Los Altos, California computer memory company founded in 1990. Center of a controversial
legal debate in 2000 regarding SDRAM and DDR SDRAM patents. But the real history goes a little
further back.
Background
Rambus was founded by Dr. Mike Farmwald, from the University of Illinois and
Dr. Mark Horowitz, from Stanford University. Though founded in 1990, Rambus only began to participate in the
Joint Electron Device Engineering Council JEDEC standards committee in 1992, though it previously
sat in at meetings. JEDEC
is an organization that seeks to collaboratively develop standards for electronics components, by
way of committee. One such committee, the 42th, sought to create new standards for Dynamic Random
Access Memory, or DRAM. Though Rambus had filed several patents regarding SDRAM (Synchronous
Dynamic Random Access Memory), it did not disclose this to JEDEC, while still participating in their
meetings. JEDEC regulations do not forbid patent owners from participating in discussion that would
involve their patents, but do however require these patents to be disclosed. Rambus now claims
it understood that only patents that had been granted, not only applied for, had to be disclosed.
JEDEC, however, states that Rambus is the only member who ever had this interpretation problem.
It is a misconception that Rambus suggested additions to the SDRAM standard
to lead it in a direction that would coincide with its patents; in fact, Rambus amended their
patent filings in order to reflect some of the suggestions raised at the meetings. In 1995, as
the first PC66/100 (these numbers reflect their rate, in Mhz) came out, Rambus removed
itself from the JEDEC committee, saying JEDEC's views on patents did not coincide with the company's
(without mentioning, however, SDRAM). Rambus, at this time, was preparing to launch its Rambus
Dynamic Random Access Memory - RDRAM. It had inked a deal with processor manufacturer Intel
to integrate its memory in Intel's chipset design over the next two years (effectively, until 1998).
The memory included would be Direct RAMBUS Dynamic Random Access Memory - DRDRAM.
DRDRAM emerges
In 1996, the fastest memory type available on the market was PC100. Given the new processors being
announced and soon to be released, and the transition from PCI graphics to AGP, a new memory
type was obviously needed. Double Data Rate, or DDR, SDRAM had been in development, as well
as SyncLink DRAM. Both these technologies would emerge in the future, but at the time, Intel
announced that what was to replace PC100 SDRAM was DRDRAM. In 1997, Rambus was floated on the stock
exchange and saw a 98% stock increase on the first day - it was the general view that soon DRDRAM
would be the dominant memory type. LG, Samsung, Mitsubishi and others all inked deals to
produce the memory.
As 1999 begun, things still looked very rosy for Rambus. One must keep in mind that Rambus does not actually produce the memory; rather, it licenses
the design to other companies, who then produce it. The line of companies licensing Rambus and ready
to produce it was long; what they needed, however, was a chipset that could use the memory. Intel's
i820 was to be that chipset, and it was expected to be available in the second semester of 1999, which,
coupled with a mass production of DRDRAM memory modules, would mean large availability of these
computers. AMD, Intel's main rival in the processor business, announced their upcoming K7 would
include RDRAM support.
Not all, however, supported the DRDRAM module as the ideal one. A group of companies, led by
chipset manufacturer VIA, supported the PC133 specification instead. The release of the i820
would be crucial to assuring a strong foothold in the market. However, the i820's release was plagued
by a host of issues, many of which related to DRDRAM, including a CMOS truncated bit error
that would prevent the data from being read from RAM. Meanwhile, Intel released the i815, which ran
on - you guessed it- PC133 SDRAM memory. "Just in case".
As more issues were ironed out and the drivers improved, the industry readied itself for the
widespread launch of the i820, with motherboard manufacturers such as ASUS and AOpen publishing
their i820 specs. VIA, meanwhile, announced its Apollo Pro133, based on PC133, and Micron
made a surprising annoucement a week before the i820 was due to be released: it was going to
work with the Pro133 rather than the i820, claiming that SDRAM's price-to-performance ratio
was much better. And then, another blow: Intel discovered a flaw in the third RIMM slot that would
cause a memory error, regardless of the slot being used or not. Hundreds of thousands of such
chipsets had already been produced, however, and they now were practically useless. Estimates
place the loss at nearly $100 million.
That old patent thing
Rambus found itself in a tricky situation. Its RDRAM was a money loser and the company needed
to do something to keep itself profitable. They broke out the law division and filed suits against
all SDRAM manufacturers, pointing to patent infringement and asking them to pay fees. A couple caved
in immediately, because they depended not on SDRAM but on RDRAM for their videogame products (Toshiba
and the Playstation 2, Hitachi and the Sega Dreamcast). Others, like Micron and Infineon
, were adamant in not
paying the licensing fees, pointing out that Rambus had obtained those patents while
violating the JEDEC's policies. When it got to court, Infineon won its case against paying
the fee and established a precedent
likely to be followed in other cases. Still, Rambus made quite a bit of money from the companies
that did pay the licensing fees, and had inked a deal with Intel that stated the processor maker
would use Rambus exclusively in 1999-2002, would obtain one million Rambus stock options,
and would not sign any memory deals with other companies.
The Pentium 4
Intel still holds on to the notion that RDRAM is the best memory specification to date; in some
aspects, it may be right. However, the large cost of RDRAM production initially hurt sales of Intel's Pentium 4
processor and allowed AMD's Athlon to gain marketshare. The Pentium 4 is now compatible with
both RDRAM (running at 800Mhz) and PC2700 DDR SDRAM. This suggests some sort of understanding
was reached between Rambus and Intel, as it clearly goes against their previous agreement.
The lower cost of DDR means it still holds
court in the price-to-performance ratio, and DDR and RDRAM each hold their own in different applications.
RDRAM's much-touted bandwidth advantage is undermined by its greater (i.e., worse)latency when
compared to DDR. Therefore, DDR memory oftentimes outperforms or matches RDRAM in real-life
applications, but at higher processor speeds, Rambus seems to gain an edge.
In June of 2002, the Federal Trade Commission filed a suit against Rambus for anti-competitive
business actions. Also in 2002, Intel announced future workstation and server products would drop
support for the RDRAM memory type, and their 2003 plans would focus on DDR-II rather than Rambus.
However, seeing that Rambus, in January of 2003, posted earnings of 6 cents per share, and RDRAM continues
to be used both in the PC market and the videogame console market, it seems, at least in the short term, that Rambus' future is secure.
Sources
- "Intel to drop support of Rambus in new CPU products". (http://www.ebnonline.com/story/OEG20020226S0040)
- Rambus (http://www.rambus.com)
- Rambus: the Story of One Company (http://www.xbitlabs.com/mainboards/rambus/)