How
is a CPU created?
It
is quite a miracle how from a pile of sand engineers are able to create
something as complicated as a processor chip like the AMD Athlon or
the Intel Pentium III. Many steps are needed to create such a device.
This article will try to explain how a chip is build from sand to silicon.
Introduction
Technological evolution towards chips started in the early
1940's. In those days integrated circuits were mostly made from Germanium.
Germanium is an element that crystallises in a diamond-like way and
has excellent semiconductor properties. However due to other properties
of germanium above 70ºC (343K) germanium devices are unusable.
Around 1950 researchers started giving up on germanium and started looking
for better solutions. The answer turned out to be silicon.
From Sand to Silicon
Silicon turned out to be an excellent element for building
chips. It has good semiconducting properties, has the ability to form
a stable controllable oxide film on it (we'll see that this is very
important) and is available about everywhere in the form of sand. High
purity silicon is obtained from two common materials: silicon dioxide
(SiO2) and carbon. In a high-temperature environment
(~2000ºC) the carbon reduces the SiO2 to elemental
silicon (Si) and CO2. Because this process results
in about a purity of 90% an extra step is required. The silicon is transformed
to SiHCl3, which can be purified by chemical means.
I won't go into this step because it is probably more interesting for
chemistry-minded people than CPU Site readers. But anyway the result
is pure silicon composed of many little (micrometer-sized) crystals.
This is called polycrystalline silicon.
Poly to Mono
Silicon needs to be mono-crystalline, that is a nearly
perfect large (10-15 cm) crystals in order to be suited for the creation
of so-called wafers. There are two ways to do this. The first way is
the Czochralski-method. This method is the most common way to create
mono-crystalline crystals. Pieces of polycrystalline silicon are melted
in a crucible. The atmosphere is usually argon in order to prevent reactions
between air and the silicon. Also the temperature is kept only a little
bit above the melting temperature of silicon (1412ºC) to prevent chaotic
turbulence effects in the melted silicon. As a final precaution the
crucible is covered with silicon dioxide, which prevents unwanted reaction
of the silicon and the material of the crucible.
A so-called seed-crystal is now inserted in the melt, while slowly rotating.
The melt is often rotated in the opposite direction. Then the seed-crystal
is withdrew out of the melt a small file of perfect mono-crystalline
silicon is form on the seed-crystal. See figure 1.
|
|
|
Figure
1, Czochralski-method
|
There is one problem
to this technique. The SiO2 from the crucible
mixes a bit with the smelted silicon. The result is extra oxygen in
the mono-crystalline silicon. This results in the silicon having a lower
electrical resistivity than pure silicon would have.
For most circuits this doesn't matter, but in those cases that it matters
there is a second method: The float-zone method.
With this method the polycrystalline silicon is not melted completely
but is held in vertical position while a melted zone (1-2 cm long) passes
from bottom to top. When the melted region is heated a seed crystal
is inserted similar to the previously described process. Because no
crucible is needed no oxygen contamination occurs which results in a
decrease of about 99% oxygen compared to the Czochralski-method. However
because this method is more expensive it is only used when high-resistivity
silicon is required.
Wafers
After the production of the mono-crystalline silicon, it
is sawed - usually with a diamond saw - into circular wafers. The wafers
are afterwards chemically edged to remove sawing damage. After the edging
the wafers are polished until a defect-free mirror-like surface is obtained.
Now the wafers are ready to start making circuits.
Oxidation
As
explained one of the most important qualities of silicon is the ability
to form a controllable oxide film on top of the silicon. This is important
to protect the silicon from contamination and even more important will
enable engineers to create circuits. More, later on this in the lithography
part. There are two ways to form an oxide layer: via dry oxide or via
steam. For those interested here are the chemical reactions (they are
quite simple actually.)
Steam requires extra safety procedures to handle the explosive H2
gas that is created, but is usually more efficient. Also because oxidation
is time-consuming high-pressure environments are often used.
Lithography
This is what it really comes down too. Now we have (almost)
perfect silicon with a nice protective layer, but it is still far away
from being an Athlon or a Pentium chip. We must now selectively remove
parts of the protective SiO2 layer in order to
create circuits on those parts. There are various ways to do this.
|
|
|
Figure
2, Lithography Step 1
|
Usually this is
done with photolithography. In this case selective removal is accomplished
by the use of a light-sensitive polymer materials. The structure of
those materials will change when exposed to ultra-violet light. The
best way is to simply show a picture, see figure 2.
First a layer of a light-sensitive polymer is created on the silicon,
simply by putting a few drops of the usually liquid material on the
wafer and then rotate the wafer very fast. Afterwards a mask (or in
this case photomask) is placed over the wafer and aligned with a microscope.
This mask contains holes as shown in figure 2. These holes are of sizes
around for instance .25 micron for a K6-2 and .18 micron for a Pentium
IIIE. So these are the values you read all about when someone tell you
a chip is x micron.
|
|
|
Figure
3, Lithography Step 2
|
The mask enables
engineers to let the UV-light change only some the parts of the resist.
After the mask is removed again the changes in the resist enable the
engineers to selectively remove some parts of the resists, because the
changed parts, for instance, respond differently to specific chemicals.
So the pattern of the mask can be transferred to the resist, see figure
3.
Now we can again use chemicals -- yes, building chips involves a lot
of chemistry -- to transfer the pattern to the protective SiO2-layer.
Again by using, for instance, chemicals that react with SiO2
but not with the resist.
|
|
|
Figure
4, Lithography Step 3
|
Afterwards the resist
is removed -- again with chemicals -- and the result is figure 4.
Now we are ready with the lithography.
Instead of chemicals it is also possible to transfer patters via electron,
ion or even X-ray beams. One company specializing in lithography-techniques
is the Dutch ASML (40% of the world-market in 1998), whose equipment
is used by AMD.
Dopant addition
Now it is time to add some extra material to the silicon
in order to form devices. There are various ways of adding a so-called
dopant material to the silicon. The most common are ion-implantation
and gaseous deposition. In the case of ion implantation the dopant atoms
are first ionised (simply said, an electron is added or removed).
|
|
|
Figure
5, Dopant Addition
|
With the help of
an electric field the ions can be pushed into the silicon. With gaseous
deposition the dopant is inserted with the help of a gas that is blown
over the material. Because both these methods only bring dopant material
to places close to the surface, an extra step is required to get the
dopant in place. Usually the device is heated to drive the dopant in:
drive-in diffusion. Well how does all of this look like in a figure?
See figure 5.
Now the protective
SiO2 layer comes into action. The dopant atoms
will not diffuse in or through the SiO2 but will
in the Si. Therefore special regions are created where extra dopant
atoms are diffused in the silicon. These dopant atoms change the electrical
properties of the device at that place. Afterwards metallic connects
can be made and the device is almost ready.
Finishing
touch
When metallic contacts -- usually aluminium although copper
technology is ready to take over in the near future -- are added to
the device the result would be like in figure 6.
Afterwards the chip, which contains many devices like in figure 6, is
usually packed inside a protective metal, plastic or ceramic cover.
So the chip you see is actually only the cover! If you would, for instance,
break off the metal plate on top of a K6-2 you'd see the real chip under
it.
Of course our sample device is not really complex. Real-life devices
usually are made with various types of dopant and those are also placed
at various places in the device - sometimes even not even connected
to the surface. And finally the concentrations vary.
|
|
|
Figure
7, MOSFET Transistor
|
In figure 7, for
those really interested, is a very commonly used device: the MOSFET
transistor.
Do not worry if you do not have the slightest idea of what this is supposed
to do, it is just to show what a bit more complicated devices look like.
The idea in this device -- for those that want to know -- is that by
changing the voltage at the places 1 to 4, the current through the channel
can be regulated.
Final Words
Of course this short article only shows the tip of the iceberg
about how a chip is created, but I think I managed to explain the basics.
For those that want to know more I could recommend the book "Device
Electronics For Integrated Circuits" by Richard S. Muller and Theodore
I. Kamins, (John Wiley & Sons, Inc. - ISBN 0-471 88758-7).
