60 MA XIMUMPC NOVEMBER 2004
the microscopic scale we’re talking
about here, you could fit hundreds of
circuit lines side by side on a single
strand of a spider’s web.)
Oh, and have we mentioned that
silicon itself isn’t a naturally occurring
material? It’s actually a component of
beach sand that is scooped up (and
divested of cigarette butts and sea-
weed) and refined. Why go through
all the hassle? Because silicon is a
tremendous semiconductor—it can
alternately allow electricity to travel
through it or be made to block electri-
cal signals.
Here’s the way the entire fabrication
process works:
➤ Silicon is extracted, purified, lique-
fied, and then grown into long cylin-
ders that are roughly 300mm (about
12 inches) in diameter. A thin wafer is
cut off this silicon ingot (think salami
log) and then heavily polished, after
which it heads for a furnace where it’s
subjected to high temperatures thatwill create a layer of silicon dioxide on
its surface.
➤ Next, a material called “photore-
sist” is layered onto the wafer, and
one of the photo masks—which con-
tains all the necessary CPU circuitry
for that layer—is placed over it. The
wafer is then exposed to ultraviolet
light, which alters the chemical proper-
ties of the photoresist in the uncov-
ered areas of the mask. The exposed
photoresist is removed and the silicon
dioxide under it is etched off.
➤ Then polysilicon (a conductor)
is deposited on the wafer and new
mask layers are made. As more and
more layers are added, the final three-
dimensional CPU package begins
to evolve and take shape. Along the
way, the exposed areas of silicon
are doped—a process somewhere
between coated and saturated—with
ions that provide positive and nega-
tive charges to help the transistors
turn on and off. (That’s the great thingabout binary circuits. No matter how
complex the overall design, their
components only need to be in one
of these two states—“off” or “on”—at
any time.)
➤ Small windows, the result of etch-
ing from what’s called “reveal masks,”
are formed through the layers dur-
ing each step of the process so they
can be interconnected as needed to
become complete circuits. Copper is
used to fill in the windows and actu-
ally make the connections. On CPUs
that pre-date the current 0.13 micron
process, the filler of choice was alumi-
num. Copper has better thermal con-
ductivity, though—it’s about 60 percent
more potent than aluminum, which
not only saves energy but also speeds
up heat dissipation.
In order to withstand the rigors of
the photo-etching process, a wafer is
initially cut thicker than it will need
to be when it becomes the final CPU.
Once the process is completed, about ËCHIP LAYERS AND
WHAT THEY MEAN
Today’s cutting-edge
processors fit millions
of transistors into a
single CPU with
only eight or
nine layers. This
illustration, adapted
from an actual
electron microscope
photograph of a CPU
cross-section,
shows how these
layers interact.
Each layer (1) is
called an interconnect
layer, wired to
transport data from
one part of the chip to
another (these layers
get thicker near the
top of the chip). The
small posts you see
between the layers
(2) are called vias,
which carry data up
and down through the
interconnect layers.
The light blue squares
denote operational1
2
illustration by Brandon Voss
areas of the CPU; the dark blue
areas are transfer lanes for the data.