Linked by Thom Holwerda on Mon 20th Feb 2012 22:53 UTC
Hardware, Embedded Systems "A group of researchers has fabricated a single-atom transistor by introducing one phosphorous atom into a silicon lattice. Through the use of a scanning tunnelling microscope and hydrogen-resist lithography, Martin Fuechsle et al. placed the phosphorous atom precisely between very thin silicon leads, allowing them to measure its electrical behavior. The results show clearly that we can read both the quantum transitions within the phosphorous atom and its transistor behavior. No smaller solid-state devices are possible, so systems of this type reveal the limit of Moore's law - the prediction about the miniaturization of technology - while pointing toward solid-state quantum computing devices."
Thread beginning with comment 507958
To read all comments associated with this story, please click here.
Moore's law
by ndrw on Tue 21st Feb 2012 04:36 UTC
Member since:

Moore's law is not about technology, it's about economy of the IC process development. That's why the growth is exponential - better chips produce more demand, more demand brings more money, more money make better chips - a positive feedback loop. The rate of growth was limited only by engineering effort needed for solving a large number of non-critical issues.

As with any exponential growth it finally ends (it must end because neither physics nor total resources depend on money), and it ends rapidly. We don't even have to go down to the atomic scale - in many applications the exponential growth has already stopped.

The Moore's law, as it was originally formulated (density of transistors doubling every X months) is still valid in applications like memories or FPGAs (which are constrained by density and are easy to scale). But in CPUs and ASICs there is hardly any performance scaling anymore, people are now focusing on incrementally optimizing performance/power instead. That's because we are no longer able to scale down the supply voltage (transconductance of MOS transistors is fixed and mismatches are growing) as we did a decade ago. For now, we can work this around by making e.g. more CPU cores on the chip (so that we can utilize larger density of transistors) but that's no longer an _exponential_ growth. Performance doesn't scale proportionally to the number of cores (IO, software parallelism) and complexity and cost grow faster than it.

Reply Score: 6