IBM and Georgia Institute of Technology are experimenting with silicon-
germanium, it is said here:
I find this interesting having just attended NANOG 37 where some
manufacturers of network devices told us in a panel that network
heat problems weren't going away unless there's a 'next big thing'
in manufacturing process.
Is this it?
Corrolary: If our routers are made of silicon-germanium, would the
CLI only operate in Deutsch?
IBM and Georgia Institute of Technology are experimenting
with silicon-
germanium, it is said here:I find this interesting having just attended NANOG 37 where some
manufacturers of network devices told us in a panel that network
heat problems weren't going away unless there's a 'next big thing'
in manufacturing process.Is this it?
Sure doesn't sound like it. In fact, it sound like they're pushing to a
high frequency regardless of the power and thermal consequences.
It also sounds like it's a single transistor. It takes a few of them to
make a router. 
I also suspsect that the community is not ready to transition to
liquid-cooled systems.
Tony
David W. Hankins wrote:
IBM and Georgia Institute of Technology are experimenting with silicon-
germanium, it is said here:I find this interesting having just attended NANOG 37 where some
manufacturers of network devices told us in a panel that network
heat problems weren't going away unless there's a 'next big thing'
in manufacturing process.Is this it?
Corrolary: If our routers are made of silicon-germanium, would the
CLI only operate in Deutsch?
Jawoll, es wuerde 
I remember my old radio days. My audion and diode receivers never
would work with silicon only with germanium diodes and transistors.
The difference is the voltage threshold where the device would start
conducting. That is 200 mV for germanium but 800 mV for silicon.
Devices running with silicon and 2.4 volts will go down to 600 mV. That
means power consumtion will drop to 1/4. The real thing is a bit more
complex but for a guesstimation ...
Cheers
Peter and Karin
IBM and Georgia Institute of Technology are experimenting with silicon-
germanium, it is said here:I find this interesting having just attended NANOG 37 where some
manufacturers of network devices told us in a panel that network
heat problems weren't going away unless there's a 'next big thing'
in manufacturing process.Is this it?
Nope, all this says is that with sufficient cooling you can go faster. What we need is going faster with less cooling.
W
Once upon a time, Warren Kumari <warren@kumari.net> said:
Nope, all this says is that with sufficient cooling you can go
faster. What we need is going faster with less cooling.
Read the article, not the headline. They got 350GHz at room
temperature (which is a lot more interesting than 500GHz a few degrees
above absolute zero).
Sure doesn't sound like it. In fact, it sound like they're pushing to a
high frequency regardless of the power and thermal consequences.
I thought their 500 Ghz number was just for rediculous press teasing,
like the people who use lHe to push AMD chips to ~10 Ghz.
The 350 Ghz 'at room temperature' insinuation is the most interesting
to me.
It also sounds like it's a single transistor. It takes a few of them to
make a router.
I haven't seen any evidence to support or contradict this, so I'll
take your word for it...
A single-transistor test on a single chip would be both ludicrous
and incomparable.
I also suspsect that the community is not ready to transition to
liquid-cooled systems.
I rather assumed 'at room temperature' implied a standard heat sink
and fan.
Perhaps there's not enough information in that article to draw a
conclusion from.
The point that I was trying to make (admittedly REALLY badly) was that this is not the 'next big thing' .
Did you read anything more than just that article?
IBMs press release is here:
http://www-03.ibm.com/technology/news/2006/0620_frozen_chip.html
and they have a video here:
http://www-03.ibm.com/technology/ets/capabilities/multimedia_tour/frozen_chip_wmv.html
This is not a new technology (IBM shipped their 100 millionth SiGe chip in around 2002 and if you look at the SONET chipset on an OC48 or greater interface chances are its SiGe), but the speed in cheap material is (Feng & Hafez achieved >600Ghz in indium doped) -- this is primarily just a bragging right though. It requires liquid helium temperatures, something that is not practical in the near term, and requires a LOT of power to achieve.
Once upon a time, Warren Kumari <warren@kumari.net> said:
Nope, all this says is that with sufficient cooling you can go
faster. What we need is going faster with less cooling.Read the article, not the headline. They got 350GHz at room
temperature (which is a lot more interesting than 500GHz a few degrees
above absolute zero).
Yes -- the previous silicon based speed record *at room temp* was 375Ghz.
Warren
> I also suspsect that the community is not ready to transition to
> liquid-cooled systems.I rather assumed 'at room temperature' implied a standard heat sink
and fan.Perhaps there's not enough information in that article to draw a
conclusion from.
There are a few bits that folks should understand: first, SiGe has been
around for awhile. It's not new. It's used when higher frequencies are
necessary, such as when building a 40Ghz modulator for an OC-768c
interface.
SiGe is more expensive, less thermally efficient, and less dense than
'standard' CMOS. So it's already headed the wrong way for most of our
applications.
Second, you should know that there are lots of folks who really are
experimenting with a single transistor. This may sound ludicrous, but
the thought here is that process improvements will eventually scale.
Thus, the conclusion that I'm leaping to is that this room temperature
transistor at 350GHz really is at room temp, but may require something
like a muffin fan all by itself. Obviously to scale that to a few
hundred million transistors in a router, you then need a few hundred
million little fans.
The breakthrough that we're looking for is a high speed, high density,
low power transistor that can be commercially scaled with good yield.
Not there quite yet.
Tony
In comparison to early-80s ECL, how do you think the scaling curve might match? I haven't found much material yet that shows any realistic projections for speed and yield ramp up for the new stuff.
--lyndon