Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : Rambus (RMBS) - Eagle or Penguin -- Ignore unavailable to you. Want to Upgrade?

To: Zeev Hed who wrote (21461)	6/2/1999 2:55:00 PM
From: Dave B	Read Replies (1) \| Respond to of 93625

Zeev, You reminded me that I wanted to post some of the EDO/SDRAM transition FUD articles from 1996. Remember that SDRAM came out in mid-1996 for PCs and within 2 years was virtually the only memory you could get on a desktop. Here's four that I found. Any of them sound familiar? #1: First, that pesky cost/performance issue... Systems get synchronous: new silicon muddies the buying waters. (EDO DRAM and SDRAM) (Technology Information) Multimedia World, July 1996 v3 n8 p20(1) Author Desmond, Michael Summary Hardware vendors are jettisoning familiar memory technologies such as DRAM and VRAM in favor of Extended Data Out DRAM (EDO DRAM) and Synchronous DRAM (SDRAM). Micron Electronics rolled out EDO DRAM in 1993 to boost the efficiency of fast-page-mode DRAM. Almost all fast Pentium systems currently use the newcomer technology. Seeking even a greater performance boost, system vendors are migrating to SDRAM, which costs negligibly more than EDO RAM but delivers a 10 percent or more performance boost. However, SDRAM's Dual In-line Memory Module (DIMM) configuration will not fit into conventional 72-pin SIMM sockets. Also, SDRAM will only work with chip sets designed for synchronous memory. In real-life tests, SDRAM delivers significant performance gains only in the realm of cache misses. The technology is sure to be improved, but until then, it is probably not worth the extra cost.

To: Zeev Hed who wrote (21461)	6/2/1999 3:07:00 PM
From: Dave B	Read Replies (3) \| Respond to of 93625

#2: More price problems (scroll down, also note Rambus references)... Samsung gives in to DRAM drop. (Samsung Electronics adjusts 16-Mbit DRAM production) (Company Business and Marketing) Electronic Engineering Times, June 3, 1996 n904 p1(2) Author Carroll, Mark; Lammers, David Summary Samsung Electronics, responding to changing market conditions, announced that it will reduce the amount of production for 16-Mbit DRAM by up to 2 million pieces per month. The move is a response to spot market exigencies, which have seen the price of 16-Mbit DRAM plunge into the $10 range and currently place the value of a megabyte of DRAM memory at about $5. Samsung will immediately trim its monthly production to 12 million units. Two other South Korean producers, meanwhile, announced expansion and investment plans. LG Semicon will expand its facilities for manufacturing 64-Mbit DRAM in several 0.35-micron product lines. Hyundai Electronics will counter by increasing its rate of production of 16-Mbit DRAM from a current level of 8 million to 14 million by the end of 1996. Despite the current pricing pressure, industry analysts are predicting a 19% increase in PC sales for 1996, with a corresponding surge in DRAM requirements. Full Text Yokkaichi, Japan - DRAM prices took another sharp slide last week, prompting Samsung Electronics to announce that it will cut 16-Mbit production by 2 million units per month. In Tokyo, sources said the 16-Mbit spot market was headed for sub-$10 territory, which would value a megabyte of DRAM memory at $5. Samsung, which has been running its factories around the clock, will shut down for two days a month and give its workers additional summer holidays in order to reduce monthly output to 12 million units, according to a spokesman. However, the other two major South Korean producers vowed to increase production of 16-Mbit DRAMs. Hyundai Electronics said it will boost monthly output from 8 million to 14 million units by year's end, and LG Semicon pledged to boost capital investments for 0.35-micron lines capable of 64-Mbit DRAM production. Gyrating market All the major DRAM makers were mulling their options last week as the spot markets in Japan, Taiwan and the United States underwent unprecedented gyrations, dropping the 16-Mbit tag to $12 and lower. DRAM prices have been under heavy pressure all year, but the slide was particularly acute in May. Against this backdrop, Toshiba Corp. last Thursday opened a mammoth memory fab here, Yokkaichi II, capable of 30,000 wafers per month at 0.35-micron process rules for 64-Mbit and shrink-16-Mbit DRAM production. "What the big U.S. PC makers have been telling us is that they will work off their DRAM inventories by mid-summer," said Koichi Suzuki, in charge of Toshiba's semiconductor operations. "But the answer to whether the DRAM market will bounce back must come from the PC makers' ability to offer new products." Peter Wolff, electronics analyst at Barings Securities, said a DRAM salesman based in the United States told him that a by-4 configured 16-Mbit DRAM sold for $10.50 on the U.S. spot market late last week, a 10 percent drop from a few days earlier. Wolff said he believes Hitachi Ltd. will cut back production in response, although Hitachi publicly said it has no such plans. "Ultimately, everyone will have to cut production," said Wolff, "but the question is, when?" An executive at a major semiconductor-equipment company said NEC Corp. has signaled it may not spend the entire <> billion it originally planned for fiscal 1996 capital expenditures, citing changes in "product mix." "The investment plans differ, company by company," the source said. "[But] overall, our business will be down 10 to 20 percent or more. We are pretty worried." Matt Cleary, semiconductor analyst for the Hong Kong brokerage HG Asia, said Hyundai is currently selling 16-Mbit DRAMs for $11 on the Taiwan spot market. Another source who tracks the Taiwan spot market added that "1-meg-by-4 4-Mbit parts are in a hopeless situation because of the drop in 16-Mbit parts. I see the price for the 16-Mbit falling all the way to $10." Wolff at Baring Securities said contract prices, including those for U.S. delivery, have been forced to matchspot-market prices. "If the spot price is $12, then the contract price is fast becoming $12," he said. Despite the gloom, Akira Minamikawa, semiconductor ana-lyst at International Data Corp. Japan, said IDC recently staged a seminar in Seoul where attendees from LG Semicon and Hyundai "were very optimistic. They want to increase their market share, and they worry that if they cut their capital spending they will eventually lose market share." However, Minamikawa predicted that the South Korean Ministry of Trade and Industry will respond to Samsung's decision to scale back production by instructing Hyundai and LG Semicon to moderate their production increases. He speculated that in Japan, MITI will keep a low profile and the major producers "will talk among themselves and decide to cut back on production. Otherwise, the price will continue to drop and there will be no profits." Minamikawa said the most efficient producers can make a 16-Mbit DRAM for $12 or less, and he believes the contract price may stabilize in the $15 range in the third quarter. Not growing Even with PC sales picking up to what IDC believes will result in a 19 percent annual rise, per-system memory usage is not growing as much as expected. Though IDC predicted that figure would rise from 15 Mbytes last year to 23 Mbytes in 1996, the actual total may be around 20 Mbytes. That 3-Mbyte shortfall, combined with a huge inventory of DRAMs, spells a continuing oversupply. Hardest hit are fast-page-mode DRAMs, said Hajime Sasaki, executive vice president of NEC. "Frankly speaking, the fast-page-mode market is shrinking, but in that area we still see a serious oversupply." Sasaki said that by year's end, 30 percent of NEC's DRAM production will be synchronous parts and by March 31, the end of the fiscal year, the SDRAM total will be 50 percent. Rambus DRAMs will account for another 10 percent, Sasaki said, particularly since NEC is supplying Nintendo Ltd. with two 18-Mbit RDRAMs for each of its new Ultra 64 game machines. Though SDRAMs now carry a 20 percent price premium over 16-Mbit extended-data-out DRAMs, sources at Taiwan motherboard makers said both NEC and Samsung have assured them of price parity by the end of the second quarter. "Price is a definite consideration on whether SDRAM will make it to the market," said Al Su, marketing manager for Holco Enterprises. "Our customers will not pay a premium, since the performance gain is minimal. Price equivalency with the EDO DRAM is essential." "Using EDO DRAM with the less expensive SRAMs for cache is not only cheap, more importantly, it is dependable;why switch to an unknown?" asked Susan Shu, marketing manager for Microstar International. She noted that "the $3 extra cost for the DIMM sockets makes SDRAM a problem to begin with." In fact, with SRAMs so cheap, few companies are shipping cacheless systems anymore. Until the system bus speeds up to a point where EDO DRAMs cannot keep up, the EDO-plus-SRAM combo will dominate, said Jim Feldham, president of Semico Research (Phoenix). "Obviously, NEC is not going to move half of its production to SDRAM as long as it has a 20 percent premium," he said. Product-planning managers at several companies said they expect it will be the second half of 1997 before SDRAMs hit mass-production levels for the desktop market. The collapse of the EDO price provides an incentive to move to faster parts, they said. Only a handful of DRAM makers now produce SDRAMs in sufficient volumes for PCs. By late 1996, desktop manufacturers will be able to draw upon five or six volume suppliers with interchangeable parts. Fujitsu produces about 800,000 SDRAMs per month, a spokesman said, with most going into Fujitsu and HAL Computer servers and workstations. Limited quantities go to the desktop market.

To: Zeev Hed who wrote (21461)	6/2/1999 3:16:00 PM
From: Dave B	Respond to of 93625

#3: Same, same, same... Mixed messages on SDRAM vs. EDO. (synchronous dynamic RAM to eclipse extended data out as industry standard, but when?) (Technology Information) Electronic News (1991), March 4, 1996 v42 n2106 p4(2) Author MacLellan, Andrew Summary Synchronous dynamic random access memory (SDRAM) could replace EDO RAM as the PC industry standard, butwhen the transition will occur and how the chips will be priced relative to one another remain unclear. By 1996,Fujitsu claims that half of its products will be SDRAM, while SDRAM chips will comprise 30% of Samsung's product line by the same date. Some believe that SDRAM will cost roughly the same as EDO by 1996, while others project that the maturity of EDO production lines will enable them to keep prices below falling SDRAM pricing. Four megabyte EDO parts typically cost $3 to make and sell for $9. EDO overtook fast page mode parts as the industry standard relatively quickly, but EDO represents only a minor change from fast page mode. Full Text Mountain View, Calif.--Unstable DRAM prices and a technology shift which could accompany the volume shipment of Intel's new PCIset chipsets later this month are sending mixed messages across the memory industry as to when and at what price synchronous DRAM (SDRAM) will surpass EDO as the main memory of choice. Market analysts and at least one industry veteran say the technology roadmap of SDRAM pioneers may be too aggressive, driven by the ever-present clamor for faster CPUs and a need to maintain market share. They claim the pledge to erase the price premium of SDRAM over EDO by year's end will be difficult given the substantial price erosion experienced by both EDO and fast page mode parts and the manufacturing and testing costs associated with SDRAM. What's more, a senior market analyst suggests the pending release of Intel chipsets for home and office PC motherboards, which carry an option for SDRAM, could send the market scurrying to make the shift to synchronous memory, triggering a further price shake-up and a repeat of the tumultuous fast page mode-to-EDO switch witnessed by the industry late last year. "All memory applications, especially graphics applications and main memory applications, are under a lot of price pressure," said Rajit Shah, VP of worldwide marketing for Mosel Vitelic [my note: MV was just hit with large tariffs for dumping chips on the U.S. market], which recently rolled out a wide-organization EDO DRAM family (EN, Feb. 26). "Given that, the elasticity of EDO DRAM to come down in price is much more than emerging technologies like SDRAM. These companies are saying they can offer the same cost structure as standard DRAM at the same technology level. If standard DRAM continue to come down drastically, I don't know if that statement is valid." SDRAM vendors are unfazed, however, and appear committed to accelerating their ramp-up and cutting prices even more deeply than originally announced. They dispute the manufacturing cost disparity cited by the analyst community, and in one case even claim to have achieved smaller die size with SDRAM than is available using existing EDO technology. Two of the more vocal SDRAM proponents, Fujitsu and Samsung, have recently reiterated statements that their synchronous technology is on course to parallel EDO prices. By the end of 1996, SDRAM will account respectively for 50 and 30 percent of the product portfolios of Fujitsu and Samsung, according to the companies. "The basic rule of thumb to get the PC people to change technology is that you must bring performance advances at equal cost, or equal performance at reduced cost," said Fujitsu Microelectronics VP of memory products George J. Robillard. "What that says is that those of us who think SDRAM is the technology of the future must recognize that any drive into the marketplace is going to have to be competitive with alternative solutions." Samsung senior product marketing manager for specialty memory Tony Grant said his company is advancing its roadmap and will introduce SDRAM into the volume PC market in 1996, not 1997 as was outlined in an earlier plan (EN, Dec. 11, 1995). Mr. Grant also called Samsung's initial strategy to reduce its SDRAM price premium to 16 percent over EDO in the first half of 1996 conservative, noting that price/bit parity is ahead of schedule. "We're not concerned," he said. "(Current DRAM prices) are not going to impact our SDRAM at all. They might accelerate (our schedule), if anything." The optimism that seems to prevail among SDRAM vendors is not echoed by many market observers, who feel any stability the DRAM industry may experience could be short-lived. SDRAM/EDO price parity--even of the stripped-down PCSDRAM and 'SDRAM-lite' varieties being pursued respectively by Fujitsu and NEC--could prove elusive in the present market environment, they claim, although there is consensus that 64M SDRAM will eventually have its day in the sun. The sheer volume of manufacturing investment could help offset lower per-unit pricing, however, especially among the larger vendors. "Samsung is in the long term on the right path," said Integrated Circuit Engineering (ICE) senior market analyst Brian Matas. "If that means they have to take a hit in coming months because EDO is priced a little lower, I think they're big enough and willing to take it." Indeed, even with 4-megabit EDO parts selling for as low as $9, vendors' profit margins, on a chip that may cost as little as $3 to make, will likely remain healthy.

To: Zeev Hed who wrote (21461)	6/2/1999 3:24:00 PM
From: Dave B	Respond to of 93625

#4: Not so much FUD as a discussion of how memory standards get set DRAM race 2000. (dynamic random access memory) Electronics Weekly, Jan 24, 1996 n1752 p26(1) Author Manners, David Summary The dynamic random access memory (DRAM) architecture standard for the 1997-2000 time period remains unknown. Intel Corp. has been urging DRAM makers to use the synchronous DRAM standard. Personal computer companies, on the other hand, may have their own preference regarding DRAM standard but their influence is minimal compared to corporate giants such as Intel and the leading DRAM manufacturers. In the DRAM supplier sector, NEC and Samsung have banded together to set a new de facto standard for synchronous DRAM. Full Text Intel has proposed to DRAM makers a new-style, simplified S-DRAM which cuts down the test time and makes it as potentially cheap as standard DRAM. Will the PC boys accept it? looks like they'll have to, says David Manners A race is on to define the standard DRAM architecture for the 1997-2000 timeframe and it remains to be seen whether it will be won by the systems people, the DRAM people, or Intel. Whether Intel is a systems company or a components company these days is uncertain, but it does not doubt the legitimacy of its role in defining the future shape of the PC and many of the components that go into it. "We work", says Intel, "with a wide variety of other suppliers in the industry to make sure they understand the upcoming capabilities of our products and plan their product offerings accordingly." So Intel has been suggesting to the DRAM manufacturers a new standard for the latest kind of fast DRAM - Synchronous DRAM - which will optimise the new speed levels of the company's micros while not adding to costs. It might be thought that the PC companies should have a modicum of a say about what sort of DRAMs they might want to use but, squeezed between the interests of $16-40bn companies like Intel and the big DRAM companies, whether the profit-starved PC boys have much clout is problematic. It might also be thought that the DRAM suppliers should have some sort of say about the sort of products they will be making. To that end, NEC and Samsung have joined forces to try and establish a new de facto standard for synchronous DRAM for which they have a 50 per cent plus world market share. Samsung is playing a cagey game not only linking with NEC, but simultaneously talking about adopting the Intel standard - even, it is rumoured, to the extent of possibly becoming a sole source supplier. It is thought that Fujitsu has also lobbied for a sole source deal with Intel to supply Fujitsu's own S-DRAM architecture. But Hitachi cannily states: "We're waiting for the JEDEC standard"; Toshiba is also waiting for an industry-wide standard to emerge. The first joint NEC-Samsung 16Mbit unified S-DRAM - called SDRAM LITE - is due to come out in June, ready for the expected upturn in demand for S-DRAM starting in the second half of this year. "If there's a de facto standard it will make it easier for PC manufacturers", says NEC. [My note: Standards help? Hmmmm.] The need for S-DRAM will accelerate when the Intel PentiumPro is upgraded next year. By using a quarter micron production process, it is expected that the internal speed of the PentiumPro will be increased to 333MHz and the external speed to 111MHz and 133MHz. This ups the speed requirement for all the components on the PC motherboard - for the first time in several years - because the external speeds of Intel's Pentium micro have been kept to 60MHz and 66MHz even though the internal speeds have climbed to 166MHz for Pentium and 200MHz for PentiumPro. It is not thought to be feasible to have the processor speed much more than three times the bus speed - unless cache memory is substantially increased - and at the top end of the PentiumPro range the 200MHz internal, 66MHz external speeds ratio is on the 3X limit. That is one powerful reason why it is expected that S-DRAM will be needed to interface with new external speeds of Pentiums and PentiumPros. Another reason is that PentiumPro is intended to be used in multiples. If several micros are accessing the main memory, then obviously main memory will have to be capable of very high data transfer rates.