Slashdot Mirror
Can New Metal-Air Transistors Replace Semiconductors and Continue Moore's Law? (ieee.org)
Will Moore's law really come to an end by 2025? Maybe not...
An anonymous reader quotes IEEE Spectrum: [R]esearchers at RMIT University in Melbourne, Australia, believe a metal-based field emission air channel transistor (ACT) they have developed could maintain transistor doubling for another two decades. The ACT device eliminates the need for semiconductors. Instead, it uses two in-plane symmetric metal electrodes (source and drain) separated by an air gap of less than 35 nanometers, and a bottom metal gate to tune the field emission. The nanoscale air gap is less than the mean-free path of electrons in air, hence electrons can travel through air under room temperature without scattering...
Using metal and air in place of semiconductors for the main components of the transistor has a number of other advantages, says Shruti Nirantar, a Ph.D. candidate in RMIT's Functional Materials and Microsystems Research Group. Fabrication becomes essentially a single-step process of laying down the emitter and collector and defining the air gap. And though standard silicon fabrication processes are employed in producing ACTs, the number of processing steps are far fewer, given that doping, thermal processing, oxidation, and silicide formation are unnecessary. Consequently, production costs should be cut significantly. In addition, replacing silicon with metal means these ACT devices can be fabricated on any dielectric surface, provided the underlying substrate allows effective modulation of emission current from source to drain with a bottom-gate field. "Devices can be built on ultrathin glass, plastics, and elastomers," says Nirantar. "So they could be used in flexible and wearable technologies."
The article also suggests ACT devices could become important in space exploration, since electrons would be unaffected by extraterrestrial vacuums and radiation.
Nirantar was lead author on a new paper published in Nano Letters, and believes that their new approach "means we can stop pursuing miniaturization, and instead focus on compact 3D architecture, allowing more transistors per unit volume."
An anonymous reader quotes IEEE Spectrum: [R]esearchers at RMIT University in Melbourne, Australia, believe a metal-based field emission air channel transistor (ACT) they have developed could maintain transistor doubling for another two decades. The ACT device eliminates the need for semiconductors. Instead, it uses two in-plane symmetric metal electrodes (source and drain) separated by an air gap of less than 35 nanometers, and a bottom metal gate to tune the field emission. The nanoscale air gap is less than the mean-free path of electrons in air, hence electrons can travel through air under room temperature without scattering...
Using metal and air in place of semiconductors for the main components of the transistor has a number of other advantages, says Shruti Nirantar, a Ph.D. candidate in RMIT's Functional Materials and Microsystems Research Group. Fabrication becomes essentially a single-step process of laying down the emitter and collector and defining the air gap. And though standard silicon fabrication processes are employed in producing ACTs, the number of processing steps are far fewer, given that doping, thermal processing, oxidation, and silicide formation are unnecessary. Consequently, production costs should be cut significantly. In addition, replacing silicon with metal means these ACT devices can be fabricated on any dielectric surface, provided the underlying substrate allows effective modulation of emission current from source to drain with a bottom-gate field. "Devices can be built on ultrathin glass, plastics, and elastomers," says Nirantar. "So they could be used in flexible and wearable technologies."
The article also suggests ACT devices could become important in space exploration, since electrons would be unaffected by extraterrestrial vacuums and radiation.
Nirantar was lead author on a new paper published in Nano Letters, and believes that their new approach "means we can stop pursuing miniaturization, and instead focus on compact 3D architecture, allowing more transistors per unit volume."
There is air and electrons in the gap so there must be quite a climate there, changing extremely rapidly.
Do you think climate change scientists care about brand stamps?
Pardon the non-English speaker, but what's a brand stamp?
Besides home and professional appliances are fast and small enough today, even scientific appliances are "good enough" today...
This reminds me of what happened with NAND (i.e. flash memory) a few years ago. Ever-smaller transistors hit a wall due to endurance problems (each one could only be reprogrammed a few hundred/thousand times), so they went back to larger transistors but started stacking them into layers. Now we're at ~96 layers, and it's expected that a few thousand layers is feasible.
The problem with layering in CPUs is how hot each layer gets, and adding new layers is unlikely to help single-core performance beyond what cache can do. So, we're going to end up with low-clockspeed (to minimize heat) thousand-core CPUs... which will actually be perfect for GPUs, not so much for that single-threaded productivity task. I could also see this being used for HBM, which is already stacked.
Corruption is convincing someone that the selfless ideal is the same as their selfish ideal.
They mean moving from 2D to 3D chips.
>"The nanoscale air gap is less than the mean-free path of electrons in air, hence electrons can travel through air under room temperature without scattering... "
And what about when not at room temperature? Seems like that little disclaimer could be what makes the whole thing impractical. A chip/board isn't going to be made up of ONLY these "metal-air" transistors, so it is going to generate a significant amount of heat or be near something that does. Plus, there is the overall environment in which the device will be used that needs to be considered. The article doesn't elaborate on this at all.
Can always wait a while and then the market will take over, allowing participants to buy at their leisure under their terms - nobody wants a chip with a ridiculous overclock governor right? Moore's law might suddenly apply again as long as making chips is profitable.
Yes. Computers are still too slow.
SLOWER TRAFFIC KEEP RIGHT
Why do Slashdot editors insist on making headlines into questions that aren't answered in the article? A headline is a super-short summary the story. The story isn't a question, so the headline shouldn't be, either.
The story is, "Researchers believe new metal-air transistors could continue Moore's Law". It isn't a debate on this belief of those researchers.
Oh, wait - this is a click bait tactic used to make something seem more interesting than it really is...
1) Moore's Law is already dead
2) Air-gap transistors have been around since mid 1990s.
3) No offense, but it is doubtful such a breakthrough would come from some university I have never heard of in Australia. Based on their Wikipedia page they are known for art and design.
In scanning tunneling microscopy under vacuum, the metal tip usually had to be a few nanometers away to observe a decent tunneling current under normal bias. Tunneling 30-nm in air? I am not sure how that makes sense. Best read the paper when I get the opportunity.
what's a brand stamp?
It's like a tramp stamp, but made with fire.
W..w..W - Willy Waterloo washes Warren Wiggins who is washing Waldo Woo.
There are plenty of areas in personal tech that could certainly use a huge jump in speed and/or density.
Virtual reality, for example - a tenfold (or more) increase in graphics processing power would make personal VR amazing instead of just fun. Standalone setups like the Oculus Go could have 4k-per-eye graphics, with high frame rate and roomscale tracking.
Transistors replaced vacuum tubes - which where exactly as described above: metal electrodes separated by a gap (vacuum not air) with a third electrode plate to control current
So now vacuum tubes replace transistors!
Wow!
Amdahl's Law proves that a single core running as fast as possible is where it's at.
This is about the laws of marketing, not physics.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
That move has happened a long time ago.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
Computers will not get much faster. You can get more cores and less power usage, but that is essentially it.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
That was my very first thought. Except it's not vaccuum. Of course neither were vaccum tubes. You had to lower the pressure to increase the mean free path. But if you could make this small enough then you could just do it right in the air. And by going to high fields you get to replace therm ionic emitters with field effect emitters. So less heat. And again to get high fields at low voltage you need to go small.
Some drink at the fountain of knowledge. Others just gargle.
Are you a bot or just slightly insane?
The answer is that soon the Internet will be running on a series of tubes.
In essence these seem to be very similar to vacuum tubes indeed. The question is, can you make digital circuits equivalent to CMOS with these? Because it's no effing good for digital circuits if you can't have zero static power consumption. I don't think you can do it with regular valves, because there is no such thing as complimentary valve, are these any different? You could certainly make TTL like circuits with these, but what good is that for CPU-s and such?
Computers will not get much faster. You can get more cores and less power usage, but that is essentially it.
Single thread problems and cores to solve them, sure, but we've moved into the thousands of cores era, just by way of GPU.
The first time I switched to a CUDA implementation of Seti@Home, I doubled my previous 10 years of calculations in 2 weeks. Hence (bonus unpredictable value of discovery aside) we might as well have not bothered.
And the upgrades I made doubled and tripled the rate again, with no end in sight.
The Seti guys would probably do better designing a proper product that automatically didn't interfere with 3D games and Netflix framerates rather than relying on manual fine tuning, which doesn't really work, and many are not capable of, or the clumsy "only when computer not in use" setting which, being a wrapper around an internal setting and not a defined feature, cuts out surfing with browser but leaves in Netflix, where you don't touch the compterand hence it is "not in use."
(-1: Post disagrees with my already-settled worldview) is not a valid mod option.
Whether it is overkill is irrelevant -- if that's the cheap standard, that's what you use, like terabyte thumb drives (if even us anymore) vs. 256 meg ones.
(-1: Post disagrees with my already-settled worldview) is not a valid mod option.
Moore's law is an observation about being able to reduce feature size via photolithography. These are bigger than existing features by a fairly big factor! The limit is no longer even the photolith, it's the wavefunction size of the electrons - eg tunneling, and ballistic transport so they won't go around sharper corners. So, like virtually all press-release "science" and any article with a question in the title, the answer is somewhere between "no" and "WTF, obviously no".
It'd be cool if you could make diodes without any forward drop for Maxwell's demon applications, but that's a pretty marginal side case.
Why guess when you can know? Measure!
Hmm, how do you make a CMOS pair out of that transistor?
I wonder what Microsoft will think up to waste all the additional compute power? Maybe preemptive aeroglass?
Moore's law is an observation about being able to reduce feature size via photolithography.
No, it isn't. It's an observation about the number of gates. It doesn't matter whether you decrease the feature size, or increase the number of layers, or just make a bigger die. All of these, of course, have happened over the years.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
(I'd do it myself but I'm abandoning my mod points in this discussion in order to make a relevant posting.)
Bantam Dominique roosters crow a four-note song. Once you've heard it as "Happy BIRTHday" you can't NOT hear it that way
Moore's Law potentially has a long way to run - because semiconductors are still only a few layers thick in the Z axis and there are a lot more doublings left before we're dealing with "chips" that are solid circuitry feet on a side. Non vacuum "vacuum tubes" are far less sensitive to high temperatures than semiconductors, so building 3-D structures of them won't have as much of a cooling problem. (You still need to dissipate all the heat, but you can let the structure get 'way hotter to encourage it to migrate out.)
Single Threading speed may be falling off its free ride on Moore's Law-like exponential scaling, as speed-of-light and electron-size leakage limits raise a wall. (Going 3-D will help some, by shortening paths, but not by a lot.) But lots of really useful computations are massively parallelizable. The should drive continued manufacture and deployment of higher-switch-count devices as the technology is developed and yields are brought up.
Bantam Dominique roosters crow a four-note song. Once you've heard it as "Happy BIRTHday" you can't NOT hear it that way
Sorry, but your graphs need work.
OTOH, even though Moore's law has hit a pause, that's happened before, and then a new technology showed up that reinstated it. The current problem with that happening is that local processing is sufficient for most current uses with current technology. Some new application will probably be needed to change that. It'll probably be called AI, but what will be meant by that is a bit unclear. One good candidate is self-driving cars. They would benefit immensely from smaller computers that were less power hungry. And there would be huge numbers of them sold.
I think we've pushed this "anyone can grow up to be president" thing too far.
As originally stated, I don't think a larger die would count as facilitating the continuation. 3-D construction would, though. And if they run enough cooler, then that would work.
I think we've pushed this "anyone can grow up to be president" thing too far.
It seems rather silly. It’s not a statement of some absolute scientific truth - nothing really depends on it holding true or not. If Moore’s Law stops being true, it’s not as if Intel or TSMC or Samsung is going to be shuttering factories because their fabs won’t work anymore. Jony Ive won’t descend into madness because he can’t make things any thinner. Nothing practical will actually change, and technological development will continue to progress.
#DeleteChrome
Virtual reality is a real possibility, but they've got to resolve the vestibular canal disagreeing with the eyes about what's happening first, so people don't get nauseous. Some people can deal with it, but most can't without a lot of training, and some never can. And among those who can, a lot don't want to. Sea sickness isn't pleasant.
I think we've pushed this "anyone can grow up to be president" thing too far.
Moore's paper is freely available. He didn't state "Moore's law" of course, but he does talk about economic factors. The graph that's usually taken to be the statement of Moore's law (at least, Wikipedia thinks so) simply shows year versus number of components per integrated function.
So there's no reason bigger dies wouldn't count, although they shouldn't be any more expensive than the older, smaller ones were.
https://drive.google.com/file/...
I've usually heard it as something about transistors per unit area. To tie it to some particular phrase he said is probably incorrect, as it was something he developed as a prediction over a number of years. It was clear at one point that he thought that 2-1/2 D construction was consistent with his prediction. But he never indicated that he thought that larger dies were significant, even though several sizes were used during his period of activity.
OTOH, I'm certain that he said something essentially similar with different words several times. And Moore's law was proclaimed as such by the technical press, so and refined statement of it shouldn't be believed. He developed it as an operating and predictive principle. It worked, so he kept using and developing it...and changing the details.
I think we've pushed this "anyone can grow up to be president" thing too far.
Alas, modern small high speed transistors are not zero static power consumption. It's a substantial problem that plays a part in the speed versus power tradeoffs.
Contribute to civilization: ari.aynrand.org/donate
Parallelizable loads are a small faction of all computing loads.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
Inapplicable quote is inapplicable. Obviously I was talking about the current tech, i.e. silicon and electricity. Nothing else is currently on the horizon (no, Quantum Computing is not going to work and even if it works, it is basically useless for most tasks), and hence we are stuck with this for the time being.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
When? The only commercially available 3D chips that I can find are memory chips, and they are just stacks of 2D chips.
They may be a small fraction, but they consume the most computing resources. The Pareto Principle and all that.
Not so much similar as identical to vacuum tubes (or valves as we call them this side of the pond). It's just that the scale is so small you can let the air in without stopping the flow of electrons.
I guess computers will now be not only smaller, faster and sexier computers, but also sound warmer with more detail. All we need now is the miniature green felt tip pen :-)
Computers will not get much faster. You can get more cores and less power usage, but that is essentially it.
That just makes me sad. Computers are still slow as shit & the software (fucking appity APPs!) is even slower. I am always waiting on something.
SLOWER TRAFFIC KEEP RIGHT
That's exactly what I said.
You need some water with that drinkypoo, or a citation, like this one: https://en.wikipedia.org/wiki/... Which is limited by photolith tech. And now, worse yet, quantum effects and limitations of silicon itself due to attempts to work around heat. They went to lower voltages to reduce the loss due to switching the effective capacity of the conductors. To do that they had to dope the transistors such that they could turn on at lower voltages. THAT resulted in transistors that also didn't turn off all the way, so unlike the CMOS of old, these draw some power even when not switching. So as density went up, heat issues reappeared. Going a little past 2d (finfets) helped somewhat, but it's not the complete answer. The limits to going further are things like ballistic electron behavior - they shoot off conductors instead of going around tight corners, and also simply fail to be contained on too-narrow tracks because their wavefunction is wider than that.
The "law" has been dead for awhile now, and so has the usual speed increase. I have no more ghz now than I did years ago, far from doubling, it's only tripled in a decade or more. We always hear about these magic new semiconductors that can go much faster, but there are zero high density chips that use them. Why? You can't make complex circuits with lithographic processes using them.
Maybe you should stick to stuff you know better. Or you might get whipped.
Why guess when you can know? Measure!