Moore's Law Alive and Well, We Will Exhaust the Periodic Table: Intel's Pat Gelsinger

[lexluthermiester]

Or just quite visibly desperate.

It's not that. Intel has a habit of pulling a rabbit out of it's hat(Pentium, Pentium3, Core2, etc..) and I sense an impending rabbit pull..

 

[Valantar]

It's not that. Intel has a habit of pulling a rabbit out of it's hat(Pentium, Pentium3, Core2, etc..) and I sense an impending rabbit pull..

But then you're talking architecture, not process. Process node development is running into a wall, and while that wall can be moved further away, it can't be removed. That's just physics. And Gelsinger's comments here were specifically about process developments. There is no rabbit to be found, no magic fix for the fact that shrinking nodes further will just become increasingly difficult as time goes by. This is where the desperation lies: this industry is built over the past decades on a rate of growth that is fundamentally unsustainable on the level of laws of nature - and the industry, blinded by said growth, is running at full speed into that wall. It's going to get ugly.

 

[AusWolf]

But then you're talking architecture, not process. Process node development is running into a wall, and while that wall can be moved further away, it can't be removed. That's just physics. And Gelsinger's comments here were specifically about process developments. There is no rabbit to be found, no magic fix for the fact that shrinking nodes further will just become increasingly difficult as time goes by. This is where the desperation lies: this industry is built over the past decades on a rate of growth that is fundamentally unsustainable on the level of laws of nature - and the industry, blinded by said growth, is running at full speed into that wall. It's going to get ugly.

There is an ultimate wall at the end of any kind of growth: the Earth. It's not just Intel, but modern-day capitalism in general that's at fault here. We can't build an economy on constant growth on a planet that's still the same size as it was millions of years ago. Trying to maintain growth at all costs is just a pipe dream that humanity will eventually have to wake up from and learn to be happy with what we have.

 

[Valantar]

There is an ultimate wall at the end of any kind of growth: the Earth. It's not just Intel, but modern-day capitalism in general that's at fault here. We can't build an economy on constant growth on a planet that's still the same size as it was millions of years ago. Trying to maintain growth at all costs is just a pipe dream that humanity will eventually have to wake up from and learn to be happy with what we have.

Yep. And that's one of the major core reasons why we're seeing the societal developments we're seeing these days: the insane sprint towards peak everything that has been going on for the past century or so is starting to, well, reach peak everything. But capitalism does this funny thing where it convinces people to sprint towards the peak, but also convinces them that there is no peak at the same time. Which, in case it wasn't rather obvious, is quite problematic. Technological development has had growth as its only target for so long that people act as if it will go on forever, yet every single development shows that it's just getting exponentially harder to do anything new at all. And, crucially, what we already have is already in many ways more than good enough, as you say.

And, of course, the rest of the world is adjusting! The performance demands of new games are dropping massively compared to a decade ago. We are seeing an increasing amount of tech aimed at making the most out of what we have, like FRS, DLSS and the like. We need to start adjusting to a future where technologies aren't going to get drastically more advanced - or, frankly, we should have done so a decade ago. 'Cause that's where we're headed. And that's fine. We just also need an economic system focused around maintaining stability and sustaining both human life and the environment at the same time, rather than seeing stability as a lack of growth and therefore inherently harmful.

 

[lexluthermiester]

Process node development is running into a wall, and while that wall can be moved further away, it can't be removed. That's just physics.

But you're ignoring chemistry, something that was mentioned elsewhere.

 

[Aquinus]

But you're ignoring chemistry, something that was mentioned elsewhere.

I think that @Valantar's point is that even if we use different materials, we're hitting physical limitations due to things like the size of atoms. We can only make things so small regardless of the materials that are used. There are very real quantum mechanical limits and it's not like we can change the size of atoms or subatomic particles.

 

[Valantar]

But you're ignoring chemistry, something that was mentioned elsewhere.

Chemistry can somewhat change the characteristics of a substance, but it will never overcome the fundamental issues of bringing silicon lithography into the sub-nm range. No amount of chemical treatments will get you there. Again: physics. Chemistry is also subject to the laws of physics, after all.

 

[AusWolf]

Yep. And that's one of the major core reasons why we're seeing the societal developments we're seeing these days: the insane sprint towards peak everything that has been going on for the past century or so is starting to, well, reach peak everything. But capitalism does this funny thing where it convinces people to sprint towards the peak, but also convinces them that there is no peak at the same time. Which, in case it wasn't rather obvious, is quite problematic. Technological development has had growth as its only target for so long that people act as if it will go on forever, yet every single development shows that it's just getting exponentially harder to do anything new at all. And, crucially, what we already have is already in many ways more than good enough, as you say.

The faster we run towards the peak, the harder we fall from the cliff, I'm afraid.

And, of course, the rest of the world is adjusting! The performance demands of new games are dropping massively compared to a decade ago. We are seeing an increasing amount of tech aimed at making the most out of what we have, like FRS, DLSS and the like. We need to start adjusting to a future where technologies aren't going to get drastically more advanced - or, frankly, we should have done so a decade ago. 'Cause that's where we're headed. And that's fine. We just also need an economic system focused around maintaining stability and sustaining both human life and the environment at the same time, rather than seeing stability as a lack of growth and therefore inherently harmful.

I think that's why things like 4K and super high FPS gaming were invented. Back in the days, 30 FPS was good enough. Now we need 300? Why? Only because game technologies aren't evolving as fast as PC hardware is. We can see it with CPUs. Any CPU paired with a mid-range graphics card can game. There's no such thing as a CPU not good enough for games anymore. Some people criticize Zen 4 for not bringing a big enough improvement in games over Zen 3. How could it when every modern game is GPU limited at basically every setting? This is why companies are desperately trying to convince us that we need that new graphics card and we need that super high refresh rate 4K monitor even though we actually don't. I'm fine with my curved 1080p monitor and 6500 XT. I'm only thinking about upgrading because I'm curious. But I'm sure that if I end up upgrading this year, or next year, I'll be happy regardless of my choice.

 

[Valantar]

I think that @Valantar's point is that even if we use different materials, we're hitting physical limitations due to things like the size of atoms. We can only make things so small regardless of the materials that are used. There are very real quantum mechanical limits and it's not like we can change the size of atoms or subatomic particles.

Exactly. We can push the wall ever so slightly further away with all kinds of innovations, but we can't make it not be a wall, so if we keep running we will run into it. The solution? Stop running, slow down, do something different instead. We do not actually need this growth.

Edit: autocorrect is fun

 

[lexluthermiester]

I think that @Valantar's point is that even if we use different materials, we're hitting physical limitations due to things like the size of atoms. We can only make things so small regardless of the materials that are used. There are very real quantum mechanical limits and it's not like we can change the size of atoms or subatomic particles.

Chemistry can somewhat change the characteristics of a substance, but it will never overcome the fundamental issues of bringing silicon lithography into the sub-nm range. No amount of chemical treatments will get you there. Again: physics. Chemistry is also subject to the laws of physics, after all.

You're both missing some context and I'm not going into that level of detail here.

 

[Valantar]

The faster we run towards the peak, the harder we fall from the cliff, I'm afraid.

Yep.

I think that's why things like 4K and super high FPS gaming were invented. Back in the days, 30 FPS was good enough. Now we need 300? Why? Only because game technologies aren't evolving as fast as PC hardware is. We can see it with CPUs. Any CPU paired with a mid-range graphics card can game. There's no such thing as a CPU not good enough for games anymore. Some people criticize Zen 4 for not bringing a big enough improvement in games over Zen 3. How could it when every modern game is GPU limited at basically every setting? This is why companies are desperately trying to convince us that we need that new graphics card and we need that super high refresh rate 4K monitor even though we actually don't. I'm fine with my curved 1080p monitor and 6500 XT. I'm only thinking about upgrading because I'm curious. But I'm sure that if I end up upgrading this year, or next year, I'll be happy regardless of my choice.

Completely agree. I can't see myself upgrading from my 5800X+6900XT in the next half decade, unless something breaks. I'll likely be moving to a higher resolution monitor from my current 1440p60 one, but that's more for work than gaming - FSR, or just running at non-native resolution in games (1080p with integer scaling should look good on a 2160p panel) will tide me over there. Considering that my Fury X worked fine for six years, it'd be a damn shame if this setup doesn't beat that.

You're both missing some context and I'm not going into that level of detail here.

No, we aren't. We're just saying that this context is fundamentally insufficient compared to the problem it is purported to solve. Silicon dopants and other treatments can't change the facts that there are physical limits to how small you can make transistors in silicon, nor that the light sources and technologies needed to etch something that small are so advanced that even the companies making the scanning machines are saying "this won't be worth the cost in a decade". Yes, they too hand-wave at "future solutions" making it cheaper, but ... those solutions are only getting harder to come by. None of this will get any easier. Ever.

 

[lexluthermiester]

Silicon dopants and other treatments can't change the facts that there are physical limits to how small you can make transistors in silicon

True. But when you change the chemistry, you change and improve the semiconductor properties, meaning the current scales no longer apply. Gallium-nitride is a perfect example of that, with it being able to reach high speeds at much lower voltage and very low heat generation as GaN has excellent semiconductor properties. When it conducts, it does so with very low resistance, and thus little waste heat. When it insulates, it does so with nearly perfection. And it's doing that at around 40nm-ish scales. Yet GaN is not sustainable on a mass market level because there isn't enough Gallium to go around in the world. It would otherwise be an excellent replacement for Silicon. However, chemistries based on Arsenic are sustainable. And when Arsenic is blended with small amounts of Tellurium, you get at potential of properties on larger scales(90nm-ish) that exceed what Silicon chemistries are capable of at current scales. The old phrase rings true, to take steps forward, we must take some steps backward. This is what I think Intel and many others are working on, and they're likely very close to commercial deployment. 2 or 3 years. Maybe 4. At that point, Moore's law will be extended greatly, likely for another 25 or 30 years.

None of this will get any easier. Ever.

That is a very defeatist attitude. It's not the kind of thinking that solves problems.

 

[Count von Schwalbe]

One thing that new methods can't do - make transistors smaller than the physical limitations of the silicon atom.
One thing that new methods can do - make transistors more efficient and clock higher at the same size.

Both Lex and Valantar are correct. Now we can just wait and see if the new methods can compete with the old on price.

 

[lexluthermiester]

One thing that new methods can do - make transistors more efficient and clock higher at the same size.

Correct. Gallium-nitride based chemistries are evidence of this.

 

[Valantar]

True. But when you change the chemistry, you change and improve the semi-conductor properties, meaning the current scales no longer apply. Gallium-nitride is a perfect example of that, with it being able to reach high speeds at much lower voltage and very low heat generated as GaN has excellent semiconductor properties. When it conducts, it does so with very low resistance, and thus little waste heat. When it insulates, it does so with nearly perfectly. And it's doing that at scales around 40nm-ish scales. Yet GaN is not sustainable on a mass market level because there isn't enough Gallium to go around in the world. It would otherwise be an excellent replacement for Silicon. However, chemistries based on Arsenic are sustainable. And when Arsenic is blended with small amounts of Tellurium, you get at potential of properties on larger scales(90nm-ish) that exceed what Silicon chemistries are capable of at current scales. The old phrase rings true, to take steps forward, we must take some steps backward. This is what I think Intel and many others are working on, and they're likely very close to commercial deployment. 2 or 3 years. Maybe 4. At that point, Moore's law will be extended greatly, likely for another 25 or 30 years.

I'll believe that when I see it - and I'll consider believing it when anyone of note makes a more concrete statement about this coming than "we're working on it, it is coming in the future". These things have been worked on for years if not decades already, and they're still nowhere near to being reality. If a technology keeps "showing promise" for decades, then it's not actually coming, and people are being strung along to fund expensive research for niche uses. Has Intel, or anyone else, actually published anything substantial relation to this? Not as far as I know, at least - but feel free to provide sources. But crucially, we can't trust that "future innovations" will fix our problems - it's too late for that. Problems are here now, today. A pie-in-the-sky solution two decades from now - that likely won't pan out as projected anyhow - is fundamentally insufficient.

There is, of course, also the "small" issue of arsenic being incredibly toxic. Moving the semiconductor industry to it as a base material is ... yeah, not unproblematic, to say the least.

Correct. Gallium-nitride based chemistries are evidence of this.

GaN is great - but it doesn't perform all that well for logic, and is better suited for other applications (like power regulation). Hence the applications it's seeing in the real world. Finding a universal or near-universal large-scale replacement for silicon that works, is inexpensive, and doesn't bring with it a host of new problems? That's rather utopian.

 

[Count von Schwalbe]

there isn't enough Gallium to go around in the world

Blame Der8aur for that

At that point, Moore's law will be extended greatly, likely for another 25 or 30 years.

The trouble is, Moore's law states that the number of transistors will double every two years. Yes, this will become practical with the new technologies, but will it become affordable? Currently, node shrinks are holding about even for the cost per transistor, not the cost per mm2 of wafer. Thus the current rising prices. If new technologies make it possible to build better semiconductors, good! However, if it is more expensive than going bigger on an existing node, than I can tell you which one customers will buy.

 

[TheoneandonlyMrK]

Yes & Moore's "law" has been dead for a while now. I think it broke sometime during middle of last decade.

The official version yes but INovation kicked in, well at AMD then Intel but still.

When they get to 1 or 2nm, what happens in the next generation?

What's in a name, f all truth in reality, it's going to Angstroms next week via Intel if you believe they're hype.

I think we're 10 years off actually atom scale transistors personally.

 

[Count von Schwalbe]

I think we're 10 years off actually atom scale transistors personally.

I doubt we will ever get there. The limit of a silicon cubic is around 1/2nm (5 Angstroms) - I doubt yields on a process of this scale would be remotely practical. For scale, TSMC N5 transistors are ~51nm wide (yes, really) which translates to around 100 cubics. I read somewhere that there are two atoms per cubic, so 200 atoms. Intel 18A should run around 360 atoms by that metric. The metal (interconnect) separation is somewhat lower, around half of that.

This is my potentially inaccurate summary of a bunch of sources.

 

[Readlight]

100 billion

 

[ARF]

1/2nm (5 Angstroms)

5 Ä is equal to 0.5 nm..

 

[Valantar]

5 Ä is equal to 0.5 nm..

View attachment 263554

... yes, that is literally what the post you quoted said? 1/2 = 0.5.

 

[lexluthermiester]

GaN is great - but it doesn't perform all that well for logic, and is better suited for other applications (like power regulation). Hence the applications it's seeing in the real world.

Not quite. It's limited to those types of deployment because it's exceptional for power channeling & regulation. Again the resource limitations.

 

[Valantar]

Not quite. It's limited to those types of deployment because it's exceptional for power channeling & regulation. Again the resource limitations.

True - but if there were applications where it would notably outperform silicon at an even remotely competitive cost, it would still most likely see adoption into those niches IMO. Gallium is rare, but still available, after all. Of course, cost is also linked to material availability - but on the other hand, sufficient performance would make that cost worth it. The trick would be finding something that improves on silicon in a similarly broad range of use cases while ideally being cheaper to produce than current cutting-edge nodes, which are already becoming too expensive. And that seems like a pie-in-the-sky proposition.

 

[lexluthermiester]

Gallium is rare, but still available, after all. Of course, cost is also linked to material availability - but on the other hand, sufficient performance would make that cost worth it.

The problem is that the moment wide-scale demand is reached, costs go through the roof. Even GaAs chemistries, which use less Gallium, would be unsustainable, which is why they were never adopted.

The trick would be finding something that improves on silicon in a similarly broad range of use cases while ideally being cheaper to produce than current cutting-edge nodes, which are already becoming too expensive. And that seems like a pie-in-the-sky proposition.

Actually those types of chemistries exist and are being tested. They've been around for decades. The problem is mass production and stability as a substrate for integrated circuits. Even Tin and Antimony have been explored but no one pushed forward because Silicon was so workable. Necessity is pushing the drive for alternatives at this point..

 

[Valantar]

The problem is that the moment wide-scale demand is reached, costs go through the roof. Even GaAs chemistries, which use less Gallium, would be unsustainable, which is why they were never adopted.

I can only assume that, this being related to the computer industry, GaAs stands for "Gallium as a service".


Edit: crap, I shouldn't be giving new ideas to these people. Please ignore.