4
$\begingroup$

If I am correct, chips cannot get much smaller because of Heisenberg's uncertainty principle. My friend and I want to perform an experiment (which is cheap, i.e. doesn't require million-dollar equipment) which shows that if a chip is too small, it will mess up. Is there any such experiment?

$\endgroup$
5
  • 5
    $\begingroup$ This might get a better answer on ee.se. This board is about computation, algorithms and the like, rather than physical hardware. $\endgroup$ May 16 '13 at 14:56
  • 3
    $\begingroup$ It seems to me that to do a meaningful experiment about the properties of very small things, you would need to produce very small things. For the values of "very small" you're talking about, this is probably expensive. $\endgroup$
    – adrianN
    May 16 '13 at 14:59
  • 1
    $\begingroup$ this is a billion dollar question related to whether Moores law is still holding (historically it has been mainly fueled by smaller gate width), and some say it may be slowing down. those who are in a good position to know, eg Intel, are not in a good position to be totally candid. but a plot of gate widths over many years esp over recent years may help give an answer. looking for that graph if anyone has seen it somewhere! another basic limit is that chips are etched optically ie with light, so light wavelengths are a key factor. $\endgroup$
    – vzn
    May 17 '13 at 17:14
  • $\begingroup$ @vzn: Moore's law has been going strong. See CPU DB: Recording Microprocessor History by Danowitz, Kelley, Mao, Stevenson, Horowitz in ACM Queue, 2012. See Figure 4. Double patterning with deep ultraviolet immersion lithography seems to be overcoming light wavelength problems. $\endgroup$ May 19 '13 at 3:05
  • $\begingroup$ thx; nice/relevant ref. there are many comparisons/metrics. if you ask me fig.3 performance vs normalized area looks sublinear esp wrt recent datapoints. thats my point that recent datapoints have to be weighed more to determine recent trends rather than curve fits over decades. a meta question "logical circuits" similar to whether this question is on topic $\endgroup$
    – vzn
    May 19 '13 at 18:40
8
$\begingroup$

I think you are right that the Heisenberg uncertainty principle is involved in the limit. I don't think we are anywhere near the Heisenberg limit. The limit we are currently approaching is that we currently compute with devices based on atoms. An atom is about 1 angstrom = .1 nm wide. We are currently making transistors that are about 22 nm wide, so we're getting close to needing to make 1 atom transistors. If we want to scale beyond that we're going to need to find ways of computing that involve storing and manipulating multiple states per atom.

The wikipedia article on fundamental limits of computation is a good starting point. I think what you really want is the Bekenstein bound.

The researchers that I know of in this area are Rolf Landauer, Charles H Bennett, and Seth Lloyd. (I think Charles H Bennett considers himself a computer scientist while the other two consider themselves physicists.) Richard Feynman was also very interested in this question and has a book called Feynman Lectures on Computation (which you can read about at http://quantum.quniverse.sk/buzek/zaujimave/p257_s.pdf). Also @PeterShor is an active member here and is the expert on quantum computation.

Here's a Scientific American article by Bennett and Landauer:

http://web.eecs.umich.edu/~taustin/EECS598-HIC/public/Physical-Limits.pdf

Here's Seth Lloyd's article in Nature: http://arxiv.org/abs/quant-ph/9908043

Here's a web page at Cambridge: http://www.sp.phy.cam.ac.uk/~SiGe/Fundamental%20Limits%20of%20Computation%20-%20Landauer%20and%20Heisenburg.html

$\endgroup$
1
  • $\begingroup$ Note that FinFETs cheat by being thin (dense in linear dimension, fast switching, etc.) but tall (helping avoid the effects of too few atoms). $\endgroup$ May 16 '13 at 20:52

Not the answer you're looking for? Browse other questions tagged or ask your own question.