by Anna Filatova
06/09/2005 | 10:38 PM
May began with a fire at the Taiwanese fab ASE Chungli, and the market showed some enthusiasm trying to repeat the 1999 scenario since the setting was roughly the same: against the background of extremely low prices, there appears a certain factor that restricts the supply (it was the earthquake on Taiwan back in 1999), and the prices shoot up by tens of percent.
They did shoot up right after the news, but the enthusiasm was rather misplaced. Everything got back to the norm even before the evening of the same day. The prices did grow up a little as a consequence, and May began at $2.41 for a 256Mb DDR400 chip. But considering that the entire beginning of May, from the 1st to 10th, is the so-called Golden Week in Asia with all its festivities, the memory prices stood motionless throughout the first decade of May. The price of a 256Mb DDR400 chip just lowered to $2.38 which was an indication of the close-to-zero activity on this market then.
After the holidays were over, the traders got back to work and pondered where to move the market further. They had little difficulty in choosing the route. In view of the low demand and low volumes, the market went the traditional way, i.e. downwards. As a result, the price of a 256Mb DDR400 chip had sunk to $2.31 by the 13th of May and stopped there for a while. A few days later it went lower still, reaching $2.25 for a 256Mb DDR400 chip and thus getting below the manufacturing cost for a majority of DRAM makers which is about $2.5 for a 256Mb chip.
So, this price reduction might have been over 6 percent in a single month, if it were not for a slight correction which returned the price back to $2.31 during the remaining week. I wouldn’t say that this changes the overall picture much, though. It’s clear that the memory makers suffered losses in April and May and now have to reduce their DRAM production or transfer their production facilities to flash. They of course prefer the second way, but it already makes no difference for DRAM: the supply of DRAM chips is expected to shrink a little in the near future and the price will rise up closer to $3 per a 256Mb chip.
The ongoing transition of the manufacturers to DDR2 intensifies the situation somewhat: the share of this memory type is approaching 30% with many major makers and will have probably reached 50% from all produced DRAM chips by the end of the year. Things going this way, with the demand remaining rather conservative and focusing on DDR, and with the manufacturers actively taking up DDR2, we may encounter deficit of DDR accompanied with overabundant supply of DDR2. A curious fact, for the first time in this year the contract prices on 256MB PC3200 modules grew, although very little, from $21.2 to $21.4, at the end of May…
Samsung is traditionally optimistic about the whole situation, promising a disastrous deficit in the next quarter and more horrible things by the end of the year. But this is just the opinion of some of the company’s employees, while the company’s official position is that there’s going to be nothing good on the market until the year is over.
There was some animation on the market of memory modules in May. That was quite natural as the prominent players were preparing for Computex. OCZ was in the head, beginning May with the release of reduced-latency PC2-6400 modules in the Platinum EB (Enhanced Bandwidth) DDR800 series, with 4-3-3-8 timings and manually tested for stability. Next, they promised to introduce 1GHz DDR2 PC2-8000 modules and even officially announced this product at the end of the month: PC2-8000 Platinum EL Dual Channel with the voltage lifted up to 2.1V. Well, the voltage is the same as with PC2-6400, but PC2-6400 has normal timings against OCZ’s new memory’s 5-5-5-15! This race of speed gives birth to almost absurd products sometimes…
A company that wouldn’t lag behind OCZ was PQI, which had become a leader among the makers of such modules. At the beginning of May, the company was rumored to be preparing 900MHz PC2-7200 modules, but PQI refuted this info in some ten days by promising to introduce 1GHz PQI28000. Unfortunately, the latencies are exactly the same as with the analog product from OCZ. On the other hand, the voltage of the modules is just 1.9V! These modules could also reach 1066MHz frequency in the single-channel mode, so we’re up to new records in the field.
Records may come from Corsair, for example, who overtook the 1GHz bar in May. The company introduced its XMS2-8000UL with the record timings for this speed: 5-4-4-9. That’s impressive, no doubt. And that’s also something really useful and reasonable.
Considering such speeds, the offers of the tardy Kingmax and Infineon come in most handy: they prepared DDR2-800 modules for Computex, and Infineon’s 800MHz modules can work without heatsinks. GeIL made an addition to its Ultra-X series: PC3200 modules of DDR SDRAM with latencies of 1.5-2-2-2! That’s a truly useful offer indeed.
Other manufacturers focused on the workstation and server market rather than on overclockers in May. Transcend released quite ordinary DDR2-667 modules of 512MB and 1GB capacities, while the other two offers were more original. Kingston introduced 4GB PC2700 modules (this standard still remains the most popular in server platforms) with support of not only ECC but also of Quad Rank (four DRAM chips on a module can be addressed simultaneously).
Infineon showed modules of a humbler functionality but these were DDR2-400 and – what was more important – 8GB capacity. The thickness of the modules is a record-breaking 4.1 millimeters, which is another confirmation of the fact that the German company remains an unrivalled leader as concerns high-density chip packaging. This leadership is paid for by the company’s clients: the retail price of the new product is something like $10,000.
In May, Infineon was spotted in another field, too. At the end of the month they announced that the company, in conjunction with IBM and Macronix, will get to developing phase-change memory – it’s when the state of a special material changes from an amorphous to a crystalline structure. DRAM isn’t concerned here at all. This new memory is going to be a competitor to flash memory, rather.
As for the DRAM realm, there was the statement from Inotera, Infineon’s Taiwanese daughter, that it was ready to begin mass production of 90nm DRAM chips. Infineon will of course be the first to receive the outcome.
Meanwhile, ProMOS with its new strategic partner Hynix are already busy developing 65nm technologies in order not to depend on the same Infineon that ProMOS had earlier licensed its current 110nm process from. But like Inotera, ProMOS will begin its 90nm production this year, having obtained the appropriate technology from Hynix.
The Koreans themselves are feeling well enough, reporting profits for a third quarter in a row: $320 millions in Q1 with sales amounting to $1.27 billions. Hynix says the profit might have been even better if it hadn’t been for the weak US dollar…
Hynix’s neighbors shouldn’t care about profits at all, having one third of the world’s DRAM market. Samsung with its production and sales volumes will be O.k. in any situation. That’s why the Korean giant can spend some time butting each other with Elpida over who’s going to supply 512Mb XDR memory for the perspective Sony PlayStation 3. In May, both the companies declared their plans on mass production of such chips, and Samsung caught up with Elpida by showing the samples.
It so happens that both these companies, Elpida and Samsung, are Rambus’s key partners. Thus, they are equals in this competition. Being a supplier of RDRAM for the first PlayStation, Toshiba may join the quarrel, too. This company is also going to carry on its partnership with Sony as a supplier of chips for the new game console. At least, Toshiba had long ago announced its intention to take up XDR manufacture seriously. So, that looks like a curious collision, and I wonder what’s going to come of it. So far Samsung seems to be on the winning side as it offers 4.8GHz 90nm chips against Elpida’s 4GHz 100nm ones.