blake:
Then there's Patent-Filing-Dates...
RMBS naysayers have been saying that RMBS doesn't have a chance in hell of proving it's case in court. There is not a week that goes by that I don't find some evidence that RMBS was working on a variety of "synchronous" memory, the "S" in SDRAM, for almost 10 years.
I wonder IMO how many RMBS NDA's were sent out over the last ten years to the semiconductor industry on either concurrent Rambus or Rambus Direct technology that have be violated to produce your basic garden variety SDRAM.
Here's another reference to RMBS's 10 year old synchronous memory design written by an author who hedges his bets RMBS's royalty debate in his article title because he wrote it before Hitachi & Toshiba settlements.
john
semi.org!OpenDocument&Highlight=0,rambus*
Rambus or Rambust?
Memory, equipment vendors uncertain about timing, depth of transition to new interface
by Bill Arnold
When it was introduced nearly a decade ago, the Rambus architecture excited technologists as a revolutionary, although much more complex, way to boost memory and logic performance by creating a new, very high-speed interface. But sometimes, revolutions don't start on schedule, and those that show up early are not always rewarded for their efforts.
Even the most optimistic forecasters are ready to admit that the market for Rambus-enabled memory chips is slow taking off. Few, however, are willing to write off the technology, even though they don't believe the memory market will provide the kind of margins needed to support the cost of licensing the interface any time soon.
Why the delay? For one, it's an open secret that DRAM manufacturers prefer the evolutionary, simpler and cheaper versions of synchronous DRAMs (SDRAMs) to the more difficult and more costly to make Rambus chips. There also are die size, packaging, connector and other issues to contend with. DRAM makers don't like paying royalties to license the architecture for Rambus when the market is already in oversupply.
In addition to memory chip makers and PC designers, test equipment makers are also playing the waiting game. Rambus requires specialized test equipment, and manufacturers invested R&D money in the 1990s to develop testers for a large Rambus market that hasn't yet materialized.
The ATE industry spent huge amounts to develop core technologies to support Rambus, and they were ready to meet the original introduction schedule, confirms Robert Graybeal, product marketing manger with Schlumberger Test & Transactions, San Jose, Calif. Business right now is okay but in a little bit of a slump compared with expectations. Schlumberger is shipping from backlog, he notes.
Last year, only 50 to 60 Rambus testers were shipped worldwide, Graybeal says. Schlumberger gets credit for shipping a handful, Graybeal noted without providing specifics. The investment in new technology development has not been wasted, despite the Rambus delay, he insists. The technology is useful in a number of high-performance applications, and Schlumberger is close to recouping its engineering costs, he adds.
"It's a small market. There aren't a lot of systems shipping right now," says Glenn Farries, high-speed memory marketing manager for Teradyne, Agoura Hills, Calif. The market hasn't ramped up large because everybody is waiting to see what happens, he adds. Most of the buying is for SDRAM testers, he says.
To compete in the limited Rambus market, tester manufacturers have adopted various strategies. Schlumberger and Teradyne, for example, make specific equipment that tests the memory, I/O and logic in what is essentially a Rambus chip set. Aehr Test Systems, on the other hand, focuses only on memory parts, including SDRAMs and double-data-rate SDRAMs, according to Carl Buck, vice president for marketing at the Fremont, Calif.-based company. Test time takes longer on memory than the other chips, he says. For that application, Aehr targets lower speed testing, the bulk of test time, working with higher speed memory testers like those from Agilent, he says.
The smaller-than-predicted volume of the Rambus tester market isn't the only potentially complicating factor for vendors. The Rambus IP company has validated eight tester companies for Rambus, including Advantest, Ando, Dark Horse, Agilent and NMS (New Millennia Solutions).
Can the now-small marketplace support eight tester-service houses? No, not really, not unless the market approaches a $100 million or more Rambus memories, observes Teradyne's Graybeal. The marketplace in the near future just isn't big enough for more than three or four vendors, he says.
The market is not big enough, at least not in the current environment, agrees Farries. A high-volume marketplace could support eight or so companies among the various test segments, such as verification, wafer probe and functional test. Memory markets are usually very high volume, he adds.
Where will the volume shipments of Rambus come from? There's also another wrinkle in the Rambus tester marketplace. Rambus Inc. has validated only five DRAM makers to produce Rambus parts, according to Avo Kanadjian, vice president of worldwide marketing, Mountain View, Calif. Infineon Technologies, Hyundai, NEC, Samsung and Toshiba have signed up and two other companies are expected to join by the end of the year, Kanadjian predicts.
The shrinking number of players dedicated to Rambus is creating a problem for tester vendors because there are only five chip makers who can purchase the equipment, Teradyne's Graybeal notes. The output from a typical wafer fab line is supported by only 35 to 40 such parallel testers, according to Aehr's Buck. Do the math and you can see that about 200 such testers can saturate the market.
Kanadjian concedes that producing the equipment to support the Rambus architecture presents a "significant challenge" for tester manufacturers. Tester vendors that stick with the Rambus market will be rewarded because they will find that their testers can be used for other high-speed parts like static random access memories (SRAMs), he says.
The market will divide into leaders and followers, Kanadjian says. DRAM leaders will focus on value-added memories such as Rambus DRAMs, whereas followers will make "legacy" parts like SDRAMs, which use more standard test infrastructure, he says.
As for the perceived lag in bringing Rambus DRAMs to market, it's a situation of "no wine before its time," according to Bob Eminian, vice president of marketing for Samsung Semiconductor, San Jose, Calif. On one hand, Rambus creates potential bandwidth before systems manufacturers can incorporate it into designs. On the other hand, it creates a major "infrastructure shift" to manufacture it all along the food chain, he says. Rambus DRAMs still aren't at the yield levels of PC100 SDRAMs, he says.
Samsung produces about 85 percent of Rambus DRAMs shipped year-to-date, Eminian says. Production is more than three million devices a month now, projected to be ramping to 10 million a month by year's end, he says.
Samsung and Toshiba are the only two Rambus chip producers ready with any kind of volume because the parts really are in the preliminary sampling stage, observes Sherry Garber, vice president, Semico Research Inc., Phoenix, Ariz. Rambus DRAMs are still very expensive to make, two to three times more costly than SDRAMs from what she's heard.
These issues and observations play out against a larger canvas in the roiling DRAM marketplace. In May, Intel recalled approximately one million motherboards containing Pentium III microprocessors and 820 chip sets designed to interface with Direct Rambus DRAMs. The problem wasn't with Rambus DRAMs but rather the lack of them. Instead, the culprit was a malfunctioning memory translator hub (MTH) designed to interface the chip set with SDRAMs. The recall revealed that only a small percentage of motherboards being shipped contained Rambus DRAMs.
The MTH was designed to augment demand for motherboards because Rambus suppliers couldn't meet demand, concedes a Rambus spokesperson. Subsequently, Intel announced that end users of the 820 boards were being asked to return them to the company they bought them from, and they will get new boards with an MTH in return. Purchasers of third-party boards that used the MTH will receive a cash refund, a credit or a replacement board. Each new motherboard will contain up to 128 megabytes of Direct RDRAM at Rambus speeds from parts Intel has apparently stockpiled.
The Rambus miscues "will feed case studies in graduate business schools for years to come," declares A. A. LaFountain III, analyst with Needham & Co., New York. He describes himself as a "non-believer in Rambus" because it offers a "modest performance gain at a much higher price."
Also in May, Intel met with some major DRAM manufacturers to discuss Rambus production plans. Although variously described, the meeting was not an emergency meeting; it was a relationship-building meeting, according to Josie Parent, product marketing engineer, NEC Electronics, Santa Clara, Calif.
The meeting underscores a potentially vexing issue: Will there be enough Direct Rambus DRAMs to meet demand for Intel's soon-to-be-announced Willamette microprocessor and companion Tehama chip set? "I don't see how there could be if you look at what's been happening," observes Semico's Garber. With SDRAMs in short supply, Intel should be concerned, she says. But it's hard to know exactly what's happening because Rambus has gag orders out so DRAM guys can't disclose their yields, she adds. "There must be 20 guys working on an MTH for Willamette," Needham's LaFountain says.
Meanwhile, market-research firm Dataquest warns that a DRAM shortage is imminent and could last for up to two years. Demand for DRAMs should start exceeding supply beginning mid-year, and it won't let up until late 2002, according to Jim Handy, director and principal analyst. During that period, the market could see allocation and long-term contract pricing becoming normal, business practices not seen since the mid-1990s.
Handy is confident that the industry is on the verge of a long-lasting supply shortage that will put PC vendors, which consume 80 percent of produced DRAMs, on allocation. Consequently, PC OEMs will soon have to start negotiating long-term pricing contracts with their DRAM suppliers, something PC OEMs had mostly dropped since DRAM prices collapsed in late 1995.
DRAMs are on allocation, and "we're seeing shortages in the third and fourth quarters," confirms NEC's Parent. But a two-year capacity shortfall seems "aggressive" to her. Nevertheless, a DRAM shortfall puts pressure on vendors in apportioning capacity, especially with PC133 SDRAMs coming on board, she says.
Complicating capacity allocation are reports from South Korea that Samsung and Hyundai are swamped trying to keep up with the burgeoning demand for flash memory chips. This is caused by the exploding market for digital information appliances, such as mobile phones, MP3 music players, mobile phones and other memory-intensive mobile devices, as demand grows worldwide.
Samsung and Hyundai are responding to that demand for digital mobile appliances by expanding production of flash memory and SRAMs at the expense of DRAMs, according to reports. But Semico's Garber says that the diverted DRAM capacity probably will come from older DRAM products like EDO (extended data out) devices.
Further clouding the horizon are three legal actions by Rambus Inc. against chip maker Hitachi based on alleged patent infringement. Surprisingly, the actions don't involve Rambus DRAMs but SDRAMs instead.
In January, Rambus filed suit in Delaware federal court seeking injunctions against the manufacture, use and sale of certain Hitachi devices. One part of the suit essentially charges Hitachi with failing to discuss with Rambus how Hitachi's non-Rambus products might infringe on Rambus patents.
In April, Rambus filed a similar suit in Germany charging infringement of its European patent affecting SDRAM and DDR SDRAM devices. It alleges that Rambus invented fundamental aspects of high-speed memory interfaces used in SDRAMs and DDR SDRAMs.
The case hinges on a narrow question, according to Peter N. Glaskowsky, senior editor of the Microprocessor Report, Sunnyvale, Calif. Did Hitachi violate Rambus's intellectual property rights? This simple question, however, has implications that extend well beyond the two companies involved, he says.
The outcome of these proceedings will be directly relevant to every company making or using synchronous DRAMs, Glaskowsky says. The Rambus patents also are likely to relate to the work of the Advanced DRAM Technology (ADT) effort, he observes.
Hitachi's most substantial argument is that JEDEC policies predating Rambus's participation require JEDEC members to disclose any patent applications that relate to the work of JEDEC committees to which these members belong, according to Glaskowsky. According to Hitachi, Rambus filed its original patent application in April 1990, more than a year before it began attending the relevant JEDEC committee meetings, but did not notify JEDEC.
In Glaskowsky's opinion, the critical question is this: Does participation in a standards committee create an implied contract among the members of the committee? Contracts require an exchange of value under agreed-upon terms. He believes this requirement was met in this case. The value lies in the cooperation among JEDEC members, and the terms of the agreement were expressed in JEDEC's policies, he says.
Glaskowsky suspects that Rambus and Hitachi will be willing to settle their differences if Rambus withdrew its claims over SDRAM and Hitachi agreed to produce and promote RDRAM. Of course, Rambus is entitled to try to enforce its patent claims on advanced memory technology, he adds.
About the Author
Bill Arnold is an independent writer concentrating on the electronics industry. Based in San Mateo, Calif., he has been regional editor for EDN News Edition; editor of Electronic Materials Report, senior editor for Electronic Business and senior editor for Electronics. Arnold has covered key issues in three industry centers: Silicon Valley, London and Washington, D.C. He has contributed articles to a number of publications in the electronics and product design and development arenas, and also has done articles on technology, trends and strategies for the Economist, Financial Times of London, Upside and Buyside. He is the author of KPMG Peat Marwick's Semiconductor Industry Handbook.
Legal Information and Terms of Use ©SEMI 2000 |
