REGISTER | LOGIN
Breaking News
News & Analysis

Intel Unveils 10, 22nm Processes

Transistor-density metric proposed
3/28/2017 08:45 PM EDT
23 comments
Page 1 / 4 Next >
More Related Links
View Comments: Threaded | Newest First | Oldest First
Left5
User Rank
Rookie
Sorry, but Intel does not have process lead
Left5   3/29/2017 9:02:25 AM
NO RATINGS
Firstly, TSMC and Samsung's 10nm (currently in production) is denser than Intel's 14nm - and that is even shown in Intel's slides.

Secondly, despite the fact that Intel want to pretend they do not exist, TSMC's and Samsung's 7nm do exist and are coming in 2018. TSMC's 7nm is as dense as Intel's 10nm. Intel will launch 10nm in very low volumns at end of 2017, but that's really only risk production - volume production will happen much later.

 

witeken
User Rank
Freelancer
Re: Sorry, but Intel does not have process lead
witeken   3/29/2017 12:07:40 PM
NO RATINGS
Coming in 2018. On what volume? Late in the year (7nm not even taped out yet). So that will still be a year lead from Intel. And how do you know it will be just as dense? Firstly, TSMC has looser metal pitch (40nm vs. 36nm). Secondly, does TSMC have the single dummy gates (20% density improvement) or the COAG (10% IIRC). So end-chip density might still be appreciably higher; time to market will be up to a year earlier; and time to market also depends on what yield a company is willing to go into manufacturing which I strongly suspect is lower for TSMC. And lastly, how will performance and power compare? For instance, Intel will be in its 2nd generation air gaps.

 

Left5
User Rank
Rookie
Re: Sorry, but Intel does not have process lead
Left5   3/29/2017 12:41:40 PM
NO RATINGS
TSMC customers will tape out 7nm designs this quarter. The volume will be high, for the iPhone to launch. Yield has to be around 90%+ for it to enter volume production.

http://appleinsider.com/articles/17/01/03/tmsc-on-track-for-volume-production-of-7nm-chips-for-iphone-ipad-in-2018

Intel only release 10nm at end of 2017 for only low volume 2 core products and there is doubt that they can even deliver that: http://marketrealist.com/2017/03/could-intels-10nm-node-face-another-delay/ 

Obviously, performance is targetted towards smartphone SoC's, not PC's, but they will have a 7nm HPC process later.

m00nshine
User Rank
Author
Re: Sorry, but Intel does not have process lead
m00nshine   3/29/2017 1:06:16 PM
NO RATINGS
Which company has the "best" processors? The one that has the "best" metric? How about the company that makes the most $$ with their chips has the best chips? Also, this http://www.taiwannews.com.tw/en/news/3121282 Also, I thought betamax was "better" than VHS...

Gondalf
User Rank
Author
Re: Sorry, but Intel does not have process lead
Gondalf   3/31/2017 7:14:25 PM
NO RATINGS
I remember you TSMC is only a foundry :). An IDM it's a lot different in needs and investiments. There is another issue, in the medium term a simple foundry is the more sensitive to the China push in silicon tech. A foundry could be dumped quickly by a cheap competitor capable to offer reasonable silicon at lower price just to gain market share. In the long term IDMs like Samsung and Intel are in a far better shape. 

Left5
User Rank
Rookie
Re: Sorry, but Intel does not have process lead
Left5   3/29/2017 2:04:49 PM
NO RATINGS
Also, while it's true Intel has tighter metal pitch (36 versus 40nm), TSMC is using a 6 track standard cell, while Intel is using 7.5 track. So, considering the other factors (dummy gates, etc.) it's not very clear which is denser.

photonic
User Rank
Author
Re: Sorry, but Intel does not have process lead
photonic   3/29/2017 3:17:27 PM
NO RATINGS
This is the story that is important, not how many companies Intel is buying.

paul.jhnson
User Rank
Rookie
Re: Sorry, but Intel does not have process lead
paul.jhnson   3/29/2017 9:16:14 PM
NO RATINGS
Intel clearly has process lead if it can deliver at 10nm with impressive 54 gate/36 metal pitch. No doubt it is giving them a distinct advantage over competitors. This is separate from layout being 6T/7.5T which in the end affects the density, but even then, I think Intel still seems to be ahead on average.

The big question is, what is the performance gain compared to the previous generation. That seems to be shrinking. Right now it doesn't sound to offer a big leap, but is to be seen.

BSFoundries (not even worth mentioning in the same article) can only dream about such densities. They better hope Intel 22FFL won't make FDSOI obsolote before they can export jobs and IP over to China. It's too much to ask for from a company where developing technology is only an afterthought. Especially when they lack the talent to do so.

22FFL rules look very easy for Intel to yield quickly, will be very interesting to watch if it performs as promised. Competition from TSMC will be fierce.

resistion
User Rank
Author
Will there be delays?
resistion   3/29/2017 9:53:02 AM
NO RATINGS
The proof is in the timely release. Will Cannonlake arrive on time?

Their Manufacturing Day announcement means they froze their 10nm process and should have at latest just started their pilot, which should be using Cannonlake as the vehicle.

rick merritt
User Rank
Author
Late responses from TSMC, GF
rick merritt   3/29/2017 11:51:15 AM
NO RATINGS
This morning I added into the story late comments on Intel's metric and 22FFL I got in overnight from GF and TSMC.

rick merritt
User Rank
Author
Do we need a standard metric or metrics?
rick merritt   3/29/2017 8:40:42 PM
NO RATINGS
Good discussion here. I'd like to focus it a bit:

For chip designers looking to compare nodes, will Intel's density metric be useful?

Are there two or three other key technical metrics needed to round out the picture? If so, what are they?

Thx

 

 

resistion
User Rank
Author
Re: Do we need a standard metric or metrics?
resistion   3/29/2017 10:59:12 PM
NO RATINGS
I think an RC-based metric would be useful.

rick merritt
User Rank
Author
Re: Do we need a standard metric or metrics?
rick merritt   3/30/2017 1:57:30 AM
NO RATINGS
@resistion:  Is there a specific formula or unit (like Intel's MTrans/mm2) you can give?

resistion
User Rank
Author
Re: Do we need a standard metric or metrics?
resistion   3/30/2017 2:03:21 AM
NO RATINGS
Might have to be something like ns/nm^2, if we use capacitance per unit length and resistance per unit length.

rick merritt
User Rank
Author
Re: Do we need a standard metric or metrics?
rick merritt   3/30/2017 11:41:40 AM
NO RATINGS
What does ns stand for?

resistion
User Rank
Author
Re: Do we need a standard metric or metrics?
resistion   3/30/2017 7:08:11 PM
NO RATINGS
Nanoseconds (RC time delay)

DataMuncher
User Rank
Author
Re: Do we need a standard metric or metrics?
DataMuncher   3/30/2017 9:14:09 PM
NO RATINGS
Intel's not doing themselves or anyone else a service with a lame FEOL measure of density. Let's get a metric that shows FEOL, BEOL and library tuning for the process.  The real measure is how well the process plus the libraries work together across a range of design types performance criteria.  Since were dreaming a bit here, I would like to see Intel libraries and process vs. ARM TSMC libraries compared on real synthesiszable designs like a large ARM core complex/SoC and a lage x86 core, targeted at the same clock rates. That's where you would have a chance to see which process was really the densest. 

rick merritt
User Rank
Author
Re: Do we need a standard metric or metrics?
rick merritt   3/31/2017 5:59:06 PM
NO RATINGS
@DataMuncher: Excellent idea!

Now can you (or some smart engineer here) take a crack at how that big idea could be boiled down into a clear formula or benchmark?

DataMuncher
User Rank
Author
Re: Do we need a standard metric or metrics?
DataMuncher   4/1/2017 9:37:19 PM
NO RATINGS
The top 10 logic semi companies, plus TSMC all do their own benchmarks as they do early pipeclenaing of their flows for new processes. Be assured that an Apple, QCOM, NVIDIA, MediaTek, HiSilicon, Spreadtrum, or LG will have tried a representative design in at least 2 of the the 3 leading edge processes/IP (TSMC, Samsung, Intel foundry), to pick the best one for their next major SoC.  Samsung, Intel, TSMC, and AMD have more commercial limitations on which processes they can try, but they do the same.

Gondalf
User Rank
Author
Re: Do we need a standard metric or metrics?
Gondalf   3/31/2017 7:29:22 PM
NO RATINGS
Not that Intel is interested much in; nearly all its revenue is done manufacturing its own products on the best process for their needs. Yes, a little of foundry could be useful but the bulk of Intel R&D is dedicated to internal products with a reasonable ASP . 

Good luck to TSMC to gain real money ion these upcoming crazy dense nodes, i have the suspect 16nm 14nm 12nm will are the sweet spot for many many years from now. Al this talk about 7nm 7nm+ 5nm is stunning, they say many things but nothing about some serious financial advantage. To shrink is easy but what about costs???

photonic
User Rank
Author
Re: Do we need a standard metric or metrics?
photonic   3/31/2017 9:21:17 PM
NO RATINGS
Spoken like someone who has given up. TSMC are total losers, right?

Gondalf
User Rank
Author
Re: Do we need a standard metric or metrics?
Gondalf   4/1/2017 8:52:04 AM
NO RATINGS
No TSMC it is not a loser, but we must to be realistic. The semiconductor market is not growing very fast at all and a day or another (early rather than later) Chinese fouundries will gain a lot of momentum. All this crazy TSMC effort in shrinking at any cost is a bad sign, they want to look better giving few in real performance increase and asking too much for less silicon area. Too bad for TSMC the design rules at finer nodes are very restrictive and the cost in designing a device is growing exponentially. So this push to 7nm or 5nm or 3nm (if ever happen), it will not translate in an industry shift to these nodes. Yes someone will go to, Apple Samsung and few others, still the bulk of industry will stay behind for economic reasons. This huge STOP in silicon shrinking "with economic advantage" will allow to new players to catch up quickly.

It is a bad situation but even with EUV we are at the end of silicon scaling with a clear economic advantage.

resistion
User Rank
Author
End in sight
resistion   3/31/2017 10:33:35 PM
NO RATINGS
So we are reaching the point where controlling the width of silicon features, e.g., fins, is already atomically quantized. The lattice constant of Si is 0.54 nm, so a 10-11 nm fin or nanowire is already atomically forced to suffer at least 10% CD variation from any process. 

Most Recent Comments
rick merritt
 
Tim R Johnson
 
ewertz
 
antedeluvian
 
ewertz
 
perl_geek
 
R_Colin_Johnson
 
perl_geek
 
R_Colin_Johnson
Most Recent Messages
11/19/2017
3:31:42 PM
Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed