Design Con 2015
Breaking News
News & Analysis

IBM Breaks EUV Throughput Record

7/30/2014 01:20 PM EDT
22 comments
NO RATINGS
More Related Links
View Comments: Threaded | Newest First | Oldest First
rick merritt
User Rank
Author
Intel has 10nm w/o EUV
rick merritt   7/30/2014 1:48:11 PM
NO RATINGS
This just in from Intel fellow Mark Bohr:

"Intel has a way to achieve good density and good cost-per-transistor on our 10 nm technology without the use of EUV. 

We are continuing to explore EUV and non-EUV options for our 7nm technology."

wilber_xbox
User Rank
Manager
Re: Intel has 10nm w/o EUV
wilber_xbox   7/30/2014 2:20:10 PM
NO RATINGS
Rick, is it using the existing methods such as unidirectional patterns, mask optimization and breaking the chip patterns into 2-3 masks etc or something more innovative.

resistion
User Rank
CEO
Re: Intel has 10nm w/o EUV
resistion   7/30/2014 4:41:19 PM
NO RATINGS
They use "pitch quartering" for 15-22 nm half-pitch lines. There can be up to four additional "cut" exposures.

wilber_xbox
User Rank
Manager
Re: Intel has 10nm w/o EUV
wilber_xbox   7/31/2014 4:29:16 AM
NO RATINGS
introducting more masks and cut masks should drive the price per die so economically EUV will be much better IMO.

resistion
User Rank
CEO
Re: Intel has 10nm w/o EUV
resistion   7/31/2014 5:03:46 AM
NO RATINGS
At 10 nm EUV doesn't have as easy opportunity as at 100 nm. The required upgrades (new mask, pellicle, high NA, resist) are weighed against multiple immersion exposures. So far not too many required overall. Beyond 10 nm has other uncertainties as well, including multiple EUV exposures.

rick merritt
User Rank
Author
Re: Intel has 10nm w/o EUV
rick merritt   7/30/2014 11:52:53 PM
NO RATINGS
@Wilbur: I did not know to ask but I will request an IBM engineer jump on and answer.

resistion
User Rank
CEO
Re: Intel has 10nm w/o EUV
resistion   7/31/2014 1:12:19 AM
NO RATINGS
I thought he was referring to Intel's 10 nm, otherwise the EUV requiring these would defeat the purpose.

wilber_xbox
User Rank
Manager
good news
wilber_xbox   7/30/2014 2:16:01 PM
NO RATINGS
This will for sure put some cheers on the face of semiconductor manufacturing industry as this will ultimately result in lower cost and complexicity. However, ASML has to go a long way to give more than 100wph that will make EUV truely production ready.

resistion
User Rank
CEO
Re: good news
resistion   7/30/2014 2:31:20 PM
NO RATINGS
Besides light source power, EUV already has other resolution-limiting issues like shot noise and multilayer angular bandwidth. And it's still nowhere near 100 WPH.

rick merritt
User Rank
Author
Re: good news
rick merritt   7/30/2014 11:55:08 PM
NO RATINGS
@resistion: 34 wafers/hour to be specific

resistion
User Rank
CEO
Re: good news
resistion   7/31/2014 1:19:44 AM
NO RATINGS
34 WPH is still below the previous 70 WPH target. 20 mJ/cm2 dose is also quite noisy below 20 nm.

TanjB
User Rank
Rookie
Re: good news
TanjB   8/1/2014 1:51:50 PM
NO RATINGS
@resistion: The EUV photon energy is around 92eV or 15 attoJ.  This implies about 5,000 photons to provide the 80 pJ necessary to expose a 20nm square.  That should not have a shot noise problem.  8 sigma would be +-12%.

Perhaps you are thinking of e-beam, where the beams are typically > 5kV so the number of electrons is a real problem for shot noise?

resistion
User Rank
CEO
Re: good news
resistion   8/1/2014 3:18:54 PM
NO RATINGS
6 sigma is 5-10% depending on feature size, that's a big impact on CD.

krisi
User Rank
CEO
Re: good news
krisi   8/1/2014 3:23:41 PM
NO RATINGS
5000 photons sounds awfully small

TanjB
User Rank
Rookie
Re: good news
TanjB   8/1/2014 6:13:15 PM
NO RATINGS
@krisi: 5,000 photons is quite a good enough number for statistical process control.  When it gets down to 500 or less things become really tricky for high reliability over huge numbers of elements which must be exposed.  This is because the delivery is a stochastic process, there is no possible feedback loop to correct it.

It is the same reason why your digicam shows noise in low light, when it may be depending on less than 100 photons per pixel.

@resistion: I cited 8 sigma, not 6, since that is the level of reliability needed to make chips with a billion elements.  So long as 88% of 5,000 is enough energy delivered to develop the resist, that is good.  Of course there is also statistical variation in the resist chemistry, and some variation in the illumination.  When a 20mJ/cm2 value is quoted that will normally allow for some variation in characterizing the resist.  If not, then add 10% to the input power to be sure to meet the threshold.

 

krisi
User Rank
CEO
Re: good news
krisi   8/1/2014 6:16:28 PM
NO RATINGS
thank you @TanjB...I agree, I had 10,000 number in my mind, which the number of photon required in my field (medial imaging) to create statistically valid image...but I guess 5,000 should do it

resistion
User Rank
CEO
Re: good news
resistion   8/1/2014 6:29:57 PM
NO RATINGS
The appropriate statistics for a billion features is +/-6sigma. And it shouldn't be 5-10% of the nominal dose. ASML data had shown this. So this concern definitely exists at 20 nm.

rmaire
User Rank
Rookie
ASML, IBM & EUV
rmaire   8/1/2014 12:43:27 PM
NO RATINGS
Imagine that a few years ago HP announced it was developing a new ultra high resolution printer capable of printing 1500 pages an hour with resolution of 10K dots per inch resolution. 
Imagine that the printer has been delayed and plagued by technology development issues, with slow progress being made to the point where customers wonder if it will ever work.
 
Then one day HP and a major customer announce that they successfully ran 637 sheets of paper through the printer in an hour as proof that it was well on its way to being a working printer. 
 
However HP and the customer conveniently leave out the fact in their announcement that the the paper went through the new printer without anything ever being printed on the paper and the paper came out as blank as it went in. In fact some of the blank paper in the output tray was recycled back to the input tray to be run through the printer again coming out blank again.
 
No toner was used and nothing showed up on the paper yet HP claimed that the new printer achieved a new "milestone",  "watershed" break through rate of 637 pages an hour.
 
Would you as an investor or analyst view this as an honest and fair representation of progress on the part of HP and its customer or would you find this misleading ?
 
Well,  that is what happened with IBM and ASML.....


rick merritt
User Rank
Author
Update story TK in about an hour
rick merritt   8/1/2014 1:54:48 PM
NO RATINGS
Stay tuned for an updated story with more of Robert Maire's criticisms of the test and IBM's responses to the criticisms.

Great discussion!

Dan Corliss
User Rank
Rookie
Re: Update story TK in about an hour
Dan Corliss   8/1/2014 4:45:36 PM
NO RATINGS
To help folks who do not understand the value of the results identified in Rick's EETimes article, I submit the following:

One of the primary challenges holding back the insertion of EUV is the source's power and reliability.  Imaging performance is not a primary concern as it is well characterized, robust and operational on IBM's NXE3300.  The upgrade to our EUV source was intended to improve its power level and reliability.  The 24 hour performance test was intended to stress those two parameters.  It was never the intent of the test to generate 637 wafer exposures.  That result was a by-product of the source operating correctly, at the increased power and reliably during the test period.  Putting resist on the wafers would have had no value whatsoever, other than to test how well our rework process was working.  At the EUV Center of Excellence in Albany, we are focused on understanding the "fundamentals of why", so that robust solutions can be developed for high volume manufacturing.  The endurance test that we performed was expected to generate a number of performance anomalies.  The anomalies would have provided us with learning opportunities to identify root causes and develop subsequent improvements to the technology.  It was an unintended output that our source performed so well.  However, the result did provide the first data point in the industry, that demonstrates that the current source technology does have the capability to achieve near term performance goals (wafer exposures per day).  Since this result was significantly better than previously reported performance, IBM decided to share this with the semiconductor industry so that all could understand the significance and adjust their EUV activities accordingly.  For IBM and our Alliance partners, this secured the EUV capability to support our 7nm technology node development.

This is my last response on this topic.  It is time for us to get back to work on the maturing of EUV technology.

Respectfully,
Dan Corliss

krisi
User Rank
CEO
Re: Update story TK in about an hour
krisi   8/1/2014 5:27:18 PM
NO RATINGS
thank you Dan for the elaborate explanation...would you be interested in presenting this technology at the emergin technologies symposium in Vancouver in 2015? preliminary program at www.cmosetr.com, kris.iniewski@gmail.com

rick merritt
User Rank
Author
Update posted
rick merritt   8/1/2014 5:40:13 PM
NO RATINGS
An updated story is now posted here: http://www.eetimes.com/document.asp?doc_id=1323353

Thanks all --Dan Corliss included -- for chiming in here.

Please join the discussion there

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