The transition from 64Mb to higher density chips is now speeding up. NEC reported that they would produce about double the number of 128Mb vs. 64Mb in the 2nd half of this year. Hitachi has jumped directly to 256Mb. Samsung is also transitioning quickly to 256Mb, while Micron is moving to 128Mb and will transition to 256Mb in 2001.
Prices of 64Mb SDRAM chips continued upward slightly in July, peaking at above $8.50, but finally stabilizing around the $8.00 range. There is some concern, however, that the increased demand was due to OEMs stockpiling inventory to prevent a repeat of last year. This could mean a sharp decrease in demand should sales turn out to be even a little soft for the Christmas buying season.
Though there is much interest in DDR and some discussion about whether DRDRAM will gain any more market share, the fact is that the vast majority of the market will still be PC100 and PC133 SDRAM for at least the next 6 months. Though the media generally focuses on the new and sexy technologies, the industry is actually still transitioning to PC133, with the expectation that at least 80% of the market will still be Single Data Rate SDRAM until after the first of the year.
The move to implement DDR is in full swing, with Micron already offering DDR modules (both registered and unbuffered) for qualified hardware engineers through Crucial Technology – and at PC133 prices. They are also offering a DDR Toolkit that includes not only data sheets and application notes, but the option to purchase DDR reference motherboards and modules to assist in design work.
NEC indicated that they will make some DDR in the 2nd half, but not in large quantities. They are still undecided as to whether sufficient DDR capable chipsets will be available to justify a faster ramp. Just in case, however, they have contingency plans to double DDR production if necessary. According to their spokesperson, the die penalty for DDR is only about 2% over SDRAM. Micron has also indicated that there is little difference in cost, as indicated by their pricing.
While the media has given a lot of coverage to DDR recently, it probably will not achieve any more than 10% market share this year, and most likely will be only half that. Nonetheless, the number of chipset vendors that have product sampling or planned is very significant, and there is very little doubt at this time that this will be the next memory standard on the PC for 2001 and beyond.
Rambus announced another DRAM licensing deal in July, this one with Oki Semiconductor, which covers SDR and DDR SDRAM. While it does put another manufacturer in their war chest, the actual revenue potential is quite small. What is most important about the agreement is that Rambus’ patent claims become stronger as a result. Still, Micron has indicated that their plans have not changed at all regarding any of their current and planned product schedules.
Samsung continues to champion DRDRAM, and has indicated that their yields are continuing to improve. At the DDR Platform Conference last month, they provided a roadmap which shows that overall yields are up to 90%, with PC800 yields at 60%. More significantly, they anticipate overall yields to improve to 98% within a few months, with PC800 reaching 70% or more. NEC also has plans to produce 128Mb DRDRAM throughout the rest of the year, but volumes are still fairly small. They anticipate higher demand for their upcoming 288Mb parts, which should be available in late September.
To this point, neither Hyundai, Infineon nor Micron have begun shipping any DRDRAM parts, and it is looking more and more like this will not occur – at least not this year. With only two chipsets able to utilize the product, one of them being all but dead and the other filling a niche market, there seems to be little incentive for anyone else to enter this market.
On the other hand, retail prices have been dropping at a rapid rate. Just 6 months ago, a 128MB DRDRAM module was almost ten times the price of an SDRAM module, but today it is just over twice the price. Though no official information is being provided regarding why this is happening, I can only guess that it is due to lack of demand.
Earlier this year I had a discussion with a Samsung spokesperson, who indicated that demand for their parts exceeded their ability to ship, with Dell receiving 80% of their production. While Dell had a special pricing agreement, everyone else had to pay the going market rate – which was quite high due to availability. While it is true that output has increased by about 500% (from approx 2 million chips per month in March to about 10 million per month today), demand obviously has not kept pace, allowing the price to drop very quickly.
With the introduction of the P4 later this year, there will be one more chipset that will accept DRDRAM, though it remains to be seen what volumes Intel will be able to achieve. If the ramp is very fast, this could drive DRDRAM prices back up to much higher levels, unless and until manufacturers can deliver additional quantities. Generally, it takes about 3 months from wafer start to finished die, so good forecasting is very important to maintain stable prices.
One other major factor in DRDRAM prices is that the die penalty for DRDRAM vs. SDRAM is approximately 35%, according to NEC – though Samsung believes they can get it down to within 5% by 2002. If prices continue to drop to within 50% of SDRAM, it will begin to look even more unattractive to manufacturers, who have had more than their fill of red ink over the past several years.
NEC is still marketing their Virtual Channel SDRAM, despite the less than glowing reviews it has received. An NEC spokesperson admitted that the 64Mb part was rushed to market too quickly, and suffered from having too few channels that were too narrow at 16 bits. The next generation, which will be 128Mb density, will have 32 bit channels, and will have twice as many of them, which should improve performance significantly.
Updates to Prior Month Reports
This should have been mentioned last month, but…
In June, a spokesperson for AMD had told me that the production capacity of Dresden was 8000 wafers per week rather than the 5000 I had previously published in an earlier update. After publishing this information, he contacted me and informed me he had misspoken, and the capacity is indeed 5000 wafers per week. I apologize for any confusion (or excitement) this might have caused.
In May, several contractors had reported that Dell was looking to build Duron systems – up to 200,000 of them. The apparent reason was the lack of availability of Intel processors, and Dell was looking at a contingency plan. These reports continued to come in throughout May and June. In July, the ‘processor crunch’ eased significantly, and PIII processors began to appear on resellers shelves – as well as in the inventory of OEMs. As a result, Dell reportedly shelved the plans for building Duron systems.
Be the first to discuss this article!