UIUC Creates World's Fastest Transistor Again

An anonymous reader writes "The University of Illinois has developed (again) the world's fastest transistor operating at over 500 GHz. They used an indium phosphide based wafer, and super-scaled dimensions. The device kind of looks like a spaceship." Milton Feng, the professor in charge of the team behind the transistor, admits that their ultimate goal is a terahertz transistor, which given their previous achievements, doesn't sound too lofty.

Comments

Depends...

[raehl]

If it takes 12 years for these new transistors to make it into commercially available processors, then it would be spot-on with Moore's Law.

Was the fastest transistor 12 years ago 3 GHz? Probably.

Re:Slightly over optimistic

[mslj]

Indeed. There are hardly enough reference points to make a reliable extrapolation. If I extrapolated backwards in the same way, I would go negative in the nineties (and that's just impossible).

How do you measure things that fast

[azpcox]

If it's the fastest transistor out there, how can you measure teh switching speeds with something slower?

Re:How do you measure things that fast

[AFairlyNormalPerson]

I'm not sure, but if they are sampling a regular frequency, they may be able to make a measurement through the "effective interference distortion". It is in quotes because I made up the term. More generally, I'm thinking about the examination of the interference pattern of two regular plane waves - the frequency of one is known and the other is two be determined.

On the other hand, I think about the digital sampling of audio and the problem of aliasing when the sound frequncy reaches the theoretical limit of proper sampling (1/2 the sampling frequency). So... maybe the question is, how is a frequncy sampled unless the sampling frequncy is twice that of the frequency being sampled?

So the only other thing I can think of is interaction with light. Maybe they can perform some sort of absorption spectrum?

I honestly have no clue, but I hope I brought up some issues for those with more knowledge than I to discuss.

-Norm

Re:Transistor Type

[G4from128k]

This isn't a FET like the transistors found in computers (and just about everything else). This is bi-polar technology that uses much more power than FET.

True, but there are technologies that combine CMOS and Bipolar for faster CPU designs (I think BiCMOS was more heavily used back in the 90s). Also IBM is working on mixed material, mixed technology that combines SiGe bipolar chips on a CMOS silicon-on-insulator wafer. You never know what those researchers will do next.

Re:Are you ready for lots of latency?

[Anonymous Coward]

Presumably, you don't need that frequency all over the CPU. It would perhaps be good enough if you could have tiny task-specialized units performing multiple local operations within the timeframe of each CPU-wide cycle. Hell, that's how computers work today already :P

Re:Misinterpreted

[wannasleep]

Just to complement what Takahashi has said, I would like to point out that:

even if you could put them into a computer (that would consume more than the rest of the building) it wouldn't go that fast, because you need to build gates with those transistors and put some of those gates together to form a path between registries. The frequency of the computer is the inverse of the time that a signal needs to go from one register to another in the slowest path in the worst case conditions

The modern FETs actually have current flowing through the gate and the leakage is actually on its way to become the primary source for power consumption. This is due to the fact that the oxide is getting thinner and thinner and it can't make it to insulate anymore

Because of the leakage problem, we will have a change in the devices, sooner or later, although we have been saying the same thing for 20 years :)

We will have lighting revolutions next

[mnmn]

Any vibrating electric signal emits radio waves. Radio waves at higher frequencies become light.

So its interesting to see the transistors gaining higher speed. Visible light is 384 to 769 THz, so the whole circuit spontaneously glows red and passes all rainbow colors to violet, then grows dark again as we speed up the circuit. This is probably the most efficient way to produce light anyway.

So we'll have blubs that will provide us with a wide spectrum of lights just as daylight and LCD monitors with insanely high resolutions and color bits

Not to mention CPUs that emit UV light at night.

/. misinformed again ... AMD transistor faster

[GISGEOLOGYGEEK]

AMD has produced a transistor that operates at 3300 Ghz ... 6.5 times faster than the supposed record holder in the above story. Sure, its a different kind of transistor, but the headline read fastest transistor, not fastest type xxx transistor.

Check It Out

Re:don't worry about it too much

[Compuser]

Why is it so hard to make all traces on PCBs coaxial?
Yes, you would have to make the PCB and traces
in one process, e.g. on a "inkjet printer" type
manufacturing process but it is very doable.
You could then easily scale lines to 1 GHz and
if you could control tolerances to a micron you
could scale much higher. Chip packaging would
get expensive but I doubt it would add more than
$100 to the price of any given chip and maybe only
a few bucks for most 6 to 12 pin chips. So your
high-end motherboard-processor(s) combo would go
up in price but insignificantly (50%). Is anyone
doing it?

Re:Improvement rate

[igny]

Try linear regression to log(x), log(y); it ll give you power of -.4, and constant ~log(3000)

log(y)=log(3000)-log(x)*.4 (approximately)

Of course I assumed specific type of dependence, and that speed goes to infinity as the size goes to 0. The speed might as well be bounded even if size 0 is reached.