Ian: regarding DRAM generations, some data is required.
Regarding your points...
< 1. 4Mb chip life was unusually long and unusually profitable. >
Unusually profitable - yes. Unusually long - no. DRAM generations last 3-4 years, averaging 3.5. This lead to the statement of 256 mb becoming mainstream in 2003. The 16k lasted 15 quarters as measured from bit crossover to bit crossover, while the 1m lasted 12. The 1m/4m crossover was in q4'91, and the 4/16m crossover happened 16 quarters later, in q4'95.
< 2. Crossover between 16Mb and 64Mb per-bit-pricing just occurred. Crossover between number of bits produced is imminent. i.e. The 16Mb chip is about to be infamous for the shortest life span ever. >
The price per bit crossover for 4/16m happened in q3'94 as the x4 version went into oversupply. However, since the part did not match what the PC guys wanted for the Pentium bus, they did not buy it. It was not until the x16 part yielded high enough in late 1995 before the market converted. In effect, the DRAM guys were profitable because they misread the market. The price crossover you referred to may be taking place with the x8 EDO part. Well, the PC guys won't buy it - they all want the x16 SDRAM, priced at about a 5-10% premium to 16m right now. But let's say the 64m will have bit crossovcer in q4'98 or q1'99 - will it be the shortest life? Perhaps, but a 12-13 quarters, it is well within the window of history.
< It looks like the same set of dynamics will cause the 256Mb chip to supplant the 64Mb chip. >
Sure, if there is a buyer and the right configuration. Keep in mind the x32 required to make the conversion before 2002 will require a costly change in packaging design, as the chip becomes too rectangular.
<3. Normally, a shrink facilitates more bits per chip. Your message implies that moving to 0.18æm conflicts with 256Mb. Really!!! >
Actually, the two have nothing to do with one another. Does the piece of equipment know what is on the wafer? The move and requirements for any technology node do not really depend on the product, or how many bits are on the chip. The techno-nerds have linked the two for convenience, but it is quite misleading. After all, several years ago, the 0.18 micron node was supposed to be the 1gb generation. That certainly is not true today, is it?
< I would expect that we'll see 256Mb chips produced in quantity by or before 2000; and that 1 Gb chips will be sampling; leading edge will be 0.13æm or smaller. >
Yes, I would expect 1gb to be sampling as well, but the term "quantity is up for some interpretation. is quantity more than 50% of bits shipped? No. 10%? Probably.
The PC market rules the mainstream product in DRAM. 16x16 256mb chips mean the only configurations you have for PCs is 128MB and 256MB. They will not accept this inflexibility in 2000.
Z |
