Frozen Chip from IBM hits 500 GHz

sideshow2004 writes "EETimes is reporting this morning that IBM and Georiga Tech have demonstrated a 500 GHz Silicon-germanium (SiGe) chip, operating at 4.5 Kelvins. The 'frozen chip' was fabricated by IBM on 200mm wafers, and, at room temperature, the circuits operated at approximately 350 GHz."

I RTFA..

[irc.goatse.cx troll]

"By comparison, 500 GHz is more than 250 times faster than today's cell phones, which typically operate at approximately 2 GHz, according to the organizations."

I think that speaks for itself.

THAT WASN'T THE POINT

[technoextreme]

Arggg read the article they said they wanted to test the theoretical limits of these chips. They know speed increases with temperature. They wanted to know how much.

Re:In other news...

[xerxesdaphat]

You do know Moore's Law relates to the number of transistors on a chip, and doesn't have anything to do with clock speed, right?

Re:THAT WASN'T THE POINT

[sco08y]

They know speed increases with temperature.

Don't you mean "decreases"?

Re:1.2mm per cycle

[ThosLives]

*blinks twice to make sure I really read that math correctly*

:
:
:
*sighs in dismay*

Re:I RTFA..

[Anonymous Coward]

Well something in your cell phone must be producing the 2Ghz! i seem to recall a simple NOT gate went at around 5Ghz last time i timed it. Im guessing the chip is just a couple of gates.

Re:1.2mm per cycle

[imbaczek]

-1 All Wrong

Re:Why?

[Serious Simon]

These are not microprocessors, and the achievement is not the amount of computing power you can get from them but the extremely high frequency of the signal they can generate. And that is not something you can increase by adding more chips!

10GHz Microwave?

[Kadin2048]

That's a pretty odd microwave then, since most of them operate at 2.45 GHz, which is chosen because of the way it causes liquid water molecules to vibrate. See this article [lsbu.ac.uk], particularly the graphs showing dielectric temperature as a function of frequency. It's pretty clear that a 10GHz microwave oven would be a lot less efficient at heating water than a conventional 2.45 GHz one, although I suppose you could choose a multiple of 2.45GHz and probably still have a functional product.

Overall, unless your goal was to build a miniature microwave (a 21st century E-Z Bake Oven?), I don't know why you'd want to use 10GHz instead of 2.4Ghz ones. The tolerances of parts in the magnetron and waveguide would have to be much tighter, I think, and this would almost certainly cause it to be more expensive.

Re:cell phones?

[ivan256]

Didn't you ever think that if you had a digital signal entering your cell phone at 2.4 Ghz, you'd need a transistor in there that could switch at least that fast? You realize that there are other types of chips than microprocessors, right?

Yes you are

[Craig Ringer]

First, mobile phones do have extremely high frequency chips in them. They have to in order to recieve and process the high frequency signals they deal with. Those high frequency chips are a fairly large part of their power draw, too - yet their draw is *tiny* compared to even the simplest CPU of that clock. Remember that clock speed means very little without a consideration of the number of transistors on the chip, energy leakage rates, and lots more I know nothing about.

You're making the erroneous equation that "chip" == CPU, which is far from the case. A phone's CPU may be clocked much lower. Even if it's integrated with the RF chip (I'm not sure this is ever done, is it?), the RF processing parts will be clock-multiplied or the CPU parts will be clock-divided to ensure sensible running frequencies.

I think you'll also find that, contrary to the assumptions made by most posters here on /., this chip is a very fast very simple unit - not a large microprocessor. I'd guess they're looking into ultra-high-speed signal processing (hence the mobile phone analogy) rather than computer CPUs here.

Re:cell phones?

[codemaster2b]

No, what you need is a signal generator that produces a 2.401 GHz signal and a mixer to produce a beat frequency. Then, you process that much slower signal. You don't work with a 2.4 GHz signal.

Two speeds: Fast enough and not fast enough

[hlh_nospam]

"By comparison, 500 GHz is more than 250 times faster than today's cell phones, which typically operate at approximately 2 GHz, according to the organizations."

And in other news, apples and oranges usually taste different.

The only question about computer speed that is important is, "Is it fast enough?" Of course, "fast enough" may change over time, and anytime you come up with a faster processor, some company like Microsoft will succeed in loading it down with bloatware. But I've got a customer who runs his company on software that I wrote for him 15 years ago, and the only reason he ever upgrades his hardware is because something breaks that is no longer available. Otherwise, the 8MHz 286 system would have been perfectly adequate.

Re:The tempurature at which books freeze

[aricusmaximus]

You clearly have not watched enough Star Trek.

Go rent old Trek seasons 1 & 2 and pay particular attention to this episode [startrek.com].

Re:Article was in EE Times!

[Enigma_Man]

Not really, because an EE would know that it's not just the RF output on a cellphone that works at 2.4 GHz, but also the signal processing unit. There is a digital system in the phone that natively controls the signal, rather than using older analog techniques. The general-purpose CPU for playing crappy java games and displaying inane text messages from your friends runs at something much lower than that, of course.

-Jesse

This doesn't mean 500 GHz CPU's

[RWalz]

I just wanted to point that out, I think some posters are thinking about it incorrectly: "The 500 GHz mark was the goal when Feng and UI colleagues received a $2.1 million, five-year grant for the project from the Defense Advanced Research Projects Agency in October. In contrast, the transistors inside the central chip of a powerful personal computer run at around 50 or 100 GHz, Feng said. The fastest that such a chip runs as a package is currently around 3 GHz." http://www.news-gazette.com/news/local/2003/01/24/ fastest_transistor_made_at_ui/ [news-gazette.com] In addition, University of Illinois broke 600 Ghz last year. http://www.physorg.com/news3662.html [physorg.com] "The speeds quoted in this article are maximum rated *switching* speeds of a single transistor. Synchronous logic designs of the type found in microprocessors involve synchronous cells (known as flip-flops) and asynchronous gates providing boolean functions on the signals passing between flip-flops. The maximum rated frequency of any design is limited by the slowest path between flip-flops and this is what the clock signal will be set at. As the paths between the clocked flip-flops are typically anywhere between 2 and 10 logic cells deep and with each one comprising 10's of transistors (usually in complementary configuration to aid switching speed), the overall figure for an ASIC design such as a uProcessor would be at least 2-4 times slower than the maximum transistor switching speed (it's not quite cumulative, because as one transistor starts switching, the voltage at the at the `gate' of the next one has already started changing causing it to start conducting, and so on). I also have a suspicion that there would be other real-world constraints such as cross-talk (noise between transistors) and thermal problems. I'd hazard a guess that a production-quality chip would be somewhere in the region of a tenth the speeds quoted here! However, these new materials and structures still make for an impressive speed gain over traditional Silicon CMOS designs." (The speeds quoted in this article are maximum rated *switching* speeds of a single transistor. Synchronous logic designs of the type found in microprocessors involve synchronous cells (known as flip-flops) and asynchronous gates providing boolean functions on the signals passing between flip-flops. The maximum rated frequency of any design is limited by the slowest path between flip-flops and this is what the clock signal will be set at. As the paths between the clocked flip-flops are typically anywhere between 2 and 10 logic cells deep and with each one comprising 10's of transistors (usually in complementary configuration to aid switching speed), the overall figure for an ASIC design such as a uProcessor would be at least 2-4 times slower than the maximum transistor switching speed (it's not quite cumulative, because as one transistor starts switching, the voltage at the at the `gate' of the next one has already started changing causing it to start conducting, and so on). I also have a suspicion that there would be other real-world constraints such as cross-talk (noise between transistors) and thermal problems. I'd hazard a guess that a production-quality chip would be somewhere in the region of a tenth the speeds quoted here! However, these new materials and structures still make for an impressive speed gain over traditional Silicon CMOS designs." (http://www.physorg.com/news3662.html)

Re:Safety tip

[ronanbear]

200mm is the diameter not the thickness

Here is a better article

[Monster_Juice]

This article has more of the details correct. http://www.newscientisttech.com/article.ns?id=dn93 68 [newscientisttech.com]

Re:Ah!

[furry_wookie]

Not quite... 500 Ghz (500 x 10^9) is a LONG WAY away from even the beginning of Infrared 3 TerraHz (3x10^12), and visible light does not start until about 430 TerraHz (4.3x10^14).

Re:Article was in EE Times!

[flynns]

Actually, an EE would know that the RF output on a cell phone is specifically NOT 2.4 GHz, but is actually 850/900/1300? MHz. See wikipedia for GSM and CDMA (fine, fine, and TDMA) frequencies.

Re:Why?

[doublebackslash]

I do believe that this is a DSP, or digital signal processor and hence the amount of information that can be had from a signal is dependent on the speed that the DSP runs at. It may seem overkill to sample a signal at 210x its frequency (assuming 2.4 GHz range), but that can allow for all manner of interesting signal encoding to help transmissions approach the Shannon limit and allow for more tunable transmission of data (meaning that you get the speed you need while investing as little energy as possible). Plus with a processor like that and a small antenna array one can setup a highly directional signal, saving more energy.

The why lies a few years away in implementation when the speed is brought down to production levels, but lets give credit to a bunch of scientists with to much funding, time, and liquid helium (?) on their hands. Bravo.

Re:I RTFA..

[artemis67]

The official QBert game online:

http://games.yahoo.com/games/downloads/qb.html [yahoo.com]

Re:1.2mm per cycle

[Pants75]

Hertz arn't in multiples of 1024. And your divide is the wrong way round. (c / (350,000,000,000)) * 1000 = 0.85654988 m / s 85cm ish. http://www.google.co.uk/search?hl=en&safe=off&q=(c +%2F+(350%2C000%2C000%2C000))+*+1000&meta= [google.co.uk] I think...

Re:Even faster at 0 Kelvins!

[Anonymous Coward]

Wasn't this the solution to the laser efficiency problem in 'Real Genius'? Ice is nice.

Re:10GHz Microwave?

[Ruie]

Overall, unless your goal was to build a miniature microwave (a 21st century E-Z Bake Oven?), I don't know why you'd want to use 10GHz instead of 2.4Ghz ones. The tolerances of parts in the magnetron and waveguide would have to be much tighter, I think, and this would almost certainly cause it to be more expensive.

Yes, but the heating would be more even.

Re:Ah!

[FireFury03]

After all, even 802.11 gear runs at 2.4GHz, that's enough to cook ones private parts after extended use of a laptop.

No, it isn't. 802.11 kit has an RF power output of around 100mW - absolute peanuts compared to your 800W microwave oven. The RF radiation from an 802.11 network isn't enough to cook anything.

What you might be referring to is the thermal output produced by a laptop, which is down to the CPU and hard drive rather than the 802.11 transmitter and that can cook your privates mostly through conduction, not radiation.

Re:Can these these chips do any calculations?

[FireFury03]

Actually they can't even do 1GHz at light speed. But that's why we have pipelining, and current generation have between 10-20 pipeline steps..

Not to mention that signals don't travel at c inside the chips. However, the signal path lengths can be decreased substantially by producing 3D integrated circuits. However, then heat dissipation becomes a real problem since there's more silicon for the heat to pass through before it gets to your heatsink. Of course this may not be a problem if your heatsink has a temperature of 4.5K :)

I'm curious how silicon reacts to these temperatures though - a lot of stuff becomes superconducting at such low temperatures.

Re:10GHz Microwave?

[Anonymous Coward]

I suppose you could choose a multiple of 2.45GHz and probably still have a functional product.

This is a common misconception about microwaves. 2.45 GHz isn't the best absobed. It's isn't some the frequency water vibrates at. Your article does a good job of explaining this all. To save you the time reading it, I'll quote the clearest line about this. The article you link to says, "The frequency for maximum dielectric loss lies higher than the 2.45 GHz (0.0817 cm-1) produced by most microwave ovens. This is so that the radiation is not totally adsorbed by the first layer of water it encounters and may penetrate further into the foodstuff, heating it more evenly; unabsorbed radiation passing through is mostly reflected back, due to the design of the microwave oven, and absorbed on later passes."

Miss a picture...

[lcde]

Good article but nothing beats a picture [dailytech.com] from This article [dailytech.com]