commentary What technologies are chip manufacturers using to create smaller and smaller chips?
The general perversity of the universe has conspired to ensure that I do not deliver the promised performance comparison between the AMD Athlon 64 and Intel "Prescott" this month. Let's just say that Murphy's Law has been having a field day and leave it at that.
However, as I was reading up on the specifications and manufacturing techniques for the two processors I realised that most of us simply take the CPU very much for granted. And, while it certainly is an enormous task in terms of person hours to design either of these two CPUs, the real trouble starts when you try to squeeze the designers' grandiose plans onto a tiny square of silicon.
A CPU has over 40 million transistors densely packed onto the relatively small die, so how small must each of those transistors be? The very latest processors have transistors that are around 50 nanometres (nm) wide -- 50 billionths of a metre. Reducing the size of the components has advantages other than simply being able to stuff more onto the chip: electrical signals propagate at a finite speed, the smaller the components and the interconnect distances the faster they change states.
Everyone by now is at least vaguely familiar with the process lithography where the pattern or feature to be etched is created by exposing the wafer coated with "photoresist" to a "shadow mask" of the pattern. The photoresist that was exposed to light is washed away and the silicon etched at these locations while the non-exposed photoresist protects the silicon from the etching process.
Conventional wisdom has it that you could not create features smaller than the wavelength of the light -- to create 50nm components one would need light of a shorter wavelength than 50nm. However the shortest wavelengths currently employed in the process are just under 200nm. How can this be so?
If the feature is smaller than the wavelength of the light, diffraction and distortion can occur (who can forget those physics experiments?). But, and this is the neat bit, the diffraction and distortion can be precisely calculated, and what's more the shape of the features on the mask can be modified to counteract the distortion or -- in some cases -- introduce distortion to create smaller artefacts than the mask itself.
| As chip manufacturers move to smaller features they must find ways to produce and employ shorter wavelength light.|
Admittedly, it is quite easy to produce short wavelength light but then stop and ask yourself "what type of lens do you use to focus the beam"? Depending on the wavelength of the light, some materials can absorb the light and be totally opaque or have no focusing effect at all. Then there is another problem: "is the photoresist going to work effectively with the shorter wavelength light"?
And here is yet another problem. Once you have etched the tiny features on the silicon you must then "wash" away the photoresist with solvents. As the features become smaller the viscosity of the solvent starts to become a problem, the thin solvent, at least on our macro scale, would act like thick treacle on such a micro scale, which would hamper its cleaning abilities. Also, as the features become smaller, the washing action must become increasingly more gentle.
I have looked at just a few of the problems that face chip manufactures and have not even touched on the new technologies of copper interconnects, stretched silicon, and doping technologies.
Yes I may be a bit of a techno freak but I still cannot help but marvel at the technology that creates my desktop PC.
Steve Turvey is Lab Manager of the RMIT IT Test Labs.
This article was first published in Technology & Business magazine.
Click here for subscription information.