|
The current photolithography process used in fabrication is an expensive one. If you notice the trends, process size has gotten smaller, but more logarithmically than linearly. Indeed, the shift to .13 micron was expensive. The actual creation of the processor is not too expensive, but the creation of the masks are horribly expensive from my understanding. While it is entirely possible to shrink smaller, I doubt there will be more than one or two future die shrinks, unless, of course, a more economical way is devised.
Neither do I think the key lies in multiprocessing. The algorithms are much more difficult (at least to me, but I know a few who are better at parallel than sequential algorithms oddly enough) and not all processes lend themselves to parallel processing. In fact, many algorithms achieve no benefit from parallelism.
I could never confirm this story, but I would not discard it: The man who created Deep Blue (whose name escapes me) did previous work on parallelism. The board almost did not give him his doctorate because they could not understand his work. They knew it worked, they knew it worked extremely well, but they could not figure out why.
Respectfully,
Demosthenes
|