backtop


Print 30 comment(s) - last by lithographer.. on Sep 29 at 10:48 AM


Mark Bohr discusses 22nm logic tech  (Source: Intel)

The P1270 22nm process is expected to scale similarly to 32nm
The company is preparing to do battle in 2011 and beyond

There have been many details made available at the Intel Developer Forum about future Clarkdale desktop and Arrandale mobile chips. Yields are very good and defect density is steadily decreasing. They will be launched very soon as part of the 32nm Westmere family.

Intel also revealed some details about production of chips using the Sandy Bridge architecture in 2010. The company's "tick-tock" strategy of alternating annual process shrinks and new microarchitecture has been quite effective. The Westmere die shrink will be the next "tick", while Sandy Bridge will be the next "tock". After that will come the 22nm Ivy Bridge shrink in 2011, followed by the new Haswell microarchitecture in 2012.

Intel showed off the first 32nm SRAM test chips at the IDF in 2007, and this year they showed us their first 22nm SRAM test chips. With the 22nm process, Intel will be able to produce four times the number of chips per wafer of the 45nm process, thus making CPUs even cheaper once the process reaches mature yields.

DailyTech received not one, but two briefings on the 22nm process by Mark Bohr, an Intel Senior Fellow and Director of Process Architecture and Integration. He works in Intel's Logic Technology Development group located at the D1D development fab in Hillsboro, Oregon. His primary responsibility is to direct process development for Intel's advanced logic technologies.

Intel is the first in the semiconductor industry to demonstrate working 22nm circuits. Their 22nm shuttle chip contains not only SRAM circuitry, but also logic and mixed signal transistors such as phase lock loop (PLL) circuits. It is a 364 Mbit sized array consisting of over 2.9 billion transistors and built using Intel's third generation of High-K/Metal Gate technology. Shuttle chips are used to test different types of logic during the development phase, but these chips will have the same transistor and interconnect features as on 22nm CPUs in the Ivy Bridge family.

Being able to demonstrate working 22nm SRAMs is an important milestone towards being able to build working 22nm CPUs. There are currently two types of SRAM cells that have been built so far. There is a 0.092 square micron cell designed for high performance CPUs, and a 0.108 square micron cell for for low voltage applications such as Atom SoC designs.

Fab D1D is currently producing 32nm chips in the Westmere family, but there is also a large portion dedicated to producing Sandy Bridge and Ivy Bridge silicon. The lessons learned will be used to develop the P1270 high performance logic process for Ivy Bridge chips, but will also lead to the development of the P1271 SoC process for 22nm Atom chips. Intel plans to speed up the introduction of P1271 so that it will be available just three months after P1270.

Intel is planning to use 193nm immersion lithography for critical layers, and double patterning for the rest. The company is unlikely to use EUV (Extreme UltraViolet) tools even for the 15nm process in 2013, but may consider them  for use in the 11nm process that will be introduced in 2015.

Intel CEO Paul Otellini has repeatedly committed to expanding the company's sales in the SoC and embedded markets as he thinks that will be the next major drivers for Intel's growth. He thinks there is a market for billions of these devices, but they must be small and cheap in order to be commercially viable. In the future, we may end up with Atom chips being produced first on the leading edge, with mobility, server, and desktop designs coming later.




Comments     Threshold


This article is over a month old, voting and posting comments is disabled

Simply incredible...
By quiksilvr on 9/24/2009 1:41:32 PM , Rating: 5
It is amazing how far computer technology has come in just 4 short years. Four years ago we had Pentium 4 with 90nm and now we cut down to 32. And now Intel is talking about 15 and 11 nm? It just boggles the mind. How small can you make it?




RE: Simply incredible...
By amanojaku on 9/24/2009 1:48:42 PM , Rating: 1
quote:
It just boggles the mind. How small can you make it?
Short, cut down... I'm not comfortable with where you're going with this... ;-)


RE: Simply incredible...
By quiksilvr on 9/24/2009 5:40:23 PM , Rating: 3
I was expecting a penis joke response. Thank you for not disappointing me.


RE: Simply incredible...
By TSS on 9/24/2009 1:51:37 PM , Rating: 3
Last i heard, 4 nanometers untill quantum tunneling effects kick in (too hard).

So while this is a big technological achievement by any means, we'll still have to wait atleast another 2 decades before we might see a revolution, instead of another evolution.

And might i add, i find it very exciting that's well within my lifespan :P


RE: Simply incredible...
By randomposter on 9/24/2009 2:07:50 PM , Rating: 1
... unless the singularity arrives before then


RE: Simply incredible...
By Omega215D on 9/24/2009 10:47:32 PM , Rating: 2
I remember reading many years back about UltraViolet lithography once the current shrinking process begins to run out of steam. What ever happened to that? And then another kind of lithography was mentioned a couple of years ago but nothing much was said about it.


RE: Simply incredible...
By BigPeen on 9/25/2009 12:13:50 PM , Rating: 2
well 193nm IS ultraviolet. All ultraviolet refers to is the light source used for the photolithography steps. The shorter the wavelength of the light used, the higher the possible resolution. You run into all kinds of other problems of course (optics, what kind of photo-resists, etc.)


RE: Simply incredible...
By flipsu5 on 9/26/2009 4:46:53 AM , Rating: 2
Well said..I think they will continue the same wavelength even for 11 nm, and beyond. The alternative of electron beam and/or nanoimprint is also very interesting, but it's still much slower, and requires more direct precision. So they may find some niche uses.


RE: Simply incredible...
By Omega215D on 9/28/2009 8:40:58 AM , Rating: 2
Yeah, thanks for that info and I found the old issue of PCWorld from 2004 talking about Extreme UVL but as you mentioned all kinds of problems can and will exist.


RE: Simply incredible...
By lithographer on 9/29/2009 10:48:17 AM , Rating: 2
As Mark said himself: The company is unlikely to use EUV (Extreme UltraViolet) tools even for the 15nm process in 2013, but may consider them for use in the 11nm process that will be introduced in 2015.
Problems are the light sources are still too weak to make them cost effective i.e. in terms of wafers processed per hour per $ of depreciation, and they have no way of ensuring the masks are clean enough yet to make high yield devices.


RE: Simply incredible...
By JTKTR on 9/24/2009 4:55:52 PM , Rating: 3
I have an old Fortran textbook that says:

"if progress in the automotive industry had been as rapid as in computer technology since 1960, today's automobile would have an engine that is less than 0.1 inch in length, would get 120,000 miles to a gallon of gas, would have a top speed of 240,000 miles per hour, and would cost $4.00."

And mind you, this book was published in 1997


RE: Simply incredible...
By TGIM824 on 9/24/2009 6:36:29 PM , Rating: 2
That reasoning would be sound if we all weighed the same as electrons.

You will need a bigger engine to get my fat ass up to 240k miles per hour.

That's another reason Moore's Law applies to semiconductors, and not clothes, food, furniture, big screen TVs, oh and Vehicles.


RE: Simply incredible...
By MrPoletski on 9/25/2009 5:16:50 AM , Rating: 2
I wish it applied to my wages...


Pretty fast
By eddieroolz on 9/24/2009 1:43:11 PM , Rating: 2
Intel's going ahead pretty fast. I wonder if AMD will ever have time to catch up.

If the last two years have been any indication, it seems like AMD will be playing catch-up for a long time. It's a shame.




RE: Pretty fast
By BladeVenom on 9/24/2009 2:19:50 PM , Rating: 2
AMD has always been behind in die shrinks.


RE: Pretty fast
By eddieroolz on 9/24/2009 6:12:59 PM , Rating: 2
What I was trying to say did not concern just the die shrinks. It also includes the performance of AMD processors and how it's been three years since they were neck to neck with Intel.


RE: Pretty fast
By segerstein on 9/24/2009 7:48:33 PM , Rating: 2
AMD would be beating Intel, if they had working 22nm Phenoms ;-)

There's only so much one can do with a Gen+1 manufacturing process, both in terms of performance and power (increasingly important for mobile applications).

I want to run Windows 7 on my Atom smartphone.


RE: Pretty fast
By segerstein on 9/24/2009 7:51:57 PM , Rating: 2
Absent another NetBurst fiasco, I don't see how AMD can come back at the top of the game...


RE: Pretty fast
By MrPoletski on 9/25/2009 5:18:03 AM , Rating: 1
...with a big fat cheque to their RnD department and a serious fab upgrade...


RE: Pretty fast
By StevoLincolnite on 9/24/2009 2:24:51 PM , Rating: 2
As far as Manufacturing process is concerned, AMD has always been "Behind", ever since I can remember Intel has been at the forefront of manufacturing processes.


RE: Pretty fast
By VeauX on 9/24/2009 2:30:10 PM , Rating: 2
That's the reason why they as trying to go FAB Less. Teh creation of GlobalFoundries etc..


RE: Pretty fast
By Spind on 9/25/2009 4:42:39 AM , Rating: 2
IBM is the pioneer though!


Skip 32nm & go for 22nm direct
By crystal clear on 9/25/2009 1:33:26 PM , Rating: 1
Given the speed & abilities of Intel's R&D & a ever weak AMD lagging far behind, its time Intel's skips the 32nm & goes 22nm.

Given the time frame that Intel has provided for 22nm chips availablity,they can afford the luxury of skiping a node & moving to the next one directly.

Besides getting themselves a lead of 2 generations over the rest,they also save huge R&D cost plus those huge crippling retooling those fabs costs.

Its a calculated risk that will pay huge returns in the future.




RE: Skip 32nm & go for 22nm direct
By Voo on 9/25/2009 6:01:37 PM , Rating: 2
Chances are good that they need to figure out how to solve all the problems that appear at 32nm before they could produce something at 22nm, so having huge R&D costs without getting any revenue out of it and letting AMD get 32nm chips while they're still at 45nm sounds like a good plan?

Dunno I think one step after the other is a better strategy. If they needed a completly new technology at 22nm it would be something different though, but afaik that's not the case.


By crystal clear on 9/26/2009 4:31:28 AM , Rating: 2
quote:
so having huge R&D costs without getting any revenue out of it


Hey those 45nm parts of the likes i5, i7, Atom besides the old generation parts like the core2 etc etc bring huge revenues for Intel & profit margins for Intel & will continue to do so for the next 2 years.

Revenues to such an extent, that it gives Intel the luxury of a massive buy back of it shares even during recessionary times.

The current Intel parts or product range beats AMD in performance & price & will ensure future financial support for the huge R&D cost involved in 22nm projects.

The great recession has ensured that rapid upgrading of harware has turned into rapid slowdown for the next 2 years.

Gone are those 3 yr upgrade plans -now its take the maximum juice out of the existing hardware.

The attitude if it works then it (upgrading) can wait is the rule of the day for companies & mainstream consumers (retail level).

A weak AMD historically & even in the future will continue to be late in their delivery of their R&D to manufacturing.

Their prices of existing products leave NO profit margins,besides bring in revenues that ensure quarterly losses for years ahead.

Their cycle of losses dont seem to end & all talk/hopes about breaking even like in the past are "NOT to be believed"

The foundry company owned by Abu Dhabi ensures endless flow of working captial,but they lack the experience/tech know how that Intel has gained over the years of manufacturing.

Intel is implementing their own version of Asset lite program,by outsourcing portion of their manufacturing.

Companies like TSMC are now taking up Intel"s Atom production & plus other Intel parts.

All those great hopes put into the foundry company (Abu Dhabi) may turn into a great dissapointment,to delivery the goods in time.

All expections of AMD delivering 32nm parts,Intel style bag on time is fantasy.


RE: Skip 32nm & go for 22nm direct
By flipsu5 on 9/29/2009 6:08:44 AM , Rating: 2
At semicon west, Intel's Yan Borodovsky did mention that after 22 nm, they could use a pitch division method where they can indeed skip two nodes at a time at least lithographically.


Roadmaps are nice and all...
By stmok on 9/25/2009 3:45:05 PM , Rating: 2
...So when is that 4Ghz quad-core CPU coming? :)




RE: Roadmaps are nice and all...
By bighairycamel on 9/25/09, Rating: 0
By IvanAndreevich on 9/26/2009 1:35:59 PM , Rating: 2
Re-read his question and then read your answer. Silly question, and the answer doesn't even match it :)


why don't they
By Randomblame on 9/25/2009 6:38:03 PM , Rating: 2
put atom cpus on the outer edge of the silicon wafers, it seems to me it would reduce waste quite a bit




"Intel is investing heavily (think gazillions of dollars and bazillions of engineering man hours) in resources to create an Intel host controllers spec in order to speed time to market of the USB 3.0 technology." -- Intel blogger Nick Knupffer














botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki