If your talking about integration of more features into the silcone die space. this will definately increase performance. But then everything will start looking like a MCU. hmmmm I think I siad this a few months ago. At some point soon I expect ram, gpu and most if not all outboard chips combined into one large die and stuck to a board. that will leave only perpheral interconect like outboard acessories etc to connect or maybe more copresseing sollutions like GPU.
did you not read what I siad ? even with light interconnects it won't massively change latencys or transmission speeds. The wall of compute power given our current computing methodologys is reaching its limits. Serial processing has hit the wall or is very very close to it. Even clockspeed increases aren't going to help.
certainly more and more things will be integrated, this has been going on forever, the first computers were room sized made out of vacuum tubes, the first microchip based computers had what we today called CPU in several seperate chips.
for example: http://en.wikipedia.org/wiki/Math_coprocessor
yes, back in the day the CPU was integer only, and you had a separate chip to perform floating point math.
I read what you said, I wasn't trying to contradict you, I was merely clarifying something.
Sure, also CRT TV's Cathode Ray Tube is basically a vacuum tube with a hole, and vacuum tubes themselves are funny shaped incandescent light bulbs without the filament. There are certainly still uses for them, but those uses are in electrical engineering, not in creating logic gates for computation.Tubes are very interesting though. did you know that norad and most TV stations still use Tubes to generate high power signal outputs and tubes are one of the few devices that are good at ultra high frequency amplification. although transistors have been catching up.
Same with SiGe. Not saying it is likely Intel is doing III-V on SOI, just saying at this time there isn't anything overtly contradictory about the rumor, it ranks high on the plausible meter.
Both Mark Bohr and Kelin Kuhn said they'll stick to conventional transistors didn't they? That was also this year. Unless they have made some breakthrough during that time... is it even possible to have a change such late in the game?
And equally positioned guys at Intel had no problem saying HK/MG was not needed...
I wonder what benefit we shall get with transistor size as interconnect latency becomes a overiding factor regardless of node size. IE the limit of the speed of electrons will become a bottle neck regardles of how small they make the transistor.
Both Mark Bohr and Kelin Kuhn said they'll stick to conventional transistors didn't they? That was also this year. Unless they have made some breakthrough during that time... is it even possible to have a change such late in the game?
http://techon.nikkeibp.co.jp/english/NEWS_EN/20100812/184966/?P=3
Says they don't know if they will switch at 15nm
You were downplaying them using FD-SOI and other more exotic tech for a while. Does it mean you are playing the same "fading and dashing the other way" too?
However, if they are intending to further pursue the medical field as well as the consumer mobile field then such a process technology transition may well be necessary if only to enable the performance of their products to be such that they can command the sort of price that enriches their gross-margin mix in the direction they know their shareholders are expecting them to.
his point is, I think, that intel lies to the press to mislead AMD, nvidia, etc; something that it HAS done before.
what do you mean the medical field? as in, implant chips into human bodies? wouldn't they have to be sealed inside something somehow anyways?
It's deeper, and more capitalistic than that, and to some extent we all do it in this business.
The real heart of the matter is IP.
The longer you can keep people (universities, suppliers, competitors, from investigating the same area of technology that you are interested in the more likely you are to plow that fertile IP field all to yourself.
Then when folks like IBM and GloFo come along years later they find themselves doing stuff like gate-first integration and so on not because of its technological superiority but because they can't get around your patents and you are unwilling to pay the royalty fees associated with them.
And it prevents upstart IP companies from coming in after the fact and gouging you in court with injunctions just because they filed a patent before you did, them knowing or suspecting in advance that someday soon you'd be trying to put that technology to work inside your billion dollar fabs.
Yes, and tiny, and powered by something that is also as tiny as possible for the same reasons. And tiny batteries means tiny power output.
thank you for clarifying, that makes a lot of sense. With IP rules being what they are it stands to reason that you would want any development you make to be kept as secret as possible... Although I wouldn't call it capitalism, IP is pretty damn antithetical to capitalism. it is literally a government mandated monopoly granted to "encourage research".
ah, I was thinking toxicity rather then power consumption
Toxicity is an issue, but its actually us that are the toxic part of the equation. We want these implants to last, and they don't last too long when exposed to the types of chemicals and environments that are found naturally in our bodies.
An implanted device would actually have to be pretty large in volume to contain enough nasty stuff as to pose a first-order health risk to its host.
Even batteries that people ingest are not corrosive enough to really cause problems, the real problems with batteries is the metal casing becoming corroded and then representing an abrasion opportunity for lancing your colon or stomach wall...then internal bleeding becomes the issues, not the exposure to lithium or cadmium.
In process development that "wall" was hit back around the time of the 90nm node.
Electron and hole mobility is a metric that is engineered into the devices, it matters to first-order. As does the wire-speed and associated capacitance.
You will see the push for 3D IC's become more and more common as a means to reduce the RMS distance between latency-bound circuits as node shrinks continue.
In regards to serial performance improvements and clockspeed increases, we need look no further than your typical extreme overclocking get together to see ample examples of clockspeed capability and serial-threaded performance improvement.
Those clockspeeds only achievable today on extreme cooling are enabled at room-temperature on tomorrow's node shrink...if the manufacturer believes that people will pay a premium for the performance.
I would be hesitant to look at the specs of commercially available processors as rendering some indication of inherent process technology or device physics limitations afoot.
When engineering collides with marketing and sales you get some intriguing results, albeit not the sort limited by physics unless you are running the LHC.
theres a whole multitude of applications for integrating computing into the medical field. say for example you have a patient that needs a precisely administered dose of a medication over a long period of time. the right chip could analyze your body chemistry, decide on a dose, and depending on the medicine, even administer it. there's also the entire field of neural analysis. there will come a time one day when we will find a way to integrate our own brains with our computing technology. imagine getting a math coprocessor implant and a solid state memory backup that you can reference with 100% accuracy on demand. it could have serious implications yes, but there ARE also diseases which affect peoples ability to convert short term memory to long term memory as well, and these people, with a memory implant of this type, could actually begin remembering things (again) and start to lead a somewhat normal life again. the even greater benefit is that if everyone were to receive these kind of implants, we would NEVER loose a single bit of a person when they die. there are so many ideas and thoughts people have that never get truly utilized because of where that person was within society in relation to what their idea meant, and upon death that persons memory could become part of the greater cognitive whole of the human race going forward. of course its also hard to say how much capacity the human brain really has for memory, as it doesnt store data in a binary manner. its possible that it will still be a long time before storage technology even catches up, but on the flipside its possible that it would lead to a massive expansion of what we can actually remember as well. and thats just the tip of the iceberg, of integrating chips into us. these things work in reverse as well dont forget. im sure there's many a computational scientist who wonder how much processing power the human brain really has, and what its capable of if you can find a way to utilize it fully.Medical field was mentioned a few times in this thread. What kind of devices are you guys thinking about?
My point is that serial increases relative to clock speed are at the limit and you will see a inverse relationship going forward.
IE a 2.6 ghz cpu running at 5.2 ghz does not scale 200% it hardly if ever scales 150%. those are dimishing returns. I expect them to actually get worse at 22nm vrs now at the current 32,40 and 45nm ranges.
if we had a 2.9 ghz cpu disapating 125w today at 40nm at 22nm its not going to be massively better. its not going to be like the jump from 90 to 45nm.
Not to mention the cooling issues.
pretty much, node shrinks have hit a limit and the technology we are using has come to the end of the road.
Even considering 3d IC designs. Thats a bridge to nowhere with massive thermal issues. Thats why no one is doing it now.
the real answer is looking at ways to fundementally change the computing market. silicon is dead. It just doesn't know it yet.
ModestGamer you probably don't know my background, I was a process technology development engineer at Texas Instruments...worked on nodes spanning from 0.5um down to 32nm with continuing government grants and university affiliations (dual employment as adjunct professor at a local university as well as process development engineer at TI) working on sub-16nm technology extensions and it is my professional opinion that silicon is most certainly not dead nor is 22nm "nearing the end".
That's not just mantra in the form of "they always say Moore's law is dead and they are always wrong"...I'm speaking from having played a role in the development of the really cool stuff that hasn't even come to the forefront of the media's attention. And that really cool stuff has legs to run another 20yrs.
Without a doubt the limits of scaling are 100% economical. The materials science and basic physics is already at our disposal for continuing to shrink circuits for decades to come. And by "shrink" I mean increasing xtor density and electrical performance, but not the kind of 1-dimensional usage of the term as used by laymen where they think of the channel narrowing, etc.
When I see "scaling is dead" arguments, and they are popular so I see them often, I just think it is such a needlessly self-defeating perspective. But there is the big question mark regarding cost and whether enough consumers can be pooled together to buy products which will justify developing the shrink in the first place.
But that is true of virtually any product on the drawing board right now in any industry. Does the TAM justify the development expense? At TI we decided it no longer did at 32nm so we shut down R&D and went fabless for CMOS logic at 45nm and smaller. (kept the fabs for the still lucrative analog stuff)
One last thing, 5nm in the traditional node-scaling label schema is where currently known/identified scaling solutions are expected to lose steam. That's if we do nothing to expand the edge of the envelope between now and when 5nm is released to production.
What is being done between now and when 5nm is released to production is improving the integration and process technology so as to reduce manufacturing costs and increase the robustness of in-field product lifetime.
edit: This thread has all kinds of good stuff in it if you are interested in more reading.
And this post from that thread definitely hits on stuff related to the OP of the current thread.
(and when you read the stuff in that thread keep in mind we only make the tip of the iceberg public information, the bulk of what is going on is never publicized for all the obvious competitive reasons...having seen and done some of the stuff that isn't public yet you'll just have to take my word for it when I say that shrinking is nowhere close to dead)
I figured it was smaller then 22nm my point "thanx for the links" was that the type of return on continued scalling will get smaller and smaller as the wall approachs, we are approaching the wall. The other side of my stance has less to do with the silicon itself and more to do with the overall design principal in use intoday computers.
Lets say that shrinking does help, to what extent ? at what extent is the serial binary model simply not going to continue to feed our ever growing need for compute power.
I think you know where that plank leads.
at some point regardless of how good the silicon is, when the wall is reached what will be left ?? thats where the deadend occurs.
do you follow what I am saying here ? we are using 1d solutions to 4d problems.
500 yrs ago mankind was doing its best to figure out how to apply all of its accumulated maritime knowledge to just get across the Atlantic ocean.
50yrs ago mankind did its best just to figure out how to apply all of its accumulated spacetravel knowledge just to get from here to the moon.
I'm not worried about the what if's when it comes to technology scaling. I participated in ITRS roadmap recommendations and formulation over time, if the experience has taught me anything its that mankind has no shortage of self-doubt and at the same time no shortage of people willing to prove the doubters wrong.
Do I know how to get from where we are today to where we will be in 20yrs? Yes.
Do I know what we will actually be doing 20yrs from now? No. Economics dictates that and today's uneconomical solutions for tomorrow will be replaced with more economical solutions before tomorrow gets here.
Do I know how to get from where we are today to where we will be in 50yrs? No. No more so than to point at the period table of elements and say mankind has yet to produce the children of tomorrow that in 50yrs time will be coming up with the crafty stuff they will need for their generation to go wherever they aim to go.
But I am confident they will do something that from our perspective is simply impossible, we are merely crossing the atlantic ocean of computing.