Intel Server Microarchitecture Roadmap until 2018 - ComputerBase

[Olivon]

http://tinyurl.com/skymont

[qcmadness]

Why Intel goes for half-node after 22nm?

[trn]

Ivybridge is so outdated already, I want 10nm

[Origin_Unknown]

Ivybridge is so outdated already, I want 10nm

better breakout the WCG time machine

[Hornet331]

Seems like intel skips the ITRS after 22nm.. next nodes would be 16nm and 11nm

[dcl]

A naive question...

Has/can Intel actually produce a 14/10nm chip at the moment? Like as a prototype?
Do they have the tech/hardware/know how to do so? Or is it just a goal at the moment?

I love the idea that there is some desktop rig deep inside Intel labs that is 5 years ahead of the curve.

[PatRaceTin]

good roadmap

[Chumbucket843]

A naive question...

Has/can Intel actually produce a 14/10nm chip at the moment? Like as a prototype?
Do they have the tech/hardware/know how to do so? Or is it just a goal at the moment?

I love the idea that there is some desktop rig deep inside Intel labs that is 5 years ahead of the curve.

It's very like that they do but it's not as functional as you are thinking and it's purely for research purposes. i cant confirm they do because i dont work there and they have not published anything on it yet.

[paulbagz]

Oh God not another roadmap thread.

-PB

[sin0822]

maybe they will start using graphene LOL

[Aberration]

It's very like that they do but it's not as functional as you are thinking and it's purely for research purposes. i cant confirm they do because i dont work there and they have not published anything on it yet.

The lithography technology does not yet exist.

[Chumbucket843]

The lithography technology does not yet exist.

e-beam should be able to do 15nm-ish. btw, there are some cutting edge techniques that have been used to create sub-20nm SRAM i.e. nanoinjection lithography.

[[XC] Synthetickiller]

I just bought a Xeon E3 and it's already old as dirt. 32nm is so thick....

In b4, "Intel won't make the dates they claim on this roadmap."

[Aberration]

e-beam should be able to do 15nm-ish. btw, there are some cutting edge techniques that have been used to create sub-20nm SRAM i.e. nanoinjection lithography.

And all that technology is used to test a new transistor or small memory node. They do not have the capability to make a full processor. That is all just software modelling of what would be capable.

With the current trend in photolithography, and Nikon bailing out of EUV, passed 22nm is going to be expensive to manufacture. Triple, maybe quadruple patterning, which drastically increases manufacturing time and cost. They are really hitting a wall of what is possible without a breakthrough.

[RussC]

And wow, INTC will be at 22nM till middle 2015. Mmmm

RussC

[Olivon]

And wow, INTC will be at 22nM till middle 2015. Mmmm

RussC

That's a server roadmap, not desktop.

[saaya]

so intel officially dropped the 12 month tick-tock cycle?
22nm and 14nm look a bit optimistic to me...
its getting harder and harder to shrink, so why does intel think it will suddenly get easier from here on?
and im curious why intel is still pushing for smaller nodes so agressively now that they have multigate finfets...

EDIT: oh and i wonder since ivy bridge slipped a quarter already, will this shorten its lifecycle or is the entire stack pushed back a quarter...
so this roadmap might already be outdated ^^

[kuroikenshi]

^^.... I know... crazy stuff there, bring back 90nm process!! and tape cpcs'!!!

[Hornet331]

so intel officially dropped the 12 month tick-tock cycle?
22nm and 14nm look a bit optimistic to me...
its getting harder and harder to shrink, so why does intel think it will suddenly get easier from here on?
and im curious why intel is still pushing for smaller nodes so agressively now that they have multigate finfets...

EDIT: oh and i wonder since ivy bridge slipped a quarter already, will this shorten its lifecycle or is the entire stack pushed back a quarter...
so this roadmap might already be outdated ^^

Servers lag a year behind consumers now...

[Tenknics]

gotta love the Saaya spin..this is where most users get confused and misinformed. Conclusion jumping isn't a safe sport, especially for you Saaya.


Intel isn't dropping tick-tock, oh eh vei

[RussC]

Servers lag a year behind consumers now...

So desktop is 1 yr ahead of server chips...that doesn't seem right to me. Still, that makes 22nM more than 2 years from now.

RussC

[Aberration]

so intel officially dropped the 12 month tick-tock cycle?
22nm and 14nm look a bit optimistic to me...
its getting harder and harder to shrink, so why does intel think it will suddenly get easier from here on?
and im curious why intel is still pushing for smaller nodes so agressively now that they have multigate finfets...

EDIT: oh and i wonder since ivy bridge slipped a quarter already, will this shorten its lifecycle or is the entire stack pushed back a quarter...
so this roadmap might already be outdated ^^

The shrinking is not so much a problem of Intel, its a problem of the vendors. If Nikon or ASML can't print it, then it's not going to happen. Nikon is practically giving up on EUV, and ASML is not having such an easy time of it.

I expect things to get MUCH lower. Which might not be a bad thing, we may see some really innovative architectures.

[Hornet331]

So desktop is 1 yr ahead of server chips...that doesn't seem right to me. Still, that makes 22nM more than 2 years from now.

RussC

With SB intel switched to that scheme, consumer products get the lower process node first and then server line follows. Ofcourse they can change it anytime they want again. But currently it looks like that. -> SB desktops beginning of 2011, for Server end of 2011 and the same seems to apply for Ivybridge -> beginn 2012 and server will come by the end of 2012 (maybe even early 2013)

They still stick to yearly tick-tock, just that servers now lag behind consumer products for approx. 1 year.

[saaya]

Servers lag a year behind consumers now...

you mean server platforms?
if servers would lag behind 1 year then there shouldnt be sandy bridge xeons, but there are, there have been for quite a while...

gotta love the Saaya spin..this is where most users get confused and misinformed. Conclusion jumping isn't a safe sport, especially for you Saaya.
Intel isn't dropping tick-tock, oh eh vei

who said intel is dropping tick tock?
id appreciate it if you read my post before criticizing me, thanks
and what exactly am i "spinning"?

The shrinking is not so much a problem of Intel, its a problem of the vendors. If Nikon or ASML can't print it, then it's not going to happen. Nikon is practically giving up on EUV, and ASML is not having such an easy time of it.
I expect things to get MUCH lower. Which might not be a bad thing, we may see some really innovative architectures.

you mean slower? that would make sense, investments in new tech have increased a lot in the past years... at some point its hard to justify those investments i think...

[Aberration]

Yes, of course slower. A new immersion tool costs about $65M, and an EUV alpha tool is like $100M. Serious money, the Immersion tools foot print was huge compared to the dry tools, and an EUV is even larger. Not to mention they are now in a vacuum, which makes service MUCH more difficult. There are times just me opening a chamber door can take a tool down for over a day. I can't imagine what needing vacuum will do. Considering Intel makes ~8-10B a year, the money per tool is not so much the problem. Its the entire cost of ownership, and throughput.

Getting a cost effective process below 22nm is looking to be extremely difficult.