Smallest Transistor in the World

Ant wrote to tell us of a story on BBC's Web site about the world's smallest transistor. The Vertical Transistor uses the thickness of a precisely-controlled layer of material, rather than light, to set the gate size, which makes for smaller circuits. With many scientists of the opinion that current transistor technology will hit a brick wall of physics soon, the vertical transistor offers a new way to get greater processing power.

Comments

small means fast, not dense

[helge]

There are different kinds of smallness for transistors. One is the size of the active part; for MOS transistors it's the channel under the gate, which is lateral (horizontal). For bipolar transistors, it's the base region between collector and emitter, where the current flows vertically. Thus bipolar transistors are (usually) called vertical.

Smallness in the active part allows higher speed (usually referred to with the parameters ft or fmax) but it also means that the transistor breaks down at a smaller voltage. That's the reason why modern digital chips are powered by ever decreasing voltage.

But the active part of the transistor is small compared to the rest of it. It must have contacts to lead the current to and from the transistor. And then the complete transistors only occupy a small area of the chips, the most of it is used for wiring.

So just because some vertical dimension of a transistor has shrunk to 50 nm doesn't mean that you can fit very many of them on a chip. That depends more on how thin wires you can make, and how many layers. No wonder that modern chips can have as many as five layers of wiring, something that was very difficult to do ten years ago.

The scientists at Bell Labs has shrunk the active
region of what appears to be an MOS transistor. It will be fast, but the number of transistors on a chip will not increase as a result.

I found an article about the transistor at
http://www.bell-labs.com/news/1999/november/15/1 .html

Re:small means fast, not dense

[bperkins]

>That's the reason why modern digital chips are> >powered by ever decreasing voltage.

I doubt many CMOS transistors are anywhere near their breakdown regime. Reduced power consumption and heat disapation is the primary reason for reducing volatge.

> And then the complete transistors only occupy a > small area of the chips, the most of it is used > for wiring.

Wiring is done in the oxide layer above the substrate. AFAIK, most of the substrate is populated with transistors. Otherwise, you increase your line length uncessisarily, which is a disaster for gate delays.

>It will be fast, but the number of transistors on >a chip will not increase as a result.

I would disagree with this. Since the size of the active region can be smaller than your photolithography line width, the transistor can be made quite a bit smaller, since you can use your smallest photolith line width on a relatively larger feature.

Re:*cough*

[orangesquid]

I built a liquid transistor once... the recipe was the following:
3 jars
lemon juice
water
salt
copper wire
aluminum foil
I used the copper wire tips and aluminum foil pieces as electrodes. I put saltwater in the center jar and lemon juice in the left and right jars. The emitter was a copper electrode in the right jar. There was a wire running from an aluminum electrode in the right jar to a copper electrode in the middle jar. The base was a copper electrode in the middle jar. There was another copper electrode in the middle jar attached to an aluminum electrode in the left jar. The collector was a copper electrode in the left jar.
I figured using different electrodes in such substances should create diode-like behavior - especially because copper+aluminum+electrolye=very sucky battery.
After connecting a voltage supply across the emitter and base (positive on emitter I think), I connected another voltage supply's positive (I think) to emitter and put an ohmmeter between the collector and that supply's negative (methinks). I noted the resistance. I then removed the first voltage supply, and noted the resistance again. Not much different.
I swapped the polarity of the voltage supplies and repeated the experiment. 70k ohms when the supply was connected, 120k when it wasn't.
Woohoo! I had a cheap, ineffective giant liquid transistor. Completely impractical ;-)
The only problem with such liquid transistors (besides them being not very efficient): the liquids tend to pick up fun little things like fungi. I had the three jars (still full) in a box in my basement a little while ago... One day, as I was cleaning up, I looked in the box... ewww...
Yet again, another nearly completely useless device pioneered by the infamous Matt Williams ;-)
If anyone repeats the experiment with even a small bit of success (try substituting other metals - it might make it more effective) please e-mail me at orangesquid@hotmail.com - I'd love to hear about it.

--theorangesquid

Heat, Pathlength and Speed

[Baldrson]

Not being a semiconductor expert, I have to wonder:
Will these transistors have heat problems?

Typically, electrically insulating materials are also heat insulating. The vertical geometry of these transistors will result in longer heat flow paths. And to make matters worse, since the heat flow paths roughly correspond to the electrical flow paths, might there be a need for greater voltage for a given clock-rate? That would mean more power going into the transistor, and therefore more heat, as well as more difficulty in getting the heat out.

Maybe its time for the chip designers to start looking at Seymour Cray's liquid convective cooling based on Florinert.

Cray had a heat dissapation problem driven by a different kind of pathlength that affects all systems as they get smaller. In his systems, geometric optimization was a first priority because the distance between semiconductors was starting to dominate the clock rates. So he had to shrink all three dimensions. Even without increasing the number of transistors, or the power or the heat conduction problems per transistor, he still had a cubed law working against heat disappation. So he started forcing an inert liquid, developed by 3M Company called Flourinert, between the circuit boards to suck the heat out. It turns out Flourinert has an exceptional heat conductivity times electrical insulation product.

An interesting side-light: Flourinert was developed to be a blood substitute. Perhaps the semiconductor systems are acquiring a circulatory system.

Long live Moore's Law

[rde]

Just when you think Moore's Law is about to reach the wall, something happens. You'd think that by now we'd know better.
This is also a rather cool discovery. It's overcome the problem with light etching and electron leaking in one go; that's impressive.

Re:Reduce it to the Max !

[dufke]

As far as I know, current P6 and K7 cores do this as well... You need branch prediction for pipelining (sp?) to work. Otherwise the pipeline would be flushed whenever there was a conditional jump in the code.

-

Smaller transistors not necessarily faster

[gargle]

Smaller transistor sizes don't necessarily make the chip run faster. The problem is, as the transistors get smaller, the wires get smaller as well. Smaller wires have higher resistance and have more capacitive contact with surrounding wires, and so at a certain point, the delays in the wires start to dominate and the delays in the chip actually increase as the chip is minaturized.

This is a big problem in the industry today. Copper wires help because they have lower resistance than Al interconnects. People are also researching optical interconnects.

Re:Vertical transistors not entirely novel

[loki7]

Does this require significant modifications to existing fab lines, or worse, entirely new ones?

My understanding of fabs is that they're generally completely replaced every few years, anyway, as process size shrinks and other parts of the manufacturing process improve.

Besides, it's not like you're going to see Athlons using this technology before Christmas. It's going to be many years (if ever) before this is actually feasible for mass production, and I'm sure the manufacturers will have plenty of time to build new fabs.

/peter

Hmmm

[Signal 11]

He's right, you know. At the current rate even nanotech will run out of options. Shall we move to using quarks and quantum physics to keep up with Moore's Law? I can just see it now - you're playing a cool-ass game of quake, and all of the sudden *bzzzt*. Oh crap, my quantum computer just "tunnelled" to another random location in the universe! Well... atleast Intel would like a computer that did that - brings new meaning to the words "forced obsolence" doesn't it? =)
--

Re:Hmmm

[rde]

At the current rate even nanotech will run out of options.
If this new transistor can be scaled (and the chip boys 'n' girls have about ten years before it's vital), then it should push Moore's law's validity into the 2020s or 2030s. By then there's bound to be some manner of nanotech that'll improve matters; that or optical computing, or quantum computing, or...

If we truly do achieve a nano solution, then no-one'll really care about Moore's law. If you can use assemblers to build the chips, a much greater precision at existing scales will be achievable, so it'll be less vital to get smaller and fuzzier.
Of course, once true nanotech is online, we'll be all less worried about the speed of quake XI and more about adjusting to living for a couple of thousand years. I hope.

A better link

[wocky]

A more informative link is here.