by Anna Filatova
09/13/2005 | 12:11 PM
August appeared this year’s quietest and sleepiest month from all viewpoints. The same is true for the memory market. There are not so many pieces of news worth mentioning, mostly we see the already existing tendencies strengthen their positions. For example, this is true for full-size server memory modules, which have been manufactured only in FB-DIMM standard by both: Elpida and Micron.
The remarkable thing is that both these companies released their 4GB memory module samples based on 1Gbit DDR2-533 chips, which turn into the most optimal solutions today, especially taking into account new more memory-hungry chipsets. At least this is true for the memory makers.
By the way, we are talking about engineering samples for their major customers here. As for the large-scale shipments of these memory solutions, Elpida and Micron will start shipping those in the end of this year at the earliest. At that time the FB-DIMM specification should finally be completely finalized.
However, as usual, there is a relatively different approach to memory modules design and development. These memory modules may be of not such big capacity and not so advanced, but quite tiny and exactly suitable for today’s very popular blade-server solutions. This exciting direction didn’t remain unnoticed in August, too. While SMART Modular and A-Data “only” launched their SO-DIMM modules on DDR2-667 chips, Micron introduced their solutions based on DDR2 chips with up to 800MHz frequency in the second half of the month.
These were the modules designed in MiniDIMM VLP (Very Low Profile) format, which were only 18.3mm high. However, Micron promised to start mass production of these solutions in the end of the year, just like A-Data Company.
However, SMART Modular solutions also do have their attraction, just like the solutions from Micron. This attraction is the memory bus width that has been reduced to 16 bit only. To tell the truth, it looks pretty strange, especially for DDR2 memory modules, and especially since they are designed with 144 pins. However, this is where the answer to our questions actually comes from: these memory modules have been designed not for computer systems, but for such peripheral electronic devices as printers, etc. As you may know, devices like that use 16bit interface and 144-pin connectors, so no additional converters will be required.
By the way, one of the devices like that has become the “best of show” in August: I am talking about the Gigabyte solution here. Namely, Gigabyte started to supply some of their mainboards with the special converters that allow using DDR memory modules in the DDR2 memory DIMM slots, so that the user acquired some kind of backward compatibility with the older memory modules. The fly in this ointment is the fact that these are mainboards designed on VIA PT880 Pro chipset, which specific features seem to have been involved here.
Anyway, this solution is targeted for those users who intend to continue using their previously purchased memory modules. As for purchasing DDR memory today, I doubt that it will be even considered. DDR2 already looks very attractive from the price prospective, due to the fact that there are more better-value chips and memory modules streaming into the market every day and tending to go far beyond all possible and impossible standards. In the beginning of August Crucial announced that they started selling their 1GHz Ballistix PC2-8000 memory modules (you can read our detailed review of this memory solution here: Crucial Ballistix PC2-8000 Memory Modules Review ). As you remember if you read our article, the timings of these memory modules are 5-5-5-15 at 2.2V voltage.
In the memory chips market a few interesting things have shaped up. According to the Q2 2005 results, Hynix regained its second position in the ranks of the world memory market, however now they seem to have faced some difficulties with the continuous growth of its production volumes. It looks like they will have to resort to help from third parties, even though they tried to reduce third-party participation to minimum, because with the shift to 90nm production process Hynix testing capacities are no longer sufficient and so is ProMOS’ success in mastering 90nm technology.
Infineon may also have some problems with further growth: their partners, such as Winbond and SMIC, transfer more and more of their capacities to flash memory and other niche products. Even Nanya has been getting much more memory orders from HP and Dell lately, so they will hardly have time to ship anything to the Germans now. Although, I don’t think Infineon should be worried about it: some time ago they started re-conservation of their Richmond fab, which will be right in time these days.
Anyway, it is very interesting to watch what’s going on: the market is growing so fast that the manufacturers can hardly handle it, mostly because of the insufficient testing capacities. Although they are really doing their best, and this is the second interesting observation this month. Within this tendency, Elpida and some other active players of the testing services market, such as Kingston, Powertech and Advantest, announced their new joint venture called Tera probe, which should become one of the largest players here. Looks like Elpida also suffers from significant shortage of wafer testing capacities.
Especially, since their satellite, Powerchip, that set into operation their second 300mm fab in the beginning of this summer, announced their intention to start building two more fabs like that! And they are going to start the construction in early next year already. Keep in mind that this is besides their fab construction in mainland China. Well, looks like they intend to make sure that the infrastructure is expanded enough in advance.
This is what I have to say about the production capacities. As for the architectures, I have to mention Elpida’s announcement in the end of August that they have completed the development of DDR3 SDRAM, following in the footsteps of Samsung and Infineon. Their 1333MHz 512Mbit chip samples should be should be shipped to the customers by the end of the year. Note that this is still happening even though JEDEC is planning to finalize and approve the DDR3 standard (which is expected to be 1067MHz only, by the way) only by 2007. It looks like the industry will be ready for transition to DDR3 in record-breaking short period of time during the entire RAM history. The only question that still remains unanswered is whether the users will be ready to accept DDR3 by 2007 and whether the demand will be there at all…
You have a life example in front of you: DDR2 has come into this market a long time ago, but it is moving forward not as fast as we have expected. Take, for instance, Samsung. They managed to go beyond their key threshold only last month: the share of DDR2 products has finally exceeded the share of DDR1 products. ProMOS is not going to be that far before the beginning of next year. I would suggest taking the DRAM Exchange estimates as a reference here: they predicted that in H1 2005 the DDR2 share was only 30% of the entire DRAM market.
Well, it looks like these are all the events for this moment. I only have to say a few words about the prices. The situation with the prices appeared no exception in the entire lazy August flow. There were no surprises here: the prices steadily and gradually dropped from $2.68 down to $2.46 for a 256Mbit DDR400 chip. By the way, speaking about the price difference between the DDR and DDR2 memory modules in August, I should say that it was only 2-5% and kept growing smaller, so that it should completely disappear in September, as experts say.
It is quite hard too make any definite forecasts these days. On the one hand, the manufacturers keep expanding their production capacities very actively, as I have already said. On the other hand, the growing demand during the back-to-school season in September will be able to eat up most of volumes. And then there is Christmas holiday season right around the corner. Nevertheless, there will hardly be any clear tendency in September. As for October-November time frame, I wouldn’t make any definite forecasts at this point.