DDR SDRAM  1. Double data rate synchronous dynamic random access memory. 2. A trailing-edge technology which has stepped into the winner's circle. 3. Strike one against Intel as in "three strikes and you are out."

DDR memory and its niche in PC history is a fascinating tale. It can't be told without involving the story of RAMBUS and its legal and technical woes. Most importantly, it is the story of a huge misstep by Intel, and perhaps the first clean sign that its time is past. Poor Wintel, who would have dreamed it would be brought down by a raggedy bunch of Internet hackers and an upstart chip maker.

Let's start by taking a look at what makes DDR special. Tweakers.com.au has a nice explanation on the page listed in Internet Resources below the story. The simple explanation for non-engineers like me is that DDR uses both the leading edge and the trailing edge of the signal carrying data. Regular SDRAM uses only the leading edge. Thus DDR doubles the effective bandwidth for data moving between itself and the CPU.

DDR is good for consumers because unlike the technology pushed by Intel the past few years, it is an open technology. Unlike RAMBUS, where everyone who made it had to pay a licensing fee in order to use it. Everyone being the consumer, in the end. DDR is also good for allowing me to make a bad pun by referring to it as trailing-edge technology. Now that Intel has given up on RAMBUS and joined the DDR parade, its life-expectancy has doubled as well. And thereby hangs a tale..

It wasn't supposed to happen this way, you know. RAMBUS has long been talked about as the solution to a problem that's been seen coming for as long as people have talked about Moore's Law. You know, the old saw about the number of transistors on a chip doubling every couple of years. In CPU terms, that also meant a doubling of the speed at which it could operate. People have known for a long time that CPUs were on a road to becoming data-starved as they got faster faster than the speed of data moving to and from them increased.

RAMBUS was the promised memory. It had sizzle, mainly because of the way it was described. Instead of being described with the standard metric which measures how quickly it can be accessed, it was described in terms of the speed of the path between the memory chip and the CPU. 500MHz sounds oh so much faster and more impressive, doesn't it?

Here is the way a newsgroup post described RAMBUS in 1993:

"This technology has been developed by the RAMBUS company. Making wide memory busses, which several chips provide various bits of data can be hard to make fast. RAMBUS solves this problem by using a small, 9 bit (data) bus with only a few control lines which runs _very_ fast (500MHz). The bus also runs at a much lower voltage swing to allow for this incredible speed. Since only one chip provides the data that is on the bus at any one time, this can be done if you are careful. The bus is used to transmit both the data, and the addresses, as well as a few other messages."

You can't blame consumers for failing to note the impact of squeezing the size of the data pipe down to a mere 8-bits. How could marketing types exploit them if they did? But you would expect more of professional chip makers like Intel. Unfortunately, the significance of that didn't seem to dawn on Intel. They invested in RAMBUS and promised to deliver RAMBUS technology starting with the Pentium III. Like a cobra being charmed, Intel saw nothing except that almost meaningless 500MHz descriptor.

As the years passed, other chinks in the RAMBUS armor began to appear. And to be noted in newsgroups. Like in this passage from a 1995 posting:

"I think the real problem is you must license the use of RAMBUS and you must implement a full custom ASIC controller with a 500MHz RAMBUS interface. This isn't easy and it won't be cheap, so don't expect motherboard manufacturers to put them in any time soon..."

So the first signs of the high prices and the technical challenges that would generate glitches for RAMBUS were already being seen by some in 1995, long before Intel decided to jettison them. There was yet to appear another massive problem for RAMBUS. A legal one. But that was a couple of years away, and RAMBUS was still on the ascendancy. In 1997, ComputerWorld picked them among the top 100 hot emerging tech companies, describing their allure as "High-speed interface technology to speed data transfer to 600M byte/sec. over narrow byte-wide data bus."

In 1998 RAMBUS was still on the rise. Announcements of its planned use abounded. Compaq, Intel, and the first of the Taiwanese manufacturers were all onboard as this clip from another ComputerWorld story shows: "In May, Vanguard International Semiconductor Corp. became the first Taiwan-based chip maker to license Direct RAMBUS for use in its 128M-bit DRAM devices. Analysts said at the time that the move could spur other Taiwanese vendors to follow suit. Intel Corp. has also given the technology its blessing, and will introduce it next year year in high-end PCs."

In 1999, the wheels began to come off the RAMBUS. As a wag noted on a newsgroup, "RAMBUS DRAM has been highly controversial, as many of you already know. It is a proprietary technology and is therefore expected to cost more than other memory types. Its technical merits, compared with competing technologies such as DDR SDRAM (Double Data Rate SDRAM) and SLDRAM (SyncLink DRAM), have been called into question, since RAMBUS DRAM essentially achieves high clock speeds by narrowing the data bus and increasing memory latency."

Then the first delay was announced by RAMBUS, a delay which also forced Intel to delay the introduction of its 820 chip set. In a PC World story circa September, 1999, it was reported that on the very eve of its promised date "Intel acknowledged memory problems associated with the RAMBUS memory design in its 820 chip set. Intel reportedly found a "memory-bit error" that curtails top memory capacity and speed."

Remember the old Ricky Lee Jones song "Easy Money"? One of the lyrics goes "One step up, one step back, one loosen her shoulder strap." So it went with chip sets for the Pentium III. Intel stepped back, VIA Technologies stepped up. As reported in that same PC World story, "Micron has already made that move -- its next-generation computer platform won't be based on Intel's upcoming 820 system chip set, but rather a VIA Technologies alternative. The move, Micron says, will save customers between $200 to $300 over systems with the 820 chip set, without sacrificing system performance."

What was it Ross Perot had to say about General Motors? Oh, yeah. He said "Revitalizing General Motors is like teaching an elephant to tap dance. You find the sensitive spots and start poking."

Old BASIC programmers may think I am POKE-ing fun at Intel's memory problems, but it really did take a lot of poking to turn the Intel elephant around. By 1999, Intel may not have gotten it yet but consumers were becoming more and more aware that buying RAMBUS and paying its licensing premium was very much like buying a pig in a poke.

A newsgroup post by Jonathan Epstein in November of 1999 summed it up this way: "A Coppermine Pentium III with a 133 Mhz Front Side Bus (and using 133 Mhz SDRAM) will perform a bit better yet. The more expensive RAMBUS memory, on the other hand, will perform (relatively) much slower than a 133 Mhz SDRAM system, and only slightly better than a 100 Mhz SDRAM system. This is because of RAMBUS's high latency."

In the fall of 2000, IDG news reported that Intel CEO Craig Barrett was trying to quiet rumors that Intel was moving to get off the RAMBUS. They quoted him as saying "Intel Corp. remains committed to introducing its forthcoming Pentium 4 processor with RAMBUS Inc.'s high-speed memory interface technology, which Intel believes is the right choice for high-end desktop PCs." But movement to DDR was already in the works.

By 2001 the legal quagmire was hindering RAMBUS as much as the technical glitches. Various members of JEDEC, and the FTC itself, have sued RAMBUS for hiding their patents on supposedly open technology jointly developed by JEDEC and RAMBUS, then forcing their technology partners to cough up major licensing bucks for their trouble. As Eric J. Sinrod noted in an article called "The Cost of Suing," going to court can be a draining experience. Sinrod noted "RAMBUS, a chip design company in Los Altos, Calif., recently reported earnings of $8.2 million for its second fiscal quarter, excluding one-time charges. Yet RAMBUS' legal fees during this same period were $7.3 million, stemming largely from litigation in different intellectual property cases involving Infineon Technologies, Micron Technology and Hyundai. Incredibly, RAMBUS' legal fees almost equaled its earnings during its second quarter."

Intel got into court itself. Having been outfoxed in the marketplace by backing the wrong technology, they sued VIA Technology for offering a chip set that provides DDR access for Pentium IVs, claiming that it violates their patent rights. RAMBUS has since prevailed in the JEDEC suit, or is at least for the moment prevailing, but then there is a pesky suit by the FTC on the same grounds to deal with as well. Win or lose in court, they have definitely lost in the real world. Even Intel is providing DDR memory support in its chip sets now. They've done so since early 2002 when they launched their 845 chip set.

What this giant whooshing sound over the past decade really represents is a mighty swing of Intel's bat completely missing the pitch. It's strike one against Intel. AMD has grown stronger and it hasn't made the same sort of disastrous error. Not yet, at least. Strike 2? Well, that looks like Intel's 64-bit processor while AMD has stroked another homerun with theirs. We'll be taking a look at those in a coming issue.

Internet Sources and Resources:

DDR Memory Performance Roundup

JEDEC Suit Against RAMBUS Dismissed

FTC charges RAMBUS with anticompetitive acts

Moore's Law

100 Hot emerging companies 1997

Compaq/RAMBUS deal aims to boost AlphaServers

Intel delays launch of high-end chip set

The Cost of Suing