Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Carbon Nanotubes Finally Outperform Silicon in Transistors (allaboutcircuits.com)
154 points by xd on Oct 14, 2016 | hide | past | favorite | 30 comments


I don't doubt that eventually our chips will be structured carbon circuits on a diamond substrate. I find it amusing that should a sentient AI arise post transition, it could reasonably think of itself as a carbon based life form :-).

The question I have not seen addressed is how we'll pattern the circuits. The ability to convert silicon into N-type and P-type material by doping is so "simple" compared to trying to induce the formation of CNTs of a particular geometry with the precision needed to develop a particular circuit. We are at the 'Bell labs' point where we can make individual (and volumetrically large) transistors out of CNTs or Graphene. But as we know from history, the revolution didn't really get underway until we could pattern a complete circuit on a substrate in the form of an integrated circuit. I'm still watching for the folks who come up with that breakthrough.


Interesting, not sure why shpx deleted their comment but it lead me to Max Shulaker who is figuring out how to build integrated circuits out of these things. Looks promising and I am surprised I missed it. Thanks shpx!

[1] https://www.researchgate.net/profile/Zuanyi_Li2/publication/...


> I find it amusing that should a sentient AI arise post transition, it could reasonably think of itself as a carbon based life form

The best of everything (X) ends up carbon-based. Life forms, computers, space towers, etc. It's just due to the carbon atom having a bunch of unusual properties.

(X) - whereas "everything" is an exaggeration used to make a point


I'm not sure if that sentiment isn't carbon chauvinism, given that the best we know so far has been carbon-based.

I'm sure the universe in its infinity has plenty of surprises in store for an species that has yet to leave their home planet.


There's no reason to expect far-flung corners of the universe to have different chemistry. We already know what the periodic table is for those unexplored regions of space, and they don't have anything cheaper and more versatile than carbon.


Maybe I'm underestimating the field, but I suspect there is still a lot to learn about chemistry in extreme environments (ultra hot, dense, etc), which would be most of the universe. Some experimentation is possible on Earth, but how much could we really learn about, say, liquid water if all we could produce were a few molecules in very restricted environments? I guess that's all kind of irrelevant until we're ready to try building transistors inside the Sun.


We already know enough about chemistry and physics to rule out the possibility of building transistors inside the sun, or any other useful permanent structure inside a star. At the opposite end of the spectrum, we're already commercially exploiting temperatures on par with deep space and extremely low pressures. Yes, there's a lot we don't know about the properties of things or conditions that are hard to keep around in large quantities in the lab. But we're not going to discover complex life or be able to engineer complex and cost-effective machines in conditions where there's so much energy that molecular bonds can't stay intact or where there's so little energy that nothing gets done at better than a glacial pace. There's surely interesting stuff left to be discovered in those fields, but we won't magically stumble across a plasma-based computing substrate that is better than what we can already do on Earth.


I'm not this confident on our knowledge of the high pressure environments inside stars.

We can rule-out unexpected simple chemistry on other planets and in the outer space, but the inside of stars is too different an environment.

Also, we surely can not rule-out anything from complex chemistry on any kind of environment. Macromolecules and non-repeating crystals are very badly understood.


We were also quite sure at one point that the earth was flat.


Asimov's response: http://chem.tufts.edu/answersinscience/relativityofwrong.htm

It is rather easy to collect sufficient evidence to conclude that the earth is not flat. The people who thought the earth was flat put very little effort into determining the shape of the earth.

Modern scientists have put a great deal of effort into determining the fundamental laws of physics. They have done a proper job of looking for evidence that the laws of physics can vary with location, and they've found none.


The most obvious check would be to go somewhere else and run some experiments. We have not done that.


Seeing as we're limited by the number of protons that can hold themselves together, I'm pretty sure we can discover most things here... at least in theory.


I'm sure that people will be able to create very fast carbon nanotube transistors (CNTFETs) in the future but that isn't currently the problem. Generally when you're creating normal silicon transistors (MOSFETs) one out of every billion or so is bad, meaning you have to fuse off some block of cache or maybe an entire core but multi-billion transistor chips are possible when you do stuff like that.

My understanding of the state of the art of CNTFETs right now is that only having one out of every 10 be bad would be a significant advance. So making 10 transistor circuits is maybe feasible but nothing larger. And even the 8086 had 20,000 transistors.

There's a lot of work that has to be done before CNTFETs are ready to be used in computers.


Silicon is not the best transistor material ever. For example, gallium arsenide can run at a switching frequency of 250 GHz thanks to faster electron velocity and higher electron mobility. https://en.wikipedia.org/wiki/Gallium_arsenide#GaAs_advantag...


You can create some pretty high frequency transistors (BJTs) with Indium Phosphide (InP) (ft > 600GHz, fmax > 1THz). See page 33 of:

http://www.ece.ucsb.edu/Faculty/rodwell/publications_and_pre...

...for examples of some 600 GHz circuits, and:

http://www.semiconductor-today.com/news_items/2014/OCT/NORTH...

...for a >1THz circuit using InP HEMTs.


Wow that's crazy.

I wonder how you even test a 1THz amplifier. Also what kind of natural noise might exist in those bands


> I wonder how you even test a 1THz amplifier.

The usual schtick is to measure the S-parameters a slower speeds and extrapolate out to the cutoff. (see chart 21 from the link above: http://www.ece.ucsb.edu/Faculty/rodwell/publications_and_pre...)

For the digital people, building a ring oscillator with N stages will cause the output to divide by N, and just make sure that f/N is in the range of your oscilloscope.

To actually measure things up in the THz range directly, there are more exotic methods: Superconducting bolometer is one that I've been involved with. But those are a PITA for a bunch of reasons.


>I wonder how you even test a 1THz amplifier.

Most likely using a mixer or some other non-linearity to down convert into the frequency range of your spectrum analyzer.


And how do you get the 1THz signal in the 1st place? Maybe upconvert as well?


Does 'Heisenberg' or some other theory limit the top frequency, if so, what's the limit?


I, for one, have a Germanium transistor in my 1964 Accutron Bulova watch. So there!

http://www.decadecounter.com/accutron/history.htm


Old joke: It's not that Germanium has a bad bandgap, it's just that room temperature is wrong.

[As it turns out, having a decent native oxide is also a winner.)


> thanks to faster electron velocity and higher electron mobility.

How does this effect energy transfer? This is the metric I assume most people care about as it is typically much faster than an electron can move.


"Best" is a funny word, because it means different things to people who care about different things. Silicon can be built with better gates / insulators than gallium arsenide, so if that's what you care about, silicon is better.



This is a really important advancement in circuits. We were reaching the physical limits of silicon and it was only a matter of time before we would have to find an alternative to continue the upward trends of transistor performance.


There are quite a few materials which outperform straight silicon, but it is hard to compete with the gigantic amount of existing silicon infrastructure, except for very niche applications.

https://en.wikipedia.org/wiki/High-electron-mobility_transis...


I thought it was innterconnect loss and heat dissipation that is keeping processor clocks slow, not intrinsic transistor speed.


Even if the science becomes feasible, this domain in encumbered with too many patents.


That'd explain why the military is funding it - patents don't apply to them!




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: