The Intel Nanometer - Lies

[shady28]

It seems that Intel measures and defines process technology different than everyone else.

See references, 22nm Intel = 26nm by the industry definition :

http://www.electronicsweekly.com/bl...nductor-blog/2013/02/the-intel-nanometre.html

"As the world now knows, there's a nanometre and then there's an Intel nanometre.
To most of us a nanometre is a nanometre but, to Intel, a nanometre measures 1.182 nanometres.

We know this because the measure of a semiconductor manufacturing process is the drawn gate length.

And, on Intel's so-called 22nm process, the drawn gate length is 26nm."



http://www.electronicsweekly.com/bl...g/2012/10/intel-has-no-process-advantage.html

"Jean-Marc Chery, CTO of STMicroelectronics points out that the drawn gate length on Intel's ˜22nm" process is actually 26nm."


http://www.eetimes.com/electronics-news/4403320/Fur-flies-over-FinFETs-and-future-in-IEDM-panel

“Any time we talk about new nodes, we should wash our mouths out with soap,” said Scott Thompson, chief technologist of startup SuVolta and a former Intel fellow. Engineers ignore traditional metrics, saying “Intel’s 22-nm node is really 26 nm, so if Intel does new math, so will we,” he said.

This helps explain why Intel keeps losing to 28nm process ARMs. x86 has some overhead, and there really isn't that much difference between Intel '22nm' and Samsung / GloFlo / TSMC 28nm.

On top of all that, Intel's 14nm node for 2014 is estimated to be - guess what - not 14nm. It's 18nm.

Meanwhile, TSMC says 20nm (a real 20nm, vs Intel's 26nm '22nm') comes online with full production in Q2 2013 :

http://www.fudzilla.com/home/item/30939-tsmc-20nm-fab-14-coming-on-line-soon


And the rumor is - Apple will be their #1 customer there.

http://www.phonearena.com/news/TSMC...r-might-land-in-next-years-iPad-first_id40826

TSMC is expected to start the A7 on 28nm, then go to 20nm later this year.

That means Intel will (in reality) be doing a 26nm process in 2013 vs TSMCs 20nm process (assuming TSMC really starts banging out 20nm A7's).

 

[R0H1T]

Let the fireworks begin ()

 

[Charles Kozierok]

As it happens, we just had a pretty lengthy discussion of these sorts of issues a few days ago. Start at post #116 and work your way through. If you want to save time, just look for posts by Idontcare.

 

[Phynaz]

Global Foundries 20nm gate pitch = 80nm
Global Foundries 14nm gate pitch = 80nm

 

[Homeles]

This helps explain why Intel keeps losing to 28nm process ARMs.

Boy are you clueless. They're losing because the underlying architecture of atom is 5 years old.

 

[shady28]

Global Foundries 20nm gate pitch = 80nm
Global Foundries 14nm gate pitch = 80nm

And Intel's range from 90-108, though many places quote intel as stating 80 :


http://semiaccurate.com/2012/12/10/intel-talks-about-22nm-soc-transistors/


Chart :
https://semiaccurate.com/?attachment_id=24493

 

[GreenChile]

Gate length is not the only metric for measuring a process node. In fact, 'industry definitions' vary between different companies. There is no strict industry wide definition.

Kind of like the definition of TDP. It depends on which company you are asking.

Intel's process technology is widely known to be the most advanced in the world so I don't know what you are trying to prove here. It kind of seems like trolling.

 

[Homeles]

It kind of seems like trolling.

It only "kind of" looks like trolling? It is trolling.

 

[nehalem256]

Boy are you clueless. They're losing because the underlying architecture of atom is 5 years old.

Also the fact that atom is still on a 32nm process anyway.

 

[Khato]

We know this because the measure of a semiconductor manufacturing process is the drawn gate length.

The author of this blog should really look at published numbers for previous process technologies. My favorite source for such being this excellent chart from Real World Technologies - http://www.realworldtech.com/includes/images/articles/iedm10-10.png?dc2136 Notice how drawn gate length parameter, LGate, has pretty much no correlation with the process node? More specifically, how it's typically a bit lower on the high performance process target and then occasionally higher for the low power process target?

Regardless, one simple way to confirm that the claim of Intel's 22nm process being correctly labeled according to industry standards in to compare minimum SRAM cell size. TSMC's 28nm HP is 0.127um^2, Intel 22nm is 0.092um^2, while TSMC's upcoming 20nm process is 0.081um^2 - Intel's pretty much exactly where they should be with their 22nm label.

Which actually does bring up an interesting quirk... While they didn't mention the finfet part at the time, Intel published SRAM cell sizes for their 22nm process in September of 2009, roughly 2.5 years before the first product using that process. But now we're down to only about 1 year before we should see the first products based on their 14nm process and they still haven't released any details?

 

[R0H1T]

It only "kind of" looks like trolling? It is trolling.

It kinda depends which side of the fence you're on ! If he's saying that Intel "lies" then yup they do(within reason) but so does everyone else, it isn't a one way street & you'd be well advised to not take it any further than that !

 

[Homeles]

It kinda depends which side of the fence you're on ! If he's saying that Intel "lies" then yup they do(within reason) but so does everyone else, it isn't a one way street & you'd be well advised to not take it any further than that !

Everyone else does it as well. This foundation for this thread is entirely comprised of ignorance.

Also the fact that atom is still on a 32nm process anyway.

Lol, that too. Good going, OP.

 

[Phynaz]

And Intel's range from 90-108, though many places quote intel as stating 80 :


http://semiaccurate.com/2012/12/10/intel-talks-about-22nm-soc-transistors/


Chart :
https://semiaccurate.com/?attachment_id=24493

That blew right over your head. Technically GloFo's 14nm is the same size as their 20nm because the gate pitch didn't change.

Your best approach to this would have been to ask what this meant instead of popping off with inflammatory accusations. You would have saved yourself from looking stupid.

We know this because the measure of a semiconductor manufacturing process is the drawn gate length.

You appear to be the only one that knows this. But it must be true because you read it on the internet - on a blog no less!

 

[shady28]

Boy are you clueless. They're losing because the underlying architecture of atom is 5 years old.

Correct about 5 years (kind of), but also meaningless and 'clueless' of you to make such a generalized statement with no context...

ARM's underlying architecture for most current chips (Coretex A8, A9, A15) is ARMv7. It was introduced in 2006. It's 2 years older than Intel's Atom.

http://web.eecs.umich.edu/~prabal/teaching/eecs373-f10/readings/ARMv7-M_ARM.pdf

"Copyright © 2006-2010 ARM Limited. All rights reserved."

https://en.wikipedia.org/wiki/ARM_architecture

"...the Cortex A-Series (ARMv7) in current high-end devices. ARMv7 includes a hardware floating point unit, with improved speed compared to software-based floating point."

So ARMs 7 year old architecture on larger process nodes is beating Intel's 5 year old architecture. Thanks for clarifying my point.

 

[ShintaiDK]

I guess these "ARM rules, Intel suck" threads is the new standard.

Just a shame the OP didnt do his homework atleast. But hey, he wouldnt be able to trigger a drama then and troll everyone.

 

[Arachnotronic]

I guess these "ARM rules, Intel suck" threads is the new standard.

Just a shame the OP didnt do his homework atleast. But hey, he wouldnt be able to trigger a drama then and troll everyone.

Yeah, what can you do? Semiconductors are too advanced for most simple brains to understand.

 

[Phynaz]

He could have gotten away with it if he didn't use the troll thread title

 

[Homeles]

ARM's underlying architecture for most current chips (Coretex A8, A9, A15) is ARMv7. It was introduced in 2006. It's 2 years older than Intel's Atom.

http://en.wikipedia.org/wiki/False_equivalence

He could have gotten away with it if he didn't use the troll thread title

He could have gotten away with it if he wasn't a known AMD apologist.

 

[rgallant]

the following reply will be written @ 28nm and 22nm tell me if you see the difference in size
.
.



thanks for reading

 

[Idontcare]

Ah yes, yet another strawman thread masquerading as the latest anti-Intel thread :\

I also heard Intel thinks Pi = 3.0 and they single-handedly forced Pope Benedict XVI to step down for refusing to exorcise the demons from Intel's USB3 chipsets.

Its all true, because you just read it, on the internet no less. That makes it like doubly or triply true, I read that on the internet too.

 

[Arachnotronic]

AMD/ARM pumpwads...unite?

 

[Markfw]

Too many complaints of this being a troll thread, so I am locking it.