Design Con 2015
Breaking News
News & Analysis

Intel Outlines 14nm, Broadwell

8/11/2014 03:36 PM EDT
18 comments
Page 1 / 4 Next >
More Related Links
View Comments: Threaded | Newest First | Oldest First
3D Guy
User Rank
Manager
Moore's law
3D Guy   8/11/2014 6:22:27 PM
NO RATINGS
Congratulations to the Intel team on a phenomenal engineering achievement. Keeping the cost per transistor going down by doing higher aspect ratio fins, boosting drive current and lowering transistor area needed is great.

I wonder, though, whether Intel can really boast that Moore's Law is alive and well... the yield challenge of high aspect ratio fins cost a year of delays (thereby the cost per transistor went down by 2x after 3 years instead of 2)... it also caused loss of face for Intel with major customers like Apple. 

rick merritt
User Rank
Author
Re: Moore's law
rick merritt   8/11/2014 9:34:26 PM
@3D Guy: As Moore himeself said, before scaling stops it will slow down

 

prousseau
User Rank
Rookie
Re: Moore's law
prousseau   8/21/2014 6:47:20 PM
NO RATINGS
I'm curious on your statement Intel use SADP, it this a fact they shared after the webcast, or your guess; I did not hear this in webcast.  SADP is not free

buprestid
User Rank
Freelancer
SRAM size
buprestid   8/11/2014 8:21:08 PM
Not bad for the SRAM size.

Intel : 0.0588 sq. um

Samsung : 0.064 sq. um ISSCC 2014

TSMC : 0.07 sq. um IEDM 2013

 

GroovyGeek
User Rank
CEO
Re: SRAM size
GroovyGeek   8/11/2014 9:37:14 PM
Buprestid You do realize you are comparing a shipping product to a conference paper presentation. Real SRAM cell sizes are substantially larger than what is advertised at conferences. I guess in a few years wwhen they get aroundto sshipping 14nm we will know.

buprestid
User Rank
Freelancer
Re: SRAM size
buprestid   8/12/2014 3:21:26 AM
NO RATINGS
For these process parameters is there anyway to compare actual numbers? Just some conference papers and some slides here and there.


Intel comes out with the Gate Pitch x Metal Pitch in November 2013 that says they are much better.

TSMC rebuts that claim saying they are equal density at 16nm node.

Intel reiterates again from the chart here they have a denser far cheaper process.

GroovyGeek
User Rank
CEO
Re: SRAM size
GroovyGeek   8/12/2014 3:35:35 AM
Yes head to head comparisons are tough, but in this case Intel is not showing a PowerPoint but an actual product. TSMC will not ship anything they define as 14nm for at least a year, and in volume even later. Historically their SHIPPING SRAM sizes have been 10-20% larger than what has been advertised in papers, probably to meet SNM and other requirements.

Gondalf
User Rank
Freelancer
Re: SRAM size
Gondalf   8/12/2014 8:57:03 AM
NO RATINGS
To be fair TSMC has never said they are equal density at 16nm node (they said will catch up at the 10nm node).

Edited.

Asking to other engineers: TSMC and Samsung cells are 1T fin, Intel cell is 2T fins so the comparison is not fair. 

Very likely both TSMC and Samsung will utilize a larger 2T cell for faster phone SOCs, still it is only a my idea. Sure the 1T cofiguaration is a limiting factor in abailable clock speed ad low power consumption.

We'll see Intel SOC process, it is unlikely an even smaller cell footprint for certain applications with 1T layout, like is happened on 22nm with the 1T 0.092um2 cell.

resistion
User Rank
CEO
scalable to 10 nm
resistion   8/11/2014 9:03:39 PM
NO RATINGS
The same SADP techniques can scale to 10 nm, but tighter gate scaling looks dangerous.

resistion
User Rank
CEO
Re: scalable to 10 nm
resistion   8/15/2014 10:59:33 PM
NO RATINGS
Anyone can estimate from the SRAM picture that Lg is 24-27 nm at 14 nm node. It's already leveling off.

alex_m1
User Rank
CEO
cost
alex_m1   8/11/2014 11:12:35 PM
NO RATINGS
I wonder how can Intel reduce transistor costs while the rest of the industry cannot ? what's different ?

resistion
User Rank
CEO
Re: cost
resistion   8/12/2014 9:31:24 AM
NO RATINGS
I don't know if it's that Intel uses SADP while foundries use LELE. SADP would restrict layouts strongly, which fits Intel's manufacturing style (they used to be a memory company). Another is they they staged their big technology transitions over many generations (45 nm high-k, 32 nm immersion, 22 nm FinFET, 14 nm double patterning). Instead of one big barrier at 2X nm.

Simon7382
User Rank
Manager
Re: cost
Simon7382   8/12/2014 8:29:52 PM
NO RATINGS
Intell is better, that simple. They are ahead of the rest of the indostry by at least 2 generations.

resistion
User Rank
CEO
Re: cost
resistion   8/12/2014 9:12:29 PM
NO RATINGS
It would be interesting to see implications for its foundry business. It may or may not translate directly.

3D Guy
User Rank
Manager
Intel has a lead?
3D Guy   8/12/2014 8:44:26 AM
NO RATINGS
A question that always comes up for me when I see Mark Bohr boasting about Intel's process technology load over TSMC:

Intel may show TSMC having higher gate pitch x metal pitch. But I think the true comparison would be cost per transistor. TSMC owns significantly cheaper fabs than Intel, thanks to much better government incentives than exist in Taiwan (vs. the US). My experience is that the cost difference can be ~30% or sometimes even more. I wonder if Intel really has a significant lead in cost per transistor?

escher
User Rank
Manager
Re: Intel has a lead?
escher   8/12/2014 11:32:18 PM
NO RATINGS
@3D guy - Intel gets plenty of incentives from the state of Oregon, Ireland and Israel. Not so sure about Arizona and New Mexico, but I imagine there must be a big carrot to keep them in these 2 dry states where water is at a premium.

HJ88
User Rank
Freelancer
Intel's Achievements
HJ88   8/12/2014 11:19:55 AM
The technical achievements by Intel at 14nm are excellent. Achieving 0.53 area scaling compared to 22nm requires incredibly complex processing technology for control of fin height, pitch, and passivation coverage.

The next step of manufacturing 50K or 80K wafers per month with high yields and high reliability is another order of magnitude in complexity. Let us hope for the electronics industry that Intel will achieve its goal of reaching high volume production in Q1 or Q2/2015.

The other vendors of FIN-based products have similar challenges to Intel, and it is realistic to expect them to have a similar time line between when the process was initially stable to when high volume of cost competitive products that also have high reliability is achieved.

Intel's progress to date is an outstanding achievement by the company and its ecosystem partners.

double-o-nothing
User Rank
Rookie
Multiple patterning cheapest after all
double-o-nothing   8/13/2014 9:47:00 AM
NO RATINGS
14 nm must have lots of double patterning and triple patterning, much more so than 22 nm. It must mean ASML and the analysts have overestimated the cost of the steps being repeated.

Radio
LATEST ARCHIVED BROADCAST
EE Times Senior Technical Editor Martin Rowe will interview EMC engineer Kenneth Wyatt.
Top Comments of the Week
Like Us on Facebook

Datasheets.com Parts Search

185 million searchable parts
(please enter a part number or hit search to begin)
EE Times on Twitter
EE Times Twitter Feed
Flash Poll