Memory Market Overview: August 2005

Summer is traditionally considered to be a slow season, but not this year and not in the memory market. Exciting details from the "Rambus vs. the rest of world" story, the end of the MHz race in the memory modules market, unexpected price growth in the memory chip market, and more in X-bit’s monthly coverage.

by Anna Filatova
08/02/2005 | 04:42 PM

Well, really hot summer weather has set in and the manufacturers as well as editors prefer to cool down in the next-door swimming-pool instead of working on the new products :) This is certainly not the reason why I didn’t right my monthly memory market overview last month (I really wish it had been the reason though :) ) Anyway, the middle of summer is usually not very rich in events, although this year doesn’t seem to be falling within this rule. If we look back at what was going on in the memory market in June and July, we will see that the memory makers didn’t even think of slowing down for the so-called “dead season”. Well, this is definitely for good.


It has already become a good tradition that the first news we report comes from Rambus. There are two major topics that can be discussed in connection to this company: the arguments with other industry members and RDRAM II aka XDR. And again Rambus managed to generate news in both of those. Firstly, they have suddenly quarreled with the company that has been almost their primary ally: Samsung has been the one supporting almost all Rambus’ initiatives lately. However, despite common gratitude Samsung was accused of violating 24 Rambus patents dealing with SDRAM, DDR/DDR2, GDDR2/GDDR3.

The response from Samsung followed almost immediately: the company filed a claim against Rambus in just a few days with practically the same accusations as Infineon, Micron and others have already resorted to. Namely, they accused Rambus of acquiring their patents illegally. However judging by the previous experience I can say that the only thing Samsung needs to win this fight is good luck. The same would also be helpful to the European Union, which seems to be paying a lot of interest to Rambus’ actions in the patent field.

Well, it looks like friendship with Samsung has come to its logical end. It’s high time Rambus started looking for new allies. The potential new victim could be IBM, which licensed XDR architecture interface from Rambus in the middle of June. This interface, XIO, is a pretty promising and interesting solution. By the way, in the middle of July they officially announced XDR2 the notorious micro-threaded architecture that makes the memory core perform according to the XIO bandwidth potential. Rambus promises to release the XDR2 chips into the market some time around 2007.

In fact it is quite evident that the next PC memory standard is going to be DDR3, especially since it is almost ready to kick in. in the middle of June already Infineon send to Intel its new 1067MHz memory modules based on DDR3 chips of their own design. The only thing we know so far about these modules is that they are still quite raw, however, this is definitely not an issue as the first samples are to start shipping only in the beginning of 2006, so Infineon has enough time to polish their products off. In fact, this should only be regarded as another R&D step. By the way, Samsung introduced 512Mbit DDR3 chips in February already, if you remember.

As for Samsung, they have been focusing on the current memory products this summer, because it is the No.1 force driving the company’s revenues and losses. In June the company started manufacturing 90nm 1Gbit DDR and DDR2 chips having once again proven its technological leadership in the memory market. In July Samsung started mass production of DDR20667 chips. Of course, these chips were also produced with 90nm technology and were available in 256Mbit and 512Mbit versions, which were the most popular. These chips were intended for the i945 based platform in the first place, however, there are some rumors that i945 will hardly be as highly demanded as it was expected to be. So, Samsung has to be a little bit more careful: the industry is overall pretty skeptical about the prospects of DDR2 for this year 2005.

Elpida certainly responded to the beginning of mass production of 90nm 1Gbit chip, although in a quite prospective manner. Namely they announced that they had just completed the development of 2Gbit 80nm DDr2 chips. The matter is though, that they will only start producing in spring 2006 at the earliest, when Samsung will undoubtedly roll out something as well. I would also like to say that although Micron started shipping first samples of 2Gbit DDR2 chips in the end of June, they claim the mass shipments to be scheduled for an even later time than those of Elpida: for Q4 2006.

All other big fishes in the memory business didn’t introduce anything new on the technology front this time: most of them were busy reorganizing their finances and production capacities. For example, Hynix is planning to pay back its 2-billion dollar stabilization loan this summer, which immediately caused tremendous protest from FTC and European Union, Hynix’ main competitors. This scandal about the state support policies still echoes in the industry: in the very end of June the world Trade Organization called back its previous decision and claimed that the sanctions applied to Hynix chips exported to the USA had been fair and correct.

The only way-out in this situation is to strengthen partnerships with other companies and to sell the chips under their brand name. the rumors say that the partnership of Hynix and STMicro has become so close, that the latter is already thinking of buying about 10% of Hynix’ stock in their joint fab in mainland China. By the way, since the memory production keeps growing, Hynix started to pay more and more attention to the chips testing facilities and increases their capacities in this field. As a result, this may have a highly positive effect on the market: the shortage will grow lower and the prices of independent companies offering memory testing services will drop down significantly.

The reduction of the memory chips production cost is a very acute topic, especially taking into account the results of Q2 2005 showing that the financial situation of most memory manufacturers got noticeably worse. Micron, for instance, lost 128 million dollars in Q2 this year. As a result, the company got negative balance for the first time during the last 1.5 years.

However, let’s return to the expansion of production capacities. The Taiwanese seem to remain traditionally active here. In summer both cut-throat competitors, ProMOS and Powerchip, announced putting in operation their second 300mm fabs. Even though with a little delay, but ProMOS started pilot production of the memory chips. So far the yields are only 60% for the 90nm process (although this is not that bad for the first pilot run, I should say), and it looks like it will take ProMOS until the end of the year to reach the 5K wafer per month production volume.

In the beginning of June, also a little later than it had been initially planned, Powerchip started pilot production on its Fab12B. I have to stress that from the technological prospective, Powerchipp is slightly behind its competitor, because they will start making 90nm chips only in the end of this year. However, even though they haven’t yet completely mastered the 90nm process licensed from Elpida, the company is already in the middle of negotiations about licensing the next-generation 80nm and 75nm production technologies.

And in the meanwhile, Inotera Memories made the following forecast: taking into account the number of wafers each fab produces and the chip yields the joint venture of Infineon and Nanya should manufacture more chips in August than Powerchip. Especially, since this efficient alliance is ahead of Powerchip already in terms of production technology, as they started mass production of 90nm chips from 300mm wafers in July already. Of course, this fact only allows them to outperform competitors: their chip yields for a wafer like that are even higher than the 100nm chip yields. By the way, Nanya is going to start building its own 300mm fab only in 2006, so it will be the last Taiwanese memory manufacturer to do it.

Well, little by little I came to the discussion of two classical topics, which I haven’t yet touched upon in this article: the memory modules and the memory prices. As for the modules, the situation is fairly simple here: OCZ rules here as always rolling out 2-3 new solutions a month. June and July were no exceptions. The remarkable thing for this period of time is that they didn’t set any new frequency records, moreover, the maximum working frequency of the new OCZ memory modules was the frequency of “only” 600MHz.

But even despite this fact, the newly released products are quite impressive as they are. OCZ announced PC-3500 Gold GX - DDR memory modules for extreme gamers with 2-2-2-5 timings; less expensive PC-3200 Gold Edition C2 – with 3-2-2-6 timings; and Value Pro product line – DDR2 PC2-4200 memory modules without any specific outstanding features (besides the traditional heat-spreader, life warranty, etc) but with a good price. In other words, I got the impression that the hunt for higher MHz has slowed down here as well, just as in the CPU sector, and the memory manufacturers started paying more attention to such eternal values as performance, price, economical factor, etc. And even the release of PC-4800 Platinum Elite Edition with 600MHz frequency and 2.5-4-4-10 timings doesn’t break this tendency. It not the 1GHz memory, anyway...

GeIL has also returned to the roots of things. They announced PC-3200 product line from GeIL ONE series featuring 1.5-2-2-5 timings and different overclocking potential: up to 2.5-4-4-7 at 600MHz for TCCD chips from Samsung and up to 2-2-2-5 at 500MHz for BH chips from Winbond. Taking into account that OCZ PC-4800 Platinum Elite Edition is also made from TCCD chips and this fact is specifically stressed by the manufacturer, I have every reason to claim that there is a general tendency to use the chip brand names and specifications for better promotion of their products, especially if these chips are known for their great overclockability.

Against the background of pretty conservative frequencies from OCZ and GeIL, Transcend looked really extreme with their new 800MHz DDR2 256MB and 512MB memory modules, which are not officially supported by any chipset out there. Nevertheless, the company targets these new products not only for the mainstream computer market, but also for servers and workstations. To tell the truth I more approve of what OCZ and GeIL are doing: Transcend should better optimize the memory modules performance for standard working frequencies. Just my two cents.

The same is probably true about Corsair: in the end of June the company launched their 1GHz memory modules. A two-module kit (2x512MB) was selling in Japanese stores for $370. Theoretically, this is not a too crazy price, especially compared with the corresponding CPU or graphics accelerator...

By the way, since we came to speak about prices... Summing up the results of the past two months we will hardly see the usual even curve going from the upper left to the lower right corner. In fact, in June nothing indicated any significant changes. Yes, the prices started off the $2.3 for 256Mbit DDR400 (because there was simply no more room for further drop), rose to $2.45, returned back to where they started and ended up at $2.4 per chip in the end of the month.

So, $2.45 for 256Mbit chip turned into the resistance level (in terms stock trading). According to the economical definitions of this phenomenon, the prices should either hit against this level and roll back if they cannot beat it, or if they hit hard enough, they will make a significant move forward. This is exactly what happened in July. The moment Samsung and Hynix (and Powerchip, too) had slightly reduced the shipments to the mass market, the stability was completely gone. They may have reduced the shipments for several reasons: maybe because they were reorganizing production capacities for flash memory products, maybe because they wanted to manipulate the market this way, maybe both. The result was outstanding: only in the first decade of July the prices sky-rocketed by over 9% to $2.64 per chip.

In the second half of the month the mainboard manufacturers started reporting unexpectedly high sales volumes in June, thus causing the prices to grow even more (as there was hope that the demand will keep growing). This was the second upsurge, a smaller one, but still: the price reached $2.73 (so that the overall price growth since the beginning of the month reached 12.8%). By the end of the month the prices got slightly corrected, so that we finished July at $2.68 per 256Mbit chip. But this is definitely none other but correction preceding another significant jump up: the memory makers keep transferring more of their production lines to flash divisions, and the demand seems really to be improving.

Well, I will keep monitoring the situation for you, so stay tuned for the next overview in about a month!