Transistors nearing the one terahertz barrier

Tossing the all-too-common "world's fastest" label on your latest gig seems to happen entirely more frequently than necessary, but researchers at the University of Illinois at Urbana-Champaign are make this claim legitimately. While we've seen those wee transistors ratchet up in speed, these gurus have shattered any previous records that may have been standing by crafting a transistor "with a frequency of 845GHz," which is "approximately 300GHz faster" that those built by "other research groups." While the terahertz barrier is arguably the "Holy Grail" of transistor speed, this leap forward doesn't leave them too far off from the ultimate goal. In addition to the pseudomorphic construction, the crew also used tinier components in order to "reduce the distance electrons have to travel, resulting in an increase of speed." Notably, the chip "only" runs at 765GHz while ticking along at room temperature, but chilling it to minus 55-degrees Celsius bumps it up to the record-holding 845GHz mark. Developers are quite pleased with the results, but as expected, aren't entirely satisfied, and seemingly can't wait to push the envelope a bit further and break their own record sometime soon.

[Bill Clark @ Dec 12th 2006 11:23AM]

someday we will think back on the quest for 1Thz and laugh... heck I remember people saying about a 1Gig HDD... that is all the memory anyone could ever need!

[joseph le brech @ Dec 12th 2006 11:49AM]

2020 :-

dude 1: Hehe remember when someone said that we would end up needing more than 1Thz, I've still got my Ahtlon 128, how about you?

dude 2: Core Octo

[Kevin @ Dec 12th 2006 11:57AM]

I didn't even know they got past 100Ghz...

Why does my computer still have a 2.4Ghz processor?
Something is wrong here.

[Tim @ Dec 12th 2006 12:29PM]

This... We're talking about 2.4ghz Phone-style things. Not 2.4ghz computer processors. Different.

[Sinbios @ Dec 12th 2006 6:37PM]

The 2.4GHz on a phone means the frequency of the signal is 2.4GHz, not that the phone's chips operate at 2.4GHz.

Figure out what you're talking about before trying to correct others.

[bloodnok @ Dec 12th 2006 12:49PM]

'course, this isn't about a processor core running at 1thz, it's about a transistor. but you athlon owners know that, dontch?

[Dean Kondo @ Dec 12th 2006 2:31PM]

Who said: "640k is all the memory anyone will ever need"

[estel @ Dec 12th 2006 2:35PM]

Wow. That's where my son-in-law teaches! He will be surprised I found out before he told me.

[chris @ Dec 12th 2006 5:26PM]

Cool, but there's just one thing... it's not a Silicon transistor. You probably won't be seeing it in your PC anytime soon.

[wiseone @ Dec 13th 2006 6:33AM]

@Dean Kondo

Bill Gates said that.

[byelaw @ Dec 13th 2006 10:52AM]

WiseOne, it was always thought to be Billy boy, but in the end that was an urban myth.... Bill never said it.

[David @ Dec 13th 2006 5:12PM]

Since I'm a semiconductor device engineer and used to work on compound semiconductor FETs and SiGe HBTs, let me translate a bit and clear up some confusion.

The number of 845GHz being quoted here is one of two figures of merit known as cutoff frequencies. Whether it's f-sub-T or f-sub-max (I'll use fT and fmax for short), I don't know until I read the paper. Due to budget cuts at my company, I didn't get to attend IEDM this year (although I'm a co-author on a paper being presented there).

Most likely, they are quoting fT. Imagine a transistor set up like an amplifier and applying a continuous signal at the input at one frequency. As you turn up the frequency, the current gain of the device drops. At some point, this gain drops to 1 and the device can no longer be said to have any useful gain. The frequency at which this occurs is fT. A similar figure of merit, fmax, can be defined based on power rather than on current gain.

These figures of merit do not represent the clock speed of a processor made with such a device. In fact, bipolar transistors aren't very useful for implementing digital logic in the first place because logic gates based on these devices draw current at all times and thus dissipate too much power. This fact plus the larger size of these devices compared with FETs means low density and not enough transistors on a chip to create a modern processor.

Instead, expect devices like this to appear in RF or mixed-signal applications. In such applications, you want fT and fmax to be much higher than the carrier frequency at which your system will work ... a factor of 3 or more is typical. Right now, you can buy HBTs made from silicon-germanium with fT values of 200GHz and fmax values of 275GHz being used to build 60 and 77GHz radio systems. This gives you an idea of the margin. SiGe HBTs have achieved fmax values of over 500GHz when chilled, by the way.

[Ed French @ Dec 15th 2006 6:58AM]

Hi,
I'd like to learn more about this, would you mind contacting me offline please?
[phone] or edward.french (at) enterprise-ventures.co.uk
Cheers,
Edward French

[Emmanuel @ Dec 16th 2006 12:19AM]

Damn David you're a master!! I'm Student @ DeVry College Studing EET, do you matter give out your MSN addy If you've though?