How will CPU processing power be increased when transistors can't be shrunk farther?

[lopri]

You mean coding?

Patents, more likely. Though I suspect the patent bubbles would bust at some point.

 

[Ajay]

The potential of graphene is pretty astonishing because you can stand the graphene on end, make it vertical, and then pack them laterally like a stack of pancakes laid on its side.

The xtor density can get amazingly high because the drain (or source) can be buried under the graphene, with the other residing over top.

Think finfet with the fin width shrunk to ~0.15nm. Crazy thin.

Interesting, Kuhn's recent presentation @ IDF show some experimental data using horizontal layouts w/graphene (pp 41 ff). This makes sense since, as I understand it, sheets of graphene are easier to manufacture.

A electron diffusion gap of 0.15 nm sounds amazing, but getting a respectable Ion current out of that would be hard (wouldn't VG need to be fairly high) - and quantum confinement could be a problem. Still, the scientists at Intel and elsewhere have become increasingly more clever about solving these types of problem as the state of the art in semiconductor technology has started putting more demand on these types of solutions.

PS, thanks for the post you made on Bravais super lattices, I was stuck in the ER for 6+ hours with only my iPad. It made for interesting reading and brought me right back to my undergrad QM studies. Suddenly, I have an urge to get my go for an M.Sc. in Physics.

 

[Idontcare]

Interesting, Kuhn's recent presentation @ IDF show some experimental data using horizontal layouts w/graphene (pp 41 ff). This makes sense since, as I understand it, sheets of graphene are easier to manufacture.

A electron diffusion gap of 0.15 nm sounds amazing, but getting a respectable Ion current out of that would be hard (wouldn't VG need to be fairly high) - and quantum confinement could be a problem. Still, the scientists at Intel and elsewhere have become increasingly more clever about solving these types of problem as the state of the art in semiconductor technology has started putting more demand on these types of solutions.

PS, thanks for the post you made on Bravais super lattices, I was stuck in the ER for 6+ hours with only my iPad. It made for interesting reading and brought me right back to my undergrad QM studies. Suddenly, I have an urge to gat my go for an M.Sc. in Physics.

It will take a while before they commercialize the perpendicular orientation, it took the hard-drive industry a while to perfect it too. (not to mention it took them a while before they publicized the fact they were working on it in the R&D labs

But they'll get there in time, it is one of those unavoidable consequences of working with something thin and flat while trying to make it become more dense "within the plane" since you are paying for the "plane" (be it a glass substrate for HDDs or a silicon wafer for CMOS CPUs).

 

[Ajay]

It will take a while before they commercialize the perpendicular orientation, it took the hard-drive industry a while to perfect it too. (not to mention it took them a while before they publicized the fact they were working on it in the R&D labs

I assume you are talking about GMR - if so, good point. And I take your second point to mean, there is more work going on in using perpendicular graphene than Intel is admitting :sneaky:

 

[cytg111]

I don't mind you asking, and I'd be happy to answer to the best of my abilities.

FWIW, I post on a forex forum (forum.mql4.com to be specific) under the username 1005phillip. As you can imagine from my CPU-related threads here, my posts over there are "enthusiast" rated (albeit towards forex rather than CPUs )

...

It is really fun though. I thought process node development was my dream job, but this is definitely more fun. A different brain teaser every day.

Cool .. I just got 'trading' and 'algos' and thought you might be heavy into some genetic algorithms of sorts .. as I can google-about some ppl actually do AI on the scripting language of MT4

 

[Idontcare]

Cool .. I just got 'trading' and 'algos' and thought you might be heavy into some genetic algorithms of sorts .. as I can google-about some ppl actually do AI on the scripting language of MT4

Yeah a lot of that stuff has been done and is already in the public domain. Which is great for people like me who best learn by reverse-engineering existing examples when we try to figure stuff out.

My approach to forex trading though is decidedly non-traditional as my background makes me have a rather unusual perspective on market models and so on. Not that this makes me special or better, just different.

MT4 and MQL4 are really easy programming languages to pick up. I taught myself and I think probably just about anyone can teach themselves how to program in MQL4. It is a lot like basic, of what I remember of programming in basic back in the 80's.

 

[cytg111]

Cool, sounds like you are having fun .. and really thats about as good as it gets for a day job . I was into a bit of trading my self some 10 years back, got these huge data sets on trading history and started modelling some neural nets around it and while doing that i came in contact with these trading forums, forgot the names, but they were huge, many people, and alot of history about people discussing stocks 'take a look at this, may be huge' etc. Me, not having a financial background, I was drowning here.. So I wrote some code to traverse the forums, who said what when, and with a few generalizations I could automate who had recommended what and put it against the historical data i had... So now i knew who got it right most of the time, with performance charts to prove it.. So i picked my horses that way, and made a buck .. Never further pursued the neuralnet angle on stocks though (other angles yes), but I know it's been done. Just another story on how things turned out-> not the way you expected it .

 

[Ajay]

The potential of graphene is pretty astonishing because you can stand the graphene on end, make it vertical, and then pack them laterally like a stack of pancakes laid on its side.

The xtor density can get amazingly high because the drain (or source) can be buried under the graphene, with the other residing over top.

Think finfet with the fin width shrunk to ~0.15nm. Crazy thin.

I'm still stuck on this because a gate width of 150 pm just blows my mind. There is a ton of research going on in Graphene. It looks like any vertical structure made up of layers of Graphene will need to be a dual layer 'composite' since Graphene has no intrinsic band gap and because there may be a need to recruit Van der Waals forces to give the structure some 'adhesive' properties.

Graphene layered with hexagonal boron nitride and others solve the band gap problem. I couldn't get the paper on creating vertical structures by recruiting Van der Waals forces easily (I have access to most publications through the school library). I recognize all the math from QM and E&M, but I can only follow along now, I've lost the ability to solve the equations by my self. But, speaking of school, I have to give this up till I finish my Masters, we'll see if I'm still hooked afterwards.

I will say, these sure are interesting times for in terms of developing and deploying new semiconductor device structures. If I were studying physics today, I'd definitely look for an applied Physic program that was strong in semiconductor device physics.

 

[Idontcare]

If I were studying physics today, I'd definitely look for an applied Physic program that was strong in semiconductor device physics.

The advanced degree you'd want is called "Chemical Physics", it combines all the grad level quantum mechanics that comes with a PhD in Physics along with all the chemistry of electronic materials that comes with a PhD in Physical Chemistry.

Combine that with a BS in materials science engineering, throw in a couple minors in mathematics and chemistry on the side, and you'll do alright for yourself as a process development engineer

I'm still stuck on this because a gate width of 150 pm just blows my mind.

And keep in mind that is just the limits of scaling if we confine ourselves to creating electronic components that physically manifest (and limit themselves to) the electronic structure of molecules.

The electronic structures allowable by quantum mechanics (structural materials, think meta-materials popularized lately by so-called cloaking devices) is really only bound by imagination, as are the electronic structures allowed in the sub-atomic length scale regime.

 

[Ajay]

The advanced degree you'd want is called "Chemical Physics", it combines all the grad level quantum mechanics that comes with a PhD in Physics along with all the chemistry of electronic materials that comes with a PhD in Physical Chemistry.

Combine that with a BS in materials science engineering, throw in a couple minors in mathematics and chemistry on the side, and you'll do alright for yourself as a process development engineer

How rude of me, first, thanks for the this info! Now, I will ramble on as usual...

BBBut, I don't know if I can handle being called a Chemist
Funny, Quite a few of Intel's fellows are Physicists, but obviously there is a need for a strong understand physical chemistry.

Oh, and I already have a degree in Physics and found out that we couldn't get a Minor in Math because it was already implied by the Physics degree (and in fact, I actually learned higher level math concepts beyond my math classes in my Junior and Senior year). The funny thing is that I'd need to learn the math all over again (but it wouldn't take me long, I'm just rusty).

 

[arsta]

New scientific discoveries and successful research projects.

 

[Cookie Monster]

It sure does sound weird when one day we will be saying "pico-meters"

Whatever the next solution(s) is, times running out.

 

[yottabit]

...Performance can be increased by one of two ways - (1) increase the rate at which work gets done (more clocks per second), or (2) increase the amount of work that gets done per clock....

I think you could probably add a 3) Reduce the amount of work that needs to get done

Both through programmers being more performance-optimization focused and by CPU makers creating new architectures

I think stagnating performance could be a reason to abandon or heavily modify x86 even at the expense of abandoning a lot of legacy support.

I also think it's pretty inevitable that a lot more emphasis will be placed on program optimization. And perhaps having a fast computer will be less about the hardware you have inside it but the software you are running (which would also carry the corresponding premium price) More standardized hardware platforms would allow for more intense optimization.

Just my theories!

 

[Idontcare]

How rude of me, first, thanks for the this info! Now, I will ramble on as usual...

BBBut, I don't know if I can handle being called a Chemist
Funny, Quite a few of Intel's fellows are Physicists, but obviously there is a need for a strong understand physical chemistry.

Oh, and I already have a degree in Physics and found out that we couldn't get a Minor in Math because it was already implied by the Physics degree (and in fact, I actually learned higher level math concepts beyond my math classes in my Junior and Senior year). The funny thing is that I'd need to learn the math all over again (but it wouldn't take me long, I'm just rusty).

Yeah, the basic premise is just that if you have the mental aptitude and natural tendency to be multidisciplinary in the sciences then it really doesn't matter what you start with because it will all come down to how you apply yourself once you are in the job.

I've worked with MIT PhD grads who couldn't come up with an innovative thought to save their lives, they were basically techs who needed to be told step-by-step what to do for their "research", and I've worked with techs who had 2-yr associate degrees who could wrap their minds around problems that were truly tricky to solve and yet they did it with aplomb.

The college degrees give you knowledge and mental training, all good and necessary, but they can't give you a work ethic or passion for your trade. Nor will passion and enthusiasm alone bridge the gap to being able to make informed credible decisions for all the right reasons.

You have to be the full package, personality wise and educationally. But the specifics of the education, outside of opening doors and getting you a job, are not all the critical as you move along in life.

My degrees were pretty much entirely irrelevant to what I was doing in real life about 5years after I left college. From then on it was all about what I had done with myself since leaving the university, education was just a checkbox to fill in at that point. And GPA/grades became even less relevant all the sooner. The only place that ever cared about my undergrad GPA was the graduate school's entrance officer, not a single living soul has ever asked me again.

 

[Idontcare]

I think you could probably add a 3) Reduce the amount of work that needs to get done

Both through programmers being more performance-optimization focused and by CPU makers creating new architectures

I think stagnating performance could be a reason to abandon or heavily modify x86 even at the expense of abandoning a lot of legacy support.

I also think it's pretty inevitable that a lot more emphasis will be placed on program optimization. And perhaps having a fast computer will be less about the hardware you have inside it but the software you are running (which would also carry the corresponding premium price) More standardized hardware platforms would allow for more intense optimization.

Just my theories!

Oh absolutely :thumbsup: I was thinking in terms of assuming you have a specific instruction to execute (A + B = C, how to get C computed in less time), but you are very much correct as well.

New instructions are created for the same reasons you touch on, FMA for example, get more done in the same time. Likewise with register renaming tricks, all ways to get the same work done while doing less work in the process.

Those performance opportunities are the very premiums we pay when we decide to use generalized processors instead of fixed hardware. Fixed hardware can always be more highly tuned for performance, both in hardware as well as in software...but at a cost of losing flexibility and possibly applicability.

A computer processor that is designed to be wicked fast at Photoshop might be a dog at Excel, so is there an incentive to alienate the Excel-facing market in developing that Photoshop-optimized processor?

Tradeoffs are just that, you trade one thing for another, and the tradeoff of fixed versus general hardware has been there since the creation of the first general-purpose processsor. That gap can be reigned in, in exchange for more performance as we see with console gaming for example, but it comes at the expense of something else - marketshare and TAM - which makes it the so-called "trivial solution" or uninteresting outcome.

What folks are really interested in is how can the performance of general processors continue to be improved upon when the existing electrical component of choice (the transistor) can no longer be effectively (economically) shrunk.

Remember it was economics, not limits of physics, that prompted electronics engineers to replace vacuum tube circuits with transistors. And IMO transistors will be supplanted for the same reason when the time comes (economics).

 

[BTRY B 529th FA BN]

May I step in and kindly drop this link here? it's probably stupid, i don't know what i'm doing

 

[Sleepingforest]

I attended a brief lecture by Chenming Hu at UC Berkeley (former CTO of TSMC) and he believes that there are several viable options for shrinking past 5nm:

• A physical (though obviously small, on the order of just a molecule or two thick) gate which closes mechanically.
• Quantum particles rather than electrons going from source to sink.
• A new, non-silicon material like graphene.
• Shrinking the distance from source to sink (or drain, whatever you want to call it. Source/sink was always catchier to me).

He only went over it at a very basic level, but I found it pretty interesting.

 

[epidemis]

I am currently self-employed in the industry of foreign currency exchange. Designing algorithms that autonomously trade (buy/sell) foreign currencies without human involvement in the process.

:O You work for evil

 

[Idontcare]

:O You work for evil

As I work for myself, your post carries intriguing implications :hmm:

Mind you I am a "PhD", so technically you should state "you work for Dr. Evil"...to which I could reply (with pinky finger firmly planted near the corner of my lips) "time for some freaking laser beams! :twisted:"

 

[ShintaiDK]

as i work for myself, your post carries intriguing implications :hmm:

Mind you i am a "phd", so technically you should state "you work for dr. Evil"...to which i could reply (with pinky finger firmly planted near the corner of my lips) "time for some freaking laser beams! :twisted:"

lol