Memory Market Overview: February 2005

We resume the monthly write-ups about the situation in the memory market. New releases, company news, prices, events. We will try to discuss all this in detail in our monthly memory market coverage!

by Anna Filatova
02/10/2005 | 09:35 PM

We’re resuming our traditional monthly reports about the well-being of the PC memory market. Since this coverage will deal with the results of the first month of the year, I will also try to summarize the results of the last year as well as make some forecasts for the entire 2005. The intrigue or rather topic for discussion here is the further penetration of DDR2 SDRAM into the market, and the main question is the extent of this penetration.


Despite the relatively good acceptance of the i915/i925 chipsets (although not as warm as it might be), the supply of DDR2 modules is not expected to be high till the second half of this year as the DDR2 share in the output of a majority of module manufacturers from Hong Kong and Taiwan is only about 1-5 percent (although Taiwanese analysts estimate the world’s share of DDR2 as 13 percent against DDR’s 72 percent).

Many manufacturers will have begun to increase the DDR2 production volume by the summer, but it still seems that it is going to take long for DDR2 to truly conquer the market. That’s natural if you recall that DDR didn’t replace SDRAM in a moment, although had more evident advantages than DDR2 has in contrast to DDR. The price difference between DDR and DDR2 is also too big today and doesn’t match the difference in their performance. So AMD may be quite right in deciding not to support DDR2 until 2006 (even though this decision is partially due to the evident technical problems).

On the other hand, the chip makers are still increasing their potential to be ready to the possible (and highly anticipated) growth of demand. This time their hopes are invested in the Sonoma. Nanya, in particular, is going to resort to near dumping, reducing the price difference between its DDR2 and DDR modules to less than 10 percent, which should arouse the interest towards DDR2. Nanya even promises a price difference of 7 percent for its Elixir series of memory modules.

The company seems to await a powerful reaction (and quite reasonably I should say) as it intends to transfer up to 50 percent of its DRAM production facilities to DDR2 (compare this to today’s 20 percent). Well, I have no doubts the demand will be growing, but the forecast that DDR2 will account for 70 percent of the world’s DRAM sales by the end of this year seems an exaggeration as least. The market is usually more conservative than enthusiastic forecasters, especially if the forecaster is one of the manufacturers involved.

Talking about over-enthusiasm, I should mention the frequency factor, too. In January Kingston showcased the fastest memory module of today. It’s 866MHz DDR2 SDRAM. These modules are so fast that they are guaranteed to work more or less correctly on one mainboard only, ASUS’s P5AD2-E. Well, Kingston isn’t planning to produce such modules in near future, but even the 750MHz modules from the same company scheduled for February have the same problem. It’s like the Lamborghini is a good car, but not in the small streets in the center of a city the modern PC is. Well, you can buy one just for the heck of it, of course…

But let’s get back to more serious things – to the fundamental parameters of the memory market at large. January, iSupply summed up the results of 2004 and found the market to have grown by a half. Nine out of the top ten manufacturers have also experienced growth by tens or hundreds percent. Quite naturally the leaders in dynamics are the outsiders like Elpida (122.8%), Powerchip (164.1%) and Elite (112.7%). Here’s the big picture:

Preliminary results of the memory market for 2004, in millions of dollars.


Company name

Revenues in 2004

Revenues in 2003

Change in %
































ProMos/Mosel Vitelic





























We congratulate Hynix who has struggled its way (using its traditional dumping) to its rightful second line in the table, although with a negligible gap towards Micron that has had the worst dynamics in the top five. It’s going to be interesting to watch the situation in the Hynix-Micron-Infineon trio develop as a conscientious analyst can hardly make any prediction about the outcome.

Note also that Powerchip’s spurt helped it to overtake the usual Taiwanese leader, Nanya, and to become the major DRAM maker on the island. In 2003 Nanya had the fifth place in the list, but now its more dexterous competitors have pushed it down to the eighth. Nanya is another object for the curious analyst to be watching intently in this year.

There’s evidence that Nanya is going to have troubles ascending up the Top Performance list (of course, if no one above Nanya has any too serious problems), at least one gets this impression looking at the forecast of the capital investments of the Taiwanese DRAM makers in this year:


Capital expenses, in millions of dollars


2005 (forecast)
















As you see, Powerchip and ProMOS keep up their tempo and even more. These data also impart some optimism about the future of Infineon (Inotera is Infineon’s and Nanya’s joint venture, and this may tell well on Nanya’s showings). The general trend is that the Taiwanese are expanding their production capacities, but this poises a few new questions.

You see, no one is waiting for any gifts from the new year. Unlike the previous year, 2005 is going to be a hard one. In December, as the first sign, the chips production volume shrunk by 2.2 percent (curiously, the production of SDRAM and DDR modules dwindled by 1.8 percent, and of DDR2 modules by 5.2 percent only).

Graver than the December results is Samsung’s prediction that the growth of the DRAM supply will exceed the growth of the demand by 1 percent in this year (45 against 44 percent growth). And this comes from Samsung who have exhibited just a childish optimism – take its forecast about a severe DRAM deficit in the fourth quarter of 2004. If Samsung doesn’t expect a deficit of memory this year, the real situation is much more serious. iSupply, for example, promises 49 and 44 percent growth of supply and demand this year, and this looks more like truth.

iSupply thinks the volume of the DRAM market in money terms will be about $27 million, i.e. only 2.6 percent above the year 2004 (compare this to the 51-percent difference between 2004 and 2003!). The agency draws a grim picture with the average price per megabyte going down by 31 percent and the manufacturers’ profits to $4 billion against $4.7 billion in 2004. They suppose there’ll be jugglery going on with the facilities between DRAM and flash which can lead to the manufacturers’ making wrong estimates of the demand on the particular product and, as a consequence, to local shortages that can push the prices up for a short while.

As for the prices, the spot market has become a poor indicator of the general trends as the major chip and module makers pay less attention to it, preferring to focus on contract shipments where the major OEMs guarantee a relatively stable demand in a world where there’s a not very favorable forecast for the entire year. Samsung, the biggest player in the market, came to the fore, saying it would change the ratio of its spot/contract market shipments from the last-year 35/65 to 20/80. Major Taiwanese companies like ProMOS and Nanya followed the leader expressing the same desire.

On the other hand, the spot prices still remain an indicator of the market’s general disposition, although January is of little interest in this respect: it started with the New Year holidays in Europe, then the disaster in south-east Asia happened, and the preparation to the Asian New Year were underway at the end of the month. So, there were few workdays in January, with all the consequences. The customers were not active, trying to find their bearings in the situation.

So when it became clear that there wouldn’t be any big laying-in of supplies before the Christmas holidays, the manufacturers and distributors began to throw away their stock, trying to reach their sales goals for January. This of course resulted in reduced prices at the end of January: from $4.02 for a 512Mb DDR400 chip at the very beginning of the year to $3.8. DDR2 has been falling down throughout the entire month (well, it has much room to fall down to), to $4.75 and $5.27 for 400MHz and 533MHz 256Mb chips, respectively. Thus, the price difference between DDR and DDR2 was 25 and 39 percent, for these two respective speeds, by the results of the month. That said, it’s obvious that DDR2-400 can’t be taken seriously, while DDR2-533 which aspires to become the mainstream still needs improving upon.

That’s enough about the prices – they just don’t deserve any more of our attention. Now we can have a look at the technologies and the news from the corporative life. The technologies are evolving, in spite of some people willing to make the others pay for it. It’s only five years since the last of the Rambus wars was waged (we’ll talk about this company shortly), but Rambus’s clone has appeared in the horizon, the Canada-based MOSAID.

In fact, MOSAID has been claiming the role of the creator of DRAM for long, but they have been doing so without that enthusiasm Rambus used to display. But Rambus has been somewhat silent, while MOSAID has gained its strength. In January we learned that Samsung gave up its position in the lawsuit that had continued since 2001 and acknowledged MOSAID’s victory and paid it the appropriate tribute. Hynix is selected as the next target – MOSAID filed a suit against Hynix’s using fundamental principles of the DRAM architecture the day after the successful conclusion of the Samsung case.

I only have two questions: why Samsung and the Japanese companies are a weak link against the blackmail of this kind (right now, MOSAID is receiving its tribute from Samsung, Fujitsu, NEC, Toshiba, Hitachi, Mitsubishi Electric, Oki Electric, Matsushita, Winbond Electronics, Sony Corporation and Royal Philips Electronics; so there’s only Samsung from the top five in this list). And how many are there such obscure DRAM inventors left that we don’t know anything about until their filing the first suit? Anyway, they know which country they should start from.

More about Samsung and Rambus: as you know, Samsung is among the major licensees of SDRAM with Rambus, not mentioning the RDRAM license. During its RDRAM affair Intel put a few million dollars in Samsung to increase the production of this type of memory. Samsung even promised then that the market volume would be 600 million RDRAM chips in the 128Mb equivalent in 2002. The whole affair crashed eventually, but Samsung and Rambus have remained allies thereafter.

As a result, at the end of January Samsung at last started to make 256Mb chips of XDR, the new memory type from Rambus which aspires for the title of the best system memory for the PC. Samsung promises to offer 512Mb XDR chips in the first half of this year that provide a bandwidth up to 12.8GB/s. That’s impressive, but XDR won’t be system memory for the PC. Samsung will probably sell its produce for Sony’s PlayStation 3 (yet another old-time partner of Rambus’s, by the way).

Now the most funny thing is Rambus’s running a new attack against the memory manufacturers as if they are not tired yet of MOSAID! California Northern District Court began to investigate Rambus’s claim about Hynix’s violating 29 of its patents. Then Rambus files a suit against GDDR3 manufacturers (Hynix, Infineon, Nanya and Inotera, a joint venture of Infineon and Nanya) who violate 18 of its patents. Rambus must have forgotten the SDRAM-concerned suit it lost to Infineon a couple of years ago. Today Nanya and Infineon stand firm and say they won’t pay any licensing fees.

That’s enough about plaintiffs and defendants, let’s better talk about the real creators of real new technologies. For example, the same Samsung has unveiled its 90nm 512Mb mobile DRAM chips for smart-phones and PDAs. It’s not a revolution, of course – just an ordinary product from the well-known generation of mobile memory, but the capacity is unusual. Samsung remains an unrivalled leader in this field. The company is going to begin the production of such chips in the second half of this year.

A good confirmation of the fact that mobile chips aren’t any news at all is Elpida’s starting to ship samples of their 110nm 256Mb DDR chips manufactured with Super Self Refresh technology that allows to reduce the DRAM refresh current in twenty times and thus to considerably reduce the power consumption of the chip, other factors being the same. Such chips are expected to come in mass quantities in March. Elpida will also glue two 256Mb chips together in a Double Density Package to ship the result as a 512Mb chip.

That’s all about the announcements from the chip makers as they were mostly busy preparing for the hard times ahead. First of all, Hynix and ProMOS have finally formed their alliance so the former will gain a preferable access to the extensive production facilities of the Taiwanese who in their turn will implement Hynix’s advanced 90nm tech process and then they both will collaborate on 65nm technology. Since the second half of the next year the production of chips under the Hynix brand will expand in China – ProMOS is going to start producing mobile DRAM chips at its Chinese fab by that time.

Powerchip seems to have enough troubles and opportunities in Taiwan. In January they reshuffled the board of directors: people from Elpida came to replace the Renesas team. That’s not much of a change, though, since Renesas belongs to the same companies as founded Elpida.

Anyway, considering the Taiwan government is going to give permission for business to only three DRAM makers there, Powerchip intends to become one of the three and thus have its patch on the continental territory, but they are seemingly not going to really manufacture memory there in near future: the company (with TSMC and Vanguard) should get additional 34 hectares of land in the Hsinchu industrial park and build their next 300mm fab there.

They will solve the money problem somehow. For example, Winbond got $250 million to build its first 300m fab. Powerchip is the more likely to have investments. Nanya is also ruminating plans on building its own 300mm fab, outside the Inotera project.

On the whole, the Taiwanese are expanding their production facilities, but there are clouds ahead in the horizon. The Chinese SMIC which has only recently gained any weight in the DRAM world is rumored to have set an absolutely dumping price of 500 dollars for its 200mm 0.35-micron wafers. That’s times below the prices from UMC or TSMC. The rumors mention the logics production, however the tendency can’t leave the memory makers untroubled.

It may happen that Infineon begot a dragon the whole region will suffer from. By the way, aren’t the rumors about the production of 20-30,000 wafers with defective 110nm DRAM chips by chance a result of the collaboration of the two companies? I hope this memory won’t get to market, also not on the modules of the new AENEON brand, oriented on regional PC makers. These modules were available at the end of the last fall in Europe and the US, and now they are present on the Asian markets. There’s nothing extraordinary about them, they are the same standard modules as in the Elixir series from Infineon’s partner Nanya. Can they come from the same production line in Taiwan?

Micron and Corsair amazed the community with their new memory modules in January. Micron began to ship samples of FB-DIMM (Full Buffered) modules on DDR2-667 chips. That’s a very appealing thing and I’m very positive about the very concept of FB-DIMM and hope it will find its way not only to servers, but also to desktop computers, but Micron itself is only going to see its products in servers no sooner than the first half of the next year.

Corsair has acted up with another of its designing caprices, introducing its Corsair XPERT series which is not only hung with a LED garland but is also equipped with an LCD display where you can run a scrolling string of text up to 68 characters long with the help of a configuration utility. Otherwise, this memory features low latencies and high frequencies. The precise parameters will be defined later on. XPERT PC3200 modules are going to come to market in February.