Memory Market Overview: March 2005

New product announcements, new memory technologies (such as ovonic unified memory), new record in terms of memory capacity and designs, and of course, our traditional prices analysis, this what we are going to talk about today in our monthly Memory Market Coverage.

by Anna Filatova
03/18/2005 | 06:51 PM

It is simply unbelievable how many news last month were connected not with the memory chips or memory technologies, but with the memory modules, which turned out pretty exotic in most cases. It can indirectly signify that the crises in the memory market grew somewhat worse this month, as there are more and more chips of bigger memory capacity produced and there is actually no demand for such great memory amounts. The major memory consumers, Windows and Office, haven’t got any newer version updates for a few years already, so their appetite for memory hasn’t grown any bigger since then.

 

As a result, there appeared a number of companies, which business model is based on offering small memory module shipments but at a larger margin, because of the “exclusive” memory module design or features. The indisputable benefits of this strategy as well as the not very promising prospects in the regular memory module market pushed some large companies, like Kingston, for instance, to start moving in the same direction. As a result, we got a pretty paradox situation: when the majority of the news in the memory market is generated by the solutions, which total share in the memory sales is relatively small.

One of the leaders of this market segment is certainly OCZ Technologies. And this time we also heard a lot from them. Last month started with the announcement of the new DDR2 product line within the Enhanced Bandwidth Platinum Edition (Platinum EB) family. These were the modules working at 667MHz frequency and boasting pretty good timing settings: 4-2-2-8. And I have to admit that this was a very pleasing fact, as the practical latencies of the DDR2 memory have always been its weak spot compared to DDR.

However, there is still room for improvement in the DDR segment as well, and OCZ is being very active here, too, especially in the end of the month. They released their PC-4000 Gold VX series from the Voltage eXtreme product family, which allowed reducing the memory timings to 2-2-2-8 by raising the voltage to 3.3V from the nominal 2.6V. Of course, they resorted to another OCZ’s brand name technology aka EL (Enhanced latency). The second big news from OCZ was their successful overclocking of the DDR memory modules to 772MHz frequency with pretty adequate timings of 3-4-4-4. Theoretically, the improvement of production technologies will still allow DDR to develop further. Besides, AMD should be pretty excited about supporting these initiatives of the memory module makers. But from the realistic point of view, all these records are more of a personal prestige matter rather than technological value.

It is interesting that the second most active memory company in February appeared Corsair, which used to be a pretty conservative company a few years ago. Nowadays the XPERT series from Corsair is very well-known among computer enthusiasts. We have already reviewed this solution equipped with large digital LEDS displaying the memory modules status and working parameters. You can check out our review here.

In the end of the month they expanded this product family with the new TwinXP1024-3200C2 product line (the memory capacity and the working speeds are evident from the name, which C2 stands for a very important memory timing parameter: CAS2). The timings for this memory type look as 2-3-3-6 for regular CPUs, and 2.5-3-3-5 for AMD Athlon 64 with the built-in DDR memory controller.

In the middle of February Corsair also reminded us of themselves with one more useful solution: 533MHz DDR2 SO-DIMM memory modules for the new mobile Intel Sonoma platform. These modules have nothing outstanding about themselves: their speed and capacity (256MB-1GB) are pretty standard, which makes them a well-balanced practical solution. A bit earlier Transcend also announced similar new products.

However, most of the news is still coming from overclocking fields. Here we also got a word from Kingmax. They announced Hard-Core DDR500, which was claimed to be working stably at frequencies over 525MHz. in the DDR2 field we hear from A-Data and GeIL. A-Data promised to start producing DDR2-667 and DDR-800 in March, although I would expect the latter solution to start appearing in mass quantities in H2 2005 at the earliest. GeIL announced in February their 533MHz DDR2 memory modules with faster timing settings of 3-3-3-8- instead of the standard 4-4-4-12.

In fact the manufacturers of overclocked memory modules may face some real problems in the nearest future. There is a rumor that Samsung may discontinue DDR TCCD memory chips with low latencies and assign the production capacities with well-established manufacturing technologies to DDR2 and XDR. As a result, OCZ and other companies will have only Micron chips to go for. Micron however, may take advantage of this situation and help Crucial take over most of this market this way.

By the way, why should Samsung be so focused on the DDR anyway? In March the company announced that they finished the development of their 512Mbit DDR3 chip, though it will appear in the market no sooner than next year and we will be able to seriously regard the DDR3 market only in 2007. In fact, this is somewhat frustrating, because the whole thing looks very attractive: the new 80nm chip is expected to work at 1066MHz frequency and only 1.5V voltage. Moreover, its power consumption should get 40% lower compared with what the current DDR2 chips working at 1.8V consume.

As for DDR2, Infineon was the No 1 company in February here. They introduced their 4GB DDR2 memory module. You may find nothing so excitingly new about it in general, except for one fact: it was only 4.1mm thick, which is a true record for monsters like that. Of course, they managed to achieve this impressive module thickness (or thinness, I should say :) ) due to MCP chips. The MCP memory technology implies that there are more than one die inside a single package. In Infineon’s case there were two 1Gbit DDR2 SDRAM chips.

Samsung went even further in terms of the memory chips capacity. In the end of February the company announced that they were beginning the production of their 2.5Gbit MCP chips with the record-breaking capacity of 2.5Gbit. Note that this is the mass production capacity. In fact, Samsung also demonstrated 3.2Gbit MCP chip. Here, however, we are talking about a “mixed’ solution: there are two 1Gbit NAND flash chips and two 256Mbit Mobile DRAM chips in a single package. This way, you can immediately guess what the market for these solutions would be: 3G cell phones. As soon as they acquire new services and functions (especially video) they should theoretically require quite a bit of memory, that’s why such memory amounts can be demanded.

In fact here we are talking about increasing the capacity of the chips designed on the already exiting architectures. However, the memory companies do not stop developing absolutely new technologies as well. In particular, in early February we suddenly saw Ovonyx Company re-emerge in the market. This company is now run by the former Micron top executive Tyler Lowrey. As far as I remember, the last time we heard from them, was about 5 years ago, when Intel invested in this company quite a significant amount of money. At that time Ovonyx was working on nonvolatile memory technology based on the phase change effect (the amorphous or crystal state of the substance, which can change if necessary, corresponds to the 0 or 1 value of the cell), and Intel considered it pretty promising in those days. They actually licensed the ovonic unified memory technology. A little later the same was done by STMicroelectronics.

Yesrs have passed since then, and what do we hear now? No, this is not about the new products announcement. In February Elpida announced that they had licensed ovonic unified memory technology in order to finish its development, although they cannot yet point out any commercial projects based on it. So, it still looks like an exciting affair to me, but maybe Elpida knows something that we don’t?

Philips didn’t license OUM, but suggested using plastic for the nonvolatile memory, to be more exact – organic field-effect transistors based on it. The interesting thing is that Intel has already managed to step in here too, as they have been working on the same project together with the Norwegian Opticom, though unsuccessfully. And Philips seems to have found a way to success, but it will definitely take time before they will be able to harvest the fruit: it is still too early to speak of any commercial value of plastic memory like that.

Toshiba decided not to experiment with any new materials and thus appear closer to the counter than anybody else out there: in February they introduced 128Mbit chip based on floating-body effect. Namely, they took advantage of the side effect that prevented further speed increase of the transistors to create very economical memory. Intel has been trying to combat this effect in its processors for a long time already, and Toshiba viewed it from a different prospective and found out that it could be very efficient. So, they managed to make it work for them and are targeting for use in built-in electronics.

However, you needn’t invent anything brand new to earn good money. On the contrary, you’d better claim your rights for the new stuff, and the manufacturers will undoubtedly give up fighting after a while and will pay. Rambus was the first one to follow this strategy. Now Mosaid is enjoying the same success. The trial started in January against Hynix ended in piece in February already. The Koreans, however, had to fulfill Mosaid’s terms and license “their” technologies. Before that Samsung had to do exactly the same thing.

Here we would like to finish with the technologies and pass over to the life of the memory market. The market felt pretty optimistic in February, I should say, at least in Taiwan. ProMOS and Winbond didn’t just invest into production expansion: they took loans, which they will inevitably have to pay back one day. ProMOS believes that they will be able to pay back 316.9 million dollars, and Winbond feels confident about their ability to handle the 253.5 million dollar debt, even though the revenues of both companies have grown noticeably smaller lately. Nevertheless, both of them are planning to spend this money on new 300mm fabs, so that the trial production could start by the end of the year already.

This is a pretty interesting strategic position taking into account the joint acknowledgement that the DRAM market is going to shrink this year from the financial point of view, although it will still grow from the quantitative viewpoint. Relatively low demand together with the continuous memory chips price drop will do their job here. This is what we will see this year in general. But the month of February already didn’t arouse any enthusiasm about the memory prices already. Firstly, there was simply no price change throughout the first half of the month because of the Chinese New Year holidays: the price for 256Mbit DDR400 chips fell from $3.8 to $3.72.

When in mid February the traders returned from holidays it still took a while before the wholesale customers woke up from their hibernation. And then here is the end of the month, when the price pressure in the market is traditionally higher, as the companies try to sell as much as possible in order to draw beautiful numbers in their monthly sales reports. As a result, in the third week of February the price of the basic 256Mbit DDr400 chip fell from $3.72 to $3.45 and by the end of March it simply crashed down to $2.85. There might be some correction in March, because despite all reasons for this price drop it is still unacceptably low, but there is definitely no reason for any long-term price growth, that’s for sure.

In this situation you should either be a dedicated optimist or you should be ready to suffer some losses now for the sake of the future gains, I don’t see any other reasonable explanation for current production expansion. Anyway, the new fabs we are talking about will be coming into service only in 2006, and by that time Longhorn may finally be out…