Processor Nodes (lastest WAN Show)

[NightEule5]

The latest WAN show was one that I didn't watch live, but in it they do touch on an interesting topic. At around 59 minutes in Linus and Luke start talking about Intel skipping from a 14nm node to 7nm, skipping 10nm. It was an interesting discussion, albeit a short one, but what really stuck out to me was when they said something about the improvement of the technology slowing down and that we'll have change what we're doing in order to improve.

 

The advancement of semiconductors is slowing down. Moore's Law is declining; no longer does it take 1 year to double component density, it's now believed to be around 18 months. I suspect it may reach 2 years, if it hasn't already. Feature size is getting so small that soon quantum mechanics may (and lost likely will) have an effect. Smaller than 7nm would experience quantum tunneling of electrons through the transistor (although some quote it as 5nm). Assuming Intel and/or AMD are able to reach 7nm, since it's generally agreed upon (I think) that 5nm would be a limit, we can't just make stuff smaller for higher performance. We're already seeing this. It's getting harder and harder to decrease the feature size and mass produce them (I believe a 3 atom wide transistor was invented, if I remember correctly, but not mass produced of course).

 

While we will inevitably hit a wall with size, there are some architectural flaws we can fix to progress. The most obvious would be the bloat involved with x86 with all the tacked on instructions. RISC V has a lot of promise in this way (which Linus mentioned). I think that's a big reason why big players in the tech industry are supporting it (like NVIDIA) on top of it being an open platform. I can't say how long it would take, or if it will ever, take over. It would take a looooong time to port something as big as Windows from x86/x64, let alone for developers to port their programs. But what happens when even Reduced Instruction Set technology becomes as refined as we can make it? Sure software also has a lot of bloat that could be dealt with (ahem... looking at you Chrome!), but we can only go so far right?

 

There are proposed solutions like carbon nano-tubes or microfluidic cooling. Quantum computing also accelerates many things and even has functions that classical cannot feasibly do. But the point at which we have to approach the problem differently is rapidly coming. Any thoughts?

 

I find this subject interesting, thanks Linus and Luke for bringing it up. I look forward to discussion if any comes.

[Master Disaster]

There's a few solutions I've heard of.

 

First one is to use a silicon germanium compound which apparently allows down to [3nm?] before suffering from quantum tunnelling.

 

The second is a bit more complicated, it's called EUV or Extreme Ultra Violet. It involves using a different coloured laser during the lithography process which again would get us down below 5nm (I've seen some quotes of 1nm) however there have been some problems with deploying EUV on a large scale. Basically it requires a metric SHIT TONNE of power as the laser is powerful enough to create plasma. By all accounts though EUV will be the preferred method and is being used in some 7nm CPUs right now.

 

https://en.m.wikipedia.org/wiki/Extreme_ultraviolet_lithography

 

Of course the ultimate goal is to move beyond Silicon entirely.

[mockedarche]

Its important to note what counts as a certain size. There was a video from gamers nexus talking about this. They stated how 10nm intel would be a larger more advanced technique compared to 7nm FF.

 found it for you guys. at 7:07 they really explain how you can't compare the two.

[NightEule5]

4 minutes ago, Master Disaster said:

There's a few solutions I've heard of.

 

First one is to use a silicon germanium compound which apparently allows down to [3nm?] before suffering from quantum tunnelling.

 

The second is a bit more complicated, it's called EUV or Extreme Ultra Violet. It involves using a different coloured laser during the lithography process which again would get us down below 5nm (I've seen some quotes of 1nm) however there have been some problems with deploying EUV on a large scale. Basically it requires a metric SHIT TONNE of power as the laser is powerful enough to create plasma. By all accounts though EUV will be the preferred method and is being used in some 7nm CPUs right now.

 

https://en.m.wikipedia.org/wiki/Extreme_ultraviolet_lithography

Alright, but wouldn't that just be delaying the problem? Technologies like this are great but they seem even harder to attain. Another thing I just though about is maybe we could try to focus on improving efficiency yield-wise. It won't make them any faster but it will certainly make them more affordable. Most likely they're doing this already though.

[porina]

Changing the CPU architecture within a similar style that we're used to doesn't help much with the problem, as the problem lies at a lower level in the silicon itself. It is the laws of physics we're running up against, and every time I fought them, I've lost.

[NightEule5]

4 hours ago, Ohsnaps said:

Its important to note what counts as a certain size. There was a video from gamers nexus talking about this. They stated how 10nm intel would be a larger more advanced technique compared to 7nm FF.

found it for you guys. at 7:07 they really explain how you can't compare the two.

I'll give that a watch, looks interesting. I know that size isn't everything, as different processes have better efficiency than others despite being larger.

 

4 hours ago, porina said:

Changing the CPU architecture within a similar style that we're used to doesn't help much with the problem, as the problem lies at a lower level in the silicon itself. It is the laws of physics we're running up against, and every time I fought them, I've lost.

The fact that we're making it so small that Quantum Mechanics actually messes with it is troublesome. Apparently using different materials decreases this further, but only to a point. It amazes me that we've gotten to the point where we're fighting an uphill battle with physics now.

[Master Disaster]

54 minutes ago, NightEule5 said:

Alright, but wouldn't that just be delaying the problem? Technologies like this are great but they seem even harder to attain. Another thing I just though about is maybe we could try to focus on improving efficiency yield-wise. It won't make them any faster but it will certainly make them more affordable. Most likely they're doing this already though.

If the 1nm claims are true then that's kind of the end anyway. I seriously doubt we'll ever reach a picometer process without a major process change, be it a new material or an entirely new way of manufacturing, more than likely both.

 

With that said, I'm fairly sure nanometre was the realm of sci fi only a decade or 2 ago so who knows.

[Worstcaster]

55 minutes ago, NightEule5 said:

The fact that we're making it so small that Quantum Mechanics actually messes with it is troublesome. Apparently using different materials decreases this further, but only to a point. It amazes me that we've gotten to the point where we're fighting an uphill battle with physics now.

This may not be so bad though.  If they can't improve the product itself then improving the process to be more efficient might be the next goal.  If they can produce the same chips cheaper and have the same quality that would be a great thing for the consumer.

[Master Disaster]

1 minute ago, Worstcaster said:

This may not be so bad though.  If they can't improve the product itself then improving the process to be more efficient might be the next goal.  If they can produce the same chips cheaper and have the same quality that would be a great thing for the consumer.

That's not how business works I'm afraid. If you're not pushing for the next big thing then you're not succeeding in the eyes of the shareholders who pay the bills.

 

The only point at which they'll be happy staying put and improving the process is when they psychically have to.

 

As long as there's a chance they can push to the next level they always will because that means more money.

[NightEule5]

Assuming we don't ever find, or it isn't possible to find a material that can have a feature size lower than 3nm or 1nm without Quantum Tunneling that's the end of making the feature size smaller, but not the end of improvement. It might take longer, but there are proposed ways to increase efficiency and performance. Nano-tech looks promising. If we could mass produce chips based on carbon nanotube transistors those would be more efficient I think. Another proposal I've seen is micro-fluid cooling. Using very tiny pumps to push fluid through the circuit or using capillary forces to move it would improve thermal efficiency rather than having a large heatsink and fan sitting on top of it. The tiny streams of liquid flowing inside would be better for transferring heat.

 

Only time will tell when manufacturers start using techniques like this, but I think they've been done in labs.

[Canada EH]

16 hours ago, Ohsnaps said:

7nm

TBH I would trust what suit guy says over long haired guy or earring guy says. But that is just me.

[mockedarche]

@Canada EH Long haired guy is honestly one of the most intelligent content producers for computer enthusiasts. He is talking to suit guy because he wants to understand why intel might be stuck at 10nm. Along with getting a better understanding about the entire process of making silicon chipsets. (note not saying this as you're mean for calling him that just stating that looks definitely aren't a way to judge this person).

[Canada EH]

5 minutes ago, Ohsnaps said:

@Canada EH Long haired guy is honestly one of the most intelligent content producers for computer enthusiasts. He is talking to suit guy because he wants to understand why intel might be stuck at 10nm. Along with getting a better understanding about the entire process of making silicon chipsets. (note not saying this as you're mean for calling him that just stating that looks definitely aren't a way to judge this person).

 

I was joking

I was stating first impressions. I felt though that he is knowledgable and that he was genuinely wanting to understand and learn. Its just a hurdle some people have to get over for the first impression. It is not like I cared much either way to actually sit and watch more of it past 30 seconds so I could come up with a response to this thread. That is why I generalized.

 

I would also have to say that once a person gets over that first hurdle of that first impression, the engagement of conversation of so called "Long Haired Dude" (sorry if you would have posted his name in the quote, I would have used his name) would have to be a bit more, because of that first impression. Just a casual observation.

 

[mockedarche]

Just now, Canada EH said:

 

I was joking

I was stating first impressions. I felt though that he is knowledgable and that he was genuinely wanting to understand and learn. Its just a hurdle some people have to get over for the first impression. It is not like I cared much either way to actually sit and watch more of it past 30 seconds so I could come up with a response to this thread.

 

Thats why I added the tidbit at the end. I highly suggest Gamer's nexus for really technical stuff. Linus tech tips is great for the general public but they don't get really dip into stuff.

[Canada EH]

5 minutes ago, Ohsnaps said:

Gamer's nexus

I do like watching their stuff, as you said they are very knowledgable. However I am not much into hardware anymore as I am good for another 10 years on that front.

 

Another person I love to watch, and if you could help me out in refinding his name, is a guy who really knows his motherboards. I posted something about him 4-8 months ago or a year. He really knows his stuff. I can't remember if he has an accent or not, but he has two names he uses for his channel. I am drawing a blank.

 

The search feature on this forum either sucks, or I just don't know how to use it. Would love to be able to search key words for a user.

 

LOL I found it fast just on YT

Just typed in motherboard VRMs

Actually Hardcore Overclocking

His accent is quite subtle, and odd to me. But thats just me. I always want to figure out where their accent comes from. His is a mix match.

[mockedarche]

@Canada EH I actually have watched him a decent amount. He is actually a bit too enthusiast for me lol. I understand it but theres a deep general knowledge and then theres so much knowledge its profession worthy.

[Canada EH]

1 hour ago, Ohsnaps said:

@Canada EH I actually have watched him a decent amount. He is actually a bit too enthusiast for me lol. I understand it but theres a deep general knowledge and then theres so much knowledge its profession worthy.

That is exactly what you want though when searching for what to buy as a consumer. Sure your eyes can glaze over when he goes on about chip part #'s and such but the information is priceless. You wont get that kind of info from Linus! TBH I only liked very few video's from Linus, the most memorable ones were where he would take apart a product, but can only do so many of those.

 

I haven't looked too much into the whole 7nm vs 10nm, but that Nexus suit guy is what I will hang my hat on. Its just a matter of implentation not so much the size, like all males right