Strained Silicon to the next level?

[lexxmac]

Prescott CPUs utilize strained silicon, where silicon is stretched onto a germanium lattice to allow increased flow of electrons. As far as I can find, IBM was the first to discover the process, or at least make it work. If you take a look at the pictures here, you'll notice that there is a very thin amount of silicon stretched onto the germanium in the transistor itself. Well, seeing as the reason germanium had to be used is that it share similar properties of silicon due to the nature of the +/-4 valence electrons it has.
I had always thought that the reason its so hard to produce carbon based semi-conductors had something to do with the amount of carbon needed and the wafers being cost prohibitive, seeing as you would essentially be dealing with diamond wafers. I'm not sure if this assumption is correct, but it seems logical. Now, imagine a transistor manufactured where silicon was the stretching element used to stretch carbon. As shown in the pictures from IBM, the amount of strained material needed is very small. So, the question is, why not use silicon to stretch carbon for the gates? Carbon is the simplest element with the properties needed, and it (to me) seems to be the logical next step to making some faster switching stransistors.

Now, go ahead, heap on the abuse and criticizm. 🙂

 

[CTho9305]

I believe there was recently an article published on diamond wafers. They're ridiculously expensive, and it takes years of research to actually get technology into the fab 😉.

 

[f95toli]

First of all: What do you mean by carbon? Carbon in what form?

Secondly, thin film processing is very tricky. In order to be able to grow one material on top of the other you need them to be lattice-matched and the interface to be chemically stable (no reaction or diffusion).

Another problem is that the materials might not be "proccess-compatible", it is a common problem that one the materials requires a processing step which destroys the other material or the interface.

 

[CanOWorms]

I think you're describing biaxial strain, which I think is what's in the market right now.

But, p-type and n-type need to be strained differently (I think biaxial increases the performance of one over the other, but some advantages too). Uniaxial strain allows the strain to be selectively compressive or tensile.

 

[lexxmac]

First of all: What do you mean by carbon? Carbon in what form?

I don't really understand what you are asking. What form? I would could say 'solid' as opposed to liquid/gas/plasma. I'm just unsure of what you meant. Carbon, as an element, shares a great many properties with silicon because it has four valence electrons (electrons in the outermost ring) and because of that you can for a lattice in a grid shape, just as you can do with silicon and germanium because those elements also have 4 valence electrons. I'll go as far as to say that it wouln't seem logical to use a radioactive or unstable 'form' of carbon for semiconductors.

 

[Description]

There are many forms of solid elemental carbon.

 

[white]

You'd have to use diamond, crystallized carbon. There are people making artificial diamonds but that technology is primitive. It would need to be highly reproducible in order for chip makers to invest billions of dollars to buy and incorporate these machines into their process. Assuming that we could even reliably deposit high quality diamond films onto other semiconductors there'd be major problems with the strain field produced by the hugh mismatch between the underlying layer and the diamond film. One reason they can use SiGe as the material to grow strained silicon onto is because the mismatch between SiGe and Si isn't too large. When the strain gets into the double digits you run the risk of producing misfit dislocations and other crystal defects. To put things into perspective, SiGe has a lattice parameter of 5.65Å and Si has 5.43Å. This results in ~4% strain. Diamond has a lattice parameter of 3.57Å which results in 58% strain using a SiGe thin film.

 

[f95toli]

I am not even sure it is possible to deposit dimond or even graphite-films and I think that is what you would need in this case.
I work with carbon films (we use it as a hard mask) but the carbon is amorphous so you can not deposit expitaxial films on top of it. We use an e-beam evaporation (HV).

Is it even possible to CVD carbon films?

Another problem with carbon is that it tends to form carbides if you heat it togheteher with another material. I have actually destroyed a sample by accidently heating it in a chamber which had not been cleaned proparly, the carbon became extremely hard and was impossible to remove (I guess there was Ti or something similar in the chamber).

 

[white]

Originally posted by: f95toli
I am not even sure it is possible to deposit dimond or even graphite-films and I think that is what you would need in this case.
I work with carbon films (we use it as a hard mask) but the carbon is amorphous so you can not deposit expitaxial films on top of it. We use an e-beam evaporation (HV).

Is it even possible to CVD carbon films?

Another problem with carbon is that it tends to form carbides if you heat it togheteher with another material. I have actually destroyed a sample by accidently heating it in a chamber which had not been cleaned proparly, the carbon became extremely hard and was impossible to remove (I guess there was Ti or something similar in the chamber).

that's an excellent point. the processing temperature would have a lower limit if a thin film of diamond were used as the semiconductor, assuming it were even possible.

 

[lexxmac]

the point is, that if possible, wouldn't that make the switching speed a great deal faster?

 

[f95toli]

The question is: Why bother?
There are other matierals (III-V) you can use if you need really fast devices, at some point we just have to accept that the Si technology is reaching its limits.

GaAs devices are much faster than Si, and as long as you are only want speed and not large-scale integration it works well. If you want even faster transistors there are other, more exotic, III-V that can be used (I think the current record is something like 150 GHz).

I might be worng but I am under the imperssion that the switching speeds are not much of an issue in the Si-industri, figuring out how to make even smaller devices wíthour to much heat and leakage currents are much more urgent problems.
If you want speed buy a HEMT or something similar.

 

[AbsolutDealage]

Originally posted by: f95toli
Is it even possible to CVD carbon films?

Yes, I remember reading about it when that whole artificial diamond thing hit. Here is the link for the wired article, the CVD info starts on the bottom of page 3.

Here is the US Patent for the CVD Diamond process that Apollo Diamond uses.