Intel embarrassed by own report

[Yousuf Khan]

One of Intel's researchers published a whitepaper concluding that the
smaller chips get, the more errors they will be susceptible to. That
however does make the pursuit of Moore's Law questionable. Also makes
Intel's bragging to the press that it's months ahead of the competition
in moving to 65-nm not so enviable.

Intel back peddles on breaking chips theory - vnunet.com
http://www.vnunet.com/vnunet/news/2138064/intel-retracts-breaking-chips

Yousuf Khan

 

[David Schwartz]

One of Intel's researchers published a whitepaper concluding that the
smaller chips get, the more errors they will be susceptible to. That
however does make the pursuit of Moore's Law questionable. Also makes
Intel's bragging to the press that it's months ahead of the competition in
moving to 65-nm not so enviable.

I learned in college that as chips got smaller, they'd get faster to a
point. Then they would get slower, less reliable, and run hotter. My
professor doubted that chips would ever get below 125nm. He also showed me a
proof that 20,000 bps was over the limit possible over a conventional phone
line.

It wasn't that he was an idiot. It's that he couldn't anticipate future
discovery. Of course we never know how the next limit will be broken. If we
did, it wouldn't be a limit. But the limit breakers have always come. On
time.

DS

 

[George Macdonald]

I learned in college that as chips got smaller, they'd get faster to a
point. Then they would get slower, less reliable, and run hotter. My
professor doubted that chips would ever get below 125nm. He also showed me a
proof that 20,000 bps was over the limit possible over a conventional phone
line.

It wasn't that he was an idiot. It's that he couldn't anticipate future
discovery. Of course we never know how the next limit will be broken. If we
did, it wouldn't be a limit. But the limit breakers have always come. On
time.

And yet the DOD already has its own "flexible fab" to manufacture chips
which still work for them:
http://www.reed-electronics.com/electronicnews/article/CA608194.html and
"DMEA has recently analyzed a small sample of the latest semiconductors and
the robustness tests resulted in 100% failure..." Of course part of the
reasons for producing old chips is so they don't have to rewrite their
software but the failures are real. After the military, who's next?

 

[Del Cecchi]

I learned in college that as chips got smaller, they'd get faster to a
point. Then they would get slower, less reliable, and run hotter. My
professor doubted that chips would ever get below 125nm. He also showed me a
proof that 20,000 bps was over the limit possible over a conventional phone
line.

It wasn't that he was an idiot. It's that he couldn't anticipate future
discovery. Of course we never know how the next limit will be broken. If we
did, it wouldn't be a limit. But the limit breakers have always come. On
time.

DS

The limit on a phone line was defined by a guy named Shannon. I would
like to think the above professor had heard of him. Now the professor
might not have concieved of the advances that new coding techniques and
massive ammounts of cheap transistors have made possible but the limit
is there, defined by shannon.

del cecchi

 

[rbmyersusa]

And yet the DOD already has its own "flexible fab" to manufacture chips
which still work for them:
http://www.reed-electronics.com/electronicnews/article/CA608194.html and
"DMEA has recently analyzed a small sample of the latest semiconductors and
the robustness tests resulted in 100% failure..." Of course part of the
reasons for producing old chips is so they don't have to rewrite their
software but the failures are real. After the military, who's next?

Electromigation doesn't seem to be much on the radar--yet. What
happens when there are only a handful of dopant atoms in a gate? A few
dopant atoms migrate, and the electrical properties of the gate change
noticeably. If it's not an important effect, it will only be because
leakage has killed scaling before it can become important.

RM

 

[Del Cecchi]

Electromigation doesn't seem to be much on the radar--yet. What
happens when there are only a handful of dopant atoms in a gate? A few
dopant atoms migrate, and the electrical properties of the gate change
noticeably. If it's not an important effect, it will only be because
leakage has killed scaling before it can become important.

RM

Electromigration has been on the radar since more years than I want
think about. For a recent example, consider the transition from
Aluminium to Copper. Resistance decrease was nice. EM limit increase
was way nicer.

And did you notice that some brilliant outfit came out with a
statistical timing analysis program for timing chip designs?

del cecchi

 

[Robert Redelmeier]

The limit on a phone line was defined by a guy named Shannon.
I would like to think the above professor had heard of him.
Now the professor might not have concieved of the advances

The real problem is that "average" SNR vs MHz graphs are
not commonly available for most media, including various grades
of phone lines or interstellar transmission (SETI).

Shannon's Limit takes [usually missing] data to apply.
The only common case where the data is available is from
Cat5+ cable certifications.

-- Robert

 

[Del Cecchi]

The limit on a phone line was defined by a guy named Shannon.
I would like to think the above professor had heard of him.
Now the professor might not have concieved of the advances

The real problem is that "average" SNR vs MHz graphs are
not commonly available for most media, including various grades
of phone lines or interstellar transmission (SETI).

Shannon's Limit takes [usually missing] data to apply.
The only common case where the data is available is from
Cat5+ cable certifications.

-- Robert

Then those guys doing DSL must have hacked the phone company or been
shooting in the dark... Or been reading BSTJ, or maybe bought a spool
of phone wire and layed it out in the back yard and done some
measurements. Sigh.

del cecchi

 

[rbmyersusa]

And did you notice that some brilliant outfit came out with a
statistical timing analysis program for timing chip designs?

As you know, I try to keep track of everything under the sun, but I
missed this one. Can you suggest a link?

RM

 

[Bill Davidsen]

I learned in college that as chips got smaller, they'd get faster to a
point. Then they would get slower, less reliable, and run hotter. My
professor doubted that chips would ever get below 125nm. He also showed me a
proof that 20,000 bps was over the limit possible over a conventional phone
line.

It wasn't that he was an idiot. It's that he couldn't anticipate future
discovery. Of course we never know how the next limit will be broken. If we
did, it wouldn't be a limit. But the limit breakers have always come. On
time.

Actually I believe the real limit of small was defined back in a
DATAMATION article in the mid to late 60's. When traces drop below a
certain size, they stop conducting reliably. The real explanation lies
in a pile of quantum physics equations, I believe the layman's
explanation was "you drop an electron in one end and it may not come out
the other." Which is dreadfully inexact but easy to understand.

In any case the limit was some 15 (from memory) orders of magnitude
below the fabs of the day, and I don't think we're anywhere near that.

People are muttering that diamond substrate will be the solution, and
perhaps for heat it will, but for leakage?

 

[daytripper]

One of Intel's researchers published a whitepaper concluding that the
smaller chips get, the more errors they will be susceptible to. That
however does make the pursuit of Moore's Law questionable. Also makes
Intel's bragging to the press that it's months ahead of the competition
in moving to 65-nm not so enviable.

Intel back peddles on breaking chips theory - vnunet.com
http://www.vnunet.com/vnunet/news/2138064/intel-retracts-breaking-chips

Yousuf Khan

<shameless plug alert>

My latest baby arrived on the showroom floor...

http://www.stratus.com/news/2005/20050606.htm

Our VP immediately cobbled together a press release referencing the Intel
white papers (there's a related Intel document out there) explaining why the
[alleged] trend should lead 24/7 mission-critical apps to our front door...

/daytripper (You got failure rate issues? We got solutions. Bring $$$ ;-)

 

[Robert Redelmeier]

Then those guys doing DSL must have hacked the phone company
or been shooting in the dark... Or been reading BSTJ, or maybe
bought a spool of phone wire and layed it out in the back yard
and done some measurements. Sigh.

They probably did all of this and more (field measurements of local
loops in the MHz range to understand impedence discontinuities).

-- Robert

 

[Hank Oredson]

The real problem is that "average" SNR vs MHz graphs are
not commonly available for most media, including various grades
of phone lines or interstellar transmission (SETI).

For phone lines see BSTJ, lots info ...

Shannon's Limit takes [usually missing] data to apply.
The only common case where the data is available is from
Cat5+ cable certifications.

--

... Hank

http://home.earthlink.net/~horedson
http://home.earthlink.net/~w0rli

 

[keith]

As you know, I try to keep track of everything under the sun, but I
missed this one. Can you suggest a link?

Web search tools are wunnerful things:

http://www-03.ibm.com/technology/news/2005/060305statisticaltiming.shtml

 

[keith]

They probably did all of this and more (field measurements of local
loops in the MHz range to understand impedence discontinuities).

Discontinuities? PLease. The fact is that the local loop has becoem far
quieter than in years past (and likely to get even more so as people ditch
wires). The gain isn't because of some new magic, but because the S/N
ratio is better than what was expected. ...not to mention that the "last
mile" isn't twenty miles these days.

 

[Robert Redelmeier]

For phone lines see BSTJ, lots info ...

In the MHz for subscriber loop? Maybe for inter-office
trunking before it was mostly replaced by fiber.

-- Robert

 

[Robert Redelmeier]

Discontinuities? PLease. The fact is that the local loop
has becoem far quieter than in years past

New or rewired loops, yes. I'm not sure how old loops
magically improve

But there still are discontinuities at splices. Maybe
not much with the newer ones, but every little bit hurts,
especially at higher frequencies.

(and likely to get even more so as people ditch wires).

I take that to mean bury. Yes. I worry a little the
aerial messenger could form a notch filter much like that
demo of uniform stapling coax ruining TV. Of course nothing
that would affect voice freqs.

The gain isn't because of some new magic, but because the
S/N ratio is better than what was expected. ...not to
mention that the "last mile" isn't twenty miles these days.

Agreed. RTs are installed more often, but many city dwellers
are on shorter but older loops.

-- Robert

 

[George Macdonald]

Electromigation doesn't seem to be much on the radar--yet. What
happens when there are only a handful of dopant atoms in a gate? A few
dopant atoms migrate, and the electrical properties of the gate change
noticeably. If it's not an important effect, it will only be because
leakage has killed scaling before it can become important.

I'd sure like to see Mr. Gaugler's "photographic evidence" before betting
the farm. Let's hope it gets published on-line.

 

[Del Cecchi]

As you know, I try to keep track of everything under the sun, but I
missed this one. Can you suggest a link?

RM

Why, certainly.

http://www.ibm.com/investor/news/jun-2005/03-06-05-2.phtml

 

[keith]

New or rewired loops, yes. I'm not sure how old loops
magically improve

Many old loops were already fairly quiet. Many have been replaced. The
ones that aren't quiet don't get a speed bump. Too bad. I had two line a
few years back. The original line wouldn't go over 26K, while the newer
one was reliably 53K.

But there still are discontinuities at splices. Maybe not much with the
newer ones, but every little bit hurts, especially at higher
frequencies.

Any discontinuity at a splice is peanuts at these frequencies. That's
like the people replacing their internal house wiring with cat-5,
expecting to see better perfromance.

I take that to mean bury. Yes. I worry a little the aerial messenger
could form a notch filter much like that demo of uniform stapling coax
ruining TV. Of course nothing that would affect voice freqs.

No, I meant get rid of their land-lines completely.

Agreed. RTs are installed more often, but many city dwellers are on
shorter but older loops.

Most of those loops are already pretty quiet. A friend's house is one of
the original (built between 1800 and 1803) structures in the village.
The loop isn't quite that old though. ;-) He has no problem at 53K. Old
wire isn't noiser than new wire (assuming twisted pair, not the individual
wires on the tree).