Less Wrong is a community blog devoted to refining the art of human rationality. Please visit our About page for more information.

How does remote Joule heating of carbon nanotubes advance singularity timelines?

-2 Post author: keefe 11 April 2012 04:15PM

 

Carbon nanotubes: The weird world of 'remote Joule heating'

Minimizing Joule heating remains an important goal in the design of electronic devices12. The prevailing model of Joule heating relies on a simple semiclassical picture in which electrons collide with the atoms of a conductor, generating heat locally and only in regions of non-zero current density, and this model has been supported by most experiments. Recently, however, it has been predicted that electric currents in graphene and carbon nanotubes can couple to the vibrational modes of a neighbouring material34, heating it remotely5. Here, we use in situ electron thermal microscopy to detect the remote Joule heating of a silicon nitride substrate by a single multiwalled carbon nanotube. At least 84%of the electrical power supplied to the nanotube is dissipated directly into the substrate, rather than in the nanotube itself. Although it has different physical origins, this phenomenon is reminiscent of induction heating or microwave dielectric heating. Such an ability to dissipate waste energy remotely could lead to improved thermal management in electronic devices6."
These experiments seem extremely important in constructing AI singularity timelines, what does lesswrong think?
What is the impact on life extension research?

Carbon nanotubes in biology and medicine: In vitro and in vivo detection, imaging and drug delivery

I don't see a line of inference free of holes, but one can imagine the sort of uses a scifi author would come up with. 

 

Comments (20)

Comment author: Vaniver 11 April 2012 04:19:21PM *  11 points [-]

I think that outside view estimates- "here's Moore's Law, this is the point at which the processing power of a human brain will cost the equivalent of $1,000 in 2012 dollars"- are way more robust than inside view estimates- "we just came up with subtechnology 734 out of ~5k necessary."

Comment author: keefe 11 April 2012 08:33:47PM *  -2 points [-]

it seems much more important than tech 734/5000 necessary... carbon nanotubes are one of the core scientific discoveries of our generation and this shows a really interesting property of them directly related to electronics development. The heat dissapation bottleneck has been the most serious issue with nanotech and much, much faster and smaller processors. When we went from faster chips to multiple cores, things became really different - parallel algorithms are inherently more difficult and tech that could reinstate an exponentiation phase is extremely significant. This is more important than that though, if substantiated it is a truly weird physics advance and there's no telling what applications it will find. The first AI or human decision support system that will offer dangerous self improvement capabilties is most likely going to be on some $10M-$100M system and the question is how do you see that coming? I mentioned moore's law as the first of many obvious and important areas this advance will impact if there is not some serious engineering bottleneck in putting it into practice, be that in moving up orders of magnitude in clock speed or providing wiring for brain implants that is small enough not to damage neural tissue and so forth. I'm seriously surprised to see a response to this advance that is not at least curious interest at an obviously related physics advance.

Comment author: Vaniver 11 April 2012 08:45:39PM 2 points [-]

I'm seriously surprised to see a response to this advance that is not at least curious interest at an obviously related physics advance.

There are two different things going on here; one of them is that nanotechnology has potential and is interesting, and the other is estimating if/when the next Singularity will occur.

The second is done best by taking the outside view. An estimate that the Singularity will happen in 2100 assumes there will be many technological improvements between then and now- both big and small- and so doesn't depend on the details- that we fixed the heating problem now and the parallelism problem ten years from now, or did them in the reverse order.

The first is interesting, but should be separated from the second.

Comment author: Vladimir_Nesov 11 April 2012 10:41:56PM 3 points [-]

the next Singularity

Ouch

Comment author: keefe 11 April 2012 09:02:42PM 0 points [-]

"Taking the outside view means using an estimate based on a class of roughly similar previous cases"

so the singularity by far is something after which we cannot predict how things are, but we're going to look at roughly similar cases for that?

I'm also an insider in this in the sense that I've been a professional software engineer for 16 years, dropped out of a phd program after passing qualification exam with a masters in compsci and eng, so yes, I am trying to imagine possible outcomes and look at trajectories and I hope other people with training on this board are doing the same.

Comment author: Vaniver 11 April 2012 09:19:32PM *  1 point [-]

so the singularity by far is something after which we cannot predict how things are, but we're going to look at roughly similar cases for that?

The comparisons people generally make are to agriculture and industrialization.

I'm also an insider in this in the sense that I've been a professional software engineer for 16 years

Okay. Part of my academic background is physics, including nanoscale physics- but if anything, being half-educated about it makes me reluctant to speculate.

For example, there's a technology under development which would use nanotubes and van der Waals forces (if I remember correctly) to do binary memory on a scale that's a massive jump from what we have now- I think the claim was they could store a petabyte in the volume of a dime. If that works, that'll be huge- you could significantly change computer architecture with the ability to store abundant memory on the same chip as the CPU, for example. But I'm reluctant to bet that it'll work until it works.

Comment author: keefe 11 April 2012 10:03:03PM -1 points [-]

So if you have a background in nanotech and I have compsci, it seems like speculation could generate ideas.

I think that as a community interested in safety, it's important we keep informed about the advancement trajectory. Understanding limitations and capabilities of fundamental science advancements also provides intelligence on companies to watch for, tech that is likely available soon and so forth.

so, why not speculate? It's almost free to scan an idea for value.

Comment author: Luke_A_Somers 12 April 2012 09:11:02PM 1 point [-]

It's interesting and I hadn't thought of it, but it's not weird. The losses are from coupling between the substrate and the carrier electrons, so it makes sense that the energy will go there.

Comment author: keefe 13 April 2012 01:46:04PM 0 points [-]

Would you talk more about the coupling between substrate and carrier electrons, that is not clear to me.

I mean it makes sense that it went somewhere nearby, but why would it transfer at all, only with these particular materials?

Why isn't it weird to you? If I got a lab report like that i'd be like ok, go ahead and rerun those experiments...

Comment author: Luke_A_Somers 14 April 2012 12:48:49AM 2 points [-]

What kind of explanation are you looking for? Are you a physicist? Have you taken physics?

Assuming you know some things but aren't a condensed-matter physicist: 1) single-layered graphene's intrinsic scattering rate is extremely low, so nearly anything else that causes scattering is going to dominate. 2) it is an exceptionally poor screener of electric fields, due to the low density of states. 3) it is entirely surface, so there is no interior region to be unaffected by boundary effects (though graphene is usually in poor mechanical contact with the substrate, this loose coupling is not weak enough to strongly suppress surface phonon scattering). Note that 'the top side, away from the substrate' is not distinct - the relevant carrier electron states straddle the center plane.

So, what happens? You've got an electron in the electrical field, accelerating. It eventually scatters. What does it scatter off of? The substrate, mainly, by one mechanism or another. Whichever it is, some energy is dissipated directly into the substrate (that's what it bounced off of), and the electron bounces off in some other direction. This electron bounce is not heat - the electrical field just goes back to pushing it forward again, and it goes - very orderly except for the isolated scattering instances.

The main ways the graphene itself gets warmer are by a) a carrier electron does manage to scatter off of a graphene lattice phonon (this is a really weak process, but it happens, and when you get rid of the substrate it dominates) ; and b) phonons from the substrate are transmitted into the graphene lattice (this is also weak because graphene binds poorly to the substrates mechanically speaking, but it's not extremely weak).

That's the summary of what's going on. It applies equally to single-walled carbon nanotubes, and to a lesser extent, multi-layered graphene and multi-walled carbon nanotubes.

Comment author: keefe 14 April 2012 05:24:31AM 0 points [-]

Thank you or the well considered response, actually helpful. You have my background about right, I published in a physicists in medicine conference and have the normal background in comparch and whatever classes I took for my math double. Definitely not a condensed matter physicist, will have to read more on phonons.

The idea that this is a hollow tube and so there is no interior region to be effected does seem intuitive. The thing that jumped out at me is that the tube itself remained cool.

I don't have a good understanding of quantum electrodynamics or phonons and that is one reason I wanted to bring this up for discussion. Some types of scattering like bremsstrahlung seem like they play a role, but it doesn't seem to explain it, from the lead scientist:

""We believe that the nanotube's electrons are creating electrical fields due to the current, and the substrate's atoms are directly responding to those fields," Cumings explains. "The transfer of energy is taking place through these intermediaries, and not because the nanotube's electrons are bouncing off of the substrate's atoms. While there is some analogy to a microwave oven, the physics behind the two phenomena is actually very different."" http://newsdesk.umd.edu/scitech/release.cfm?ArticleID=2657

The normal mechanism for heating in electric current transmission as I understand it, the electrons are bouncing off other atoms causing them to vibrate. So we make a transistor or a wire and we a pass a current through it, the atoms inside get hotter and we then dissipate that heat. They don't appear to think this is what is going on here.

It seems like the electrons go through the nanotube wire and the energy kind of jumps from the current to the tube to the substrate it's laying on without accumulating much inside the wire itself.

They claim this is a weird game changer which every scientist wants to find, so either that's hype or it's legit. It seems like you have a good understanding, this is in a discussion area, what's your opinion, is this a big discovery that is going to lead to multilayer chips orders of magnitude faster or is it just a fluke thing?

Comment author: Luke_A_Somers 16 April 2012 02:56:14PM 1 point [-]

Phonons are the quanta of lattice vibration. Vibrations are one kind of imperfection in a lattice.

Electrical carriers (electrons and holes) do not scatter off of individual lattice atoms; they scatter off of lattice imperfections. These imperfections can be defects, or phonons, or other carriers.

On to the claims in the article... as before. Usefulness? I'm not really sure. If you're always using these nanotubes to carry currents, their entire environment will heat up (the heat will leak into them eventually), so it seems like it would only really help in cases where it needs to carry a current spike.

Comment author: keefe 18 April 2012 06:35:08PM 0 points [-]

If yi would have to do more reading to understand the lattice stuff, it seems reasonable though.

As far as usefulness, the idea I had was you could layer a substrate to dissipate the heat really well. My limited understanding is 85% of the heat jumps the wires somehow and is absorbed by the substrate, which could be engineered arbitrarily. This is important because cnt are very good electrical conductors so you could pair them with a good head absorbing substance and achieve separation of heat and current in ways we have not seen before, which one could speculate as a way to restart moores law progression of speed.

Comment author: Manfred 11 April 2012 06:06:05PM 1 point [-]

The overall impact seems small to clock speeds, since it's speculative and there are several possible ways the same thing could be done. And the impact to clock speeds on singularity timelines is itself pretty small, since once you have working code, making it faster or raising money for more hardware seems like a much smaller problem.

Comment author: keefe 13 April 2012 01:50:51PM 0 points [-]

I think people underestimate the intrinsic computational complexity in solving even relatively simple pattern recognition tasks. There are also all sorts of algorithms where you do a heuristic search through some big space and it's particularly interesting to note that a lot of programs for finding proofs or optimizing code are in this class. Anybody who thinks computers today are fast doesn't write enough code.

I have an intuition this particular tech(or related) is going to advance us to the next exponentiating phase of a stacked sigmoid advancement curve that eventually leads to ai.

Comment author: keefe 13 April 2012 01:57:06PM 0 points [-]

Curious if people would be willing to articulate negative sentiment on this piece?

It seems like we should all see advancements like this as a way of training our intuition about how the tech tree will go and also make efforts to do outreach into important communities as they are growing. If graphene transistor and remote cooling cpus eclipse efforts in parallelism or biological computing, then researchers in that field have a lot of influence to spread to users as well as developers.

Also, to most people this is a highly counterintuitive phenomenon and some people I hung out with for a while used to talk a lot about the utility of physics intuition.

Some hypotheses for negative affect are that the story is not relevant or interesting (strong evidence this is not the case), that those people who have the technical expertise to discuss it don't want to discuss it off secure channels or...

Comment author: Viliam_Bur 13 April 2012 02:51:42PM *  1 point [-]

My reaction: You have posted a link to some invention, without explaining why it is related to singularity more than any other random technical invention. Sure, there are three hyperlinks where I can get more information, but I would like to get some explanation from the article itself. Or maybe you just use "how does X advance singularity timelines?" as a synonym for "X is cool"; I don't know, and you didn't help me avoid this suspicion.

What would I like to find in articles like this? A short explanation of context (for people who know nothing about carbon nanotubes, which is not a typical LW topic), and explanation of why do you think this invention is exceptional (even when compared with hundred other inventions made and published in the same year). Something like this -- the following text is completely made up, just to better illustrate what I mean:

Estimates of Singularity timeline are often based on the Moore's law. However, in recent years the progress in computer speeds has slowed down. Computers are not getting faster anymore. Their increased power is mostly gained by adding more processor cores, which is not the same as making faster cores. Also the electric power consumption increases linearly with the number of cores, so even if in 2050 we get a computer capable of simulating a human brain, it would require more power than is produced by Sun. One possible solution is to avoid an integrated-circuit approach and build the computer directly, atom by atom. (Just like we got a thousandfold increase in capacity and speed by replacing vacuum tubes by integrated circuits in 1960s.) To make it possible, we need a material that is able to do -- blah blah blah -- but material with such properties is not known yet. However, recent experiments with carbon nanotubes suggest that they are similar to what we need. More details about physical properties of carbon nanotubes are here: [1], [2], [3].

Any you don't have to put "Singularity" in the name of the article. "Recent advances in carbon nanotubes research" would work fine.

Comment author: keefe 13 April 2012 07:36:16PM 0 points [-]

Or maybe you just use "how does X advance singularity timelines?"

Very much not this. I was a visiting fellow at singinst and discussed timelines with many people. I still feel some level of ethical obligation to provide a more complete analysis as I was actually converted to have some worry about this recursive self improvement, though I tend to worry more about IA than AI (even if just because of the IA-->AI path) I'm also poking a little bit at the LW codebase again and wanted to try to stimulate a discussion and explore the site. I was looking for actual discussion of the impact of this on people's timelines, how they are updating on this kind of evidence.

Any you don't have to put "Singularity" in the name of the article.

It's interesting to note that I put singularity and timelines plural, implying this trajectory towards "methusalarity" or whatever you want to call it. I linked the other article to show there is some evidence that this particular advance might very well unlock a lot of tech along a lot of interesting, rapidly accelerating change technologies.

A short explanation of context

I posted the most relevant abstract. If they spent some fairly large time writing that, why should I assume I can do better? If you check out the link, this was published in nature nanotech, so that is pretty big news. I linked to papers on tissue engineering, nanotech used internally. That's three pieces of a puzzle that we're all trying to get a better handle on.

Something like this

I think that's a good introductory description, but I thought people on this site already know all this. I think they know moore's law has stalled and that we're looking to increase it, that we're right on the cusp of the computational power required to search solution spaces near the human scale (watson, go ais, deep blue, etc)

So, what techs will change that space... carbon nanotubes and graphene wires are taken pretty seriously. If our mutual goal is to get ahead of this problem, it seems we should all take the time to do the basic reading and you come to certain concepts - heat dissipation, scaling out rather than up, 3D chips, transistor sizes, nanotechnology, graphene and carbon nanotubes, cloud computing, bot nets, cortical columns, human computer interface, neuronal destruction due to wire thickness and so forth.

It is indeed kind of mean that I ask how this piece fits into everyone's puzzle without disclosing my puzzle, but I do worry about security and that sort of thing and this was a good experiment on how people would respond, thank you for your insights, need to run back to code...

Comment author: Viliam_Bur 14 April 2012 10:16:46AM 1 point [-]

Thanks for explanation. Now my model is that you consider carbon nanotubes very important technology for increasing computational power (more than many other hardware-related technologies), so they are also a very important component in calculating Singularity timeline. Makes sense, though I lack the knowledge necessary to discuss it.

I posted the most relevant abstract. If they spent some fairly large time writing that, why should I assume I can do better?

Because they were writing for a different audience -- for people who already know a lot of context.

Now it's your choice whether you want to discuss with an average LessWronger, in which case you should provide more content, or you want to discuss only with people who are experts in some area, which is completely legitimate, but perhaps you should state it more explicitly.

It is indeed kind of mean that I ask how this piece fits into everyone's puzzle without disclosing my puzzle, but I do worry about security and that sort of thing and this was a good experiment on how people would respond

That's good. I was hoping you will not get the impression that we are somehow biased against discussing singularity or carbon. :D

Comment author: keefe 14 April 2012 02:14:15PM 0 points [-]

I think it tends to be most useful to do things that support multiple different plans, so I had a lot of motives for putting this up here. I don't have a lot of time, so that's how I try to roll anyway...

Here is what my motivations were :

  • In general play with the site - I did a little bit of work on it in the summer of 2010 and I was talking about making a contribution again, so I wanted to play with the interface and understand the code better.

  • I enjoyed my time at singinst, but was one of the rare people that had never read lesswrong when I arrived there (having found the post on hacker news) so it was an effort to reach out to the community

  • I deliberately didn't write much about a topic that I have reason to believe should be interesting to this community, so it was a good level set for the community's tech level, curiosity and so forth. Definitely more open to stuff than the average forum, but was surprised, would have thought people would be all over this as a weird idea.

  • definitely not just looking for expert opinions, I do think CNTs are very important and I want to encourage you and everyone else to let your imagination run wild a bit and say what you think, if it's too dangerous to discuss fine, but I doubt most think that...

  • for experts that want to set me straight or point out some obvious reason these techs are not useful, let me know but since we are already using CNTs to replace thermal paste for heat sinks so can we do a full chip etc...