Friday, June 02, 2000
For over a few years, the PC industry has been looking for a new kind of memory to elevate itself to new level of performance. Intel announced exclusive support of Rambus memory 4 years ago and has since put most of its eggs in the basket. After several technical delays, it finally shown off the 820 and the 840 Rambus enabled chipset at last Comdex Las Vegas Show (November 1999).
During the time, the JEDEC Committee (an industry standard setting body) and its members has been working diligently developing a new evolutionary type memory called DDR (Double Data Rate). In the roll out summit held at the San Jose Double Tree Hotel last week (May 18, 2000), the DDR memory was made official. That signifies the official market entry of the new type of memory.
In order to predict which memory standard will win out at the market place, one must look from all angles to determine the technical merit, the marketing effort and the economic of the new memory.
From The Technical Aspect:
Does 800Mhz bring more performance than 266Mhz?
For the general consumer, that is always a misconception. Rambus is a serial bus technology that compresses data from the 64bit CPU bus to an 8 bit narrow bus. In doing so, it executes 8 bit transmission 4 times at higher frequency to complete the 64 bit data transfer. The DDR memory bus, on the other hand, does 64 bit parallel transmission all at one time. Therefore, the effective bandwidth of the 800Mhz Rambus is 1.6 Gb/s while the 133Mhz clocked DDR is 2.1 Gb/s bandwidth. That means DDR memory can bring significantly higher performance. This is evident by the bench mark comparison of “DDR vs. Rambus”.
800Mhz is hot as a pistol!
As a radio engineer in my early years of electronic design, I learned that anything VHF (50-250Mhz) can be touched and feel. However, anything above 500Mhz is getting into black magic and is not always predictable. We have already learned that the Rambus memory modules and the motherboard require specially made (28 ohm) impedance PC board. It also requires a specific layout dictated by Intel provided example board. Motherboard manufacturers were also told not to make any modification on the Rambus channel. Even with all the precautions, we still hear of technical delays and recalls on Rambus systems.
Infrastructure problem (or the lack of it)
In the history of electronic memory, supporting infrastructure was the key to success. In the case of Rambus memory, it requires special PC board, special CSP assembly, ball grid soldering, and X-ray inspection. Most of all, it also requires expensive unproven test equipment. A custom heat spreader is also needed to relief the heat concentration problem. At the semi-conductor level, it has extra circuitry and memory arrays that make it difficult to overcome the yield curve.
On the contrary DDR is an evolutionary technology that can reuse most of the existing infrastructures built for the regular SDRAM. The PCB impedance is the regular 60 ohm. Testing can be performed with slightly modified SDRAM testers. Since the die difference is only at the output stage, the same identical die can be used for both SDR and DDR with a simple mask option. Due to the reasonable 266 Mhz frequency, logic vendors have no problem in providing new support logic chips like differential clock, PLL, buffer, and SSTL interface converters. Tester vendors had also joined in to announce their DDR ready testers.
There are other compatible processors and chipsets
AMD and other X86 compatibles can be a threat, too. AMD has recently announced that it’s mobile processor is now standing at 60% of the mobile market. Its Athlon is gaining momentum with good yield from their 0.18uM factories in Germany and in Austin. With the introduction of the low cost socket-A version of the Athlon processor, they will give the Intel Celeron processor a run for the money. Intel, on the other hand is due to introduce it’s new Willamette and Timna. These new processors will have problem with memory since they were designed to work with Rambus memory directly. They do have the original option to also work with SDRAM through an MTH (memory translation hub). With the latest trouble discovered in the MTH, Intel will probably delay this popular option.
There is also a new X86 compatible processor on the market. The Transmeta Crusoe processor taken alternative architecture and use a special internal language together with Linux operating system. It has, therefore, avoided the conflict and license requirement with Intel. It was designed to be most suitable for Internet Appliance and Mobile applications. Co-incidentally, Transmeta has also chosen DDR memory as its primary memory technology.
Micron Technology has been showing its DDR enabled prototype motherboard since September, 1999. By now, we would assume that most of the technical problems has been worked out. Last week, Micron finally announced that it will be selling a version of its Samurai DDR enabled chipset to work with the AMD processors. Sources said that apparently that Micron has obtained the proper license through alliance with a major foundry partner.
AMD had also shown off their 750 DDR Enabled chipset which will work with both registered and un-buffered DDR modules.
The DDR enabled chipsets made by VIA Technologies, the Taiwan fabless manufacturer, will be working with both the Intel and the AMD processor. VIA has already demonstrated in its PC133 chipsets that it can work with both camps of processors without their endorsement. VIA reviewed at last week’s Summit that they are on the way to ship 50% of market requirement on chipset. This message had caught processor manufacturers in surprise. You see that microprocessor manufacturers have no definite knowledge of how many of their microprocessor is required. Instead, only the chipset people can tell that how many of which kind of motherboard will be produced through the number of chipsets delivered.
Well, It Is Largely A Marketing Game Also
Look at swelled stock values
If anybody is asked about marketing and PR in the hi-tech world, you can bat on Rambus and Intel among the top spenders. You can see that Rambus stock price has gone from the $20’s to the $200. Intel’s book value has swelled through the high-tech venture capital investments instead of from their operational profit.
Big advertising dollar brings heavy influence
Big companies that have multi-million dollar media budget can usually get publishers to go their way. For the last year, the PC user public heard the Rambus memory message very loudly. Some even became Rambus share holders. On the other hands, Intel had put their money on the Intel Developer Forum (IDF) twice a year to boast their memory roadmap to their distributors and resellers.
Last week’s DDR Summit was partially sponsored by CMP publications including Electronic Business News and EE Times. As Steve Cholas puts it “The job of the press is to tell the truth as it is. We have the obligation to report on both DDR technology just as much as on Rambus technology.”
When a company owns the memory technology license, the chipset manufacturing and the microprocessor supply, it can easily jiggle around the component prices to make the memory price look attractive at the penalty of the other components. News circling last week indicated that several DRAM manufacturers were asked to reduce their price on Rambus memory chips and to increase supply quantities.
DDR marketing campaign had just started
If last week’s DDR Summit was a marketing kick off, they sure did a very good job. When Todd Rovazzine of Silicon Tech started this idea, he had some difficulty to recruit the first few sponsors. After he got Micron Technologies , CMP Publications and DDRtester.Com as premiere sponsors, the snow ball just started rolling. The industry was turned-on. Not only major players wanted to get in, but also other second ti’er companies were also flocking to jump onto the boat too. As Todd said “ All at a sudden, they realized that DDR is on the way to success and they wanted to be part of success.”
Anyway, the crowd had filled the room at the meeting site. Vendors were racing to show off their DDR related products at the mini-exhibit. The Webcast was so attractive that late comers had ran into bandwidth problem and had audio dropped off.
At the DDR Summit, all major chipset companies announced working prototypes. They are Micron, AMD, VIA and Ali. DDR supporting motherboard manufacturers included all the major players in Taiwan and China. That could represent over 70% of the motherboard production in the world.
Major DDR marketing campaign is yet to come
Seeing the willingness of the players in the DDR industry to allied together, more marketing activities will definitely be planned in the future. AMII (memory industry’s marketing and infrastructure building arm) is taking the center stage to co-ordinate DDR marketing activities. Industry suppliers will be united together to put up major promotional programs. Future marketing target will target more towards educating end-users and local dealers. Do look forward for creative programs that will blow your socks.
Price, Price and Price
“In my 20 years of dealing with memories, there came many creative technologies” said Desi Rhoden, chairman of JEDEC and president of AMII. “But when the consumer has to make the decision, the lowest price with reasonable performance had always won.” “If you look at the evolution path of DRAM, EDO, PC66, PC100 and PC133, the change over did not occur until the new memory had reached price parity,”
Today, we are looking at DDR memory has all the possibility to reach price parity in the near future. Semico (a research firm) has projected DDR cost parity by mid-2001. We are also seeing that Rambus memory not likely to be cost competitive in the foreseeable future. Indeed, Rambus memory might never reach price parity due to the royalty, die size, high frequency test and the heat dissipation.
820 chipset were made at the PIII lines with 0.18uM process
This just do not make sense. Why would a company dedicate its precious high-end fab that produces $700 Patium III microprocessors to make a chipset that can only fetch $35.00? Particularly at a time when some PIII’s are in allocation?
Infrastructures all in place
Desi Rhoden declared that all DDR infrastructures are in place now. For the DRAM manufacturers, there is no new processing equipment to purchase. Most of the first ti’er manufacturers has already converted their memory dies to accommodate both SDR and DDR chips. The DDR enabled chipsets are almost ready for production. There are no special setup for motherboard manufacturers. The memory module common Gerbers are now available. The memory module manufacturers require no new equipment to assemble the modules. All these will keep the price down at the transition period.
Considering all the facts and trends, I predict DDR enabled motherboards will come into the market in early Q4, 2000. It will quickly spread like a wild fire. Aggressive marketing will lead to early adoption from the curious users that want to reward themselves with a high tech Christmas gift.
Once the systems are proven by the pioneer users, the corporate user will start to adopt in the first half of 2001. DDR based servers stations will sink into the market place by mid 2001. By year-end 2001, DDR will be at an attractive price for the value end users. The year of 2002 will see a mass conversion from SDRAM to DDR memory.
On the other hand, I predict Rambus will retreat by mid 2001. Intel will finally yield to the customer’s choice. Other microprocessor companies will have bigger share of the market. New revolutionary processors will take up large part of the Net Appliance market and that consumers will benefit from the high performance of DDR in their PC’s and Appliances.
By: Cecil Ho/ CST Marketing
Copyright © 2000 CST, Inc. All Rights Reserved