by Anna Filatova
11/13/2003 | 01:52 AM
The last month brought no changes into the assortment of available memory modules. The cap has only climbed beyond “PC4000” and the progress is rolling on at a high rate. OCZ launched its PC4200 Premier series on 533MHz chips (against the standard frequency of 400MHz) in early October.
The same company improved those modules a week later and added to them the “EL” prefix (Enhanced Latency). The latency was 3-4-4-8 clocks in the first case, and became 2.5-4-4-7 after the improvement. Curiously enough, both variants feature a solid design, with sober-looking aluminum or copper heat-spreaders. After the parrot-like colors we’ve got used to seeing lately, the new models from OCZ are a treat for the tired eye.
And this was not the end. After another week, OCZ drew the spotlight again, bettering the latencies and issuing its EL DDR PC-3500 Platinum Limited Edition. The timings formula is close to perfection – 2-3-6-1! The speed is of course lower than in the first two cases, “only” 3.5GB/s. Let me repeat the well-known fact once again, though. The difference between the performance of PC3500 and PC4200 modules is negligible and is better examined under a strong microscope.
In fact, it was only OCZ that released overclocker modules in October. Other manufacturers were trying to find their fortune with niche modules that work at standard frequencies. Transcend sampled 512MB PC2100 modules of the MicroDIMM form-factor that make the nightmare about the notebook with 1GB of memory, a reality. The same company boasted its 1GB PC3200 modules for mass PCs, never mentioning who might need such capacities.
Corsair and Smart Modular preferred to release registered modules in October. Corsair announced two sets of paired 512MB modules (with and without ECC) for the Athlon 64 with the appropriate speed, PC3200. Smart Modular targeted server platforms where PC2700 is the most widespread specification and the capacity of 1GB per module is much needed.
Overall, the module makers did nothing extraordinary to promote their products in October. Well, we didn’t expect that, either. The DRAM market is all calm and quiet, running on its schedule filled in for the next couple of years (Gartner promises a strong deficit for 2004-2005 years, though. They are wrong – supply will still be higher than demand). This memory has long become an exchange product, the only point of interest is the oscillation of the price.
In October as well as in previous months, this market seemed to have gone through the stormy days. Even the prices were in a stupor, against the background of minimal shipment quantities. The regularly chanted mantra, “The user doesn’t need such capacities and speeds”, is starting to work at last. The average price for a 256MB DDR400 chip remained at the point of $4.6-4.7 throughout the month. The price would go down by about twenty cents in the per-day graph, but it got back anyway by the end of October (these are purely traders’ activities, though, having nothing to do with the global market situation). The price of 266 and 333MHz chips is following the old trend: the average prices for 256MB chips went up by 5-10 cents in October, getting more closely to the DDR400 ceiling. On October 30, they were $4.49 and $4.54, respectively.
But if the intrigue has gone from here, it should appear somewhere else, shouldn’t it? I mean we are at an exchange and the traders have to earn their living by shuffling the prices about. It was the good old PC133 SDRAM that moved the market on in October. One 256Mbit chip of such memory costs now $4.95 compared to $4.6 at the beginning of the month. This is already more than PC3200 DDR costs! It cannot be avoided, though. The demand for SDRAM is still higher than for DDR, while the supply is low. The chip manufacturers overestimated the market of high-performance PCs, but neglected the household appliances field where SDRAM is a wanted product today. Underestimation provoked the deficit we witness today.
Considering the “back-to-school” season brought about no agitation and, consequently, no price growth, we can extrapolate this for the Christmas season ahead. We will hardly see any significant growth of the demand for memory. Since the memory makers are trying to get their financial results up by throwing more of their products into the market, we are more likely to observe a reduction of DDR SDRAM prices in November.
By the way, how have our manufacturers been doing? This October was a month of both joy and sorrow to the Taiwanese. Nanya and ProMOS calculated their sales results for September and found them to be lower than in August because of the price reductions and the deteriorating demand. The same ProMOS with Winbond and Powerchip were on the successful side returning back to profits, according the third quarter results. Nanya is not included. The ex-locomotive from Taiwan is in a slight minus. However, Nanya is said to be getting an order for memory from the NEC computer division. This may be well for Nanya’s sales. It was actually very interesting that NEC approached Nanya for memory instead of NEC’s own daughter company, Elpida.
Elpida has been somewhat unlucky recently. ProMOS managed to get along without Elpida in building its second 300mm fab, notwithstanding the signed memorandum for mutual understanding and cooperation on 0.10micron and finer production technologies. The Taiwanese preferred two unnamed-yet partners instead.
Still, Elpida has a loyal (yet) ally, Powerchip, which is making profits and boosting the sales volumes (225% more compared to the September of 2002). In October, Powerchip laid the keystone into the foundation of its second 300mm fab. The supported manufacturing technological process is 0.10micron, the planned output – 40,000 wafers monthly. Powerchip is going to be the world’s fourth memory maker after the fab is launched – right after Samsung, Micron and Infineon.
High hopes… Elpida was once aiming at the first place, too, and where is it now? They do promise to boost the outcome of their 300mm Hiroshima fab to 15-16,000 wafers per month. That’s even ahead of the schedule they had revealed before. By the way, the announcement about the Hiroshima fab was made at the ceremony devoted to the beginning of the Powerchip fab construction, where Elpida’s orders will always be welcome.
Yeah, Taiwan is steadily gaining its ground, providing production facilities for nearly everyone. Infineon is going to invest $600 million into the island in the next three years. The biggest share of the money comes to the joint venture of Infineon and Nanya – Inotera Memories.
There are peninsulas in this world, which try to be independent of this island, however. The Koreans don’t want to be dependent on anyone. Samsung and Hynix both prefer to rely on their own forces solely. They do it well, though. Hynix is even in holiday spirits now: the results of the last quarter, summed up in October, indicate a sales growth of 27% against the previous quarter and an operational profit! And it seemed like Hynix had forgotten the word “profit” altogether, having erased it from its vocabulary. That’s the effect the protective duties can make…
From the technical point of view, Hynix has nothing new to offer. Well, no one seems to do nowadays. Intel is testing DDR2 products for compatibility with its upcoming platforms. October, Hynix’ 0.11micron 512Mbit chips passed the tests. This is no record, though. Elpida sampled 0.10micron 1Gbit DDR2 chips working at 533MHz instead of 400MHz in October. Mass production is scheduled to start in the beginning of the next year. Hynix works by the same schedule, too.
Well, who said users need 1Gbit? Considering the consumption rate, most of the users will be quite satisfied with a couple (two channels!) of 256-512MB modules on 256Mbit chips. That’s exactly what Micron successfully tested with Intel in October – 0.11micron DDR2 chips. The company is not sluggish in technology, having sampled 1Gbit DDR2-400 and DDR2-533 chips. It’s just not necessary for a mass computer system to have so much of memory.
There is only one traditional refuge, niche markets. Notebooks come first. Looking into the future, Micron is already sampling SO-DIMM DDR2 modules. They are intended for the next generation of Centrino – Sonoma – to arrive in the second half of 2004. The platform is going to be quite outstanding, and the memory should match it. So, DDR2 is the only option here. At that, it’s better to get the memory certified with Intel as soon as possible. That’s what Micron has been doing lately. But there is a lot of time left yet and others will surely catch up. Let’s wait for them.
Micron’s fantasies didn’t stop at that. At the beginning of the last month, the company sampled first 288Mbit RLDRAM II chips – 400MHz DDR with low latencies (average cycle time is only about 20ns) intended for telecommunication equipment, yet another niche market appealing for the memory makers.
The last month had its intrigue, though. Rambus is returning into the main PC memory market. No, Intel hasn’t yet announced its chipsets to support the new standard from Rambus. So far, Rambus only revealed its plans on making its standard, XDR DRAM, into the basic PC memory by 2006. The clock-rate of the chips is going to be about 3.2-6.4GHz, with a bus width of 128bit and a bandwidth of up to 100GB/s. It sounds sweet like hell. In its time, however, 800MHz RDRAM was most impressive against 100MHz PC100 until the problems with heat dissipation, price and other small things arose.
Yeah, it is all very beautiful in theory, just like everything Rambus has offered in the last couple of years, but I rather doubt Intel would go for another try with this company. So, Rambus remains a kind of honorable theoretician (regrettably, maybe). So far, the company is waiting for the end of the suit with Infineon, after which it will get its fees for the SDRAM produced by the Germans.
By the way, Rambus is not the only innovator. As it turned out in October, Sun Microsystems is also developing a new type of memory for the PC. The company voiced its “throughput computing” paradigm, which is to become the ultimate solution of the memory bandwidth problem. Sun considers the low memory bandwidth one of the narrowest bottlenecks in today’s systems and suggests using up to 8 physical cores in one memory chip. The cores could work on different data streams in parallel. Interestingly enough, mass shipments are promised for the same 2006 year. It would be interesting to watch these two technologies collide, but…what will Intel say?
Infineon, waiting for the court decision, is focusing on MRAM. The future of MRAM is being discussed less nowadays, since the capacities and speeds of such chips grow too slow. It was lately rumored that Infineon, collaborating with IBM, also decided to enroll Axon Technologies with its Programmable Metallization Cell memory (PMCm) technology. Specialists say (for a hundredth time already!) that this could be a replacement for both DRAM and flash. First products based on this technology are going to appear…Yes! in 2006.
Motorola, on the other hand, doesn’t plan on anything. They just do their work, being the pioneers in the field. They produced 0.18micron 4Mbit MRAM chip available in samples for their clients. Recalling that the company offered only a 0.60micron 1Mbit chip a year ago, the progress is impressive. But of course, there is no talk about competition with DRAM or flash.