SI
SI
discoversearch

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: Alan Hume who wrote (22957)6/19/1999 2:49:00 AM
From: unclewest  Respond to of 93625
 
alan,
i think you are on to something there. delaying coppermine until november times the release with the scheduled completion of micron's rdram production ramp.
at the same time, it just might result in an earlier release of camino. we know that a lot of computer and peripheral companies have working rambus boxes. does that mean rambus enabled chipsets are already shipping for slower processor speeds?
unclewest



To: Alan Hume who wrote (22957)6/19/1999 2:59:00 AM
From: unclewest  Respond to of 93625
 
Date: 06/18 19:16 EST

K7 Renews AMD Hope Of Challenging Intel

Jun 18, 1999 (Tech Web - CMP via COMTEX) -- Maybe it's because the
design team's leader and many of its members hail from Digital
Equipment's Alpha team, where technical derring-do prevailed. Maybe
it's the culture at Advanced Micro Devices, where nearly irrational
perseverance in the face of Intel's far larger resources is a way of
life. Or maybe it's the simple fact that people like an underdog -- and
a little price competition.

Whatever the reason, many say they are openly hoping that AMD's K7
processor, due to be formally unveiled later this month, will be a
blockbuster.

"The K7 will be a success; it will become AMD's bread and butter," said
Donny Chien, a manager at motherboard giant First International
Computer, in Taipei, Taiwan.

But analysts and battle-hardened competitors are less sanguine, warning
that even though the K7 processor itselfmay be a technical success,
creating the surrounding infrastructure to support the MPU could be a
challenging task, particularly in the commercial market.

The K7 is the company's most powerful and complex assault on Intel's
X86 juggernaut yet. The 120-person K7 team completed the design in 17
months, said Dirk Meyer, chief architect of the K7 and a veteran of MPU
designs at Intel and Digital Semiconductor. With 500-, 550- and 600-MHz
versions already being sampled to many personal computer OEMs, the K7
appears set to move into the real world. And AMD has a system bus,
licensed from Digital, which adds to the technical merits of the
processor.

At a recent dinner sponsored by MicroDesign Resources, in Sebastopol,
Calif., Meyer said that the K7 outperforms Intel's Pentium III in a
variety of benchmarks, and runs 40 percent faster on floating-point
tasks.

"We've had sample K7s along with reference boards since April," said an
engineer at a major Taiwan motherboard manufacturer, who asked not to
be identified. "Currently, we have both 550- and 600-MHz sample CPUs.
We are now on our fourth revision of K7 mainboards, and we expect to go
into mass production in August."

AMD and U.S. OEM customers have provided help to the Taiwanese in the
design of what is the first high-performance, non-Intel board in recent
history. "We are getting good support from both AMD and our OEM
customers in dealing with the new bus and CPU," said the engineer. "We
also have ex-Digital employees who are familiar with the EV6 bus
architecture. And we Taiwanese are no slouches when it comes to
mainboard design."

AMD's own Iron Gate chip set is now sampling, but third-party parts
have yet to show up for OEMs to test-drive, said an R&D manager at a
U.S. PC maker. When those chip sets emerge, it is believed all the
first-generation ones will support the existing mainstream PC-100
synchronous DRAM, later migrating to 133-MHz SDRAM. Acer Labs has
discussed plans for a highly integrated K7 chip set, aimed at low-cost
business PCs, but so far has not given OEMs spec sheets on the product.

In Taiwan, several core-logic design houses have alternatives to the
AMD core logic in the works. "We will sample a K7 core-logic product at
the end of June," said an executive at Via Technologies. "We will offer
a mass production product about two months later. We are definitely
working on it, but it is not our first priority."

A Pentium III chip set is, but that may change if Intel keeps turning
the thumbscrews on Taiwan's core-logic vendors. Intel recently forced
Via to withdraw samples of a PC133 chip.

Also, the K7 alternative emerges as Intel is pressuring OEMs to build
systems using the Direct Rambus memory architecture for the fourth
quarter -- traditionally the biggest selling quarter, thanks to both
Christmas buyers and commercial IT managers expending their year-end
budgets. However, Intel's Camino chip set, its first to support Rambus,
has been delayed and Intel has had to roll out a reduced-speed 600-MHz
version before it will be able to field a full-speed 800-MHz version.

"I have to do a Camino system to stay on Intel's good side and to
protect myself from competitors that will also have the design," said
one PC R&D manager. "As for K7, I need to have a program on the back
burner just in case it really takes off."

A marketing manager at the up-for-sale Cyrix division of National
Semiconductor said it is all too easy to underestimate what it takes to
launch an entire platform, with chip sets and motherboards that differ
from the mainstream Intel design. Not all of the Taiwanese companies
have the engineering depth required to design multiple chip sets and
motherboards, this source said.

An engineer at a U.S-based PC manufacturer said he believes the board
layout will be challenging, partly because of the faster EV6 bus.

Socket 370 For AMD? As attractive as the 200-MHz EV6 bus may be, with
its point-to-point connections and synchronous source clocking, it does
not have the volumes Intel's Socket 370 approach enjoys for the
consumer market. And that raises the question of whether AMD will try
to field a Socket 370 version of the K7, despite apparent legal
barriers.

AMD recently settled two historic legal disputes with Intel. One gave
it the right to make compatible X86 processors, but the other --
involving AMD's right to clone the MMX instruction set -- limited AMD's
ability to field parts that used the Pentium's P6 bus, and that could
plug into Intel-standard motherboards.

For the consumer market, Intel has turned to Socket 370, which is
electrically compatible with the P6 bus. Many expect that AMD will be
forced to put a Socket 370 part on the market, partly to take advantage
of the far larger volumes, and lower costs, associated with the Socket
370 infrastructure.

AMD's official position is that it has no need to do a slower P6-based
bus when it has a 200-MHz bus in hand. "Why would we want to go
backwards, to the P6 bus?" said an AMD executive.

Mike Feibus, a principal at Mercury Research, in Scottsdale, Ariz.,
said the EV6 bus "is out there and shipping ahead of the K7. It is a
proven bus. But there is no doubt that developing a Socket 370 version
of the K7 would be a smart thing to do. There doesn't seem to be any
design activity in that area that we know of, and it may be that AMD is
not willing to take a chance, given the two cross-licensing agreements
with Intel.

"They might go ahead and take a shot, and hope the Federal Trade
Commission backs them up. It would be a lot easier to sell a chip that
just plugs in [to a P6 motherboard]," Feibus said. "But the 'thou shalt
not' [in the Intel-AMD agreement] must be very strong."

Going in a different direction from Intel brings "risks as well as
rewards," Feibus said. But this time around, AMD is far from alone.

When Intel was forced to respond to the aggressive pricing of the K6
processor, many PC OEMs realized that keeping AMD healthy was to their
own advantage. Feibus recalled that in the first quarter of 1998,
Hewlett-Packard fielded a 200-MHz Pentium-based system to the consumer
market, while Compaq Computer put a 233-MHz-based K6 box on the market
at the same time. The Compaq machine won in the consumer market,
proving that "megahertz, not the brand of the processor, is the main
selling point. That had a profound effect," he said, "and I think some
PC OEMs will belly up to a similar challenge by taking the K7 into the
commercial space."

Meanwhile, one of AMD's partners, the Alpha Processor (API) subsidiary
of Samsung, will detail plans on Monday at PC Expo in New York to use
the core-logic AMD has built for its K7 in new low-cost motherboards
built around the Alpha 21264 processor. API will also tip word of an
upcoming merchant chip set, of its own design, that will support either
K7 or Alpha in single- and dual-processor servers that use a high-end,
133-MHz double-data-rate SDRAM memory subsystem.

"People take for granted the infrastructure Intel has created - the
sheer amount of technology they have put out that enhances their CPU
sales," said Gerry Talbot, chief technology officer for Alpha
Processor. "There are unique things like chip sets, cache interfaces,
and power supplies that we have to nurture together. AMD is our partner
in a big way."

API will roll out a Slot B processor-module interface in the next week
that is mechanically similar to Intel's Slot 2 server interface but
uses the K7/Alpha EV6 processor bus. The company also plans to launch
chip sets supporting K7 and Alpha for four- and eight-way servers a
year from now.

The API chip sets will sport a unique interconnect design that allows
users to plug multiple interfaces into their high-end memory subsystem,
opening a door for uses in high-end routers and switches where API is
already beginning to court design wins. That effort could ultimately
take both the Alpha and K7 chips into networking and telecom
territories.

API is not alone in laying a high-end path for the K7. Startup Poseidon
Technology, in San Jose, Calif., led by Apple and Exponential veteran
Rick Shriner, is developing eight-way multiprocessing chip sets for K7
due to ship by year's end.

The commercial market has long been a nonstarter for AMD, filled with
end users who played it safe with an Intel-only purchasing policy.
Hewlett-Packard, for example, canceled a line of AMD-based commercial
desktops more than two years ago when it found it faced an uphill
battle selling the machines to IT managers.

One manager said "AMD has never had a server chip before, so there's a
question hanging over them there," the PC engineer said. "And we always
have doubts about whether AMD can really manufacture their latest
processor in the volumes we need."

Indeed, the company's historic manufacturing problems, and questions of
whether AMD has the resources to keep its momentum going forward, hang
over the K7.

Keith Diefendorff, an analyst at MicroDesign Resources, noted that the
die size of the K7 is a relatively large 184 square millimeters at
0.25-micron process rules.

Despite the cost pressures in the consumer market, Intel has managed to
keep its overall average selling price relatively stable in recent
months. The ASP difference -- $227 for Intel's overall line of MPUs,
and only $78 for AMD -- means that Intel has plenty of financial
resources to pour into its upcoming Katmai and Willamette processors.

Feibus cited an equally daunting statistic from Mercury Research's
quarterly report on PC processors. In the fourth quarter of 1998, AMD
had 16.1 percent of the microprocessor market, but that fell to 13.6
percent as Intel lowered the prices on its Celeron and AMD's
manufacturing stumbled.

Will the K7 reverse that slide? Feibus said "the personal computer OEMs
see AMD as the last bastion of competition to Intel, and I think they
will support the K7 partly for that reason. For AMD, it's theirs to
lose."

Additional reporting by Mark Carroll and Rick Merritt

-0-

Copyright (C) 1999 CMP Media Inc.




To: Alan Hume who wrote (22957)6/19/1999 3:03:00 AM
From: unclewest  Read Replies (1) | Respond to of 93625
 
Date: 06/18 23:52 EST

Knights of the DRAM table

Jun. 18, 1999 (Electronic Buyers News - CMP via COMTEX) -- Chip makers
are still on their quest for the Holy Grail of DRAM-namely, the
value-added memory that will release them from the bloodletting in
commodity devices. But this Grail may prove as elusive as the one in
Arthurian legend.

That's good news for OEM purchasers, who've reveled in cheap memory
for the past three years. As long as the world is swimming in commodity
memory, OEMs can count on continuing their current buying strategies.

Speaking to EBN in Korea, Yoon Woo Lee, president and chief executive
of Samsung's semiconductor business, reiterated his company's goal of
moving away from commodity DRAMs toward higher-margin memory products.
He told EBN that Samsung will hold 64-Mbit DRAM production at about the
current level of 15 million to 20 million per month, while moving
rapidly to more profitable 128-Mbit and Direct Rambus chips.

NEC, another DRAM power, is ramping up DRAM output, but focusing on
higher-priced 128-meg and embedded memory. And Toshiba, as a principal
supplier of the new DRAM chip for Sony's upcoming PlayStation 2 game
console, is casting its lot heavily with Direct Rambus.

Commodity DRAMs continue to be a chip-industry anathema. The pricing
freefall has resumed, after a brief leveling-off at the end of last
year, with the perennial culprit being a soaring chip oversupply.
Vendors have hoped that the DRAM market might eventually stabilize,
with a balance of supply and demand, but those hopes keep slipping. And
though the earlier consensus was that the light at the end of the DRAM
tunnel might come in this year's second half, the view is now shifting
to year's end or early 2000.

In the meantime, the pot of value-added DRAM gold at the end of the
rainbow continues to prove evanescent.

True, 128-Mbit DRAMs carry a price premium, but volumes are just now
starting to ramp up. Since 128-megs are made on the same lines as
64-Mbit devices, the major memory companies will simply shift the
production mix as demand for the new chips increases. With the Big Boys
in full sway in the die-shrink race and many upping capacity, there's
no reason to think that 128-megs won't be in ample, if not
overextended, supply when the market takes off.

As for Direct Rambus, its dice are 15% to 20% larger than SDRAM dice.
Since Direct RDRAM initially will divert production from existing SDRAM
lines, any big shift to Direct Rambus could gobble a lot of excess
silicon.

The biggest threat for OEMs is not the so-far futile attempts by DRAM
makers to shift out of commodity devices; rather, OEMs may feel the
impact if some suppliers finally get tired of the game and drop out of
the market altogether. On the other hand, if the big power players in
DRAMs continue their massive ramp-up of output, the dropouts might
never be missed.


-0-

By: Jack Robertson
Copyright 1999 CMP Media Inc



To: Alan Hume who wrote (22957)6/19/1999 3:10:00 AM
From: unclewest  Respond to of 93625
 
see toshiba/sony paragraph...in bold

Date: 06/18 23:52 EST

Japan's DRAM makers stray from PC mkt.

Jun. 18, 1999 (Electronic Buyers News - CMP via COMTEX) -- Tokyo-
Renewed pricing pressure in the commodity DRAM arena is causing a
schism in Japan's semiconductor market, as chip makers here wrestle
with their long-term commitment to PC main memory.

Four of the country's top DRAM manufacturers-Fujitsu, Hitachi,
Mitsubishi, and Toshiba-are significantly reducing their exposure to
the volatile PC-memory sector and will redirect efforts toward high-end
computing, consumer-electronics, and communications products.

Only NEC Corp.-Japan's leading DRAM supplier and the world's
third-largest memory-chip maker-will continue to sell most of its
devices into mainstream PCs, according to a series of interviews EBN
conducted here last week with senior executives from the nation's
semiconductor industry.

By staying the course, NEC hopes to hang in as one of a handful of
very large DRAM vendors able to subsist on razor-thin profit margins by
churning out tens of millions of commodity devices. For the rest of
Japan Inc., which once dominated the PC main-memory market, it's the
end of an era.

"I don't know who can get profit [from the PC market]," said Ryusuke
Hoshikawa, executive vice president of Fujitsu Ltd.'s Semiconductor
Group, and chairman of San Jose-based Fujitsu Microelectronics Inc.

According to Hoshikawa, Fujitsu hopes to cut PC-related DRAMs by as
much as 90% during the current fiscal year, ending March 31, 2000. He
said the company is shifting rapidly to non-PC memory chips, which now
account for more than 50% of all Fujitsu DRAM sales. As a result, the
company's DRAM production bit rate will decline next year, he said.

Hideo Inayoshi, deputy general manager for strategic planning at
Hitachi Ltd.'s Semiconductor and Integrated Circuits Division, said
fewer than half the company's DRAMs next year will be sold into desktop
PCs. Hitachi's bit growth is expected to increase by about 75%, roughly
in keeping with the 70% growth that has served as a historic industry
average. "We will focus on DRAMs and 256-Mbyte modules for high-end
applications, such as servers and workstations, and 128-Mbit DRAMs for
notebooks," Inayoshi said.

Mitsubishi Electric Corp., meanwhile, said that while it will remain
in the PC market, it will use its position to advance into areas such
as servers, workstations, and communications platforms, in addition to
consumer electronics. "[The PC] DRAM business is so risky, but DRAM
technology is required in [system-level integration]," said Koichi
Nagasawa, president of Mitsubishi's semiconductor group. "To stay in
DRAMs, we need a certain amount of volume production."

And at Toshiba Corp., where 80% of all DRAMs made today are sold into
PCs, exposure to the market will be slashed to about 20% of output by
early 2001, according to Yoshihide Fujii, general manager of strategic
planning for the company's semiconductor unit.

Of Japan's top chip suppliers, NEC is the only company not actively
distancing itself from the PC market. The company sells about 70% of
its DRAMs to the likes of Hewlett-Packard and IBM, and believes that by
ramping up output and servicing the PC, workstation, and server needs
of large OEMs, it will emerge from the DRAM industry's pricing
imbroglio with its market ranking unscathed.

"So far, the PC is our biggest market, and we mainly sell to big PC
companies, which also use a lot of memory for workstations," said
Keiichi Shimakura, associate senior vice president of NEC.

Having invested in the construction of an 8-inch joint-venture DRAM
fab in Shanghai, China, last year, capital spending this year will
decline to about $1 billion, down from approximately $1.25 billion in
fiscal 1998, according to Hajime Sasaki, chairman of NEC. Sasaki said
volume production at the site is at 5,000 wafers per month and will
double by the end of the year. In all, NEC said bit growth will
increase 80% in 1999, about the same level as last year.

Additionally, NEC expects that its proprietary Virtual Channel Memory
technology will be used as a DRAM core in portable PCs because of its
low power and relatively high speed.

Weakened by the rampant defenestration of commodity DRAM prices,
which have slid another 18% in the past two weeks, all of Japan's
suppliers-NEC included-are using precious capital resources to develop
application-specific DRAMs while broadening their ASIC, flash-memory,
microcontroller, microprocessor, and other non-DRAM semiconductor
design efforts.

Fueling this transition is a wave of consumer-electronics devices,
such as digital still cameras, televisions, set-top boxes, and DVD
players, although whether these markets can support the production
plans of every DRAM supplier remains to be seen.

Toshiba, for example, is riding high on the business it expects to
secure through sales of Direct Rambus DRAMs into Sony Corp.'s
PlayStation 2, to be released late this year in Japan. Toshiba worked
with Sony to develop the PlayStation's 64-bit MotionEngine CPU. The
game machine uses two 128-Mbit Direct RDRAMs, and Toshiba expects to
supply half of that demand.


To meet the market's needs, Toshiba said it will expand production at
Dominion Semiconductor, its Manassas, Va.-based joint fab operation
with Motorola Inc., and will increase its outsourcing agreement with
Winbond Semiconductor Corp. in Taiwan.

Despite its commitment to DRAM, Toshiba stressed a well-balanced
approach in which memory ICs-flash and SRAM as well as DRAM-make up no
more than 25% of the company's product portfolio. "We need to be strong
in memory and processors to be competitive in the system LSI market,"
Fujii said.

Mitsubishi hopes to maintain a critical mass of DRAM manufacturing by
shifting to other types of consumer products. Nagasawa also sees
embedded DRAMs growing rapidly in a wide range of consumer goods.
Mitsubishi is counting heavily on its joint chip-development program
with Matsushita as a means to switch its DRAM focus away from the PC
market.

In addition to diversifying into non-PC markets, several Japanese
chip makers are reining in capital spending. Although it plans to
increase production, Toshiba will cut its semiconductor capital
spending in 1999 by about 8%, to a little less than $1 billion. Fujitsu
is projecting a 30% drop in spending, to approximately $475 million.
Mitsubishi is planning about $375 million in new semiconductor capital
spending this fiscal year, flat with 1998. And Hitachi will hold its
chip capital expenditures to last year's levels, about $580 million.

According to Inayoshi, Hitachi is considering a resumption of DRAM
outsourcing to LG Semicon Co. Ltd., which is in the process of a merger
with the chip division of Hyundai Electronics Industries Co. Ltd. LG
had been a longtime DRAM contract manufacturer for Hitachi, supplying
the company's customers with as many as 2 million units per month. But
outsourcing levels have "dropped to zero" as a result of the prolonged
price pressure in the market, Inayoshi said.

Nagasawa said Mitsubishi will continue to outsource 45% to 50% of its
total DRAM production to Taiwan's Powerchip Semiconductor Corp., and
will transfer 0.18-micron process technology to Powerchip in the third
quarter.

Fujitsu's Hoshikawa said his company would also continue outsourcing,
but he was less definite about which foundry it would use. "We may need
to make some changes," he said, without elaborating. Fujitsu is using
Taiwan Semiconductor Manufacturing Co. Ltd. as its principal foundry,
and according to a number of industry sources has signed with Acer
Semiconductor Manufacturing Corp., also in Taiwan.


-0-

By: Jack Robertson and Andrew MacLellan
Copyright 1999 CMP Media Inc.



To: Alan Hume who wrote (22957)6/19/1999 3:14:00 AM
From: unclewest  Respond to of 93625
 
Date: 06/18 23:55 EST

Hon Hai takes aim at RIMMs -- Taiwanese co. developing Rambus-compliant connectors

Jun. 18, 1999 (Electronic Buyers News - CMP via COMTEX) -- Adding its
name to the growing list of connector suppliers offering parts that
support the Direct Rambus DRAM architecture, Hon Hai Precision Industry
Co. Ltd. is developing connectors for Rambus in-line memory modules
(RIMMs).

Under an agreement signed with Rambus Inc., Taiwan-based Hon Hai will
develop the RIMM connectors, as well as RIMM-continuity modules and
module heat spreaders for Rambus memory systems.

"With the addition of Hon Hai, the availability of Rambus
infrastructure components worldwide gets a big boost," said Subodh
Toprani, general manager and vice president of Rambus' industry-
enabling division. "OEMs, no matter where they are, now have another
source of supply for connectors."

Hon Hai, which operates in the United States as Foxconn International
Inc., Sunnyvale, Calif., last year ranked 11th in worldwide connector
sales and first in sales in the Pacific Rim, according to Fleck
Research Inc., Santa Ana, Calif. The company lists each of the top 10
PC and motherboard manufacturers among its customers, including Compaq
Computer, Hewlett-Packard, IBM, and Intel. About 70% of Hon Hai's
$632.5 million in connector sales in 1998 were for PCs.

With its agreement to manufacture the RIMM connectors, Hon Hai
becomes the fourth major connector supplier preparing for a strong
surge in demand for parts compliant with Rambus' memory-bus
specifications. AMP, FCI Electronics Worldwide, and Molex have also
ramped up into high-volume production after developing the products for
about a year.

So far, AMP and Molex's parts have passed Rambus' validation-testing
requirements to show they meet the impedance requirements that enable
transmission of data on the 800-MHz Direct DRAM interface, with FCI
expected to follow shortly.

Hon Hai, which had been developing its RIMM connectors for several
months prior to signing the agreement with Rambus, expects to have its
parts in full compliance with Rambus specifications before the end of
the summer, and to be shipping the parts in volume by early fall,
according to Jeff Pan, the company's RIMM connector product manager.

"Our plan is to ramp up and support about 300,000 per month by the
end of July, and probably double that by the end of August," Pan said.
"We expect, probably starting in September, that the volume will be up
much more significantly. After we pass the validation testing, I
believe we need about two months to allow our customers to try them out
and to use our products to test their designs."

Once the company completes the validation process, it plans to shift
the bulk of its RIMM-connector manufacturing to its facilities in
China, Pan said.

Neither Hon Hai nor any of the other suppliers currently tooling up
for mass production of RIMM connectors has expressed concerns or backed
off from manufacturing plans because of Intel Corp.'s delayed launch of
the "Camino" chipset.

AMP, for example, already has the capacity in place to churn out 800,
000 parts per month, and has plans under way to boost that to 1.5
million per month by September.


-0-

By: Richard Richtmyer
Copyright 1999 CMP Media Inc.



To: Alan Hume who wrote (22957)6/19/1999 3:37:00 AM
From: unclewest  Read Replies (2) | Respond to of 93625
 
dated 6/10/99 do not recall seeing this posted here.

developer.intel.com



To: Alan Hume who wrote (22957)6/19/1999 3:46:00 AM
From: unclewest  Respond to of 93625
 
not sure this got over here either...

electronicnews.com