Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/on-50th-anniversary-of-shockley-semiconductor-question-on-7nm-problems.10739/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021370
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

On 50th Anniversary of Shockley Semiconductor question on 7nm problems.

S

smeyer0028

Guest
I studied physics as a Stanford undergraduate back in
the late 1960s. I knew William Shockley from taking his Freshman
seminar. I noticed various recent commemorations of the founding
of Shockley Semiconductor. I knew a different Shockley than the
current mythology and think the book "Broken Genius: The Rise and
Fall of William Shockley" is factually inaccurate and unfair to the
real physicist William Shockley.

I do not understand semiconductor physics or process technology.
I wonder why there seem to be so many problems moving to 7nm
semiconductor fabrication. I even read things like "7nm transistor
drives are too weak." I wonder what the problem is. Is it related
to understanding of semiconductor physics or is it related to
materials science?

There is an interesting recent controversy in physics that
Shockley anticipated in the late 1960s involving hidden momentum.
The controversy is related to interpretation of quantum physics and
also to the properties a quantum computer will have if such computers
can be built. A materials engineer named M. Mansuripur claimed the
electrodynamics of hidden momentum contradicts special relativity.
A story appeared in Science ("Textbook Electrodynamics May
Contradict Relativity", News&Analysis section, Science Vol. 336(2012),
404, 2012). The claim was incorrect. There were numerous following
articles and papers showing Mansuripur was wrong. The interesting
part is that in the late 1960s Shockley anticipated the problem
(Shockley, W. and James, R. "Try Simple Cases" Discovery of "Hidden
Momentum" Forces on "Magnetic Currents." Phys. Rev. Lett. 18,
876, 1967). It is related to quantum physics because it is not
clear how to measure using a point magnetic dipole.
 
"I wonder why there seem to be so many problems moving to 7nm
semiconductor fabrication."

Maybe you should be wondering about why certain individuals keep insisting that there are GENERIC problems moving to 7nm? What is their agenda?

Some companies have made the transition just fine, some have not made the transition at all gracefully.
And those companies that have not performed well have a vested interest in creating a story that the transition is "impossibly difficult" rather than admitting that they did a terrible job of performing their version of the transition...
 
I wonder why there seem to be so many problems moving to 7nm
semiconductor fabrication. I even read things like "7nm transistor
drives are too weak." I wonder what the problem is. Is it related
to understanding of semiconductor physics or is it related to
materials science?

It may not be so much a technical barrier as one of development costs.

With rising development costs, there are fewer players moving on the next node (7nm at present). This has the effect of raising the barrier higher for the next node in development costs for those remaining players.

Actually with GF out, there is only one player left (TSMC) without negligible foundry market share (actually it has more than 55%). The other remaining player in advancing foundry nodes (Samsung) had less than 8% foundry market share in 2017.

Intel is in a unique position and seems to have most explicitly shown the symptoms of advancing node difficulties, since 14nm node. It may be argued from the Intel camp that their going from 22nm to 14nm is equivalent to foundry 28nm to 10nm, without going through a 16/14nm intermediate node. Likewise Intel's 10nm is a little more aggressive than foundry's 7nm. So it would seem they had self-imposed development difficulties.

One aggravating factor is the increased number of patterning steps, which has gotten worse since 16/14nm node. At the same time, EUV technology which provided hope for reducing those steps still has gaps for high volume manufacturing.
 
Interesting question.

It is difficult to compare nodes now that the naming has changed and what I call the Apple Effect. Back in the day nodes would be delayed if they did not meet performance or yield targets. With the yearly Apple cadence that is no longer the case and with FinFETs it seems we can name nodes whatever we want. Cost is a significant barrier for sure (see GF announcement) and you can blame that on Apple as well because they can write some VERY big checks that keep TSMC on a very aggressive process delivery schedule.

TSMC/Apple do what we used to call half nodes which used to be a node shrink. It can be expensive because only high volume customers will use it but TSMC has certainly mastered half nodes with Apple's help. Fabs and equipment are reused and the half node yield learning is critical if you are chasing Moore's Law.

TSMC and Apple did a 10nm node for the iPhone 8/X which was optimized for 7nm. TSMC will also "introduce" EUV at 7nm in preparation for 5nm EUV. Again, without Apple this would not be possible.

Samsung of course can keep up, remember they are the #1 semiconductor company. Intel could certainly keep pace if they had better management. GF, UMC, SMIC really never had a chance unfortunately.

So today it is TSMC and Samsung leading the way. Maybe the new Intel CEO will make a difference, but then again the last new Intel CEO made it worse so it could go either way.

Just my opinion of course.
 
Last edited:
I studied physics as a Stanford undergraduate back in
the late 1960s. I knew William Shockley from taking his Freshman
seminar. I noticed various recent commemorations of the founding
of Shockley Semiconductor. I knew a different Shockley than the
current mythology and think the book "Broken Genius: The Rise and
Fall of William Shockley" is factually inaccurate and unfair to the
real physicist William Shockley.

I do not understand semiconductor physics or process technology.
I wonder why there seem to be so many problems moving to 7nm
semiconductor fabrication. I even read things like "7nm transistor
drives are too weak." I wonder what the problem is. Is it related
to understanding of semiconductor physics or is it related to
materials science?

There is an interesting recent controversy in physics that
Shockley anticipated in the late 1960s involving hidden momentum.
The controversy is related to interpretation of quantum physics and
also to the properties a quantum computer will have if such computers
can be built. A materials engineer named M. Mansuripur claimed the
electrodynamics of hidden momentum contradicts special relativity.
A story appeared in Science ("Textbook Electrodynamics May
Contradict Relativity", News&Analysis section, Science Vol. 336(2012),
404, 2012). The claim was incorrect. There were numerous following
articles and papers showing Mansuripur was wrong. The interesting
part is that in the late 1960s Shockley anticipated the problem
(Shockley, W. and James, R. "Try Simple Cases" Discovery of "Hidden
Momentum" Forces on "Magnetic Currents." Phys. Rev. Lett. 18,
876, 1967). It is related to quantum physics because it is not
clear how to measure using a point magnetic dipole.
smeyer,

In physics it is also possible for an object to approach the speed of light, however we are well aware of the barriers to do that at a macro level. Likewise, for semiconductors as your try and reduce the feature size towards 0.0nm you run into quantum and atomic-scale barriers. We do not know how to fabricate single Atom semiconductor structures. In my lifetime there has been an incredible explosion in transistor complexity from one transistor per chip all of the way up to billions of transistors per chip, so now that progress is slowing way down because it simply isn't cost effective unless you are producing millions or billions of chips for a single application.
 
I do not understand semiconductor physics or process technology.
I wonder why there seem to be so many problems moving to 7nm
semiconductor fabrication... I wonder what the problem is. Is it related
to understanding of semiconductor physics or is it related to
materials science?

Ignoring the lithography issues for a minute: Our transistors are not as good as they "should" have been if you'd extrapolated the 1970-2006 trend.

For device physics reasons, since 130 nm, we haven't been able to reduce the supply voltage sufficiently, while being able to strongly switch the transistors on -- this is basically the "weak drive current" bit -- and off (we desire a low leakage current).

The strained-silicon mobility boost, FinFETs, SiGe channel materials etc. have been brought in to try and improve the MOSFET performance, but add complexity and cost.

The materials science challenge accompanies this. e.g. you have to be able to maintain desired strain during processing, and manage defects at Si SiGe interfaces.

Furthermore, variability generally gets worse moving to a new node, as a consequence of the atomic granularity of dopants.

(Then you have the materials challenges of trying to introduce Cobalt interconnect, and the required barrier layers etc.)

Moving to a new node is very technologically and economically demanding.
 
Back
Top