Katherine - Demand drives shrinks, not the other way around. For DRAMs, which are by far the largest single segment (especially in unit terms, less so in dollar terms), demand is very elastic. If you make them cheaper, people will buy more of them: you can never have too much memory.
I humbly disagree. There are, IMO, two very independent drivers of shrinks. The first is a desire to get faster, lower power chips (i.e. your 'demand'). The second is to get more chips on a wafer and to produce more chips in the same amount of time (i.e. cheaper). I think item #2 predominated in 97 and 98 (How else do you explain the much more rapid than normal rate of shrink in last few years? Was demand that much greater?) Then, having shrunk anyway to save money, why not put more on the chip and thus reduce the number of chips and costs further. I.e. a decrease in the number of chips shipped can indicate either a decrease in demand, or excessive shrinking and it is impossible to tell which without more data. The first is bad since it may well continue indefinitely (e.g. PC saturation), but the second is a transient factor which I suspect will run out of steam as the ability to do cheap shrinks dries up.
In any case, I am not sure we really disagree on the larger picture, since we both believe that the number of chips produced isn't a great indicator of real demand since there are too many confounding influences. I think the best indicator of health is ASP per chip or even better $ per transistor. If the latter is declining much more rapidly than normal, it is a good bet that the manufacturers aren't going to invest in more capacity.
Clark
|
