Leakage Problems Solved going smaller

[Sxotty]

According to the inquirere here http://www.theinq.com/?article=25512
The leakage problems have lessened when going to 65nm and evenmore going to 45nm. This makes no sense to me, but perhaps it is true. If it is in fact that case wow! That will really give the business a boost, it seemed that CPUs were stagnant and we had little hope for real performance increases, but if this pans out there could be a massive performance leap again.

Think happy thoughts here people, from what several sources have told the INQ, the leakage problem is solved, and I mean solved, not lessened. This will be a massive gain for Intel, and unless AMD and IBM can match it, it will pretty much hand it the mobile space, not to mention anything else where power matters.

 

[ANova]

If this is true, I wonder how the dual core Pentium Ms are going to perform on 65/45 nm.

 

[Sxotty]

Well yeah, I think that the Inq was very shortsighted tho by saying "hand it mobile space"

It will hand it the whole CPU industry if others cannot match it IMO.

But then I believe that otheres will be able to come close to matching whatever exactly it is that allowed this. Hopefully AMD won't have to liscence it

 

[hoom]

Seems odd that it can't be applied back to 90nm or 65nm though.

I think I'll believe it either when I see other people saying it or I see it in action.

I wonder if it could be some kind of wave interference type thing where 90nm & 65nm just happen to have the wrong spacing (constructive interference while at 130nm & 45nm it gets destructive interference) or something?

 

[Killer-Kris]

This is all actually old news, trigate, high-k, and I can't remember what but there was a third option Intel was looking at. That's not to say that leakage won't become a problem again as the process continues to shrink, it's just solved for the mean time.

And I think the reason it isn't "back-ported" to the larger processes, is likely because the whole process would have to be redone (killing yields), and not to mention that you'd have to do a new tapeout of your processor design. So it's easier to just do it all at once on a new process with a new design. That way it's just part of the regular cost, instead of adding on to it. If that all makes any sense.

 

[The_Wolf_Who_Cried_Boy]

As Killer-Kris said this isn't a sudden Inq. revelation.
Intel have been explaining 45nm will be a big improvement for several years now. The big fix is changing the gate dielectic from silicon oxide to exotic zircon alloys and somesuch.
Browse through here for a bit of substance to Charlie's fluff.
http://www.intel.com/technology/silicon/micron.htm#high

...and unless AMD and IBM can match it...

Um, yeah, right, it's not like IBM invented SOI or silicon straining or anything, what a shock it will be for them....

 

[3dilettante]

I recall AMD slides discussing metalized gates and raised fin transistors for future process nodes. It could keep Intel's competitors within the same ballpark (better or worse, I don't know), though without more details it would be difficult to say how tight the race could be.

 

[Frank]

I don't believe it's solved. That's pretty much impossdible, quantum tunneling being what it is. It might be much improved, but that's it. To solve it entirely, you would need to use that quantum tunneling effect to create your gates, which wouldn't be transistors or FET's as we know them. Which might be what they mean, but I doubt it. That would be entirely new technology, who knows how to build a processor out of that?

The current steppers use lasers with a wavelength that is larger than half the smallest size of the structures, so the tolerances are much larger than needed for the fine control needed to create the quantum wells, dots and tunnels. The problem with using smaller wavelengths is that you cannot focus them with lenses anymore, and that there aren't any good mirrors to reflect them. And you need a plasma with a temperature like the Sun to produce the needed radiation.

So let's wait for the new generation of steppers and chips before we proclaim it solved.

 

[PC-Engine]

http://www.necel.com/en/process/lowpower_overview.html

 

[bloodbob]

High-k gates also mean lower clock speeds which = less preformance. Digiguru we haven't hit qutuam tunnel effects atm it simply classical resistance problems.

 

[KimB]

I don't believe it's solved. That's pretty much impossdible, quantum tunneling being what it is. It might be much improved, but that's it. To solve it entirely, you would need to use that quantum tunneling effect to create your gates, which wouldn't be transistors or FET's as we know them. Which might be what they mean, but I doubt it. That would be entirely new technology, who knows how to build a processor out of that?

Edit: yeah, just read up on leakage currents in semiconductors, and bloodbob's right, tunneling isn't the issue, not yet...

Actually, now that I think about it, tunneling effects will serve to effectively place an absolute barrier on the distance between conducting lanes. Basically, the amount of tunneling goes exponentially with the distance. So if tunneling starts to become a factor, you just can't get any smaller at all or it'll suddenly be a huge component.

Resistance, on the other hand, is linear with distance, and thus its onset is much more gradual and manageable by changing things like the materials used.

And you need a plasma with a temperature like the Sun to produce the needed radiation.

Er, there are many ways to create high-frequency coherent light other than plasmas. You can bombard an appropriate surface with electrons (similar to how TV's work). You can apply large accelerations to charged particles to get them to radiate (The Advanced Light Source over at Lawrence Livermoore uses this to produce X-ray radiation for experiments). If an appropriate semiconductor can be found with the right band gap, you can build an LED-based laser.

Note: the first two methods above just create the light: you have to filter it to make it coherent, which means that they're very energy inefficient. Plasma is the same way, of course.

 

[Frank]

Edit: yeah, just read up on leakage currents in semiconductors, and bloodbob's right, tunneling isn't the issue, not yet...

Yes, but it's ruled by uncertainty. The effect is already there, but "tunneling" means all of the electrons decide to pass the insulator together. Individual electrons already do so.

But the idea was, that you cannot solve leakage as long as there is tunneling around the corner.

Actually, now that I think about it, tunneling effects will serve to effectively place an absolute barrier on the distance between conducting lanes. Basically, the amount of tunneling goes exponentially with the distance. So if tunneling starts to become a factor, you just can't get any smaller at all or it'll suddenly be a huge component.

Resistance, on the other hand, is linear with distance, and thus its onset is much more gradual and manageable by changing things like the materials used.

Yes. But the only way to solve it would be to use superconductors and almost perfect insulators, until you encounter tunneling.

Er, there are many ways to create high-frequency coherent light other than plasmas. You can bombard an appropriate surface with electrons (similar to how TV's work). You can apply large accelerations to charged particles to get them to radiate (The Advanced Light Source over at Lawrence Livermoore uses this to produce X-ray radiation for experiments). If an appropriate semiconductor can be found with the right band gap, you can build an LED-based laser.

Note: the first two methods above just create the light: you have to filter it to make it coherent, which means that they're very energy inefficient. Plasma is the same way, of course.

Yes, but you need many Watts of it, as the mirrors aren't very efficient: 30-40% at most. And you need many of them. And that wafer has to be processed fast enough to satisfy the customers. While things like electron beams do work to etch individual structures, they are much too slow.

 

[Geo]

Errrm. . .net-net, did y'all just say "dual-core is a temporary phenomenon" ??

 

[Frank]

Well, we've got some generations left to go. 90nm is working, albeit problematic, 65nm can probably be done, 45nm will be the end of the way it's worked so far. That, or 30nm, will be a whole new ballgame.

 

[KimB]

Yes, but it's ruled by uncertainty. The effect is already there, but "tunneling" means all of the electrons decide to pass the insulator together. Individual electrons already do so.

Er, that's a statistical impossibility, unless you're talking about a breakdown of the material between conducting lines, and that's a much bigger problem than simply not being able to clock high enough.

Yes. But the only way to solve it would be to use superconductors and almost perfect insulators, until you encounter tunneling.

Superconductors won't help at all. The primary thing that you need to do to prevent tunneling is increase the distance between lanes. Failing that, reducing the voltage can help somewhat. Or you can increase the potential barrier between the conducting lanes, and I'm not sure going with higher-resistance materials is the best way to do that. It may be better to make use of electrical phenomena instead, but I'm not sure off the top of my head how you'd go about doing that.

Yes, but you need many Watts of it, as the mirrors aren't very efficient: 30-40% at most. And you need many of them. And that wafer has to be processed fast enough to satisfy the customers. While things like electron beams do work to etch individual structures, they are much too slow.

Correct, but this is true of plasma-based lasers as well. You just need a high-luminosity source, and all of those that I mentioned can be high luminosity if you throw enough power at them. I would expect, though, that 30%-40% power efficiency to be a fantastically high estimate for any of these techniques of producing a laser (except the semiconductor one....but it's limited by the properties of materials, and so may not go into the UV range...it wasn't that long ago that we discovered how to make blue LED's, let alone UV). I would expect the efficiency of the plasma, electron acceleration, and electron bombardment techniques to give closer to 1%-5% power efficiency, likely on the low end of that.

 

[Frank]

The 30-40% efficiency is for one mirror.

 

[KimB]

The 30-40% efficiency is for one mirror.

With what setup? Because that sounds really doubtful the way I'm thinking about the efficiency.

 

[Frank]

The efficiency of the radiation source is very low, like you said. But to focus that radiation onto a mask, they cannot use lenses. Because that high-ultraviolet radiation will zip right trough those lenses without changing direction. So, they need mirrors. But most mirrors have the same problem. They think they can make mirrors with many layers of a special coating that will reflect up to 40% of the radiation, but probably less. And you need at least 9 mirrors.

 

[KimB]

The efficiency of the radiation source is very low, like you said.

That's what I thought

Because that high-ultraviolet radiation will zip right trough those lenses without changing direction.

Er, actually glass is pretty opaque to UV radiation Whether or not lenses will work is just a question of materials used. Mirrors are clearly another option, and manufacturers must weigh the costs involved. UV radiation behaves no differently in materials that have different indices of refraction than visible light. It's just that the index of refraction is frequency-dependent, as is the opacity (If I remember correctly, copper is transparent to UV, for example).

So, they need mirrors. But most mirrors have the same problem. They think they can make mirrors with many layers of a special coating that will reflect up to 40% of the radiation, but probably less. And you need at least 9 mirrors.

I'm wondering what these mirrors are for, and what this coating is for. The only direct reason I can think that you can't make a mirror with very high reflectance would be because you're attempting to use that mirror to fitler frequencies, which is possible with a very thin coating of transparent material.

Of course, some of the mirrors will be there solely to focus the light from the source into a beam (though some light sources are automatically beam-like from the start, like the accelerated electron source). And I'm not sure why you couldn't get 100% reflectance for a particular frequency from your chosen mirror.

 

[Frank]

Yes, you're right that those extreme UV beams are absorbed by almost any material. My mistake. That's why they cannot use lenses, and even mirrors are a problem, as they have to reflect it, not absorb. Here is an article that explains it.

I read a very nice explanation from ASML (who are developing those steppers), but it was in Dutch (as it's a Dutch company). Anyway, they were pretty sure they could pull it off, but it is a major problem.