What comes after node shrink?

[Arthur Hanson]

With node shrink coming to an end in the not to distant future, what technologies does the SemiWiki community see coming to replace it or that semi/nanotech will go in entirely new directions and if so what technologies will be coming on? Does anyone see a company coming out of left field with revolutionary technologies or applications?

 

[alphaguy]

As far as I can remember, node shrink has been "coming to an end in the not too distant future" at least since the 1 micron node. Somewhere around 30 years ago.

I'm sure it has slowed since then, and will end some day. But at this point I'm not betting on whether it will happen in my lifetime.

 

[Portland]

Tmsc is working on 2nm. 3in is projected to be 2022.

 

[jmlobert]

As far as I can remember, node shrink has been "coming to an end in the not too distant future" at least since the 1 micron node. Somewhere around 30 years ago.

In contrast to 30 years ago, however, we *are* approaching physical limits. Once you get to the size of a single molecule, there is no more shrinking. The SiO2 molecule, for example, has a van der Waals diameter of 0.3 nm. I very much doubt that we will get beyond 1 nm feature size. The line edge roughness would be tremendous. I am not convinced that we will get to 2 nm or that 3 nm will be very succesfull, we shall see.

 

[count]

Well 3nm no longer means 3nm feature size as the node naming conventions don't have anything to do with reality anymore. So I'm sure TSMC will be successful on it's "3nm" process. That said, each node shrink is exponentially more expensive than the last, and it seems that we will likely reach an economic limit before we reach a physical limit. GloFlo threw in the towel at 7nm leaving only 3 companies left pursuing leading edge processes. I'd be willing to bet there will be another drop out at 3nm (Either Intel or Samsung) and things will stall shortly after. I could see Intel going fabless and outsourcing it's volumes to either TSMC or Samsung.

 

[Portland]

I think Intel is a broken company. It's not all node size architecture is important as well, you look at things like ufs storage and they seem lost. Intel was the king but now I'm wondering about the quality of Intel's products.

I can't see the campus is Arizona or oregon going fabless and being successful. 1. They can't find people willing to work there. 2. People in those countries have a reputation for being dipshits. Relationships was one of their biggest faults.

 

[Arthur Hanson]

Anyone have any thoughts on layering or some other technology replacing node shrink, possibly a whole new architecture? Any thoughts on how DNA or materials with optical characteristics may bring change?

 

[count]

I think Intel is a broken company. It's not all node size architecture is important as well, you look at things like ufs storage and they seem lost. Intel was the king but now I'm wondering about the quality of Intel's products.

I can't see the campus is Arizona or oregon going fabless and being successful. 1. They can't find people willing to work there. 2. People in those countries have a reputation for being dipshits. Relationships was one of their biggest faults.

They will have to figure it out if they want to stay in business

Anyone have any thoughts on layering or some other technology replacing node shrink, possibly a whole new architecture? Any thoughts on how DNA or materials with optical characteristics may bring change?

I'm going to use an analogy of automobiles, where between the late 1800s up to around the 1970s, power and fuel efficiency both rose at a steady rate, and most of that had to to with improvements in engine design. But after that, for a long time, automobiles continued to get better but it's not because they became more powerful. They continued to get more fuel efficient, safer, and more driver comfort features were added. No-one would dispute that today's cars aren't better than cars were in the 1970s, but they aren't faster.

I think something similar will happen in computing, where CPU performance will top out in a few years, but computers will still get incrementally better due to other improvements in architecture and maybe smaller improvements in process, and I think that's probably going to how it'll be for a little while after we reach the economic/physical limits of node shrinks. I think there will ultimately be some other new technology that drives human innovation forward in the future, although it may not be computing.

 

[Arthur Hanson]

They will have to figure it out if they want to stay in business



I'm going to use an analogy of automobiles, where between the late 1800s up to around the 1970s, power and fuel efficiency both rose at a steady rate, and most of that had to to with improvements in engine design. But after that, for a long time, automobiles continued to get better but it's not because they became more powerful. They continued to get more fuel efficient, safer, and more driver comfort features were added. No-one would dispute that today's cars aren't better than cars were in the 1970s, but they aren't faster.

I think something similar will happen in computing, where CPU performance will top out in a few years, but computers will still get incrementally better due to other improvements in architecture and maybe smaller improvements in process, and I think that's probably going to how it'll be for a little while after we reach the economic/physical limits of node shrinks. I think there will ultimately be some other new technology that drives human innovation forward in the future, although it may not be computing.

They will have to figure it out if they want to stay in business



I'm going to use an analogy of automobiles, where between the late 1800s up to around the 1970s, power and fuel efficiency both rose at a steady rate, and most of that had to to with improvements in engine design. But after that, for a long time, automobiles continued to get better but it's not because they became more powerful. They continued to get more fuel efficient, safer, and more driver comfort features were added. No-one would dispute that today's cars aren't better than cars were in the 1970s, but they aren't faster.

I think something similar will happen in computing, where CPU performance will top out in a few years, but computers will still get incrementally better due to other improvements in architecture and maybe smaller improvements in process, and I think that's probably going to how it'll be for a little while after we reach the economic/physical limits of node shrinks. I think there will ultimately be some other new technology that drives human innovation forward in the future, although it may not be computing.

They will have to figure it out if they want to stay in business



I'm going to use an analogy of automobiles, where between the late 1800s up to around the 1970s, power and fuel efficiency both rose at a steady rate, and most of that had to to with improvements in engine design. But after that, for a long time, automobiles continued to get better but it's not because they became more powerful. They continued to get more fuel efficient, safer, and more driver comfort features were added. No-one would dispute that today's cars aren't better than cars were in the 1970s, but they aren't faster.

I think something similar will happen in computing, where CPU performance will top out in a few years, but computers will still get incrementally better due to other improvements in architecture and maybe smaller improvements in process, and I think that's probably going to how it'll be for a little while after we reach the economic/physical limits of node shrinks. I think there will ultimately be some other new technology that drives human innovation forward in the future, although it may not be computing.

 

[Arthur Hanson]

Do you see advanced packaging playing a more than minor role giving us increased performance out of current technologies at low cost since we would just be extending the technologies we already have? In short, just packing more power in a smaller space and using energy management processes to keep the power problems in check?

 

[count]

Do you see advanced packaging playing a more than minor role giving us increased performance out of current technologies at low cost since we would just be extending the technologies we already have? In short, just packing more power in a smaller space and using energy management processes to keep the power problems in check?

Yeah, I definitely see packaging as an avenue where there is still room for improvement, especially when you have the industries moving in the direction of chiplets. Silicon photonics are also in interesting area given this trend, where you need to have these chiplets that are physically further away from each other but still need to communicate with each other at high speed. You can start to imagine the need for some kind of advanced packaging with an integrated silicon photonic communication bus enabling a group of chiplets to work together.

 

[aayres]

After node shrink there is a high possibility of 3D monolithic going mainstream. I am skeptical about exotic materials HVM. I also think there is a great chance of seeing a hybrid cmos + quantum. Quantum will probably always require cryogenics, so I see those chips running in huge data centers. When your device execute a high load task, it will assign it to the data center (something like google stadia).

 

[Portland]

Intel seems more interested in automotive than .... or really focused on automotive. They want to buy a transportation app for a $1 billion.

 

[SunnyLining]

Anyone have any thoughts on layering or some other technology replacing node shrink, possibly a whole new architecture? Any thoughts on how DNA or materials with optical characteristics may bring change?

I think DNA technologies or some kind of molecular machine approach will get scaling going again in approximately 10 years. Feel free to round that up to 100 if you like.

As for microfabricated optics, certainly, it could do a lot to improve networking speeds, but it won't help with increasing density. We are currently using mlrrors that barely reflect the "extreme ultraviolet" light (a.k.a. X-rays) to make the 7nm, and soon 3nm, transistors. That's a strong hint that optics won't work at those sizes! Also, if you want photonics to take over from electronics in digital logic, it's worth wondering how easy it will be to get an optical switching device to work as fast as those 3nm transistors TSMC is working on, especially if you want this switcher to be made of silicon. My guess is that this will be a real challenge.

I don't have any more optimism for things like multi-layer. For that to keep Moore's Law on track, we'd need to be able to double the stack layering every 18 months. And at the same time, we'd have to constantly reduce power and increase speed with the same size transistors, so I think we're going to need to invest in the search for magical fairies or rainbow pooping unicorns or something to provide the magic needed to defy physics to make that happen.

Our industry is entering the dreaded "mature technology" era. Excitement in the immediate years ahead will be driven by things that once required DoD cost plus budgets falling down to the consumer level, which will include the requisite cutsie-putsie name changes. It's already happening: "chiplet" sounds a lot cuter than "multi-chip module". Changing shape and color will drive sales more and more. Antiquing couples 20 to 50 years from now will be buying computers because they just don't make them that way anymore... meaning the shape and color!