Transistor Breaks Speed Record

Thanks to funding by the U.S. Defence Advanced Research Projects Agency, the record for the world's fastest transistor has been broken once again by U.S. researchers. The new transistor, made of iridium phosphide and iridium gallium arsenide, has a frequency of 845 GHz (845 billion cycles a second) beating last year's speed record of 704 GHz. 845 GHz is achieved at -55 C while at room temperature (25 C), the transistor operates at 765 GHz.

The scientists at University of Illinois Urbana-Champaign are credited for the achievement and they believe that a tetrahertz, one trillion hertz, is now within reach. Speed is, however, not the only achievement; engineering techniques have also improved, allowing even smaller parts for the transistor (the transistor in question is about 12.5 nanometres wide).

This just goes to show you, the potential for much faster computers and more secure wireless communications is constantly evolving.

News source: CBC News

Report a problem with article
Previous Story

Next OS X rumored to feature new GUI

Next Story

ATI Catalyst 6.12 for Windows

24 Comments

Commenting is disabled on this article.

does anyone know what the theoretical limit is of the transistors currently used by AMD and Intel, under lab conditions... cooled & isolated to just one ... etc. That would give me more of a comparison.

I don't know since I don't work in the Oregon R&D facilities but I have heard they have gotten then TINY compared to units today. I'll keep you in suspense and won't spill any numbers...

Indeed analogue devices,

for example, the 5 GHz 802.11a band.

Also think of the power consumption (power is proportional to clock speed). Then you also need the front side bus, and memory (RAM, HDD, L2cache, etc...) access speed. You simply can't put in a 1 THz chip into your LGA775 slot and expect it to go that fast. There are many other bottle necks in the computer that must be solved first.

Remember L2 cache is DRAM which isn't made up of transistors like registers are.

As for your kids with 3DTV, I doubt it, we've had colour TV for several decades, and high definition for about 7 years and that is only starting to become affordable now. The next twenty years (when you will likely have kids) will be spent developing HDTV in the market, and new display technologies such as OLED, and others.

Two points on the speed here:

1. This is a single transistor. To get an operational unit, you have to get a lot of transistors to operate at the same time, synchronized, which slows them all down.

2. These kinds of high-speed transistors often end up in analog devices, not digital. They're in radio systems to allow for higher frequencies to be used -- the faster the switching speed, the higher the frequency that can be used.

Amazing! The future will be very interesting. It's scary, my dad mentioned to me how he never saw color tv when he was growing up and when I grew up I did, and he said that my kids will probably see something different than what we see today, I'm guessing 3DTV (no glasses needed) will be the next big thing, which comes out in 2008/9. But what after that, holographic tv?

17% increase in speed, but lowering the temp by 80 C. I could live with a chip only running at 83% (like a 3Ghz running at 2.5Ghz without the need for a noisy fan or in this case a supply of liquid nitrogen).

Faster transistors are not always a solution, but multi core together with multi core smart applications is the way to go.

Exactly.

Also, I'd like to point out that "tetrahertz" is not a word. The correct term should be terahertz. (Terabyte, teraflop, come on, guys.)

To address a few comments in here about why Intel can't just buy this technology...

First, it's not a matter about cost because Intel invests BILLIONS in R&D. Think about this... Fab32 (the new fab in Chandler, AZ that is under construction) broke ground November 2005. The estimated investment cost of the fab construction and all of its tools is about $3.8 billion. You're talking about a HUGE investment that will not even begin to see a return until 2-3 years later when the factory is running at full capacity. So there is no question that Intel can absorb the cost of purchasing research like this...

Secondly, technology like this is still in R&D. This is not something you can just buy and all of a sudden throw into high volume manufacturing. I don't expect you to understand what it takes to get the technology behind a fab up and running for high volume but all I can say is that me working at Intel, it still amazes me to this day how it all comes together. It takes a TREMENDOUS effot to perfect your process technology and tool sets (which by the way also have to go through R&D to develop a tool to handle this type of technology) in order to go into high volume manufacturing. Intel makes millions and millions of units in a short period of time.

The step from a labratory test such as the one the article describes into high volume manufacturing is an ENOURMOUS one! Don't beleive me? If you live in Phoenix, I'll give you a tour of the fab and you'll see why...

Still, I'll bet that you'd love to get a close look at how these University of Illinois guys did what they did. I know I would.

Quote - ECEGatorTuro said @ #5
To address a few comments in here about why Intel can't just buy this technology...

First, it's not a matter about cost because Intel invests BILLIONS in R&D. Think about this... Fab32 (the new fab in Chandler, AZ that is under construction) broke ground November 2005. The estimated investment cost of the fab construction and all of its tools is about $3.8 billion. You're talking about a HUGE investment that will not even begin to see a return until 2-3 years later when the factory is running at full capacity. So there is no question that Intel can absorb the cost of purchasing research like this...

Secondly, technology like this is still in R&D. This is not something you can just buy and all of a sudden throw into high volume manufacturing. I don't expect you to understand what it takes to get the technology behind a fab up and running for high volume but all I can say is that me working at Intel, it still amazes me to this day how it all comes together. It takes a TREMENDOUS effot to perfect your process technology and tool sets (which by the way also have to go through R&D to develop a tool to handle this type of technology) in order to go into high volume manufacturing. Intel makes millions and millions of units in a short period of time.

The step from a labratory test such as the one the article describes into high volume manufacturing is an ENOURMOUS one! Don't beleive me? If you live in Phoenix, I'll give you a tour of the fab and you'll see why...

As an Intel Intern in the UK I have to agree 100%
It's amazing just how much work goes into it.

hmm its a bit confusing.. if we get a spped of 800+ GHz, then why cant Intel or AMD come up with processor more then 3GHz.. 4 or 5GHz ? if I am not mistaken theres a issue with the speed transistors reaching or something.. but not sure.. if someone can throw some light on it, will be really good :).

The clock speed of a processor is not the same as the frequency of the transistor. It takes a lot of transistors to perform simple actions and just because a single transistor can do 800+ ghz doesn't mean that a cpu clock can go from 3ghz to 4ghz in just a snap.

Quote - tony-inpo said @ #3
How many FLOP does it do though? That's what is really interesting

It's still just a single transistor, so it's not really practicable to do any computing with it.

why dont intel just buy this technology and develop it and release it say in 3yrs, a 845ghz is a HELL of alot faster than the current 3ghz core 2 duo's!!!

because this technology is in research and can't be mass produced yet. Offcourse intel, amd, samsung and every chip maker out there are looking for ways to make their transistors faster and smaller, what did you think?

This was just a single transistor and they still had to have it at -55 C to operate. If you can think of a way to pack hundreds of millions of them together on a tiny die and run them at 845GHz with normal cooling I'm sure Intel would love to speak with you. :P

It's a 3-5 based material, it's hella hard to implement that into the current group 4 based semiconductor industry, even if the device is reliable at room temperature.

However, uber-fast materials like this can get implemented into Telecom much faster.