The nonlinearity of the curve is why it's smart to take the long view. The understood footnote to Moore's Law is "on average".
Each discontinuity corresponds either to the introduction of a new chipmaking technology, or a disruptive change in market conditions. The heat problem tells you that the current state-of-the-art has hit a wall; I haven't lost faith that the next jump in technology will reset the curve and free up some more headroom.
It might help to update the precise expression of Moore's Law: "the number of transistors per square inch on integrated circuits [will] double every year since the integrated circuit was invented." That's a pretty parochial definition when you think about it, almost self-referential. Why should a feature of a particular technology--the transistor--be the unit of measure? Why not a functional measure like the number of logic gates? Maybe even that's too parochial, being an architectural assumption; how do you fit quantum or photonic processors into that picture? I'd say that benchmark-based density metrics are most objective--teraflops/cubic centimeter, say (I deliberately used a 3-dimensional physical unit, because areal units are also tied to a particular technology--the wafer).
Geek Speak » in reply to re: Moore's Law is not a continuous curve
punctuated equilibrium, heat, and bear markets
The whole thread (40 posts)
- Moore's Law has finally failed ?
- Most computers today are limited by DISK SPEED
- Maybe not!
- Intel press release on failure to increase processor speed.
- I think the next big push will be multi-processor chips.
- Why any faster?
- Diminishing returns
- re: Moore's Law is not a continuous curve
- It's more of a guideline... (n/t)
- Not the physics, but the markets