by Anna Filatova
04/12/2003 | 11:16 PM
The leader remains unchanged. Samsung, which retained the leading position in the world’s memory market since the end of last year, now announced that they begin the mass production of 1GB DDR II memory modules built of 512Mbit chips with 533MHz frequency, that is, complying with the PC4300 specification. We only wonder what the word “mass” means here? And where can you use these modules in mass volumes now? Anyway, a number of manufacturers are going to roll out products supporting DDR II this year, so we can’t but be happy about Samsung’s willingness to rush ahead. Moreover, as Samsung claims they can raise the frequency to 667MHz (PC5400) with absolutely no or very little effort.
Meanwhile, the number two racer, Micron, running notably behind, prefers to make up for it with higher capacities. Tin March the company announced the first shipments of 4GB registered PC2100 DIMM modules and also showed the first samples of 0.11micron 512Mbit DDR400 chips, which, unlike the modules, can become a potential hit of the year. Not now, of course, but by the time and annual results are announced. We hope that the 0.11micron process, into which Micron put so much money and effort, will help the company out of its present troubles: according to the recent quarterly report they suffered quite a big loss of $619 million.
Otherwise, Micron can follow the fate of its might-have-been purchase, currently being hunted from all sides. In the end of March we learned that the European Commission found Hynix guilty of dumping and set a 35% protective duties for its products. It’s impossible to be competitive in the today’s market with such a burden on your shoulders, so Hynix leaves the European market in April. Well, there are two more big places: Asia and the USA. But in the latter a similar suit is underway. So, the company has prepared a kind of safety net by investing extra $100 million into its Oregon fab to transfer it to the 0.13micron process. No custom duties can act against goods produced directly on place, by the definition.
Still, Hynix is going down, greeting Nanya that is coming from below and at a much quicker pace. The Taiwanese Nanya has already entered the Top Five of world DRAM manufacturers, ousting such an experienced player as Elpida. And it doesn’t seem to be slowing down at all. Nanya is going to come over to the 0.11micron manufacturing process by the end of this year, not in early 2004, as it has been planned before. And this process is going to become very (vitally!) important in the near future.
Nanya is also surprisingly fast in reacting to what is happening in the world. Some time ago, it pioneered the transition to DDR SDRAM, and now it … involves a part of its production facilities into manufacturing of SDRAM that has become more expensive today. It doesn’t spurn off new products, though: NVIDIA has contracted Nanya to ship graphics memory chips. This means that Nanya can compete at equal terms with the traditional NVIDIA’s supplier, Samsung. Well, what else could you imagine considering their huge investments into business and R&D.
More to the point: Nanya said that co-operation with Infineon will allow spending more on research & development and it’s going to be ready for 0.09micron technology in 2004. Infineon has been a sort of throwing out its technological potential to all who are willing to catch: the Chinese SMIC licensed the 0.11micron process and also technologies to bake those 300mm wafers. That’s good for SMIC as they are now building a 300mm fab.
In fact, Infineon has a lot to share on the matter. The company at least retains the technological parity, having announced the first samples of 512Mbit DDR II chips supporting 533MHz and 667MHz frequencies. This is definitely not the “mass-produced” DDR II modules from Samsung, but still we are sure by the time there is mass demand, Infineon with partners will have got ready to meet it.
Right now, the company is seriously considering the move to a new location. Switzerland, Singapore and the USA are possible places for Infineon to land. The company has representative offices in the latter two countries, while they consider Switzerland in case Infineon decides to remain European, but save on taxes, which seem to be really high in Germany today.
Unfortunately, legal costs are legal costs in any country and the legal battle between Infineon and Rambus has gone too far. Some international standardization bodies interfered into the process trying to speed it up a little. While courts of various instances speak out contrary opinions, the fate of patents that belong to Rambus and lie at the foundation of a number of present-day and future standards remains obscure. Overall, the new parties involved took Infineon’s side. As well as the US judge James Timoni who accused Rambus of unpremeditated evidence concealment (he said they had destroyed all internal documentation). Not a big thing, actually, but still no pleasure for Rambus.
As for pleasures, Rambus should turn to Japan where its long-time partner, Elpida, bought the Yellowstone license. We can take a guess what they might need it for. Elpida did quite well in graphics memory area lately. In March they announced their first DDR II chip in GDDR2-M variant. Of course, Yellowstone doesn’t belong there yet, but it might be a nice idea to combine the two technologies.
(On the other hand, there are problems with GDDR2 as all manufacturers try to push their own version of the standard. It may be that the industry passes by this phase and goes directly to GDDR3, which is approved unanimously and is now drafted by JEDEC. Still, this doesn’t have much to do with GDDR2-M as this graphics memory for mobile devices is a private matter of ATI and Elpida and they won’t look back at GDDR2).
Elpida has just one simple task ahead: maintain normal sales volumes. In the last year, the company sales volumes dropped by 30%, from $874 to $615 million. It should be better this year thanks to the integration with Mitsubishi that sold DRAM chips for $362 million last year; the collaboration with Powerchip, which is now governed by Elpida, not Mitsubishi, and sells to Elpida half of its products (and Powerchip is planning to double the total output in 2003). Summing it all up, we think Elpida can win back its fifth place by the time the annual results are summed up.
One more thing to be mentioned: the agreement with Kingston to become an OEM memory modules maker for Elpida since Q2. It’s quite possible that the modules will be made of Powerchip’s chips. Elpida will only have to stick its brand on and try to sell the end product.
Kingston, meanwhile, extends its “extreme” HyperX module series by introducing surprisingly traditional PC3200 models and improving latencies in the currently available PC3000 and PC3500.
The competitors were better in March, however. TwinMOS made a 512MB PC3700 module, while GeIL rolled out paired 500MHz PC4000 modules for dual-channel configurations. That’s extreme as hell. Kingmax brought some merriment into the house showing modules made of multi-colored chips. They are aesthetes, aren’t they? Although they say it’s just to fight fakes that imitate brand modules like ValueRAM and JetRAM.
But it is another well-known memory overclocker, OCZ Technology, that was the most spectacular of all in March. Not wasting their time on trifles, they announced memory modules with up to 6.4GB/s bandwidth. 800MHz chips? No, the chips are ordinary, 400MHz ones. OCZ is just the first memory modules maker to use the QBM technology from Kentron that allows doubling the effective memory bus bandwidth. A tiny and low-cost controller built into the module makes a 4.2GB/s module from 266MHz chips, a 5.4GB/s module – from 333MHz chips and 6.4GB/s – from 400MHz chips.
These are numbers for standard configurations, in case of dual-channel ones we should multiply them by two. And we are going to get to this arithmetic quite soon. Upcoming VIA KT600 and PT600 will support single- and dual-channel configurations of QBM modules. Bravo, Kentron! To tell you the truth, we were quite skeptical about anybody going to produce such modules.
For more information about QBM technology from Kentron please see our article called “Quad Band Memory Technology from Kentron: Shorter Way to Higher Memory Bandwidth?”.
That’s about all there was in the market in March. There is only one topic left: prices. They stopped in their downward motion, and that’s it. Prices were at the same level throughout the month; this specifically concerns contract prices. Anyway, we guess memory prices won’t go up in the next two or three months.
Moreover, the manufacturers’ assortment policy suggests that we can get ready for further DDR SDRAM price reduction. From this quarter on, all the majors are going to increase PC2700-PC3200 chips share in their production volumes. This will bring the price of these memory types down, and it will definitely affect the price for PC2100 modules and chips. Overall, it’s going to go rather down than up.