10nm chips? How many atoms per transistor?

[Beamin]

Is that the limit of thing where uncertainty starts to take over? Where will things go when you can't go smaller? 3d chips? Whats the min number of atoms to make a transistor?

[T3sl4co1l]

FWIW, an atom is about 0.2nm across, give or take.

Tim

[srce]

Whats the min number of atoms to make a transistor?

One. It was done a few years ago.

[Circlotron]

FWIW, an atom is about 0.2nm across, give or take.

Tim

How much space is there between atoms? That is to say, how may atoms side by side to make up 10nm?

[Beamin]

Whats the min number of atoms to make a transistor?

One. It was done a few years ago.

How? it must have all kinds of supporting circuitry. An atom really only has a ground state and a few oxidation states. Don't see how you could make a transistor that relies on the flow of electricity for that. That also doesn't seem practical more like a science experiment.

I know around 50 atoms the size is dominated by uncertainty and you can't rely on it behaving like its supposed to  because of tunneling. 

[T3sl4co1l]

FWIW, an atom is about 0.2nm across, give or take.

Tim

How much space is there between atoms? That is to say, how may atoms side by side to make up 10nm?

None, that's the average radius when stacked up in a crystal or stuck together in a molecule.  For more info, check out: atomic bond length, silicon lattice constant, diamond cubic crystal structure, and so on.

So, we're talking a fraction of a million atoms in one of these transistors, give or take.  (Remember 10nm is the feature size, while a transistor is a few feature-sizes across.  The thickness or height can be less, however.)

Tim

[mac.6]

remember that 10nm is not the size of the transistor, but the size of the minimal feature.
Things are pretty complicated when comparing different node/factory processes.

[apis]

Silicon, for example, forms a dimond cubic crystal structure; which means it has a repeating pattern of cells that look like this:

The side of such a cube is 0.54 nm. (As one can see, the shortest distance between two atoms is smaller than that though.)

[TimNJ]

Below somewhere like 22nm, fabs stopped using the real nm values.
10nm simply means it's better than 14nm, and 14nm simply means it's better than 22nm.

Interesting..but why? Because it's all the investors understand?

But from a technical standpoint, why can't they pinpoint the actual feature size?

[TimNJ]

But from a technical standpoint, why can't they pinpoint the actual feature size?

They can, and they provide such info in PDK, but investors want to see Moore's law to continue.

Nice. So hypothetically we might actually be at an 18nm, but the investors we'll think we're at "10nm"?

[Circlotron]

If what the investors get told is only approximately the truth, is one company’s investor “10nm” the same as another company’s investor “10nm” ?

[Beamin]

Some doping agents in silicon are as little as 1/1,000,000. That would become a problem when working with 20 silicon crystal cubes as shown above and very hardtop control. To think at one point vacuum tubes were fiddley small to make.

[NiHaoMike]

Some doping agents in silicon are as little as 1/1,000,000. That would become a problem when working with 20 silicon crystal cubes as shown above and very hardtop control. To think at one point vacuum tubes were fiddley small to make.

I remember reading about nanoscale vacuum tubes. What came of those?

[T3sl4co1l]

Some doping agents in silicon are as little as 1/1,000,000. That would become a problem when working with 20 silicon crystal cubes as shown above and very hardtop control. To think at one point vacuum tubes were fiddley small to make.

I remember reading about nanoscale vacuum tubes. What came of those?

No one's figured out cathode wear. Seems when you apply enough energy to rip electrons out, sometimes the atoms come too.

Tim