in the second half, demand for new systems may slump unless compelling new technology is introduced. Enter Rambus, tripping, stumbling, reeling, perhaps, but coming nonetheless. If enough of the Rambus-enabled systems are on the market by midyear, it could spark a revival for a PC industry desperate for something new to offer its customers.
At the beginning of this decade, when Gordon Moore sat down to listen to a new concept of memory architecture from Rambus founders Mike Farmwald, Mark Horowitz and venture capitalist Bill Davidow, I doubt if anyone imagined that at the end of the decade the industry would be straining to give birth to the Rambus architecture.
But here we are. Or aren't. Not enough of the Direct Rambus DRAMs are shipping. A few vendors have respec-table yields, but most DRAM vendors are barely yielding. LG Semicon, one of the early leaders, is totally distracted by its forced merger with Hyundai. NEC has part of its attention on VCM (virtual channel memory); and Micron, IBM and many others have been dragged most reluctantly to the Rambus party.
And that's probably too bad. A half-hearted approach to improving yields and creating a success does no one any good.
Intel is being forced to put its S-RIMM option into action, a worst-case scenario. The S-RIMMs might allow system vendors to ship boxes, but since a PC100 S-DRAM lacks the separate control and data buses that are part and parcel to the Direct Rambus architecture, performance with an S-RIMM-stuffed system may be awful. Noise on all those traces won't please the FCC either.
The best answer, of course, is to plow ahead, improve RDRAM yields quickly, get some great boxes out to the corporate customers so they can evaluate and probe, and then ramp like hell in the fourth quarter. That will take big capital investments (Korean fab investments this year will be down slightly, according to the Korean Development Bank), and total commitment to the Rambus approach, neither of which is in abundance right now.
_______________________________________________________
ISSCC Analysis: Memory bottleneck continues to haunt designers
02/19/99 eetimes.com
__________________________________________________
Suppliers differ over Rambus DRAM plans
06/28/99 eetimes.com
Prior to the Camino delay, industry pundits projected that the equivalent of as many as 200 million 64-Mbit Direct RDRAM chips would ship this year - about 10 percent of all DRAM bit sales. Post-delay estimates are fluid, but lie in the range of 60 million 64-Mbit equivalents, said Mark Ellsberry, vice president of marketing for the semiconductor division of Hyundai Electronics America (San Jose, Calif.), the U.S. subsidiary of Hyundai Electronics Industries Co. Ltd. But several analysts said even that estimate is too high.
"We will start production about the first quarter of 2000, but maybe the schedule will be delayed because the forecast from the Intel [chip set] side has become a little ambiguous," said Koichi Nagasawa, a vice president and director at Mitsubishi, and group president of the company's semiconductor business. "My guess . . . is that Intel is waiting on chip-set delivery until the [Direct RDRAM] supply is enough for the market. If it's not enough, Intel would have a lot of inventory."
"We're trying to determine exactly what's going on," said Jan du Preez, vice president of memory product for Infineon Technologies AG in Cupertino, Calif. "We're getting very conflicting information coming from the companies that own the [Rambus] technology and from our customers."
Even among the most progressive Rambus supporters, however, yields remain an issue. A number of suppliers expect 30 percent or more of their fourth-quarter Direct RDRAM yields to be in the 600-MHz range
"Our customers are not interested in a 600-MHz Rambus," he said.
But while Lee asserted that the majority of Samsung's Rambus parts will be above 700 MHz in 1999, he still believes it will be about six months before yields of the optimum 800-MHz speed grade "reach acceptable levels." NEC and Toshiba confirmed similar production-yield curves, with most other suppliers falling in somewhere behind.
_________________________________________________
Toshiba braces for battle on memory front
07/27/99 eetimes.com
Brown reiterated Toshiba's strategy of scaling back sharply on price-pressured 64-Mbit DRAMs to concentrate on higher margin 128-Mbit chips, Direct Rambus DRAM, and embedded DRAM. He said that by the end of the year, Toshiba will be producing only about 1 million 64-Mbit chips a month, while ramping up 128-Mbit production to 4 million to 5 million units a month.
Toshiba also expects to accelerate its target of producing a 20-million-gate ASIC with embedded memory, and has scaled back its introduction schedule to three years from the originally projected five years, Brown said.
"The limiting factor is not process technology, it is tool drivers" to design such a chip, he said. This system-on-a-chip device could have as much as 64 Mbytes of DRAM, 256 Mbits of SRAM, a MIPS Technology RISC processor core, and various configurations of analog, synthesized logic, and high speed I/O, he said. _______________________________________________
Next year, Hyundai estimates that only about 7 to 17 percent of all DRAMs shipped will be RDRAMs, hampered by continuing high costs for packaging and testing.
"To produce one-million RDRAMs a month requires an investment of $10-to-$20 million. Hyundai currently makes about 40 million 64-Mbit DRAMs per month, and to convert our capacity to the Rambus memories would cost a minimum of $400 million and perhaps as much as $800 million. On the other hand, in the case of PC133 we already have the testers in place. But there is no doubt that, with Intel behind it, Rambus will become established. Hyundai is behind it."
Sherry Garber, a senior vice president at Semico Research Corp. (Phoenix), said recent events have led her to question whether the Rambus architecture will ever really get off the ground. The cost of manufacturing an RDRAM remains too high, and the market is shifting toward commodity desktops at a very rapid pace, she said. That makes Rambus a niche market even into 2001.
Garber said Semico's latest estimates call for 182.9 million Rambus memory units to be shipped next year, nearly all at the 128-Mbit density. That will account for only 6.2 percent of total DRAM units shipped. In 2001, according to Semico, that ratio will increase to only 7.8 percent.
By then the 256-Mbit generation will start to become a factor. "SDRAM manufacturers already are coming down the learning curve on 256-Mbit DRAMs for the server market, and Rambus won't have a 256-Mbit density part for quite some time" Garber said. "Can they be ready with a 256-Mbit RDRAM by 2002? That's the fear among the RDRAM vendors."
_____________________________________________________
IDF: Intel bows to call for PC133 SDRAMs
09/02/99 eetimes.com
Bowing to pressure from system OEMs and DRAM vendors alike, Intel Corp. announced this week that it will support PC133 synchronous DRAMs with an Intel chip set that the company said won't be ready until the first half of 2000.
The announcement here at the Fall Intel Developer Forum (IDF) was acknowledgement of what many have been saying for a year: that the cost of manufacturing, packaging and testing Rambus DRAMs is too high for most of the cost-sensitive PC industry.
To that end, Intel and Rambus Inc. (Mountain View, Calif.) announced the formation of the Rambus Implementers Forum, <http://www.eet.com/conferences/idf/story/OEG19990831S0080>a committee of DRAM makers charged with finding ways to reduce costs. Among their biggest tasks will be deciding how to cut the number of memory banks on each RDRAM from the current two sets of 16 dependent banks, a source of die size bloat.
Intel fellow Peter MacWilliams said system OEMs began pressuring Intel to support PC133 SDRAMs in March, after Intel said progress in its Camino chip set and RDRAM availability would not support a summer commercial rollout. Since then, DRAM makers have rallied behind the PC133 and double-data-rate SDRAMs in a big way.
Janet Ramkissoon, technology analyst at Quadra Capital Inc. (New York), said Intel and Rambus failed to pay sufficient attention to the very real manufacturing challenges RDRAMs entail. "I think we're left with a malaise. Intel says its PC133 chip set won't be ready until the first half of next year, which leaves a six-month gap for many of the system OEMs." Ramkissoon said the Rambus program "is in a disastrous situation as far as costs go. . . . Too many people listen at IDF or Hot Chips or wherever, [yet] don't consider the manufacturing issues presented by some of their new technologies."
MacWilliams said the forum may come up with one or more commonly accepted RDRAM designs with fewer memory banks, which would then be supported by iterations of the Intel chip sets. One suggestion is to trim the number of memory banks per chip to a single set of 16 dependent banks; another calls for four independent banks. The final decision "is up to each vendor, and over time Intel would build chip sets that would support different approaches," he said.
Three years ago, Rambus engineers thought adding banks would allow the Rambus technology to offer on-chip concurrency, fetching bits from multiple banks at the same time. The goal was to reduce first-access latency, an issue that dogged RDRAM's performance in the early going.
Geoff Tate, chief executive officer of Rambus, said that over the past few years, Rambus and the RDRAM vendors have come to realize that having so many on-chip memory banks made it more difficult to efficiently lay out the redundancy bits needed to keep yields at respectable levels. Adding redundancy bits for each bank added to the chip's die size, and thus increased costs. Cutting redundancy bits can also raise costs by reducing yields.
"If we knew then what we know now about the die size impact, we would have decided on fewer banks. But the issue is not just reducing the number of banks, it is making sure there is compatibility among the various DRAM vendors," Tate said.
Tate said he expects the memory-bank issue to be decided in the next few months. As vendors create new mask sets for shrink versions of the various 128-Mbit RDRAM implementations, the reduced-bank design would be implemented. Dell is expected to put a line of Rambus-based desktops on the market in October, and Bell said he expects demand to be strong from big corporations that use PCs in large networks. As companies buy new desktops based on the Windows 2000 operating system, most will opt for Rambus models that will provide performance headroom for three years on average, he said.
Bell said Dell estimates that by mid-2000, fully half the DRAMs it buys will be Rambus parts.
One advantage of Rambus' extra memory bandwidth, Bell said, is that IT managers can run data backups, virus checkers and other "touches" to the networked PCs while the user continues to work with applications.
An IT manager could run a Laplink connection to a desktop and still leave the user with 85 percent of the system performance during a data backup, he said. A similarly configured system with SDRAM would leave only 67 percent of the available performance to the user, causing many to try to reboot their systems. |
