Why don't they just Make C/GPUs out of silver?

Why don't they just Make C/GPUs out of silver?

We're never going to get over this performance plateau without it

Other urls found in this thread:

en.wikipedia.org/wiki/List_of_thermal_conductivities
twitter.com/NSFWRedditVideo

Because silver isn't a semi-conductor

Silver will probably have more issues right now than whatever they use now. Silver is incredibly easy to ionize and polarize, and as far as I know, the 'plateau' is due to quantumn mechanics interfering with our classical knowledge of electrodynamics because of the tiny scales of the components we're making and using. Silver, then, would probably have even more interference, and likely work worse, even though it's a better conductor, which is the only thing that I can think of that would make you think that it would be better for processing units.

This is one of the reasons why silver atoms were used in the Stern-Gerlach experiment, which if you don't know, is quite intereseting and well worth a look-up if you have the time.

Wow, I came here to shitpost about silver being expensive as fuck, but I actually learned something today.

Native and isotope forms of copper, aluminum, silver, gold, tungsten, and a dozen other materials are all used in semiconductors to carry and regulate electrical signals in the back end of a chip leading up to the source and drain wells.
There is no plateau like you're trying to imply. Semiconductors are not made out of any single material like you're trying to imply.

If you're meaning that silicon as a substrate material is inherently limiting clockspeeds of switching devices, you're still pretty far off the market since silver is a conductor, not a semiconductor. Its not something you'd replace silicon with. There is nothing about silicon itself as a substrate material thats keeping clock speeds of high performance chips around 4-5ghz, this is entirely transistor topology.

>quantum mechanics
Stop regurgitating things you don't understand. Electrons leak through a channel due to insufficient electrical resistance to keep them in place. This is an incredibly simple principle, one that has been well understood for the better half of a century.
It is the short channel effect. Eventually the channel becomes short enough that no planar or even FinFET gate can effectively control the channel region.
If you have to dress something up with vague popsci buzzwords then you don't understand it.

You didn't really learn anything.

I love you Albedo!

Albedo is worst girl and a waste of screentime.

I love you too, ma pĂȘche!

>Electrons leak through a channel due to insufficient electrical resistance to keep them in place.
Not him but I thought this was due to electron tunnelling, is that not the case?

Electron tunneling is the term you'll see used at a populist level, but is totally inaccurate given its definition. Electrons have no problem passing through most materials. They'll readily leave the channel of a transistor of any size, it is a normal percentage of leakage current. It only becomes a significant issue as short channel effect is more severe in smaller devices.

Imagine trying to stop a marble from rolling down a slope via friction. Smaller transistors represented by steeper slopes. Eventually you get to a point where your slope is entirely vertical, and friction thereby has little to no effect in practice.

In sufficiently small devices you could have an effective gate surface area many times greater than the channel length, but it still can't adequately control those electrons and keep them in place. Things like SiGe channels, other means of isolation, different transistor structures all address the isssue.

Everything is well understood within the realm of classical physics. Electrical resistance isn't some quasi scientific meme property that the reddit graphene community thinks it is.

That makes a lot of sense. Thanks for the detailed explanations.

There's same reason why we're not make them out of diamond.
1) It's expensive.
2) There're much better materials for it (such graphene etc.)
en.wikipedia.org/wiki/List_of_thermal_conductivities

Aren't switching losses where all the heat is rather than resistive losses?

The only thing I don't quite understand is why I can buy GaN mosfets already but not processors.

Rude.

Heat is the waste product of a transistor switching, yes. Though a transistor switches by way of exerting a field of electrical resistance. Hence the name field effect transistor.

Gallium nitride is a viable III-V material for semiconductors, but the process of fabbing logic using it as a base is radically different. New nodes coming online cost several billion dollars now. No one is going up up and change everything unless there is an immediate dire need.

>Gallium nitride is a viable III-V material for semiconductors, but the process of fabbing logic using it as a base is radically different. New nodes coming online cost several billion dollars now. No one is going up up and change everything unless there is an immediate dire need.

How much of an impact would GaN make on logic chips assuming you had mature fabs for it? The GaN fets have a huge reduction in switching losses it's about 10x less to turn them on and they switch faster.

Would you get large chips with lower TDP or the same size with a big jump in clock speed?

>How much of an impact would GaN make on logic chips assuming you had mature fabs for it?

It would come down to what exactly was implemented using it. If you bake it down to the simple metric of power per transistor at a given switching frequency then you could make the generalized claim of X being 10 times more efficient than Y, but its never that straight forward. I'm impossible to say what a hypothetical process would yield. Even foundry PR departments have the tact to base their claims on validated test chips, even if they are small simplistic circuits, instead of going by theoretical parametrics.

Kill yourself, shit stain.

Impolite.

Gas yourself, shit for brains.

Not nice. Gimme more of your (You)s.

...

How do you make a programming language out of silver ?

If you made transistors small enough that'd be a problem. But not yet.

Yes.

The answer is because the old men who knew how to create new processors died a few years ago. Everyone in CPU manufacturing business just swear they know how to write in assembly while sweating (pro tip: no one does)

Which old men?

Is it even possible for it to compete with silicon off the bat? Wouldn't they have to progressively reduce the nanometers like they have with silicon?

Electronic old men

>Wouldn't they have to progressively reduce the nanometers like they have with silicon?
Not necessarily.
A lot of what has happened over the last few decades in photolithography has been applying and refining theory. It was very much learn as we go, come up with new theories, test them, and hope to somehow profit.
In this areas scaling and all the ways we've facilitated it have been doubly served the purpose of improving vital performance and cost metrics while also ironing out the methodology which would serve as the building blocks for the next upcoming node.

When we switch to a new substrate material we won't be starting from zero. Its inevitable that new tooling will be created, but most existing tooling will be adapted, and design methodologies employed today will likewise be adapted. Research done is sort of an investment. It can always end up paying dividends.
IE we could have some self assembling crystal matrices form 1mm nanowires on molybdenum disulfide or another exotic fancy lab material.

I don't know shit about how the fuck processing units are made, all I was stating that the only explanation of a 'plateau' in processing speeds that I've heard is because of unexpected behavior at the molecular level. This is the only guess that I could make given OPs vagueness. I'm definitely no expert on any circuits or electronics in general, but I have a good background in quantumn mechanics and I din't say anything inherently wrong.

From what I remember, semiconductors have special properties, and like you said, are made of specific layers of various materials with specific individual properties. I do not believe that silver is a widely used component in semiconductors as it is too easily ionizable.

I also wasn't talking about gates at all, and of course electrons will leak through a gate without having to rely on quantumn mechanics to explain that. If you try to claim that two channels that are near each other don't interfere with each other with EMR or quantumn interferance, then you are a fool. As chips become smaller, it would make sense that there are less and less space between 'forbidden' regions.